Introduction to Applied Physics

Course code 31IAP

J.H. Deijkers
H. Gelderblom
A.J.M. Mackus
H.C.J. Mulders
C. Visser

Department of Applied Physics and Science Education

email: 31IAP@[Link]

2024
All figures without a reference are own images, made by the teachers of this
course.
Contents
1 Variables and dimensions 3
1.1 Dependent and independent variables . . . . . . . . . . . . . . 3
1.2 Dimension analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Strategies for mastering physics . . . . . . . . . . . . . . . . . . 11
1.4 Sketching graphs of functions . . . . . . . . . . . . . . . . . . . 14
1.5 Mathematical notation . . . . . . . . . . . . . . . . . . . . . . . 19

2 Vectors, coordinates and motion 22

2.1 Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2 Lines and planes . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3 Coordinate systems . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4 Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

3 Evolution equations 38
3.1 Water tank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2 Heating/cooling object . . . . . . . . . . . . . . . . . . . . . . . 42
3.3 Differential equations . . . . . . . . . . . . . . . . . . . . . . . . 44
3.4 Newton’s second law: dynamics . . . . . . . . . . . . . . . . . . 52
3.5 Electrical circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.6 Exponential decay and characteristic timescales . . . . . . . . 63

4 Physical models 65
4.1 Defining a physical model . . . . . . . . . . . . . . . . . . . . . . 65
4.2 Nondimensionalization of evolution equations . . . . . . . . . . 66
4.3 Simplifying evolution equations . . . . . . . . . . . . . . . . . . 72
4.4 Scale models and fluid mechanics . . . . . . . . . . . . . . . . . 76

5 Elementary oscillations and mechanical waves 78

5.1 Harmonic oscillations . . . . . . . . . . . . . . . . . . . . . . . . 78
5.2 The pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.3 The mass-spring system . . . . . . . . . . . . . . . . . . . . . . . 87
5.4 Effect of damping . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.5 Effect of forcing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.6 Population model . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.7 Mechanical waves . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6 Graphical and numerical analysis of evolution equations 105
6.1 Graphical analysis of evolution equations . . . . . . . . . . . . 105
6.2 Numerical analysis of evolution equations . . . . . . . . . . . . 119

7 Analysis of stationary points and their stability 123

7.1 Stationary points and how to find them . . . . . . . . . . . . . 123
7.2 Analysis of the stability of stationary points . . . . . . . . . . 125

8 Non-linear dynamics: from order to chaos 137

8.1 The trouble with non-linear systems . . . . . . . . . . . . . . . 137
8.2 Bifurcations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
8.3 Catastrophe & hysteresis . . . . . . . . . . . . . . . . . . . . . . 141
8.4 Non-linear oscillations . . . . . . . . . . . . . . . . . . . . . . . . 146
8.5 Limit cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
8.6 Chaos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
8.7 Summary of important concepts . . . . . . . . . . . . . . . . . . 165

9 Exercises 167
9.1 Exercises chapter 1. . . . . . . . . . . . . . . . . . . . . . . . . . 167
9.2 Exercises chapter 2. . . . . . . . . . . . . . . . . . . . . . . . . . 175
9.3 Exercises chapter 3. . . . . . . . . . . . . . . . . . . . . . . . . . 179
9.4 Exercises chapter 4. . . . . . . . . . . . . . . . . . . . . . . . . . 190
9.5 Exercises chapter 5. . . . . . . . . . . . . . . . . . . . . . . . . . 194
9.6 Exercises chapter 6. . . . . . . . . . . . . . . . . . . . . . . . . . 204
9.7 Exercises chapter 7. . . . . . . . . . . . . . . . . . . . . . . . . . 209
9.8 Exercises chapter 8. . . . . . . . . . . . . . . . . . . . . . . . . . 215

A References and recommended literature 223

A.1 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
A.2 Recommended literature . . . . . . . . . . . . . . . . . . . . . . 223

B Information about [Link] 225

B.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
B.2 A peek under the hood: setting starting values . . . . . . . . . 226
B.3 A peek under the hood: calculating . . . . . . . . . . . . . . . . 226
B.4 Visualizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227

C Greek alphabet 229

Introduction
In this course you will learn to investigate dynamic processes in physics (and
in other disciplines) using various mathematical and analysis tools. We will
give an introduction to all sorts of physical (and non-physical) systems based
on basic principles. Especially vibrations, waves, instabilities, catastrophe
theory and chaos are topics to be discussed. Figure 0.1 gives an overview of
the main elements of the course.

Interpretation

Variables, dimensions, Evolution

nondimensionalization Mathematical equation
Physical reality Analysis
formulation

Physical phenomena Mathematical tools Analysis tools

• Elementary oscillations • Vectors • Analytical solutions
• Growth and decay • Polar coordinates • Numerical solutions
• Electrical circuits • Differential equations • Phase diagrams
• Mechanical waves • Taylor series • Equilibrium states
• Non-linear oscillations • Linearization • Stability analysis
• Chaotic behavior • Insight

Figure 0.1: Overview of the course.

As a starting point, we will show that by making a list of all relevant variables
in a certain system based on their dimensions, we can derive relationships
between the dependent variables (those that describe the dynamic behavior
of the system) and the independent variables (those that define the system,
e.g. lengths, mass, gravitational acceleration). Such an analysis is elegant
and often very effective. Next, we will introduce the use of vectors, as well
as Cartesian and polar coordinate systems to describe motion in space. This
will serve as an administrative framework that will be used in many future
courses.

Then we will focus on a precise, mathematical formulation of the system,

such that a description of the dynamic behavior becomes possible. What we
want to obtain in particular is a description of the time dependence of the
dependent variable(s). For example, think of the time-dependent excursion
of a swaying pendulum, or the current in an electric circuit. We call this the
setup of the evolution equation for the dependent variables.

1
Besides a focus on physics, we want to make several mathematical tools ‘op-
erational’ by applying them in these analyses and descriptions. For example,
things like the aforementioned vector calculus & coordinate systems, sketch-
ing functions, simple differential equations, Taylor series, linearization, etc.
will be introduced and treated in an elementary manner, in order to subse-
quently apply them in the analysis of the aforementioned physical systems.
We do not aim to present derivations and formal proofs of these mathemati-
cal matters - this will happen in the mathematics courses that will run more
or less in parallel with this course.

As the overview illustrates, the mathematical formulation allows for the use
of a variety of analysis tools to obtain better understanding of the system.
The acquired insights thereby help in the interpretation of the physical real-
ity.

Many (physical and non-physical) systems show - when they are brought out
of their equilibrium state - oscillatory behavior. We will pay special attention
to this: besides linear behavior at small oscillation amplitudes we will also
investigate non-linear oscillations at larger amplitudes.

In addition, we will dive into the phenomenon of ‘instability’ and into catas-
trophe theory, where sudden changes in the behavior of the system are being
analyzed. These phenomena show up in numerous systems, varying from
simple mechanical systems to biological populations, economical and socio-
logical systems, and even to the dynamics of an arms race between two hostile
superpowers. Special attention will be given to non-linear oscillations and
to chaos.

These lecture notes are partly based on text written by the previous teachers
of the course Variables, Dimensions & Dynamics, GertJan van Heijst and
Léon Kamp.

Sanne Deijkers, Hanneke Gelderblom, Adrie Mackus, Hjalmar Mulders

August 2024

2
1 Variables and dimensions
In this chapter we will delve into the analysis of various physical systems.
To this end we will not use a detailed description of the dynamic behavior
governed by physical principles or ‘laws’, but instead use information about
the relevant variables in the system and their dimensions. We will start with
several examples of physical systems, but we will also focus on non-physical
systems later on. In addition, we will provide a framework for mastering
physics in a systematic way and discuss how to sketch graphs of functions.

1.1 Dependent and independent variables

When discussing physical phenomena, we often start with describing a phys-
ical system. This can, for example, be an object in a gravitational field,
a photon interacting with a surface of a specific material, or the weather
in a small region. With a system we mean a part of the physical universe
that is under study. By limiting ourselves to the system, the influence of
the environment of the system is ignored, mostly to simplify the situation.
When analyzing a system we mean to identify the variables that define a
system, and to find relevant relationships between those variables. An im-
portant question is typically which variables are constant (i.e., do not vary
in the physical system) and which variables are dependent on other variables.

As an example, we consider the well-known pendulum, consisting of a point

mass m connected with a thin cord or rod of length l to a suspension point
O (see figure 1.1).

O
g

θ
l

Figure 1.1: The pendulum.

3
Under the influence of gravity the mass can sway back and forth around
its equilibrium position (θ = 0). We assume that the cord (or rod) is rigid
and massless, that there is no air resistance, and that the friction in the
suspension point O is negligible.
When we give the pendulum an excursion from its equilibrium position and
release it, the mass will oscillate around the equilibrium position. The ‘ideal’
pendulum would keep on swaying infinitely long, with a certain frequency ω.
When there is friction in the system (and this will of course always be the
case to some degree), the motion will be damped and after some time the
mass will be back in its equilibrium position and stay there.

We can describe the position of the pendulum (mass + rod) with the angle θ
between the pendulum and the vertical (see figure 1.1). During the motion
this angle θ changes continuously, from positive to negative values. In other
words: the angle is a function of the time: θ = θ(t).
We could measure this time dependence with an experiment quite easily.
If we would then plot the angle θ (vertical axis) as function of the time t
(horizontal axis) we obtain the function θ(t) in a graph form (see figure 1.2).
This graph of θ(t) shows periodic behavior: the mass sways back and forth
between the angles θ = θ0 (initial excursion) and θ = −θ0 with a certain period
T.

θ
θ0

t
-θ0

Figure 1.2: The excursion angle θ of the pendulum as a function of the time
t.

What are the relevant variables (or parameters) in this physical system?
Figure 1.1 shows: mass m, pendulum length l, and the gravitational acceler-
ation g. The ‘behavior’ of the pendulum is described by the excursion angle

4
θ, which shows periodic behavior. A characteristic quantity for this oscilla-
tion is the timescale T, i.e., the period of the motion. Sometimes we prefer
to use the angular frequency, which is defined as ω = 2π/T .
The physical parameters m, l, g we call the independent variables, whereas
the period T (or the frequency ω = 2π/T ) and the angle θ are dependent vari-
ables that describe the dynamic behavior of the system in the non-equilibrium
state.

We distinguish the variables as follows:

Independent variables do not dependent on other variables. They are

typically constants in the defined physical system, and describe inher-
ent properties of (components of) the system, such as:

- mass
- sizes (length, area, volume)
- gravitational acceleration
- heat capacity
- density

Physical constants can also be considered as independent variables,

such as:

- the speed of light c

- the Planck constant h.

Dependent variables depend on one or more independent variables, and

describe the dynamic behavior (e.g., the reaction to an excursion from
the equilibrium position, or to a driving force) of the system, such as:

- typical timescale, e.g. period

- oscillation frequency
- timescale of the damping
- resulting velocity

Experiments are typically about the study of an dependent variable, by in-

vestigating the influence of a certain independent variable. For example, you

5
can do an experiment to find out how the period T (dependent variable) of
a pendulum depends on the rod length l (independent).

The time t does not depend on other variables and can therefore be consid-
ered as independent variable. One of the first experiments to characterize
a physical system is typically to measure a dependent variable as function
of time, as in figure 1.2. Since the variable time does not really describe
inherent properties of a system, it is not considered in dimension analysis as
treated in the next section.

1.2 Dimension analysis

We are now going to address the question: can we establish a relationship
between the period T and the independent variables m, l, g?

We could make an analysis of the mechanical system by looking at the forces

and resulting accelerations and so on. We will do this later, but for now
we will be using another approach: we are going to look at the dimensions
(units) of the independent and dependent variables.

For the mass we use kilograms [kg], a length is expressed in meters [m], the
time in seconds [s], and an acceleration in [m s−2 ]. So, our inventory is:

• independent variables:
m [kg]
l [m]
g [m s−2 ]
• dependent variables:
T [s]
θ [-]

The dimension of the period T is [s], so no mass unit [kg] occurs there. Since
the only variable that has this dimension [kg] is m, we can conclude that the
period T will not depend on the mass m.

The only independent variable that contains the dimension [s] is the gravi-
tational acceleration g, with dimension [m s−2 ]. So, we can conclude that

6
 √
T ∼ 1/ g,

where ‘∼’ is the proportionality symbol.

The dimension of the right hand side of this equation is [s/m1/2 ], whereas
the period has dimension [s]. The only way we can change the right hand
side to something that also has this dimension [s], is by combining it with
the only other variable that has the dimension [m]: the length l. This gives
us:

√
T∼ l/g. (1.1)

So, the period T (the characteristic timescale of the motion) is proportional

to the combination (l/g)1/2 , which also has the dimension [s].
By using this analysis based on the dimensions of the variables in the physi-
cal system we are able to derive a relationship between the dependent variable
and the independent variables that define the system.

Example: a leaking barrel

Consider a barrel filled with water, as sketched in figure 1.3. The water is
flowing out of the barrel with a velocity V because of a hole at the bot-
tom. Since this hole has a diameter that is small compared to the horizontal

ρ
H
V

Figure 1.3: The leaking barrel.

diameter of the barrel, the water level decreases very slowly. So, in good
approximation, we can view the height H of the water in the barrel as a
constant. The density ρ of water has the dimension of mass per unit volume,
so [kg m−3 ]. We now want to derive a relationship between the velocity V

7
(the only dependent variable in this problem) and the independent variables
ρ (density), g (gravitational acceleration), and H (water level).
The variables in this problem have the following dimensions:

• independent variables:
ρ [kg m−3 ]
g [m s−2 ]
H [m]
• dependent variable:
V [m s−1 ]

Given the dimensions of V and ρ (the latter contains the mass dimension
[kg]), V will not depend on ρ. That leaves us with g and H, and the only
combination that yields the dimension [m s−1 ] is the following:

√
V ∼ [m s ]
−1
gH. (1.2)
So, the velocity is not dependent on the density ρ, and scales with the square
root of the water level H.

This example illustrates again how powerful this analysis based on dimen-
sions is: we are able to derive relevant relationships without knowing the
dynamic details of the system!

Recap: we are performing the following steps:

• determine which variables (physical parameters) are relevant;
• distinguish between independent and dependent variables;
• determine the dimensions of each variable;
• determine, based on the dimensions, the relationship between the de-
pendent and independent variables.
NB
The constants of proportionality in equations (1.1) and (1.2) can only be
determined by performing an exact mathematical analysis of the system in
question.

8
1.2.1 Systematic dimension analysis
So far it was relatively straightforward to find a relationship between the vari-
ables of a specific system. However, when multiple variables are involved, it
can sometimes become quite hard to perform dimension analysis. We will
now introduce a systematic method to derive relationships between variables
using a mass-spring system as the example.

Consider the mechanical system consisting of a mass m connected to a

spring (see figure 1.4). In the equilibrium position the mass is positioned
as sketched.

g
α

m
Figure 1.4: The mass-spring system.

In this case the weight of the mass (being mg) is in equilibrium with the
upward pointing force of the spring. In good (linear) approximation we can
write the magnitude of the spring force as:

spring force Fs = α × extension, (1.3)

where α is the so-called spring constant. The dimension of the force is [N] =
[kg m s−2 ] and that of the extension is [m]. Consequently, the spring constant
has the following dimension: [kg s−2 ].
If we pull downwards on the mass, the spring force will increase and therefore
will also grow bigger than the weight mg. Upon release the mass will begin

9
moving upwards, and will subsequently begin a vertical oscillation. In the
absence of friction this oscillation will be undamped, and the vertical position
of the mass will progress similar to the excursion angle θ(t) of the pendulum
(see figure 1.2).
So, the oscillation of the mass-spring system is also characterized by a timescale,
the period T of the oscillation.
In this mechanical system we can distinguish the following variables and di-
mensions:

independent variables
m [kg]
g [m s−2 ]
α [kg s−2 ]
dependent variable
T [s]

Question: how does the period T of the mass-spring system (see figure 1.4)
depend on the independent variables m, g, and α?

As a starting point, we can write the relationship as a function with the

independent variables as factors with powers:

T ∼ ma1 ⋅ ga2 ⋅ αa3 , (1.4)

in which a1 , a2 , a3 are constants. Our task becomes to find those constants.

We are going to use that the left hand side should have the same dimension as
the combination of variables at the right hand side, which gives the following
equation:
a a
[s] = [kg] 1 ⋅ [m s−2 ] 2 ⋅ [kg s−2 ] 3 .
1 a
(1.5)

This can be worked out as three equations corresponding to each of the three
units:
[kg] ∶ 0 = a1 + 0 + a3
[m] ∶ 0 = 0 + a2 + 0 (1.6)
[s] ∶ 1 = 0 − 2a2 − 2a3
These are 3 equations with 3 unknowns. Working out this system of equations
gives the solution

10
 a1 = 1/2, a2 = 0, a3 = −1/2, (1.7)

corresponding to the relationship:

√
m
T ∼m 1/2
⋅g ⋅α
0 −1/2
.= . (1.8)
α
We can easily check whether this relationship is correct
√ by evaluating the
√m
dimension of the result. The dimension of α is [ kg s−2 ] = [s].
kg

- Movie1 - Dimension analysis: mass-spring system

1.3 Strategies for mastering physics

One of the goals of this course is to learn to systematically approach prob-
lems in physics. A useful method to do physics exercises in a systematic way
is the ISEE approach.2 The letters in ISEE stand for the main steps of the
approach, which are Identify, Set up, Execute and Evaluate. While such a
step-wise approach may appear unnecessary for simple problems, you will
experience that the approach can help you to determine how to get started
and to more quickly obtain the required insights to tackle a problem.

I - Identify

In this step you identify all relevant information that is needed to solve the
problem. Try to answer the questions: What is the target variable? What
are known variables? What are relevant concepts (e.g., dimension analysis,
energy conservation) you can use as the starting point? Can you estimate
what the answer roughly should be? How do you expect the physical system
to behave when one of the variables is changed?

1
As supplementary study material, you can find videos on Canvas in which we briefly
explain important concepts.
2
The use of the ISEE approach is inspired on the book “University Physics” by Young
and Freedman.

11
S - Set up

To set up the problem, it always helps to make a drawing of the situation.

Based on the identified concepts, you can select which equations you need to
use to solve the problem. Check whether those equations, together with the
known variables, can result in the target variable.

E - Execute

In this step you do the actual math to solve the problem.

E - Evaluate

Although the Execute step should already give you the answer, you are not
done yet until you have checked whether your answer is correct. Is your
answer consistent with the estimations and expectations formulated during
the Identify step? Does your answer have the correct dimension? In case
your answer is an algebraic expression, you can evaluate what happens if you
take extreme values for one of the variables. Also check whether you found
a realistic value. For example, if you find a velocity that is larger than the
speed of light, you for sure made a mistake somewhere.

Connected to the theme of this chapter, one of the most common ways of
evaluating your answer is to do dimension analysis as a consistency check.
The basic idea here is as follows: if an equation describes a physical process,
then all terms separated by a +, −, = or ∼ sign must have the same dimensions.
Consider for example that you found for the pendulum
√
g
T∼ . (1.9)
l
At first sight, nothing seems wrong with this equation. However, since the
time at the left hand side has dimension [s], g is expressed in [ms−2 ], and l
in [m], equation (1.9) basically tells you that
1
[s] = [ ] (1.10)
s

12
which is clearly not correct. In addition, you can also check whether ex-
ponents and arguments of special functions such as sine and cosine do not
have a dimension. For example sin(meters), or 2seconds are incorrect. We
would strongly encourage you for each expression you write down to con-
vince yourself that it is dimensionally correct. This habit will help you to
quickly identify mistakes in your models or answers to (exam) questions.

ISEE example: pendulum

We can use the systematic dimension analysis method to derive the relation-
ship between the period T and the independent variables m, l and g of a
pendulum.

I - Identify

The target variable is the period T and the known variables are m , l and g.

S - Set up
First, identify the dimensions of the dependent variable and the
independent variables to find a relationship between them. Our inventory is:
• independent variables:
m [kg]
l [m]
g [m s−2 ]
• dependent variables:
T [s]
θ [-]
The next step is to apply systematic dimension analysis and write the
relationship as follows:

T ∼ ma1 ⋅ la2 ⋅ g a3 , (1.11)

where a1 , a2 and a3 are constants.

Find the constants a1 , a2 and a3 . This can be done by using that the left
hand side should have the same dimensions as the combination of variables
at the right hand side. This gives:

13
 a
[s] = [kg] 1 ⋅ [m] 2 ⋅ [ms−2 ] 3 .
1 a a
(1.12)

E - Execute

By looking at the dimensions [kg], [m], and [s] independently we can derive
the following set of equations for the coefficients a1 , a2 , and a3 :

[kg] ∶ 0 = a1 + 0 + 0
[m] ∶ 0 = 0 + a2 + a3 (1.13)
[s] ∶ 1 = 0 + 0 − 2a3

These are 3 equations with 3 unknowns. Solving this set of equations gives
the solution:

a1 = 0, a2 = 1/2, a3 = −1/2, (1.14)

Corresponding to the relationship:

√
l
T ∼ l1/2 g −1/2 = . (1.15)
g

E - Evaluate
√
Firstly, the dimension should be checked. The dimension of gl is
√
[ mms−2 ]=[s], which is correct. The found relationship corresponds to
equation (1.1). From the result we can conclude that T is not dependent on
the mass m as expected.

- Movie - Dimensional Analysis: pendulum

1.4 Sketching graphs of functions

The graphical representation of data is of great importance in many fields,
such as mathematics, sciences, engineering, technology, finance, and many

14
other areas. Presenting data using plots is the method of choice to illustrate
relationships between different quantities or variables. An example is fig-
ure 1.2 in which the excursion angle of a pendulum is shown as a function of
time. Although plots are quite often drawn using a computer or a graphical
calculator, one should not underestimate the value of being able to draw a
plot by hand. Making a plot helps you to gain insight into the behavior of
the function under study.

Many types of plots exist, but here we restrict ourselves to plots of (math-
ematical) functions of just one variable, say y = f (x), where x is the inde-
pendent variable, f denotes the function, and y is the dependent variable.
The idea of a graph of this function f (x) is to draw two axes, an x-axis and
a y-axis that are perpendicular to each other. The intersection of these two
axes is the so-called origin where (usually) x = 0 = y. Each axis may be
provided with tick marks that indicate numerical values of the two variables
x and y. The functional dependence between x and y is then illustrated by
a line consisting of the set of all points having coordinates (x, y) = (x, f (x)).

Except for trivial or very simple (basic) functions, it makes sense to first
investigate some properties of the function before graphing it. Here are the
most important properties one can evaluate in order to appreciate the local
and global behavior of the function:

(1) Find the intercepts of any kind. A graph will have a y-intercept at
points (0, f (0)) and x-intercepts are found by setting f (x) = 0 and
solving for x.

(2) Find any asymptotes. Remember that x = a is a vertical asymptote if

the graph increases or decreases without bound as x moves in closer
and closer to x = a. The line y = b is a horizontal asymptote if the
graph approaches y = b as x increases or decreases without bound.

(3) It sometimes also makes sense to investigate a function’s behavior as

x → ±∞. This so-called asymptotic behavior might be a straight line
that the function is tending towards in these limiting cases.

There are a couple of additional properties of the function that one might
want to investigate in order to show a reasonable accurate sketch of the graph
of the function:

15
 (4) Determine all intervals where f (x) is increasing (f ′ (x) > 0)3 or de-
creasing (f ′ (x) < 0). Here f ′ (x) denotes the derivative of f (x) with
respect to x. Moreover, determine the maximum and minimum values
(the so-called extrema) of the function at those locations where the first
derivative changes sign and f ′ (x) = 0.

(5) The function might be even or odd. A function is even if its graph is
symmetric with respect to the y-axis, i.e. f (−x) = f (x) for all x. A
function is odd if its graph is symmetric with respect to the origin, i.e.
f (−x) = −f (x) for all x.

Once one has sufficient information about the behavior of the function, it can
be graphed. Note that you do not necessarily have to perform all 5 steps.
As soon as you understand the trend of the graph, you can start drawing it.

Special functions

In physics we often encounter special functions. Some of these special func-

tion you already know: sin(x), cos(x), exp(x), ln(x). The exponential func-
tion exp(x) is often written as ex .

The functions cosh(x) (hyperbolic cosine) and sinh(x) (hyperbolic sine) may
perhaps be new to you. These functions are defined as follows:

exp(x) + exp(−x) exp(x) − exp(−x)

cosh(x) = and sinh(x) = . (1.16)
2 2
Table 1.1 gives an overview of the derivatives of these special functions and
the corresponding integrals.

3
The prime denotes the derivative of the function, see §1.5.

16
 Table 1.1: Derivatives of special functions and related integrals.
d
sin(x) = cos(x) ∫ cos(x) dx = sin(x) + C
dx
d
cos(x) = − sin(x) ∫ sin(x) dx = − cos(x) + C
dx
d
exp(x) = exp(x) ∫ exp(x) dx = exp(x) + C
dx
d 1 1
ln(x) = ∫ x dx = ln(x) + C
dx x
d
sinh(x) = cosh(x) ∫ cosh(x) dx = sinh(x) + C
dx
d
cosh(x) = sinh(x) ∫ sinh(x) dx = cosh(x) + C
dx

ISEE example: Sketching a graph

Exercise: Sketch a graph of the mathematical function f (x) = x cosh(x) in

the range −∞ < x < ∞.

I - Identify

The function f (x) can be seen as the product of two functions f (x) = g(x)h(x)
with g(x) = x and h(x) = cosh(x).

S - Set up
The function g(x) = x can be sketched as straight line through the origin
with slope 1. The function h(x) = cosh(x) is a special function as defined in
equation (1.16). It can be helpful to memorize the behavior of the special
functions listed in table 1.1. For the sake of the example we will analyze the
behavior of h(x).

By sketching both g(x) and h(x) in the same graph, the function f (x) can be
obtained by multiplying the y-values of g(x) and h(x) for different x-values.

17
E - Execute

e0 + e0
(1) h(0) = =1
2
The y-intercept is (0,1).

ex + e−x
cosh(x) = = 0 does not have a solution, so there is no x-
2
intercept.

(2) h(x) has no asymptotes along the x-axis.

(3) lim h(x) = ∞ and lim h(x) = ∞.

x→∞ x→−∞

(4) h′ (x) = sinh(x) = e −e

x −x
2
Solving h′ (x) = 0 gives a minimum value at x = 0.
(minimum because of the asymptotic behavior evaluated at (3)).
h(0) = 1 so the minimum is (0,1).
x < 0: h′ (x) < 0 → decreasing
x > 0: h′ (x) > 0 → increasing
e−x + ex
(5) h(x) is an even function because h(−x) = = h(x), meaning that
2
the graph is symmetric with respect to the y-axis.

This is enough information to sketch the graph. See figure 1.5 for the result.

Now the graph for f (x) can also be sketched by multiplying the y-values of
g(x) and h(x) for different x-values, see also figure 1.5.

E - Evaluate

The graph can be checked by evaluating some of the properties (1-5) for the
graph f (x) itself. For example, the sketched graph is odd, consistent with
the evaluation f (−x) = −f (x).4

4
Note that since h(x) is even and g(x) is odd, f (x) must be odd.

18
 Figure 1.5: The functions f (x), g(x) and h(x).

In this course, when we say sketching a graph we expect you to draw a graph
that contains all details of the graph that are relevant for understanding the
underlying physics. Make sure to show a tidy graph rather than presenting
a sloppy or messy one. Use a ruler to draw the axes, put tick marks on each
of the two axes and show numerical values along the axes. Also make sure
that it is clearly stated which quantity is actually shown along each of the
two axes (and what the dimension of that quantity is, if any). Also indicate
important characteristics of the graph in the sketch, such as the maxima or
asymptotes. For example, in figure 1.5 the slope of the graph at x = 0 is
indicated.

1.5 Mathematical notation

In the text above we have been using Lagrange’s notation for writing deriva-
tives (i.e., using f ′ (x) for the first derivative). While studying physics you
will notice that there are various ways of notation for derivatives. Since you

19
will need to be able to read text books and scientific literature, it is helpful to
recognize the differences. So, instead of being fully consistent in the courses
of our bachelor program, we will introduce the various notations here, such
that you will learn to use and appreciate the various notations.

We consider here a quantity y as the dependent variable and a quantity x as

the independent variable.
y = f (x). (1.17)
The symbol f (the function) denotes the function rule that describes how y
depends on x.

We often use the notation of Gottfried Leibniz for derivatives:

dy
. (1.18)
dx
Higher derivatives are written as:
d2 y d3 y dn y
, , ... . (1.19)
dx2 dx3 dxn
The derivative of a function can be written in different ways in Leibniz’s
notation:
df (x) df d
= (x) = f (x). (1.20)
dx dx dx
Lastly, the value of the derivative in point x = a is noted as:
R
dy dy RRRR
(a) = R .
dx RRRR
(1.21)
dx
Rx=a

In many cases, more simpler expressions can be written by using Lagrange’s

notation instead. In that case, a derivative is indicated by a prime:
f ′ (x), f ′′ (x), f ′′′ (x). (1.22)
Beyond the third order derivative, either Arabic (in parentheses) or Roman
numerals (typically in lower case) are used:
f (4) (x), f iv (x). (1.23)

20
A benefit of this notation is that you can also define the nth derivative of a
function:

f (n) (x). (1.24)

Sometimes the math can be written even more compactly when using New-
ton’s notation, which is also referred to as the dot-notation. In this notation
you place a dot above the dependent variable to indicate differentiation with
respect to the independent variable. This notation is most commonly used
when the time is the independent variable:
...
ẏ, ÿ, y . (1.25)

For completeness we also have to introduce Euler’s notation, although less

extensively used as the previous options. In this case, a differential operator,
denoted with the letter D is used to indicate differentiation of a function f .
Higher order differentiation is written using an Arabic numeral in superscript:

Df, D2 f, D3 f, ...Dn f. (1.26)

As a summary, all these notations refer to the same second order derivative:

d2 y d dy d d ′
2
= ( ) = ÿ = ( ẏ) = f ′′
(x) = (f (x)) = D2 y. (1.27)
dx dx dx dx dt
Since we often use Greek letters when defining variables, it can be useful to
memorize the Greek alphabet, as shown in appendix C.

21
2 Vectors, coordinates and motion
In this chapter we will discuss how to describe the motion of objects in
space. To do this we need to establish a suitable administrative framework,
or maybe a few, so that we have a way to quantitatively describe points in
space.
First we will discuss vectors,5 which are an essential concept to describe many
things that cannot be captured by numbers (scalars) alone. Many concepts
later in these lecture notes and in later courses will use vectors. Then we
will discuss a selection of coordinate systems that fit the most frequently
occurring types of motion. These coordinate systems will be useful again in
e.g. the courses 31MCA and 31EMA. With these frameworks established,
we will finally come to the discussion of motion of objects through space.

2.1 Vectors
2.1.1 Definition
A vector is different from a number. You can think of a vector as a geometric
object that has a magnitude and a direction. Most often it is represented by
an arrow, as shown in figure 2.1, to form a directed line segment: the length
symbolizes the magnitude and the direction is indicated by the arrow. A
vector is distinct from a scalar, which is only a number without a direction.

B C D
A

⃗ B,
Figure 2.1: Vectors A, ⃗ C⃗ and D
⃗ with different magnitudes and directions

We can shift a vector in such a way that the start changes, but the length
and direction remain the same. This new arrow is regarded as another rep-
resentative of the same vector. The vectors in figure 2.1 each have a different
length and direction, so they are different vectors.
5
Parts of this chapter are based on the module “Vectors in two and three dimensions”,
from the Department of Mathematics and Computer Science, TU/e.

22
2.1.2 Representation
To distinguish vectors from scalars, we unfortunately use various representa-
tions, mostly modifications of letters. It is important to familiarize yourself
with all of those representation, because you are most certain to come across
all of them during your studies and in all kinds of scientific literature.
Examples include A, A, ⃗ A, ⇀
A.

The representation of the vector in space can be done by an arrow as shown

in figure 2.1. Besides an arrow you can write a 2-dimensional vector as a
column of 2 numbers:
v
v⃗ = ( 1 ) , (2.1)
v2

where v⃗ represents a vector that points v1 units in the x-direction and v2

units in the y-direction.
In three dimensions a vector is represented by a column with three compo-
nents:

⎛w1 ⎞
w⃗ = ⎜w2 ⎟ , (2.2)
⎝w3 ⎠

where the third component indicates the amount of units the vector points
in the z-direction.
Another way to represent the vector is the row representation: v⃗ = (v1 , v2 ).
In literature you can also find different types of brackets: v⃗ =< v1 , v2 > or
v
v⃗ = [ 1 ].
v2
⃗ or simply by w and can be calculated
The length of a vector is denoted by ∣∣w∣∣
from its components:
√
⃗ = w1 2 + w2 2 + w3 2 .
∣∣w∣∣

2.1.3 Multiplication by a scalar

Vectors can be multiplied by a scalar. This multiplication only affects the
magnitude of the vector, but not the direction. The only exception is when
a vector is multiplied by a negative number, in which case the direction of

23
 v λv
the vector is reversed. Multiplying v⃗ = [ 1 ] results in λ⃗v = [ 1 ], where λ is
v2 λv2
a scalar. This vector λ⃗v is called a scalar multiple of v⃗.

Calculation rules
For every vector v⃗ and scalars λ and µ holds:

v ) = (λµ)⃗
• λ(µ⃗ v

2.1.4 Adding and subtracting

To add two vectors u⃗ and v⃗, the procedure is simple. Place the tail of one
vector on the head of the other vector (without changing the direction of
the vector) and draw the new vector u⃗ + v⃗, as shown in figure 2.2(a). To
subtract two vectors, the procedure is similar, except now you add u⃗ and
−1⋅⃗
v = −⃗
v as shown in figure 2.2(b), which is also the reason why we discussed
multiplication by a scalar first.

y y

u’

u +v v’ u-v
v v
u u
0 0 x
x

(a) (b)

Figure 2.2: (a) Vector addition of vector u⃗ and v⃗, (b) vector subtraction of
vector u⃗ and v⃗.

In the column representation you can add two vectors by adding the individ-
ual corresponding components:

⎛v1 ⎞ ⎛w1 ⎞ ⎛v1 + w1 ⎞

v⃗ + w⃗ = ⎜v2 ⎟ + ⎜w2 ⎟ = ⎜v2 + w2 ⎟ . (2.3)
⎝v3 ⎠ ⎝w3 ⎠ ⎝v3 + w3 ⎠

24
It is not possible to add or subtract two vectors with different dimensions
(i.e. different number of components).

Calculation rules
For all vectors u⃗, v⃗ and w⃗ holds:

• v⃗ + w⃗ = w⃗ + v⃗

u + v⃗) + w⃗ = u⃗ + (⃗
• (⃗ ⃗
v + w)

Furthermore, there also rules for which scalar multiplication as well as vector
addition plays a role:

⃗ = λ⃗
v + w)
• λ(⃗ v + λw⃗

• (λ + µ)⃗
v = λ⃗
v + µ⃗
v

2.2 Lines and planes

2.2.1 Lines
We can define a chosen origin O in space by the null-vector 0, ⃗ which has a
0
length 0 and no direction. In 2D the vector is defined as 0⃗ = ( ) and in 3D
0
⎛0⎞
⃗
as 0 = ⎜0⎟.
⎝0⎠
Each unique vector starting in origin O defines a unique point in space. This
also holds the other way around: each point in space defines a vector from
the origin to that specific point.

A line connects two points to each other. In secondary school you learned
that you can write a line in in 2D in the form of

y = ax + b, (2.4)

where a represents the slope or gradient of the line and b is the y-intercept.
More generally you could define a line in 2D as
(y1 − y0 )
(y − y0 ) = (x − x0 ) (2.5)
(x1 − x0 )

25
connecting the points (x0 , y0 ) and (x1 , y1 ). In this representation the slope
(y1 − y0 )
is represented by and the offsets are given by x − x0 and y − y0 .
(x1 − x0 )

a + λv

v
a

Figure 2.3: Parameter representation of straight line with supporting vector

⃗ and direction vector v⃗. Each point of the line can be obtained by adding a
a
specific scalar multiple of v⃗ to a
⃗.

Besides using an equation as we did above, we can express lines using a

parameter representation, as illustrated in figure 2.3. The scalar multiples
x⃗ = λ⃗
v of a vector v⃗, with v⃗ ≠ 0, run through all points (vectors) on a straight
line l through the origin. If we consider a second vector a ⃗, then for any scalar
λ,
x⃗ = a
⃗ + λ⃗
v (2.6)
⃗ and parallel with l (see figure 2.3).
is a point on a straight line m through a
In this notation, we referred to
l ∶ x⃗ = λ⃗
v, (2.7)
and
m ∶ x⃗ = a
⃗ + λ⃗
v (2.8)

26
as parametric representations, or vector representations of lines l and m, re-
spectively. For both lines, v⃗ is the so-called direction vector. The vector a
⃗ is
a supporting vector. Both the direction vector and the supporting vector are
not uniquely defined. Any multiple of v⃗ could also be used. Furthermore,
⃗, any vector on line m could serve as an alternative supporting vec-
besides a
tor.

How do we describe a line through two points, let’s say points P and Q? We
can use the vectors from the origin O to each of the points, p⃗ from O to P
and q⃗ from O to Q. The vector from P to Q is given by q⃗− p⃗ and can be used
to indicate the direction of the line. The vectors p⃗ and q⃗ both end on the
line, and can therefore function as supporting vector. This gives as possible
vector representation of the line:

x⃗ = p⃗ + λ(⃗
q − p⃗). (2.9)

For λ = 0 we get p⃗ and for λ = 1 we get q⃗.

