You are on page 1of 95

Mathematical Modelling 1

Paul Papatzacos
Stavanger University
Last updated: August 17, 2005
Contents

Part 1. Continuous media 1

Chapter 1. Introduction to continuous media 3


1.1. Definition of a continuum 3
1.2. Kinematics in a continuum 6
1.3. Forces in a continuum 8
1.4. The tensor notation 13
1.5. The Reynolds transport theorem 14
1.6. The Du Bois-Reymond lemma 16
Problems 16

Chapter 2. The conservation laws 19


2.1. Introduction 19
2.2. Conservation of mass 19
2.3. Conservation of momentum 20
2.4. Conservation of energy 22
2.5. Summary: The Laws of Conservation in differential form 24
2.6. The Laws of Conservation in integral form. Applications 24
Problems 27

Chapter 3. Application: some elementary traffic problems 29


3.1. An elementary model of traffic flow 29
3.2. Solution of the first order wave equation 30
3.3. Two applications of the traffic flow model 37
Problems 39

Chapter 4. Fluid models 43


4.1. Ideal fluids 43
4.2. Viscous fluids – the Navier-Stokes equations 44
Problems 48

Part 2. Dimensional analysis and applications 57

Chapter 5. Measurements, units, dimensions 59


5.1. Introduction. The SI system of units 59
5.2. Concerning secondary quantities 62
5.3. On the arbitrariness of primary quantities 70
5.4. Some specialized systems of units 72
Problems 74
Appendix: Derived SI units with special names 75

Chapter 6. The Pi-theorem 77


6.1. Unit-free equations 77

Part 3. Well-posed problems with second order PDEs 69


iii
iv CONTENTS

Chapter 6. Standard equations and well-posed problems 71


6.1. Introduction 71
6.2. Start and boundary conditions 72
6.3. Well-posed problems/not well-posed problems 74
6.4. Hadamard’s example of a problem which is not well-posed 75

Chapter 7. The Laplace equation 77


7.1. Properties 77
7.2. Well-posed problems with the Laplace equation 79
7.3. Extensions 80
7.4. Example 1. Electrostatic shield 80
7.5. Example 2. Conductive sphere in an electrostatic field 81
Problems 85

Chapter 8. The wave equation 89


8.1. Linear acoustics 89
8.2. The wave equation in one-dimensional space 92
8.3. Uniqueness and stability theorems 93
8.4. The balloon problem 95
Problems 100

Chapter 9. The diffusion equation 105


9.1. Thermal conduction in solids 105
9.2. General theorems 106
9.3. A well-posed problem without initial conditions 108
9.4. Kelvin’s calculation of the age of the Earth 109
9.5. Some comments on propagation by diffusion 111
Problems 112

Part 4. Calculus of variations 117

Chapter 10. Bernoulli’s problem in variational calculus 119


10.1. Bernoulli’s brachistochrone problem 119
10.2. The fundamental lemma of the calculus of variations 120
10.3. The Euler-Lagrange equation 121
10.4. Bernoulli’s problem 123
10.5. Natural boundary conditions 124
10.6. Variational problems with several dependent variables 125
Problems 126

Chapter 11. Isoperimetric problems 129


11.1. Maximum enclosed area: formulating the problem 129
11.2. An elementary variational problem with a constraint 130
11.3. Other isoperimetric problems 132
Problems 134

Chapter 12. Several independent variables 137


12.1. Hamilton’s Principle 137
12.2. Variational problems with two independent variables 138
12.3. The vibrating string 140
12.4. Generalized solutions 140
Problems 146

Reminders and extensions 149


CONTENTS v

Chapter 13. Reminders and extensions 151


13.1. Differentiation under the integral sign 151
13.2. Problems on extrema with constraints 151
13.3. The divergence theorem and some extensions 152
13.4. Proof strategies 153
Bibliography 157
Part 1

Continuous media
CHAPTER 1

Introduction to continuous media

After introducing the concept of ’a continuum’ this chapter will conclude with
two theorems: the Reynolds transport theorem and the Du Bois-Reymond lemma.

1.1. Definition of a continuum


We shall, in chapters 1, 2, and 4, look at a method of modelling a simple fluid
(gas or liquid). “Simple”, in the previous sentence, means that the molecules of
the fluid are electrically neutral, approximately spherical and that they do not react
chemically with each other. The resulting model describes the fluid in a way that
is detailed and precise enough to cover all needs in, say, industrial applications.
The model is based on the laws of physics governing the mass, momentum, and
energy of a system, and it contains one major assumption, namely that the fluid
is a continuous medium (also called a continuum). This is called the continuum
assumption (or, sometimes, the continuum hypothesis). We shall sometimes call
the resulting model the continuum model.
The continuum asumption is essential because it allows the use of theorems of
mathematical analysis. However, fluids consist of molecules in a state of perpetual
chaotic motion. Modelling a fluid as a continuum then raises some important
questions. We shall call them question one, question two, and so on, for ease of
reference.
Question one: what is the definition of a continuum? We have assumed that
the concept is intuitively understandable, but it could be useful to have a definition.
Answer to question one: a medium is a continuum if (i) given a point P
belonging to the medium at time t, and (ii) given any closed surface S surrounding
P at time t and not touching a solid boundary, then all the points inside S at time
t belong to the medium.
Note that the closed surface mentioned in this definition is virtual. Note also
that any closed surface in usual three-dimensional space contains an infinite num-
ber of points. The definition states that all these points must belong to the medium,
which implies that the continuum is an abstraction: no medium is a continuum
since any closed surface taken in a real medium at any given time will contain
infinitely many points that are in the vacuum between molecules.
Let us state, before formulating the next question, how we shall use the ab-
straction of a continuum. We shall consider the continuum as a material system
obeying the laws of physics and, to describe its evolution in time we shall intro-
duce two quantities, ρ and v, depending on space and time, and defined as density
and velocity. These two quantities will then lead to the definitions of mass and
momentum for a region of the continuum. The application of the laws of classi-
cal mechanics for mass and momentum to such a region will then lead to a set of
equations which make the calculation of ρ and v possible. As we shall see, ther-
modynamics will allow the calculation of a number of other characteristics of the
continuum, like its pressure, temperature, etc. We shall thus be able to calculate,
at any place and at any time, all the quantities characterizing the continuum, thus
3
4 1. INTRODUCTION TO CONTINUOUS MEDIA

making the model applicable to a number of fields: the design of airplanes, cars,
buildings, etc.
Question two: in view of the fact that the real fluid consists of molecules, can
a model of a fluid as a continuum be at all valid?
Answer to question two: we can not know before we have tried, but it is
worth building a continuum model because it is reasonnable to assume that it will
be valid in cases of flow around objects whose linear dimensions are extremely
large compared to the linear dimensions of molecules and to the distances be-
tween molecules. Let L be a length characterizing the flow around an object. (In
the case of an airplane, one would choose L to be the thickness of the wing. In the
case of flow in a pipe, L is the diameter of the pipe.) For the simple fluids we are
considering, the average molecular distance ℓ is large as compared to the molec-
ular “diameter”. We can thus express a first condition of validity as the following
inequality:

L ≫ ℓ. (1.1)

Note that ℓ can be obtained when one knows the chemical composition of the
fluid and its mass per unit volume, but a quick estimate follows from the fact that
a cubic centimetre of any gas, at the usuall conditions of everyday life, contains
about 1019 molecules. This gives a volume per molecule equal to 10−19 cm3 , and
taking twice the cube root of this one gets an average distance between molecules
of about 10−6 cm. In practice, we accept that equation (1.1) is satisfied when L is
larger than about hundred to a thousand times ℓ.
Question three: in view of the fact that the real fluid consists of molecules
in chaotic motion, in what way does the description of the continuum provide a
description of the real fluid?
Answer to question three. (We shall use considerable more space answering
this question than the previous ones.) It is essential, in order to answer the question,
to keep in mind the extremely large molecular density quoted above. In statistical
studies, the behavior of a population can usually be understood by looking at a
few quantities that are averages of individual quantities. We thus consider a real
fluid and define a density and a velocity at a point Q and at a time t by averaging
molecular masses and momenta over a large number of molecules. It may be
helpful to imagine that the averaging is done by an instrument, and to think of
such an instrument in general terms, as something that interacts with the fluid over
a certain region in space. We shall call this region the averaging region, and we
can think of it as approximately spherical or cubical, having a volume V and being
centered at the point Q. The size of the averaging region can then be characterized
by just one length which is of the order of V 1/3 . We introduce coordinate axes
and we call x the vector from the origin to Q. Then the real-fluid density ρr f and
real-fluid velocity vr f are defined as follows:

1 X
ρr f (x, t) = mi , (1.2)
V
V
1 1 X
vr f (x, t) = m i vi , (1.3)
ρr f (x, t) V
V

P
where m i is the mass of the i -th molecule, vi its velocity and V means that we
sum the masses and momenta of all the molecules that are inside the averaging
region at time t.
A number of remarks are in order.
1.1. DEFINITION OF A CONTINUUM 5

• Equations (1.2) and (1.3) assume that molecules can have different mas-
ses, so that the definitions apply to cases where we have a mixture of
different chemical components.
• Nothing has been stated yet about the size of V , although it is assumed
that it is in some sense “large enough” so that the averaging region con-
tains a large number of molecules. A clarification of this most important
point will be given presently.
• Fluid velocity is defined indirectly, namely through momentum, and us-
ing the definition of density. According to equation (1.2), the total mass
inside the averaging region is ρr f V . We define the velocity vr f (x, t)
by saying that the total momentum in the averaging region is ρr f V vr f .
Expressing
P the total momentum as the sum of the molecular momenta
m
V i i v we then get equation (1.3). The reason for such an indirect
definition is that there is a law regulating the change of momentum, not
the change of velocity. It is then important to have a definition of ve-
locity that takes proper care of possible changes in momentum in the
averaging region. Note that such a subtle definition is only necessary if
we have a mixture of molecules with different masses. If all molecules
have the same mass, then equation (1.3) leads to vr f being equal to the
average of the molecular velocities.
We now turn to the size of V . It is reminded that the molecules of a fluid are in a
perpetual state of movement and mutual collisions. An important concept in this
context is the mean free path λ. It is the average distance travelled by a molecule
between two successive collisions. If we can approximate the collisions between
molecules by the collision between spheres of diameter d then an approximate
formula for λ is [18]
1
λ= √ ,
2πnd 2
where n is the number density. Taking the atmosphere as an example, then one
can use d ≈ 3 × 10−10 m, while n depends on the height above sea level. Typical
values are λ ≈ 6 × 10−8 m at sea level and 20◦ C, and λ ≈ 0.1 m at an altitude of
about 100 km.
The linear dimension V 1/3 of the averaging region must be much larger than
the mean free path to ensure that most molecules undergo a large number of colli-
sions so that their momenta are effectively randomized:
V 1/3 ≫ λ. (1.4)
(For a further discussion of randomization, see the answer to question four below.)
Inequality (1.4) gives a lower bound for V 1/3 . There is an obvious upper bound:
the averaging region must be much smaller that the solid objects around which the
flow takes place. Somewhat more precisely, the linear dimension of the averaging
region must be much smaller than the length L introduced in the answer to question
two:
V 1/3 ≤ L. (1.5)
We see that inequalities (1.4) and (1.5) are both satisfied if and only if
L ≫ λ. (1.6)
We can now complete the answer to question three. The description of the con-
tinuum provides a description of the real fluid which is such that the details of the
chaotic molecular motion are eliminated by the averaging procedures of equations
(1.2) and (1.3).
Question four: are there any criteria that one can use to judge the applicability
of the model in a given situation involving a real fluid?
6 1. INTRODUCTION TO CONTINUOUS MEDIA

Answer to question four: there are two criteria, inequalities (1.1) and (1.6).
The first inequality makes sure that the averaging volume contains a large number
of molecules, so that averaging is statistically meaningful. The second inequality,
inequality (1.6) is more subtle. It originates from inequality (1.4), which is justi-
fied by the necessity to randomize. This necessity itself comes from the fact that
we are going to make a field theoretical model. This is, by definition, a model
where the quantities characterizing it (density, velocity, . . . , in the present case)
are determined locally, in both space and time. This means that, for example, the
density is calculated at x and t, as a function of x and t exclusively: in particular,
ρ(x, t) does not depend on what happened at earlier times. One usually expresses
this by saying that the system has no memory. Randomization is essential in eras-
ing memory as the following example shows. Immediately after reflecting off a
wall, a molecule has a velocity pointing away from the wall, so it can be said to
“remember” the position of the wall: it is behind it. After colliding four or five
times with other molecules, its velocity points in any direction with a probabilty
that is almost independent of the direction, and the memory of the position of the
wall is lost.

1.2. Kinematics in a continuum


The term kinematics denotes the description of motion without referring to its
cause. We have seen in the previous section that the continuum model is a field
theory, meaning that all quantities in the model depend on space and time alone,
i.e., they are locally determined. We shall first look at some practical conventions
about notation, especially concerning vectors. Then we shall define what we mean
by a material particle in a continuous medium.
1.2.1. Notation: axes, coordinates, vectors. Any model or theory contains
a certain number of symbols for the dependent and independent variables in the
model. In the continuum model, we shall need symbols for coordinates in space,
for time, density, velocity, energy, and so on. Some of these quantities are so-called
scalars (time, density, . . . ) while others are vectors (velocity, force of gravity,
. . . ). The choice of symbols is dictated by tradition (density is almost universally
denoted by ρ) and by practical considerations, often dictated by the requirement
that the meaning of a symbol must be easy to remember.
We begin with introducing the notation that will be used almost consistently in
these lecture notes for the space coordinates. Coordinates depend on two choices:
the origin, and the orientation of the axes. Both choices depend on the problem one
x3 is considering. For a general exposition such as this one, and for the building of
the continuum model, origin and orientation are arbitrary, the only usual restriction
being that the coordinate system should be right-handed.
The origin being chosen we shall call the three unit vectors i1 , i2 , i3 . We
break with the tradition of calculus books where these vectors are denoted i, j, k,
i3 because it is wasteful of letters. Further and most importantly, the three coordinate
axes having i1 , i2 , i3 as unit vectors will be called the x 1 -axis, the x 2 -axis, and the
i2 x2 x 3 -axis. See figure 1.1. Here again, the departure from the traditional x, y, z is
explained by the need to minimize the number of letters used.
i1
For vectors and their components we shall use the following rule: the compo-
x1 nents of a vector A on the unit vectors i1 , i2 , i3 are denoted A1 , A2 , A3 .
Figure 1.1: Notations for unit vectors and co- As a consequence of the above, we say that an arbitrary point P in space
ordinate axes. has coordinates x 1 , x 2 , x 3 . We denote the vector beginning at the origin of the
coordinate system and ending at P by x. The velocity of the fluid is traditionally
denoted v. Its components are denoted v1 , v2 , v3 . The dependence of v on space
and time is made explicit in various ways, depending on context, on the degree of
1.2. KINEMATICS IN A CONTINUUM 7

explicitness we want to achieve, and on the space we have available on the page.
The most compact way to indicate this dependence is
v = v(x, t).
The most detailed is
v1 = v1 (x 1 , x 2 , x 3 , t)
v2 = v2 (x 1 , x 2 , x 3 , t)
v3 = v3 (x 1 , x 2 , x 3 , t).
We return to indexes and their uses in section 1.4. Symbols are introduced and
defined in these lecture notes as the need arises, and we shall try to keep to the
conventions exemplified above.
1.2.2. Particles in a continuum. According to the presentation in section
1.1, we shall make a model where the real fluid is idealized as continuous matter,
and we shall concentrate on a field theoretical description. The reason for a field
theory is that it is the easiest to obtain and that it answers directly most questions
that are asked about the fluid in practical applications. We shall now look at one
question, the answer to which is mostly useful to our intuition, but which is not
directly answered by the model.
Suppose we perform a thought experiment where we put a tiny amount of
color in a small neighborhood around a point in the continuum (say inside a radius
of 0.1 mm). We would like to follow this colored matter in time. To be more
specific, we assume that we have a continuum model, that we can simulate the
coloring of a small part of the continuum around some point at time zero, and that
we can solve the model and thus obtain density and velocity as a function of space
and for all times larger than zero. Can we say where the colored matter is at some
time larger than zero? The answer is no, not directly. That is because we have a
field theoretical description and we only know what value the velocity has at each
point and each time. In other words, we know how the matter moves, but not where
it moves.
To answer the question of where matter moves we have to perform additional
calculations on the results given by the model. Suppose then that we know v(x, t)
for all relevant x and t. To follow the center of the colored speck in time we have
to introduce a set of time-dependent coordinates for this point. We denote them
x 1 (t), x 2 (t), x 3 (t). The velocity of the point is then a vector with components
d x 1 /dt, d x 2 /dt, d x 3/dt. But the velocity is also given by v1 , v2 , v3 . We can then
obtain the movement of the center of the colored speck by solving the following
set of coupled first order differential equations:
d x1
= v1 (x 1 , x 2 , x 3 , t)
dt
d x2
= v2 (x 1 , x 2 , x 3 , t) (1.7)
dt
d x3
= v3 (x 1 , x 2 , x 3 , t),
dt
with
x 1 (0) = ξ1
x 2 (0) = ξ2 (1.8)
x 3 (0) = ξ3 ,
giving the position of the center of the speck at t = 0. Assuming that equations
(1.7) and (1.8) have a solution, we can visualize this solution as a curve in space,
8 1. INTRODUCTION TO CONTINUOUS MEDIA

parametrized by the values of time (see figure 1.2). We can then interpret this
curve as the path of a material point particle.
In fact, we can define a material particle in a continuum as the point whose
path is given by the solution of equations (1.7) and (1.8). Some important details
must be checked, ensuring that the particle we have defined in this manner has the
properties that a classical material particle should have. Most important among
these is that its path is continuous and that it is uniquely determined by the initial
position. We shall not go into this subject here but refer the reader to the literature
on systems of differential equations (and to the literature on dynamical systems).
See for example Braun [7], where it is shown that the desired properies exist if the
Figure 1.2: Particle path. functions v1 , v2 , v3 are differentiable.