Example
If we consider the points (2,3,-1) en (3,4,1), a direction vector for the lines
through these points is:

⎛3⎞ ⎛ 2 ⎞ ⎛1⎞
v⃗ = ⎜4⎟ − ⎜ 3 ⎟ = ⎜1⎟ . (2.10)
⎝1⎠ ⎝−1⎠ ⎝2⎠

and a parametric representation is:

⎛2⎞ ⎛1⎞
x⃗ = ⎜ 3 ⎟ + λ ⎜1⎟ . (2.11)
⎝−1⎠ ⎝2⎠

It is important to realize that an equation and a parametric representation

are two different ways of describing a line. We just learned that a line l in
two dimensions can be represented by a parametric representation:

x a v
l ∶ x⃗ = ( ) = ( 1 ) + λ ( 1 ) (2.12)
y a2 v2

27
If we want to describe the line l with an equation, we have to eliminate the
parameter λ from the two equations:
⎧
⎪
⎪ x = a1 + λv1 ,
⎨ (2.13)
⎪
⎩ y = a2 + λv2 .
⎪
Multiplying the first equation by v2 and the second by v1 and then subtracting
them gives:
v2 x − v1 y = v2 a1 − v1 a2 (2.14)
in which λ is eliminated. After dividing by v1 , this equation can be rewritten
as:
v2 v2
y = x + (a2 − a1 ) (2.15)
v1 v1
which may look more familiar, as it has the form y = ax + b.

In 3 dimensions, a line can be described using a parametric representation

x⃗ = a
⃗ + λ⃗
v , which is equivalent to:

⎛x⎞ ⎛a1 ⎞ ⎛v1 ⎞

⎜y ⎟ = ⎜a2 ⎟ + λ ⎜v2 ⎟ .
⎝z ⎠ ⎝a3 ⎠ ⎝v3 ⎠

⎛a1 ⎞ ⎛v1 ⎞
⃗ = ⎜a2 ⎟, and has direction vector v⃗ = ⎜v2 ⎟.
This line goes through a
⎝a3 ⎠ ⎝v3 ⎠

2.2.2 Planes
In three dimensions we do not only have lines, but we also have planes.
Planes can also be described using vectors. In contrast to a line, we need
one supporting vector and two direction vectors, as illustrated in figure 2.4.
These two direction vectors should not be a scalar multiple of each other.
Furthermore, each of the direction vectors needs its own independent pa-
rameter in order to describe every possible point in the plane. The plane U
going through the origin and with direction vectors u⃗ and v⃗ is given by the
parametric representation:
U ∶ x⃗ = λ⃗
u + µ⃗
v

28
 u
u v
v
a
O O

Figure 2.4: Left: plane through the origin, right: plane through supporting
⃗ both with direction vectors u⃗ and v⃗.
vector a

⃗ and the same direction

and is parallel to plane V with supporting vector a
vectors u⃗ and v⃗:

V ∶ y⃗ = a
⃗ + λ⃗
u + µ⃗
v.

Similar to lines, direction and supporting vectors are not uniquely defined
per plane.

In three dimensions, plane V can be described as:

⎛x⎞ ⎛a1 ⎞ ⎛u1 ⎞ ⎛v1 ⎞

⎜y ⎟ = ⎜a2 ⎟ + λ ⎜u2 ⎟ + µ ⎜v2 ⎟ .
⎝ z ⎠ ⎝ a3 ⎠ ⎝u3 ⎠ ⎝v3 ⎠

A plane in three dimensions can also be described by a linear equation, for

which we again have to eliminate the parameters. We will illustrate that
with an example:

29
If we consider a plane V :

⎛ x⎞ ⎛ 2 ⎞ ⎛1⎞ ⎛2⎞
⎜y ⎟ = ⎜ 1 ⎟ + λ ⎜−1⎟ + µ ⎜−1⎟ .
⎝z ⎠ ⎝−2⎠ ⎝0⎠ ⎝−1⎠

For any vector x⃗ should hold:

⎧
⎪ x = 2 + λ + 2µ,
⎪
⎪
⎪
⎪
⎨ y =1−λ−µ (2.16)
⎪
⎪
⎪
⎪
⎪
⎩ z = −2 − µ.
If we add the first two equations we get x + y = 3 + µ. We can add this to the
last equation to obtain

x + y + z = 1. (2.17)

2.3 Coordinate systems

In this course we will look at two coordinate systems to describe the position
and motion of objects in space: the Cartesian coordinate system and the
polar coordinate system. The coordinate system of choice depends on what
is most practical, given the problem under consideration.

2.3.1 Cartesian coordinate system

The Cartesian coordinate system is by far the most popular coordinate sys-
tem. It defines a rectangular grid in space as well as an origin of the coordi-
nate system. From this origin it is possible to assign unique coordinates to
every point in space. The system works well in 1D, 2D and 3D, especially for
motion along the grid lines of the system. The Cartesian coordinate system
is well covered by the secondary school curriculum, so we will not discuss
this further.

2.3.2 Base vectors

Base vectors are vectors of magnitude 1 that point in the (positive) direction
of each dimension of the coordinate system. Base vectors are therefore also
perpendicular to each other.

30
The base vectors are usually labeled î, ĵ, k̂, for the x, y and z direction,
respectively.
Using the knowledge about adding vectors discussed previously, this means
that every vector can now be described as the sum of 3 vectors

A⃗ = Ax î + Ay ĵ + Az k̂. (2.18)

2.3.3 Polar coordinate system

For two dimensional spaces, it is sometimes useful to consider a coordinate
system other than the Cartesian system. Many times, when the motion is
not along the direction of the cartesian axes, and particularly when it is
not straight, cartesian coordinates become difficult to work with. It is then
more convenient to use a coordinate system, that has axis-directions that
match what you’re trying to describe. One option to consider is the polar
coordinate system. This system is particularly useful for situations where
circular motion is described. Both cartesian and polar coordinate systems
are shown in figure 2.5.

Figure 2.5: Coordinate lines for Cartesian and polar coordinates.

In polar coordinates, the position in the x, y plane is described by two coordi-

nates, that are not x and y. Instead, the distance to the origin r and the angle
θ between position vector r⃗ and the x axis are used. The r-direction is called
the radial direction and the θ-direction is called the azimuthal direction.

31
So, for any given point in space (x, y) an equivalent set of coordinates (r, θ)
exists that perfectly describes the same point. This defines the relation be-
tween x, y, r and θ (for positive values of x):
√
r = x2 + y 2 (2.19)
y
θ = arctan (2.20)
x
This also leads to the inverse relation:
x = r cos θ (2.21)
y = r sin θ (2.22)
We can also define base vectors for the polar coordinate system: r̂ and θ̂ for
the r and θ directions.
- Movie - Coordinate systems

2.3.4 Cylindrical and Spherical coordinates

In this course we do not encounter Cylindrical or Sperical coordinates, but
they are nevertheless very useful and used often in science. Cylindrial co-
ordinates are essentially just polar coordinates expanded with a cartesian
z-coordinate. They are useful for things that have a cylindrical geometry.
Spherical coordinates are like polar coordinates, but expanded with a second
angle that gives the angle relative to the vertical axis.

2.4 Motion
Now that we have established the administrative framework to describe the
position of an object in space, it is time to start using it to describe motion.
In this course, we will focus specifically on kinematics, which describes the
motion as it happens, but does not cover why it happens (due to what forces,
as covered in dynamics).

2.4.1 Displacement
Describing a position in space is something we can now easily do by using a
vector r⃗. To describe a displacement we need to describe two positions.
For example, imagine an object that is moved from a position r⃗1 to a position
r⃗2 . The displacement S⃗ is then given by r⃗2 − r⃗1 (see figure 2.6).

32
 z

r2 S

r1
0
y

Figure 2.6: The displacement S⃗ is given by r⃗2 − r⃗1 .

2.4.2 Velocity and acceleration

To start off the discussion about velocity and acceleration, it is good to repeat
the well known definitions for average velocity, instantaneous velocity, aver-
age acceleration and instantaneous acceleration, for motion in 1 dimension.
The average velocity is defined as
x(t2 ) − x(t1 )
vav = . (2.23)
t2 − t1
The instantaneous velocity is given by
x(t + ∆t) − x(t) dx
v = lim = . (2.24)
∆t→0 ∆t dt
Similarly, the average acceleration can be defined as
v(t2 ) − v(t1 )
aav = (2.25)
t2 − t1
and the instantaneous acceleration can be given as
v(t + ∆t) − v(t) dv d2 x
a = lim = = . (2.26)
∆t→0 ∆t dt dt2
Likewise the velocity v can be found by integrating the acceleration a

v = ∫ adt (2.27)

33
and similarly the position x can be found by integrating the velocity v

x = ∫ vdt. (2.28)

Besides the position x, we can calculate the path length s, i.e., the traveled
distance, of this mass by integrating the absolute value of v over time t:
dx
s = ∫ ∣v∣ dt = ∫ ∣ ∣ dt. (2.29)
dt
If we graph the position, velocity and acceleration as function of time we can
see the same relations as given above. Figure 2.7 shows these graphs in case
of a constant acceleration.
dx dv
x v dt a
dt

∫vdt ∫adt

0 t 0 t 0 t

Figure 2.7: Graphs depicting x, v and a as function of t for constant acceler-

ation.

To expand the 1D definitions to motion in 3D, we will now rewrite the above
equations in terms of 3D position vectors.

r⃗(t + ∆t) − r⃗(t) d⃗r

v⃗ = lim = (2.30)
∆t→0 ∆t dt
For r⃗ = (x, y, z) you can describe the x component of the velocity as
x(t + ∆t) − x(t) dx
vx = lim = (2.31)
∆t→0 ∆t dt
If we remember our base vectors for the Cartesian coordinate system, we
can also write the instantaneous velocity as a sum of the component velocity
vectors
d⃗
r dx dy dz
v⃗ = = î + ĵ + k̂ (2.32)
dt dt dt dt
34
Similarly we can describe the instantaneous acceleration

d⃗
v dvx dvy dvz d2 r⃗
⃗=
a = î + ĵ + k̂ = 2 (2.33)
dt dt dt dt dt
Of course, it is also possible to describe our motion in polar coordinates. You
can use figure 2.8 to find the velocity and acceleration in polar coordinates.
In exercise 9.2.7 you will show that the radial and azimuthal component of
the velocity are given by:

vr = ṙ,
(2.34)
vθ = rθ̇.

Similarly, in exercise 9.2.8 you will deduce that the radial and azimuthal
component of the acceleration are:
2
ar = r̈ − r (θ̇) ,
(2.35)
aθ = rθ̈ + 2ṙθ̇.

2.4.3 Projectile motion in the x, y-plane

To describe the motion of a projectile flying through space, a 2D coordinate
system is enough if the object moves on a straight line horizontally. We
will choose the x-coordinate in such a way that it aligns with the horizontal
displacement of the object.
The motion of the projectile can then be separated in a horizontal (x) and
a vertical component (y), that are independent of each other. The displace-
ment vector is given by x⃗ = (x, y), the velocity vector by v⃗ = (u, v) and the
⃗ = (ax , ay ). Due to dynamics, that will be treated in
acceleration vector by a
the following chapters, the motion will obey the following two equations:

ax = 0 (2.36)
ay = −g (2.37)

Integrating this twice with respect to time gives

x(t) = at + b
} (2.38)
y(t) = − 21 gt2 + ct + d,

35
Figure 2.8: Mass with velocity (a) expressed in Cartesian coordinates, (b)
expressed in polar coordinates, (c) expressed in both coordinates systems and
(d) in Cartesian coordinates projected on the radial direction.

where a, b, c, and d are four (= 2 × 2) integration constants. Using the

following (four) initial conditions
t = 0 ∶ x = x0 , u = u0
} (2.39)
t = 0 ∶ y = y0 , v = v0
we can determine these constants, resulting in
x(t) = x0 + u0 t
} (2.40)
y(t) = y0 + v0 t − 21 gt2 .
These x and y-coordinates of the position x⃗ indicate that the mass follows a
parabolic trajectory.

36
We will show this for the case where v0 = 0 (the bullet is shot in a purely
horizontal direction, with initial velocity u0 ≠ 0). In this case (2.40) reduces
to
x(t) = x0 + u0 t
} (2.41)
y(t) = y0 − 12 gt2 .

To find the equation for the trajectory, we are now going to ‘eliminate’ the
time t. This will describe the relationship between x and y for the trajectory
of the projectile. In doing so, we create a single equation that describes
the path of the motion, but we lose all the information about the timing
of the motion. We will achieve this by writing the first equation (2.41) as
t = (x − x0 )/u0 , and substituting this in the second equation (2.41):
g
(y − y0 ) = − 2
(x − x0 )2 . (2.42)
2u0

If we now define the new coordinates (x′ , y ′ ) as (x′ , y ′ ) = (x − x0 , y − y0 ), i.e.

the coordinates that give the position relative to the initial position (x0 , y0 ),
we get
g ′2
y′ = − x . (2.43)
2u20

This mathematical expression describes a parabola, see figure 2.9.

(x0,y0)

0 x

Figure 2.9: Parabolic trajectory of a projectile.

37
3 Evolution equations
In this chapter we want to make a start with the analysis of the dynamic
behavior of a (physical) system by considering the time dependence of the
dependent variables. The type of equations we use to study this time-
dependence are called evolution equations, as they describe how the system
evolves as time progresses. We will formulate evolution equations for several
examples, ranging from thermodynamics to electrical circuits.

3.1 Water tank

Consider a water tank (see figure 3.1) that is being filled by a filling tube.
Water is flowing through this tube with a constant velocity U into the tank.

U
A

ρ
h
A0

Figure 3.1: Water tank being filled.

The system is characterized by the following variables:

U [m s−1 ] : velocity of the inflowing water

A [m2 ] : cross section of the filling tube
A0 [m2 ] : cross section of the tank
h [m] : height of the water level
ρ [kg m−3 ] : density of water
The total amount of water that flows through the tube per unit time we call
the volume flow rate, which is defined as
ϕV = U A [m s−1 ].
3
(3.1)
The corresponding mass flow rate is
ϕm = ρU A [kg s ].
−1
(3.2)

38
The water volume in the tank

V = A0 h, (3.3)

increases with the water supply (ϕV ). Apparently, the volume V , and there-
fore the water height h, are dependent on time:

V = V (t), h = h(t). (3.4)

In other words, these quantities both are functions of the time. An example
of the course of the water height h(t) in time is illustrated graphically in
figure 3.2. The question now is: how can we determine an expression for the
course of h(t) at a given volume flow rate ϕV ?

h
h(t)

h0

0
t

Figure 3.2: The course of the water height h as a function of the time t.

The volume flow rate ϕV of the added water results in an increase of the water
volume in the tank. During a time span ∆t the water volume increases from
V (t) to V (t + ∆t) according to:6

V (t + ∆t) = V (t) + ϕV ∆t. (3.5)

We can also write this as

∆V ≡ V (t + ∆t) − V (t) = ϕV ∆t. (3.6)

Figure 3.3 shows the water tank again, but now with the variables we just
defined.
6
Note that this expression is only correct when ϕV is constant.

39
 Φv

ΔV = V(t+Δt) -V(t)
h(t+Δt)
h(t)

Figure 3.3: During a time span ∆t the volume flow rate ϕV results in a
volume increase ∆V.

We choose the time span ∆t to be small, but ≠ 0, so we can divide the left
and right hand side of equation (3.6) by ∆t:

∆V V (t + ∆t) − V (t)
= = ϕV . (3.7)
∆t ∆t
dV
In the limit ∆t → 0, the left hand side is equal to the derivative , so:
dt

∆V dV (t)
lim = = ϕV (3.8)
∆t→0 ∆t dt

With this we have derived the evolution equation for V (t): this differential
equation (see §3.3) describes the change in volume V (t) in time resulting
from the incoming water flow rate ϕV .
In the case ϕV = constant this equation can simply be solved for V (t) by
integrating the left and right hand side over t:
dV
∫ dt = V (t) = ϕV t + C. (3.9)
dt
The integration constant C can be determined with an initial condition

t = 0 ∶ V = V0 , (3.10)

which results in

V (t) = V0 + ϕV t. (3.11)

40
We can also perform the integration of equation (3.8) between two bound-
aries, e.g., from t′ = 0 to t′ = t. The result then is
t′ =t dV ′ t′ =t t
∫t′ =0 dt = V ∣ = V (t) − V 0 = ϕV t′
∣ = ϕV t, (3.12)
dt′ t′ =0 0

and this is (of course) in agreement with equation (3.11).

As mentioned earlier in equation (3.3), V (t) = A0 h(t), so we can rewrite
equation (3.11) to an expression for the water height:

V0 ϕV
h(t) = + t. (3.13)
A0 A0
With ϕV = U A (see equation (3.1)) this becomes:

UA
h(t) = h0 + t, (3.14)
A0
where h0 = h(t = 0) is the water height at t = 0. The water height h increases
linearly (for constant ϕV ) with t, so similar to what we saw earlier in figure
dh
3.2. The slope of this graph for h(t) is equal to the derivative , and this
dt
quantity describes the change of h in time:
dh ϕV U A
= = . (3.15)
dt A0 A0

Φv
vsur

h(t)

Figure 3.4: The velocity vsur with which the water surface rises is equal to
dh
.
dt

41
So, this quantity is equal to the velocity vsur with which the water surface
rises:
dh
= vsur , (3.16)
dt
as is illustrated in figure 3.4.

3.2 Heating/cooling object

We are now going to look at a problem of a thermodynamic nature: a solid
object is being heated or cooled by a heat flow from/to its surroundings (see
figure 3.5).
The object has a heat capacity Cw (dimension [J K−1 ], i.e. Joule [J] per Kelvin
[K]) that describes the increase in temperature (in Kelvin [K]) resulting from
an addition of heat (expressed in Joule [J]).
A heat flow rate ϕw , i.e. an amount of heat per unit of time [J s−1 ], results
dT
in a temperature change per unit of time according to
dt
dT
ϕw = Cw . (3.17)
dt
Note the similarity of this expression to equation (3.8).

At a constant heat flow rate ϕw this equation changes into

dT ϕw
= = const, (3.18)
dt Cw

T(t)
Φw

Figure 3.5: A solid object that receives a heat flow rate ϕw from the surround-
ings will experience an increase in average temperature.

42
of which the solution can be found by integration:
ϕw
T (t) = t + C. (3.19)
Cw
The integration constant C can be determined by applying the initial condi-
tion T (t = 0) = T0 , which results in C = T0 .

Alternatively, a heated object can dissipate heat to the surroundings, e.g.

through the flow of air (convection). In good approximation the heat flow
rate to the object (which is thus negative) can then be written as

ϕw = −α(T − T∞ ), (3.20)

where T and T∞ are the temperature of the object and the surroundings,
respectively, and α is the so-called heat transfer coefficient [J s−1 K−1 ]. With
this equation (3.17) becomes
dT
Cw = −α(T − T∞ ). (3.21)
dt
In order to solve this equation it is useful to introduce the ‘over temperature’
T ≡ T − T∞ . Since

dT dT
= (3.22)
dt dt
we can rewrite equation (3.21) with this to

dT α
= − T. (3.23)
dt Cw
This evolution equation can be solved (using separation of variables, ex-
plained below in §3.3.5), and has the solution:
α
T = T − T∞ = C exp (− t) . (3.24)
Cw
Again, the constant C follows from the initial condition T (t = 0) = T0 result-
ing in C = T0 − T∞ . With this the solution becomes:
T (t) − T∞ α
= exp (− t) . (3.25)
T0 − T∞ Cw

43
Apparently the temperature T (t) shows an exponential decay to the final
(surrounding) temperature T∞ . The timescale (i.e. lifetime) associated with
this decay is
Cw
τ= . (3.26)
α
This characteristic timescale also follows from an analysis of the dimensions
of the relevant independent physical parameters in this problem, see exercise
9.3.18.

3.3 Differential equations

In the previous sections we looked at the evolution equations for the volume
of a water tank and the temperature of an object. The rates of change of the
water volume and of the temperature were defined in the form of differential
equations:
dV
= ϕV (3.27)
dt

dT
= −α(T − T∞ ) (3.28)
dt
In this section, we discuss some basic properties of differential equations from
a mathematical perspective. Here will introduce some relevant concepts and
terminology that we need to continue our discussion on evolution equations
in the next section. In addition, we will provide some guidelines for solving
differential equations.

Let us start with the definition. A differential equation is a mathematical

expression that contains terms with one or more of the derivatives of an
unknown function.
The word equation implies that this unknown can be found by solving the
equation in question. Note that we are not solving for a variable, but we want
to find a function as the solution. For example, in equation (3.27) the func-
tion V (t) is the unknown and our task is to determine this function explicitly.

44
We discussed above that in case ϕV is constant, the solution can be found
easily by integrating the left and right hand side over t, as demonstrated in
equation (3.9).
Please note that simply integrating the left and right hand side of a differ-
ential equation only works if the right hand side is a constant. If we for
example consider the simple differential equation
dx
= x, (3.29)
dt
then integrating the left and right hand side over t gives:
dx
∫ dt dt = ∫ x(t) dt. (3.30)

Since we do not know the unknown function x(t) at this point, the integral
at the right hand side cannot be evaluated. Instead, this differential equation
can be solved using the method of separation of variables, as discussed below
in §3.3.5, giving as a solution:

x(t) = et , (3.31)

Please check for yourself by filling in equation (3.31) in equation (3.29) that
this is indeed a solution.

Besides separation of variables, there are many techniques to solve differ-

ential equations analytically. As a first step towards identifying a suitable
technique to solve differential equations, in the next sub-sections we will
categorize differential equations by discussing some of their properties (e.g.,
order, homogeneity, linearity, autonomy).

Contrary to the previous examples, most differential equations in physics

(and also in other branches of science) can not be solved analytically. By
‘analytically’ we mean that the solution of the differential equation can not be
found in terms of known mathematical functions (e.g., the exponential func-
tion of equation (3.25)). In that case we must use the help of the computer
in order to determine an approximation of the solution of the differential
equation, or resort to other types of analyses, as will be discussed in chapters
6 and 7.

45
In this course we predominantly refer to differential equations for functions
of one variable. We then speak of an ordinary differential equation. Ordinary
differential equations exist in many ‘shapes and sizes’. Let us give a number
of examples of differential equations for a function x(t):

d2 x
+ x = 0, (3.32)
dt2

dx
+ 5x = 0, (3.33)
dt

d2 x
− 9x = 0, (3.34)
dt2

d4 x dx
+ + 2tx = 0, (3.35)
dt4 dt

d2 x
+ sin(x) = 0, (3.36)
dt2

dx
x − (x2 + 1) = 0, (3.37)
dt

d2 x
+ 2x = sin(t). (3.38)
dt2
Using these examples we are now going to discuss several properties of or-
dinary differential equations. As mentioned above, we do this to determine
the method we will employ to solve the differential equation or to study the
behavior of the solutions.

46
3.3.1 Order of a differential equation
The order of a differential equation is a natural number that gives the highest
derivative in the equation. So, equations (3.33) and (3.37) are 1st order since
the highest derivative in these two equations is a 1st order derivative. Equa-
tion (3.35) is a 4nd order differential equation and all the other differential
equations are of the 2nd order.

Sometimes (for reasons that will become clear in chapter 6) differential equa-
tions of the 2nd or higher order are written as a system of first order differ-
ential equations. Consider for example equation (3.32), and let x1 = x. We
introduce a new function x2 (t), which is defined by

dx1
x2 (t) = . (3.39)
dt
With this new function we can also formulate (3.32) as follows:
⎧
⎪ dx1
⎪
⎪
⎪ = x2 ,
⎪ dt
⎨ (3.40)
⎪
⎪
⎪ dx2
⎪
⎪ = −x1 .
⎩ dt
We call this a system of two coupled 1st order differential equations. If we
eliminate the function x2 (t) from these two equations (by differentiating the
first equation with respect to t and then using the second equation) we find
equation (3.32) again. In general, a differential equation of the order N can
be written as a system of N coupled 1st order differential equations.

3.3.2 Homogeneous versus inhomogeneous

Contrary to equations (3.32)-(3.36) equation (3.38) contains a term in which
the function x(t) does not occur explicitly, namely in the right hand side of
the equation. Therefore, we call equation (3.38) an inhomogeneous differen-
tial equation. Also equation (3.37) is an inhomogeneous differential equation,
because of the constant term (that is 1) at the left hand side. Equations
(3.32)-(3.36) can be categorized as homogeneous differential equations.

47
3.3.3 Linear versus non-linear
A differential equation is linear if the dependent variable (e.g., x in the ex-
ample) appears only to the 1st power, and no products of the dependent
variables (or its derivatives) or special functions are included in the equa-
tion. Examples of non-linear terms are higher powers (x2 , x3 ), products such
as xy in coupled differential equations (e.g., in the Lotka-Volterra model dis-
cussed in chapter 5), or sin θ in the non-linear pendulum equation (7.16).

In order to check whether or not an inhomogeneous differential equation is

linear, we solely look at the homogeneous part of the equation in question,
that is by excluding all terms in which the function x(t) that we are looking
for does not occur. For equation (3.38) this means that the sine on the right
hand side can be replaced by a 0.

Non-linear equations are the rule rather than the exception in physics and
are notorious for being hard to solve. Quite often, the computer will have
to help in order to learn something about the solutions, or the solutions can
be studied graphically, as will be discussed further in chapters 6 and 7. On
the other hand, the solutions of non-linear differential equations can show
extraordinarily interesting and intriguing behavior, as we will see later in
chapter 8.

Using the theory on Taylor series it is possible to ‘convert’ non-linear differ-

ential equations into linear equations, as will be discussed in §5.2. In this
context we speak of the linearization of the differential equation. An im-
portant advantage of the linearization of a non-linear differential equation is
that the linearized version, contrary to the original equation, can generally
be solved analytically in terms of well-known functions.

3.3.4 Solutions of linear differential equations

Equation (3.33) can simply be solved in terms of the exponential function
exp(t) = et . Check that the function x(t) = e−5t is a solution. However,
x(t) = 2 ⋅ e−5t also is a solution. In fact: e−5t multiplied by any random con-
stant A, x(t) = A ⋅ e−5t is a solution to equation (3.33). This is a signature
property of linear differential equations.

48
So if an equation has an infinite number of solutions, which is the solution
that we are looking for? That is determined by additional constraints we
impose on the function. For example, such a constraint can be that we are
looking for a solution of equation (3.33) in a certain region 0 ≤ t < ∞ and
that for t = 0 the function has to have a certain value, e.g. 1. When we
demand that x(0) = 1 the constant is set and with that the solution has
been determined uniquely: x(t) = e−5t . Such a constraint (or constraints)
that is imposed on the function that we are looking for is called a boundary
condition. When the independent variable is the time, we tend to speak of
initial conditions instead of boundary conditions.

In principle, equation (3.32) also has an infinite amount of solutions. This

equation has the solutions sin(t) and cos(t). Any linear combination of these
solutions
x(t) = A cos(t) + B sin(t), (3.41)
is also a solution. Generally, when solving a second order differential equation
like (3.32), the solution is a linear combination of two solutions, and constants
A and B can be found by satisfying two initial conditions. To argue why the
solution should be a linear combination of two solutions, let us consider the
linear second order differential equation:
d2 x dx
a 2 + b + cx = 0. (3.42)
dt dt
As will discussed in more detail in chapter 6, the evolution of a second order
system can be predicted when knowing both x and dx dt (e.g., the position and
the velocity) for a specific moment. This means that we can uniquely define
the system by providing the differential equation and two initial conditions,
e.g., x(t = 0) = x0 and dx
dt ∣t=0 = v0 .
If we have found two solutions x = x1 (t) and x = x2 (t) that both satisfy
equation (3.42), we can check
x = Ax1 (t) + Bx2 (t) (3.43)
is also a solution of equation (3.42) for any combination of constants A and
B. If x2 is not just a multiple of x1 , by determining the constants A and B
using the two initial conditions, equation (3.43) uniquely satisfies the differ-
ential equation and the initial conditions, and is therefore the solution.

49
3.3.5 Autonomous versus non-autonomous
We call a differential equation autonomous when the independent variable
does not occur explicitly in the differential equation.
Consider the following system of N coupled 1st order differential equations
for the N dependent variables x1 , x2 , . . . , xN :
dx1
= f1 (x1 , x2 , . . . , xN ),
dt
dx2
= f2 (x1 , x2 , . . . , xN ),
dt (3.44)
⋮
dxN
= fN (x1 , x2 , . . . , xN ).
dt
Since the right hand sides, the functions f1 , f2 , . . . , fN , do not explicitly de-
pend on the independent variable t we call the system of differential equa-
tions (3.44) autonomous. Out of the equations (3.32)-(3.38) two are non-
autonomous: the differential equations (3.35) and (3.38), since in these equa-
tions the independent variable (in this case t) occurs explicitly.

In situations for which the independent variable is the time, an autonomous

system is also referred to as a time-invariant system. Note that many laws
in physics are autonomous differential equations, meaning that those laws
also hold at any time in the past or the future (i.e., they are time-invariant).

Solving differential equations using separation of variables

A 1st order autonomous differential equation can in principle be solved ana-

lytically, using a method we refer to as separation of variables (also known as
the Fourier method). For this purpose, consider the following general form
of such a differential equation:
dx
= F (x), (3.45)
dt
with a not further specified function F (x). We then rewrite this equation as
follows:
dx
= dt. (3.46)
F (x)

50
Next, we integrate both on the left and right hand side of the = sign:
1
∫ F (x) dx = ∫ dt = t − t0 , (3.47)

where t0 is a random integration constant. Equation (3.47) is an implicit

expression for the solution x(t) (provided that the integral over x can really
be calculated). The following example shows how this works in practice.

Example

We are looking for the solution to the differential equation

dx
= x(2 − x). (3.48)
dt
Note that this is indeed an autonomous differential equation, as t does not
occur explicitly in the equation.

By using the above, we know that the solution to this equation is determined
by
1
∫ x(2 − x) dx = ∫ dt = t − t0 . (3.49)

The integral over x can easily be calculated (also check this yourself!):

1 1 1 1 1
∫ x(2 − x) dx = ∫ 2 ( x − x − 2 ) dx = 2 {ln ∣x∣ − ln ∣x − 2∣} (3.50)

With this equation (3.49) becomes

√
⎛ x ⎞
ln ∣ ∣ = t − t0 . (3.51)
⎝ x−2 ⎠

If we assume that 0 < x < 2, we thus obtain

e2t
x(t) = 2 . (3.52)
e2t + e2t0

51
The integration constant t0 is determined by the initial condition for x(t). If
we demand, for example, that x(t = 0) = 1 then the solution becomes

e2t et et
x(t) = 2 = 2 = , t ≥ 0, (3.53)
e2t + 1 et + e−t cosh(t)

which satisfies the previously assumed condition 0 < x < 2.

- Movie - Separation of variables: example

More in general, a differential equation for an unknown function f (t) can be

separated if it can be expressed in the form:

df (t)
= g(t)h(f (t)) (3.54)
dt
in which g and h are functions. Rearranging the terms gives:

df (t)
= g(t)dt, (3.55)
h(f (t))

which can be simplified further using x = f (t) to:

dx
= g(t)dt. (3.56)
h(x)

This differential equation can be solved by integrating the left and right hand
side separately.

- Movie - Separation of variables: method

3.4 Newton’s second law: dynamics

After this rather long mathematical section, we can now finally go back to
discussing physics. We all know Newton’s second law of motion: usually
formulated in the following form:

F = ma, (3.57)

52
which means that the force F that needs to be supplied to give a mass m an
acceleration a is equal to m ⋅ a. In physics we often use Newton’s second law
as a starting point for formulating evolution equations.

A more precise formulation is

F = change (in time) of momentum,

where momentum = mv, with v the velocity (in the direction of the force):

d
F= (mv). (3.58)
dt
In many cases we can assume m is a constant, such that equation (3.58)
becomes:
dvx d2 x
Fx = ma = m =m 2, (3.59)
dt dt
where Fx is the force in the direction (x) of the motion of the mass.

[dimension check: force ∼ [N] = [kg m s−2 ] = [kg] × [m s−2 ] → correct]

In case multiple forces act on a mass, the sum of the forces should be taken.

d2 x
∑ Fx = m . (3.60)
dt2
In this summation, the direction of the forces should be considered. Forces
directed in the positive x-direction are taken positive and forces directed in
the negative x-direction are taken negative.

The force F that is exerted on a mass is a vector, with components in the x

and y-direction:

F = (Fx , Fy ). (3.61)

We can therefore decompose ‘Newton’s law’ (3.58) in the x and y-direction,

where the x-component has already been given by equation (3.60). This gives

53
 Ff M v(t)

Figure 3.6: Rolling car being slowed down by the frictional force Ff .

us:
d2 x ⎫⎪
x-direction ∑ Fx = m ⎪
dt2 ⎪ ⎪
⎪
⎬ (3.62)
d2 y ⎪⎪
⎪
y-direction ∑ Fy = m 2 . ⎪ ⎪
dt ⎭
As said, we will use Newton’s second law for formulating evolution equa-
tions. During secondary school you already used Newton’s second law to
solve dynamic problems. By using the symbol a for acceleration, instead of
2
the second derivative of the position x with respect to the time ddt2x , it had
been camouflaged that you were in fact dealing with a differential equation
(to keep it simple).

Because of the 2nd order derivative, you generally obtain a 2nd order differen-
tial equation, as discussed further in chapter 5. Here we will give an example
of using Newton’s second law to formulate a 1st order differential equation.

Example: rolling car

A car (with mass M ) moves across a horizontal road with velocity v, see
figure 3.6. At time t = 0 the engine is turned off and the car rolls further
without being powered. The car is now subject to a (rolling) resistance force
Ff that is proportional to the velocity:
Ff = −αv, (3.63)
where α is the coefficient of friction [kg s−1 ]. The question now is: can we
derive an expression for the (decreasing) velocity v(t) of the car?
Newton’s law given by equation (3.58) now looks like the following:
dv
M = −αv. (3.64)
dt
54
This differential equation has the following solution (found using separation
of variables):
α
v(t) = C ⋅ exp (− t) , (3.65)
M
where C is a random constant. With the initial condition v(t = 0) = v0 ,
equation (3.65) becomes:

v(t) = v0 ⋅ e−t/τ , with τ = M /α. (3.66)

So, the velocity decreases exponentially in time, with a characteristic timescale

τ . This timescale τ could also have been derived with an analysis based on
the dimensions of the independent variables (α, M ) in this problem, as will
be done in exercise 9.3.18.

ISEE example: Formulating an evolution equation

Consider the following exercise:

A car with mass m is driving through a curve with a radius of curvature R.
The engine of the car provides a force in the direction of motion of Fe . Simul-
taneously, the car experiences a (rolling) resistance force of Ff = −αvθ , where
α is the coefficient of friction and vθ is the velocity of the car. Formulate the
evolution equation for the motion of the car.

I - Identify

The answer to a question “formulate the evolution equation” should be a

differential equation that describes the dynamic behavior of the physical sys-
tem. Note that it is not asked to solve the evolution equation.

Known variables are the radius of curvature R, the force in the direction of
motion Fe , the (rolling) resistance force of Ff = −αvθ , and the coefficient of
friction α. The velocity of the car vθ is a function of time vθ = vθ (t), but
its time-dependence is not given. However, as discussed above, vθ is not the
target valuable, since the question only asks for the evolution equation.

The given radius of curvature hints that the motion is along a circular path.
Therefore the motion along the curve can be described conveniently in polar

55
 Figure 3.7: Rolling car in a curve with radius R.

coordinates. Newton’s second law, equation (3.60), can be used as a starting

point for formulating the evolution equation. No information is given about
forces in the radial direction, so we will focus on the azimuthal direction. We
can use equation (2.35) for the acceleration in azimuthal direction aθ . Since
the radius of the movement R is constant, we can use ṙ = 0 and r = R.

What do we expect for the behavior of the physical system? Since there is a
force in the direction of movement (Fe ) and a force in the opposite direction
(Ff ), their relative values should determine whether the car decelerates, has
a constant velocity, or accelerates in the curve.

S - Set up
An important step is to make a drawing, see figure 3.7. Newton’s law should
be evaluated in the azimuthal direction. In the drawing, we chose the direc-
tion of the car counterclockwise (indicated with the velocity vθ ), such that
the movement is in the positive θ-direction. Following that choice, the sum
of the forces in the azimuthal direction is given by

ΣFθ = Fe − Ff = Fe − αvθ . (3.67)

E - Execute

Newton’s second law gives:

∑ Fθ = Fe − αvθ = maθ = m(Rθ̈ + 2ṙθ̇) = mRθ̈ (3.68)

56
The evolution equation is:
Fe − αvθ
θ̈ = (3.69)
mR

E - Evaluate

It is always valuable to check whether the dimensions at the left and right
side of the equal are the same:

1 kgms−2 − kgs−1 ms−1 1

[ 2] = [ ] = [ 2], (3.70)
s kgm s

which is correct.

Furthermore, the answer reveals that if Fe > αvθ , the car accelerates in the
curve (θ̈ > 0). The car has a constant angular velocity (θ̈ = 0) for Fe = αvθ ,
and decelerates (θ̈ < 0) for Fe < αvθ , in agreement with our expectation.

3.5 Electrical circuits

As a next topic, we are going to discuss how to describe electrical circuits.
We limit ourselves to direct-current (DC) circuits, that is circuits in which
there is no change of the direction of the current. We will first review some
general concepts regarding electrical circuits, after which we introduce Kirch-
hoff’s rules, as a starting point for formulating evolution equations.

As you remember from secondary school, electrical circuits are described

using currents and potential differences. The current I [A]=[C/s] is the
amount of charge flowing through an area per unit time:

dq
I =∣ ∣, (3.71)
dt

in which the vertical bars indicate that the absolute value is taken.