1.3. Forces in a continuum


In chapter 2 we will apply physical laws to a fluid with the objective of find-
ing equations which describe the motion of the fluid in general. These laws are
mathematical expressions for the conservation of three central quantities in an ar-
bitrary1 physical system: mass, momentum, and energy. In order to apply these
laws to a fluid, it is expedient to consider a finite volume of the fluid, thus creating
a well-defined physical system to which conservation laws can be applied.
The formulation of the laws of conservation requires that all forces acting on
the physical system are known. There are two types of force to be considered for
a system consisting of a fluid inside a finite volume: volume forces and surface
forces. Volume forces originate from outside the system, gravity being the most
common example. Surface forces are internal. They act on the surface of the
system and are due to the surrounding fluid.
This section (that is section 1.3) is divided into three subsections. Subsection
1.3.1 looks in more detail at the technical means of limiting, and consequently
defining, a physical system where the laws of conservation can be applied. Sub-
section 1.3.2 looks at the volume forces, and subsection 1.3.3 at the surface forces.
1.3.1. Auxiliary surfaces. Fluids are normally characterized by the fact that
they adopt the shape of the container. In the mathematical expression of a given
three-dimensional problem the container will be modelled as one or more outer
surfaces subjecting the fluid to given conditions (so-called boundary conditions).
We will be returning to these conditions later. At the present stage of modelling
it is useful to introduce what we call auxiliary surfaces. These may be of two
different types: Closed or open.
• Closed surfaces are introduced when we want to define a limited quantity
of the fluid as a physical system.
• Open surfaces are introduced when we want to study internal forces in a
fluid. Since intermolecular forces exist, we must expect the presence of
forces between different parts of a fluid. This requires the introduction of
surfaces to separate the fluid into well-defined parts, and the subsequent
identification of the forces acting between those parts.
It is important to note that these auxiliary surfaces do not actually exist. Therefore,
they do not hinder the free movement of the fluid as physical boundaries would.
Like other mathematical aids, these surfaces disappear when they have served their
purpose: we will see that they do not appear in the final equations which describe
the fluid.
1It is important to note here that the fluids we will be observing during this course are not quite
that arbitrary. For instance, they do not contain free electrical charges. The presence of charged parti-
cles would require us to apply the law of charge conservation.
1.3. FORCES IN A CONTINUUM 9

1.3.2. Volume forces. Consider that part of the fluid which, at time t, is
within a closed surface S and has volume V . There exist forces, such as grav-
ity, which originate outside the fluid itself and which are of long range. These
forces act, therefore, at all points in the fluid and are called volume forces. We
will use the most common volume force, that is gravity, to derive a generalR expres-
sion. Referring to figure 1.3 we can see that the fluid Rwithin S has mass V ρ d V .
Consequently, the force acting on this volume is −i3 g V ρ d V . For a large volume
(a part of the atmosphere, for example) we must take into account that gR is depen-
dent on position so that a more correct expression for the force is −i3 V ρg d V .
By denoting the components of the force as Fi and by introducing the vector g
with components
g1 = 0, g2 = 0, g3 = −g, (1.9)
we can write Z Figure 1.3: Gravity as an example of volume
Fi = ρgi d V. (1.10) force.
V
We see that ρgi is the force of gravity per unit volume. The resulting volume force
can be written in the general form
Z
Fi = fi d V, (1.11)
V
where fi is force per unit volume. The Coriolis force and electromagnetic forces
are other examples of volume forces. See exercise 4.12.
1.3.3. Surface forces. We are now going to look at the internal forces in
a fluid, that is forces caused by interactions within the fluid itself. We will be
looking at their origins and properties.
1.3.3.1. Stress in a fluid. We introduce a small auxiliary surface d S and ob-
serve how the fluid on the one side of d S interacts with the fluid on the other side.
We find two different types of interaction:
• One type of interaction is due to inter-molecular forces. Two electrically
neutral and spherically symmetrical molecules attract each other with a
force which acts along the line between the centers of mass and which
has a very short range.2
• Another type of interaction is due to the transfer of momentum, caused
by molecules crossing the surface d S. The resulting forces are also of
short range.
We will call both forces described above particle forces. Both have short range,
and Newton’s third law applies to both of them.3
Regarding the mathematical modelling of the force resulting from the interac-
tion of the fluid on the one side of d S with the fluid on the other side, one adopts
the following approach. Let P be a point in the middle of d S (see figure 1.4) and
let n be a unit normal to d S at P. Let D1 denote the fluid in the vicinity of P, on Figure 1.4: Notations used in the definition
that side of d S to which n points, and let D2 denote the fluid on the other side. of surface forces.
D1 exerts a force dK on D2 equal to the resultant of the particle forces around
d S. Since the particle forces have a very short range, consider an ideal situation
and assume that the range equals zero. It follows that the resultant dK involves
2This force varies proportionally to r −7 , where r is the distance between the centers of mass, as
long as r is somewhat greater than the molecular diameter. When the distance between the centers of
mass decreases the force will become repellent and vary like r −13 . Intermolecular forces of this kind
are often called van der Waals forces. They are due to the electric charges of the electrons and occur
when the electrically neutral molecules come close to each other and act as electric dipoles.
3We remind the reader that Newton’s third law states that k = −k where k is the force
ab ba ab
acting on particle b from particle a, and kba is the force on particle a from particle b.
10 1. INTRODUCTION TO CONTINUOUS MEDIA

particles which are in d S, and is thus proportional to d S: dK = t d S. Vector t is a


stress, that is a force per unit area.
This means that the stress is a local property of the fluid, in the same way
as its local density or velocity. From the way it was introduced we arrive at the
following definition.
D EFINITION 1.1 (Stress in a fluid). Given a surface element d S in a fluid,
with unit normal n at a point P on d S, the stress t at P across d S is equal to the
local force per unit area that the fluid which is on the side of d S where n points,
exerts on the fluid which is on the other side.
It is natural to assume that the stress depends on the position x of the point
P, on time t, and on n. This assumption was first expressed by Cauchy and is
therefore called Cauchy’s principle. The dependence on x and t is implied and is
never included in the expression, while the dependence on n is usually indicated:
we write t(n) to denote the stress exerted by the fluid which is on the side where n
points, on the fluid which is on the opposite side, so that (see figure 1.4):
t(n) d S = the force which D1 exerts on D2 ,
t(−n) d S = the force which D2 exerts on D1 .
Since the particle forces satisfy Newton’s third law it follows that the stress is an
odd function of the unit normal:
t(−n) = −t(n). (1.12)
We are now able to write a general formula for the resultant surface force:
Z
K i = ti d S, (1.13)
S
where ti is the i -th component of t(n) in an arbitrarily chosen Cartesian coordinate
system.
We will now look at the most significant property of the stress and at its con-
sequences.
1.3.3.2. The principle of local equilibrium of stresses. We will demonstrate
that the stresses are in local equilibrium even when the fluid is in motion. To
make this statement more precise we first need to write the law of conservation of
momentum for a fluid. This law states that the derivative with respect to time of
an object’s momentum is equal to the sum of all the forces acting on the object.
To apply this law to a fluid we introduce an auxiliary surface S. This is closed and
contains a quantity of fluid, with volume V , which we will now treat as an object.
n
t(n) dS The momentum can easily be found by considering the volume within S as
consisting of infinitesimally small partial volumes d V . See figure 1.5. At a point
dV
dS P in the middle of d V the density is ρ and the velocity v such that the momentum
rvi dV of the fluid in d V is ρvd V , and the momentum of all the fluid within S is a vector
fi dV P with components Z
S
Pi = ρvi d V.
Figure 1.5: Part of a fluid within a closed sur- V
face S. d S is an infinitesimally small part of S Regarding the forces acting on that part of the fluid within the surface S, we already
and d V is an infinitesimally small part of V . know that there is a volume force given by formula (1.11) and a surface force given
by formula (1.13). The law of conservation of momentum is therefore:
Z Z Z
d
ρvi d V = f i d V + ti d S. (1.14)
dt V V S
If V has characteristic length L, then we can write
V = k1 L 3 , S = k2 L 2 ,
1.3. FORCES IN A CONTINUUM 11

where k1 and k2 only depend on the shapes of V and S (they are in other words
independent of L). The left-hand side of (1.14) is therefore proportional to L 3 as
is the first term on the right-hand side, while the second term on the right-hand
side is proportional to L 2 . Dividing both sides by L 2 and letting L → 0 gives
Z
1
lim ti d S = 0.
L→0 L 2 S

This equation shows that the surface forces are in local equilibrium at all times
since, less precisely, it states that
Z
ti d S = 0 when S is infinitesimally small. (1.15)
S

We are now going to look at the consequences that this local equilibrium has for
the stress t.
1.3.3.3. The stress tensor. Consider equation (1.15) written for a specific sur-
face S, constructed in the following manner: At an arbitrary value of time, we
consider a point O in the fluid and introduce a Cartesian coordinate system with
origin at O. We choose the points A, B, and C on the axes so that the distances
O A, O B, and OC are infinitesimally small. The tetrahedron OABC (see figure
1.6) has an infinitely small characteristic length and we can write equation (1.15)
where S is the union of the four surfaces O BC, O AC, O AB, and ABC. Making
use of the fact that the spatial dimensions are infinitely small, we ignore t’s varia-
tion from place to place within the tetrahedron. Since time is also a given constant,
t(n) is the only dependence we need to consider.
Consequently, we can say that the fluid outside the tetrahedron subjects the
fluid on the inside to the following stresses:
• t(−i1 ) across surface O BC,
• t(−i2 ) across surface O AC,
• t(−i3 ) across surface O AB, and
• t(n) across surface ABC.
Note that n is the unit normal to surface ABC, pointing out of the tetrahedron.
Denoting the areas O BC, O AC, O AB, and ABC as A1 , A2 , A3 , and A, and
using property (1.12), we can write equation (1.15):

At(n) − A1 t(i1 ) − A2 t(i2 ) − A3 t(i3 ) = 0.


Figure 1.6: Stresses in a tetrahedron.
With the notation
n = n 1 i1 + n 2 i2 + n 3 i3 ,
we see that
A1 = An 1 , A2 = An 2 , A3 = An 3 ,
so that

t(n) = n 1 t(i1 ) + n 2 t(i2 ) + n 3 t(i3 ). (1.16)

We now introduce the Cartesian components of t(i1 ), t(i2 ) and t(i3 ):

t(i1 ) = σ11 i1 + σ12 i2 + σ13 i3 ,


t(i2 ) = σ21 i1 + σ22 i2 + σ23 i3 , (1.17)
t(i3 ) = σ31 i1 + σ32 i2 + σ33 i3 ,
12 1. INTRODUCTION TO CONTINUOUS MEDIA

where each σi j is a given function of x and t. The substitution of (1.17) in (1.16)


gives:

t(n) = (n 1 σ11 + n 2 σ21 + n 3 σ31 )i1


+ (n 1 σ12 + n 2 σ22 + n 3 σ32 )i2
+ (n 1 σ13 + n 2 σ23 + n 3 σ33 )i3 .

This means that the coordinates of t(n), defined by

t(n) = t1 i1 + t2 i2 + t3 i3 ,

can be expressed as

t1 = n 1 σ11 + n 2 σ21 + n 3 σ31 ,


t2 = n 1 σ12 + n 2 σ22 + n 3 σ32 , (1.18)
t3 = n 1 σ13 + n 2 σ23 + n 3 σ33 .

We have in other words arrived at the significant result that the stress in a medium,
across a surface element with unit normal n, is given by a linear transformation
of n. The following matrix-form of equations (1.18) is perhaps a more familiar
illustration of the linear transformation in question:
 
  σ11 σ12 σ13
t1 t2 t3 = n 1 n 2 n 3 σ21 σ22 σ23  . (1.19)
σ31 σ32 σ33

The quantities σi j are the components of the stress tensor σ . In the mathematical
literature equation (1.19) is often written

t = n · σ. (1.20)

We shall see in section 1.4 that another way to write equations (1.19) and (1.20) is

t j = n i σi j . (1.21)

This is the so-called component version.


It is worth noting that little can be said about σi j without taking a closer look
at the substance we are modelling. In chapters 2 and 4 we will see that in many
situations we can use
   
σ11 σ12 σ13 −p 0 0
σ21 σ22 σ23  =  0 − p 0  , (1.22)
σ31 σ32 σ33 0 0 −p

where p is the pressure in the fluid. In such cases the stress t acts along the unit
normal n, since inserting (1.22) in (1.21) gives

t = − pn. (1.23)

Substances are today classified into two sets: the ”non-polar” for which σi j = σ j i
and the ”polar” for which this symmetry does not apply (see also chapter 2, in the
section containing equation (2.9)). In chapter 4, we will be looking at important
examples of non-polar fluid models and will then show how to derive σi j . Solids
have to be treated separately, for which we refer to books on the theory of elasticity
or mechanics of materials.
1.4. THE TENSOR NOTATION 13

1.4. The tensor notation


The quantities σi j are the components of a tensor. Tensors are mathematical
quantities having well-defined properties which will not be discussed in detail here.
The reader is referred to, for example, [2]. However, it is useful to familiarize
ourselves with some simple rules and symbols.
Models involving vectors and tensors will normally exhibit expressions like
equation (1.21), which looks like the scalar product of two vectors a : (a1 , a2 , a3 )
and b : (b1 , b2 , b3 ):
3
X
a · b = a1 b1 + a2 b2 + a3 b3 = ai bi . (1.24)
i=1
Note that (1.24) can also be written
3
X
a·b= ajbj.
j =1

The index which is summed over is called the dummy index. In nearly all calcu-
lations involving tensors and vectors the dummy index appears twice in the same
term. The summation sign is therefore superfluous and can be omitted. Equation
(1.24) is thus written
a · b = ai bi .
This is the summation rule (sometimes called Einstein’s summation rule): a re-
peated index indicates summation. This short formulation of the rule is often mis-
understood and it should be modified to: an index indicates summation whenever
it is repeated in the same term4 or at the same letter. Some examples will clarify
this. Consider the following expressions:
a i + b i + ci = d i , (1.25)
b j k dk + a j = M j , (1.26)
Riilm + Tklm Bk + Ulm = Vlmi Wi . (1.27)
The index i in expression (1.25) does not indicate summation. Even if it appears
four times, it is not repeated in the same term. Actually, expression (1.25) states
that the indicated equality holds for all values of the index i (where it is supposed
to be known that i can take the values 1, 2, 3 only). So that expression (1.25) is a
shorthand version of
a 1 + b 1 + c1 = d 1 ,
a 2 + b 2 + c2 = d 2 ,
a 3 + b 3 + c3 = d 3 ,
Indexes such as i in expression (1.25), j in expression (1.26), and l and m in
expression (1.27), are called free indexes. As we have just seen with the case of
expression (1.25), free indexes must, if the expression where they appear is to make
sense, appear once and only once in each term of the expression; in that case they
indicate that the expression is true for all the allowed values of the index . This
also means that
ai + bi + c j = di

4To avoid further misunderstandings, let us define the word “term”. It is an identifiable group in
an equation or expression that participates in an addition or a subtraction with other terms. Expression
A + BC − E/F + (J K )2 − L(M + N ) has five terms. The last term consists of two factors, the second
of which consists of two terms.
14 1. INTRODUCTION TO CONTINUOUS MEDIA

is meaningless, unless one is reasonably sure that c j is a typographical error, easily


corrected by changing j to i .
Expression (1.26) has one index indicating summation, namely k, and a free
index, namely j . It might be of some interest to the beginner to see the “un-
wrapped” version of the equation. The safest way to perform the unwrapping is
to do it in two steps, where the first step consists in performing the summation
explicitly:
b j 1 d1 + b j 2 d2 + b j 3 d3 + a j = M j .

The second step is to explicitly write the three equation resulting from the free
index j taking the value j = 1, j = 2, j = 3 in succession:

b11d1 + b12 d2 + b13 d3 + a1 = M1 ,


b21d1 + b22 d2 + b23 d3 + a2 = M2 ,
b31d1 + b32 d2 + b33 d3 + a3 = M3 .
Expression (1.27) has two summation indexes: i (in the first term on the left-hand
side and in the only term on the right-hand side) and k (in the second term on the
left-hand side). The expression has two free indexes, l and m, appearing once and
only once in each term. The unwrapping of this equation is left to the reader.
We now introduce two symbols which are of great help in calculations, namely
the Kronecker-symbol δi j , and the Levi-Civita-symbol ǫi j k . These are defined as
follows:

1 if i = j
δi j = (1.28)
0 if i 6= j,

 0 if two indexes are equal
ǫi j k = +1 if (i j k) is an even permutation of (123) (1.29)

−1 if (i j k) is an odd permutation of (123),
It can be shown that
ǫi j k ǫlmk = δil δ j m − δim δ j l . (1.30)
With the help of the Levi-Civita symbol one can write
(a × b)i = ǫi j k a j bk . (1.31)
The same symbol gives, for curl V
(∇ × V)i = ǫi j k ∂ j Vk , (1.32)
with the practical and space-saving notation

∂j ≡ . (1.33)
∂x j
For later use we introduce a corresponding notation for the time variable:

∂t ≡ . (1.34)
∂t
1.5. The Reynolds transport theorem
Reynolds’ transport theorem will serve as a useful tool when studying conser-
vation laws.
T HEOREM 1.1. Consider a continuous medium in motion. Let S(t) be a
closed, smooth surface at time t and let V (t) be the volume inside S(t), such
that S(t) and V (t) follow the flow of particles (see figure 1.7). Let φ(x, t) be a
property of the fluid and let v(x, t) be the velocity field of the fluid.
1.5. THE REYNOLDS TRANSPORT THEOREM 15

If φ and the components of v are continuous and differentiable with respect to


the arguments x 1 , x 2 , x 3 and t, then
Z Z  
d ∂φ
φ(x, t) d V = + ∇ · (φv) d V. (1.35)
dt V (t ) V (t ) ∂t
where n is the unit normal of S(t) at x, pointing outwards.
Proof (Direct proof, see section 13.4.1.) The proof below is taken from [23].
A somewhat more ”sophisticated” proof can be found in [2]. We apply the defini-
tion of the derivative:
Z
d
φ(x, t) d V
dt V (t ) Figure 1.7: The displacement of the fluid
Z Z  which is inside S(t), between times t1 and
1
= lim φ(x, t + 1t) d V − φ(x, t) d V . t2 .
1t →0 1t V (t +1t ) V (t )
By referring to figure 1.8 we can see that
Z Z Z Z
(· · · ) = (· · · ) − (· · · ) + (· · · ),
V (t +1t ) V (t ) I III
so that
Z n
d S(t)
φ(x, t) d V vDt
dt V (t ) dS
Z Z  I II III
1
= lim φ(x, t + 1t) d V − φ(x, t) d V dS vDt
1t →0 1t V (t ) V (t )
Z Z 
1 n S(t + Dt)
+ lim φ(x, t + 1t) d V − φ(x, t + 1t) d V .
1t →0 1t III I Figure 1.8: The displacement of the fluid
The first limit on the right hand side can be re-written as follows which is inside S(t), between times t and
Z Z  t + 1t (broken line). S(t) consists of the
1
lim φ(x, t + 1t) d V − φ(x, t) d V parts S1 (t) (thin line) and S2 (t) (thick line).
1t →0 1t V (t ) V (t )
Z The shaded area at the top right has volume
φ(x, t + 1t) − φ(x, t) d V = d S (v·n) 1t and the shaded area at the
= lim dV
V (t ) 1t →0 1t bottom left has volume d V = −d S (v · n) 1t.
Z
∂φ
= d V,
V (t ) ∂t
since φ is continuous and differentiable. This gives us the following preliminary
result:
Z Z
d ∂φ
φ(x, t) d V = dV
dt V (t ) V (t ) ∂t
Z Z 
1
+ lim φ(x, t + 1t) d V − φ(x, t + 1t) d V . (1.36)
1t →0 1t III I
When 1t is small, volumes I and III are small and the corresponding d V are as
follows (see figure 1.8):
For region III: d V = d S (v · n) 1t,
For region I: d V = −d S (v · n) 1t,
so that
Z Z
φ(x, t + 1t)
lim dV = φ(x, t) (v · n) d S,
1t →0 I I I 1t S (t )
Z Z1
φ(x, t + 1t)
lim dV = − φ(x, t) (v · n) d S,
1t →0 I 1t S2 (t )
16 1. INTRODUCTION TO CONTINUOUS MEDIA

where S1 (t) and S2 (t) are the parts of S(t) lying inside and outside S(t + 1t) ,
respectively, (see figure 1.8). With reference to equation (1.36) we can now see
that
Z Z 
1
lim φ(x, t + 1t) d V − φ(x, t + 1t) d V
1t →0 1t III I
Z Z
= φ(x, t) (v · n) d S + φ(x, t) (v · n) d S
S (t ) S2 (t )
Z 1
= φ(x, t) (v · n) d S
S(t )
Z
= ∇ · (φv) d V,
V (t )

where the last equality results from the application of the divergence theorem.
Equation (1.35) is thereby proved.
End of proof
In chapter 3 (section 3.2.4) we will be looking at a situation where formula
(1.35) cannot be used directly because φ is not continuous.

1.6. The Du Bois-Reymond lemma


The Du Bois-Reymond5 lemma is used in the next chapter to derive the laws
of conservation in their differential form.
T HEOREM 1.2. Assume a space V with n dimensions and a continuous func-
tion f (x) where the vector x has components x 1 , x 2 , . . . , x n . If D is an arbitrary
region in V and if
Z
f (x) d n x = 0 (1.37)
D
then f (x) ≡ 0 for all x.
Proof (Proof by contradiction, see section 13.4.2.) Assume that f 6= 0 at an
arbitrary point P0 , and that it is > 0 (or < 0). Since f is continuous, there is a
region D0 around P0 where f > 0 (or < 0). Since D is arbitrary, let us choose
D = D0 . We get
Z
f (x) d n x > 0 (or < 0).
D
This contradicts the assumption (1.37). Therefore f (x) is equal to zero for all x.
End of proof

Problems
Problem 1.1 In a continuous medium it is given that
x
v= ,
1+t
where v is the particle velocity, and that x = ξ when t = 0. Find the particle paths.
Problem 1.2 M is the mass of the earth and R its radius. G is the gravitational
constant. Find the acceleration which a particle is subjected to due to gravity,

5Those interested in the history of science will search in vain for two French scientists named
Du Bois and Reymond: Du Bois-Reymond is the name of one (and only one) German mathematician
(1831–1889).
PROBLEMS 17

as a function of height h above the surface. Give a simplified expression when


h/R ≪ 1. Numerical values:
M = 5.973 × 1024 kg,
R = 6.371 × 106 m,
G = 6.672 × 10−11 m3 s−2 kg−1 .
Problem 1.3 A mass of air with volume 106 m3 is at 60◦ north and is moving
northwards at 20 m/s. Assume the density of the air to be 0.5 kg/m3 and find
the Coriolis force. Assume there is a pressure gradient, positive in a north-south
direction and equal to 5 × 10−3 Pa m−1 , and assume that the mass of air is cubical
in form. Estimate the force due to the pressure gradient and compare it to the
Coriolis force.
Problem 1.4 An arbitrarily shaped solid object is immersed in a liquid through-
out which the velocity field is equal to zero. For such a case it has been experi-
mentally demonstrated that t(n) = − pn where p is the pressure (see also chapter
4). Assume that the liquid has constant density ρ. Choose a coordinate system
with origin at the surface of the liquid and with the x 3 -axis pointing downwards.
It is known that p = p0 + ρgx 3 (where p0 is the pressure at the surface and g the
acceleration due to gravity). Derive Archimedes’ law of buoyancy. See also [11],
exercise 20, page 796.
Problem 1.5 Assume that all indexes take values from the set {1, 2, 3}. Also
assume that the summation rule applies (indexes occurring twice in the same term
are summed from 1 to 3).
(1) One of the following two equations is meaningless. Which one?
ai j x j = b j , ai j x j = bi .
Decompress the correct equation.
(2) Decompress the following equations:
di j u j + ak Aik = Fi ,
∂ Rk
n j Ai j = G i + ai .
∂u k
Problem 1.6 Assume that the summation rule applies (indexes occurring twice
in the same element are summed from 1 to 3).
(1) Find the numerical values of δii , and ǫi j k δ j k .
(2) Given the vectors A and B with components Ai and Bi , (i = 1, 2, 3), what
are ǫi j k A j Ak and δi j Ai B j ?
(3) Show that
A × (B × C) = B(A · C) − C(A · B),
∇ · (A × B) = B · (∇ × A) − A · (∇ × B),
∇ × (∇ × A) = ∇(∇ · A) − (∇ · ∇)A,
A × (∇ × A) = ∇(A2 /2) − (A · ∇)A.
Problem 1.7 Derive the transport theorem in a space of one dimension.
Hint: First show that
Z x2 (t ) Z x2 (t )
d ∂φ d x2 d x1
φ(x, t) d x = d x + φ(x 2 (t), t) − φ(x 1 (t), t) .
dt x1 (t ) x 1 (t ) ∂t dt dt
Then show that this can be written
Z x2 (t ) Z x2 (t )  
d ∂φ ∂(φv)
φ(x, t) d x = + d x.
dt x1 (t ) x 1 (t ) ∂t ∂x
CHAPTER 2

The conservation laws

2.1. Introduction
In this chapter we will use the laws of physics to construct equations between
the various quantities which characterize a continuous medium. The physical laws
we shall use are primarily the conservation laws and it is important to note that
we shall restrict ourselves to classical physics. In particular, we shall not consider
the relativistic effects arising from the equivalence of mass and energy which are
significant at high speeds. Consequently, we are looking at media in which speeds
are low relative to the speed of light. It is also assumed that the medium carries no
electrical charge.
In the following sections we shall investigate a continuous medium in motion
and apply the laws of physics to a limited part of the medium, more precisely that
part contained within a closed surface S(t) with volume V (t), where S(t) follows
the flow (see figure 1.7). By applying the laws which express the conservation
of mass, momentum and energy, we will arrive at a set of equations with quite a
broad area of validity. In a later chapter these equations will be used to construct
models for gases and liquids.

2.2. Conservation of mass


We consider a fluid which consists of only one chemical component so that
there are no chemical reactions taking place: at no point in the medium is fluid
created or destroyed . The mass contained within S(t) therefore remains constant
so that Z
d
ρ(x, t) d V = 0, (2.1)
dt V (t )
where ρ(x, t) is the density of the medium (mass per unit volume). We now as-
sume that ρ and v are continuous and apply Reynolds’ transport theorem. Equation
(1.35) gives Z  
∂ρ
+ ∇ · (ρv) d V = 0.
V (t ) ∂t
Since V (t) is an arbitrary volume the Du Bois-Reymond lemma gives
∂ρ
+ ∇ · (ρv) = 0, (2.2)
∂t
which is often called ”the equation of continuity”. Note that using the tensor nota-
tions, and especially the notations (1.33) and (1.34), this can be written
∂t ρ + ∂i (ρvi ) = 0.
The source concept. Equation (2.1) presupposes that mass is neither created Figure 2.1: Example of the introduction of
nor destroyed in the medium. Source effects are, however, important in practice, mass across a boundary surface.
and we will now look at two ways of describing such effects.
Mass can be introduced or removed across a boundary surface. If there is a
boundary surface S (figure 2.1) and if mass is introduced or removed through S,
then (2.1) still applies inside the medium. The conditions at the surface itself can
19
20 2. THE CONSERVATION LAWS

then be taken into account by applying so-called boundary conditions. We will


return to this topic in chapter 6.
Mass can be introduced or removed through sources inside the medium. (It is
also possible for mass to be created or destroyed if the medium consists of several
chemical components which can react with each other, but we shall not consider
this case here.) Consider a system where a known amount of fluid with density ρ
is pumped in or out at a point x and at a time t, per unit volume and unit time: let
this quantity be called r (x, t) (r < 0 if mass is pumped out). The conservation of
mass for this system can then be expressed as follows:
Z Z
d
ρ(x, t) d V = r (x, t) d V.
dt V (t ) V (t )
It easily follows that (2.2) is replaced with
∂t ρ + ∂i (ρvi ) = r. (2.3)
In this equation, r is called the source term and ρvi is called the flow of density
term. We see that the flow of density is proportional to the velocity and, to empha-
size this fact, say that it is convective.

2.3. Conservation of momentum


Newton’s law for that part of the medium within S(t) can be written
Z Z Z
d
ρvi d V = fi d V + ti d S, (2.4)
dt V (t ) V (t ) S(t )
where f i and ti represent the components of volume force per unit volume and
surface tension respectively. We have seen in section 1.3.3.3, that ti = n j σ j i .
Equation (2.4) then becomes
Z Z Z
d
ρvi d V = fi d V + n j σ j i d S. (2.5)
dt V (t ) V (t ) S(t )
Applying the necessary assumptions of continuity, the left-hand side of the equa-
tion can be re-written using formula (1.35). With reference to the tensor notation
(section 1.4) we can write
∇ · (ρvi v) = ∂ j (ρvi v j ),
so that the left-hand side of (2.5) becomes
Z Z
d  
ρvi d V = ∂t (ρvi ) + ∂ j (ρv j vi ) d V. (2.6)
dt V (t ) V (t )
The surface integral on the right-hand side of (2.5) can be re-written using the
divergence theorem. Note that, for an arbitrary vector A with components A1 , A2
and A3 the divergence theorem is writen
Z Z
Ajnj dS = ∂ j A j d V,
S(t ) V (t )
so that Z Z
n j σji d S = ∂ j σ j i d V, (2.7)
S(t ) V (t )
by considering σ1i , σ2i and σ3i to be the components of a vector Ai . Applying
(2.6) and (2.7) to (2.5) gives
Z
 
∂t (ρvi ) + ∂ j (ρv j vi ) − ∂ j σ j i − fi d V = 0,
V (t )
and the Du Bois-Reymond lemma gives
∂t (ρvi ) + ∂ j (ρv j vi − σ j i ) = f i . (2.8)
2.3. CONSERVATION OF MOMENTUM 21

This is the differential equation for the law of conservation of momentum. Note
that we have, formally speaking, the same type of equation as in (2.3) and we can
interpret the quantities as follows:
ρvi = momentum density,
ρvi v j − σ j i = flow of momentum density,
f i = source of momentum density.
Note that the flow of momentum density consists of two elements of which only
one is convective: momentum ρvi is transported at velocity v j . The second el-
ement consists of the stress tensor σi j , which can thus be regarded as a non-
convective flow of momentum density.
It is also possible to write a conservation equation for angular momentum.
One can show that this equation results in
σi j = σ j i , (2.9)
if one assumes that changes in angular momentum can be attributed solely to the
momentum of volume and surface forces (see problem 2.1). This assumption ap-
plies for so-called ”non-polar” media and we will be looking at some examples of
such media in a later chapter. Polar media are not covered in this course.

The Stokes operator. The Stokes operator or the substantial derivative for a
medium in motion with velocity v(x, t) is defined as:
D ∂
= + v · ∇, (2.10)
Dt ∂t
or, with the symbols (1.33) and (1.34):
D
= ∂t + vi ∂i .
Dt
In this chapter we have used the spatial description (see section 1.1) and we can
easily demonstrate that operator (2.10) occurs when we differentiate with respect
to t along the path of a particle. Let E(x, t) (for the sake of simplicity index E
is omitted) be a space and time dependent property, and let us assume that we are
measuring the property along the path of a particle x = x(t). The time derivative
with respect to E along the path is
d ∂E ∂E d x i
E(x(t), t) = +
dt ∂t ∂ x i dt
∂E ∂E DE
= + vi = .
∂t ∂ xi Dt
We are now going to look at some of the operator’s properties. The proofs are left
to the reader.
• If ϕ(x, t) and ψ(x, t) are two arbitrary functions:
Dϕψ Dψ Dϕ
=ϕ +ψ . (2.11)
Dt Dt Dt
• If ϕ(x, t) is an arbitrary function and if ρ(x, t) satisfies the equation (2.2)
(no source term):

∂t (ϕρ) + ∂i (ϕρvi ) = ρ . (2.12)
Dt
The law of conservation of momentum can then be written
Dvi
ρ = fi + ∂ j σ j i . (2.13)
Dt
22 2. THE CONSERVATION LAWS

• By letting ϕ = ρψ in equation (1.35) this can be written


Z Z
d Dψ
ρ(x, t)ψ(x, t) d V = ρ d V, (2.14)
dt V (t ) V (t ) Dt
if ρ satisfies the equation (2.2).

2.4. Conservation of energy


2.4.1. Conservation of energy in general. In order to express conservation
of energy we have to take into account the internal energy of a fluid as defined in
thermodynamics. Some preparatory steps must however be taken before we can
apply thermodynamic concepts to fluid mechanics.
Firstly, internal energy is extensive: it increases as the size of the system
increases. There is no natural way of limiting the size of a fluid, especially a fluid
in motion. For this reason, all extensive variables for a fluid always refer to the
unit mass.
Secondly, internal energy is normally defined for a system in equilibrium,
that is, a system where all intensive variables such as temperature and density are
uniform through space. This is not the case for a fluid in motion, but it is still
possible to use general laws of physics to arrive at a meaningful definition of local
internal energy per unit mass. For example, we have no problems in defining the
density at x at t as the mass of fluid in a small volume 1V at x at time t, divided
by 1V . This definition is independent of whether the fluid is in equilibrium or
not — we simply need to ensure that 1V is small enough to allow us to ignore
any gradients in the density. By referring to the general law of conservation of
energy it is also possible to define the internal energy of 1V as the internal energy
measured after the fluid in 1V has been isolated and reached a state of equilibrium,
without it having exchanged mechanical or thermal energy. Other thermodynamic
variables, such as temperature and entropy per unit mass, can then be defined in
terms of density and internal energy per unit mass by using equations of state. See
reference [5], especially section 3.4 for a discussion of this.
Thermodynamics also tells us that two bodies with different temperatures ex-
change heat energy. Since variations, especially in temperature, will occur in a
fluid in motion, we can expect heat energy to flow from place to place in a fluid,
not necessarily convectively. In addition to the internal energy per unit mass E,
we therefore introduce a vector q to represent the energy flow due to heat transfer
in the fluid (energy per unit area and unit time; energy flowing into V (t) is taken
to be positive).
We can now express conservation of energy. The energy within a closed sur-
face S(t) consists of a kinetic part (v 2 /2 per unit mass) and a thermal part, E per
unit mass, and is equal to
Z  2 
v
ρ + E d V.
V (t ) 2
The integral form of the equation for the conservation of energy is therefore
Z  2  Z Z Z
d v
ρ + E dV = f i vi d V + ti vi d S − qi n i d S. (2.15)
dt V (t ) 2 V (t ) S(t ) S(t )
The first integral on the right-hand side represents the effect of the volume forces
while the second integral represents the effect of the surface forces. Here, ti vi =
n j σ j i vi . The negative sign in front of the final integral is due to the definition of
q above, and also to the definition of n as the unit normal which points out of the
volume.
2.4. CONSERVATION OF ENERGY 23

Applying the standard assumptions relating to continuity we can re-write the


left-hand side of equation (2.15), making use of equation (2.14), while the right-
hand side can be re-written using the divergence theorem. Using the Du Bois-
Reymond lemma, it then follows that
 
D v2
ρ + E = f i vi + ∂ j (σ j i vi − q j ). (2.16)
Dt 2
This is the differential form of the law of conservation of energy. We are now
going to look at some variants of this equation.
2.4.2. An equation of equilibrium for total energy. We set
v2
ε= + E, (2.17)
2
and re-write (2.16) by using (2.12):
∂t (ρε) + ∂ j (ρεv j − σ j i vi + q j ) = f i vi . (2.18)
This equation has the same form as (2.3) and (2.8): it is an equation of continuity.
We interpret the quantities in (2.18) as follows:
ρε = energy density,
ρεv j − σ j i vi + q j = flow of energy density,
f i vi = energy source.
2.4.3. An equation of equilibrium for internal energy. Kinetic energy is
a result of momentum, so we expect to be able to derive a law of conservation
of kinetic energy from the law of conservation of momentum, equation (2.13).
Performing a scalar multiplication of both sides of (2.13) by v, gives:
Dvi
ρvi = fi vi + (∂ j σ j i )vi . (2.19)
Dt
Equation (2.11), however, shows that
 
Dvi 1 D D v2
ρvi = ρ (vi vi ) = ρ .
Dt 2 Dt Dt 2
Further,
(∂ j σ j i )vi = ∂ j (σ j i vi ) − σ j i ∂ j vi ,
so that (2.19) can be written
 
D v2
ρ = f i vi + ∂ j (σ j i vi ) − σ j i ∂ j vi . (2.20)
Dt 2
This is the law of conservation of kinetic energy. We can see that equations (2.16)
and (2.20), give
DE
ρ = σ j i ∂ j vi − ∂ j q j , (2.21)
Dt
which is the law of conservation of thermal energy.
By using equation (2.12), equations (2.20) and (2.21) can be re-written as
equations of continuity. For example, the equation of continuity for thermal energy
becomes
∂t (ρ E) + ∂ j (ρ Ev j + q j ) = σi j ∂ j vi , (2.22)
which shows that σi j ∂ j vi is a source of thermal energy. This element, which is
the result of internal stresses within the medium, also appears as a source term in
equation (2.20), but with the opposite sign. In other words, the work of the internal
stresses reduces the kinetic energy and increases the thermal energy correspond-
ingly.
24 2. THE CONSERVATION LAWS

2.5. Summary: The Laws of Conservation in differential form


We have applied the laws of conservation of mass, momentum and energy,
and arrived at

∂t ρ + ∂ j (ρv j ) = 0, (2.23)
Dvi
ρ = fi + ∂ j σ j i , (2.24)
Dt
DE
ρ = σ j i ∂ j vi − ∂ j q j . (2.25)
Dt
Remember that the conservation of angular momentum gives σi j = σ j i . Given f i ,
σi j , and qi , we then have five equations for the five unknowns ρ, E, and vi . The
volume force f is usually known while σi j and qi are model dependent. Experi-
ence shows that in the most realistic liquid and gas models the σi j depend on the
pressure p and the velocity components vi , while the qi depend on temperature
T . (Explicit expressions for σi j and qi using p, vi , and T are called constitu-
tive relations.) This means that (2.23), (2.24), and (2.25) generally contain seven
unknowns, p, ρ, T , E, and vi , so we need two more equations. These are the
so-called equations of state

f ( p, ρ, T ) = 0, (2.26)
g(E, ρ, T ) = 0, (2.27)

taken from thermodynamics.