Charge flowing through a material consists of electrons carrying a negative

charge, or alternatively by transport of vacancies (known as holes) carrying
a positive charge. The direction of the current is defined as the direction

57
in which positive charges flow. This means that when considering a flow
of electrons through a metal wire, the current I is defined in the direction
opposite to the actual transport of electrons.

A circuit should include a source of electromotive force (emf), which main-

tains a potential difference. Note that this a poorly chosen term, as it is not
a force, but has the unit Volt [V] instead. A Volt corresponds to 1 Joule of
work on every Coulomb of charge that is transported ([V]=[J/C]). Examples
of sources of emf are batteries, power supplies and solar cells. When drawing
a source of emf in a circuit, the terminal marked with a + has a higher poten-
tial than the terminal marked with a −. In the electrical circuit the current is
directed from the positive terminal of the emf to the negative terminal (from
+ to − ). The electrons actually move in the opposite direction.

The main other components of electrical circuits that we consider here are
the resistor R [Ω] and the capacitor C [F]. When a circuit contains multiple
resistors, it is often useful to define an equivalent resistance. When resistors
are in series, the equivalent resistance is given by:

Req = R1 + R2 + R3 + ... (3.72)

For resistors in parallel holds:

1 1 1 1
= + + + ... (3.73)
Req R1 R2 R3

These equivalent resistances can be used for very simple circuits, but in many
cases we will have to use Kirchhoff ’s rules to describe the currents and po-
tential differences in more complex circuits. To this end, we need to define
two new terms: a junction and a loop. A junction is a point in an electrical
circuit where three or more wires are connected. Any closed conducting path
containing wires and electrical components is a loop. Figure 3.8 shows an
example of an electrical circuit consisting of several junctions and loops.

Kirchhoff ’s junction rule defines that the sum of the currents at a junction
is zero:
N
∑ In = 0 . (3.74)
n=0

58
 Junction a
Loop 3

R1 R2 R3
Loop 1 Loop 2
+ +
ε1 ε2

d
Junction b

Figure 3.8: Electric circuit containing three loops and two junctions.

The junction rule follows if you consider that there should be conservation of
electric charge. Kirchhoff’s junction rule is analogous to a water-pipe junc-
tion, for which the flow rate of water leaving the junction should be equal to
the flow rate of water entering the function. When using Kirchhoff’s junction
rule, the direction of the current should be taken into account. See the recipe
below for more details on how do apply Kirchhoff’s rule correctly.

Besides the junction rule, there is Kirchhoff ’s loop rule which states that the
sum of the potential differences V [V] in any loop is equal to zero:
N
∑ Vn = 0 . (3.75)
n=0

The loop rule is a consequence of that the electrostatic force is conservative.

When going around a loop (passing potential differences across various com-
ponents), you will have to reach the same potential when you arrive back at
the starting point.

Recipe: Applying Kirchhoff ’s rules

1. Make a sketch of the circuit. Label all components (resistors, capaci-
tors, emfs) with numbers.

59
 2. Also label all currents. If a current is unknown, you can assume a direc-
tion. You will find a negative sign for the current if the actual current
is in the opposite direction to what you have assumed. Already take
Kirchhoff’s junction rule into account as much as possible to minimize
the number of currents (e.g., label a current as I1 + I2 ).

3. Write down an equation for each junction based on Kirchhoff’s junction

rule.

4. For every component determine which terminal has a higher potential

and label that end with a + sign (and the other terminal with a − sign).

5. Define loops in the network and assign a direction. The direction of

the loop does not need to be the same as the direction of the currents.
Continue defining loops until each component is considered in at least
one loop.

6. By going around each loop in the assigned direction, evaluate the po-
tential differences, and work out equations according to Kirchhoff’s
loop rule. Any source of electromotive force, as well as all potential
differences over resistors and capacitors have to be taken into account,
considering the sign conventions as illustrated in figure 3.9.

7. Algebraically solve the equations to find the unknowns.

(a) (b) (c) (d)

Loop Loop Loop Loop

- -
I + I
+ - + - +

+ε -ε +IR -IR

Figure 3.9: (a-d) Sign conventions for potential differences when evaluating
a loop. You can label each terminal of emf, resistor or capacitor with a + or
− sign, depending on the direction of the current (from + to −). When going
around in a loop, the potential difference ∆V is positive if you go from − to
+, and negative if you go from + to −.

60
 t=0s

+-

R C

Figure 3.10: Electric circuit consisting of a resistor, a capacitor and an elec-

trical switch. At t = 0, the switch is closed, at which moment the capacitor
starts to discharge.

We are now going to discuss how to use Kirchhoff’s rules as a starting point
for formulating an evolution equation for the current in an electrical circuit.
As an example, we will look at the discharging of a capacitor. Figure 3.10
shows an electrical circuit containing a capacitor and a resistor. The capaci-
tor initially contains a charge Q0 [C], and is discharged through the resistor
by closing the switch at t = 0 s.

The potential differences over these two components are:

VRes = IR (3.76)

q
VCap = . (3.77)
C
The potentials at the terminals of the capacitor are indicated, from which
the direction of the current can be deduced, as indicated in figure 3.11. If we
evaluate a loop in counterclockwise direction, we get:
N
q
∑ Vn = IR − =0 (3.78)
n=0 C
We can use that the current is the rate of change of the charge (equation
(3.71)). However, do we need to use I = +dq/dt or I = −dq/dt? We know

61
 Loop
I +-

R C

Figure 3.11: The electric circuit of figure 3.10 with the direction of the current
and the loop indicated.

that the charge q should decrease when the current flows; the capacitor is
losing charge. So, the current is equal to the negative rate of change of the
charge at the capacitor. This gives the evolution equation for the charge on
the capacitor plate:
dq q
=− (3.79)
dt RC
Again, this evolution equation can be solved using separation of variables,
resulting in an expression for the time-evolution of the charge in the circuit:

t
q = Q0 e− RC (3.80)

We find an exponential decay for which a characteristic time scale can be

defined (see also exercise 9.3.18):

τ = RC. (3.81)

Check for yourself that the product of the resistance and the capacitance
indeed has the unit [s].
The evolution equation for the current can be found by simply taking the
derivatives of equation (3.80):

dI d2 q Q0 − t
=− 2 =− e RC . (3.82)
dt dt (RC)2

62
3.6 Exponential decay and characteristic timescales
In the previous sections, we have seen 3 examples of the differential equations
of similar form:
dT α
=− T (3.83)
dt Cw

dV α
=− V (3.84)
dt M

dq q
=− (3.85)
dt RC
All these physical systems can be formulated in the general form:
dN
= −CN, (3.86)
dt
with C a constant. This equation describes that the rate of change of the
function N (t) is equal to a negative constant times the function itself. Such
equations describe systems in which there is exponential decay. The solution
to the general equation is:

N (t) = N0 e−Ct , (3.87)

where N0 is given by the initial condition: N0 = N (t = 0).

To answer the question: when is the equilibrium temperature reached, or

when does the car stand still (while avoiding the answer ‘never’), a charac-
teristic timescale τ = C1 is typically calculated. At t = τ the quantity has
decreased to Ne0 and at t = 2τ to Ne20 (check this yourself). Using a simi-
lar differential equation to model radioactive decay or toxicology, typically a
half-life is defined:
ln(2)
t1/2 = = τ ln(2) (3.88)
C
Note that also if there is no exponential time-dependence, it is often possible
to define a characteristic timescale. For example, dimension analysis as dis-
cussed in chapter 1 can be performed to find a combination of variables that

63
has the dimension second. Alternatively, similar to how the constant C in
equation (3.87) should have the dimension [1/s] because the argument of the
exponent should be dimensionless, sometimes you can easily find a (combi-
nation of) variables that has the dimension [s] by evaluating a mathematical
expression.

Concluding remarks

The similarity of equations (3.83)-(3.84) illustrates that very different phys-

ical systems can often be modeled using the same mathematical expression.
In a way the main physical systems treated in this chapter can be described
in a similar way: The water tank of §3.1 is a reservoir of water characterized
by the volume V which is supplied with a volume flow rate ϕV . The object of
§3.2 is a reservoir of heat characterized by the temperature of the object T
which is supplied by a heat flow rate ϕw . The rolling car of §3.4 is basically a
reservoir of velocity which is dissipated (i.e., a negative supply) by a friction
force. When discharging a capacitor as in §3.5, you consider a reservoir of
charge q which is dissipated by a resistor R.

In this chapter, most of the examples involved first order differential equa-
tions, except for the ISEE example about the car in a curve. In chapter 5
there will more attention for second order differential equations.

64
4 Physical models
One of the main skills of a physicists is to capture the complex reality of natu-
ral and technological phenomena in a physical model. A model is a simplified
representation of a physical system, which can have many different forms. A
scale model can for example be used to study the behavior of a large object,
while a conceptual model describes a complex situation with abstractions of
elements of the real world. Examples of physical models discussed so far are
the pendulum for a swing or a clock, and the mass-spring system, which can
serve as model for vibrations of bonds in molecules.

In the last part of the previous chapter it was shown that very different physi-
cal systems can sometimes be described with similar mathematical equations.
In this chapter, we will discuss the simplification of evolution equations using
the method of nondimensionalization in order to recognize such similarities.
In addition, it will be demonstrated how nondimensionalization can help to
judge the importance of different terms in an evolution equation.

4.1 Defining a physical model

When formulating a physical model, the first step is to describe the physical
system to be studied. As briefly discussed in chapter 1, a physical system is
a part of the universe that is under study. We need to distinguish between
what we consider part of the physical system and what is the environment.
As an example, let’s consider that we are making a model for the climate
in the Netherlands. In this case, it won’t be sufficient to only consider data
collected in the Netherlands, since at least the weather systems at the Eu-
ropean scale and the trans-Atlantic Gulf stream need to be considered. On
the other hand, day-to-day temperature data measured in the Pacific can
likely be ignored, while data from the mainland of Asia can be considered as
the environment (e.g, affecting boundary conditions). To define the physical
system, you also have to decide what a relevant timescale for the study is.
Climate is about how the weather changes over decades, so changes at the
timescale of seconds are not relevant for the big picture.

A very simplistic climate model would be to consider the surface of the earth
and just one layer of the atmosphere. The state of both of these “objects” can
be described with the temperature as the dependent variable. You can write

65
down evolution equations for temperature of the surface and the one-layer-
atmosphere, similar to equation (3.21). Instead of just one heating/cooling
term at the right hand side of the equation, you will have to consider: (i )
heating by the radiation of the sun, (ii ), the reflection of light at the surface,
which in turn heats the atmosphere, (iii ) the loss of heat to the surrounding,
et cetera. You understand we end up with a complicated differential equa-
tion with many terms. Since there is interaction between the surface and the
one-layer atmosphere, you will have two coupled differential equations. Now
think about that the atmosphere consists of a troposphere, stratosphere,
mesosphere, and thermoshere, which all have different characteristics, while
there is also lateral transport of heat as a result of e.g. wind or the Gulf
stream.

Even simple models typically consist of complex (coupled) differential equa-

tions with many terms. The study of such a physical model requires an
assessment of how important the different terms of the equation are relative
to each other. By neglecting the less important terms, it is often possible to
perform analysis of the system, arriving at physical insights. An important
step towards assessing the important of different terms is to nondimension-
alize the evolution equation, which is main topic of the next section.

4.2 Nondimensionalization of evolution equations

The key idea of nondimensionalization is that we simplify the evolution equa-
tion such that it only contains dimensionless variables. A dimensionless
variable is a variable that does not have any dimension, i.e. it has no units,
sometimes expressed as [-]. Dimensionless variables are often ratios of dimen-
sional quantities. An example is the ratio A/A0 in the water tank problem
described in §3.1. Since both A and A0 have dimensions [m2 ], their ratio is
just a number and does not have a dimension. The advantage of numbers
is that you can compare them to each other. Whereas we cannot compare
areas to temperatures, one can compare a ratio of areas A/A0 to a ratio of
temperatures (e.g. T0 /T∞ in §3.2).

An evolution equation can be nondimensionalized by following a simple

recipe. We will first give that recipe here and then illustrate the approach
with a few examples.

66
Recipe: Nondimensionalization of evolution equations
1. Make an inventory of all the variables and their dimensions.

2. Scale the dependent variable and the main independent variable (e.g.,
the time) using constants or ‘typical’ values defined in the problem,
for example the initial temperature or the size of an object: Ã = AA0 ,
where à is a dimensionless variable. Independent variables that serve
as physical constants in the problem do not need to be rescaled.

3. Substitute A = ÃA0 into the evolution equation and simplify the equa-
tion. The evolution equations become dimensionless: instead of many
dimensional variables, the equation now only contains a few dimension-
less variables.

4. Also nondimensionalize the initial / boundary conditions.

5. Typically the simplified nondimensionalized evolution equation con-

tains (a ratio of) physical constants in front of one of the terms. This
is a dimensionless parameter that can be used to describe the behavior
of the system. Check whether this parameter indeed does not have a
dimension.

As an example, we will nondimensionalize the evolution equation for the

rolling car of the previous chapter:
dv α
=− v (4.1)
dt M
Our inventory of variables is:

• independent variables:
time t [s]
mass M [kg]
coefficient of friction α [kg s−1 ]
initial velocity v0 [m s−1 ]
• dependent variables:
velocity v [m s−1 ]

67
We need to nondimensionalize the dependent variable velocity v and the in-
dependent variable time t. For the rest, M , α and v0 are constants defined
in the problem.

The initial condition v(t = 0) = v0 can be used to rescale the dependent vari-
able v via ṽ = vv0 . To rescale the time, we need a combination of variables
with the dimension of [s]. A look at the inventory reveals that M /α has the
unit [s], so we can use t̃ = M
α
t.

Substituting v = v0 ṽ and t = α
M t̃ in the evolution equation 4.1 gives:

dṽv0 α
= − ṽv0 . (4.2)
d(α/M )t̃ M

This can be simplified to (check this yourself):

dṽ
= −ṽ. (4.3)
dt̃
Finally, we also have to nondimensionalize the initial condition: v(t = 0) = v0 .
We can again use v = v0 ṽ, giving:

v0 ṽ(t = 0) = v0 . (4.4)

Note that when t = 0, t̃ should also be zero. This gives as the nondimension-
alized initial condition:

ṽ(t̃ = 0) = 1. (4.5)

In this case, all the constants cancel out when we simplify the equation, so
we can skip step 5 of the recipe.

- Movie - Nondimensionalization: rolling car

Some mathematics: differentiation with respect to another variable

In a nondimensionalized differential equation there is a derivative of a di-
mensionless dependent variable with respect to a dimensionless independent

68
variable. This means that when we nondimensionalize a differential equation,
d
we have to change the variable in the differential operator, e.g, from to
dt
d
. How can we do that? You are familiar with the chain rule:
dt̃
df df dt̃
= (4.6)
dt dt̃ dt
For the re-scaling of the independent variable we use other variables that
do not depend on that specific independent variable (i.e., not on the time t
dt̃
in this case). Consequently, the derivative should be a constant factor.
dt
α dt̃ α
In the previous example, we used t̃ = t, which gives = and thus
M dt M
dv α dv
= . Note that this is exactly the same result as what we obtained
dt M dt̃
by just substituting t̃ = M
α
t into the differential operator.

Example: heating and cooling of an object

Consider again the heating or cooling of a solid object that is described by

the evolution equation

dT̄
Cw = −αT̄ , (4.7)
dt
with initial condition

T̄ (t = 0) = T0 − T∞ . (4.8)

We are asked to nondimensionalize these equations. In this example we will

follow the ISEE method.

I - Identify

There are six variables in this problem:

69
 • independent variables:
time t [s]
heat capacity Cw [J K−1 ]
heat transfer coefficient α [J s−1 K−1 ]
ambient temperature T∞ [K]
initial temperature T0 [K]
• dependent variable:
over temperature T̄ = T − T∞ [K]

We need to rescale the variables over temperature T̄ and time t by their

typical values. Note that Cw , α, T∞ and T0 are constants in the problem
and therefore do not need to be rescaled.

S - Set up
For the temperature, the most logical choice would be to rescale T̄ by the
initial over temperature T0 − T∞ . For t, there is only one other variable that
has a dimension that contains seconds, namely α (which has units [J s−1
K−1 ]). To construct a typical timescale we therefore choose Cw /α (verify
for yourself that this ratio has indeed the dimension [s]). The dimensionless
variables now read
T̄ αt
T̃ = , t̃ = . (4.9)
T0 − T∞ Cw

E - Execute

We now want to nondimensionalize equations (4.7) and (4.8).

From equation (4.9) we can express T̄ in terms of T̃ as T̄ = T̃ (T − T∞ ) and

t in terms of t̃ as t = Cw t̃/α and substitute these expressions into equations
(4.7), (4.8) to find

d(T0 − T∞ )T̃
Cw = −α(T0 − T∞ )T̃ , (4.10)
d(Cw t̃/α)
(T0 − T∞ )T̃ (t = 0) = T0 − T∞ . (4.11)

70
Bringing the constants in front of the derivative we obtain
α(T0 − T∞ ) dT̃
Cw = −α(T0 − T∞ )T̃ , (4.12)
Cw dt̃
(T0 − T∞ )T̃ (t = 0) = T0 − T∞ . (4.13)
Next, we divide both the left-hand and right-hand side of equation (4.12)
by α(T0 − T∞ ) and the left-hand and right-hand side of equation (4.13) by
(T0 − T∞ ) to arrive at
dT̃
= −T̃ , (4.14)
dt̃
T̃ (t = 0) = 1. (4.15)

E - Evaluate

Hence, instead of six, the problem now contains only two variables: T̃ and
t̃. To check our results, we can confirm that the left- and right-hand sides of
equations (4.14) and (4.15) are all dimensionless. In addition, we can check
that T̃ and t̃ are dimensionless.

From equation (4.9) we can immediately see that the effect of increasing the
heat transfer coefficient on the temperature of the object is the same as the
effect of decreasing the heat capacity. Likewise, if both the heat capacity
and the heat transfer coefficient are doubled, the temperature of the object
remains unchanged.

The system of equations (4.14), (4.15) is of a standard mathematical form

for which the solution can easily be found
T̃ = exp (−t̃). (4.16)
In fact, you will recognize that the decelerating car problem discussed above
has exactly the same form (equation 4.3) with solution
ṽ = exp (−t̃). (4.17)
Hence, nondimensionalization of the equations allows us to reduce the phys-
ical problem to a basic mathematical structure and emphasizes similarity
between different physical problems.

71
4.3 Simplifying evolution equations
In the previous examples we dealt with relatively simple models, for which
you may have concluded that nondimensionalization is strictly speaking not
a crucial step when analyzing the system. However, real-world problems of-
ten consist of a combination of many coupled physical processes, as discussed
in § 4.1. We write down the evolution equation(s) to model the balances be-
tween the different mechanisms. If we are lucky, not all of these are equally
important, and some terms may be neglected all together. Nondimension-
alization allows you to assess the importance of each term beforehand and
bring the equations into a standard mathematical form that one can then
(potentially) solve.

As an example, we consider an object falling through air that is released from

height H. Similar to the car of §3.4, the object experiences an resistance force
opposite to its direction of movement (see figure 4.1). In motion through air,
this resistance force is referred to as the air drag. For relatively high speeds,
the air drag has a quadratic dependence on the speed:
Fdrag = Dv 2 , (4.18)
in which D is the drag coefficient. Using Newton’s law, we get for the evolu-
tion equation:
dv
m = mg − Dv 2 . (4.19)
dt
We are now going to nondimensionalize this equation, such that we can an-
alyze the falling object by considering different situations. We will do that
by following the ISEE approach.

I - Identify

Six variables were defined in this problem:

• independent variables:
time t [s]
height H [m]
mass m [kg]
gravitational acceleration g [m s−2 ]
drag coefficient D [kg m−1 ]

72
Figure 4.1: Free body diagram showing the two forces acting on the object
along the y axis. Note that the positive y-direction is taken downwards, such
that the movement is in the positive y-direction.

• dependent variable:
velocity v [m s−1 ]

To nondimensionalize this evolution equation, we need to rescale v and t.

Note that the variables H, m, g and D are all constants in this problem.

S - Set up
From√ this inventory we find that the vtime t can be made dimensionless via
t̃ = Hg t, and the velocity v via ṽ = √gH . Please note that these choices are
not unique.
√ For example, the time can alternatively be made dimensionless
using t̃ = Dg
m t (see footnote below).

E - Execute
√ √
Inserting t = Hg t̃ and v = gH ṽ in the equation gives:

√ √
d( gH ṽ)
m √ = mg − D( gH ṽ)2 . (4.20)
d( H/g t̃)

If we bring the constants in front of the derivative and work out the squared

73
term then we get:
√ √ g dṽ dṽ
m gH = mg = mg − DgH ṽ 2 . (4.21)
H dt̃ dt̃
Dividing left and right with mg results in:
dṽ DH 2
=1− ṽ = 1 − αṽ 2 , (4.22)
dt̃ m
DH
where α = is a dimensionless parameter that describes the behavior of
m
the system.7

E - Evaluate

The left- and right-hand sides of equation (4.22) are dimensionless. Further-
−1
more we can check that α [ kg mkg m ]=[-] indeed does not have a dimension.

The dimensionless parameter α is basically a ratio that determines the im-

portance of drag force (∼ D) relative to inertia (∼ m), and can therefore be
used to simplify the evolution equation.

In case α >> 1, meaning DH >> m, the evolution equation simplifies to

dṽ
= −αṽ 2 . (4.23)
dt̃
This differential equation can be solved using separation of variables (see
§3.3). As a (dimensionless) initial condition we consider ṽ(t̃ = 0) = 1. This
gives for the solution:
1
ṽ(t̃) = . (4.24)
1 + αt̃
A graph of the evolution of the velocity is given in figure 4.2. The velocity
asymptotically approaches zero when time progresses. We can convince our-
selves that this solution is correct by thinking about specific examples. A
√
7
In case we would have used t̃ = Dg m
t instead, the simplified evolution equation would
√ dṽ 2
be: α dt̃ = 1 − αṽ . Please check for yourself that the analysis of the different cases gives
the same conclusions.

74
feather is an example of an object characterized by a large value for the drag
coefficient D, such that α >> 1 holds. A feather can fall initially with some
speed, but it then decreases to almost zero because of the dominating drag
force. It also makes sense to have asymptotic behavior of the velocity such
that it never becomes equal to zero.

Figure 4.2: Graphs of the dimensionless velocity as a function of the dimen-

sionless time for the two considered cases.

In case α << 1, meaning DH << m, the evolution equation simplifies to

dṽ
= 1. (4.25)
dt̃
We arrive at the evolution equation for free fall of an object without friction.
Together with the initial condition ṽ(t̃ = 0) = 0, the solution of this simplified
evolution equation is:

ṽ(t̃) = t̃. (4.26)

This case represents the famous experiment of Italian scientist Galileo Galilei.
He is supposed to have dropped two spheres with the same volume (i.e., the
same D) and different masses from the Leaning Tower of Pisa to experimen-
tally demonstrate that objects fall with the same gravitational acceleration,

75
independent of the mass. He thereby established that Aristotle’s theory on
gravitation is incorrect. From the analysis above we learn that it is crucial
for this experiment to use two heavy spheres described with a relatively low
drag coefficient, representing m >> DH, otherwise the mass m is part of the
evolution equation. In other words, he could not have performed the exper-
iments with two feathers with similar sizes.

4.4 Scale models and fluid mechanics

In the previous section it was demonstrated that the analysis of a physical
system can be modeled by considering different cases using a dimensionless
number as the control parameter. By using either a large or a small value
for the dimensionless parameter, certain terms can be neglected, thereby re-
ducing the problem to a simpler situation.

As discussed, a model is a representation of a physical system that is for

example too complex to directly perform observations or experiments. Al-
though it is oftentimes not doable to capture the full complexity of a physical
system, good practice is to validate the model by carrying out experiments
for specific conditions of the physical system.

When a physical phenomenon or system is too large or too small to investi-

gate on lab scale, a scale model can be defined, for which nondimensional-
ization plays an important role. If the response of a system depends only on
a specific combination (dimensionless parameter) of all variables, one could
change individual dimensional variables (such as the size of an object), while
keeping the dimensionless parameter constant and thus not changing the
evolution equation. When you change an individual dimensional variable in
this case, you also have to change the other dimensional variables included
in the dimensionless parameter in order to keep the dimensionless parameter
constant. This is an extremely powerful concept that is used a lot in, e.g. the
airplane industry, where wind tunnel tests could be performed with much
smaller scale models of a real airplane.

Fluid mechanics is field of physics in which dimensionless parameters are

frequently used to describe complex fluid behavior. Such dimensionless pa-
rameters are often referred to as characteristic numbers. A large collection of

76
different dimensionless parameters has been defined in the literature. Some
relevant examples to highlight are:

The Reynolds number:

ρuL
Re = , (4.27)
µ
ρ [kg m−3 ] is the fluid density
u [m s−1 ] is the flow speed
in which
L [m] is a characteristic length scale
µ [kg m s ] is the dynamic viscosity of the fluid.
−1 −1

The Reynolds number is a measure for the ratio between inertial and viscous
forces, and it is often used to distinguish laminar from turbulent flow. Low
Re number flows correspond to laminar flow, i.e., smooth flow without mix-
ing, while turbulence can develop in high Re number flows.

The Knudsen number:

λ
Kn = , (4.28)
L
λ [m] is the mean free path
in which
L [m] is a characteristic length scale.
The mean free path represents the average distance a molecule can move
before changing its direction as a result of collision(s) with other molecules.
The Knudsen number can be used to decide whether a fluid can be considered
as a continuum, or whether it should be described with statistical mechanics.

The Mach number:

u
M= , (4.29)
c
u [m s−1 ] is the flow velocity with respect to an object or boundary
in which
c [m s−1 ] is the speed of sound.
You may recognize the Mach number from descriptions of airplanes passing
the sound barrier. This happens at a speed corresponding to M = 1. Mach
numbers of 1.2 – 5.0 are typically referred to as supersonic speeds.

77
5 Elementary oscillations and mechanical waves
Many dynamic systems will, when they are brought out of their equilibrium
state, show behavior that repeats itself: periodic or oscillatory behavior.
Simple examples of systems that display this behavior are the pendulum and
the mass-spring system. When the amplitude of the oscillation is sufficiently
small, the governing equations are linear and the system shows harmonic
oscillations. In that case we can usually find analytical, mathematical solu-
tions for the behavior of the system. Another type of dynamic behavior is the
behavior of populations, which can be captured in models. Furthermore, the
formation of mechanical waves, propagating disturbance within a medium,
is an additional type of dynamical behavior. In this chapter we will first
describe different harmonic oscillations and their evolution equations. Later
in the chapter we will discuss population models, mechanical waves and their
properties.

5.1 Harmonic oscillations

Imagine a ball in a spherical bowl rolling back and forth (see figure 5.1(a)).
The ball will accelerate going from the top to the equilibrium point at the
bottom of the bowl and will decelerate going to the other top. This motion
repeats itself over and over. Such dynamic systems that show periodic (or
oscillatory) behavior are called harmonic oscillators. The definition of an
harmonic oscillation is an oscillation that shows repetitive behavior with
respect to the equilibrium position. In the special case where the restoring
force is linearly dependent on the displacement, the oscillation is called simple
harmonic oscillation.
In simple harmonic motion the total mechanical energy (kinetic + potential
energy) of the system is conserved, which is why the motion can keep re-
peating itself. In the example of the ball in the bowl, the kinetic energy K
increases when moving from the top towards the bottom, while the potential
energy U decreases. When the ball moves to the other top, K decreases and
U increases. When its kinetic energy becomes zero, the ball stops, returns
and will roll back down due to gravitational forces.

At the ‘top’, the horizontal displacement of the ball x (see figure 5.1(a))

78
(a) (b) E

E=K+U
x K

U
-A A x

Figure 5.1: (a) Schematic representation of a ball in a spherical bowl rolling

back and forth. (b) Kinetic energy K, potential energy U and total energy E
as function of displacement x for simple harmonic motion.

reaches its maximum displacement, which is called the amplitude A of the

motion, where K = 0 and U = Umax . At the stable equilibrium position,
x = 0, K = Kmax and U = 0. The kinetic, potential and total energy E in
such a simple harmonic motion are displayed schematically in figure 5.1(b)
as function of the displacement x.

5.2 The pendulum

The pendulum is one of the most well-known oscillating systems: it consists
of a rigid and very thin, light rod of constant length l with a point mass m
attached to its end, while the other end is attached to a point O where it
is allowed to pivot, so that the pendulum can move freely back and forth in
one vertical (two-dimensional) plane (see figure 5.2).

The evolution equation, also called the equation of motion, for the swaying
pendulum can be derived from an analysis of the forces exerted on the mass.
Since the mass at the end of the rod moves on a circular path, we use polar
coordinates here for the force analysis. The advantage of this approach is
that one of the coordinates, r, stays constant during the motion. As figure
5.3 shows, we can decompose the gravitational force (the weight Fg ) in a
component mg cos θ in the radial direction (pointing in the direction away
from the suspension point O) and a component mg sin θ in the azimuthal
direction (parallel to the circular trajectory of the moving mass).

79
 O
g

θ
l

Fz = mg

Figure 5.2: The mathematical pendulum.

We assume that the component mg cos θ is compensated by the tension in the

pendulum’s rod. Furthermore, we assume that frictional effects are negligibly
small. Then, the azimuthal component mg sin θ of the gravitational force will
result in an acceleration of the mass along the circular trajectory (in the θ-
direction). Using equations (2.35) Newton’s second law in polar coordinates
is given by
⎡ ⎤
⎢ ⎥
⎢ 2⎥
⎢
Fr = m ⎢r̈ − r (θ̇) ⎥⎥ ,
⎢ ²⎥
⎢ ⎥
⎣ term A ⎦ (5.1)
⎡ ⎤
⎢ ⎥
Fθ = m ⎢⎢rθ̈ + 2ṙθ̇ ⎥⎥ .
⎢ °⎥
⎣ term B⎦

Terms A and B in equation (5.1) are the centrifugal acceleration and the
Coriolis acceleration respectively. Since the acceleration of the pendulum is
in the azimuthal direction, we will use Fθ here.

The rod is assumed to be rigid and of constant length. Thus, the radius of
the circular motion is constant and equal to the length of the rod, i.e. r = l

80
 O
g

θ
l

Direction of motion

mg sin
mg cos

Fz = mg

Figure 5.3: Force analysis of the swaying pendulum.

and ṙ = 0. Under these conditions Newton’s second law in the azimuthal

direction takes the following shape:

mlθ̈ = −mg sin θ , (5.2)

i.e.:
d2 θ g
+ sin θ = 0 . (5.3)
dt2 l
Such equations of motion are other examples of evolution equations. This
equation of motion for the pendulum is a second order differential equation.
As you can see we need only one equation to describe the entire motion of the
pendulum. When using Cartesian coordinates we would have needed two!
The term sin θ is a non-linear function of the excursion angle θ, and this
forms a complication when solving this equation of motion. If we limit our-
selves to small excursion angles θ we can approximate the term sin θ and
linearize the equation to allow for a simple solution. This approximation can
be done by using a Taylor series.

81
Some mathematics: Taylor series around x = 0
A Taylor series or Taylor expansion is an approximation of a function in
and around a certain point in terms of a power series of the independent
variable(s). Here we will show the Taylor expansion around x = 0, while in
§7.2.3 Taylor expansions around an arbitrary point will be explained.
Consider a function y = f (x) and let us investigate if we can approximate
this function in the vicinity of x = 0 by using a polynomial of the N th order
(with N a natural number). So, we write

N
f (x) ≈ TN (x) = c0 + c1 x + c2 x2 + . . . + cN xN = ∑ cn xn . (5.4)
n=0

The question now is how we can determine the coefficients c0 , c1 , ⋯, cN such

that the polynomial indeed is an adequate approximation of the function
f (x) (at least in and around x = 0). Without proving why (this will be done
in the mathematics course Calculus C, 2WCB0), we pick the aforementioned
coefficients as follows:

c0 = TN (x = 0) = f (x = 0),
dx x=0 = 1 dx x=0 = f (x = 0),
c1 = 11 dT N
∣ 1 df
∣ ′

1 d2 TN 1 d2 f
c2 = 1⋅2 dx2 ∣x=0 = 1⋅2 dx2 ∣x=0 = 2 f (x = 0),
1 ′′

1 d3 TN 1 d3 f (5.5)
c3 = 1⋅2⋅3 dx3 ∣x=0 = 1⋅2⋅3 dx3 ∣x=0 = 6 f (x = 0),
1 ′′′

⋮ ⋮
1 dN TN 1 dN f
cN = N ! dxN ∣x=0 = N ! dxN ∣x=0 = 1 (N )
N! f (x = 0),

with N ! = 1 ⋅ 2 ⋅ 3 ⋅ 4 ⋅ . . . ⋅ N and 0! = 1. In words this says that the first to the

N th derivative of the function f (x) evaluated in x = 0 are equal to the first
to N ! times the N th coefficient of the polynomial TN (x) in x = 0.

We now have the following expression for our approximation of the function
f (x) around x = 0:

f (x) ≈ ∑N n=0 cn ⋅ x
n

= f (0) + 1 f (0) ⋅ x + 1⋅2

1 ′
f (0) ⋅ x2 + 1⋅2⋅3
1 ′′ 1
f ′′′ (0) ⋅ x3 + . . .
(5.6)
+ 1⋅2⋅3⋅...⋅N f (0) ⋅ x
1 (N ) N

= ∑N 1 (n)
n=0 n! f (0) ⋅ xn ,

82
with

dn f
f (n) (x) = . (5.7)
dxn
It seems plausible that as we include more terms in this Taylor expansion
(so if we increase N ), we get a more accurate approximation of the function
in question and this is often the case, but not always (more about this in
Calculus).

Above, we talk about the N th order Taylor series of a function. In most

cases we want a specific approximation, for instance the 2nd order Taylor
series. This means N = 2 and you include the 0th, 1st and 2nd term in the
polynomial:
2
1 (n) 1 1 1
T2 (x) = = ∑ f (0) ⋅ xn = f (0) (0)⋅x0 + f (1) (0)⋅x1 + f (2) (0)⋅x2 . (5.8)
n=0 n! 0! 1! 2!
Important to remember is that 0! = 1 and that the 0th order derivative of a
function is the function itself.

Example

Consider the function y = f (x) = sin(x). According to equation (5.6), the

Taylor expansion of this sinusoidal function around x = 0 then becomes

sin(x) ≈ sin(0) + dx {sin(x)}∣x=0 ⋅ x + 2 dx2 {sin(x)}∣x=0

d 1 d2
⋅ x2 + . . .
(−1)m 2m+1
(5.9)
=0+1⋅x+ 2 ⋅ 0 ⋅ x − 6 ⋅ x + . . . = ∑m=0 (2m+1)! x
1 2 1 3 M
.

We see that this expansion only contains odd powers of x. This observation
is no surprise since the sine function is an odd function of x. Figure 5.4 shows
the sine function together with three Taylor expansions (N = 1, 3, and 5 or
M = 0, 1, and 2). It can clearly be seen that the approximation improves as
we include more terms in the Taylor expansion. The most basic, non-trivial
approximation for the sine function around x = 0, i.e. ∣x∣ ≪ 1, is (N = 1 or
M = 0), is

sin(x) ≈ x for ∣x∣ ≪ 1. (5.10)

83
 1st order
f(x)= sin(x) Taylor series
of f(x)
5th order
3rd order
Taylor series
Taylor series
of f(x)
of f(x)

Figure 5.4: (a) Function sin x (solid), (b-d) sin x with its Taylor series of
equation (5.9) where (b) M = 0 (N = 1) (dash), (c) M = 1 (N = 3) (small
dash) and (d) M = 2 (N = 5) (dot).

Since we now approximate the sine function in the vicinity of x = 0 with a

straight line, we say that the sine function has been linearized around x = 0:
it is approximated by a linear function.

A linearized function means that the function is approximated using a Taylor

series of the 1st order, N = 1. This includes the 0th and 1st order term. In
this example of the sine function the 0th term is equal to 0 and the 1st order
term is equal to x.

- Movie - Taylor series

Back to the pendulum equation. If we limit the approximation of the non-

linear term in equation (5.3) to the 1st order we can apply the following
approximation or linearization:

84
 sin θ ∼ θ when ∣θ∣ << 1 . (5.11)

With this linearization equation (5.3) reduces to

d2 θ g
+ θ=0 . (5.12)
dt2 l
Equation (5.12) is the evolution equation of the pendulum for small angles
of θ and it is a 2nd order differential equation.
By solving equation (5.12) an expression of the angle θ as function of time is
found.

Some more mathematics: solving second order differential equa-

tions
In this course we will encounter different 1st and 2nd order differential equa-
tions. Solving 1st order differential equations was discussed in chapter 3.3.
The general solution to a 2nd order differential equation is a linear combi-
nation of two solutions. Boundary or initial conditions determine the final
specific solution.
Here we will limit ourselves to solving two different types of 2nd order dif-
ferential equations by using standard solutions.
The first type is given by:

d2 y
+ c2 y = 0 . (5.13)
dx2
Both a sine function and a cosine function are a solution to equation (5.13).
You can check this by filling in y = sin(cx) or y = cos(cx) in equation (5.13).
Thus, the general solution to equation (5.13) is given by:

y = A sin(cx) + B cos(cx) , (5.14)

with A and B integration constants. The sine and cosine function are linearly
independent and that is why we can add them for the general solution.