In practice, situations occur in which variations in temperature in the fluid can
be ignored. T then becomes a parameter only, and the energy equation (2.25) is
superfluous. Equations (2.23), (2.24) and (2.26) then form a complete set for the
five unknowns: p, ρ, and vi .
Finally, it should be noted that it is necessary to make use of additional laws
of conservation, constitutive relations and equations of state if the medium carries
electrical charges and/or consists of several chemical components.

2.6. The Laws of Conservation in integral form. Applications


In the preceding sections we have looked at the laws of conservation in differ-
ential form. The complete set of equations (see section 2.5) is quite complicated.
On the other hand, they provide a very detailed description of the flow. There
are many cases where we are not interested in a detailed description, but where
we rather have use for approximate values of certain quantities which we believe
to characterize a specific situation. For example, the quantity of liquid flowing
through a pipe per unit time, or the mean value of the frictional force acting on
an object in a wind tunnel. In these cases, a method using a so-called control sur-
face can be applied. This is normally a closed surface in the fluid, and the volume
within it is called the control volume. The control volume is virtual, that is, it
does not obstruct the flow: It is an abstract construction which allows us to exploit
simplified assumptions about the flow.
First, we will see how the conservation laws can be re-written by integrating
them over a control surface. Then, by using an example, we will see how the
integral equations can be used to calculate approximate results. We will confine
ourselves to situations where temperature changes are insignificant, so that we
only need to consider the equations for the conservation of mass and momentum
(see section 2.5). See references [19, 41] for an exposition which includes the
conservation of energy.
2.6. THE LAWS OF CONSERVATION IN INTEGRAL FORM. APPLICATIONS 25

Let S be a closed control surface and let V be the volume contained within it.
we use equation (2.3) for mass conservation, and integrate it over V :
Z Z Z
∂t ρ d V + ∂i (ρvi ) d V = r d V.
V V V
Since
R V does not depend on time, the first term on the left-hand side can be written
d
dt V ρ d V while the second term can be re-written using the divergence theorem.
This gives Z Z Z
d
ρ dV + ρv · n d S = r d V, (2.28)
dt V S V
where n is the unit vector of S, which points out of the volume. Similarly, inte-
grating the equation for the conservation of momentum gives:
Z Z Z Z
d
ρvi d V + ρvi v · n d S = fi d V + ti d S, (2.29)
dt V S V S
where (see section 1.3.3.3) ti = n j σ j i .
Equations (2.28) and (2.29) are equivalent to equations (2.3) and (2.8) but they
are more useful together with certain types of approximation. This is illustrated in
example 2.1 below and in problems 2.3–2.5. In order R to interpret these equations
let us first look at the left-hand side terms of type S J · n d S. In equation (2.28),
J is a flow of mass density, in equation (2.29), J is a flow of momentum density.
These elements represent so-called fluxes, that is, they give the quantity of mass
or momentum density which crosses surface S (per unit time). Note that, since
surface S is virtual, the flow enters through certain parts of the surface (namely
where J points in towards V ) while the flow exits through the other parts of the
surface (where J points out of V ). Since n always points out of V , J · n is positive
when the flow is out of V and negative when the flow is into V . It follows that the
last two equations above can be interpreted in the following way:
     
Time derivative Flow of mass Flow of mass
 of the mass in + out of the − into the =
the control volume control volume control volume
 
Mass added
 per unit time to  (2.30)
the control volume

for the conservation of mass, and


     
Time derivative Flow of momentum Flow of momentum
 of the momentum in + out of the − into the =
the control volume control volume control volume
 
Sum of all the forces
 acting on the medium  (2.31)
in the control volume

for the conservation of momentum. Note that the law of conservation of momen-
tum is a vector equation which must be expressed along specified axes.
Many applications look at so-called steady-state flows. These are types of
flow where the quantities which describe the fluid (density, velocity, . . . ) are in-
dependent of time. The derivatives with respect to time on the left-hand side of
equations (2.28)–(2.31) then drop out. The following example shows how these
equations can be applied.
E XAMPLE 2.1: Consider the steady-state flow of a liquid in a curved pipe
between cross-sections a and b (see figure 2.2). The control surface is
represented in the figure by the broken line. The axis of the pipe (XY
26 2. THE CONSERVATION LAWS

in the figure) is a curve in the (x 1 , x 2 )-plane. Let n be the unit normal


pointing out of the control surface, and more specifically, let na and nb
be the unit vectors at the cross-sections a and b. It is given that na has
components [−1, 0, 0] and that nb has components [cos θ, sin θ, 0]. It is
also given that the cross-sections have the same area, A.
We are going to find an approximate expression for the force which
the liquid exerts on that part of the pipe which is between cross-sections
a and b, based on the following assumptions.
• The stress t(n) (see section 1.3.3) is given by
t = − pn.
We will see in chapter 4 that this is often a useful assumption in
cases where viscosity effects can be ignored. p represents the pres-
sure of the liquid. It is also assumed that p = pa at cross-section
a, and p = pb at cross-section b, where pa and pb are two positive
Figure 2.2: Curved pipe between cross-
constants.
sections a and b. The figure shows a section
• The density ρ is constant.
of the pipe along the plane of the pipe’s axis.
• v = −va na at cross-section a and v = vb nb at cross-section b,
The section of the control surface is repre-
where va and vb are two positive constants.
sented by the broken line. The coordinate
• The weight of the liquid can be ignored.
axes are also shown, except for the x3 -axis
Equations (2.28) and (2.29) become
Z
which is perpendicular to the plane in the fig-
ρv · n d S = 0, (2.32)
ure. S
Z Z
ρvi v · n d S = ti d S. (2.33)
S S
Considering equation (2.32) first, we can calculate the integral as the sum
of the integrals over the partial areas Sa , Sb and Sr where the indexes
refer to cross-sections a and b and to the inner surface of the pipe. With
these assumptions, it is easy to see that,
Z Z Z
ρv · n d S = −ρva A, (...) = ρvb A, (...) = 0.
Sa Sb Sr
Applying this to equation (2.32) gives
va = vb ≡ v.
The velocity of the liquid is therefore the same at both cross-sections
due to conservation of mass. Turning to equation (2.33), we can again
calculate the integrals as sums of integrals over Sa , Sb and Sr . Looking
at the integrals on the left-hand side, we get
Z Z Z
ρv (v · n) d S = ρv 2 Ana , (...) = ρv 2 Anb , (...) = 0,
Sa Sb Sr
while the integrals on the right-hand side are
Z Z Z
t d S = − pa Ana , (...) = − pb Anb , (...) = F,
Sa Sb Sr
where F is the force which the pipe exerts on the liquid. This gives
F = ( pa + ρv 2 )Ana + ( pb + ρv 2 )Anb ,
or, using the known values of the components of na and nb ,
F1 = −( pa + ρv 2 )A + ( pb + ρv 2 )A cos θ,
F2 = ( pb + ρv 2 )A sin θ.
PROBLEMS 27

The force which the liquid exerts on the pipe is then −F.

Problems
Problem 2.1 Assume that the time derivative of the angular momentum in a
medium is equal to the momentum of the volume and surface forces exclusively.
Show that the stress tensor σi j is symmetrical.
Problem 2.2 Prove equations (2.11), (2.12), and (2.14).
Problem 2.3 A primary jet of water with area A p runs out into a secondary jet of
water in a pipe with uniform cross-section A (see figure 2.3). At cross-section b
the primary and secondary jets are completely mixed. Make the following assump-
tions: (i) the stress t(n) (see section 1.3.3) is given by formula t = − pn, where a b
p is the water pressure (see chapter 4); (ii) at cross-section a the velocity of the
P vs i vb i
primary jet is equal to v p i where v p is a positive constant and i is a horizontal unit vp i
vector (see figure 2.3), while the velocity of the secondary jet is equal to vs i where
vs is a positive constant; (iii) at cross-section b the velocity is equal to vb i, where i
vb is a positive constant; (iv) the pressure is equal to a constant pa at cross-section S
a (in both jets), and a constant pb at cross-section b; (v) the density of the water ρ Figure 2.3: See problem 2.3.
is a constant; (vi) the weight of the water can be ignored.
Express vb and pb − pa as functions of A p , A, v p , vs , and ρ. Find numerical
values given that
A p = 5 × 10−3 m2 ,
A = 6 × 10−2 m2 ,
v p = 30 m/s,
vs = 3 m/s,
ρ = 103 kg/m3 .
Problem 2.4 During tests carried out on a jet motor the velocity of the air at the
cross-section of the inlet is equal to va i where i is a horizontal unit vector (see
figure 2.4). The velocity of the gases at the cross-section of the outlet is equal to
vb i. Assume that va and vb are constant and that vb > va . Fuel is introduced so
that the mass of fuel per unit time, m, is a fraction ǫ of the mass of the incoming
airflow. The areas at the motor’s inlet and outlet are both equal to A. Assume that:
(i) the stress t(n) (see section 1.3.3) is given by formula t = − pn, both in the air
at the inlet and in the gases at the outlet, where p is the pressure (see chapter 4);
(ii) the pressure is equal to a constant pa at the inlet and to a constant pb at the
outlet, and pa = pb ≡ p; (iii) the density of the air at the motor’s inlet is equal to
a constant ρa ; (iv) the weight of the air and of the gases can be ignored.
Find the force F which the arm supporting the motor must be able to with-
stand by using the control volume enclosed by the broken line in figure 2.4. (The
”insides” of the motor are not shown in the figure. They consist of parts which
ensure a transfer of forces from the gas to the motor and vice versa.) Hint: Di-
vide
R the surface S of the control volume into three parts: Sa , Sb and Sr , and put
Sr i d S = Fi . Find the numerical value of |F| given that:
t
va = 150 m/s, Figure 2.4: See problem 2.4.

vb = 1000 m/s,
ǫ = 0.02,
A = 0.2 m2 ,
ρa = 1.0 kg/m3 .
28 2. THE CONSERVATION LAWS

Problem 2.5 A water reservoir is drained into a canal of uniform width b (see
figure 2.5; the width of the canal, at right angles to the plane of the figure, is not
shown). A region of still water can be observed just behind the fall of water, with
height H . Calculate H by using the control volume as shown in figure 2.5 and by
making the following assumptions: (i) the flow of water in the canal has height
h and velocity V i, where V is constant and i is a horizontal unit vector (see the
figure); (ii) the density of the water is constant and equal to ρ; (iii) the stress t(n)
(see section 1.3.3) is given by formula t = − pn, where p is the pressure (see
chapter 4); (iv) the air pressure at the surface of the water, both in the region of
still water and in the canal, is equal to constant p0 , and the water pressure at a
depth z from the surface is equal to p0 + ρgz, where g is the acceleration due to
Figure 2.5: See problem 2.5. gravity; (v) the fall of water from the reservoir is vertical.
CHAPTER 3

Application: some elementary traffic problems

3.1. An elementary model of traffic flow


Consider a traffic queue along a road without junctions so that vehicles can
neither join nor leave the queue, that is, the number of vehicles in the queue is
constant. Let ρ denote the traffic density (the number of vehicles per unit length).
We are going to construct a continuous model for traffic flow, with ρ(x, t) a func-
tion of position x and time t.
Assigning the coordinates x F (t) and x S (t) to the first and last vehicle in the
queue, respectively, we can write
Z x F (t )
d
ρ(x, t) d x = 0, (3.1)
dt x S (t )
which expresses that the number of vehicles in the queue is conserved. This equa-
tion is analogous to equation (2.1), so the assumption that ρ is continuous gives us
(see equation (2.2))
∂ρ ∂
+ (ρv) = 0, (3.2)
∂t ∂x
where v(x, t) is the speed of the vehicle at position x and time t. We will use the
denotation
j = ρv, (3.3)
for traffic flux (the number of vehicles passing a specific point per unit time).
Equation (3.2) has two unknown functions, ρ and v. There are no other ob-
vious conservation laws so that, in order to progress, we shall have to introduce
a relationship between ρ and v. Such a relationship should be based on observa-
tions of relevant traffic situations. However, the more realistic a model, the more
complicated the calculations. It is therefore customary to start with a model which
takes into account the most obvious observations, but which otherwise is as mathe-
matically simple as possible. This will hopefully provide us with an understanding
of the most significant features with the aid of simple calculations.
It is an acknowledged fact that speed decreases with density and that v = 0 at
maximum density, that is when the vehicles are standing bumper to bumper. Let
ρ M be the maximum traffic density and let V0 be the highest permitted speed1. The
simplest relationship between v and ρ is then
v = V0 (1 − ρ/ρ M ). (3.4)
This gives v = V0 when ρ = 0, that is when there are no vehicles on the road! A
more realistic model would be to let the speed be equal to V0 for all ρ less than
a value ρm . Such a model, however, would certainly lead to more complicated
calculations, which makes it more sensible to keep the expression (3.4), provided
it does not have untenable consequences. Equations (3.3) and (3.4) then give Figure 3.1: Car flux as a function of car den-
sity.
j = V0 ρ(1 − ρ/ρ M ). (3.5)
1With a vehicle length of 4 m ρ is equal to 0.25 vehicles per metre. Typical values of V are 50
M 0
or 60 kilometres per hour.

29
30 3. APPLICATION: SOME ELEMENTARY TRAFFIC PROBLEMS

This function is shown in figure 3.1. We note that j is a function of ρ only,


j = j (ρ), (3.6)
and we will, in the rest of this section and in section 3.2, look at the consequences
of (3.6).
The differential equation (3.2) can be written
∂ρ ∂j
+ = 0. (3.7)
∂t ∂x
Substituting from equation (3.6) gives
∂ρ ∂ρ
+ c(ρ) = 0, (3.8)
∂t ∂x
where
c(ρ) = j ′ (ρ). (3.9)

Here j denotes the derivative of j . From equation (3.5) it follows that
c(ρ) = V0 (1 − 2ρ/ρ M ). (3.10)
Equation (3.8) is a non-linear differential equation for ρ(x, t). It is often called the
first order wave equation due to features associated with its solution. In the next
section we will show how this equation can be solved as an initial value problem,
that is, by assuming that ρ is known when t = 0.