85
The other type is given by:
d2 y
− c2 y = 0 . (5.15)
dx2
Both an exponential function with a positive argument and an exponential
function with a negative argument are a solution to equation (5.15). Again,
check for yourself whether this is true. Thus, the general solution to equation
(5.15) is given by:
y = Ae+cx + Be−cx , (5.16)
with A and B (different) integration constants.

Equation (5.12) is of the same type as equation (5.13), and therefore the
general solution is a linear combination of sines and cosines, just like equation
(5.14).
For the case where the pendulum is released at t = 0 from a starting position
θ = θ0 with zero initial velocity the following initial conditions apply:
t=0 ∶ θ = θ0 , θ̇ = 0 . (5.17)
The solution that satisfies these initial conditions is
θ(t) = θ0 cos ωt (5.18)
with the angular frequency
√
g
ω= . (5.19)
l
Apparently, for small excursion angles θ, the pendulum shows a purely har-
monic oscillation, with a period
√
2π l
T≡ = 2π . (5.20)
ω g
√
This relationship T ∼ l was already determined experimentally by √ Galileo
Galilei in 1602. Furthermore, we already found the relationship T ∼ l/g in
chapter 1 using dimensional analysis. We will show in chapter 8 that the rel-
atively simple behavior of equation (5.20) is quickly lost when the excursion
angle θ is no longer small.

86
5.3 The mass-spring system
Another well-known oscillatory system is the mass-spring system. This sys-
tem consists of a point mass m connected to a spring, where the mass can
move back and forth along the x-axis (see figure 5.5(a)).
The mass-spring system is another example of a dynamic system where we
can use linearization to make the problem suitable for an analytical descrip-
tion. The linearization is applied frequently in the analysis of dynamic sys-
tems as long as it concerns small excursions from a (stable) equilibrium state
(more on equilibrium states will follow in chapter 7).

F
F(x) F(x)
m
x
O
x

(a) (b)

Figure 5.5: (a) Mass-spring system, with mass m is connected to a spring.

(b) Graphical interpretation of the spring force F (x) as a function of the
displacement x.

The equilibrium state of the point mass - where the spring is neither pressed
nor stretched, and therefore exerts no force - we call x = 0. Upon a displace-
ment x the spring exerts a force F (x) in the negative x-direction. In general
the spring force F (x) can be quite a complicated function of x, depending on
the detailed elastic properties of the spring, such as the example sketched in
figure 5.5(b). However, in most cases F (0) = 0, since x = 0 is the equilibrium
state by definition. If we neglect frictional effects the equation of motion for
the mass is the following:
d2 x
m = −F (x) . (5.21)
dt2
The function F (x) can now be expanded in a Taylor series around x = 0:
x2 ′′ x3
F (x) = F (0) + xF ′ (0) + F (0) + F ′′′ (0) + ... (5.22)
2! 3!
87
An example of the function F (x) and different orders of its corresponding
Taylor series are shown graphically in figure 5.6. Often, it is true that the
higher the order of the Taylor series, the more accurate the approximation
is (not always!).

F(x) 1st order

Taylor series
of F(x)

5th order
Taylor series
of F(x)
3rd order Taylor
series of F(x)

Figure 5.6: (a) Graphical interpretation of the spring force F (x) (solid) as a
function of the displacement x, (b-d) F (x) with its Taylor series where (b)
N = 1 (dash), (c) N = 3 (small dash) and (d) N = 5 (dot).

If we limit ourselves again to small displacements ∣x∣ with respect to the

equilibrium state, we can approximate (5.22) by

F (x) ≃ F (0) + xF ′ (0) , (5.23)

and since F (0) = 0 this becomes

F (x) ≃ αx , (5.24)

where α = F ′ (0) is the so-called (positive) spring constant. Equation (5.24)

may be familiar to you as Hooke’s law.

88
Because of this linearization the equation of motion (5.21) simplifies into
d2 x
m = −αx , (5.25)
dt2
valid for small ∣x∣. With
√
α
ω= (5.26)
m
this equation can be written as
d2 x
+ ω2x = 0 , (5.27)
dt2
and this is of the same type as equation (5.13). Thus, the general solution is

x(t) = A cos ωt + B sin ωt , (5.28)

where A and B are constants that can be determined from the initial condi-
tions for x(t = 0) and ẋ(t = 0).
Apparently, when the displacements from the equilibrium state x = 0 are
small, this system also shows harmonic oscillations with an angular frequency
ω. The period of an oscillation is T = 2π/ω, and the frequency f (i.e. the
number of oscillations per unit of time) is then f = ω/2π.
For the initial conditions x(t = 0) = x0 and ẋ(t = 0) = 0 the constants in
equation (5.28) are equal to A = x0 and B = 0, and therefore

x(t) = x0 cos ωt . (5.29)

From this the instantaneous velocity ẋ(t) and the acceleration ẍ(t) of the
point mass follow directly:

ẋ(t) = −x0 ω sin ωt

(5.30)
ẍ(t) = −x0 ω 2 cos ωt .

The evolution of x(t), ẋ(t), and ẍ(t) is represented graphically in figure 5.7.

We can see that the displacement x(t) and the velocity ẋ(t) are exactly 90○
out of phase, whereas the acceleration ẍ(t) has an opposite phase to x(t).

89
 x
x0

(a) t

x
x0ω

(b)
t

x
x0ω2

(c)
t

Figure 5.7: Graphical representation of the course of the (a) normalized dis-
placement x/x0 ,(b) velocity ẋ/x0 ω, and (c) acceleration ẍ/x0 ω 2 .

5.3.1 Kinetic and potential energy

We can also consider the mass-spring system in terms of energy. This can
give us insight in the behavior of the system without explicitly solving the
evolution equation.
Starting again from the equation of motion for the mass-spring system equa-

90
tion (5.21) in its general form, we can rewrite the spring force F (x). The
total force f (x) is equal to minus the spring force −F (x) and the total force
is in the direction of decreasing potential energy, therefore
dV
f (x) = − , (5.31)
dx
where the function V (x) is the potential energy that is stored in the spring
at a certain displacement x.
This also becomes clear when we integrate equation (5.31) over x:
x
∫0 f (x̃) dx̃ = −V (x), (5.32)

in other words: by performing work (force integrated over distance x) poten-

tial energy is stored in the spring.
We now write equation (5.21) with equation (5.32) as

dV
mẍ + = 0. (5.33)
dx
We would like to see if we can rewrite this expression into the form dE/dt,
since the goal is to learn about the energy of the system. Therefore, we want
to rewrite the potential energy term into a time derivative in order to make
it of the same form as the first term. To this end, we multiply both terms of
equation (5.33) with ẋ, which gives:

dV
mẋẍ + ẋ = 0. (5.34)
dx
We are going to rewrite this into two time derivatives with a trick worth
remembering. Using the chain rule, we can write (please check for yourself
that these are correct):

d 1 2 dV dx d
ẍẋ = ( ẋ ) and = [V (x(t))] , (5.35)
dt 2 dx dt dt
and rewrite equation (5.34) to

d 1 d dE
[ mẋ2 + V (x)] = [Ekin + Epot ] = 0 → = 0. (5.36)
dt 2 dt dt

91
In other words: the total energy consisting of the sum of kinetic energy Ekin
and potential energy Epot
1
E = Ekin + Epot = mẋ2 + V (x) (5.37)
2
of the mass-spring system is conserved. When this energy is conserved the
system is called conservative. In figure 5.1 the conservation of the total en-
ergy is depicted graphically as well by the line at a constant horizontal height
depicting total energy E.

5.4 Effect of damping

So far, we have neglected frictional effects. In real life, oscillating systems
such as the pendulum and mass-spring system will experience friction. As
you know, the pendulum will sway slower and slower over time and the mass
on the spring will also move slower and slower over time.
If we take the example of the mass-spring system in figure 5.5 again and
assume that the mass is subject to friction, the equation of motion changes.
In the most basic description the friction is assumed to be opposing the
motion and proportional to the velocity ẋ of the mass, introducing an extra
force −γ ẋ in the x-direction, where γ is a positive constant.
The equation of motion of (5.25) is then extended to
d2 x dx
m 2
= −αx − γ , (5.38)
dt dt
again under the restriction of small displacements ∣x∣ ≪ 1. This equation can
also be written as
d2 x dx
2
+k + ω2x = 0 , (5.39)
dt dt
where k ≡ γ/m and ω is again defined as in equation (5.26). It can be proven
(see exercise 9.5.6) that the general solution to this equation is
⎧ √ ⎫
− 12 kt
⎪
⎪ 1 ⎪
⎪
x(t) = Ce cos ⎨ (ω 2 − k 2 ) t − D⎬ , (5.40)
⎪
⎪ 4 ⎪
⎪
⎩ ⎭
under the condition that the damping is not too big (i.e. 14 k 2 < ω 2 ).

92
The constants C and D can be determined from the initial conditions for
x(t = 0) and ẋ(t = 0).
This solution shows that the amplitude of the oscillation decreases exponen-
tially with time, more specifically with a factor exp(− 12 kt). An example of
this damped oscillation can be seen in figure 5.8.

Figure 5.8: Example of a damped harmonic oscillation and the decaying ex-
ponential function.

An additional effect of the damping is a decrease in the oscillation frequency

from ω to (ω 2 − 14 k 2 )1/2 .
We can again derive an evolution equation for the total energy in the damped
mass-spring system by multiplying the terms in the equation of motion of
(5.38) by ẋ:
mẍẋ + αxẋ + γ ẋ2 = 0 (5.41)
which we can rewrite into
d 1 1
[ mẋ2 + αx2 ] + γ ẋ2 = 0, (5.42)
dt 2 2
which can also be written as
dE
= −γ ẋ2 < 0. (5.43)
dt
93
In other words, the total energy E = Ekin + Epot of this damped system de-
creases, meaning that the system is no longer conservative.

5.5 Effect of forcing

Let us now look at a situation where the mass in figure 5.5(a) is not only
brought in motion by the spring force −αx, but also by a set driving force
F0 cos Ωt, which varies harmonically with frequency Ω. In absence of damping
the equation of motion for small ∣x∣ then becomes

d2 x
+ ω 2 x = a cos Ωt , (5.44)
dt2
with a = F0 /m.
This equation was already studied in 1739 (!) by Euler, and he found that
interesting things happen when the forcing frequency Ω is roughly equal to
the natural frequency ω of the mass-spring system: in that case resonance
occurs, where the oscillations have a monotonously increasing amplitude (see
figure 5.9 (b)). Figure 5.9 shows the solution of equation (5.44) for the cases
Ω = 0.8ω and Ω = ω; in this last case the resonance can clearly be seen.

Figure 5.9: Solutions x(t) of equation (5.44) for (a) Ω = 0.8ω and (b) Ω = ω.

94
It can be shown (see exercise 9.5.7) that for Ω = ω
a
x(t) = t sin ωt (5.45)
2ω
is a solution of equation (5.44) that satisfies the initial conditions x(t = 0) = 0
and ẋ(t = 0) = 0. From equation (5.45) we see that the amplitude of the
resonant oscillation increases linearly with the time t (figure 5.9(b)).

5.5.1 Combination of damping and forcing

When some damping is applied to a forced system, the oscillation amplitude
will initially still grow, but after some time a state can be reached with con-
stant amplitude, depending on the amount of damping and forcing. In the
case of a constant amplitude there is a balance between the driving force and
the damping force. The course of x(t) at damped resonant forcing is shown
in figure 5.11.

Figure 5.10: Course of the solution x(t) with damped resonant forcing.

95
5.6 Population model
So far we introduced the pendulum and a mass-spring system as simple
mechanical systems that show periodic behavior. However, there are many
natural systems that can show periodic behavior. One example we will treat
here is the population model of Lotka and Volterra. This model describes
the competition between two animal species.
The model was initially developed for explaining significant fluctuations in
the amount of fish caught in the Adriatic Sea in the period 1905 - 1923. It
was inspired on the observation that the amount of big and small fish caught
showed variations with the same frequency, but slightly out of phase. The
model can also be used more generally for the analysis of any system where
a competition between two parties (e.g. prey and predator) is present.

For simplicity we will first assume that there are no natural enemies. In
this case the animals can reproduce to their heart’s content. We can then
expect that the increase of the population size N1 (= number of animals) is
proportional to N1 :

dN1
= αN1 , (5.46)
dt
where α is the growth factor [s−1 ]. This differential equation has an expo-
nential solution:

N1 (t) = Ceαt , (5.47)

where the constant C is determined using the initial condition N1 (t = 0) =

N10 :

N1 (t) = N10 eαt . (5.48)

For the case where α > 0 (sufficient food, healthy, etc.) the population
grows exponentially, whereas for α < 0 (insufficient food, diseases, etc.) the
equation resembles the equation for exponential decay (see equation (3.86))
and a decrease in the population size N1 occurs.

96
The typical timescale of this biological system is τ = 1/α. We could have
obtained this result with an analysis of the dimensions of the relevant in-
dependent variables in the system as well. Since N1 is dimensionless (an
‘amount’ is a number, so [-]) only the growth factor α remains, which has
the dimension [s−1 ]. So, the timescale of this system will have the following
proportionality relation: τ ∼ 1/α.

We can go one step further and assume that the species is threatened by
predators. In a first approximation we could say that the decrease in the
population N1 is proportional to N1 , so that equation (5.46) becomes:

dN1
= αN1 − βN1 (5.49)
dt
where α(> 0) is the growth factor and β(> 0) the mortality as a result of the
predators. The solution to this equation then is:

N1 (t) = N10 e(α−β)t . (5.50)

When α = β the population is stable (growth = 0), when α > β the popula-
tion increases, whereas when α < β the number of animals (N1 ) decreases.

We can make the population model even more realistic by assuming that
the mortality of the animals also is determined by the amount of predators
(N2 ), so: β = β(N2 ). In biological population dynamics the following is often
assumed:

dN1
= α1 N1 − γN1 N2 , (5.51)
dt
and for the predators (N2 ):

dN2
= −α2 N2 + γN1 N2 . (5.52)
dt
Here, α1 is the growth factor of the population N1 , α2 the mortality of the
population N2 , and γ a constant ‘interaction’-factor. The behavior of N1

97
and N2 is illustrated as function of time in figure 5.118 . We can see that
the population of both the prey and predators changes periodically. The
oscillations are out of phase with respect to each other, this is due to the
response they have as a result of the interaction.

Figure 5.11: Plots (a) of the solutions x(t) = N1 (t) and y(t) = N2 (t) and (b)
of the (N1 , N2 ) phase diagram of the Lotka-Volterra model (5.51,5.52) for
α1 = 2, α2 = 1, γ = 0.1 with (x0 , y0 ) = (10, 2) and (x2 , y2 ) = (10, 20).

When the predators do not eat their prey, so when γ = 0 or when N1 = 0 (all
preys are dead), equation (5.52) has a simple solution: N2 = N20 e−α2 t , which
implies that the population of the predators will decrease.

5.7 Mechanical waves

So far, we described different kinds of periodic motion where we looked at
the displacement at a certain point. However, it can be the case that the
disturbance of the medium/material takes time to propagate through the
material, think about a very long spring for the mass-spring system. With
disturbance we mean local periodic behavior around an equilibrium point in
a material.
The waves that travel within a medium are called mechanical waves. A wave
is an example of periodic motion around an equilibrium point. Particles in
this medium move up and down or from left to right around this equilibrium,
8
These plots have been made with a variant of the Matlab program XTPHASE. Matlab
programs and descriptions of those programs can be found on Canvas.

98
but they do not show a net displacement. By contrast, the pattern itself dis-
places. Examples of mechanical waves are sound waves (a disturbance of
air), water waves, waves on a rope and waves in a spring.

Let us consider a mass-spring system once more. As shown in figure 5.12,

the spring is vertically aligned with the mass on top and a rope is now
attached to the mass. The other side of the rope is connected to a point in
the far distance. As the (block of) mass moves up and down, a wave travels
over the rope. The mass itself undergoes a simple harmonic motion up and
down, as does each individual point on the rope. However, the disturbance
pattern itself travels from left to right on the rope and thereby forms a wave.
This type of mechanical wave for which the direction of the disturbance is
perpendicular to the direction of the wave is called a transverse wave.

v
A

up
&
down λ

Figure 5.12: Block of mass on top of the spring moves up and down in a
simple harmonic motion, creating a sinusoidal wave in the rope traveling to
the right. The rope is connected to a point far to the right. The transverse
wave has an amplitude A, a wavelength λ and a wave speed v.

Periodic waves like the one just described have a repeating pattern. A pe-
riodic wave has the following properties: wavelength λ, spatial distance be-
tween two disturbances; wave speed v, speed with which the disturbance
pattern travels; and frequency f , frequency with which the pattern repeats
itself. These properties are related to each other in the following way:

v = λf. (5.53)

In this course we will assume that the wave speed is only dependent on the
mechanical properties of the system, and that upon changing λ, f changes

99
and vice versa. When considering light waves, as you will do in the Optics
course (31OPT), the wave speed will also depend on the frequency.

Besides transverse waves, there is another type of periodic waves, namely

longitudinal waves. The medium in a longitudinal wave is not moving up
and down, but is being compressed and dispersed. Now the direction of the
disturbance is parallel to the direction of the wave.
Suppose there is a long tube filled with a gas and on one end there is a piston.
If the piston moves into the tube, the gas is compressed near the piston. The
compressed region pushes against next region and so forth. If the piston
moves back again, the region near the piston is dispersed. Now imagine
that the piston is moving back and forth in a simple harmonic motion. A
longitudinal wave will travel through the tube.
The wavelength is defined as the distance between two points that are both
fully compressed. The particles in a medium that is undergoing a longitudinal
wave are moving back and forth in the same harmonic motion as the driving
force (here the piston). Again, the wave speed is related to the speed of the
wave itself and not to the motion of the individual particles.

5.7.1 Mathematical description of a wave

Periodic waves, both transverse and longitudinal, can be described math-
ematically. Let us reconsider the example of figure 5.12. As any periodic
motion can be described as a sum of sines and cosines (you will learn more
about this in the course Vector Analysis (32VAN), we look at the most basic
description (a cosine) here.

The height y of a point on the rope is a function of both time t and coordinate
x. At the x = 0 the point is moving up and down with period T (or frequency
f ). At t = 0 point x = 0 is at the maximal height, y = A. A quarter of the
period later, at t = T /4, the point is at the equilibrium height y = 0. Another
quarter period later, at t = T /2, the point is at the lowest height, y = −A,
and so on. The displacement of the point at x = 0 is given by
2π
y(x = 0, t) = A cos(ωt) = A cos ( t) , (5.54)
T
with angular frequency ω = 2πf = 2π/T .

100
Since the wave is traveling in the +x-direction with wave speed v, the motion
of a point at position x at time t is the same as the motion of the point at
x = 0 at time t − x/v. The full mathematical description of the wave, the
wave function, then becomes
x t
y(x, t) = A cos [2π ( − )]
λ T
x (5.55)
= A cos [ω ( − t)]
v
= A cos (kx − ωt) ,

with wave number k = 2π/λ.

Equation (5.55) gives information on what the vertical position of points on

the rope is in time and in space. In order to understand the behavior further
we could look at the transverse velocity and acceleration and at the slope
and curvature of the rope. For this we will need partial derivatives.

Some mathematics: partial derivatives

Besides functions of one variable, we can also differentiate functions of mul-
tiple variables using the following. An example of a function of multiple
variables is the temperature in a large room. The temperature will depend
on the place and height, so T = T (x, y, z). If you want to know how T
changes when you walk in the x-direction you will need the derivative in the
x-direction, ∂T
∂x .
Suppose we now have a function of two variables f (x, y) and we want to
differentiate this with respect to the variable x. We keep the other variable
y constant (so y then has a set value) and apply the normal differentiation
recipe. The result is known as the partial derivative of the function f with
respect to the variable x. We use the following notation for this derivative:
∂f
. (5.56)
∂x
The ‘curved’ d (∂) denotes the (partial) differentiation of the function f with
respect to one of the independent variables (in this case x) where all other
independent variables (in this case y) are kept constant.

101
Example
√
Consider the function f (x, y) = x2 y + y. For the partial derivatives of this
function we find
∂f ∂f 1
= 2xy, = x2 + √ . (5.57)
∂x ∂y 2 y

The partial derivatives of the wave function with respect to t are related to
changes of the y(x, t) in time. So, the first and second partial derivatives of
the wave function with respect to t represent the transverse velocity vy and
the transverse acceleration ay , respectively, i.e. how the local rope height
changes in time. On the other hand, the partial derivatives with respect
to x are related to the shape of the rope, i.e. how the rope height changes
in space. The first and second partial derivative of the wave function with
respect to x represent the slope and curvature of the rope, respectively.
In fact, the second partial derivatives with respect to x and to t are related
by the following relationship:

∂ 2 y(x, t) 1 ∂ 2 y(x, t)
= 2 , (5.58)
∂x2 v ∂t2
which is called the wave equation.
The wave equation is one of the most important equation in physics. You
can show (we will not do that here) that not only equation (5.55) satisfies
the wave equation, but all functions describing a periodic wave, regardless of
the shape, satisfy the wave equation, since you can express each shape as a
sum of sines and cosines and equation (5.58) is linear.
One outstanding example of the wave equation is in the theory of electro-
magnetism. Maxwell showed that both the magnetic and electric field satisfy
the wave equation. The resulting wave speed is actually the speed of light,
proving that light is an electromagnetic wave.

5.7.2 Determining the wave speed

Light is an example of an electromagnetic wave, with a wave speed of 3.00⋅108
m/s. Sound is another example of a wave, but the wave speed is much lower,
∼344 m/s. The wave speed is an important property of any wave. Waves

102
traveling on a rope, like in figure 5.12, also have a wave speed: a speed with
which the disturbance propagates.
We have mentioned before that the wave speed depends on the mechanical
properties of the system. In general, the wave speed does not depend on the
wavelength and frequency. If one would change the frequency, the wavelength
would change, and the wave speed would remain the same. Wave speed can
be calculated based on the mechanical properties of a system, as we will show
now.
If we consider the exemplary example of a transverse wave on a rope, the
mechanical properties of the system are the tension T in the rope and the
mass per unit length µ. To determine the relation between v and T, µ, you
can consider the forces on a rope segment ∆x starting at x and ending at
x + ∆x, see figure 5.13, and apply Newton’s second law to this small segment
of rope.
Since the motion is transverse (particles are only moving up and down), there
is no net force in the x-direction, only a net force in the y-direction. If the
rope is at rest there is only the tension T acting on both sides of the segment
in opposite directions. If the rope is displaced, this horizontal component of
the total force stays constant, since there is no net force in this direction. In
the y-direction there will be a net force, resulting in ay .

F
Fy

T
rope

T
Fy
F Δx

x x + Δx

Figure 5.13: Force analysis of forces on segment of rope.

It can be derived that the net force in the y-direction on a small portion of
the rope ∆x is given by
∂y ∂y
Fy = T [( ) − ( ) ], (5.59)
∂x x+∆x ∂x x

103
where the subscript of the derivative denotes the value for which the deriva-
tive is evaluated.
This net force can also be derived using Newton’s second law.

∂ 2y
Fy = ∆may = µ∆xay = µ∆x , (5.60)
∂t2
where we used that the mass of the segment is ∆m = µ∆x.
Equation (5.59) and equation (5.60) can be equated, and by dividing both
sides by ∆x while considering an ultra-small portion of the rope, the equation
becomes:
∂ 2 y(x, t) ∂ 2 y(x, t)
T = µ . (5.61)
∂x2 ∂t2
As you can see, equation (5.58) and equation (5.61) are identical. Both
equations describe the same wave motion, so we can derive that the wave
speed is
√
T
v= . (5.62)
µ

In general the speed of a mechanical wave is a trade-off between a restoring

force, trying to get the system back in equilibrium (in this case the tension)
and inertia resisting this return to equilibrium (in this case the mass (or mass
density)) of the rope.

104
6 Graphical and numerical analysis of evolu-
tion equations
Often, the (set of) evolution equation(s) that describes a certain physical
system is too complicated to solve analytically. By that, we mean it is not
possible to write down the solution on paper in terms of known functions,
such as x(t) = x0 cos ωt. However, we still can get information from these
equations about the behavior of the system. In this chapter we will discuss
methods to investigate the behavior of evolution equations and their solu-
tions, without actually solving these equations explicitly. To this end, we will
make use of so-called phase diagrams and/or graphical means to interpret
evolution equations, together with numerical methods to construct approx-
imate solutions. Frequently, the information we gain from these methods is
even more valuable than the exact analytical solution to the set of evolution
equations!

6.1 Graphical analysis of evolution equations

Consider first- and second-order evolution equations in their most general
form
dx
= f (x, t), with x(0) = x0 , (6.1)
dt
d2 x
= f (ẋ, x, t), with x(0) = x0 , ẋ(0) = ẋ0 , (6.2)
dt2
These equations describe how x will evolve over time, in other words, they
describe the system’s dynamics. Instead of trying to solve these equations
explicitly for x(t), we are, from a physics point of view, often interested in
a much more practical question, namely: what eventually happens to x as
time evolves? In particular, as illustrated in figure 6.1, does x

(a) tend to infinity, i.e. ∣x∣ → ∞.

(b) tend to some finite limit, i.e. ∣x∣ → c, with c a constant.

(c) evolve into a regular oscillation, such that it keeps repeating in cycles,

(d) do none of these things, but instead fluctuate erratically while remain-
ing bounded.

105
 (a)
x
(b)

x0 (c)

(d)

Figure 6.1: Four different potential outcomes for the long-time evolution of
x(t) for a certain system.

Indeed, we are often more interested in the qualitative behavior of the system
as time evolves than in the exact analytical solution at each point in time. To
identify this qualitative behavior, we can start to think about the differential
equations given by equations (6.1), (6.2) in a different, geometrical way. We
will now discuss how this graphical analysis works for first- and second-order
systems.

6.1.1 First-order systems

First-order systems are often easier to analyze than second-order systems.
Let us therefore start with first-order systems, as given by equation (6.1), to
introduce you to the basic concepts of the graphical analysis of differential
equations.
Recall that the graphical interpretation of the derivative dx dt is that it gives
you the slope of a tangent line to x at a certain point in time. Hence, even
if we do not know x(t) explicitly, equation (6.1) still gives us a recipe for its
rate of change in each point! In the examples below we will show you how to
make use of this recipe to extract information about the behavior of solution
x(t).
First, we will discuss how to use information about the rate of change of x
to construct a direction field of the solution x(t), from which we graphically
reconstruct the solution. Then, we will turn to a more basic, but also more
abstract graphical representation of the behavior of the solution x(t), namely
a phase diagram.

106
Direction fields Consider the first-order system that describes the cooling
of a hot object, or heating of a cold object, as discussed in §3.2, which in
dimensionless form is given by

dT̃
= −T̃ , (6.3)
dt̃
which is an autonomous system as the right-hand side does not explicitly
depend on time.
For the sake of generality, we replace the symbol T̃ by x and drop the tildes
in the discussion that follows, giving
dx
= −x. (6.4)
dt
Let us pretend we do not know the analytical solution to this system (even
though in this case we know it is given by x(t) = x0 exp (−t) of course), but
would like to learn how the hot object cools down by graphical inspection.
We start by drawing the grid (t, x), as is shown in figure 6.2(a). Each grid
point (t, x) then corresponds to a position x that our system occupies at a
time t. We now proceed by drawing in each grid point a little arrow with a
slope dxdt . Following the rule given by equation (6.4), we know exactly what
this slope should be, as for each point (t, x) equation (6.4) gives us the value
of dx
dt . All the arrows together form the vector field or direction field that
corresponds to equation (6.4). Note that for each point in time the direction
field looks exactly the same; i.e. if you draw a horizontal line (constant x,
varying t) in figure 6.2(a), you see that the vectors along the line are identical.
The reason for this similarity is the fact that time does not explicitly appear
in equation (6.4), and therefore the slope of the arrows only depends on the
positions x on the grid.
We can use the direction field to graphically construct the solutions x(t) to
equation (6.4). Each initial condition x(0) = x0 corresponds to a point (0, x0 )
in the direction field. The arrow in a grid point indicates the direction in
which the solution will evolve as time proceeds. If we follow the arrow, we
end up in the next grid point, where again the local arrow indicates how to
get to the next point, and so on. Hence, we can graphically construct the
solution by following the arrows, which point in the direction towards where
x evolves over time.

107
 x x

t t

(a) (b)

Figure 6.2: Direction fields for (a) the first-order autonomous systems given
by equation (6.3) and (b) the first-order non-autonomous system given by
equation (6.5). The lengths of the arrows signifies the magnitude of the ve-
locity ẋ. The solid lines represent the exact analytical solutions to equations
(6.4) and (6.5), respectively, for various initial conditions.

We now extend the analysis to the first-order non-autonomous system

dx
= −x + t, (6.5)
dt
with x(0) = x0 , which is a non-autonomous system as the right-hand side
explicitly depends on time. Similar to the autonomous system, we construct
the slope field using equation (6.5), see figure 6.2(b). Now, as equation (6.5)
is non-autonomous, the time-axis does contain important information: each
time-line (i.e. the vectors along a horizontal constant-x line vary) is different.
By tracing out the paths indicated by the arrows, we can again graphically
construct the solutions to equation (6.5) for any initial condition x0 .

One-dimensional phase diagrams There is a simpler way to graphically

illustrate how x evolves over time than the method described above. As
discussed in the beginning of this chapter, we are often not interested in the
exact value of x for each t, but in to where x evolves eventually. Instead of

108
drawing a vector in each grid point (t, x) and tracing out the solution curves,
it would be sufficient to sketch how x will evolve as time progresses. Such a
sketch of behavior of a solution is the purpose of a phase diagram.
To construct such a diagram, let us start again with the autonomous first-
order system (6.4). In figure 6.3(a) we first plot the rate of change ẋ = f (x)
for each x. The y-axis of this plot now tells you for each value of x what
is the magnitude, and, importantly, the direction of ẋ. In other words, it
tells you in what direction x will evolve over time; if ẋ > 0 the solution will
move towards larger x, hence in the positive x-direction, whereas if ẋ < 0 it
will move towards smaller x, i.e. in the negative x-direction. If ẋ = 0 the
solution will remain stuck at its present location and stay there forever. For
that reason, such a point is called a stationary point.
We mark this qualitative information about the fate of x by arrows on the
x-axis, where arrows pointing to the right imply evolution towards larger
x (i.e. ẋ > 0), and arrows pointing to the left evolution towards smaller x)
(ẋ < 0). The points where ẋ = 0 and the system will remain indefinitely (in
this example that happens in x = 0) are marked with a solid dot. As you
can see, x will always evolve towards this point as time progresses; when the
system is at a position with larger x it will move to the left until it reaches
the stationary point, whereas when it is at smaller x, it will move to the right
until it arrives at the stationary point. We therefore say that the solution
x(t) is attracted to the stationary point. Such a point is called a stable
stationary point or attractor and is marked by a solid dot in figure 6.3.
In figure 6.3(b) we show a similar plot, but now for the basic population
model discussed in §5.6 (where we replaced the symbol N by x)
dx
= x. (6.6)
dt
In this case, x will always move away from the stationary point x = 0, which
is now an unstable stationary point or repeller, marked by an open circle in
figure 6.3(b). Indeed, for x > 0 the population will start to grow and move
away from the stationary point (note that the case x < 0 does not represent
a physical situation when equation (6.6) is used to describe a population
model).
The plots shown in figure 6.3(a) and (b) in fact still contain some redun-
dant information. To illustrate the qualitative behavior of the systems they
describe, we could just show a single line denoting all positions x, where
we mark the stationary points and the behavior of the system around these

109
 (a) (b)
x x

O x O x

(c) x (d) x

Figure 6.3: Plots illustrating the qualitative behavior of (a) the heating or
cooling of an object as given by evolution equation (6.3), and (b) a growing
population as given by equation (6.6). (c) and (d): Phase lines that contain
all qualitative information about the evolution of these first-order autonomous
systems.

points (i.e. if the solution will move towards or away from the stationary
points) with lines and dots, as shown in figures 6.3(c) and (d). For any point
on the line, the arrows indicate in what direction the system will move as
time evolves. Hence, the line contains all qualitative information that is re-
quired to predict how a system evolves over time. Such a line is called a
phase line, and is a one-dimensional example of a phase diagram. Note that
to construct the phase line and decide on the directionality of the arrows and
the position(s) of the stationary point(s), we used the information contained
in on the y-axis of plots in panels (a) and (b).

Phase space For the first-order autonomous systems illustrated in figure

6.3, we could analyze the behavior of the solution at any point in time using
a single piece of information: the position x. We can therefore say that x
defines the state of the system: x is all the information we need to know
to predict the future of the system, given its present. Indeed, if we know
where the system is at a certain point in time, we can predict where it is
going from the evolution equation: one first-order differential equation that

110
describes how the state of the system x evolves in time. Qualitatively, this
information is contained in the phase lines shown in figure 6.3(c) and (d).
These lines, containing all x, span what we call the phase space of the system.
For the first order systems given by equations (6.3) and (6.6) this space is
one-dimensional, and the phase diagram illustrating this space is indeed a
single line.

Two-dimensional phase diagrams For non-autonomous systems, such

as the one given by equation (6.5), we cannot collapse all the information
needed to qualitatively describe the system’s behavior onto a single line.
Indeed, for non-autonomous first-order systems we cannot predict the future
by just knowing the position x, as the magnitude of dx dt also depends explicitly
on time. If we would draw the phase line of the system in a similar way as
the plots in figure 6.3(c) and (d), the arrows and stationary point would
be wiggling around as time evolves. Hence, to predict the future state of
the system we need to know both x and t. Indeed, the state of a first-
order, non-autonomous system is two-dimensional. To better reflect this
two-dimensional nature of the system, we can use a trick (discussed in §3.3)
to recast equation (6.5) into a set of two first-order differential equations, one
per state variable. To this end, we let x1 = x but now introduce x2 = t, such
that the system equivalent to equation (6.5) is given by the following set of
first-order equations
ẋ1 = −2x1 + x2 ,
{ (6.7)
ẋ2 = 1,
where we set x2 (0) = 0 as initial condition to ensure that indeed x2 = t.
The advantage of this way of writing is that equation (6.7) now clearly reflects
the two-dimensional nature of the system, with one evolution equation per
state variable (x1 and x2 ). The change of variables also allows us to visualize
the phase space in a two-dimensional phase diagram (or phase plane) where
all the trajectories appear frozen and arrows no longer wiggle in time; see
figure 6.4(b) (which, in contrast to figure 6.4(a) cannot be reduced to a single
line that contains all information).

6.1.2 Autonomous, second-order systems

We now turn to the geometric interpretation of second-order systems, as
given by equation (6.2). For the sake of simplicity, and to allow for visual

111
 x

(a) (b)

Figure 6.4: Phase diagrams of (a) the autonomous first-order system (6.4)
and (b) the non-autonomous first-order system (6.5). Note that for this sys-
tem the phase diagram is two-dimensional phase diagram and identical to the
direction field of figure 6.2(b).

inspection of solutions on a two-dimensional plane, in the following we restrict

ourselves to autonomous second-order systems, i.e. systems without explicit
time dependence.
As an example, consider the mass-spring system from chapter 5, which is
described by the second-order autonomous evolution equation

ẍ = −ω 2 x. (6.8)

To predict where the mass will move in the next instant, we not only need to
know the position of the mass, but also whether it was, from that position,
extending or compressing the spring, and with what speed it was doing so
(i.e. we need to know its velocity and direction of motion). Indeed, to know
the state of this system at any given time and hence be able to predict its
future, we do not only know where the mass is but also where it is going. In
mathematical terms that means we need to know x and ẋ, which implies that
the phase space of an autonomous second-order system is two-dimensional.
To better reflect this two-dimensional nature of the system, we can rewrite
equation (6.8) into a set of two first-order differential equations, one per state

112
variable. To this end, we let x1 = x, as before, but now we introduce a second
state variable x2 = ẋ, such that the system reads
ẋ1 = x2 ,
{ (6.9)
ẋ2 = −ω 2 x1 .

The reconstruction of the solution x(t) to equation (6.7) by graphical means

via the direction fields (see §6.1.1) directly is not feasible; one would need to
draw a three-dimensional graph (with an x-, a ẋ- and a time-axis) and get
additional information on what locations (x, ẋ) the system can visit as time
evolves. However, we can graphically analyze the solution in the (x, ẋ)-plane
by constructing its phase diagram. We will now look at how to draw the,
inherently two-dimensional, phase diagram (or phase plane) for autonomous
second-order systems.

6.1.3 Phase plane

As for the first-order systems discussed above, we would also like to be able to
construct the phase diagram without solving the evolution equations explic-
itly. To this end, we first rewrite the second-order system into two first-order
equations, one per state variable, such as in the mass-spring example given
by the equations (6.9).
The set of first-order equations (6.9) shows the rate of change of the state
variables x1 and x2 at each point in space. Hence, to each point (x1 , x2 )
we can assign a vector (arrow) (x2 , −ω 2 x1 ) that represents the way these
coordinates will change over time; i.e. in each location (x1 (t), x2 (t)) the
arrow will point in the direction of the new location (x1 (t + ∆t), x2 (t + ∆t))
where the system will move in the next instant (t + ∆t). We can use these
vectors to sketch the phase diagram, as shown in figure 6.5(a). For example,
let us consider the vector field on the x1 -axis. On that axis, x2 = 0, such that
we find, using equation (6.9), that ẋ1 = 0 and the vector field on the x-axis is
given by (0, −ω 2 x1 ). Hence, as illustrated in figure 6.5(a), the vectors point
downwards on the positive x1 -axis, and increase in length (corresponding to
a higher magnitude of the velocity ∣ẋ2 ∣) with increasing x1 . On the negative
x1 -axis, the vectors point upwards, and increase in length with decreasing x1 .
Similarly, the vector field on the x2 -axis is given by (x2 , 0) and points to the
right when x2 > 0 and to the left when x2 < 0. If we fill in some additional
points, we can see that the vectors trace out ellipses in phase space, as shown
in figure 6.5(b).