3.2. Solution of the first order wave equation


Equation (3.8) is often called quasi-linear because it is linear in the derivatives
of ρ, but not in ρ itself since c is dependent on ρ. Equation (3.8) where c(ρ) is
replaced by c0 , a constant, is the first order linear wave equation. In order to
achieve an understanding of the first order wave equation, we will start by solving
the linear case. This is done in section 3.2.1. The general case, that is where c
depends on ρ, is addressed in section 3.2.2.
3.2.1. The first order linear wave equation. We are going to look at the
solution of the following initial value problem:
∂ρ ∂ρ
+ c0 = 0, (3.11)
∂t ∂x
ρ(x, 0) = f (x), (3.12)
where c0 is a given constant and f is a known function. An elementary method of
solving this problem is by applying the following theorem:
T HEOREM 3.1. Let φ(x 1 , x 2 ) be continuous and differentiable. Then
∂φ ∂φ
= (3.13)
∂ x1 ∂ x2
if and only if
φ = ψ(x 1 + x 2 ), (3.14)
where ψ(u) is an arbitrary continuous and differentiable function.
Let us prove the theorem.
Proof First we demonstrate sufficiency (by direct proof, see section 13.4.1),
that is, we assume that equation (3.14) is true and show that equation (3.13) fol-
lows. Using the chain rule we find
∂φ ∂φ
= ψ ′ (x 1 + x 2 ), = ψ ′ (x 1 + x 2 )
∂ x1 ∂ x2
where ψ ′ is the derivative of ψ. Hence we see that (3.13) is true.
3.2. SOLUTION OF THE FIRST ORDER WAVE EQUATION 31

Then we demonstrate necessity (by direct proof, see section 13.4.1): we as-
sume that equation (3.13) is true and show that equation (3.14) follows. We intro-
duce two new independent variables y1 and y2 by letting
 
y1 = x 1 + x 2 x 1 = (y1 + y2 )/2
⇐⇒
y2 = x 1 − x 2 x 2 = (y1 − y2 )/2.
In principle φ is now a function of y1 and y2 . In order to show that equation
(3.14) follows from equation (3.13) we only need to show that equation (3.13)
implies that φ is a function of y1 only, that is, that ∂φ/∂y2 = 0. We calculate
this partial derivative by representing the dependence of φ on y1 and y2 by φ =
φ(x 1 (y1 , y2 ), x 2 (y1 , y2 )) and by using the chain rule:
∂φ ∂φ ∂ x 1 ∂φ ∂ x 2
= +
∂y2 ∂ x 1 ∂y2 ∂ x 2 ∂y2
 
∂φ 1 ∂φ 1
= + −
∂ x1 2 ∂ x2 2
= 0,
so that (3.14) is true.
End of proof
Let us now use the theorem to solve the initial value problem expressed in r f(x) f(x-c0t)
equations (3.11) and (3.12).
By substituting f(X)
f(x-c0t)
x1 = x
x 2 = −c0 t,
it is clear that equation (3.11) can be written X X+c0t
∂ρ ∂ρ x-c0t x
= ,
∂ x1 ∂ x2 Figure 3.2: Graphical representation of the
and, according to the theorem above, ρ is an arbitrary continuous and differentiable solution to the initial value problem ex-
function of x 1 + x 2 , that is, pressed in equations (3.11) and (3.12), for
c0 > 0 and for an arbitrary t > 0. The arrow
ρ(x, t) = ψ(x − c0 t).
has length c0 t.
The function ψ can now be determined in terms of the initial condition, equation
(3.12):
ρ(x, 0) = ψ(x) = f (x).
So the solution of the initial value problem specified by equations (3.11) and (3.12)
is
ρ(x, t) = f (x − c0 t). (3.15)
When c0 > 0, ρ plotted against x for different values of t > 0 is as shown in figure
3.2: the f (x − c0 t) curve is obtained by a translation of the f (x) curve parallel
to the x-axis and to the right, where the translation vector (shown as a bold line in
the figure) has magnitude c0 t.
3.2.2. The quasi-linear first order wave equation. We are now going to
look at the initial value problem of the quasi-linear first order wave equation (3.8):
∂ρ ∂ρ
+ c(ρ) = 0, (3.16)
∂t ∂x
ρ(x, 0) = f (x), (3.17)
where c and f are differentiable functions.
It is useful to introduce a three-dimensional set of axes where the coordinates
are (t, x, ρ), as shown in figure 3.3 (top). The solution of the initial value problem
32 3. APPLICATION: SOME ELEMENTARY TRAFFIC PROBLEMS

will then be a surface in this set of axes. For the time being, all we know about
this surface is its curve of intersection with the (t = 0)-plane since this curve
represents the given initial value, equation (3.17). At an arbitrary time t1 , ρ as a
function of x is a curve in the (t = t1 )-plane. This curve is unknown at present.
r We are now going to solve equation (3.16) by using the following trick. Let
r(x,0)=f(x)
K1 be a curve2 x = x(t) in the (t, x)-plane with the property
x=X+C(X)t1 dx
= c(ρ). (3.18)
P r=f(X) dt
Along this curve (3.16) can be written
f(X)
K ∂ρ d x ∂ρ dρ
0 + ≡ = 0.
X X+C(X)t1 x ∂t dt ∂ x dt
Q
K1 In other words ρ is constant along K1 . It is important to establish the value of
f(X) this constant, and this can be done as follows. In general, K1 will intersect the x-
axis, and we call X the abscissa of the point of intersection (see figure 3.3 (top));
t1
the vertical line through this point intersects the curve ρ(x, 0) at point P, and the
R height of P is the required value of the constant, namely ρ(X, 0) = f (X). Since
t ρ is constant along K1 c(ρ) is also constant along K1 and from equation (3.18) it
x=X+C(X)t1
r=f(X) follows that K1 is a straight line with equation
r x = X + c(ρ(X, 0)) t. (3.19)
r(x,0)=f(x)
Let us set
c(ρ(X, 0)) = c( f (X)) ≡ C(X). (3.20)
f(X)
Note that C(X) is known since both f and c are known functions. Equation (3.19)
can therefore be written
x = X + C(X) t. (3.21)
X X+C(X)t1 x
For a given X this is the equation of a straight line. This line will generally intersect
Figure 3.3: Construction of the solution to the (t = t1 )-plane at a point, point R in figure 3.3. Since ρ is constant along K,
the initial value problem (3.16) and (3.17). ρ = f (X) at R. The point Q on the vertical through R, at a height equal to f (X)
The two-dimensional representation (at the above the plane, is then such that RQ gives the value of ρ at x = X + C(X)t1 and
bottom) results from the projection of the t = t1 . In other words, we have arrived at the solution of the initial value problem
(t = t1 )-plane on to the (t = 0)-plane, par- in parametric form:
allel to the t-axis. The arrow (at the bottom) x = X + C(X)t
is of a length equal to C(X)t1 .
(3.22)
ρ = f (X),
where the parameter is X. This is so because the curve can be drawn for any given
value of t. Proving that this is the only solution, however, is not within the scope
of this course. See for example reference [45, chapter 5].
The above exposition is relatively abstract. It is therefore useful to carry out
a control calculation which shows that equations (3.22) really do provide the so-
lution of the problem expressed in equations (3.16) and (3.17). Such a calculation
has the added advantage, as we shall see, of showing that c and f have to be
differentiable, which is not made clear in the exposition above.
We start by showing that (3.22) satisfies (3.17). The representation (3.22)
becomes
x=X
ρ = f (X),
when t = 0. This is equivalent to ρ = f (x) (t = 0), which is the same as (3.17).

2K is the projection of a “characteristic curve” K of equation (3.16). K is the line PQ in figure


1
3.3 (top).
3.2. SOLUTION OF THE FIRST ORDER WAVE EQUATION 33

Let us now show that (3.22) satisfies (3.16). In order to calculate the required
partial derivatives of equation (3.16) we first find d x and dρ from equations (3.22):
d x = (1 + C ′ (X)t)d X + C(X)dt
dρ = f ′ (X)d X,
which makes it clear that C ′ and f ′ , the derivatives of C and f , must exist for all
values of X relevant to the problem. The first equation then gives
d x − C(X)dt
dX = .
1 + C ′ (X)t
Using this in the equation for dρ gives
f ′ (X) f ′ (X)C(X)
dρ = ′
dx − dt
1 + C (X)t 1 + C ′ (X)t
and thus
∂ρ f ′ (X) ∂ρ f ′ (X)C(X)
= , =− .
∂x 1 + C ′ (X)t ∂t 1 + C ′ (X)t
Finally we arrive at
∂ρ ∂ρ f ′ (X)C(X) c( f (X)) f ′ (X)
+ c(ρ) =− + ,
∂t ∂x 1 + C ′ (X)t 1 + C ′ (X)t
where the right hand side is exactly equal to zero since c( f (X)) = C(X).
1
Turning to the physical significance of solution (3.22), a geometrical interpre- 0
tation of the solution is shown in figure 3.3. The bottom part of the figure makes t=
0.8
it clear that the construction is a generalization of the simple translation we saw
in the linear case. Instead of all points on the initial value curve moving at the 0.6 .5
same speed c0 , each point has its own speed c(ρ), where ρ is equal to the point’s t=0
ordinate. Let us look at some examples. 0.4
t=1
E XAMPLE 3.1: Solve the initial value problem
0.2
∂y ∂y
+ c(y) = 0,
∂t ∂x (3.23) 0 0.2 0.4 0.6 0.8 1
y(x, 0) = f (x),
√ Figure 3.4: Solution of the initial value prob-
where c(y) = y and f (x) = x. lem in example 3.1.
Taking into account that the dependent variable is denoted by y, we
can see from equations (3.16), (3.17), and (3.22) that the solution can be
written √
x = X + Xt
y = X,
Figure 3.4 shows y(x, t) plotted against x for three values of t.
E XAMPLE 3.2: Solve the initial value problem 1
5
t=0.5

t=
0.2

1.0 0
t=
t=

∂y ∂y
+ c(y) = 0, 0.5
∂t ∂x (3.24)
y(x, 0) = f (x),
where c(y) = 1 − 2y and f (x) = kx where k is an arbitrary constant. -1 -0.5 0.5 1
x
As in example 3.1, the solution can be written
-0.5
x = X + (1 − 2k X)t
y = X, -1
Since x and y are linearly dependent of X, we can easily arrive at the
Figure 3.5: Solution of the the initial value
Cartesian representation of the solution by eliminating X. This gives
problem in example 3.2.
k
y(x, t) = (x − t). (3.25)
1 − 2kt
34 3. APPLICATION: SOME ELEMENTARY TRAFFIC PROBLEMS

Given t, the solution is therefore a straight line which intersects the x-


axis at x = t, where the angular coefficient, k/(1 − 2kt), depends on t.
Figure 3.5 shows y(x, t) plotted against x for four values of t and k = 1.
Note that all the lines go through the point y = 1/2. The reason for
this is that c = 0 when y = 1/2.
With reference to later applications it is of interest to look at the case
k → ∞. The initial condition, y(x, 0) = f (x) = kx, is then the y-axis,
or the line x = 03. Letting k go to infinity in (3.25) one gets
t−x
y= . (3.26)
2t
which is the solution of the initial value problem (3.24) with c(y) =
1 − 2y and x = 0 as the initial condition. This problem is treated in
more detail in example 3.3.
E XAMPLE 3.3: Solve the initial value problem
∂y ∂y
+ c(y) = 0,
∂t ∂x (3.27)
with the initial condition x = 0.
and with an arbitrary differentiable c(y).
The method involving (figure 3.3, and formulas (3.22)) cannot be
applied, the underlying reason being that x = 0 is not a function. From
example 3.2, however, we know that the solution can be found as the
limit of the case y = kx when k → ∞. We will see that it is possible to
c(y)t derive equation (3.26) directly.
y By applying the most important property of the solution, namely
that every point on the initial curve has a horizontal velocity c(y) de-
pending on the point’s ordinate y, and by referring to figure 3.6, we see
that when we start along the curve x = 0, then at t > 0 we are on the
curve
x = c(y)t. (3.28)
x
Figure 3.6: See the solution of the initial
It can be verified by substitution that this satisfies equation (3.27). Set-
value problem (3.27).
ting c(y) = 1−2y in equation (3.28) we easily find that y = (t −x)/(2t):
this coincides with the solution which we found by a limit operation in
example 3.2.
The examples above show that when the initial function is a straight line and c is
linear, then the solution at a later time is also a straight line. In the next example
we will look at a somewhat more complicated initial function, and show that the
displacement is accompanied by a change of form.
E XAMPLE 3.4: Given the initial value problem
∂y ∂y
+ c(y) = 0,
∂t ∂x (3.29)
y(x, 0) = f (x),
where c(y) = 1 − 2y and


 0, x < −1,

x + 1, −1 ≤ x < 0,
f (x) = (3.30)

 −x + 1, 0 ≤ x < 1,

0, x ≥ 1,
find y(x, t) where t = 0.25, 0.5, and 0.75.

3It is easy to “derive” x = 0 from y = kx when k → ∞ by rearranging y = kx as y/k = x.


3.2. SOLUTION OF THE FIRST ORDER WAVE EQUATION 35

The function f is not differentiable for x = −1, x = 0, and x = 1.


It is therefore not certain that the method used above will apply. Remem- y 1
ber that this method involves using formulas (3.22) or, more generally,
using the fact that every point (x 0 , y0 ) on the initial curve has a velocity t=0
c(y0 ) so that, at time t1 , it has moved to (x 0 + c(y0)t1 , y0 ). We will first
apply the method and then address the question of its legitimacy. -1 0 +1 x
The function f consists of straight lines and c is linear. From pre-
1
vious examples we know that y(x, t) will consist of straight lines for all
values of t. The easiest way to find y(x, t1 ) is to find the positions of the t=0.25
three points which, at t = 0, have coordinates (−1, 0), (0, 1), and (1, 0):
-0.75 -0.25 1.25
• Point (−1, 0) has ordinate 0. Its velocity is therefore c(0) = 1 and,
at time t1 , it will be on the x-axis, with abscissa −1 + t1 . 1
• Point (1, 0) has ordinate 0. Its velocity is therefore c(0) = 1 and, at t=0.50
time t1 , it will be on the x-axis, with abscissa 1 + t1 .
• Point (0, 1) has ordinate 1. Its velocity is therefore c(1) = −1 and,
-0.5 0 1.5
at time t1 , it will be at point (−t1 , 1).
The function y(x, t) is drawn for different times in figure 3.7. A change 1
of form occurs because the various points on the curve move at different
t=0.75
speeds. Note that this change of form creates an unphysical situation:
y(x, t) is ambiguous for all t > 1/2. In figure 3.7, for example, the
curve for (t = 0.75) gives three different values of y for one single -0.75 -0.25 1.75
value of x in the range −0.75 < x < −0.25. This problem is addressed Figure 3.7: Solution of the initial value prob-
in section 3.2.3. lem (3.29) drawn for different times.
We now have to answer the following question: Can the use of this
method be defended when f (x) is continuous but not differentiable for
all x? Let us start by stating the problem more precisely: the “defects”
of y(x, 0) at x = −1, x = 0, and x = 1 are transferred to y(x, t) at
x = −1 + t, x = −t, and x = 1 + t, respectively. Looking at the
(x, t) plane, this means that y(x, t) will not be differentiable when the t

+1
(x, t) coordinates are on one of the three lines4 shown in figure 3.8. This

x
t=
t=

means that neither ∂y/∂t, ∂y/∂ x, nor the differential equation itself in
-
x

1
(3.24) exists on these lines.

1
Even so, the answer to the question is yes, but, unfortunately, a full
x-
explanation is not within the scope of this course. Suffice it to say that a t=
non-continuous and/or non-smooth function is a solution of a differential -1 0 1 x
equation if it is a solution of an integral equation which can be derived
from the differential equation. It is then called a weak or generalized
-1
solution. The reader is referred to the mathematical literature, for exam-
ple [44, 45]. An introduction to weak solutions, although with respect to
a different differential equation than the one in this chapter, is given in
section 12.4 of this compendium. Figure 3.8: In the initial value problem (3.29),
with initial value given by (3.30), the differen-
3.2.3. Shock front. We will now look at the effect illustrated in example 3.4, tial equation is not defined on the three lines
namely the fact that an ambiguity occurs in the solution because c is a function shown.
of the dependent variable. With reference to the traffic flow problem we denote
the dependent variable ρ and assume that c is a decreasing function of ρ. The
conclusions we arrive at can easily be transformed to other types of variation.
When c is a decreasing function of ρ high values of ρ will travel more slowly
than low values of ρ so that after a sufficiently long period of time such effects as

4These are “characteristic curves”. See footnote 2 on page 32.


36 3. APPLICATION: SOME ELEMENTARY TRAFFIC PROBLEMS

shown in figure 3.9 will occur. The curve (b) is unphysical since it gives several ρ
values for one single value of x.
Since it is not very likely that the model is wrong, the ambiguity has to be
eliminated in a way which makes sense physically. Figure 3.7 in example 3.4 gives
us an indication of the physical event that takes place: namely that a discontinuity
occurs at t = 0.5. A discontinuity is usually interpreted as a shock front, which
enables us to say that the ambiguity in curve b means that such a front has arisen
or, in other words, that ρ in reality is discontinuous and that curve b has to be
interpreted as shown in figure 3.10. The curve abcdefg has to be replaced by abfg.
a b Let us digress slightly to answer a question which may have arisen in the
r
reader’s mind: since the model indicates the occurrence of a shock front, why
is this not expressed directly, in other words, why does the model not produce a
discontinuous function? The answer is that an initial value problem consisting of
equations (3.16) and (3.17) with continuous c and f functions can not produce
discontinuities since it presupposes that ρ is such that equation (3.16) is defined,
that is ∂ρ/∂ x and ∂ρ/∂t are defined, and consequently, that ρ is continuous and
differentiable. (A further question: what about the discontinuous derivatives in
x
example 3.4? Answer: those discontinuities were imposed on the solution through
Figure 3.9: Ambiguous solution (curve b) as a non-smooth f . This resulted in a need to redefine the solution.)
the result of a decreasing c(ρ). Curve a In order for the interpretation shown in figure 3.10 to be valid there has to
shows ρ(x, 0). For the shaded areas, see be a physically meaningful way to determine the position of the vertical line bf.
the text following equation (3.31). Let us first, with reference to figure 3.9, note that if the a-curve has equation ρ =
f (x) and the b-curve is given by equations (3.22), then the shaded areas are equal
(assuming that equation 3.20 applies, namely that C(X) = c( f (X))): R .. proving this
is relatively simple, see problem 3.2. Let us use the notation . ρ d x for these
areas, using suitably selected integration boundaries.
With reference to the traffic flow model we know that the “area”
Z +∞
ρ(x, t) d x = B, (3.31)
−∞
is equal to the number of vehicles. Remember that it is not possible for vehicles
to drive on or off so that B is independent of time. Equation (3.31) therefore
applies both to the a-curve and the b-curve, with the same value of B. The physical
stipulation that the number of vehicles is the same at all points in time forces us to
choose bf in figure 3.10 so that the two shaded areas in the figure are equal. The
r discontinuous ρ function constructed in this way then satisfies equation (3.31).
f
The expression “shock front” implies that the discontinuity itself moves. In
e section 3.2.4 we will confirm this by deriving a general expression for the velocity
of a shock front. In section 3.3.2, we will look at an example of shock formation
d g and shock displacement in a flow of traffic.
c
b 3.2.4. The velocity of a shock front. In this section we are going to find the
a velocity of a shock front as a simple expression involving the values of density and
x1 s x2 x velocity immediately before and after the front.
Consider a medium in one-dimensional space and assume that the medium
Figure 3.10: The interpretation of the un-
is between the coordinates x 1 and x 2 > x 1 . The medium is in motion, so that
physical b-curve in figure 3.9 using a shock
x 1 and x 2 depend on time. Assume that the medium’s density ρ has one (and
front. The physically correct curve is drawn
only one) discontinuity at x = s and that this discontinuity moves in such a way
with a thick line.
that x 1 (t) < s(t) < x 2 (t). The abfg-curve in figure 3.10 is an example of the
representation of ρ as a function of x for a given t.
Equation
Z x2 (t )
d
ρ(x, t) d x = 0, (3.32)
dt x1 (t )
3.3. TWO APPLICATIONS OF THE TRAFFIC FLOW MODEL 37

expresses the fact that the mass of the medium between x 1 and x 2 is conserved.
We can apply this equation of conservation to derive the velocity of the shock
front. We cannot use Reynolds’ transport theorem directly since ρ is discontinuous
within the bounds of integration. We therefore write
Z x2 (t ) Z s(t ) Z x2 (t )
d d d
ρ(x, t) d x = ρ(x, t) d x + ρ(x, t) d x.
dt x1 (t ) dt x1 (t ) dt s(t )
where ρ is now continuous within each region of integration on the right hand side,
so that the transport theorem can be applied. We know (see problem 1.7) that, in a
one-dimensional space, the transport theorem is equivalent to theorem 13.1. This
enables us to write
Z s(t ) Z s(t )
d ∂ρ ds d x1
ρ(x, t) d x = d x + ρ(s(t), t) − ρ(x 1 (t), t) . (3.33)
dt x1 (t ) x 1 (t ) ∂t dt dt
We see that in this formula ρ(s(t), t) is the density immediately before the shock
front (point b in figure 3.10), and we put ρ(s(t), t) = ρ− . We also see that
d x 1 /dt = v1 , the velocity of the medium at x 1 . By using the notation ṡ for the
derivative of s with respect to time, and by simplifying ρ(x 1 (t), t) to ρ1 , equation
(3.33) can be written
Z s(t ) Z s(t )
d ∂ρ
ρ(x, t) d x = d x + ρ− ṡ − ρ1 v1 . (3.34)
dt x1 (t ) x 1 (t ) ∂t