113
 x2 x2
(a) (b)

O x1 x1

Figure 6.5: Construction of the phase diagram for the mass-spring system
via (a) the vector field given by equation (6.9) or (b) via the calculation of
trajectories as given by equations (6.14) or (6.21) (method A and B in the
main text).

Next to this graphical reconstruction, one can also calculate the phase-space
trajectories exactly, either from the evolution equations (method A) or from
energy conservation (method B) as we will show now.

A. Calculating the trajectories in the phase diagram from the evo-

lution equation Recall that the purpose of the diagram is to show how
the state variables evolve. Hence, in the phase diagram we would like to plot
trajectories x2 (x1 ) to show how x2 evolves as a function of x1 . To plot these
trajectories, we do not need to know the explicit time dependence of x1 (t)
and x2 (t). To find the trajectories x2 (x1 ), we will therefore eliminate time
from the equations. Using the chain rule we write
dx2 dx2 dx1
= , (6.10)
dt dx1 dt
or
dx2
dx2 ẋ2
= dt = . (6.11)
dx1 dx1 ẋ1
dt
After substitution of equation (6.9) we obtain

dx2 ẋ2 ω 2 x1
= =− (6.12)
dx1 ẋ1 x2

114
and therefore
dx2
x2 + ω 2 x1 = 0. (6.13)
dx1
Note that the time t no longer explicitly appears in equation (6.13). Integra-
tion of equation (6.13) over x1 gives (see §3.3):
1 2 1 2 2
x + ω x1 = C, (6.14)
2 2 2
with C an integration constant. This equation describes a set of ellipses in
the phase plane with coordinates x1 and x2 ; see figure 6.5(b). To see why
equation (6.14) represents an ellipse, note that we can recast this expression
into the standard equation for an ellipse centered at the origin
x21 x22
+ = 1, (6.15)
a b
if we set a = 2C and b = 2C/ω 2 . To indicate the time evolution of the system,
we draw arrows in the elliptic trajectories. These arrows are drawn with the
help of equation (6.9): x1 increases in time for x2 > 0 (positive velocities),
and decreases for x2 < 0 (negative velocities).
Similarly, for a system of the form

ẍ = ω 2 x. (6.16)

one could show (see exercise 9.6.7) that the trajectories in the phase plane
are given by
x21 x22
− = 1, (6.17)
a b
which corresponds to a set of hyperbolas.
Figure 6.6 shows that important information about the physical behavior
of the mass-spring system can directly be obtained from the phase diagram
and we do not have to plot (and derive!) x1 (t) and x2 (t) individually. For
example, we can learn from the phase diagram that:
• A periodic motion corresponds to a closed trajectory in the phase dia-
gram. From the plot in figure 6.6(a), we see that we start from point a
at t = 0, where x1 = x0 and x2 = 0 and from there move to point b, in a

115
(a) (b)
x2 x ω
0

d
c a&e c
-x0 x0 x1 x1=0
b d
b
-x0ω a&e

Figure 6.6: (a) Phase plane for the two-dimensional, second-order au-
tonomous mass-spring system given by equation (6.8).(b) Corresponding po-
sitions of the mass, where the arrow indicates the direction of motion (com-
pression or extension of the spring). The labels a−d refer to the position in the
phase plane in panel (a). The positive x1 -direction is defined as downwards,
where extraction of the spring corresponds to positive x1 and compression of
the spring corresponds to negative x1 .

clockwise direction as indicated by the arrow. From b we move to point

c, then point d and then point e, which happily coincides with point
a. Hence, from the phase diagram we immediately see that the mo-
tion repeats itself, corresponding to the mass-spring system oscillating
back and forth (see figure 6.6(b)). For the mass-spring system without
damping, the shape of this curve is an ellipse (with lengths of the half
axes x0 and x0 ω). At a given ω the size of this ellipse is determined by
the initial displacement x0 .

• For any value of x1 , two values of x2 are possible (except for the ex-
trema), which correspond to the compressive and extensile motion of
the spring from any position in space (see figure 6.6(b)). Vice versa,
for every value of x2 only two values of x1 are possible. We can also see
very clearly: a big displacement/stretch ∣x1 ∣ implies a smaller velocity
∣x2 ∣ and vice versa. Indeed, kinetic and potential energy are continu-
ously being converted into one another, and their sum is constant, as
we will see below.

• The extrema x2 = 0 (point a, c and e in figure 6.6) correspond to the

turning points of the motion, where the mass reverses direction and
goes from extensional motion towards compression or vice versa. The
extrema x1 = 0 (point b and d in figure 6.6) correspond to the positions

116
 where the spring is at its rest length and where there is no net force
acting on the mass.

• The origin represents a special point: all other trajectories circle around
it like the eye of a tornado. A mass positioned in this point would not
move, since ẋ1 = ẋ2 = 0, but stay in there forever. Indeed, if we position
the mass such that the spring is at its rest length and do not give it
any initial velocity, it will not move anywhere. Hence, the origin is a
stationary point.

B. Calculating the trajectories in the phase diagram from energy

conservation For conservative systems such as the mass-spring system
(recall §5.3) energy conservation provides an important relation between po-
sition x1 and velocity x2 , namely
1
mx22 + V (x1 ) = E, (6.18)
2
with E the total energy of the system. One can use this relation to construct
trajectories in the phase diagram, as we will now illustrate.
For the linear spring with force F (x) = αx (see equation (5.24)) which motion
is described by equation (6.8), the potential energy is given by
1
V (x1 ) = αx21 , (6.19)
2
and the energy conservation equation reads
1 1
mx22 + αx21 = E. (6.20)
2 2
This result corresponds to the expression of equation (6.14) with C = E/m,
for the trajectories in the (x1 , x2 ) phase plane. Hence, along each trajec-
tory in phase space the total energy E is conserved. Typically, the value of
constant E in equation (6.21) follows from the initial conditions x2 (0) and
x1 (0): E = 12 mx22 (0) + 12 αx21 (0). We can then use the conservation equation
(6.20) to find an expression for the trajectories
√
2
x2 = ± (E − 21 αx21 ). (6.21)
m

117
This method of constructing a phase diagram is very powerful, since it can
also be used when only a rough description of the system’s potential energy
V (x1 ) is known, as will be further discussed in the course Theoretical Clas-
sical Mechanics.

x x x
(a) (b) (c)

x x x

Figure 6.7: Phase diagrams for (a) a damped mass-spring system, (b) a
forced mass-spring system, and (c) a mass-spring system with damped reso-
nant forcing such that it reaches a stable oscillation amplitude.

Let us now compare the phase diagram of a conservative system to that

of a system where energy is not conserved: a damped mass-spring system.
Due to the damping (see §5.4) the total energy of the system decreases over
time. The trajectories in the phase space therefore now no longer form closed
curves (which would indicate a periodic motion) but spiral inwards to the
final state (x1 , x2 ) = (0, 0) due to the loss of energy, see figure 6.7(a). By
contrast, if the mass-spring system is not damped but forced (hence, energy
is put into the system), the trajectories spiral outward, see figure 6.7(b). If
one forces the oscillator but also adds some damping to it, the mass-spring
system may eventually reach a stable loop in which it continues oscillating
with a fixed amplitude, as shown in figure 6.7(c). In summary, we observe
that, since energy is no longer conserved in case of damping or forcing, the
phase diagram no longer consists of closed loops only.

6.1.4 Phase space for higher-dimensional systems

For the autonomous first-order systems discussed above, we saw that the
phase space is one-dimensional; once we knew the value of x, we could iden-
tify where the system would move in the next instant using the evolution
equation for x, e.g., equation (6.3). For non-autonomous first-order sys-
tems we also need to know the time t, which renders the phase space two

118
dimensional, with one evolution equation per state variable, such as equa-
tion (6.7). The phase space of an autonomous second-order system is also
two-dimensional; to know how the system evolves one needs to know both
position and velocity, see equation (6.9). We can generalize this discussion
to evolution equations of N -th order. An autonomous N -th order evolution
equation can be converted into N first-order evolution equations, and has
an N -dimensional phase space. For a non-autonomous system we add one
dimension to the phase space to take care of the time-dependence; and N -th
order non-autonomous evolution equation therefore results in an (N + 1)-
dimensional phase space where the state variables are described by N + 1
first-order evolution equations.
Consider for example the forced mass-spring system described in §5.5 given
by

ẍ + ω 2 x = a cos Ωt. (6.22)

To know the evolution of this system starting from a certain point we need
three numbers: x, ẋ, and t. Indeed, to predict the future, we need to know
where the mass is, how fast it is moving and in what direction, and at what
point in time we are to determine the amount of forcing it gets. The evolution
equations for the three state variables describing this 3D system are given
by
ẋ1 = x2 ,
ẋ2 = −ω 2 x1 + a cos Ωx3 , (6.23)
ẋ3 = 1.

6.2 Numerical analysis of evolution equations

Next to the graphical tools discussed above, we have another means to ana-
lyze the behavior of a set of evolution equations without solving them exactly:
numerical analysis on the computer. Numerical integration is a huge subject,
but we restrict ourselves here to the very basics, to give you a glimpse of how
the method works. We will focus on the numerical integration of first-order
equations as given by

ẋ = f (x, t), with x(0) = x0 . (6.24)

Consider again the vector plot of figure 6.2 for the system given by equation
(6.4). If we locate ourselves in a certain initial value x0 and follow the arrow

119
starting in that point, we could trace out the solution curve. Numerical
integration on the computer works in a similar way. Suppose we start at
t = t0 in a point x0 . If we follow the arrow, some time ∆t later we will have
moved by an amount f (x0 , t0 )∆t, because distance = rate × time. Hence the
new position x(t0 + ∆t) ≈ x0 + f (x0 , t0 )∆t. Of course that is not the exact
position but only an approximate, since our velocity will change a little bit
during the time step. However, if we choose ∆t small enough, that error will
be small. Let us call this approximate position x1 = x0 +f (x0 , t0 )∆t. Likewise,
after the second time step, we will be in a position x2 = x1 + f (x1 , t1 )∆t. We
can generalize this method by writing that after n + 1 time steps, we are in
position

xn+1 = xn + f (xn , tn )∆t. (6.25)

This rule is the simplest numerical integration rule we know and termed
Euler’s explicit method. Let us see how this works out for a first-order au-
tonomous system such as given by equation (6.3). In the exercises you will see
how this method is applied to first-order non-autonomous, and second-order
systems.

Example: first-order autonomous system Consider the first-order au-

tonomous system
dx
= −cx. (6.26)
dt
Numerical integration of equation (6.26) gives us the following integration
rule

xn+1 = xn − cxn ∆t. (6.27)

Let us take c = 2, use x0 = x(t = 0) = 1 as starting point and apply this rule
to find the future positions using a time step of ∆t = 0.2, as shown in table
6.1. A continuation of this manually constructed table can be found in the
Excel sheet example62.xls9 . The first row of this sheet contains the initial
conditions, similar to the first row of table 6.1, each following row contains
the result of the application of integration rule (6.27) to the previous row.
On Canvas you will find more Excel sheets where other, more complicated
systems are integrated numerically.

120
Table 6.1: Numerical integration of equation (6.26) compared to its exact
solution for c = 2.
x t numerical solution exact solution

x0 0 1 1

x1 0.2 1 − 2 × 1 × 0.2 = 0.60 e−0.4 ≈ 0.67

x2 0.4 0.60 − 2 × 0.60 × 0.2 = 0.36 e−0.8 ≈ 0.45

x3 0.6 0.36 − 2 × 0.36 × 0.2 = 0.22 e−1.2 ≈ 0.30

x4 0.8 0.22 − 2 × 0.22 × 0.2 = 0.13 e−1.6 ≈ 0.20

x5 1.0 0.13 − 2 × 0.13 × 0.2 = 0.078 e−2 ≈ 0.14

The exact and numerical solution are also plotted in figure 6.8. The top
line shows the exact solution and the black squares the approximate values
obtained from Euler’s method. As you can see, the numerical approxima-
tion quickly becomes inaccurate. We could improve the approximation by
decreasing ∆t to a very small number. However, decreasing ∆t comes at
a cost. Taking smaller (and hence more) time steps requires an increasing
amount of computational power. However, with the current powerful com-
puters this is not the most important issue. The main problem is that each
time step carries a round-off error, as computers do not have an infinite ac-
curacy. For example, to manually calculate the approximate values of the
exact solution depicted in table 6.1 we rounded them off to two significant
digits. Computers do something similar. As these round-off errors occur in
each time step, they rapidly accumulate as the number of time steps taken
increases.

More accurate integration schemes The main reason Euler’s explicit

method is so inaccurate is that it approximates the derivative only at the
9
This file can be found on Canvas.

121
 x 1.0

0.8

0.6 Exact solution

0.4 Numerical approximation

0.2

0.0
0.0 0.2 0.4 0.6 0.8 1.0

Figure 6.8: Exact solution (solid line) and numerical approximation (black
squares) to equation (6.26) with c = 2, x0 = 1 and ∆t = 0.2.

left end of the time interval tn to tn+1 . Other integration schemes, such as
the improved Euler method or the fourth-order Runge-Kutta method have a
better and more accurate way to approximate the derivatives. A discussion of
these methods is beyond the scope of our course, but we mention the names
of these methods here such that you are aware of their existence. Euler’s
explicit method as discussed above, however, does contain the essence of
numerical integration and is the most basic scheme, which is why we focus
on that method here.

122
7 Analysis of stationary points and their sta-
bility
So far we have used graphical and numerical methods to inspect the behavior
of a system qualitatively. Often, we are also interested in some quantitative
information about the system. In particular, we would like to know whether
there are any special points in the phase space where the system could remain
stuck forever, and how quickly the system will evolve towards these points
as time progresses. These special points are called stationary points. We will
now discuss how to find these points and how to examine the behavior of
the solution around them. Again, this information will be obtained without
explicitly solving the evolution equations on their entire domain!

7.1 Stationary points and how to find them

A stationary point (sometimes also called an equilibrium point) is defined
as a point where the system remains stationary, i.e. motionless. Once the
system has reached such a point, it will never be able to move out and it
will stay there forever. In the analysis of the long-term behavior of evolution
equations, stationary points are therefore very important to identify. For
autonomous first-order systems

ẋ = f (x) (7.1)

the stationary points xe are given by

f (xe ) = 0, (7.2)

such that in these points the velocity ẋ is zero. As a consequence, once the
system is in such a point, it cannot leave. Verify for yourself that for the
system (6.3) the stationary point is found in T̄e = 0. At this point, the object
has the same temperature as its surroundings and hence it will not heat up
or cool down.
As discussed in chapter 6, an autonomous second-order system, such as equa-
tion (6.8) can always be written into two first-order systems. Hence, the most
general form of an autonomous second-order system is

ẋ1 = f (x1 , x2 ),
{ (7.3)
ẋ2 = g(x1 , x2 ).

123
The stationary points x1,e , x2,e of such a system are given by

f (x1,e , x2,e ) = g(x1,e , x2,e ) = 0. (7.4)

Note that, for a second-order system to be in its stationary point, the right-
hand side of both equations (7.3) needs to be equal to zero simultaneously,
such that the system has a zero velocity and a zero acceleration. Likewise,
for N -th order systems the stationary points can be found by equating the
right-hand sides of the set of first-order evolution equations for the state
variables to zero simultaneously.

Example: stationary points of a mass-spring system Consider the

mass-spring system given by second-order equation (6.8), which can be rewrit-
ten as a set of two-first order equations

ẋ1 = x2 ,
{ (7.5)
ẋ2 = −ω 2 x1 .

We are asked to find its stationary points. To this end, we will follow the
ISEE method.

I - Identify

The stationary points of equation (7.5) are the points where both the posi-
tion x1 and the velocity x2 of the mass do not change over time (i.e. the mass
has no velocity and no acceleration).

S - Set up
If x1 and x2 are not changing, then it must hold that ẋ1 = x˙2 = 0.

E - Execute

Using ẋ1 = 0 we find, upon equating the right-hand side of the first equation
of the set (7.5) to zero, x2,e = 0. Using ẋ2 = 0 we find, upon equating the
right-hand side of the second equation of the set (7.5) to zero, x1,e = 0. Hence,
the stationary point of equation (6.8) is given by (x1,e , x2,e ) = (0, 0)

124
E - Evaluate

Physically, the point (0,0) in phase space corresponds to the spring being at
its rest length and the mass having no velocity. Indeed, in this point there is
no force acting on the mass to accelerate it, while at the same time the mass
has no velocity. Hence, the point (0,0) indeed corresponds to a stationary
point.
By contrast, if only one of the two conditions were fulfilled, the mass would
not remain stationary. For example, if only the velocity ẋ1 = 0, but the mass
would still be in a position where the spring is extended or compressed (such
as in the extrema at locations a and c in figure 6.6(b)), the mass would still
have an acceleration. As a consequence, the mass would not remain station-
ary but simply pass through the point where ẋ1 = 0 and move to the next
position. On the other hand, if only the acceleration ẋ2 = 0 but the mass
would still have a finite velocity, it would simply pass through the point and
move to a new position. This situation would correspond to location b and
d in figure 6.6(b), where the mass passes through these points on its way
up or down. Only if we would position the mass in these locations without
giving it any velocity, it would remain stationary, showing that we indeed
need x1 = x2 = 0 in the stationary point.

7.2 Analysis of the stability of stationary points

As we have seen, a system placed in a stationary point will stay there indef-
initely. In practice, we are interested in knowing whether a system is able
to reach these stationary points, and if so, how much time that will take.
To this end, we are going to analyze the stability of the stationary points.
A stationary point is defined to be stable if small disturbances of the sys-
tem away from the stationary point will decay in time, or at least remain
small, causing the system to remain in or close to the stationary point. A
stationary point is unstable if any small disturbance will grow into a large
one, causing the system to move away from the stationary point. Later on,
we will make these definitions about stable and unstable stationary points
a bit more precise, but let us first develop some physical intuition for these
concepts.

125
7.2.1 Stable and unstable behavior
A classic example to illustrate the (in)stability of stationary states is that of
a ball in a hilly landscape. Consider the landscape shown in figure 7.1. The
evolution of the position of the ball in this landscape is of course described
by Newton’s second law. From figure 7.1, we observe that there are two
stationary points or equilibrium positions for the ball: in the bottom of the
valley (position A) or exactly on top of the hill (position B). If the ball is
placed in one of these positions at zero initial velocity, it will stay there
indefinitely as no force is acting on the ball.

B g

Figure 7.1: Ball on an hilly landscape in its (A) stable and (B) unstable
equilibrium states.

However, of course there is a clear difference between the two positions: if

the ball is slightly disturbed, either by pushing it to a position just outside
the stationary point, or by giving it a tiny initial velocity, the behavior of
the ball will differ dramatically between point A and B. Indeed, when small
disturbances are imposed these will barely have an effect on the ball in posi-
tion A, but in position B the smallest disturbance will cause the ball to roll
away. We therefore say that point A is a stable equilibrium point, whereas
point B is unstable. This difference has important consequences in practice,
where small disturbances (think of some wind, or tiny vibrations) are always
present: as time evolves, the ball will eventually end up in the stable equi-
librium point A, while it will never be found in point B. Even if you would
be able to place the ball with infinite precision exactly in point B, the ball
will quickly roll off as soon as it experiences the tiniest disturbance.
This reasoning might seem elementary and trivial, but in fact we can use the
same approach to analyze the stability of equilibrium states of more com-
plex systems. The dynamic behavior of complex systems often cannot be
described in a simple manner. In these situations it is useful to first map

126
all possible stationary states, and then analyze which of these are stable and
which are not. Below we will discuss two approaches to perform such a stabil-
ity analysis: qualitatively, via a graphical analysis in the phase diagram, and
quantitatively, via a so-called linear stability analysis. The former method
is often quick and intuitive to apply, whereas the latter is more robust and
has the additional feature that it provides us with an estimate of how much
time the system will take to reach a certain stationary point, or how quickly
it will run away.
Both approaches to stability analysis can be described by a basic ‘recipe’.
For clarity we first summarize the steps of each recipe, and then show you
how to apply these recipes by discussing two physical examples in detail; a
first- and a second-order system.

Recipe: graphical stability analysis

1. Determine the stationary points of the system.

2. Draw the phase diagram and mark the stationary points.

3. Draw arrows indicating how the system will move in between the sta-
tionary points.

4. From the directionality of the arrows, identify for each stationary point
whether it is

• stable: arrows are directed towards the stationary point, and the
solutions are hence attracted to it. We denote such a point by a
in the phase diagram.
• unstable (arrows are directed away from the point, and solutions
are repelled). We denote such a point by a # in the phase diagram.
• neutrally stable as solutions are neither attracted to nor repelled
from the point. We denote such a point by a in the phase
diagram.
• half-stable: arrows are directed towards the stationary point and
the solutions are attracted from one side, but repelled away from
the stationary point on the other side. We denote such a point by
a#G (if the point is stable when coming from the left) or #
H (if the

127
 point is stable when coming from the right)in the phase diagram.
In this course, we will not focus on this hybrid situation, but we
mention it here for the sake of completeness.

Recipe: linear stability analysis

1. Determine the stationary points xe of the system.

2. Analyze the behavior of the solution to the evolution equation close to

the stationary points. To this end, consider a small perturbation δx ≪ 1
around xe , where δx is a function of time (i.e. the perturbation can grow
or decay over time!). If we fill in the point xe + δx into the evolution
equation we arrive at an evolution equation for the perturbation δx(t).
Linearize the resulting equations by using the fact that δx is very small.

3. Substitute in the values for xe in the linearized evolution equation for

δx(t) and solve for δx(t).

4. Analyze your result: if δx grows over time it means that the solution
is unstable to small perturbations around xe , as it will move away.
If δx decays over time it means that the solution is stable to small
perturbations. If δx neither grows nor decays, the solution is neutrally
stable to small perturbations.10

7.2.2 First-order system: Simple population-growth model

A basic model that describes the growth of a population of fish in a pond is
given by
N
Ṅ = rN (1 − ), (7.6)
k
10
For those who are interested: in half-stable points you find that if δx > 0 the pertur-
bation decays with time and if δx < 0 it grows, or vice versa.

128
where N (t) is the population of fish at time t, and r and k are positive
constants describing the growth rate and the so-called carrying capacity, re-
spectively. For small N ≪ k, the model predicts that the growth rate of
the fish population is proportional to r. However, for populations larger
than the carrying capacity k, the growth rate actually becomes negative and
the population will decrease, modelling the effect of overcrowding (lack of
food and/or space) in the lake. We are asked to (a) determine the station-
ary point(s) of this model system and analyze their stability using (b1) the
graphical method and (b2) a linear stability analysis. To this end, we will
follow the ISEE method.

(a) Stationary points

I - Identify

The stationary points of equation (7.6) are the points where the number of
fish N does not change over time.

S - Set up
If N is not changing, then it must hold that Ṅ = 0.

E - Execute

Using Ṅ = 0 gives
Ne
rNe (1 − ) = 0, (7.7)
k
from which we find two stationary points, namely Ne,1 = 0 and Ne,2 = k.

E - Evaluate

Physically, the two stationary points make sense: Ne,1 corresponds to the
situation where there are no fish in the lake, and hence no growth can occur.
Point Ne,2 corresponds to the carrying capacity, meaning that if there are

129
less fish the population will grow, but if there are more the population will
decay due to overcrowding.

(b) Stability analysis

I - Identify

We need the stationary points from (a) to analyze their stability. There are
two approaches to analyze the stability: (b1) the graphical method and (b2)
a linear stability analysis.
(b1) Graphical stability analysis

S - Set up

The system is given by equation (7.6). The phase diagram of this first-order
system will be a one-dimensional phase line.

E - Execute (following the recipe for graphical stability analysis)

2. To construct the phase line we first plot Ṅ as a function of N . The

x-axis of this plot will form the eventual phase line. We use the infor-
mation on the y-axis to work out the directions of the arrows around
the stationary points on the phase line (x-axis). This procedure is sim-
ilar to the what we used to draw the phase lines of system (6.4) in
figure 6.3.
The plot of Ṅ as a function of N is shown in figure 7.2. The N -axis
forms the phase line, where we marked the two stationary points (the
zero-crossings on the N -axis, where Ṅ = 0).

3. The direction of the arrows can be found from the sign of Ṅ : for Ṅ > 0
the arrow points towards the right, to larger N , whereas for Ṅ < 0 it
points towards the left, to smaller N .

130
 N

O k/2 k N

Figure 7.2: Plot of equation (7.6), where the x-axis corresponds to the phase
diagram (phase line) of the population model, and the y-axis is used to derive
the direction of the arrows and find the stationary points depicted on the
phase line.

4. From the direction of the arrows on the phase line we immediately

see that the left stationary point Ne,1 = 0 is unstable: as soon as the
system moves slightly to the right it will continue growing and will
never return. We can also see that Ne,2 = k is stable: the arrows point
towards this stationary point, if the system moves slightly to the left
or right it will move back to the stationary point at N = k.

(b2) Linear stability analysis

S - Set up

We will consider a small perturbation δN around the stationary points Ne .

E - Execute (following the recipe for linear stability analysis)

2. Substitute N (t) = Ne +δN (t), with δN ≪ 1, into the evolution equation

(7.6) to find
d (Ne + δN ) Ne + δN
= r (Ne + δN ) (1 − ), (7.8)
dt k

131
 which we can work out a step further by realizing that Ne is just a num-
ber, and hence does not depend on time, and expanding the brackets
on the right-hand side, to find
2 2
˙ = r (Ne − Ne − Ne δN + δN − Ne δN + δN ) ,
δN (7.9)
k k k k
which is an evolution equation for the perturbation δN (t). We can
reorganize this equation slightly and linearize it into

˙ = rNe (1 − Ne ) + rδN (1 − 2 Ne ) ,
δN (7.10)
k k
where we have dropped the term proportional to δN 2 , since for small
δN ≪ 1 this quadratic term will be even smaller and can therefore safely
be neglected. Note that by dropping the quadratic term, equation
(7.10) is now linear in δN . Upon close inspection of equation (7.10) we
see that we can further simplify the right-hand side: the first term on
the right-hand side will be zero by definition, since Ne is a stationary
point. The resulting linear evolution equation for small perturbations
δN around the stationary points Ne hence becomes

˙ = rδN (1 − 2 Ne ) .
δN (7.11)
k
Note that this result is still general and holds for all stationary points
Ne .11
3. We now solve equation (7.11) for each stationary point, by substituting
the specific values for Ne into the equation. In Ne,1 = 0 we find
˙ = rδN,
δN (7.12)
which has solution
δN = c0 exp (rt), (7.13)
with c0 an initial condition. In the stationary point Ne,2 = k equation
(7.11) reduces to
˙ = −rδN,
δN (7.14)
11
We deliberately keep the result general here, to avoid having to derive two evolution
equations, one per stationary point.

132
 with solution
δN = c0 exp (−rt), (7.15)
again with c0 an initial condition.
4. According to solution (7.13), any perturbation in the number of fish
above Ne = 0 will grow exponentially over time. Hence, we classify
this equilibrium point as unstable. By contrast, perturbations around
Ne = k decay exponentially, according to solution (7.15), rendering that
equilibrium point stable.

E - Evaluate

Physically, the stability of the two stationary points makes sense: (point
Ne,1 ) as soon as there are a few fish, they will start reproducing, causing the
population to grow, rendering this stationary point unstable; (point Ne,2 ) if
there are less fish than the carrying capacity of the lake the population will
grow, but if there are more the population will decay due to overcrowding,
so this stationary point is stable to small perturbations.
The conclusions from the linear stability analyses are the same as the con-
clusions from the graphical analysis, as they should. However, from solutions
(7.13, 7.15) we additionally get quantitative information about how quickly
the population will grow or decay away from or towards its stationary points.

7.2.3 Second-order system: the non-linear pendulum

Consider the second-order evolution equation describing the motion of a pen-
dulum that we derived in §5.2
θ̈ = −ω 2 sin θ. (7.16)
We will analyze the stability of the equilibrium point(s) for the pendulum
using linear stability analysis. In the exercises you will be asked to analyze
the stability of these points graphically.

I - Identify

We have to find the stationary points of equation (7.16) and identify their
stability using a linear stability analysis.

133
S - Set up
Equation (7.16) is a non-linear, second-order autonomous equation. The sta-
tionary points are given by θ̈ = θ̇ = 0. Indeed, for the mass not to move and
hence be in the stationary point, both the velocity and the acceleration of
the mass need to be zero.

E - Execute

1. To find the stationary points we need to solve θ̈ = sin θ = 0, which gives

at θe,1 = 0, and θe,2 = π. In these points the acceleration of the mass is
zero. For the mass to be in the stationary point, it also needs to have a
zero velocity. We fulfill that condition by setting θ̇ = 0 simultaneously;
i.e. if we place the mass in the point θe,1 or θe,2 with zero initial velocity,
the mass is in a stationary point and will hence remain there forever.

2. We consider a small deviation around the stationary points θ = θe + δθ,

with δθ ≪ 1 and substitute that into equation (7.16) to find
¨ = −ω 2 sin (θe + δθ).
δθ (7.17)

Making use of the fact that δθ ≪ 1 we can linearize equation (7.17)

around the point θ = θe using a Taylor expansion. In §5.2 we discussed
how to expand a function f (x) around x = 0. Importantly, here we
need to expand around an arbitrary, nonzero point.

Some mathematics: Taylor series around an arbitrary point x

The Taylor series of equation (5.6) (see §5.2) for a function y = f (x)
around the specific point x = 0 can be generalized to a Taylor series of
this function around an arbitrary point x = a as follows

N
1 (n)
f (x) ≈ ∑ f (a) ⋅ (x − a)n . (7.18)
n=0 n!

134
With this expression we can approximate the given function y = f (x)
locally, i.e. around the point x = a, for ∣x − a∣ ≪ 1. The linearization
of y = f (x) around x = a is then given by a straight line through the
point (a, f (a)) with a slope f ′ (a); i.e.

f (x) ≈ f (a) + f ′ (a)(x − a). (7.19)

The expression (7.18) gives the general expression for a Taylor expan-
sion around any point a in space. However, people often get confused
and mix up this expression with a Taylor series around x = 0. To avoid
mistakes, some people find it easier to perform a coordinate transfor-
mation that moves the point x = a to the origin before performing the
Taylor expansion. The coordinate transformation involving a trans-
lation of the old coordinates x to the new coordinates x′ is given by

x′ = x − a (7.20)

such that the new coordinate x′ is equal to zero when the old coordi-
nate x is equal to a. The Taylor series of the function f (x′) around
x′ = 0 is now equal to the Taylor series of function f (x) around x = a,
where x′ = x − a; see exercise 9.7.1.

We now use the Taylor series (7.18) to expand sin θ around the point
θ = θe (which is non-zero for θe,2 !). We deliberately do not yet fill in
the values we found for θe but keep the expansion general, to find

sin θ ≈ sin θe + cos θe δθ (7.21)

(see exercise 9.7.2), and hence

¨ = −ω 2 (sin θe + cos θe δθ) .
δθ (7.22)

We stress once more that this result is valid around arbitrary points
θe . The reason for keeping the result general and not filling in the
stationary points directly is that we can now use our Taylor expansion
(7.22) in both stationary points, which saves some time (compared to
writing to expansions, one for each stationary point).

135
 3. We now solve equation (7.22) for each stationary point. In θe,1 = 0 we
find
¨ = −ω 2 δθ,
δθ (7.23)
with solution
δθ = c1 cos ωt + c2 sin ωt, (7.24)
with c1 and c2 constants that are determined by some initial conditions.
In θe,2 = π we find
¨ = +ω 2 δθ,
δθ (7.25)
with solution
δθ = c1 exp ωt + c2 exp −ωt, (7.26)
with c1 and c2 again constants that are determined by initial conditions.
4. From equation (7.24) we observe that around θe,1 = 0 perturbations
neither grow nor decay but will keep oscillating. We therefore call the
point θe,1 = 0 neutrally stable, or elliptic, as we will see in exercise
9.7.5. By contrast, the perturbations around θe,2 (see equation (7.26))
contain an exponential growth term that will cause the perturbations
to grow without bound, rendering this stationary point unstable. Such
a stationary point is also called a hyperbolic point, somewhat related to
the shape of the trajectories in phase space around such a point (see
exercise 9.7.5).

E - Evaluate

The two stationary points correspond to the positions where the pendulum
hangs straight down (θe,1 = 0) or points straight up (θe,1 = π), and does not
have any velocity (such that θ̇ = 0 and the pendulum cannot simply pass
through the points θe ). The first stationary point was found to be neutrally
stable. The physical interpretation of this behavior is that, once perturbed
by giving it a small displacement or velocity, the pendulum will keep swinging
back and forth around its vertically downward pointing position at a small,
but constant, amplitude. In practice, there will always be some damping (for
example in the pivot point) causing the oscillation amplitude to gradually
decay, in line with our experience. The second stationary point (pendulum
pointing straight up), is unstable, as we know from experience.

136
8 Non-linear dynamics: from order to chaos
So far, we have mainly been concerned with linear systems. We only had
a glimpse of the behavior of non-linear systems such as the Lotka-Volterra
model in chapter 5. In this chapter we will see how the non-linearity can
give rise to complicated dynamic behavior: bifurcations, catastrophe, hys-
teresis, limit cycles, and chaos. We will show how these phenomena arise
in physical systems by giving a number of examples, such as the Van der
Pol equations describing the response of an electric circuit, and the Lorenz
equations that provide a basic model for atmospheric circulation. Note that
this final chapter is more conceptual than the previous chapters. You do not
need to memorize all details of the examples that are given. Instead, the fo-
cus should lie on understanding the concepts as well as on how to practically
construct a bifurcation diagram. To help you, the important concepts are
highlighted in italics, and summarized at the end of the chapter.

8.1 The trouble with non-linear systems

Most non-linear systems are impossible to solve analytically, and one has to
resort to the methods discussed in chapters 6 and 7 to gain some insight into
their behavior. Typical non-linear terms in the evolution equations describing
these systems are higher powers (2 and beyond), products, or functions of
the state variables xi . Examples are the product term N1 N2 in the Lotka-
Volterra model (5.51,5.52) or the function sin θ in the non-linear pendulum
equation (7.16).
But why are non-linear systems so much harder than linear ones? The key
difference is that linear systems can be broken down into parts, where each
part can be solved separately and then recombined to get the response of
the entire system. The fact that a linear system is exactly equal to the sum
of its parts allows for powerful simplifications of complex problems, and the
application of mathematical techniques such as superposition, Laplace and
Fourier transforms, which you will learn all about in later courses. Non-linear
systems are so nasty because they do not allow for these simplifications. In a
non-linear system, the separate parts interact, cooperate or compete, giving
rise to unexpected outcomes. Many physical phenomena in nature are in
fact non-linear, from changes in our climate, epileptic seizures, to turbulent
motion in fluid flow. So we better prepare ourselves how to handle non-
linearity. . . .

137
8.2 Bifurcations
As we saw in chapter 7, to observe a certain state of a system in nature, this
state should not only obey the evolution equations describing the system,
but it should also be stable against small perturbations. However, for many
non-linear systems, stationary points and their stability are not fixed. As
we will see, stationary points can be created or destroyed, become stable or
lose their stability depending on the value(s) of certain system parameter(s).
These major qualitative changes in the system’s dynamics (i.e. its temporal
evolution) upon variation of a system parameter are called bifurcations.

A classical example of a system that undergoes a bifurcation is the so-called

‘elastica’ by Euler (1744): the buckling of an elastic strut that is pressed
on by a force F (see figure 8.1). As long as the force on the strut remains

Figure 8.1: When the longitudinal force F on an elastic strut is increased

beyond a critical force Fc , the strut undergoes a bifurcation: the (a) straight
state is no longer stable and the strut will (b) bend to the left or (c) to the
right.

smaller than a certain critical value Fc the strut remains straight, and small
disturbances will not change this state. Hence, the straight state is stable.
However, when F is increased gradually, the ‘straight’ state will, at F > Fc ,
suddenly become unstable, causing the strut to buckle into one of the bent
states shown in figure 8.1(b,c). This qualitative change in the equilibrium
state of the system upon the change of control parameter F is called a bifur-
cation. Euler showed that the bifurcation point Fc is given by
π2
Fc = B , (8.1)
L2
138
where L is the strut length and B the rigidity. The parameter B is fully
determined by the material properties and the shape and size of the strut’s
cross section.
To quantitatively show the response of the strut to an increase in the com-
pressing force, we plot the excursion amplitude xa of the strut as a function
of F ; see figure 8.2. This amplitude remains zero as long as F < Fc , but starts
to increase gradually with increasing force F > Fc ; xa will become positive
when the strut bends to the right, and negative when it bends to the left,
see figure 8.2(b). This type of graphical representation of the behavior of the
system is called a bifurcation diagram.
Strut bending to the right
(a) (b) x (c)
F a
d

Fc F
xa 0
F

xa
x
Strut bending to the left

Figure 8.2: (a) When a force F > Fc is exerted on the strut its length shortens
by an amount d. The excursion amplitude of the strut is xa . (b) Bifurcation
diagram of the excursion amplitude xa of the strut versus the applied force
F . (c) Illustration of the bifurcation of panel (b) in the ball-on-hill analogy.