In the same way we can derive that


Z x2 (t ) Z x2 (t )
d ∂ρ
ρ(x, t) d x = d x + ρ2 v2 − ρ+ ṡ, (3.35)
dt s(t ) s(t ) ∂t
where ρ+ is the density immediately after the shock front (point f in figure 3.10).
Equations (3.34) and (3.35) can now be used to express the equation for the
conservation of mass (3.32) in the following form:
Z s(t ) Z x2 (t )
∂ρ ∂ρ
dx + d x + ρ2 v2 − ρ1 v1 + (ρ− − ρ+ )ṡ = 0. (3.36)
x 1 (t ) ∂t s(t ) ∂t
r(x,0)
This equation applies for arbitrary values of x 1 and x 2 . Let us choose such values rM
that s − x 1 and x 2 − s are infinitely small. The integrals in equation (3.36) are
thereby negligible while ρ2 can be replaced by ρ+ and ρ1 can be replaced by ρ− .
Similarly, v2 and v1 can be replaced by v+ and v− , the velocities immediately after
and before the shock front. Then equation (3.36) gives
0 x
j+ − j−
ṡ = , (3.37) r(x,t)
ρ+ − ρ−
rM
where j = ρv.

3.3. Two applications of the traffic flow model


3.3.1. Starting at a green light. In this section we will apply the model de- -V0t 0 V0t x
veloped in section 3.1 and the results from section 3.2 to the following problem: Figure 3.11: The “starting at a green light”
An infinite queue of cars, of maximum density, that is ρ = ρ M , is waiting at a problem. Thick line: Assumed traffic density
traffic light. Find how the queue develops after the light has changed to green. at t = 0 (top), and calculated traffic density
We position the x axis so that the traffic light is at x = 0 and let the queue at t > 0 (bottom).
occupy the full length of the negative side of the axis such that, when possible, all
the vehicles will drive towards the right, that is towards x > 0. Additionally we
choose t = 0 to be the time at which the light changes to green.
38 3. APPLICATION: SOME ELEMENTARY TRAFFIC PROBLEMS

We are going to solve the initial value problem represented by equations (3.16)
r(x,0)
rM and (3.17) with an initial condition which, according to figure 3.11 (top) can be
written

ρ M , x ≤ 0,
ρ(x, 0) = f (x) = (3.38)
0, x > 0.

-a 0 x Let us use equations (3.22) where C(X) is given by equation (3.20) and c is given
by equation (3.10). For all X ≤ 0 we get f (X) = ρ M and C(X) = c(ρ M ) = −V0 ,
Figure 3.12: The “starting at a green light”
so that the whole horizontal plateau which ρ(X, 0) consists of for X ≤ 0 has
problem. Continuous variants of the as-
moved a distance V0 t to the left after time t. We easily find that the horizontal
sumed traffic density at t = 0: the a-value
plateau of ρ(X, 0) for X > 0 has moved a distance V0 t to the right after time t.
of the broken curve is less than the a-value
This means that ρ(x, t), plotted against x for t > 0, is as shown in figure 3.11
of the solid curve.
(bottom).
We see that the density is not determined in the interval −V0 t < x < V0 t.
This is not due to a serious defect in the solution, but is a result of careless applica-
tion: In section 3.2.2 it was specified that formulas (3.22) apply to differentiable f
functions, which is not the case for function (3.38). To avoid the problem we need
to make a smooth function f , for example as shown in figure 3.12. The smooth
transition can be achieved in many ways, for example by using a sine function
enabling us to replace equation (3.38) with

 ρM , x ≤ −a,
C ρ(x, 0) = f (x) = ρ M sin2 π2ax , −a < x ≤ 0, (3.39)

0, x > 0.
r The problem with the smooth transition is that it introduces a complication we
rM
would have liked to avoid: the idealized representation in figure 3.11 (top) is at-
tractive in its simplicity and we would prefer to keep it. We note that a smaller
p0 d q0 a-value gives an f which is still everywhere differentiable, while the resulting
function (shown with a broken line in figure 3.12) resembles more the idealized
-a 0 x version. This leads us to introduce the continuous curve C shown by the thick line
Figure 3.13: The initial condition for the in figure 3.13, and which consists of three straight lines: the two horizontal lines
“starting at a green light” problem. A smooth making up the discontinuous function f and an additional vertical line. Note that
function f (broken line) and curve C (solid the distance δ between a point p0 , chosen on the smooth function f , and the point
line). q0 on the vertical part of C and at the same height as p0 , can be made as small as
we like by choosing a suitably small a.
Consequently it is reasonable to assume that the simplest way to formulate the
“Starting at a green light” problem is: solve the set of equations (3.16) and (3.17)
where c is given by equation (3.10) and f is given as the curve C. The solution is
r then relatively simple to construct, bearing in mind examples 3.2 and 3.3 above.
rM Since C consists of straight lines, ρ(x, t) plotted as a function of x will also consist
of straight lines. It is easy to see that the traffic density at time t is as given in figure
3.14. (Note that the solution is a so-called weak or generalized solution: see the
last paragraph in example 3.4.)
-b -V0t 0 V0t x A series of questions can be answered using this model. One of the simplest
is the following: A vehicle is at x = −b in the queue at a traffic light when
Figure 3.14: The “starting at a green light” the light changes to green. How long does the driver have to wait before he can
problem. The traffic density at time t, corre- drive? The answer can easily be obtained by looking at figure 3.14, because we
sponding to a density at t = 0 given by curve know, according to equation (3.4), that the vehicle’s velocity is equal to 0 as long
C in figure 3.13. as the density is equal to ρ M , and this, according to figure 3.14, lasts as long as
V0 t < b. Therefore, the driver has to wait a time t = b/V0 before he can drive.
With V0 = 50 km/t ≈ 14 m/s we can see that, if we are 100 m away from the
traffic light, we have to wait approximately 7 s after the light has changed to green,
before we can drive.
PROBLEMS 39

3.3.2. Stopping at a red light. Now consider the following problem: An


infinite queue of traffic, with constant density ρi < ρ M , has to stop at a red light.
Make the following assumptions: (1) the traffic light is at x = 0 and changes to
red at t = 0; (2) originally, the traffic is flowing in the positive x direction; (3) all r(x,0)
the vehicles on the positive x axis when the light changes to red disappear (we are rM
not interested in them). We want to find the change in density over time for the
queue on the negative x axis.
We must, here also, solve an initial value problem consisting of equations
ri
(3.16) and (3.17) where c is given by equation (3.10). We have to decide on an
initial condition, that is ρ(x, 0) or f . 0 x
The model is such that the velocity is zero when ρ = ρ M . We can therefore Figure 3.15: The initial condition for the
simulate a red light at x = 0 and t = 0 by assuming that a queue with density “stopping at a red light” problem (section
ρ = ρ M for all x > 0 has suddenly appeared at t = 0: consequently ρ(x, 0) is the
3.3.2).
discontinuous function, equal to ρi for x < 0 and ρ M for x > 0. We have learned
from section 3.3.1 that we need a continuous curve, so we add a vertical part as
shown in figure 3.15.
The initial condition is a curve consisting of straight lines and, as in section
3.3.2, the solution is easy to construct when bearing in mind examples 3.2 and 3.3.
It is sufficient to find the positions, at time t, of the two points on the curve in
figure 3.15 with coordinates (0, ρ M ) and (0, ρi ). At time t, these points turn out
r(x,t)
to be at, (−V0 t, ρ M ) and (V0 (1 − 2ρi /ρ M )t, ρi ), respectively, as shown in figure
3.16 (top).
rM
In figures 3.15 and 3.16, ρi has been chosen smaller than ρ M /2 so that V0 (1 −
2ρi /ρ M )t > 0. If ρi were greater than ρ M /2, V0 (1 − 2ρi /ρ M )t would be to the
left of the origin, but still to the right of −V0 t so that the conclusion is the same,
independently of the size of ρi : there is a shock front at position s such that the ri
shaded areas are equal. Due to the symmetry of curve ρ(x, t), it is easy to find s: -V0t s 0 V (1-2rr 0
i

M
)t x
it is at the middle of the interval [−V0 t, V0 (1 − 2ρi /ρ M )t]:
ρi r(x,t)
s(t) = − V0 t. (3.40)
ρM
rM
See figure 3.16 (bottom). Before proceeding, let us check that the answer agrees
with formula (3.37):
j (ρ M ) − j (ρi ) −ρi V0 (1 − ρi /ρ M ) ri
ṡ = =
ρ M − ρi ρ M − ρi s 0 x
−ρi
= V0 , Figure 3.16: Traffic density (t > 0) for the
ρM “stopping at a red light” problem (section
which agrees with formula (3.40). 3.3.2).
In other words, our first result is an obvious one, namely that a region will
form behind the traffic light with maximum density ρ M . The model shows further
that the horizontal extension of this area increases linearly with time, and that the
rearmost boundary of the area is a shock front travelling at a velocity equal to a b
−(ρi /ρ M )V0 .
Most of the problems below apply the traffic flow model to simple situations.
In addition, in problem 3.7, the initial value problem consisting of equations l1(r) l2(r)
r
(3.16) and (3.17) is applied to a flow type which has important applications in
porous media, namely the Buckley-Leverett flow.

Problems x
Problem 3.1 Function y(x, t) satisfies Figure 3.17: See problem 3.2.
∂y √ ∂y
+ y = 0,
∂t ∂x
40 3. APPLICATION: SOME ELEMENTARY TRAFFIC PROBLEMS

with


sin(π x), 0 ≤ x ≤ 1,
y(x, 0) =
0, otherwise.

r Draw y(x, t) for t = 1 and for t = 1.5.


rM Problem 3.2 Assume that, in figure 3.9, the a-curve has equation ρ = f (x)
and that the b-curve is represented by equations (3.22), where C(X) = c( f (X)).
Show that the shaded areas are equal. Hint: Look at figure 3.17: first show that
l1 (ρ) = l2 (ρ) and use this to demonstrate that the areas are equal.
Problem 3.3 Refer to example 3.4. Assume that t > 1/2 and find s(t), the shock
front position as a function of time.
-L 0 x Problem 3.4 Refer to the “starting at a green light” problem in section 3.3.1.
Figure 3.18: See problem 3.5. (1) Find the movement, that is position as a function of time, of a vehicle which
is to the rear of the queue at position −X (X > 0) when the light changes to
green at t = 0.
(2) Given that the highest permitted speed is 14 m/s and that the light remains
green for 30 s, what is the largest value of X for which the car only just passes
the light before it changes to red?
Problem 3.5 There is a traffic light at x = 0. At t = 0 the traffic density is given
S1(x,0) by figure 3.18. The light changes to green at t = 0 and stays green until the last
Sb vehicle has passed. Find the speed of the last vehicle for t ≥ 0. At what time has
this speed reached a value equal to 90% of the maximum speed?
Numerical values: L = 100 m, V0 = 14 m/s.
Problem 3.6 Refer to the notation used in section 3.1. A lorry is driving at a
Sa speed of 3V0 /4. Behind the lorry a queue has formed, with uniform density and
length L. The lorry turns off at a car park and we assume that we can find the traffic
density as the solution of an initial value problem represented by equations (3.16)
0 x
and (3.17) where c(ρ) = V0 (1 − 2ρ/ρ M ). Remember the assumption behind
Figure 3.19: See problem 3.7, question 5. this expression for c: the traffic density v is dependent on the density through
v = V0 (1 − ρ/ρ M ). We want to find the traffic density as a function of time after
the lorry has left the road. We will look especially at how the speed of the last
vehicle in the queue changes over time (assuming no overtaking).
F1(S1) Assume that the lorry drives off at x = 0 and t = 0.
(1) Find ρ(x, 0) as a function of x.
(2) Find ρ(x, t) as a function of x for a relatively small value of t, let us say such
that V0 t < L. What is the speed of the last vehicle in the queue? Hint: the
last vehicle in the queue is at a shock front.
(3) The last vehicle in the queue drives at the constant speed of the shock front
(previous question) as long as t is less than a specific value t1 . Find t1 and
draw ρ as a function of x for t = t1 .
(4) Assume that t > t1 and find the shock front’s, or the last vehicle’s, position
as a function of time.
(5) Find the shock front’s speed as a function of time.
Sa Sb S1 (6) Given that L = 0.5 km and V0 = 80 km/t, for how long after the lorry has
driven off the road will the last vehicle in the queue drive before it reaches a
Figure 3.20: See problem 3.7, question 5.
speed V2 = 75 km/t (assuming no overtaking)?
Problem 3.7 The Buckley–Leverett flow.
The flow of two non-miscible liquids (identified below as liquid 1 and liquid
2) in a one-dimensional porous medium where gravity and capillary pressure can
PROBLEMS 41

be ignored, is described by the following equations:


−K k1 ∂ p
v1 = , (3.41)
φS1 µ1 ∂ x c1(S1)
−K k2 ∂ p
v2 = , (3.42)
φS2 µ2 ∂ x
∂ ∂
(φρ1 S1 ) + (φρ1 S1 v1 ) = 0, (3.43)
∂t ∂x
∂ ∂
(φρ2 S2 ) + (φρ2 S2 v2 ) = 0, (3.44)
∂t ∂x c1(Sb)
S1 + S2 = 1. (3.45)
Here vi , µi , ρi , Si and ki are the velocity, viscosity, density, saturation, and c1(Sa)
relative permeability of the i liquid, respectively. K and φ are the permeability
and porosity of the medium, and p is the pressure in the liquids. Sa Sb S1
The following quantities are introduced:
Figure 3.21: See problem 3.7, question 5.
u 1 = φS1 v1 , u 2 = φS2 v2 , (3.46)
u = u1 + u2. (3.47)
It is assumed that K , φ, ρi and µi are given constants and that k1 and k2 are given
functions of S1 and S2 . It is further assumed that S1

u 1 = given constant at x = 0, Sb
(3.48)
u 2 = another given constant at x = 0.
The problem contains five unknowns, namely v1 , v2 , S1 , S2 and p. These are
unknown functions of x and t. In what follows, we will focus our attention on
saturation S1 .
(1) Using equations (3.43), (3.44), and (3.45) show that
Sa
∂u
= 0, (3.49)
∂x
in other words that u is independent of x. tc1(Sa) tc1(Sb) x
(2) Show that it follows from (3.48) that u is independent of t.
(3) By eliminating ∂ p/∂ x between equations (3.41) and (3.42), show that Figure 3.22: See problem 3.7, question 5.

u 1 = f 1 u, u 2 = f 2 u,
where
k1 /µ1
f1 = , f2 = 1 − f1 .
k1 /µ1 + k2 /µ2 S1
Note that f 1 and f2 are functions of S1 only.
Sb
(4) Show that it follows from equations (3.43) and (3.49) that
Sf
∂ S1 ∂ S1
+ c1 (S1 ) = 0, (3.50)
∂t ∂x
where
d F1
c1 (S1 ) = ,
d S1 Sa
and
u
F1 (S1 ) = f 1 (S1 ).
φ tc1(Sf) x
(5) Assume that S1 (x, 0), F1 (S1 ) and c1 (S1 ) are given by figures 3.19, 3.20, and
Figure 3.23: See problem 3.7, question 6.
3.21, and show that it follows from equation (3.50) that S1 (x, t) is given by
figure 3.22.
42 3. APPLICATION: SOME ELEMENTARY TRAFFIC PROBLEMS

(6) A shock front occurs. We position the shock front as shown in figure 3.23
F1(S1) where the two shaded areas are equal. In this way we make the area bounded
to the left by x = 0, at the top and to the right by the thick line, and below by
the line S1 = Sa , has the same area as the shaded region in figure 3.22: We
want the two areas to be equal because they represent the amount of liquid
characterized by Sa ≤ S ≤ Sb that has passed x = 0 over time t.
(a) Show that S f in figure 3.23 is such that
F1 (S f ) − F1 (Sa )
F1′ (S f ) = ,
S f − Sa
where F1′ = d F1 /d S1 . Hint: The equality of the areas in figure 3.23
implies that
Sa Sf Z Sf
S1
c1 (S1 ) d S1 = c1 (S f )(S f − Sa ).
Figure 3.24: See problem 3.7, question 6b. Sa
(b) Show that S f can be obtained graphically as shown in figure 3.24.
(c) Find the velocity of the shock front. Draw a sketch of the shock front,
1t and 21t later than the time t which applies for figure 3.23. Hint:
the velocity of the shock front is constant.
Problem 3.8 A plane shock front travels through a fluid at velocity vs i where i
Shock front is the unit normal of the shock front. The densities and velocities within the fluid
before and after the shock front are ρ1 , v1 i and ρ2 , v2 i respectively (see figure
r1 r2 3.25). The broken line in the figure indicates the control volume to be used in the
v1i v2i calculations below. It is assumed that ρ1 , ρ2 , v1 , and v2 are constant within this
Area Area
A vsi A control volume.
(1) Find the mass M inside the control volume using the quantities indicated in
i the figure.
(2) Show that
ds ρ2 v2 − ρ1 v1
x1 s x2 x =
dt ρ2 − ρ1
Figure 3.25: See problem 3.8. using equation (2.28).
CHAPTER 4

Fluid models

In this chapter, which is a continuation of chapter 2, the word ”fluid” will be


used as a generic term for gases and liquids. It was pointed out in chapter 2 (see
section 2.5 in particular) that the stress tensor σi j and the energy flow due to heat
transfer qi are model dependent. Here we are going to look at two fluid models
where σi j and qi are ”derived” from certain assumptions about the model’s domain
of validity and on the basis of known physical laws.

4.1. Ideal fluids


Consider a fluid at rest (zero velocity everywhere) in a state of stable mechan-
ical equilibrium. Under these conditions, a familiar physical law states that the
fluid inside a closed surface S is subjected to stresses from the fluid outside S,
acting in the direction normal to S and inwards. In other words
t(n) = − pn (4.1)
where n is the unit normal pointing out of S. The quantity p is defined as the
pressure of the fluid. We then see that (see equation (1.21))
σi j = − pδi j . (4.2)
Assuming further that the fluid is in thermal equilibrium, that is that no heat trans-
fer takes place, then:
qi = 0. (4.3)
A simple fluid model can be constructed by assuming that equations (4.2) and
(4.3) apply even when the fluid is in motion. Such a model could then be used in
situations with small velocities, or rather, where the velocity gradients are small
since these are responsible for the stresses and the generation of heat. Using (4.2)
and (4.3) in equations (2.23) to (2.25) gives:
∂t ρ + ∂ j (ρv j ) = 0, (4.4)
Dvi
ρ = f i − ∂i p, (4.5)
Dt
DE
ρ = − p∂i vi . (4.6)
Dt
As indicated in section 2.5, these equations have to be supplemented by equations
of state.
Fluids satisfying equations (4.4) to (4.6) are called ideal fluids. Their viscosity
and heat conductivity are equal to zero. Equation (4.5) is called the Euler equation
(1755).
To the above equations one must add the condition which applies when the
fluid meets a solid body. This condition is that the fluid cannot penetrate the solid.
On the surface of the solid the following then applies
v(x, t) · n = v0 (x, t) · n, (4.7)
43
44 4. FLUID MODELS

where v(x, t) is the fluid’s velocity and v0 (x, t) is the surface velocity at point x
and time t. When the body is stationary in the chosen system of coordinates the
condition above is v · n = 0.

4.2. Viscous fluids – the Navier-Stokes equations


4.2.1. The viscosity tensor. The simple model described in the previous sec-
tion cannot explain the force acting on a solid body when a fluid flows around it.
The air resistance on a vehicle, for example, is calculated to be zero. This is known
as the “d’Alembert paradox”, the proof of which can be found in several books:
see for example references [5, 26]; for a less general proof see problem 4.9. This
does not mean that the model is useless, and in chapter 8 we will, in fact, use it to
study acoustic waves. It simply means that it cannot be used in situations where
the frictional forces are significant since the model does not take such forces into
account. This can be seen from equation (4.1), which states that, if d S is a ficti-
tious surface, the fluid on one side of the surface will affect the fluid on the other
side only through a force normal to d S.
In order to create a model where frictional forces are present we have to ensure
that t also has a component in d S. That is, we have to generalize the expression
for σi j given by equation (4.2). To do so we postulate that
σi j = − pδi j + σi′j , (4.8)

and specify that σi′j must vanish when the velocity gradients vanish, so that we
recover the simple model from the previous section.
It is easy to see that, if σi′j is to account for the stresses due to friction, it can
only depend on the velocity gradients. Figure 4.1 shows two fluid particles close to
each other, with different velocities, v1 and v2 . Simplifying the physical picture a
little, imagine that two layers of fluid are in contact with each other, one layer with
velocity v1 , the other with velocity v2 . It is only when the two velocities differ that
friction along the common surface and hence a stress in d S will occur (d S and its
unit normal are shown in grey in the figure).
Consequently, σi′j is a function of the gradients ∂m vn , which is zero when the
gradients are zero. The simplest model we can set up is, of course, linear, where we
assume that σi′j is a linear combination of the velocity gradients. Even with this
simplification the expression for σi′j is quite complicated unless we also assume
that the fluid is isotropic, that is that the fluid properties do not depend on any
Figure 4.1: Two fluid particles close to each specific direction. It can be shown, using theorems of tensor analysis, that isotropy
other, with different velocities. The solid lines implies that
represent other particle paths. A small sur-
∂vi ∂v j ∂vl
face and its normal are shown in grey. σi′j = a +b + cδi j , (4.9)
∂x j ∂ xi ∂ xl
where a, b, and c are constants which characterize the fluid. These must be de-
termined experimentally. Formula (4.9) can be seen as the lowest order in the
McLaurin series of a function of the velocity gradients. The model we are in the
process of constructing can therefore be said to be of the first order in the velocity
gradients. The model for ideal fluids of the previous section is of zeroth order.
We can easily see that the constants a and b in (4.9) have to be equal, as
a consequence of the fact that σi j = σ j i (see equation (2.9) and the associated
comments). See problem 4.1 which also gives a physical interpretation of equation
a = b. Expression (4.9) then becomes:
 
′ ∂vi ∂v j ∂vl
σi j = a + + cδi j .
∂x j ∂ xi ∂ xl
4.2. VISCOUS FLUIDS – THE NAVIER-STOKES EQUATIONS 45

This is usually re-written using two other parameters, µ and ζ :


 
′ ∂vi ∂v j 2 ∂vl ∂vl
σi j = µ + − δi j + ζ δi j . (4.10)
∂x j ∂ xi 3 ∂ xl ∂ xl
σi j is called the stress tensor and σi′j the viscosity tensor. The parameter µ is called
shear viscosity or simply viscosity while ζ is sometimes called volume viscosity.
Both µ and ζ can be shown to be positive (see [25], sections 16 and 49). Regarding
the physical interpretation of ζ , see [25], section 81.
Fluids whose flow can be described in terms of the viscosity tensor above are
called Newtonian. Most natural fluids (water, hydrogen, oxygen, and so on) con-
sisting of molecules of approximately spherical form are Newtonian. The number
of non-Newtonian fluids in industry is increasing. Their viscosity tensor is far
more complicated than described in expression (4.10) above. For models of non-
Newtonian fluids and their properties, refer to the literature on rheology [4, 6].
4.2.2. Conservation of momentum. The law of conservation of momentum
for our model is found by writing (2.24) with σi j given by (4.8) and (4.10). Note
that ∂ j σ j i must be calculated, so that we need to know whether µ and ζ depend on
the coordinates. Generally, it can not be ruled out that they depend on pressure and
temperature, but in many applications they are considered to be constants. When
µ and ζ are constant, equations (2.24) give
Dvi ∂p ∂ ∂  µ  ∂ ∂vl
ρ = fi − +µ vi + ζ + , (4.11)
Dt ∂ xi ∂x j ∂x j 3 ∂ x i ∂ xl
or, in vector form,
Dv  µ
ρ = f − ∇ p + µ∇ 2 v + ζ + ∇(∇ · v). (4.12)
Dt 3
If the fluid is incompressible, ρ is constant and equation (2.23) gives
∂ j v j = ∇ · v = div v = 0,
so that the vector equation (4.12) can be simplified to
Dv
ρ = f − ∇ p + µ∇ 2 v, (4.13)
Dt
which is often called the Navier-Stokes equation (or equations, depending on the
context, since (4.13) is a vector equation, that is a system of three equations).
It is worth noting that:
(1) The equation for the conservation of momentum for ideal fluids (that is
equation (4.5)) is invariant when time is inverted: when t is replaced
by −t, vi changes to −vi and the equation as a whole keeps its form.
The model for ideal fluids therefore gives a time-reversible image of
the flow. The equation for the conservation of momentum for viscous
fluids, however, is not invariant under time invertion: equation (4.13),
for example, becomes
Dv
ρ = f − ∇ p − µ∇ 2 v.
Dt
Hence, when viscosity is not equal to 0 the description of the flow is no
longer time-reversible.
(2) It has been shown experimentally that practically speaking all viscous
fluids ”adhere” to solid bodies they come into contact with. The only
well-known exceptions are gases with extremely low density. At the
body’s surface the following equation applies:
v(x, t) = v0 (x, t),
46 4. FLUID MODELS

where v(x, t) is the fluid velocity and v0 (x, t) is the surface velocity at
point x and time t. This equation is used as a boundary condition in the
solution of equation (4.12) or (4.13).
(3) The viscosities µ and ζ have dimension (see chapter 5, especially section
5.2.3):
[µ] = [ζ ] = (Mass)(Length)−1 (Time)−1 ,
= (Pressure)(Time).
In SI units they are given in Pa s (Pascal second). A more commonly
used unit is Poise (P) (in honour of Poiseuille, see problem 4.3), where
1 P = 0.1 Pa s
The viscosity of water is approximately 0.001 Pa s or 0.01 P.
(4) Equation (4.12) or equation (4.13) can easily be written in Cartesian
coordinates (see (4.11)). A transformation to other coordinate systems
is not so easy and the methods used to do this are out of the scope of this
course. See references [25] and [19]. See also the problems at the end
of this chapter.
4.2.3. The law of conservation of energy. In addition to equation (4.8), with
σi′j given by (4.10), Fourier’s law of heat conduction is used:
q = −k∇T, (4.14)
where T is temperature and k thermal conductivity. Equation (2.25) then becomes
DE
ρ = − p∇ · v + ∇ · (k∇T ) + 8, (4.15)
Dt
where the ”dissipation function” 8 is given by
∂vi
8 = σ j′ i . (4.16)
∂x j
As mentioned above, equations (4.12), (4.15), and the equation for the conserva-
tion of mass have to be supplemented by the equations of state (2.26) and (2.27).

4.2.4. Example of a complete system of equations. As already mentioned


above (see section 2.5), a complete description of a fluid in motion includes the
conservation equations and the equations of state. In this chapter we have seen
that the equation of momentum conservation contains two viscosities, µ and ζ , and
that the equation for the conservation of energy contains the thermal conductivity
k. In the most general case these three parameters, called transport coefficients,
will be functions of pressure p and temperature T , and these functions have to be
included in the system of equations which describe the fluid.
As a simple, but still realistic example of a total system of equations, we are
now going to look at the case where the fluid is a perfect gas. A perfect gas is a
model where the gas is assumed to consist of spherical molecules which do not
interact in any other way than by collisions. Gases like helium, argon, or nitrogen
behave approximately like perfect gases for pressures up to 50 atmospheres and
temperatures up to 100 ◦ C [18]. Statistical physics enables us to calculate the
transport coefficients. One finds that [32]:
r
5 mk B T 15 k B
ζ = 0, µ= 2
, k= µ,
16d π 4 m
where d and m are the diameter and mass of the molecules and
k B = 1.38 × 10−23J/K
4.2. VISCOUS FLUIDS – THE NAVIER-STOKES EQUATIONS 47

is the Boltzmann constant (the units are Joules per Kelvin, see chapter 5). For most
gases
d ≈ (from 0.5 to 1.0)10−9m.
A good approximation of the mass m can be obtained by multiplying the proton
mass
m p = 1.67 × 10−27 kg,
by the number of nucleons (protons and neutrons) per molecule.
Using the notation
ei j = 12 (∂i v j + ∂ j vi ),
1 = eii = ∂m vm ,
we can write the conservation equations as follows:
∂t ρ + ∂i (ρvi ) = 0,
Dvi
ρ = f i − ∂i p + ∂ j [2µ(ei j − 13 1δi j )],
Dt
DE
ρ = − p1 + 2µ(ei j ei j − 31 12 ) + ∂i (k∂i T ).
Dt
The equations of state are
p = (k B /m)ρT, E = 32 (k B /m)T.
The factor 3/2 in the expression for E is replaced by 5/2 if the gas molecules have
two atoms, or by 3 if they have more than two atoms. The reason for this is that
gases like N2 and CO2 can behave like perfect gases even though the assumption
of spherical molecules does not hold.

4.2.5. Example of a simplified system of equations. A complete descrip-


tion of a fluid in motion is quite complicated from a mathematical point of view.
When addressing a specific flow problem it is therefore useful to be able to deter-
mine whether assumptions can be introduced so as to simplify the mathematical
process.
The following assumptions have proved to be applicable to quite a few types
of flow (see for example problem 4.3 below):
• the temperature is a given constant (isothermal flow);
• the fluid density is a given constant (incompressible flow);
• the viscosity µ is a given constant.
The first two assumptions make the thermodynamic description of the fluid irrele-
vant. As shown in section 4.2.2, from the last two assumptions it follows that the
equations for the conservation of mass and momentum become
∇ · v = 0, (4.17)
Dv
ρ = f − ∇ p + µ∇ 2 v. (4.18)
Dt
Because of equation (4.17) the viscosity tensor (see equation (4.10)) reduces to
 
∂vi ∂v j
σi′j = µ + . (4.19)
∂x j ∂ xi
Equation (4.18) is, as already pointed out, called the Navier-Stokes equation. Equa-
tions (4.17) and (4.18) make up a system of four equations for the four unknowns
p, v1 , v2 , and v3 . Some of the problems below are aimed at solving these equa-
tions.
48 4. FLUID MODELS

Problems
Problem 4.1 Use equations (4.8), (2.9), and the symmetry of the Kronecker delta
to show that σi′j = σ j′ i . Then show that constants a and b in equation (4.9) are
equal.
Now show that σi′j is identical to zero (that is, that the viscosity effects dis-
appear) when it is assumed that the fluid rotates like a solid body with constant
angular velocity. Hint: let v = ω × x (see chapter 5 in [15]).
Problem 4.2 The Bernoulli and Torricelli theorems.
(1) Write the law of conservation of kinetic energy for an ideal fluid.
(2) Assume that the fluid has constant density ρ0 , and show that
 
D v2
ρ0 = vi f i − vi ∂i p.
Dt 2
(3) Assume that f i is due to gravity and show that
f i = −ρ0 ∂i 9, 9 = gx 3 .
(4) Now assume that ∂ p/∂t = 0 (steady-state flow) and show that
v2
ρ0+ p + ρ0 gx 3 = constant
2
along a particle path. This is Bernoulli’s1 theorem (1738).
(5) A container with an open tap (see figure 4.2) is being refilled so that the height
h is constant. Apply Bernoulli’s theorem to show that the velocity at B is
p
v = 2gh
(This is Torricelli’s theorem (1643). It has, incidentally, been derived a long
Figure 4.2: See problem 4.2. time before Bernoulli’s theorem). Hint: consider a particle moving from A to
B (figure 4.2) and assume that p A = p B .
Problem 4.3 Poiseuille flow.
A liquid is flowing through a horizontal cylindrical pipe with circular cross
section (radius R). We are going to use the simplified model (section 4.2.5) with
isothermal and incompressible flow and with constant viscosity. That is, we are
going to use equations (4.17) and (4.18) where v, as usual, is the velocity, p the
pressure and µ the viscosity.
Due to the problem’s geometry it is necessary to introduce cylindrical coor-
dinates (figure 4.3), with the z axis along the axis of the pipe. The components
of v along ur , uθ and uz are denoted by vr , vθ and vz . The following additional
assumptions are made:
• gravity is negligible,
• the flow is stationary (∂v/∂t = 0),
• v is along the z axis (vr = vθ = 0),
• vz does not depend on θ ,
• vz is equal to zero at the pipe wall.
These assumptions lead to a description of the flow which agrees with reality as
Figure 4.3: Cylindrical coordinates (r, θ, z).
long as the Reynolds number (see chapter 6, and question 6 below) is less than
ur , uθ and uz are orthogonal unit vectors.
approximately 2000.
(1) Use the additional assumptions above to show that equation (4.17) implies
that vz is a function of r only. Hint:
1 ∂ 1 ∂vθ ∂vz
∇·v= (r vr ) + + ,
r ∂r r ∂θ ∂z
1Daniel Bernoulli (1700–1782), son of Johann Bernoulli. For information about the latter, see
footnote 1 on page 119.
PROBLEMS 49

in cylindrical coordinates.
(2) In cylindrical coordinates, vector equation (4.18) gives the following:
!
∂vr ∂vr vθ ∂vr ∂vr vθ2
ρ + vr + + vz − =
∂t ∂r r ∂θ ∂z r
   
∂p 1 ∂ ∂  vr  1 ∂ 2 vr ∂ 2 vr 2 ∂vθ
− +µ 2 r3 + 2 + − ,
∂r r ∂r ∂r r r ∂θ 2 ∂z 2 r 2 ∂θ
 