Recipe: bifurcation diagrams

A bifurcation diagram is used to graphically illustrate the appearance and
disappearance of new equilibrium states of a system and the (changes in)
stability of these states as a system parameter is varied. On the x-axis of
such a diagram one typically plots the system parameter that is varied, and
on the y-axis the equilibrium states. The line style is used to mark the
stability of the equilibrium states: a solid line means a stable equilibrium, a
dashed line unstable. Consider for example the first-order system

ẋ = r + x2 , (8.2)

139
where r is the system parameter, which can be positive, negative or equal to
zero. To plot the bifurcation diagram, we use the following steps:

1. Calculate the stationary states of the system as a function of the system

parameter. For the system given by equation (8.2), the stationary
states are given by ẋ = 0, and hence r + x2e = 0. Hence the number
√ value of r; for r < 0 we have two
of stationary states depends on the
stationary states, namely xe = ± r, for r = 0 we have one stationary
state xe = 0, and for r > 0 we have no stationary states.

2. Analyze the stability of these stationary states as a function of the sys-

tem parameter. For the example system (8.2) you will do this analysis
graphically in exercise 9.8.1. The resulting phase lines are shown in
figure 8.3.

(a) r<0 (b)

r=0 (c) r>0
x x x

Figure 8.3: Phase lines corresponding to the system described by equation

(8.2) for three different values of system parameter r.

3. Plot the stationary states as a function of the system parameter. Use

the line style (solid for stable, dashed for unstable) to mark the stability
of the stationary states. For the system described by equation (8.2) this
step implies that we plot xe as a function of r. The final plot is shown
in figure 8.4.

unstable x

stable

Figure 8.4: Bifurcation diagram of the system given by equation (8.2).

140
From the bifurcation diagram of the strut show in figure 8.2(b) it is clear that
the system’s ‘equilibrium state’12 xa bifurcates at F = Fc into two possible
solutions: the upper branch corresponds to excursions xa > 0, whereas the
lower branch describes excursions xa < 0. Each of these branches is stable
to small perturbations. The dotted line (xa = 0, for F > Fc ) describes an
unstable solution branch: a small disturbance would push the system out
of the equilibrium xa = 0, the system would jump to one of the two stable
branches (xa > 0 or xa < 0) pretty much instantaneously. Note that the
symmetry of the physical system (equal chance of the strut buckling to the
left or to the right), is reflected in the symmetry of the bifurcation diagram
around the F -axis.
The bifurcation behavior shown in figure 8.2(b) has a nice analog in the
ball-on-a-hill configuration that we discussed in the context of stability of
equilibrium states in chapter 7. As shown in figure 8.2(c), one can think of a
ball placed in a valley located at xa = 0 for F < Fc ; in that position the ball is
stable to small perturbations around this position, and will oscillate around
xa = 0 when perturbed. At F = Fc the valley splits into two valleys with a
hill in between. For F > Fc the equilibrium state xa = 0 is now on top of
the hill, and therefore unstable, while there are two new stable equilibrium
states at xa > 0 and xa < 0.
Bifurcations occur in many physical systems. Prominent examples are the
transition from laminar to turbulent flow upon increase of the flow rate in
the field of fluid physics, or the onset of stimulated laser emission in optics.

8.3 Catastrophe & hysteresis

In the situations discussed so far a smooth change of a system parameter
resulted in a smooth change in the state of the system. For example, a
gradual change in the force on the strut beyond Fc resulted in the gradual
transition of the system into one of the buckled states. However, in many
physical systems it can happen that a gradual change of a system parameter
results in a sudden change of the system state: in that case we speak of a
catastrophe.
12
We use the term ‘equilibrium state’ loosely in this example as we did not provide the
full mathematical description of the strut; xa is merely used as a representation of the
equilibrium state of the system.

141
An example of catastrophic behavior we see in a vertically set up, thin elastic
metal wire (e.g. brake cable of a bike) that is being clamped from below (see
figure 8.5).

LS g
LC

Clamping points

(a) (b) (c) (d) (e)

Figure 8.5: Catastrophe and hysteresis behavior of a clamped vertical elastic

wire. The clamping points are represented by the black squares. (a) If the
top clamping point is lowered by a length L < Ls the wire is stable to any
perturbation and will always remain in upright position. (b) If Ls < L < Lc
the wire will be stable to smaller perturbations, but collapse for larger ones.
(c) For L > Lc a catastrophic jump occurs upon the slightest perturbation:
the wire collapses. (d) Raising the top clamping point again to Ls < L < Lc
shows that this system displays hysteresis: the raise does not cause the wire
to jump back into its original upright position, for that only occurs when (e)
L < Ls .

At a distance L below the free end the wire is clamped at a second point. At
short lengths L the wire will remain straight: its mechanical stiffness makes
sure the wire is stable in this state and will move back to this state when it
gets perturbed; hence, the wire is stable in its upright position. Now we are
going to increase the length L gradually. Until a certain critical length Ls the
wire will jump back to the vertical state, even when we apply big disturbances
to this equilibrium state (see figure 8.5(a)). When L is increased gradually
even further (> Ls ) we can keep the wire stable in the vertical position until
a second critical length L = Lc , but only when the disturbances are very
small (see figure 8.5(b)). When the free wire length L > Lc even the slightest
disturbance will result in a sudden collapse of the wire (see figure 8.5(c)). So
for L > Lc a small change in the system parameter L results in an abrupt
change in the state of the system: a catastrophe. The term catastrophe is
motivated by the fact that a minor change in the system parameter has major
consequences. You can imagine that if such a jump occurs in the equilibrium

142
state of e.g. a bridge, it can have catastrophic consequences!
Now, from the state L > Lc (see figure 8.5(c)), we are going to decrease the
wire length L gradually by slowly shifting the top clamp upwards. It turns
out the wire does not immediately return to the vertical state at L < Lc
(see figure 8.5(d)): upon small disturbances the wire remains hanging. Only
when the wire is shortened to L < Ls it jumps from the collapsed state back
to the vertical state at the slightest disturbance (see figure 8.5(e)). This irre-
versibility in behavior (the way down is different from the way up) is called
hysteresis.

The behavior of this mechanical system is represented graphically in the

bifurcation diagram of figure 8.6. In this diagram a represents a measure for

a
L

0
LS LC L

a
(a) (b)

Figure 8.6: (a) The state of the clamped wire is measured by the deviation
a of the tip from its vertically upward position. The distance between the
tip in upright position and the top clamping point is L. (b) Bifurcation dia-
gram showing a as function of system parameter L. For L > Lc the system
will undergo a catastrophic jump from the vertical state to one of the stable
collapsed states.

the deviation of the wire tip from the vertical state (a = 0) as function of
distance L between the tip position in vertical state and the top clamping
point.
This bifurcation behavior stands in a striking contrast to the bifurcation of
the buckling strut (figure 8.1) as it allows for sudden jumps in the system
state and irreversibility in the system’s behavior upon a change of parameters.
To see how these catastrophic jumps and hysteretic behavior appear, we are

143
going to analyze the behavior of the system using figure 8.7. The initial state

a R

0
P Q L

R’

Figure 8.7: Illustration of the catastrophic jumps and hysteresis loop that can
occur for a system with a bifurcation diagram similar to the one in figure 8.6.

of the wire a = 0 with L < Ls (see figure 8.5(a)) is described by a point on the
horizontal axis between O and P . Upon increase of L beyond the first critical
value Ls this equilibrium state is still stable for small disturbances, and the
system state is described by a point on the horizontal axis between P and
Q (see figure 8.5(b)). Only at L = Lc (the point Q) the state a = 0 looses its
stability and the system suddenly jumps to the state R or R′ (dependent on
the nature of the inevitably present, small disturbances); a catastrophe (see
figure 8.5(c)). When the system has ended up at R the state of the system
will, upon a gradual decrease of L, remain on the top stable solution branch,
somewhere in between R and S (see figure 8.5(d)). Upon further decrease of
L, the state of the system will suddenly jump at L = Ls from the top stable
solution branch (point S) to point P on the equilibrium branch a = 0 (see
figure 8.5(e)). The arrows in the figure denote how the state of the system
changes upon gradual increase and decrease of the system parameter L. On
the interval Ls ≤ L ≤ Lc a so-called hysteresis loop (P QRSP ) occurs. In a
hysteretic system, the reversibility is lost: the way forward (increasing L) is
different from the way back (decreasing L).
The possibility for catastrophic jumps and hysteresis are a characteristic for
a system that shows multiple stable solution branches, such that the system
has more than one stable state for the same value of the system parameter.
In the example of the clamped wire, for Ls < L < Lc the system is stable
in the vertically upright position (a = 0), or in the position bend to the left

144
(a < 0), or to the right (a > 0).
A key example where catastrophic jumps and hysteresis occur is in our cli-
mate. So-called ‘tipping points’ in our climate are points where, when a
system parameter such as the amount of CO2 in the atmosphere is gradually
changed, the system undergoes a catastrophic jump to a new equilibrium
state, that cannot simply be undone by lowering the CO2 levels again; the
system has reached a new stable equilibrium branch. Here indeed, a small
change in the system parameters may lead to a catastrophic change in the
fate of the system. Tipping points are therefore a major topic of study for
climate researchers.

8.3.1 The role of imperfections in catastrophe behavior

The buckling strut of figure 8.1 was an example of a system with perfect
left-right symmetry, reflected by the x-axis symmetric bifurcation diagram
shown in figure 8.2(b). In this system, a gradual change in the system pa-
rameter resulted in a gradual change in response: if the force on the strut is
gradually increased beyond the bifurcation point Fc , the unbend state gradu-
ally changes into one of the buckled configurations and the system continues
to move along one of these stable branches. Here no sudden changes of state
occur like they did for the bending ruler in figure 8.5.
In reality, however, if you carry out the experiment you will find that the
strut will buckle more easily in one direction than in the other. The reason
is that physical systems are often subject to small imperfections (in case
of the strut, for example some material imperfections or asymmetry in the
fixation point). These imperfections lead to a loss of symmetry, and subtle
changes in the bifurcation diagram. These subtle changes can have dramatic
consequences for the system’s behavior as they will allow for catastrophic
jumps and hysteresis.
In figure 8.8 we show the consequence of an imperfection in the bifurcation
diagram of the buckling strut (see also figure 8.2(b) for comparison). This
time, upon a gradual increase of F the system will naturally follow the upper
solution branch for F > Fc . If we deliberately push the system such that the
state jumps to the lower stable branch and then let F decrease gradually, we
see that at a certain smaller value of F (but still > Fc ) the system suddenly
jumps back to the original state in the top branch, and subsequently goes
back along this branch until it reaches the stable equilibrium xa = 0. Gen-
erally this catastrophic jump is small when the imperfection of the system

145
 Strut bending to the right
xa

F
0

Strut bending to the left

Figure 8.8: Bifurcation diagram for a strut with some imperfections: the strut
is more likely to bend towards the right than towards the left. To push the
system to the lower stable branch you need a big disturbance. When the force
on the strut is then decreased, the system will move along this branch until
it undergoes a catastrophic jump back to the stable upper branch.

(i.e. the symmetry breaking) is small, but it does occur!13

8.4 Non-linear oscillations

In chapters 5 and 7 we have seen that second (or higher) order systems can
show oscillations, i.e. periodic motions in time, and that these oscillations
show up as closed orbits in the phase diagram (see §6.1). We discussed the
pendulum as an example of such a system, and showed how we can use Taylor
series to linearize the pendulum equation around its two stationary points
to investigate its behavior using a linear stability analysis; see §7.2.3. The
results of this analysis allowed us to sketch the behavior of the system in the
phase plane close to the stationary points; see exercise 9.8.6 and figure 8.9.
To learn about the behavior of the pendulum in between the stationary
points, we have to embrace the full non-linearity of the pendulum equation
(7.16). Fortunately this conservative system allows for construction of the
13
If you would like to read more about the branch of science called catastrophe theory we
refer you to a nice review paper by Zeeman: E.C. Zeeman - Catastrophe Theory, Scientific
American 234(4) p. 65-83 (1976).

146
 θ

o θ
-

Figure 8.9: Phase diagram of the non-linear pendulum showing its behavior
close to the stationary points.

phase space trajectories from energy conservation (see §6.1.3B), by plotting

the trajectories for different values of the total energy E, as shown in fig-
ure 8.10. The physical interpretation of this phase diagram is as follows.
The stationary points θ = 0, 2π, . . . are neutrally stable; small perturbations
will cause the pendulum to oscillate around these positions with a constant
(small) amplitude. As the initial perturbation, and hence the total energy,
increases, the orbits around the stationary points and hence the oscillation
amplitude grows. At a critical total energy E = mgl the trajectories just join
the saddles at θ = ±π, . . . . These unstable saddle points represent the pendu-
lum at rest in its vertically upright position. At E = mgl the pendulum just
has enough energy to come to rest exactly at the inverted position. For larger
energy the pendulum will swing over the top and make full cycles around the
suspension point. These oscillations are non-linear and cannot be described
by the single sine or cosine that we used for the simple, small-amplitude
harmonic oscillations around θ = 0, as you will see in exercise 9.8.9.
The two-dimensional phase plane as depicted in figure 8.10 is a little bit
misleading, in the sense that the large-amplitude full revolutions of the pen-
dulum are periodic, but do not show up as closed orbits there. In fact, the
points θ = −π and θ = π are not separate points, as suggested by the phase-
plane plot, but coincide. A more natural way to plot the phase diagram is
therefore to wrap the phase plane around the surface of a cylinder, such that
θ = −π and θ = π indeed correspond to the same point; see figure 8.11. Now,
all trajectories are closed curves: the trajectories going around the cylinder
correspond to full motion around the suspension point, whereas closed curves
around the point (θ, θ̇) = (0, 0) describe the small oscillations back and forth

147
 θ

- θ

Figure 8.10: Complete phase diagram of the non-linear pendulum. We ob-

tained the trajectories making use of energy conservation (see §6.1.3B).

around θ = 0.

Figure 8.11: The periodic phase plane of figure 8.10 is wrapped around a
cylinder such that the points θ = −π and θ = π overlap.

148
8.5 Limit cycles
The non-linear oscillations of the pendulum described above show up in the
cylindrical phase space as a set of closed orbits. However, non-linear systems
may also posses closed orbits that appear in isolation (i.e. without any other
closed orbits nearby). Such an isolated closed orbit is called a limit cycle.
This name comes from the fact that all trajectories that are sufficiently close
to a limit cycle either spiral away from it or towards it as time evolves; see
figure 8.12.

x2 x2 x2

x1 x1 x1
(a) (b) (c)

Figure 8.12: Three examples of isolated closed orbits in phase space: tra-
jectories either spiral (a) towards or (b) away from a limit cycle, or (c) a
combination of both.

Limit cycles are very important scientifically. The reason for that is that if
all nearby trajectories are attracted to a limit cycle, such as in figure 8.12(a),
they represent a periodic motion that is self-sustained: even in absence of
any forcing, these systems will evolve into oscillations. Some important ex-
amples of such behavior are the beating of the heart, daily rhythms in body
temperature or the detrimental self-excited oscillations in bridges or airplane
wings. In §8.5.2 we will discuss a famous example of limit-cycle oscillations
in electronic circuits.
Given the importance of limit cycles, scientists are often concerned with
identifying the presence of isolated closed orbits in the systems they study.
Unfortunately, to proof that a limit cycle must exist based on the evolution
equations of a system alone is often very hard, if not impossible, and only
suspicions, for example based on numerical simulations, remain. One theo-
rem that can sometimes be used to prove the existence of closed orbits (or, as

149
we will see later, rule out the existence of chaos) is the Poincaré-Bendixson
theorem.

8.5.1 Poincaré-Bendixson theorem

The discussion on the existence of limit cycles or chaos rapidly becomes very
mathematical and advanced, and way beyond the scope of our course. The
purpose of this paragraph is therefore merely to give you a glimpse of what
theorems to rule out or establish the existence of such complex dynamics
look like, rather than a rigorous treatment of the subject.
The Poincaré-Bendixson theorem states that the dynamic behavior that is
possible in the phase plane (i.e. in a two-dimensional system) is rather lim-
ited:
If there exists a trajectory that is confined within a closed, bounded region
in the phase plane, then it must either
(a) end in a stationary point, or
(b) return to the point from which it began (and therefore form a closed
orbit), or
(c) approach a limit cycle.
These cases are schematically represented in figure 8.13. The implication of

x2 x2 x2

x1 x1 x1

Figure 8.13: Illustration of the Poincaré-Bendixson theorem: The three pos-

sible fates of a trajectory that is confined to a closed, bounded region in the
phase plane: (a) it ends in a stationary point, (b) it returns to its starting
point, (c) it ends in a limit cycle.

this theorem is that if we can find a closed, bounded region within the phase

150
plane that does not contain any stationary points and in which a trajectory
remains trapped, then there must be a limit cycle within this region.
The formal proof of the Poincaré-Bendixson theorem requires tools from ad-
vanced mathematical analysis and will not be discussed here. Roughly, the
proof involves the (highly non-trivial) construction of a so-called ‘trapping
region’ in the phase plane, where the velocity vector across all boundaries
is pointing into the region, such that the trajectories inside it will remain
trapped forever. Next, we can argue that under these conditions the trapped
trajectories must end in a limit cycle by using the following property: tra-
jectories in the phase plane cannot cross themselves. To see why that is the
case, consider the most general form of a two-dimensional system:

ẋ1 = f (x1 , x2 )
(8.3)
ẋ2 = g(x1 , x2 ),

which has stationary points f (x1 , x2 ) = g(x1 , x2 ) = 0. From equation (8.3)

we can derive that
dx2 g(x1 , x2 )
= (8.4)
dx1 f (x1 , x2 )

which expresses that in every point (x1 , x2 ) of the phase plane the direction
dx2 /dx1 of the trajectory has one single value, unless the point in question is
a stationary point (for which both f = 0 and g = 0 simultaneously). As a con-
sequence, in a two-dimensional system trajectories cannot cross; if they did
cross the derivative in equation (8.4) would be multi-valued, and trajectories
would be able to move in two directions at once. The only exception to this
rule are the stationary points (where nothing moves); there trajectories can
cross themselves, or another trajectory.
The notion that trajectories cannot cross has important implications for our
trajectories in the trapping region: since this region does not contain any
stationary points and the trajectories will remain confined inside this region
at all times, they have no choice but the spiral towards a limit cycle to prevent
them from crossing each other or themselves.
The results of the Poincaré-Bendixson theorem are only applicable in a two-
dimensional phase space, i.e. the phase plane. In higher dimensional systems
a radically different behavior can occur: trajectories can wander around in
a bounded region forever, without settling into a stationary point or closed
orbit. The fate of these trajectories can be extremely sensitive to the initial

151
conditions from which they start. This type of aperiodic behavior that is
extremely sensitive to initial conditions is called chaos, as we will discuss in
§8.6. A highly important corollary of the Poincaré-Bendixson theorem is that
chaos cannot occur in two-dimensional systems, where the fate of trajectories
confined to bounded regions is simple; recall figure 8.13.

8.5.2 The Van der Pol oscillator

A famous example of a two-dimensional physical system where a self-sustained
limit-cycle oscillation occurs is the electronic system first developed by the
Dutch mathematician and physicist Balthasar van der Pol. For his employer
Philips in Eindhoven, van der Pol studied the oscillatory behavior of an elec-
trical circuit with a triode vacuum tube. This circuit is shown in figure 8.14.
It is the triode that makes this system interesting, because of its non-linear

Figure 8.14: The electrical circuit studied by van der Pol. Reproduced from
figure 11.3 of (1).

behavior. The (dimensionless) electrical current x in this system is described

by the Van der Pol equation

ẍ + ε(x2 − 1)ẋ + x = 0 , (8.5)

where ε is a positive constant.

We can see directly that the parameter ε determines the extent to which
non-linearity is of importance: when ε = 0 equation (8.5) describes the lin-
ear oscillator, but the non-linear term x2 ẋ becomes more important as ε
increases. In fact, the term ε(x2 − 1)ẋ acts as a negative damping for ∣x∣ < 1,
causing oscillations to decay, and as a positive damping for ∣x∣ > 1, causing
oscillations to grow in amplitude. As a consequence, the system eventually

152
settles into self-sustained oscillations where the energy that is pumped in
exactly balances the energy that is dissipated during one cycle14 ; see figure
8.15. In the (x, ẋ) phase diagram, this periodic motion is visible as a limit

Figure 8.15: Example of a self-excited oscillation, the solution to equation

(8.5) for ε = 0.1.

cycle, as shown in figure 8.16: the phase trajectories in the vicinity of the ori-
gin spiral outwards to a closed curve (drawn with a thick line). By contrast,
trajectories starting from ‘large initial conditions’ spiral inwards towards the
same cycle. Since the limit cycle is a closed curve, it corresponds to a periodic
oscillation in time.
Figure 8.17 shows the numerically determined (x, t)-plots and their corre-
sponding (x, ẋ) phase diagrams of equation (8.5) for the cases ε = (a) 0.1 ,
(b) 1.5, and (c) 10, all with the initial conditions x(0) = 0 and ẋ(0) = 0.01.
For small ε (figure 8.17(a)), the phase diagram looks elliptical and the oscil-
lation in x(t) is approximately harmonic. Upon an increase in the value of ε
we see that the system tends to a periodic oscillation much faster, but this
oscillation does not look harmonic any longer (see figures 8.17(b,c)) and the
shape of the limit cycle deviates more and more from an ellipse. In figure
8.17(c) we see that in every cycle x slowly decreases from the value 2 to
the value 1, and then quickly jumps to the value −2. In the second half of
the cycle x gradually increases from −2 to the value −1, and then quickly
jumps back to the value +2, after which the cycle repeats itself. The bigger
the value of ε, the more abrupt the jumps are and the larger the oscillation
period is.
14
This intuitive reasoning can be made more rigorous; mathematicians have proven that
the Van der Pol equation has a unique, stable limit cycle for each ε > 0.

153
Figure 8.16: The phase diagram of the Van der Pol oscillator (8.5) for ε = 1.
Image taken from (Van der Pol, 1926)(3).

Oscillations of the form shown in figure 8.17(b,c) are called relaxation os-
cillations. The physical interpretation of the asymmetry in the oscillation
is that the dynamic system gradually absorbs energy during the oscillation,
and then subsequently quickly releases it: in this phase the system relaxes.
This behavior can be observed best by doing the computer simulation (us-
ing a modified version of the Excel-file Lotka Volterra, or Matlab program
POLPHASE) yourself.15

15
These Matlab programs and a description of the programs can be found on Canvas.

154
Figure 8.17: The behavior of the Van der Pol oscillator (8.5) for ε = 0.1
(a), 1.5 (b), and 10 (c) with initial conditions x(0) = 0 and ẋ(0) = 0.01.
Calculations are done using the Matlab program POLPHASE (a modified
version of XTPHASE).

8.6 Chaos
The phase-space of the Van der Pol equation is two-dimensional. As discussed
in §8.5.1 the long-term evolution of a non-linear two-dimensional system will
always be predictable. We now turn to systems with a phase-space of three-
dimensions or higher. While the Poincaré-Bendixson theorem states that
the fate of trajectories in a two-dimensional system is limited (they end
in a stationary point, a closed orbit or a limit cycle), in three-dimensional
systems that is no longer the case. In three- or higher dimensional systems
the Poincaré-Bendixson theorem no longer applies, and chaotic behavior can

155
occur.
Let us first specify what we mean by chaos. Even though there is no uni-
versally accepted definition of chaos, people have agreed on the following
characteristics:

1. The motion of the system is irregular, i.e. it never repeats itself as time
evolves, and it also does not settle down to stationary points.

2. The motion of the system is highly sensitive to the initial conditions

from which the motion starts. As a consequence, trajectories that start
out closely together in phase space separate from each other exponen-
tially fast as time progresses.

3. The system is deterministic, which means there is no randomness or

noise involved in its parameters or input. The irregular behavior there-
fore comes from the non-linearity of the system, not from noisy driving
forces.

The third condition implies that if we know the initial conditions with infinite
precision, we can predict all future states of the system from its evolution
equations. However, the issue is that in practice we are never able to reach
such precision, and due to condition 2, any small deviation in initial condi-
tions will lead to vastly different outcomes in the long-term.
We now illustrate the beauty and complexity of chaotic motion in two exam-
ple systems.

8.6.1 A cubic harmonic oscillator

One of the ‘simplest’ dynamic systems that shows chaotic behavior is the
harmonically forced mass-spring system of figure 5.5, but now with non-
linear spring characteristics. With a cubic spring force F (x) ∼ x3 and linear
damping (∼ ẋ) the (dimensionless) equation of motion becomes:

ẍ + k ẋ + x3 = A cos Ωt . (8.6)

Here, k is the damping coefficient, A the forcing amplitude, and Ω the forcing
frequency. Note that equation (8.6) might be a second-order differential
equation, but it is non-autonomous. If we rewrite equation (8.6) into an

156
autonomous system of first-order differential equations we arrive at
ẋ = y
ẏ = −ky − x3 + A cos Ωt (8.7)
ṫ = 1 .
Since corresponding phase space is three dimensional, the Poincaré-Bendixson
theorem does not apply (see §8.5.1), and the system can potentially show
chaotic behavior, as we will see.
The behavior of the solution of equation (8.6) is illustrated in figure 8.18.
This figure shows the numerically determined solution for the parameter
values k = 0.01, Ω = 1, and A = 7.5 for two different initial conditions: the
thick line (a) for x(0) = 3 and ẋ(0) = 3, and the thin line (b) for x(0) = 3.003
and ẋ(0) = 3. The irregular course of the solution x(t) is obvious in both

Figure 8.18: Numerically determined solutions to equation (8.6) for k =

0.01, Ω = 1, A = 7.5 with initial conditions (a) x(0) = 3, ẋ(0) = 3 and
(b) x(0) = 3.003, ẋ(0) = 3. Calculations are done using the Matlab program
NXT.

cases: the non-linearity of the equation of motion (8.6) apparently gives rise
to irregular oscillations that never perfectly repeat the same pattern. With
that comes the sensitivity to the initial conditions: although the values of
x(0) only differ 1%, the solutions (a) and (b) for these initial conditions,
show radically different behavior after 5 oscillation periods of the forcing
term (A cos Ωt).
We can also map the behavior of the solution by plotting it in the (x, ẋ) phase
diagram: see figure 8.19. It can clearly be seen that the trajectory in this
two-dimensional phase-plane crosses itself time and time again, each time in

157
Figure 8.19: The chaotic oscillations of the non-linear periodically forced os-
cillator given by equation (8.6). Phase diagram constructed using the Matlab
program XTPHASE.

a different point, but the trajectories do remain confined in space (i.e. they
do not wander off to infinity). The system shows chaotic behavior. The
trajectory drawn in figure 8.19 is actually a projection of a three dimensional
trajectory on the x, y-plane: the chaotic phase trajectory actually comes out
of the plane of the drawing (in the positive t-direction), and does not cross
itself in the three-dimensional phase space.
In this example the three-dimensionality of the system arises from the fact
that the second-order evolution equation describing it is non-autonomous.
In the next example we are going to focus our attention to an autonomous
system with a three dimensional phase space.

8.6.2 The Lorenz equations

Several decades ago the meteorologist Lorenz introduced an over-simplified
model of convection flow in the atmosphere that became world-famous for its
complicated dynamics. A sketch of atmospheric convection is shown in figure
8.20: the medium in between the horizontal plates is brought into motion by
thermal forcing (a temperature difference T1 − T2 ).
An important parameter in this problem is the dimensionless Rayleigh num-
ber, which describes the ratio between the destabilizing upward buoyancy
forces (∼ g∆T ) and the stabilizing viscous forces in the heated fluid. When
the Rayleigh number gets above a certain critical value, the convection can
organize in a certain pattern, such as rolls and hexagonal cells. At even
higher values of the Rayleigh number the cell patterns become unstable and

158
Figure 8.20: Thermally-driven convection flow in between two horizontal
plates. (a) The initially ordered motion can at a sufficiently large Rayleigh
number transform into (b) convection rolls.

the convection flow can become turbulent, showing chaotic motion.

In itself the description of this motion is complicated: the equations for the
flow and heat transport form a system of three coupled non-linear partial
differential equations. By applying several simplifications Lorenz brought
these equations for convection in the cell regime back to the following set of
(dimensionless) coupled ordinary differential equations:

ẋ = −σ(x − y)
ẏ = rx − y − xz (8.8)
ż = xy − bz ,

where σ, r, and b are positive constants. In these equations x, y, and z are

measures for the flow velocity, the temperature difference between the up-
ward and downward flowing fluid, and the average convective heat transport,
respectively. Furthermore, σ is a parameter that depends on the properties of
the fluid (the dimensionless Prandtl number, the ratio between the viscosity
and thermal diffusivity), r is the Rayleigh number, and b gives the horizontal
wave number of the convection cells. In his study, Lorenz set σ = 10 and
b = 8/3 and kept these values constant, while he studied the behavior of the
system (8.8) for different values of the parameter r.
Lorenz discovered that the simple-looking deterministic system of equations
(8.8) could show extremely erratic dynamics. He discovered that for a wide
range of parameters, the solutions to this system oscillate irregularly and
never repeat themselves, while they do remain bounded to a finite region in
phase space. This region turned out to have very peculiar properties, as we
will see below.

159
Lorenz tried to analyze the set of equations (8.8) as far as possible using
standard techniques, similar to what we have discussed our course so far.
Step-by-step, he eliminated all known possibilities for the long-term evolution
of the system, until he entered terra incognita. We will follow some of his
steps to reveal the hidden secrets of the set of equations (8.8).

Non-linearity The system of differential equations (8.8) is non-linear. The

two non-linearities are caused by the terms xz and xy, which give a coupling
between the three equations. Equations (8.8) form a dissipative system. The
damping is described by the terms −σx and −bz: it can easily be seen that
for y = 0 the x- and z-equations reduce to

ẋ = −σx , ż = −bz (8.9)

and show exponentially damped solutions.

Phase space The Lorenz equations are an autonomous set of three first-
order differential equations. The state of this system is given by a point
(x, y, z) in the three-dimensional phase space. The so-called ‘velocity in the
phase space’ v ≡ (ẋ, ẏ, ż) is unambiguous: trajectories in this phase space
can never cross themselves or others! However, in the two-dimensional phase
plane, crossings are possible, and the Poincaré-Bendixson theorem does not
apply. Lorenz could prove that all trajectories will remain trapped within
a finite ellipsoidal volume in the three-dimensional phase space, which will
have important implications for the long-time behavior of the system, as we
will see later.

Stationary states and their stability The equations (8.8) have several
equilibrium states (x0 , y0 , z0 ). In exercise 9.8.12 we will show that

• For r < 1 the system only has one stationary state, namely (x0 , y0 , z0 ) =
(0, 0, 0).

• For r > 1 three stationary states exist:

(x0 , y0 , z0 ) = (0, ⎫
⎪
√0, 0) √ ⎪
⎪
= ( √ b(r − 1), b(r
√ − 1), r − 1) ⎬ (8.10)
⎪
⎪
⎪
= (− b(r − 1), − b(r − 1), r − 1) . ⎭

160
The physical interpretation of these stationary states is as follows: for r < 1
(small Rayleigh number, i.e. the viscous forces dominate the upward buoy-
ancy forces due to the heating of the fluid) the heat transport happens via
conduction (there is no flow in this case). When r > 1 convection occurs,
and the heat transport then happens via flow. For r < 1 the system has one
stationary state (x0 , yo , z0 ) = (0, 0, 0), whereas three stationary states exist
for r > 1. Hence, the quantity r is a bifurcation parameter that changes the
number and qualitative behavior of the stationary states.
You can prove, using what we have learned in chapter 7, that the stationary
state (x0 , y0 , z0 ) = (0, 0, 0) is stable for r < 1, and unstable for r > 1. You
will set up this proof in exercise 9.8.12. Moreover, Lorenz proved that for
r < 1 every trajectory will approach the stationary point (0, 0, 0) and hence
no limit cycles or chaos can occur. The other two stationary points listed
in expression (8.10) are (linearly) stable for 1 < r < rc , with rc = σ(σ+b+3)
(σ−b−1) . In
figure 8.21 the bifurcation diagram for the Lorenz system is shown. Beyond
the point r = rc there are no stable stationary points or limit cycles. So what
happens when the stable branches loose their stability beyond r > rc ? In that
range of the phase space, some peculiar behavior arises, as we will see below.

0
1 r = rc r

Figure 8.21: Sketch of the bifurcation diagram for the Lorenz equations (8.8).

Sensitivity to initial conditions Lorenz used numerical simulations to

determine how the system would evolve over time for r > rc . Two typical

161
numerically determined x(t) solutions of equations (8.8) are shown in figure
8.22. As is clear from the figure, the temporal evolution of these solutions
is irregular (aperiodic) in shape, and moreover, it is extremely sensitive to
initial conditions. The x(t) solutions in figure 8.22 only differ 1% in their
initial value for x(0), and show radically different behavior for larger t values.
In the figure this divergent behavior can be observed for roughly t > 14. If
we make the difference in initial conditions smaller by a factor 100 (so the
absolute values then are 5 and 5.00005) the solutions stay similar for only a
slightly longer time: yet at t > 16 they diverge again. Indeed, for r > rc the
Lorenz equations show chaotic behavior.

Figure 8.22: Numerically determined x(t) solutions of equations (8.8) for

σ = 10, b = 8/3, r = 28 (hence r > rc = 24) with initial conditions y(0) = z(0) =
5, x(0) = 5.000 (thick line) and x(0) = 5.005 (thin line). Calculations are
done using the modified Matlab program NXT.

It is obvious that the chaotic oscillations of the system such as plotted in

figure 8.22 are sensitive not only to small changes in the initial conditions, but
also to (small) inaccuracies in the numerical calculations. Every numerical
calculation gives an approximation of the real (analytical – if this would
exist) solution, and therefore implies inaccuracies by definition. Numerical
calculations of the behavior of non-linear system therefore need to be done
with care! The calculations in figure 8.22 have been performed with so-called
‘double precision’ and a Runge-Kutta method with a time step h = 0.001. As
a test the calculations have also been performed with time steps h = 0.002
and h = 0.0005, and since this did not give a significantly different result
we can assume that the solutions of figure 8.22 are accurate for the interval
0 < t < 20. Calculations with a much bigger time step, e.g. h = 0.01, showed

162
substantial deviations, even in the limited interval 0 < t < 20.
Lorenz saw that the chaotic behavior of the equations (8.8) has important
consequences for the long-term predictability of of chaotic systems. Lorenz
(1963) writes:

When our results concerning the instability of non-periodic flow

are applied to the atmosphere, which is ostensibly non-periodic,
they indicate that prediction of the sufficiently distant future is
impossible by any method, unless the present conditions are known
exactly. In view of the inevitable inaccuracy and incompleteness
of weather observations, precise very-long-range forecasting would
seem to be non-existent.

Chaos on a strange attractor We now turn our attention again to the

(r, x)- bifurcation diagram shown in figure 8.21. As we have seen, for r > rc no
stable stationary branches exist that could attract the trajectories in the long
run. So where do the trajectories in phase space go for r > rc ? To visualize
this behavior we plot two-dimensional projections of the three-dimensional
phase space for a fixed value of r > rc in figures 8.23 for the (x, z)-plane, 8.24
for the (x, y)-plane.

Figure 8.23: The projection of the solution trajectory of equations (8.8) on the
x, z-plane, for σ = 10, b = 8/3, r = 28 (hence r > rc ) and the initial condition
(x, y, z) = (5, 5, 5) on t = 0. Calculations are done using the Matlab program
NPHASEXT.

Lorenz discovered that the trajectories in phase space follow a wonderful

path that resembles a pair of butterfly wings. As time evolves, the state of

163
Figure 8.24: The projection of the solution trajectory of equations (8.8) on
the x, y-plane, for r = 28 and the initial condition (x, y, z) = (5, 5, 5) on t = 0.
Calculations done using NPHASEXT.

the system remains ‘dwelling’ on the butterfly wings, without ever settling
into an equilibrium state. For example, in figure 8.23 we see a trajectory that
starts near the origin, and first moves around the unstable point in the right
half of the phase plane, before it suddenly jumps to the left, where it spirals
around the left unstable point for a while, before jumping back to the right,
and so on. The state of the system keeps on changing in this way, without
ever repeating itself. The trajectory seems to be crossing itself several times,
but this caused by the projection. Recall that a trajectory of the autonomous
system given by equations (8.8) can never cross itself, or another trajectory,
in the three-dimensional phase space. In three dimensions, the ‘wings’ of the
butterfly may look like two surfaces that merge into one, but in fact they
are not: Lorenz showed that if you zoom in carefully, each of these surfaces
consists of another set of two surfaces, which in turn consists of yet another
set of two surfaces, and so on. Hence, the butterfly has what is called a
fractal structure.
The motion of the trajectories on the butterfly wings turns out to be highly
sensitive to the initial conditions; i.e. from what point in phase space the
trajectory starts. This sensitivity means that two trajectories that start close
to each other in phase space rapidly diverge as time evolves, and end up in
completely different positions on the butterfly. However, at all times they
will remain confined in space (i.e. they do not leave to infinity), as if attracted
to the butterfly wings. An attracting object in the phase space that displays
high sensitivity to initial conditions is called a strange attractor. The Lorenz

164
attractor is one of the most famous types of strange attractors that appear
in chaotic systems, but many more have been discovered since.
Chaos is a fascinating phenomenon in non-linear system dynamics, but it
is not only a scientific curiosity. Applications of chaos can be found in a
wide range of fields, ranging from analyzing the turbulent motion in flu-
ids, making weather predictions, understanding of cardiac arrhythmia, stock
market predictions, to the use of synchronized chaos for encrypting messages
in communication.

8.7 Summary of important concepts

In this chapter we have seen how non-linearity can give rise to complex
dynamic behavior of a system.
We discussed bifurcations: major qualitative changes in the temporal evolu-
tion of a system that happen when a system parameter is varied. When this
parameter reaches a certain critical value, the bifurcation point, the number
of stationary points and/or their stability changes. These changes can be
visualized in a bifurcation diagram, where the stationary state(s) are plotted
as a function of the system parameter, and the stability of the stationary
states is indicated by the line style (dashed or solid).
When a gradual change in a system parameter leads to an abrupt change in
the state of a system, we call that a catastrophe; the system suddenly jumps
to a new state that is far from the original one. Catastrophic jumps give rise
to hysteresis in a system: irreversible behavior, where the system responds
differently to an increase in the value of the system parameter than to a
decrease by the exact same amount. This irreversible behavior can be illus-
trated by a hysteresis loop in the bifurcation diagram. Catastrophic jumps
and hysteresis are found in systems that have multiple stable equilibrium
states for one value of the system parameter. They are often induced by im-
perfections that break the symmetry and make it more likely for the system
to be in one stationary state than in the other(s).
For two- and higher dimensional non-linear systems, trajectories in the phase
diagram may show a limit cycle: an isolated, closed orbit where all nearby
trajectories will spiral towards or away from. A limit cycle in the phase
diagram will give rise to (damped or forced) oscillations of the system in
time. A theorem that can sometimes be used to prove the existence of limit
cycles is the Poincaré-Bendixson theorem. The Poincaré-Bendixson theorem
states that the fate of trajectories in a two-dimensional system is limited: if

165
a trajectory is confined to a bounded region, it either ends in a stationary
point, moves in a closed orbit, or approaches a limit cycle. In three- or
higher dimensional systems the theorem no longer applies and chaos can
occur. Chaotic motion is irregular (in the sense that it does not repeat
itself) and extremely sensitive to the initial conditions. Chaos is a direct
result of the non-linearity of the system: if a system starts from two only
slightly different initial positions in the phase diagram, this difference may be
amplified in a non-linear way such that the system moves towards two vastly
different locations as time evolves, making the two trajectories diverge.
In some circumstances the trajectories separate but remain attracted to cer-
tain region on the phase diagram such that they stay within a finite volume.
If such an attracting region in the phase diagram has a high sensitivity to
initial conditions (i.e. two points that start close by end up at vastly different
locations on the attractor at later times) it is called a strange attractor.

166
9 Exercises
The level of the exercises is categorized with stars:

• 8: The first time may cause some sweat, but after a few practises, this
should be done with your eyes closed.

• 88: Requires some physical insights and multiple steps to find the
right answer.

• 888: Finishing these exercises correctly, results in a great feeling of

pride.

We introduce this categorization in this year’s version of the lecture notes

for the first time, so we appreciate your feedback on this change.

9.1 Exercises chapter 1

9.1.1 Dimension check 8
Check whether the following equations are dimensionally correct.

(a) W = F d cos θ, in which W is the work done by a force F over a distance

d and the angle between F and d is θ.
∆P h ̵
(b) ̵ is the reduced
= , in which P is momentum, x displacement and h
∆x 2
Planck constant [J s].

(c) z = Ae(x+ct) , in which x is the position.

2

d2 x dx
(d) + + C = 0, where x is a position and t is the time.
dt2 dt
x − ut
(e) x′ = √ , in which x and x′ are positions, t is the time, u is the
1 − u2 /c2
velocity and c is the speed of light.

167
9.1.2 Dimension check 2 8
Determine in the given equations the dimension of variable(s):
GmM
(a) the gravitational constant G in F = , in which F is the gravita-
r2
tional force and r the distance between masses m and M .
(b) Ω in x = a cos(Ωt), where t is the time.
(c) γ in E = γv 2 with E [J]=[kg m2 s−2 ] the energy and v a velocity. (How
did you check your answer during the evaluate step?)
d2 θ
(d) α and β in + (α + β cos(T ))θ = 0, in which θ [-] is an angle.
dT 2
0.450g
(e) ϵ in ϵ < , in which ωmax is an angular frequency and g is the
ωmax 2
gravitational acceleration.

9.1.3 Dimension analysis 8

Use the method of systematic dimension analysis (see equation (1.5)) for
deriving the results of equation (1.2) on page 8 for the leaking barrel.

9.1.4 Nozzle 88
A fireman holds the nozzle of a fire hose, from which a forceful stream of
water emits (cross section A, velocity V ), see figure 9.1. The water has a
density ρ, and the flow can be considered frictionless. In order to keep the
nozzle under control the fireman has to exert a force F in the direction of
the stream.
Derive using dimension analysis a relationship between the force F and the
variables ρ, V , and A.

Figure 9.1: Nozzle.

168
9.1.5 Fountain 88
In a fountain water flows upwards, against the gravity (g), see figure 9.2.
The velocity at the base of the fountain equals V .

Figure 9.2: Fountain.

The water has a density ρ, and all frictional effects can be neglected.
Derive using dimension analysis the relationship between the height of the
fountain H and the other relevant variables.
- Movie - Exercise: Fountain

9.1.6 Newton’s falling apple 88

An apple with mass M falls from a tree with height H. This phenomenon
gave Newton the idea of the gravitational force, which is related to an accel-
eration g. Derive using dimension analysis an expression for the falling time
t as a function of the relevant parameters in this problem.

9.1.7 Insects, birds, and aeroplanes 88

Insects, birds, and aeroplanes have in common that they are equipped with
wings, and with that they have the ability to travel through the air. The lift
force Flift that a wing can generate is related to the length dimension L of
the wing, the velocity V with which the object flies through the air, and the
density ρ [kg m−3 ] of the surrounding air. Derive using dimension analysis
an expression for the lift force Flift as a function of the relevant parameters
in this problem.

169
9.1.8 Water waves 88
A wave goes across a horizontal layer of water (depth H), see figure 9.3.
Since the gravitational acceleration g plays an essential role in this motion,
the wave usually is denoted by a ‘gravity wave’.

Figure 9.3: Gravity wave.

(a) Derive using dimension analysis the relationship between the wave ve-
locity V and the relevant variables g and H.

The wave is sinusoidal and has a wavelength λ. By measuring the vertical

movement of the free surface at a certain location, we can also characterize
the wave with a frequency f .
(b) Derive a relationship between these wave variables V , λ, and f .

9.1.9 Water cannon 88

A riot policeman aims the jet of a water cannon at the shield of a demon-
strator, see figure 9.4.
The water jet has a cross section A, and the velocity of the water in the jet
is V . The water has a density ρ, and flows frictionless. The demonstrator
has to exert a force F in order to keep the shield at the same position.

Derive using dimension analysis the relationship between the force F that the
demonstrator has to exert and the relevant physical variables in this problem.

9.1.10 Electrical circuit 88

Electrical circuits often consist of three elements; resistors, coils, and capac-
itors. A well known example of a very simple network is the LC-circuit that
consists of a capacitor with capacitance C and a coil with self-inductance L

170
 Figure 9.4: Beam of water.

connected in parallel (see figure 9.5). When at a certain time the capacitor
has stored a certain amount of charge, it will discharge via the coil. Because
of the self induction of the coil this discharge process will be opposed and
this will even cause the charge in the circuit to flow back to the capacitor and
therefore the capacitor will charge again. If one would measure the electrical
current (dimension [A]) in the circuit, oscillating behavior would be observed
since the capacitor periodically charges and discharges.

Derive using dimension analysis the relationship between the oscillation time
T of the alternating current in the LC-circuit and the relevant variables of
this electrical network. N.B. The dimensions of self-induction and capaci-
tance are

L [kg m2 A−2 s−2 ]

C [kg−1 m−2 A2 s4 ].

9.1.11 Electrically charged particle in a magnetic field 88

An electrically charged point particle with charge q (dimension: Coulomb=[A s])
and velocity v is subject to a Lorentz force FL in an externally applied mag-
netic field B (dimension [kg s−2 A−1 ]). Derive using dimension analysis the

171
 L C

Figure 9.5: Electrical LC-circuit.

relationship between the Lorentz force and the relevant physical variables in
this problem.

9.1.12 Wave speed 88

Transverse waves traveling on a rope have a wave speed determined by the
mechanical properties of the rope, namely the tension force T in the rope
and the mass per unit length µ.
Derive using dimension analysis the relationship between the wave speed v
on a rope and the relevant mechanical properties.

9.1.13 Graphing functions 8

Sketch a graph of the following mathematical functions:

(a) f (x) = x+1 , −∞

x
< x < ∞, x ≠ −1

(b) f (x) = sin(x), −∞ < x < ∞

(c) f (x) = cos(x), −∞ < x < ∞

(d) f (x) = ln(x), 0 < x < ∞

(e) f (x) = ln(∣x∣), −∞ < x < ∞, x ≠ 0

(f) f (x) = sinh(x), −∞ < x < ∞

(g) f (x) = cosh(x), −∞ < x < ∞

172
 (h) f (x) = x+1 , −∞
x2
< x < ∞, x ≠ −1
(i) f (x) = x−1 , −∞
3x+6
< x < ∞, x ≠ 1
(j) f (x) = x2 −9 , −∞
9
< x < ∞, x ≠ ±3
(k) f (x) = 1+ex , −∞
ex
<x<∞
(l) f (x) = 1 + x1 , −∞ < x < ∞, x ≠ 0
√
(m) f (x) = x + 1, −1 < x < ∞

9.1.14 Differentiation 8
Determine the following derivatives. You can use:
• Product rule: d
dx [f (x) × g(x)] = f (x)g ′ (x) + f ′ (x)g(x).

• Quotient rule: d
dx [ fg(x)
(x)
]= f ′ (x)g(x)−f (x)g ′ (x)
(g(x))2
.

• Chain rule: d
dx [f (g(x))] = f ′ (g(x)) g ′ (x).
(a) f (x) = sin(px), p is a parameter
(b) f (x) = sin(x) cos(x)
√
(c) f (x) = 9 − x2 , − 3 ≤ x ≤ 3
(d) f (t) = exp (t2 )

(e) f (z) = tan(z) = sin(z)

cos(z)

9.1.15 Integration 8
Determine the following integrals. For some of these integrals, the method of
substitution can be useful. This method is the integral version of the chain
rule, which can be written in the integral form as:

∫ f (g(x))g (x)dx = f (g(x)) + C (9.1)

′ ′

Let u = g(x). Then: du

dx = g ′ (x), which in differential form is du = g ′ (x)dx.
This gives:

∫ f (g(x))g (x)dx = ∫ f (u)du = f (u) + C = f (g(x)) + C. (9.2)

′ ′ ′

173
 (a) ∫−2 (2 + 5x)dx
2

9√
(b) ∫0 9 − x dx
1
(c) ∫1/3 1t dt

(d) ∫ exp(ax) dx, a is a parameter

(e) ∫ sin(px) dx, p is a parameter

(f) ∫ sin2 (x) dx

(g) ∫ sin(x) cos(x) dx

(h) ∫ bax dx, b and a are parameters and b > 0 (Hint ∶ bax = exp [a ln(b)x] )

9.1.16 Special functions 8

Show that
d d
cosh(x) = sinh(x), sinh(x) = cosh(x)
dx dx
and

cosh2 (x) − sinh2 (x) = 1.

174
9.2 Exercises chapter 2
9.2.1 Adding and subtracting vectors 8
⃗ for which holds
(a) Give an expression for w,

⎛2⎞ ⎛5⎞
⎜7⎟ + w⃗ = ⎜ 2 ⎟ .
⎝3⎠ ⎝−1⎠

(b) Take two random, non-equal in length and/or direction, vectors u⃗ and
v⃗ and draw

• u⃗ + v⃗
u + 2⃗
• 3⃗ v
• u⃗ − v⃗
u − 3⃗
• 5⃗ v

9.2.2 Lines 8
Determine a vector representation of the line through

1. (1, 5, 8) and (0, −3, 5)

2. (1, 1) and (3, 2)

9.2.3 Planes 8
Determine a vector representation of the plane

1. Through the points (1, 1, −2), (1, 0, 3) and (−1, −3, 2)

⎛4⎞ ⎛1⎞
2. That contains the line x⃗ = ⎜−1⎟ + µ ⎜ 1 ⎟ and the point (5, 2, 6).
⎝3⎠ ⎝−2⎠

9.2.4 Vector expressions 1 8

⃗ y) = (3, 4) and
Express the following vectors in polar coordinates (r, θ): A(x,
⃗ y) = (−1, 4)
B(x,

175
9.2.5 Vector expressions 2 8
⃗ θ) = (3, π)
Express the following vectors in Cartesian coordinates (x, y): A(r,
⃗
and B(r, θ) = (−1, 2 )
π

9.2.6 Velocity in polar coordinates 1 88

Derive using trigonometry and figure 2.8 that

vr = vx cos θ + vy sin θ,
(9.3)
vθ = −vx sin θ + vy cos θ.

9.2.7 Velocity in polar coordinates 2 88

Show that

vx = ṙ cos θ − rθ̇ sin θ,

(9.4)
vy = ṙ sin θ + rθ̇ cos θ.

And show that with equations (9.3) and (9.4) you can find equations (2.34).

- Movie - Coordinate systems: velocity

9.2.8 Acceleration in polar coordinates 8

Show in a similar manner as above that the acceleration vector a = (ar , aθ )
is given by equations (2.35).

9.2.9 Moving mass 1 8

Given is the time varying position x(t) of a mass moving along the x-axis:

x(t) = x0 + γt2 ,

where x0 is the initial position and γ a constant. Determine the velocity v

and the acceleration a of the mass in the x-direction. Display the evolution
of x(t), v(t), and a(t) graphically.

176
9.2.10 Moving mass 2 8
A mass m moves with a velocity v(t) along the x-axis and its position is
given by

x(t) = x0 + βt + γt2 ,

where x0 is the initial position and β and γ are constants. Determine the
corresponding expressions for the velocity v(t) and the acceleration a(t).
Display the evolution of x, v, and a as a function of the time graphically.

9.2.11 Oscillating mass 8

A mass m moves back and forth harmonically along the x-axis and its position
is given by

̂ sin ωt,
x=x

where x̂ is the amplitude and ω the frequency of the oscillating motion.

Determine the corresponding expressions for the velocity v(t) and the accel-
eration a(t). Display these graphically.

9.2.12 Moving mass 3 8

Given is the velocity v(t) = ̂
v sin ωt with which a mass moves along the x-axis.

(a) Determine the position x(t) of this mass, where x(t = 0) = x0 .

(b) Determine the path length s of the mass in between t = 0 and t = T /2,
where T is the period of the oscillating motion.

9.2.13 Moving mass 4 88

A mass m moves in the x, y-plane, where the velocity is denoted by a velocity
vector v = (u, v) with velocity components u and v in the x and y-direction,
respectively. Sketch the (evolution of the) velocity vector v (in magnitude
and direction) for the following cases:

177
 (a) u=̂
v, v =̂
v.
√
(b) u=̂
v, v =̂
v 3

(c) u=̂
v, v =̂
vt

u=̂
v sin ωt
(d) }
v=̂
v sin ωt

u=̂
v sin ωt
(e) }
v=̂
v cos ωt

9.2.14 Moving mass 5 88

Given is the problem of the mass that is moving in the x, y-plane, from the
initial position (x0 , y0 ) and with an initial velocity (u0 , v0 ), for which it has
been derived that:
x(t) = x0 + u0 t
}
y(t) = y0 + v0 t − 12 gt2 .

(a) Give a graphical representation of x(t) and y(t).

(b) Derive an expression for the trajectory of the mass.

(c) Determine the time t = t∗ at which the mass is (back) at y = y0 .

9.2.15 Projectile motion 88

Calculate the location where a projectile lands after being launched from
⃗ y, z) = (0, 0, h) with an initial velocity v⃗ = (vx0 , 0, vz0 ), where
location A(x,
the acceleration of gravity is given by −g.

178
9.3 Exercises chapter 3
9.3.1 Water tank 88
Consider a water tank, where the input tube is below the free surface of
the water in the tank, and where the volume flow rate into the tank varies
harmonically in time:

ϕV (t) = A0 U (t) = A0 V̂ sin ωt,

where A0 is the cross section of the tube, V̂ is a velocity amplitude and ω a

frequency.

(a) Determine the expression for the water height h(t) with h(t = 0) = h0 .

(b) Sketch the evolution of ϕV (t) and h(t) in one graph as a function of t.
Discuss the result.

9.3.2 Leaking water tank 1 88

Consider a leaking water tank, see figure 9.6. The water flows (frictionless)
from the water tank through an opening (cross section A) at the bottom.
The exit velocity is U . Determine the water height h(t) as a function of the
time t for the case U = constant = U0 and h(t = 0) = h0 .

Figure 9.6: Leaking water tank.

179
9.3.3 Leaking water tank 2 88
See the previous exercise, but now for the case where the exit velocity is
dependent on the water height in the tank:

√
U= 2gh.

- Movie - ISEE example: water tank 2

9.3.4 Linear versus non-linear 8

√
(a) Check that the function x(t) = F (t) = exp(2t) − 1 is a solution to
equation (3.37) on page 46 and check that for example x = 2 ⋅ F (t) is
not a solution. In other words, differential equation (3.37) is non-linear.

(b) Discuss why equation (3.36) on page 46 is non-linear.

9.3.5 First order differential equation 8

Consider the following 1st order differential equation:
dx
= x(2 − x)t. (9.5)
dt
(a) Is this equation linear or non-linear?

(b) Is this equation autonomous or non-autonomous?

(c) Show that the solution with x(0) = 1 is given by

exp (t2 )
x(t) = 2 . (9.6)
1 + exp (t2 )

9.3.6 Order of a differential equation 8

Give the order of the following differential equations, and answer the addi-
tional questions:

180
 dy
(a) + sin(t)y = 0
dt
d2 y dy
(b) + +y =0
dt2 dt
Rewrite this differential equation into a system of 1st order differential
equations.
d2 y
(c) + y3 = 0
dt2
d4 y
(d) +y =0
dt4
Rewrite this differential equation into a system of 1st order differential
equations.

ẋ = y
(e) {
ẏ = −x
Derive one single differential equation for x(t) from this system.
⎧
⎪ ẋ = y
⎪
⎪
(f) ⎨ = −z
ẏ
⎪
⎪
⎪
⎩ ż = 2(x + y)
Derive one single differential equation for x(t) from this system.

9.3.7 Linear versus non-linear 8

Are the following differential equations linear or non-linear? Motivate your
answer.
dy
(a) + sin(t)y = 0
dt
d2 y
(b) + sin(y) = 0
dt2
dy
(c) + sin(ty) = 0
dt
d2 y dy
(d) + +y =0
dt2 dt
d2 y dy
(e) + + e−t y = 0
dt2 dt
181
 d2 y
(f) + y3 = 0
dt2
du
(g) = u2 + 1
dt
d2 x
(h) + tanh(9x) = 0
dt2
d2 x
(i) + ln(x + 1) = 0
dt2
dx
(j) = −xt
dt
Can you solve this differential equation, i.e. find the function x(t)?

(k) ẋ = cx + t2 , where c is a parameter

θ̇ = y
(l) {
ẏ = − sin(θ)
Derive one single differential equation for θ(t) from this system.

ẋ = y
(m) {
ẏ = −x − (x2 − 1) y
Derive one single differential equation for x(t) from this system.

9.3.8 Separation of variables 8

Solve the following differential equations using the method of separation of
variables.
dT α
(a) = − T; T (t = 0) = T0
dt Cw
dx
(b) = x; x(t = 0) = 1
dt
dN
(c) = αN ; N (t = 0) = N0
dt
dv
(d) = βv 2 ; v(t = 0) = −1
dt

182
9.3.9 Drag 88
At high velocities V the resistance of a driving car is dominated by drag (or
air resistance). In good approximation the drag force is proportional to V 2 :

Fdrag = −DV 2 ,

where D is the drag coefficient [kg m−1 ].

At t = 0 the engine is turned off and the car with mass M rolls further with
a (decreasing) velocity V (t) with V (t = 0) = V0 .

(a) Derive the evolution equation for V (t).

(b) Solve the evolution equation for V (t).

(c) Show that the characteristic timescale T in this expression for V can
also be determined from a dimension analysis of the independent vari-
ables in this problem.

9.3.10 Decelerating bike 88

Consider a bike with mass M moving over a horizontal road with velocity
V . At time t = 0 the cyclist stops pedaling and the bike rolls further before
it eventually comes to rest. Assume the resistance force is given by
1
Fw = −kV 2 ,

with k the friction coefficient (k > 0).

(a) Write down the evolution equation for velocity V .

(b) Determine the dimensions of k.

(c) Solve the evolution equation for V assuming that V (t = 0) = V0 .

9.3.11 Railway turntable 888

In rail terminology, a railway turntable is a device for turning railway rolling
stock, usually locomotives. This particular one, as shown in figure 9.7, has
been upgraded with a special non-linear spring attached at the point of ro-
tation and connected to the locomotive (mass m [kg]) on the turntable. The

183
spring force is given by Fs = kx2 , with k the spring constant (k > 0), and the
spring has a length of zero in rest.
The angular velocity for this device is constant and given by θ̇ = ω [s−1 ].
A locomotive is located at a distance from the center of the turntable r [m].
The position is described in polar coordinates.

Figure 9.7: Railway turntable.

(a) Derive the units of k.

(b) Show that, by using Newton’s second law, the radial component of the
evolution equation for the position of the locomotive is given by
k 2
r̈ − ω 2 r + r =0
m

(c) Give the evolution equation for the azimuthal component of the posi-
tion of the locomotive.

9.3.12 Kirchhoff 1 88
For the circuit shown in figure 9.8, calculate:

(a) the current I1 .

(b) the current I2

184
 14.0 V
e + - f

4.0 Ω I2
- + I1
b c
6.0 Ω I3
10.0 V

a d
2.0 Ω

Figure 9.8: Circuit 1

9.3.13 Kirchhoff 2 88
Under steady-state conditions, calculate in the multiloop circuit shown in
figure 9.9:

(a) the current I1 .

(b) the charge Q in the capacitor.

9.3.14 Kirchhoff 3 88
The electric circuit of figure 9.10 contains four resistors: R1 , R2 , R3 and Rv ,
where Rv is an adjustable, variable resistance. In addition, there is an emf
with voltage V .

The resistances can be considered as given. The currents are for convenience
already labelled, but should be considered unknown.

Initially, the variable resistor is set to Rv = 0 Ω.

(a) Give the equation for the loop that runs from the emf, via R1 and then
via R2 back to the emf.

185
 4.00 V
- +
d e

I3 5.00 Ω I3
3.00 Ω
c f
I2
I1 5.00 Ω I1
+ -
b g

8.00 V
I=0
- + + -

a h
3.00 V 6.00 μF

Figure 9.9: Circuit 2

(b) Determine the current I1 , in the case that Rv = 0 Ω.

Now the variable resistor is set to an unknown value Rv .

(c) Give a relationship between the four currents: I1 , I2 , I3 , Iv .

(d) Determine the current I1 for this case.
(e) Under what condition for R2 , R3 and Rv is I1 = 3V
R1 ?

9.3.15 Charging a capacitor 888

For this exercise we consider a similar electrical circuit as discussed when
treating the discharging of a capacitor in §3.5, but now it also contains an
electromotive source ξ (see figure 9.11). At t=0 s, there is no charge in the
capacitor, i.e., q(t = 0) = 0
(a) Use Kirchhoff’s loop rule to formulate the evolution equation for the
charge in this electrical circuit.

186
 I1 R1 Rv Iv R2 I2 R3 I3
V
+ -

Figure 9.10: Circuit 3

(b) Use separation of variables to solve the evolution equation.

(c) Make graphs of the charge q and the current I as a function of time.
(d) When can the capacitor be considered to be fully charged?

closed
ε switch
+ -

R C

Figure 9.11: Electric circuit in which a capacitor is charged.

9.3.16 Langmuir model 888

In the field of surface science, the interaction of molecules or atoms with a
surface is studied under vacuum conditions. We consider a surface that has

187
N surface sites per unit area, i.e., positions where a molecule can adsorb.
The coverage θ is the fraction of sites on which a molecule is absorbed. A
molecule can only adsorb on empty sites, which scales with ∼(1-θ). The
number of molecules impacting on the surface (leading to adsorption) can be
described with a flux Γ [m−2 s−1 ].

(a) Make an inventory of the variables and their dimensions.

(b) Write down the expression for how the coverage θ changes during a time
interval ∆t as a result of the molecules adsorbing on the surface. (Hint:
find the relationship between the variables based on their dimensions).

(c) Give the evolution equation for the coverage of molecules on the surface.

(d) Solve the evolution equation, assuming θ(t = 0) = 0 as the initial con-
dition.

When adsorbed on the surface, a molecule can also escape from the surface
again, in process referred to as desorption. Such a desorption reaction scales
with the coverage (∼ θ), and is characterized with a so-called rate constant
k [s−1 ], which is basically a probability for whether desorption occurs.

(e) Add a term accounting for the desorption to the evolution equation
derived in (c).

9.3.17 Solar sail 888

A possible method of spacecraft propulsion for future space missions is to
make use of the momentum of photons from sunlight impacting on a large
mirror. In analogy to the role of wind in sailing, such a mirror is called a
solar sail.
Each photon has a momentum pp [kg m s−1 ], of which we assume a fraction
η [-] is transferred to the spacecraft when it is reflected on the solar sail. The
force on the spacecraft as a result of the incoming photons is perpendicular to
the solar sail, as shown in figure 9.12. The number of photons impacting on
the mirror can be described with a flux Γ [m−2 s−1 ]. We consider a spacecraft
with mass m [kg] and a mirror with area A [m2 ].

(a) Make an inventory of the variables and their dimensions.

188
Figure 9.12: The force executed on the spacecraft is a consequence of the
photons impacting on the solar sail.

(b) Give an expression for how the momentum of the spacecraft psc changes
during a time interval ∆t as a result of the light captured on the sail.

(c) Calculate the force Fs exerted on the sail.

(d) Give the evolution equation for the position of the spacecraft.

(e) Solve the evolution equation, assuming x(t = 0) = x0 and v(t = 0) = 0

m/s as the initial conditions.

9.3.18 Characteristic timescales 88

Use dimension analysis to find the characteristic timescales for:

(a) the cooling of an object, based on the variables defined in §3.2.

(b) the discharging of a capacitor, based on the variables defined in §3.5.

Hint: [F]=[kg−1 m−2 A2 s4 ]; [Ω]=[kg m2 A−2 s−3 ].

(c) the slowing down of a rolling car, based on the variables defined in §3.4.

For exponential growth, the term e-folding indicates a timescale τ during

which an exponentially growing variable increases with a factor e.

(d) Give the differential equation for exponential growth. Give an expres-
sion for the doubling time in terms of the e-folding timescale τ .

189
9.4 Exercises chapter 4
9.4.1 Nondimensionalizing evolution equations 88
Consider the decelerating bike of exercise 9.3.10.
(a) What would be an appropriate typical velocity scale to nondimension-
alize velocity V ?
(b) What would be a typical time scale to nondimensionalize time t?
(c) Nondimensionalize the evolution equation for V that you found in ex-
ercise 9.3.10 including the initial condition.
Consider again the leaking water tank of exercise 9.3.3.
√
(d) Show that A0 /A h0 /2g has the dimension [s].
(e) Nondimensionalize the evolution equation for h(t) derived in exercise
9.3.3 choosing an appropriate characteristic time and length scale.
(f) Compare the result of question (e) to the result of question (c). What
do you observe?

9.4.2 Atmospheric pressure 888

The pressure p [Pa]=[kg m−1 s−2 ] in the atmosphere of the earth changes with
altitude h [m] according to:
dp −mgp
= , (9.7)
dh kT
with m [kg] the average mass of the air molecules, g [m s−2 ] the gravitational
acceleration, k [JK−1 ]=[kg m2 s−2 K−1 ] the Boltzmann constant, and T [K] the
temperature. The pressure at the earth surface, p(h = 0) = p0 .
We assume that the temperature decreases linearly with the altitude T (h) =
T0 − Lh with T0 the temperature at the earth surface and L [K m−1 ] a pro-
portionality constant, resulting in the differential equation:
dp −mgp
= . (9.8)
dh k(T0 − Lh)
(a) Nondimensionalize evolution equation (9.8).
(b) Solve the evolution equation to find the expression for p(h).

190
9.4.3 Nondimensional electrical circuit 888

R L

Figure 9.13: Electrical circuit containing a resistor R, a capacitor C and an

inductance L.

We consider an electrical circuit that contains besides a resistor R and ca-

pacitor C, an inductance L, as shown in figure 9.13. An induced voltage in
a coil as a consequence of the inductance is given by:
dI d2 q
V =L =L 2 (9.9)
dt dt
at t = 0, there is a charge Q0 in the capacitor.
(a) Make an inventory of the variables and their dimensions.
(b) Use Kirchhoff’s loop rule to write down the evolution equation for the
charge in the circuit.
(c) What is an appropriate characteristic timescale to nondimensionalize
the independent variable t?
(d) What is an appropriate typical value for the charge to nondimension-
alize the dependent variable q?
(e) Nondimensionalize the evolution equation for the charge in the circuit
derived in (a). Note that this also requires nondimensionalizing the
initial condition.
(f) Simplification of the nondimensionalized evolution equation gives a di-
mensionless parameter. Check that this parameter indeed does not
have a dimension.

191
9.4.4 Nondimensional mechanical system 888
In chapter 5, we will deduce a differential equation for a mass-spring system
with damping:
d2 x dx
m 2
+γ + αx = 0, (9.10)
dt dt
in which γ is the damping constant and α is the spring constant. At t = 0,
the mass is at position x0 .
(a) Deduce from equation (9.10) what the dimensions of γ and α are.
(b) What is an appropriate characteristic timescale to nondimensionalize
the independent variable t?
(c) What is an appropriate length scale to nondimensionalize the depen-
dent variable x?
(d) Nondimensionalize the evolution equation for this mass-spring system.
(e) Compare the result of question (d) to the result of question 9.4.3(e)
What do you observe?
(f) Study the analogy between electrical and mechanical systems by filling
out the table below.
Characteristic Mechanical system Electrical system
Independent variable time t
Dependent variable
Inertia mass m
Rate of change current I
Resistance Resistance R
Stiffness spring constant α

9.4.5 Here comes the sun 888

Consider an object in the atmosphere of the earth that is heated when it
receives daylight, which we model with a heat flow rate:
ϕw,1 = ϕ0 sin2 ωt (9.11)
The object continuously loses heat to the surrounding ϕw,2 according to equa-
tion (3.20). The initial temperature is T0 .

192
(a) Make a graph of ϕw,1 as a function of time. Define what ω should be.

(b) Give the evolution equation for the temperature of the object.

(c) Nondimensionalize the evolution equation. See equation (4.9) for how
to nondimensionalize the time t, instead of using ω.

(d) Simplification of the nondimensionalized evolution equation gives a di-

mensionless parameter, which we indicate here with β. What does this
dimensionless parameter represent?

(e) Solve the evolution equation for β >>1 and discuss your result.

(f) Solve the evolution equation for β <<1 and discuss your result.

193
9.5 Exercises chapter 5
9.5.1 Taylor series around x = 0 8
Expand the following functions in a Taylor series up to order N = 3 around
x = 0:

(a) f (x) = exp(x)

(b) f (x) = cos(x)

(c) f (x) = sin(αx), α is a parameter

(d) f (x) = sinh(x)

(e) f (x) = cosh(x)

(f) f (x) = tan(x) = sin(x)

cos(x)

9.5.2 Linearization 88
Linearize the following non-linear differential equations for ∣x∣ ≪ 1 (i.e., write
the non-linear term as a Taylor series) and then solve:

(a) d2 x
dt2 + sin(x) = 0

(b) d2 x
dt2 + tanh(−16x) = 0

(c) dx
dt = ln(x + 1)3

(d) d2 x
dt2 dt + 2 sin(x) = 0 (Hint: try the solution x(t) ∼ e for the
+ 3 cos(x) dx pt

linearized equation, where p is a parameter that has to be determined.)

(e) ẍ + (x2 − 1) ẋ + x = 0 (Hint: define x(t) = exp ( 2t ) y(t) in the linearized

equation and derive a differential equation for y(t).)

194
9.5.3 The simple pendulum 88
Consider the pendulum (see figure 5.2 on page 80), of which the motion can
be described conveniently in polar coordinates, see figure 5.3 on page 81.

(a) Determine the r and θ-components of Newton’s Law F = dt d

(mv) for
the oscillating mass m. [If you do this correctly you find equation (5.3)
on page 81 for the θ-component].
(b) Show that the linearization of the pendulum equation (5.3), page 81,
results in equation (5.12), page 85. Which equation do you obtain when
you also include the next term of the Taylor expansion of sin θ?
(c) Determine the general solution of the linearized pendulum equation
(5.12). Determine the integration constants for the following initial
conditions:
(i) t = 0 ∶ θ = θ0 , θ̇ = 0;
(ii) t = 0 ∶ θ = 0, θ̇ = Ω;
(iii) t = 0 ∶ θ = θ0 , θ̇ = Ω.

9.5.4 The linear oscillator 8

The general solution to the equation (5.25) on page 89 for the linearized mass-
spring oscillator (see figure 5.5) is given by (5.28) on page 89. Determine the
constants A and B for the following initial conditions:
(i) t = 0 ∶ x = x0 , ẋ = 0;
(ii) t = 0 ∶ x = 0, ẋ = v0 ;
(iii) t = 0 ∶ x = x0 , ẋ = v0 .

9.5.5 Relativistic and classical kinetic energy 88

The relativistic kinetic energy Erel,kin is given by
m 0 c2
Erel,kin = mc2 − m0 c2 = √ − m0 c2 (9.12)
1 − ( vc )2

where m0 is the mass in rest. For small speeds v (i.e. v

c ≪ 1) the kinetic
energy can be described by the classical expression.

195
 (a) Give the classical (normal) expression for the kinetic energy Eclass,kin of
a mass.
(b) Use a Taylor expansion to approximate equation (9.12) for small speeds.
Hint: use ( vc )2 = x around x = 0.
(c) Show that the relativistic kinetic energy Erel,kin in equation (9.12) is
the classical kinetic energy Eclass,kin if v ≪ c. Hint: assume that the
mass is equal to m0 at small speeds.

9.5.6 The damped oscillator 88

(a) Show that by using the trial solution x(t) = Ce−αt cos (Bt − D) the
general solution x(t) for the damped oscillator – described by equation
(5.39) on page 92 – is given by equation (5.40):
⎧ √ ⎫
1 ⎪
⎪ 1 2 ⎪
⎪
x(t) = Ce − 2 kt
cos ⎨ (ω − k ) t − D⎬ ,
2
⎪
⎪ 4 ⎪
⎪
⎩ ⎭
for the case 14 k 2 < ω 2 .
Hint: it is useful to substitute the trial solution x(t) = Ce−αt cos (Bt − D)
in equation (5.39) and then choose the parameters α and B such that
the trial solution is in fact a solution to equation (5.39).
(b) Investigate the properties of this damped oscillator using the excel file
[Link]. Explain what differences you can observe by changing
the damping term k.

- Movie - Exercise: the damped oscillator

9.5.7 The periodically forced oscillator 88

(a) Verify that for the resonance frequency Ω = ω
a
x(t) = t sin ωt
2ω
is a solution for the periodically forced oscillator, described by equation
(5.44) on page 94, with initial conditions x(0) = ẋ(0) = 0. Also derive
the expression for the velocity ẋ(t) and the acceleration ẍ(t).

196
 (b) Investigate the behavior of equation (5.44) using the excel file pendu-
[Link], for example for the cases Ω = 0.8ω and Ω = ω (see figure 5.9).
Describe the behavior you observe.

9.5.8 Simulation damped & forced oscillator 8

Investigate and describe the behavior of the damped and forced mass-spring
system using the excel file [Link].

(a) Set the forcing term to zero (A = 0) and investigate the influence of
damping by using different values of k as input (as in exercise 9.5.6(b)).

(b) Set the damping to zero (k = 0) and investigate the influence of forcing
by using different values of A and Ω as input. The value Ω = ω will
result in resonance (as in exercise 9.5.7(b)).

(c) Now combine damping and resonant forcing (Ω = ω). Try different
values for k and A and investigate when the damping is dominant, when
the forcing is dominant and when damping and forcing are balanced.

9.5.9 Nondimensionalization 88
(a) Nondimensionalize the linearized pendulum equation as given by given
by equation (5.12).

(b) Nondimensionalize the linearized equation of motion of the mass-spring

system given by equation (5.25).

(c) Compare the answers to (a) and (b) to each other. What do you notice?

9.5.10 The pendulum: energy balance 1 88

Consider the undamped pendulum of figure 5.2 on page 80, of which the
equation of motion is given by equation (5.3) on page 81, which we now
write as

θ̈ + ω 2 sin θ = 0. (9.13)

During the oscillations θ(t) of the mass, kinetic energy Ekin is continuously
being converted into potential energy Epot and vice versa.

197
 (a) Give the expression for the kinetic energy Ekin of the swaying mass.

(b) Derive the expression for the potential energy Epot (θ) that the mass
has at a certain excursion θ.

(c) Set up the equation that describes the energy balance.

(d) Derive this equation for energy conservation again, but now by multi-
plying the terms in equation (9.13) by θ̇ and rewriting this in the shape
of d[⋯ ]/dt = 0. Hint: use θ ⋅ θ̇ = dt ( 2 θ ).
d 1 2

(e) Does the result found in (d) correspond to what has been found in (c)?

9.5.11 Conservation of energy 1 888

We reconsider the mass-spring system (see §5.3). The linearized oscillator
equation is given by

mẍ + αx = 0. (9.14)

We would like to investigate the energy of the system, for that we will use
the linear oscillation equation and apply several mathematical ‘tricks’.

(a) Rewrite this second-order differential equation into a system of two

first-order differential equations. Hint: use ẋ ≡ v.

For the next step we would like to add the two equations, preferentially in
such a way that the right-hand side of the equality sign is equal to 0. So, we
multiply the equation for ẋ on both sides of the equality sign with αx and we
multiply the equation for v̇ on both sides with mv. The right hand side of
each of the equations from (a) now are the same but have an opposite sign.

(b) Show that the two equations from the system in (a) can then be rewrit-
ten into
d 1
(hint: x ⋅ ẋ = ( x2 ))
dt 2
d 1 2
( αx ) = αvx, (9.15a)
dt 2
d 1 2
( mv ) = −αxv. (9.15b)
dt 2

198
As said, we would like to add the two equations (9.15a) and (9.15b) to each
other.

(c) Verify that thus it is found that:

d 1 2 1 2
( mv + αx ) = 0. (9.16)
dt 2 2

(d) Explain that we can conclude that

1 2 1 2
mv + αx = E, (9.17)
2 2
where E is a positive constant.

(e) What is the dimension of this constant E? What is the physical mean-
ing of result equation (9.17)?

9.5.12 Conservation of energy 2 888

The analysis given in exercise 9.5.11 can be generalized as follows: Consider
the following second-order differential equation:
dU
mẍ = − , (9.18)
dx
where U (x) is some known function of x.

(a) Show that this equation can also be written as

ẋ = v, (9.19a)
1 dU
v̇ = − . (9.19b)
m dx

(b) Try to deduce, in a similar way as in exercise 9.5.11, that

1 2
mv + U (x) = E, (9.20)
2
dU dU dx dU
with E a constant once again. Hint: use ⋅ ẋ = ⋅ =
dx dx dt dt
199
9.5.13 The pendulum: energy balance 2 88
Consider the expressions for the kinetic energy Ekin and for the potential
energy Epot (θ) as derived in exercise 9.5.10.

(a) Give the expressions for the kinetic energy Ekin (θ) and for the potential
energy Epot (θ) (see exercise 9.5.10).

(b) Use a Taylor expansion to approximate the potential energy for small
excursions θ.

(c) Set up the equation that describes the energy balance of a pendulum
in terms of θ using the approximation.

(d) Compare the found energy balance to the energy balance of a spring
given by equation (5.37) on page 92 with V (x) = 21 αx2 . What are the
similarities and differences?

9.5.14 The Lotka-Volterra model 8

Argue why the Lotka-Volterra model, described by equations (5.51) - (5.52),
on page 97 is more realistic than for example the following population model:

dN1 ⎫
⎪
= α1 N1 − β1 N2 ⎪
⎪
⎪
dt ⎪
⎬ (9.21)
dN2 ⎪
⎪
⎪
= −α2 N2 + β2 N1 , ⎪
⎪
dt ⎭

9.5.15 Stationary Lotka-Volterra 8

Under certain conditions the population of preys (N1 ) and predators (N2 )
according to the Lotka-Volterra model, see equations (5.51)-(5.52), can be
stationary, i.e., the growth of both populations can equal zero.

(a) Determine the values of N1 and N2 for this equilibrium state.

(b) The same question (a), but now for the simplified Lotka-Volterra model
given by equation (9.21).

200
9.5.16 Model for disease epidemic 88
For modeling the evolution of a disease epidemic we can create a model based
on the Lotka-Volterra model given by equations (5.51)-(5.52) with α1 = 0:
dx ⎫
⎪
= −cxy ⎪
⎪
⎪
dt ⎪
⎬ (9.22)
dy ⎪
⎪
⎪
= +cxy − by, ⎪
⎪
dt ⎭
where
• x(t) = number of individuals in the population that are at risk of being
infected; y(t) = number of people that are infected at a time t;
• b = coefficient (> 0) that determines how fast y decreases as a result of
cure (or death) of infected people;
• c = coefficient (> 0) that describes how fast the infection spreads.

(a) Show that the number of infected people has to decrease when x(t =
0) < b/c.
(b) Show that when x(t = 0) > b/c the number of infected people increases
monotonically until x reaches the value of b/c. Give a physical expla-
nation for the observed behavior of the solutions x(t) and y(t).
(c) How could the model (9.22) be made more realistic?
(d) Nondimensionalize the evolution equations.

9.5.17 Water waves 8

You observe that a duck in a pond is moving up and down due to water on
the surface of the water. The distance from the highest point to the lowest
point is η and it takes time τ . You observe that the highest points are a
distance δ apart.
(a) Calculate the wave speed.
(b) Calculate the amplitude of the wave.
(c) Give a mathematical description of the wave.

201
9.5.18 Graphing transverse waves 8
A transverse wave on a rope has wave speed v, amplitude A and wavelength λ.
At time t = 0 and at position x = 0 the displacement is maximum downwards.
The wave is traveling in the positive x-direction.

(a) Give an expression for the frequency, period and wave number.

(b) Sketch the wave function as function of x at time t = 0, and indicate

the amplitude and wavelength in the graph.

(c) Sketch the wave function as function of t at position x = 0, and indicate

the amplitude and period in the graph.

9.5.19 Partial derivatives 8

Determine the partial derivative with respect to each of the independent
variables of the following functions:

(a) f (x, y) = y sin(px) + 3y, p is a parameter

(b) f (x, y) = sin(x) cos(x) cos(y)

√
(c) f (x, y, t) = x2 − t + exp(5y)

(d) f (x, y) = xy

(e) g(v, w) = v ln w

(f) h(k, l, m) = m sinh(km) + 2l3 m1 exp(−k)

9.5.20 The wave equation 8

Confirm the wave equation by taking both the second partial derivative to-
wards t and towards x of y(x, t) = A cos [ω ( xv − t)] and filling this in in
equation (5.58) on page 102.

202
9.5.21 Wave speed on a rope 888
Use figure 5.13 on page 103 to show that

(a)

∂y
F1y = T ( ) , (9.23)
∂x x
and
∂y
F2y = T ( ) . (9.24)
∂x x+∆x
Hint: the ratio Fiy /T should be equal to the magnitude of the slope of
the rope in that point.
F 2 y(x,t)
(b) Show that ∆xy is equal to T ∂ ∂x2 for very small ∆x by using the
equations above.

9.5.22 Wave on connected ropes 888

In order to create a longer rope, three pieces, each with length L, are con-
nected to each other. The long rope now has a length of 3L. Each of the
pieces of rope have a different mass. Piece one has mass m1 , piece two
m2 = 4m1 and piece three m3 = m1 /4. The entire rope is under tension F .
Calculate the time t it takes for a transverse wave to travel over the entire
rope.

203
9.6 Exercises chapter 6
9.6.1 Graphical analysis of an autonomous first-order system 88
Consider the system
dx
= sin(x).
dt
(a) Sketch the direction field for this system for −π ≤ x ≤ π.

(b) Draw the solutions x(t) for initial conditions x(0) = 0, and x(0) = π/2.

(c) What is the dimensionality of the phase space of this system?

(d) Sketch the phase line for this system, indicate the stationary points
and use arrows to illustrate the behavior of the system away from the
stationary points.

9.6.2 Graphical analysis of a non-autonomous first-order system

88
Consider the system
dx
= −x2 + t.
dt
(a) Sketch the direction field of this system on the domain 0 ≤ t ≤ 4,
−2 ≤ x ≤ 2.

(b) Draw the solutions x(t) for initial conditions x(0) = 0, x(0) = 1 and
x(0) = 2.

(c) What is the dimensionality of the phase space of this system?

204
9.6.3 Phase space 8
Reformulate the following second-order differential equations into a set of
first-order autonomous differential equations:
d2 θ g
(i) + sin θ = 0
dt2 l
d2 x
(ii) + ω2x = 0
dt2
d2 x k dx
(iii) + + ω2x = 0
dt2 l dt
d2 x
(iv) + ω 2 x = a cos(Ωt)
dt2
d2 x dx
(v) 2
+k = at2
dt dt

9.6.4 Phase diagram of an autonomous second-order system 1 8

Consider the linear oscillator of equation (5.27) with the solution (5.29) for
x(t) (pages 89 and 89). The corresponding solutions for ẋ(t) and ẍ(t) are
given by equation (5.30) on page 89.

(a) Now construct the x, ẋ phase diagram by determining the values of x

and ẋ at the times t = 0, T /8, T /4, 3T /8, T /2, 3T /4, T (where T = 2π/ω
is the oscillation period), and plotting these in the x, ẋ-graph.
(b) Construct the x, ẍ phase diagram in the same way.

9.6.5 Phase diagram of an autonomous second-order system 2

888
Consider the linearized pendulum equation (5.12) on page 85. The harmonic
solution to this equation describes a closed curve in the θ, θ̇ phase plane. We
are going to construct the trajectories in the phase plane directly from the
evolution equation, without using the exact analytical solution.
(a) Prove that the trajectories in the θ, θ̇-plane have the shape of an ellipse.
To this end, follow method A described in §6.1.3.
(b) In the θ̇, θ̈ phase plane the evolution of the system is also described by
a closed curve. Determine the shape of this trajectory.

205
 (c) What does the evolution of the solution look like in the θ, θ̈ phase
plane?

9.6.6 Conservation of energy in the phase diagram 88

Verify that the result of exercise 9.5.11 determines the phase portrait as
shown in figure 6.6 on page 116.

9.6.7 Phase diagram of an autonomous second-order system 3

88
Consider the system

ẍ = ω 2 x. (9.25)

(a) What is the dimensionality of the phase space of this system?

(b) Rewrite the system (9.25) into a set of first-order equations.

(c) Show that the trajectories in the phase plane of this system are given
by hyperbolas; i.e. show that the shape of the trajectories is given by

x21 x22
− = 1, (9.26)
a b
and determine the constants a and b.

(d) Sketch the phase diagram and use arrows to indicate the direction of
the trajectories.

9.6.8 RLC circuit 888

Consider an RLC circuit, with R the resistor, L the inductor and C the
capacitor. The equation for the electrical current I in the circuit is given by
I
LI¨ + RI˙ + = 0,
C
with L, R, C > 0.

(a) Rewrite the system into a set of first-order differential equations.

206
 (b) Roughly sketch the trajectories in the phase plane, in a similar way as
what is done in figure 6.5 in §6.1.3.

(c) Determine the exact shape of the trajectories for the case where R = 0.

(d) Argue whether this system will show periodic oscillations or not.

9.6.9 Numerical integration of a non-autonomous first-order sys-

tem 8
In this exercise we will study the numerical integration of the first-order
system
dx
= f (x, t)
dt
for the following cases:

1. f (x, t) = λ,

2. f (x, t) = λt,

3. f (x, t) = λx,

4. f (x, t) = λxt,

where λ is a constant.

(a) Determine the analytical (exact) solutions to these differential equa-

tions.

(b) Determine the solutions numerically, by hand, using the explicit Euler
integration method, for three time steps ∆t. To this end, take ∆t = 1,
λ = 1, and initial condition x(0) = 0.

(c) Compare the numerical solutions to the exact solutions for the same
time steps by tabelling and plotting the values (see the example in table
6.1 and figure 6.8). To this end, take λ = 1, x0 = 1 and use a time step
∆t = 1.

207
9.6.10 Numerical integration of a second-order system 88
In this exercise we will numerically integrate the mass-spring system given by
equation (6.8), which is a second-order, autonomous system. At a first glance
it may not be clear how to numerically integrate this system using Euler’s
method, following the integration rule given by equation (6.25). However,
we can use a trick:

(a) Rewrite equation (6.8) into a set of first-order differential equations for
the state variables x1 (position) and x2 (velocity).

(b) Write down the expressions for position and velocity in the next time
step tn+1 in terms of the values in the previous time step tn using Euler’s
explicit method.

(c) Tabulate the numerical values of x1 and x2 for the first three time steps
∆t. To this end, take ω = 1, ∆t = 1, x1 (0) = 1, and x2 (0) = 0.

208
9.7 Exercises chapter 7
9.7.1 Taylor series 1 8
The Taylor series of a function f around x = 0 is given by (recall §5.2)
N
1 (n)
f (x) ≈ ∑ f (0) ⋅ xn . (9.27)
n=0 n!

We now want to evaluate the Taylor series of f in a point x = a.

(a) Perform a coordinate transformation x′ = x − a and Taylor expand your

new function f (x′ ) according to equation (9.27).

(b) Compare your result to equation (7.18).

9.7.2 Taylor series 2 8

Consider the function

f (θ) = sin(θ). (9.28)

We want to Taylor expand this function around an arbitrary point θe .

(a) Explain why the expansion f (θ) ≈ θ is incorrect.

(b) Give the first two terms of the correct expansion of f around θe .

(c) Give the first two terms of the expansion of f around θ = 0 and θ = π.

(d) Make a sketch of f (θ) on the domain [0, 2π]. In your sketch, add the
two linearizations of f derived under (c).

9.7.3 Stability analysis of a first-order system 88

Consider the system
dx
= x2 − 1. (9.29)
dt
(a) Determine the stationary points of this system.

(b) Sketch the (x, ẋ) phase diagram.

209
 (c) Using the phase diagram, determine whether the stationary points
found under (a) are stable or unstable.

You will now perform a linear stability analysis to determine the stability of
each stationary point, to practice this procedure and confirm your answer to
(c). To this end, we will consider a small perturbation δx ≪ 1 around each
equilibrium point xe and determine whether this perturbation will grow or
shrink over time.

(d) Substitute the expression x = xe + δx into equation (9.29) and linearize

its right-hand side to find a linear equation for δx.

(e) Solve the resulting equations for δx (you will have one equation per
stationary point) and analyze the result: does the small disturbance
grow or shrink over time?

(f) What do you hence conclude about the stability of each stationary
point?

9.7.4 Linear stability analysis of a second-order system 88

Consider the system

θ̈ = sin (θ − 1). (9.30)

You will analyze the stability of the stationary points of this system using a
linear stability analysis. To this end, perform the following steps:

(a) Determine the stationary points on the domain θ ∈ [0, 2π].

(b) Consider a small perturbation δθ ≪ 1 around each stationary point

θe , by substituting θ = θe + δθ into equation (9.30) and linearizing its
right-hand side.

(c) Solve the resulting equations for δθ and analyze your results: does the
disturbance grow over time or not?

(d) What do you hence conclude about the stability of each stationary
point?

210
9.7.5 Non-linear pendulum 888
(a) Rewrite the pendulum equation (7.16) into a set first-order equations.
What is the dimensionality of the phase space for the pendulum?

(b) Determine the stationary point(s) and make a sketch of pendulum in

this/these point(s).

(c) Why is it physically irrelevant to mention points beyond the domain

[0, 2π⟩?

(d) Sketch the shape of the trajectories in the (x1 , x2 )-phase diagram.
Clearly mark the location(s) of the equilibrium point(s). Add arrows
to indicate the directionality of the curves.

(e) Using your plot, evaluate the stability of the equilibrium point(s).

(f) Why are these points also called elliptic and hyperbolic points do you
think?

(g) Using the results of the linear stability analysis of the pendulum equa-
tion in §7.2.3, namely equation (7.23) and (7.25), determine the exact
shape of the trajectories in the phase space close to the two equilibrium
points.

9.7.6 The Lotka-Volterra population growth model 1 88

In this exercise we will consider again the Lotka-Volterra model discussed in
§5.6, which was given by two first-order equations

Ṅ1 = α1 N1 − γN1 N2
(9.31)
Ṅ2 = −α2 N2 + γN1 N2 .

In this exercise we will graphically analyze the stability of the stationary

points of the Lotka-Volterra model.
(a) Determine the stationary points of the model (9.31).

(b) What is the dimensionality of the phase space for this system?

(c) Draw the (N1 , N2 )-phase plane and mark the stationary points (with-
out trajectories).

211
 (d) Sketch the directionality of the paths that lie on the N1 - and N2 -axes.
(e) Indicate, by roughly drawing arrows with the right direction, the tra-
jectories around the stationary points.
(f) From the direction of the trajectories in phase space, what do you
conclude about the stability of the stationary points?

9.7.7 The Lotka-Volterra population growth model 2 888

In this exercise we will analyze the stability of stationary points of the Lotka-
Volterra model using a linear stability analysis.
(a) Consider small perturbations around the stationary points (N1,e , N2,e )
determined in the previous exercise and derive evolution equations for
the perturbations δN1 , δN2 . To this end, substitute the expressions
N1 (t) = N1,e + δN1 (t),
(9.32)
N2 (t) = N2,e + δN2 (t),
into the Lotka-Volterra equations (9.31).
(b) Show that the equations obtained under (a) can be linearized to find
˙ 1 = (α1 − γN2,e )δN1 − γN1,e δN2 ,
δN
(9.33)
˙ 2 = (−α2 + γN1,e )δN2 + γN2,e δN1 .
δN

(c) Simplify equations (9.33) by filling in the stationary points and solve
them.
(d) From your results under (c), what do you conclude about the stability
of the stationary points? Compare this conclusion to your results of
the graphical stability analysis in the previous exercise.
(e) Show that the solutions starting from the vicinity of the stable sta-
tionary point show oscillations around this point with a period T =
2π/(α1 α2 )1/2 .
(f) Derive expressions for the shape of the trajectories for the perturba-
tions δN1 , δN2 in the phase diagram, and demonstrate why one of the
stationary points is also called an elliptic point, while the other is called
a hyperbolic point.

212
 (g) Modify the excel file [Link] for the Lotka-Volterra model
with α1 = α2 = 1 and γ = 0.1. Verify what has been found in (a) - (c) by
performing simulations, e.g. with initial conditions (N1 (0), N2 (0)) =
(9,9), (7,7), (2,2), etc. Also give a physical explanation for the observed
behavior of the system.

9.7.8 Mass-spring system 888

Consider a mass-spring system as depicted in figure 5.5(a), where x(t) is the
(time dependent) displacement of the mass (from its rest position x = 0),
m is the mass, and α is the spring constant. Apart from the linear spring
force, also an external force F (t) is acting upon the mass. This external force
depends on the displacement of the mass according to:

F (t) = β [x(t)] ,
2

where β is a positive constant.

(a) Show that the motion of the mass is governed by

mẍ = −αx + βx2 . (9.34)

√
(b) Using t̃ = ωt, with ω = α/m, as dimensionless time, derive the follow-
ing dimensionless form of equation (9.34)
d2 x̃
= x̃(x̃ − 1), (9.35)
dt̃2
where x̃ is the dimensionless displacement of the mass.

(c) Determine the stationary points of this (dimensionless) equation (9.35)

and investigate their stability.

(d) Give an expression for the (dimensionless) kinetic energy of the mass.
From hereon we discard the tildes in equation (9.35) and rewrite this equation
as follows
dV
ẍ = − . (9.36)
dx
(e) Determine the function V (x) assuming that V (x = 0) = 0.

213
(f) Sketch the function V (x) and show from this plot that equation (9.35)
has two stationary points. Can you judge from the plot of V (x) which
of these stationary points are stable and which are unstable? Motivate
your answer.

(g) Multiply all terms in equation (9.36) with ẋ and then rewrite the result
dt = 0, where E is the sum of kinetic
into the energy conservation form dE
and potential energy.

214
9.8 Exercises chapter 8
9.8.1 Bifurcation diagram 1 88
Consider the system (8.2).

(a) What are the stationary points of this system for r < 0, r = 0 and r > 0?

(b) Graphically analyze the stability of these equilibrium states (see chap-
ter 6).

(c) What is the dimensionality of the phase space for this system?

(d) Construct the phase diagrams of the system for a case where r < 0,
r = 0 and r > 0.

(e) Draw the bifurcation diagram of this system. Indicate the stability of
each branch by using dashed lines for unstable and solid lines for stable
branches.

9.8.2 Bifurcation diagram 2 88

Consider the system
dx
= rx − x2 . (9.37)
dt
(a) Determine the equilibrium points of this system.

(b) Plot ẋ) versus x for r < 0, r = 0 and r > 0 and use this plot to construct
the phase lines for these three cases.

(c) Using your sketches, determine the stability of each equilibrium point
for the cases r < 0, r = 0, r > 0.

(d) Draw the bifurcation diagram for this system. Indicate the stability of
each branch by using dashed lines for unstable and solid lines for stable
branches.

215
9.8.3 Bifurcation diagram 3 888
Consider the following nonlinear, first-order system

ẋ = rx − sinh (x), (9.38)

where r is a positive constant and sinh (x) = 12 {exp (x) − exp (−x)}.

(a) For 0 ≤ r < 1 this system has only one stationary point, namely x = 0.
Investigate the stability of this equilibrium point using a linear stability
analysis.

(b) Show that for r > 1 there are three equilibrium points one of which
still is x = 0. Investigate the stability of all three equilibrium points
by graphical analysis. (Hint: Plot the two functions on the right-hand
side of equation (9.38) in one figure and evaluate the sign of ẋ as a
function of x).

(c) Show that the system described by equation (9.38) undergoes a bifur-
cation at x = 0 as r is varied and plot a bifurcation diagram.

9.8.4 Catastrophe and hysteresis 88

Consider again the bucking strut of figure 8.1 as discussed in §8.2;

(a) Roughly sketch the bifurcation diagram of the strut. In other words:
plot the excursion amplitude xa of the strut as a function of the com-
pression force F . Mark the bifurcation point Fc .

(b) Consider the situation where the strut has an imperfection, such that it
is more likely to bend towards the left when compressed than towards
the right. Modify the bifurcation diagram accordingly.

(c) Now the strut is compressed with a force F > Fc but we give it a push
such that it is forced to bend towards the right(i.e. its unfavorable
state). From this bent state, we slowly decrease the compression force
back to zero. Sketch the path taken by the strut in the bifurcation
diagram.

(d) Explain (using your previous answers) why the imperfection of the strut
allows for hysteresis in this system.

216
9.8.5 The non-linear pendulum 1 88
Consider the non-linear pendulum described by equation (5.3) in §5.2, and
imagine that we give it an initial angular velocity Ω from its initial position
θ = 0.

(a) Write down the initial conditions that correspond to this situation.

(b) Introduce the dimensionless time t̃ = t/(l/g)1/2 and formulate the prob-
lem (pendulum equation + initial conditions) in a dimensionless form.

(c) Show that the equation can be written as a system of two first order
differential equations:
dθ
θ̇ = = y,
dt̃
dy
ẏ = = − sin θ,
dt̃

(d) Using the chain rule we write:

dθ dθ/dt̃
= = ⋯.
dy dy/dt̃

1 2
Show that after integration we get: y = cos θ + constant.
2
(e) Determine the constant by applying the initial conditions.

(f) Consider the situation θ → π (pendulum pointing upwards).

√ Show that
the pendulum can go through this point when Ω > 2 g/l.

9.8.6 The non-linear pendulum 2 8

Consider again the dimensionless pendulum equation derived in the previous
exercise. In this exercise we will use the file [Link] to investigate
the properties of the non-linear pendulum.

(a) Study the motion of the pendulum for an initial excursion θ0 = 3.124139 (=
179○ ). Compare the oscillation period with those for the cases θ0 = 178○
and θ0 = 179.5○ .

217
Now suppose that the pendulum is subject to a friction force. This friction
proportional to the angular velocity and therefore given by Ff = kdθ/dt.
(b) Show that by including the friction force non-dimensional pendulum
equation becomes
θ̈ + k̃ θ̇ + sin θ = 0 (9.39)
and determine k̃.
(c) Using the initial conditions θ(0) = 0, θ̇(0) = Ω̃, with Ω̃ = Ω(l/g)1/2 ,
investigate the influence of the damping term on the motion of the
pendulum using the excel file.

9.8.7 Limit cycles 1 88

(a) Can a limit cycle occur in systems with a one-dimensional phase space?
Why (not)?
(b) Can a limit cycle occur in a linear system? Why (not)?

9.8.8 Limit cycles 2 888

The trajectory of a point in the Cartesian x, y-plane is governed by the
following evolution equations
ẋ = x(1 − x2 − y 2 ) − y, (9.40)
ẏ = y(1 − x2 − y 2 ) + x. (9.41)
(a) Reformulate the above system in terms of polar coordinates. To this
end, make use of the relation between the Cartesian coordinates (x, y)
and the polar coordinates (r, θ), and rewrite the equations in the form
ṙ = F (r, θ), θ̇ = G(r, θ) and determine the functions F (r, θ) and G(r, θ).
(b) Determine the stationary point(s) of the equation for ṙ.
(c) Determine the stability of the stationary points using a graphical anal-
ysis.
(d) Using your previous answers, argue why any solution of equation (9.40,
9.41) eventually tends to a limit cycle in the (x, y)-phase space.
(e) What is the shape and location of this limit cycle?

218
9.8.9 The Van der Pol oscillator 8
Investigate the behavior of the Van der Pol oscillator – described by equa-
tion (8.5) in §8.5.2 – by making a modified version of the excel file Lotka-
[Link]. Investigate the behavior of equation (8.5) for different values
of ε. Use for example the input values x(0) = 0, ẋ = 0.01 and the time step
dt = 0.01. Is the final oscillation very sensitive to the initial conditions x(0)
and ẋ(0)? How does the oscillation period T depend on ε? NB: upon in-
creasing values of ε the size of the time step dt will have to be decreased
in order to still obtain accurate results. Check how for ε = 10 the period T
depends on the time step dt.

9.8.10 The forced cubic oscillator 88

The forced cubic oscillator is given by equation (8.6) in §8.6.1. The oscillator
is chaotic and sensitive to the starting conditions, see figure 8.18.

(a) Give a physical interpretation of each of the terms of equation (8.6).

(b) Convert equation (8.6) into a system of autonomous 1st order differen-
tial equations.

(c) Give an explanation for the strong dependence on the starting condi-
tions.

(d) In figure 9.14 different numerical solutions are shown for different start-
ing conditions. Describe the different behaviors.

(e) Investigate the behavior of this system by using the excel file pendu-
[Link] (you have to modify the file yourself).

9.8.11 The forced Van der Pol oscillator 88

Consider the forced Van der Pol oscillator:

ẍ + ε(x2 − 1)ẋ + x = A cos ωt, (9.42)

with A and ω the amplitude and frequency of the forcing.

(a) Give a physical interpretation of each of the terms in equation (9.42).

219
Figure 9.14: (x, t)-diagram obtained numerically with NXTWAIT for differ-
ent starting conditions, with k = 0.08, Ω = 1 and A = 0.2.

(b) Convert equation (9.42) into a system of autonomous 1st order differ-
ential equations.

(c) Describe the behavior of the forced Van der Pol oscillator seen in figure
9.15, 9.16 and 9.17: explain which of the figures shows a limit cycle, a
(potential) strange attractor or period doubling (i.e. a periodic motion
where the oscillation period doubles).

9.8.12 Lorenz equations 888

Consider the Lorenz equations (8.8) as discussed in §8.6.2. In this exercise
we will analyze the stationary points of this system.

(a) Show that (xe , ye , ze ) = (0, 0, 0) is a stationary point.

(b) Determine the other stationary points (if any) of the Lorenz equations
for r < 1.

(c) Determine the other stationary points (if any) of the Lorenz equations
for r > 1.

220
Figure 9.15: Behavior of the forced Van der Pol oscillator for A = 1, ϵ = 0.5
and ω = 1. Starting conditions are x(0) = 0.01 and ẋ(0) = 0.

Figure 9.16: Behavior of the forced Van der Pol oscillator for A = 1, ϵ = 0.5
and ω = 0.5. Starting conditions are x(0) = 0.01 and ẋ(0) = 0.

From a linear stability analysis around the stationary state (xe , ye , ze ) =

(0, 0, 0) we can find that the perturbations (δx, δy, δz) around this stationary

221
Figure 9.17: Behavior of the forced Van der Pol oscillator for A = 1, ϵ = 0
and ω = 0.5. Starting conditions are x(0) = 0.01 and ẋ(0) = 0.

point behave according to

δx(t) = c1 eλ1 t + c2 eλ2 t ,

δy(t) = c3 eλ1 t + c4 eλ2 t ,
δz(t) = c5 e−bt ,

with c1 to c5 some constants, and

1 √
λ1,2 = {−(1 + σ) ± [(σ + 1)2 − 4σ(1 − r)]} ,
2
with σ, b and r positive constants. Similar to Lorenz, we will take σ = 10 and
b = 8/3 in this exercise, and study the influence of r.

(d) Using these results, determine the stability of the stationary point
(0, 0, 0) for r < 1 and for r > 1.

As discussed in §8.6.2, the other stationary points will remain stable up to

r = rc = σ(σ+b+3)
(σ−b−1) .

(e) Based on your answers above, sketch the (r, xe )-bifurcation diagram
for the Lorenz equations for 0 < r < rc .

222
A References and recommended literature

A.1 Bibliography

References
[1] D.J. Acheson – From Calculus to Chaos. Oxford University Press, https:
//[Link]/oclc/38062718

[2] O. Reynolds – An Experimental Investigation of the Circum-

stances Which Determine Whether the Motion of Water Shall
Be Direct or Sinuous, and of the Law of Resistance in Par-
allel Channels. The Royal Society Publishin, [Link]
org/stable/109431, [Link]
[Link]/doi/10.1098/rspl.1883.0018.

[3] B. van der Pol – On relaxation oscillations. Phil. Mag. 2, 978 - 992 (1926),
[Link]

[4] P. Martien, S.C. Pope, P.L. Scott & R.S. Shaw – The chaotic behaviour
of the leaky faucet. Phys. Lett. 110 A, 399-404 (1985), [Link]
org/10.1016/0375-9601(85)90065-9

[5] H. Aref – Stirring by chaotic advection. J. Fluid. Mech. 143, 1-21 (1984),
[Link]

[6] Georg-Johann, Mercator series, (2010), obtained from https:

//[Link]/wiki/File:Mercator_series.svg#
/media/File:Mercator_series.svg, accessed 14/02/2019.

A.2 Recommended literature

In this course, we have focused on
• Elementary oscillations of linear and non-linear systems;

• Instability, bifurcations, and catastrophes;

• Non-linear oscillations and chaos.

223
A lot of the material has been inspired on the book
D.J. Acheson – From Calculus to Chaos
Oxford University Press (1997).

This book is recommended for those that want to read more about this sub-
ject matter!

Unfortunately, most books on this subject are quite specialized and not very
accessible for first year students, because of the (still) limited knowledge of
all mathematics involved. Books that give a nice and readable introduction
to non-linear systems and chaotic behavior are (in increasing order of degree
of specialization):

J. Gleick – Chaos
Abacus (1993) [this book has also been translated into Dutch]

E.N. Lorenz – The Essence of Chaos

UCL Press (1993)

G.L. Baker & J.P. Gollub – Chaotic Dynamics: an Introduction (2nd ed.)
Cambridge University Press (1996).

This is only a small selection from the vast amount of literature (both spe-
cialized and popular scientific) that has been published on this subject.
On the subject of ‘catastrophe theory’ a lot has been written as well. A nice
review paper (mentioned before in chapter 8) is:

E.C. Zeeman – Catastrophe Theory

Scientific American 234 (4), 65-83 (1976).

Slightly ‘classical’, but very readable.

There also are a lot of websites on the field of non-linear systems. There you
can find information on advanced research in this field, but also on the ele-
mentary principles of non-linear oscillations, bifurcations, chaotic dynamics,
etc.

224
B Information about [Link]

B.1 Introduction
The excel file [Link] is a tool for your study and enjoyment, that
uses the Euler method to numerically solve the following differential equation:

g
ẍ + k ẋ + sin x + A cos Ωt = 0 (B.1)
l
For many of you this equation will be recognizable as the differential equation
that describes a pendulum. If you do not recognize this equation, please have
another look at chapter 5.
This pendulum-model here is a rather elaborate version of the pendulum. It
is not linearized; it has some damping (the k ẋ-term) and it has a harmonic
driving force (the A cos Ωt-term), which can make it behave in very interesting
ways.
Having all these terms built in to this file may seem intimidating at first,
but it is ultimately quite helpful, because all the constants can be set freely,
enabling the study of a rich set of behaviors of this pendulum system.
Moreover, the freely setting of constants, also includes the freedom of setting
them to 0, which – of course – makes them drop out of the equation entirely,
putting us back into the world of a much simpler equation, whenever we desire
to relax ourselves by enjoying a simpler depiction of the physical reality.
So, this luxurious version of the equation does not force you to work with
a complex equation if you do not want to. It does, however, allow you to
switch on extra bits of physically relevant terms in the equation, to reveal
entire new worlds of exciting behavior.
The file provides immediate graphic output that visualizes the behavior of
the differential equation: It gives a plot of x against t, as well as ẋ and ẍ
against t. If you look closely, you will be able to see the more or less 90° phase
difference.
It also gives a phase diagram: ẋ versus x. And finally it gives a vector plot.
Play around with the settings and see what interesting stuff you can make
happen. See if it is numerical in origin or depicts ‘real’ physics.

225
B.2 A peek under the hood: setting starting values
When we take a closer look at the excel file at hand, at the very top of sheet
1 we first see 4 columns that are labeled t, x′′ , x′ and x. We will come back
to these columns in a bit.
Further to the right (column F onwards) we see relevant constants that are
given a numerical value. We will list all of them here:

• dt [s]: this sets the time step for the calculation. It needs to be small
enough to make the assumption (of linearity and stability) true16 .

• g [m s−2 ]: this sets the gravitational acceleration. This one is generally

set to 9.81, but becomes very interesting to play with for those who are
interested in extraterrestrial pendula.

• l [m]: this sets the length of the pendulum.

• k [rad s−1 ]: This sets the strength of the damping.

• x’(0) [s−1 ]: This sets the starting value for ẋ(0).

• x(0) [rad]: This sets x(0).

• Forceterm A: Sets the strength of the harmonic forcing.

• Forceterm Ω: This sets the angular frequency of the harmonic forcing

term.

B.3 A peek under the hood: calculating

Now that all these values are set, we can use the Euler method to calculate
where the system wants to go (and will go) given a known position. And
since we have just defined the starting position, why not start there?
We will use the previously given equation in a slightly modified form:
g
ẍ = −k ẋ − sin x − A cos Ωt (B.2)
l
16
I know that some aspiring physicists then promptly decide that the best way to proceed
is to make dt super super small, just to be extra safe. For those, I will add that small
steps mean many steps and therefore, longer calculation time. Anyway, something about
balancing the need for a quick answer and the need for an accurate answer. The joy of
engineering. But, do play around with this setting to see its effect.

226
This form of the equation allows us to calculate ẍ(0), i.e. ẍ at time t = 0
using nothing but the constants as defined in the previous subsection. And
since for t = 0 we now have x , ẋ and ẍ, we might as well put them all in a
table. This is exactly what can be seen in the first numerical row (so really
the second row) of the first 4 columns of sheet 1.
Now, how does one get from the first numerical row to the second numerical
row? That is very easy, if you are willing to assume that everything is linear,
which we are, because we have made dt small enough for that to be true. So,
what does that linearity mean? Well, it means that we can use the Euler
method:

ẋ(t = dt) = ẋ(t = 0) + ẍ(t = 0) ⋅ dt (B.3)

and that

x(t = dt) = x(t = 0) + ẋ(t = 0) ⋅ dt (B.4)

These two equations require only information that we already have and thus
allow us to fill the second numerical line of the table. And since we now have
the second numerical line of the table filled, we can use the same procedure
to fill the third numerical line. And so on, and so forth, until we have had
enough17 .
Excel has some neat tricks to fill tables with values based on values in pre-
vious rows. It seems therefore that the program Excel was made for doing
calculations like these, but it wasn’t. I highly recommend you to take a look
at how the formulas in this excel file implement this calculation as described
in this manual.

B.4 Visualizations
Now that we have filled our (by now enormous) table, we want to assess the
information that is hidden in these many many numbers. To do that, we will
represent this information graphically, because our brains are much better at
recognizing patterns in graphs than in tables. What graphs are most helpful
for your purpose, is for you, the engineer, to decide.
17
I had enough after 10 000 lines, as I am sure some of you will have seen by now. Feel
free to expand the table as far as you like.

227
For your convenience, this file provides a number of graphs: of x against t,
as well as ẋ and ẍ against t.
While these graphs are interesting, there are – perhaps – better ways to see
what is happening to this system. First, there is a phase diagram, which plots
ẋ against x, which makes a lot of sense in this particular physical context,
because the potential energy scales with x2 and the kinetic energy scales with
ẋ2 .
The sum of those two energies then corresponds to the mechanical energy of
the system, which is conserved (constant) if there is no damping or driving.
That means that in that case we expect to see an ellipse in the phase diagram.
The phase diagram is also very helpful to see if the motion of the system
repeats itself or not.
The final graph that is provided for your convenience is the vector plot (direc-
tion field).18 The vector plot shows the direction in which the system wants
to progress at various coordinate points in the x-ẋ-space. By following the
arrows, you can draw the phase diagram that would result from any given
starting point in the x-ẋ-space. Play with the starting values to see how the
curve looks in the vector plot.
If you modify this excel file to model the behavior of other exciting differential
equations, you may well arrive at a point where you need other visualizations.
Even for this system, there may be situations where you want to make some
aspect of the behavior visible that are not currently covered by the provided
graphs. Have fun exploring the possibilities.

18
Vector plots are very difficult to make in Excel, but with some trickery it worked.

228
C Greek alphabet

We often use Greek letters in physics and mathematics. Therefore, it can be

useful to memorize the Greek alphabet, see table C.

Table C.1: Greek alphabet.

Alpha αA Nu νN
Beta βB Xi ξΞ
Gamma γ Γ Omicron o O
Delta δ∆ Pi πΠ
Epsilon ϵ ε E Rho ρϱP
Zeta ζ Z Sigma σΣ
Eta ηH Tau τ T
Theta θ ϑ Θ Upsilon υ Υ
Iota ιI Phi ϕφΦ
Kappa κK Chi χX
Lambda λ Λ Psi ψΨ
Mu µM Omega ωΩ

229