∂vθ ∂vθ vθ ∂vθ ∂vθ vr vθ
ρ + vr + + vz + =
∂t ∂r r ∂θ ∂z r
   v  
1 ∂p 1 ∂ 3 ∂ θ 1 ∂ 2 vθ ∂ 2 vθ 2 ∂vr
− +µ 2 r + 2 + + 2 ,
r ∂θ r ∂r ∂r r r ∂θ 2 ∂z 2 r ∂θ
 
∂vz ∂vz vθ ∂vz ∂vz
ρ + vr + + vz =
∂t ∂r r ∂θ ∂z
   
∂p 1 ∂ ∂vz 1 ∂ 2 vz ∂ 2 vz
− +µ r + 2 + .
∂z r ∂r ∂r r ∂θ 2 ∂z 2
Use the additional assumptions to simplify these equations.
(3) Find vz . Hint: from F(r ) = G(z) it follows that F and G are both equal to
an arbitrary constant.
(4) Find the mean value of vz using the definition v̄z = Q/(π R 2 ρ) where Q is
the mass of liquid passing an arbitrary cross section of the pipe per unit time.
Answer:
p A − p B R2 − r 2 p A − p B πρ R 4 p A − p B R2
vz = , Q= , v̄z = .
l 4µ l 8µ l 8µ
where p A and p B are the pressure values at cross sections A and B, and l is
the distance between these cross sections.
(5) Find the z component Fz of the frictional force which the liquid exerts on a
unit length of pipe. Hint: the stress t at an area d S of the pipe is given by the
equation ti = n j σ j i where n j are the components of the unit normal to d S
pointing towards the liquid. The stress tensor in cylindrical coordinates is
∂vr
σ11 = − p + 2µ ,
∂r
 
1 ∂vθ vr
σ22 = − p + 2µ + ,
r ∂θ r
∂vz
σ33 = − p + 2µ ,
∂z
 
1 ∂vr ∂vθ vθ
σ12 = σ21 = µ + − ,
r ∂θ ∂r r
 
∂vθ 1 ∂vz
σ23 = σ32 = µ + ,
∂z r ∂θ
 
∂vz ∂vr
σ31 = σ13 = µ + ,
∂r ∂z
where the directions along the unit vectors ur , uθ and uz are denoted by 1,2
and 3, respectively.
Answer: Fz = ( p A − p B )π R 2 .
(6) Generally, a body in contact with a fluid will be subjected to a force F (for
example air resistance on a plane). A standard notation is F = 21 ρV 2 SC,
where ρ and V are representative quantities for the fluid density and the rel-
ative velocity between the body and the fluid, S is a representative area for
50 4. FLUID MODELS

the body, and C is a dimensionless vector. (See chapter 5, especially section


5.2.3, for the term dimensionless.)
For Poiseuille flow it is natural to use V = v̄z and to define the body as a unit
length of the pipe such that S = 2π R, which gives
Fz = ρ v̄z2 π R C z .
Show that
C z = 16/Re,
where Re = 2Rρ v̄z /µ is a dimensionless combination called Reynolds’ num-
ber.
Problem 4.4 A viscosimeter consists of two concentric cylinders with radii r1
and r2 (r1 < r2 ) and height h. The cylinders are free to rotate about their common
axis. The space between the cylinders is filled with the fluid whose viscosity is to
be measured. A motor rotates the innermost cylinder about the axis at a constant
angular velocity ω, while an applied moment of force Fr2 keeps the outer cylinder
stationary (see figure 4.4). The rotation of the inner cylinder is started. When
the fluid’s motion has reached a time independent regime, F is measured and the
viscosity is calculated by the formula
r22 − r12
µ= F.
4πωr12r2 h
The problem consists in deriving this formula by applying the simplified model of
section 4.2.5 where temperature, density and viscosity are constant. This means
that we base our solution on equations (4.17) and (4.18) where v is the velocity p
the pressure and µ the viscosity.
We make the following additional assumptions:
• Gravity is negligible.
• The flow is stationary (∂v/∂t = 0).
• The base and top locks of the viscosimeter have a negligible effect on the flow
between the cylinders so that pressure and velocity in a plane which is normal
to the common axis are independent of the position of the plane. Thus flow is
two-dimensional.
(1) We introduce polar coordinates with the notation defined in figure 4.3 when
ignoring the z axis. It is assumed that pressure is a function of r only. Make
corresponding assumptions for vr and vθ (the velocity components along ur
and uθ ). Show that the velocity satisfies equation (4.17), which in polar coor-
dinates is written
1 ∂ 1 ∂vθ
Figure 4.4: See problem 4.4. ∇·v= (r vr ) + .
r ∂r r ∂θ
Specify the boundary conditions and find the velocity.
The Navier-Stokes equations in two dimensions and in polar coordinates are
!
∂vr ∂vr vθ ∂vr vθ2
ρ + vr + − =
∂t ∂r r ∂θ r
   v  
∂p 1 ∂ 3 ∂ r 1 ∂ 2 vr 2 ∂vθ
− +µ 2 r + 2 − 2 ,
∂r r ∂r ∂r r r ∂θ 2 r ∂θ
 
∂vθ ∂vθ vθ ∂vθ vr vθ
ρ + vr + + =
∂t ∂r r ∂θ r
   v  
1 ∂p 1 ∂ 3 ∂ θ 1 ∂ 2 vθ 2 ∂vr
− +µ 2 r + 2 + 2 ,
r ∂θ r ∂r ∂r r r ∂θ 2 r ∂θ
when gravity is equal to zero and density and viscosity have constant values.
PROBLEMS 51

(2) Calculate the moment of force acting on the outer cylinder and derive the
formula for µ above.
It is given that the stress component along uθ is
 
∂  vθ  1 ∂vr
σ = −µ r + .
∂r r r ∂θ
The formula gives the stress which the fluid inside the cylinder with radius r
exerts on the medium outside.
Problem 4.5 Consider a liquid flowing down an inclined plane. The plane forms
an angle α with the horizontal and the liquid is of constant thickness h (see figure
4.5, which also shows the position of the axes, except the x 2 axis which is normal
to the plane of the figure and points into the paper). The liquid is assumed to be
unlimited along the x 2 axis. We are going to use the simplified model in section
4.2.5: that is, temperature, density and viscosity are constant and the liquid is
described by equations (4.17)–(4.19) where vi are the velocity components, p is
the pressure, ρ the density, and µ the viscosity. In equation (4.18) we substitute
f = ρg where g has the components
g1 = g sin α, g2 = 0, g3 = −g cos α,
and g is the acceleration of gravity.
(1) Assume that vi = (v(x 3 ), 0, 0) and that p = p(x 3 ). Show that v satisfies Figure 4.5: See problem 4.5.
equation (4.17). Show that p and v satisfy
∂ 2v ρg sin α
=− ,
∂ x 32 µ
∂p
= −ρg cos α.
∂ x3
(2) Find a boundary condition for p and find p.
(3) Given that the viscous stress acting on the (x 3 = h) plane is equal to zero,
show that
∂v
= 0 when x 3 = h.
∂ x3
(4) Given that the liquid velocity is equal to zero in the (x 3 = 0) plane, find v(x 3 ).
Problem 4.6 In order to facilitate the measurement of the volume of water per
unit time Q flowing in a canal, an underwater barrier is constructed. The thickness
of the water d is measured at an arbitrary point near the barrier and we record the
difference h in the height of the water surface at the point of measurement and a
point further up the canal where velocity is approximately equal to zero (see figure
4.6). One then uses the formula
p
Q = Ld 2gh
where L is the width of the canal and g is the acceleration due to gravity. Show
that the formula is correct by assuming that:
• The water velocity V at the point of measurement is a constant (that is, inde-
pendent of y and z).
• V can be found by applying Bernoulli’s theorem (see problem 4.2) to a parti-
cle path along the surface of the water.
Hint: The control volume method is not applicable. The first assumption makes Figure 4.6: See problem 4.6.
it possible to write Q in terms of L, d, and V . The second assumption makes it
possible to find V .
Problem 4.7 A liquid flows between two coaxial pipes of circular cross-section.
The radii are R1 and R2 (R1 < R2 , see figure 4.7). We apply the simplified
model of section 4.2.5: temperature, density and viscosity are constants and we
52 4. FLUID MODELS

use equations (4.17) and (4.18) where v is the velocity, p the pressure and µ the
viscosity. Assuming that gravity can be ignored, that the flow is steady-state and
that the velocity is along the axis of the pipe and dependent only on its distance
from the axis of the pipe,
(1) show that the components of the velocity along the axis are
" #
pA − pB 2 2 R22 − R12 r
R2 − r + ln ,
4µl ln(R2 /R1 ) R2
where p A and p B represent the pressure at two points A and B, and l the
distance between these points. Hint: equations (4.17) and (4.18) in cylindrical
Figure 4.7: See problem 4.7. coordinates are given in problem 4.3.
(2) Find the mean value of the velocity along the axis of the pipe by defining it as
Q/(π R22 ρ − π R12 ρ) where Q is the mass of liquid passing an arbitrary cross
section per unit time.
(3) Find the component along the axis of the pipe of the frictional force the liquid
exerts on a unit length of the two coaxial pipes. Hint: the components of the
stress tensor in cylindrical coordinates are given in problem 4.3.
Problem 4.8 A pipe filled with a liquid rotates about its own axis with velocity
of rotation ω. The pipe is cylindrical with a circular cross section and is closed
at the bottom. The cylinder axis is vertical. We assume that the liquid itself has
a rotating motion. The purpose of the problem is to see what kind of rotating
motion the simplified model in section 4.2.5 leads to when additional assumptions
are introduced into equations (4.17) and (4.18).
We want to find a solution which describes a liquid in time-independent rota-
tion and assume therefore that
• ∂v/∂t = 0,
• vz = vr = 0,
• vθ and p are independent of θ ,
• vθ is independent of z.
We use the same notation as in problem 4.3. The notation for the cylindrical coor-
dinates is shown in figure 4.3.
(1) Show that equation (4.17) is satisfied. Hint:
1 ∂ 1 ∂vθ ∂vz
∇·v= (r vr ) + + ,
r ∂r r ∂θ ∂z
in cylindrical coordinates.
(2) Equations (4.18) (the Navier-Stokes equations) are given in cylindrical co-
ordinates in problem 4.3 for the case where gravity can be ignored. Here,
however, we do not ignore gravity, but if we assume that it acts along the z
axis we can still use the first two equations in problem 4.3 as they are, while
the third becomes
 
∂vz ∂vz vθ ∂vz ∂vz
ρ + vr + + vz =
∂t ∂r r ∂θ ∂z
   
∂p 1 ∂ ∂vz 1 ∂ 2 vz ∂ 2 vz
− ρg − +µ r + 2 + ,
∂z r ∂r ∂r r ∂θ 2 ∂z 2
where g represents acceleration due to gravity.
Use the assumptions above to simplify the Navier-Stokes equations.
(3) Show that vθ = ωr . What significance does this have for the liquid?
(4) Find the pressure p in the liquid, given that p = p0 when r = 0, z = h.
(5) Assume that the upper part of the pipe is open to the atmosphere, that p = p0
on the surface of the liquid, and that the surface goes through point r = 0, z =
h. What is the form of the surface?
PROBLEMS 53

Problem 4.9 d’Alembert’s paradox.


Consider the time-independent flow of an ideal, incompressible fluid around S
vai vbi
a body, inside a long and straight pipe of uniform cross-section. We are going to
calculate the component along the axis of the pipe of the force F which the body is pa F pb
subjected to, by using the control volume method (see section 2.6). The selected r i r
control volume consists of the region between two closed surfaces, S and 6. The
area S consists of the internal surface of the pipe, Sr , as well as two lids Sa and Sb , Sa Sr Sb
at a large distance from the body. Sa and Sb are both plane, with normal vectors
parallel to the axis of the pipe. The area 6 consists of the body’s surface. See x1
figure 4.8 where S and 6 are drawn with a broken line.
The following assumptions are made: (i) the weight of the fluid can be ig-
nored; (ii) since the fluid is ideal the stress t(n) is given by the formula t = − pn, x3
where p is pressure (remember that n is the unit normal pointing out of the control x2
volume); p is a constant equal to pa at Sa and a constant equal to pb at Sb ; (iii) the
fluid density is a constant ρ; (iv) the fluid velocity at S can be written vi where i Figure 4.8: See problem 4.9.
is a unit vector parallel to the axis of the pipe, and the values of v at Sa and Sb are
two constants, denoted by va and vb , respectively.
(1) Show that conservation of mass in the control volume gives va = vb .
(2) Show that conservation of momentum in the control volume gives
F3 = ( pa − pb )A,
R
where A is the area of the cross section of the pipe. Hint: F = 6 pn d S.
(3) Use Bernouilli’s theorem (see problem 4.2) and show that F3 = 0.
Problem 4.10 A liquid is situated between two coaxial parallel, horizontal discs
(see figure 4.9). The discs are circular with radius R. The vertical distance between
them, h, is much smaller than R so that the liquid cannot escape. The upper disc is
rotated at a low velocity of rotation ω = constant, the lower is held stationary by a
moment of force T .
The purpose of the problem is to show that by measuring T , the viscosity
coefficient µ of the liquid can be calculated by using a formula derived in the
framework of the simplified model of section 4.2.5, namely
2hT
µ= .
πω R 4
Because of the geometry of the problem cylindrical coordinates are introduced.
Equations (4.17) and (4.18) in these coordinates are given in problem 4.3 (gravity
is assumed to be negligible in this problem, too). The position of the axes is shown
in figure 4.9; see also figure 4.3. Since a liquid of viscosity not equal to zero
“adheres” to solid bodies it comes into contact with, the following applies Figure 4.9: See problem 4.10.

vr = 0, vθ = ωr, vz = 0, when z = h,
vr = 0, vθ = 0, vz = 0, when z = 0.
Assume that
• The flow is independent of time.
• The flow is layered, that is
vr = 0, vθ = r f (z), vz = 0,
where f (z) is an unknown function.
• The pressure p is independent of θ .
(1) Show that equation (4.17) is satisfied.
(2) Give the boundary conditions for f , that is, the values of f (0) and f (h).
Look at the second Navier-Stokes equation and use the assumptions above to
54 4. FLUID MODELS

calculate f . Show that


ωr z
vθ = .
h
(3) The stress t, which the liquid exerts on the bottom disc, is given by
ti = n j σ j i ,
where n is the unit normal to the disc pointing towards the liquid. Directions
1, 2, and 3 referred to by the indexes in the equation above, are along ur , uθ ,
and uz ,respectively, and σi j in cylindrical coordinates is given in problem 4.3.
Show that
t1 = 0, t2 = µωr/ h,
and that the total momentum of the stress forces about the z-axis is
πωµR 4
T = ,
2h
Figure 4.10: See problem 4.11. A water jet
so that the viscosity can be calculated as specified at the beginning of this
directed at a flat object can in some cases
problem.
exert a force on the object towards the jet
(4) Now look at the first and third Navier-Stokes equations and use the assump-
(grey arrow).
tions made about velocity and pressure to simplify these equations. Show that
the resultant equations are contradictory. The model is therefore inconsistent.
Give a brief evaluation of the assumptions, without mathematical expositions.
Problem 4.11 A water jet, directed at an object, can in some cases attract the
object (see figure 4.10). We are now going to look at a simple model which can
explain this, and which is based on the fact that the attraction is best observed
when the water adheres to the object so that there is no splashing.
Consider a flat object and that part of the jet (shaded in figure 4.11) which has
thickness h and whose velocity is parallel to the object. In this volume of water
we imagine that the flow is layered, and that the different layers are parallel to
the plane of the object (see figure 4.11, top). In each layer the velocity vectors
are radial (see figure 4.11, bottom). Assume that the water is an ideal fluid (zero
viscosity) so that pressure and velocity do not vary in a direction at right angles to
the layers. Assume also that the water density is constant.
(1) Show that r v = rb vb where rb and vb = |vb | are defined in figure 4.11, and
where v is the magnitude of the velocity vector v for r between ra and rb .
Hint: choose a control surface S and use the fact that the mass inside the
control surface is conserved.
(2) Use Bernoulli’s theorem and show that pressure p at r is given by
  r 2 
ρv 2 b
p = pb + b 1 − ,
2 r
where pb is the pressure at rb . Assume that pb is equal to atmospheric pres-
sure and give a simple argument (without calculations) explaining why the
object is subjected to a force directed downwards when referring to figure
4.11 (top).
Figure 4.11: See problem 4.11. The shaded
Problem 4.12 Navier-Stokes in a rotating system. The Ekman layer.
part of the water jet is between two cylinders
Consider the simplified model of section 4.2.5: the temperature, density and
with a joint axis (shaded). The object is the
viscosity are constant and the liquid is described by equations (4.17)–(4.19) where
thick black line.
v is the velocity, p the pressure and ρ the density.
When the volume force is due to gravity we can write (see section 1.3.2)
f = ρg, where g = −gi3 , (4.20)
where g is acceleration due to gravity and i3 points vertically upwards.
PROBLEMS 55

Formula (4.20) assumes that the system of basis vectors (i1 , i2 , i3 ) is inertial.
In order to describe flows where the rotation of the earth is assumed to be signifi-
cant it is necessary, instead, to assume that the system (i1 , i2 , i3 ) is fixed in relation
to a point on the earth’s surface and therefore rotates with an angular velocity ω
(see figure 4.12). It can then be shown [15] that the expression for f must be written
 
f = ρ g − ω × (ω × x) − 2ω × v , (4.21)
where g is the same vector as in equation (4.20) (right) and x = x 1 i1 + x 2 i2 + x 3 i3 .
The first term between the parentheses on the right hand side is due to gravitation,
the second term is a centrifugal acceleration, and the third term is the Coriolis
acceleration. Figure 4.12 shows ω and the basis vectors i1 , i2 , and i3 . From the
figure we can see that
ω = ω sin φ i2 + ω cos φ i3 , where ω = |ω|.
(1) Show that f can be written
 
f = ρ∇ g · x + 21 |ω × x|2 − 2ρω × v, Figure 4.12: See problem 4.12. Movable co-
ordinate system at latitude φ; i3 is along the
and that equation (4.18) can therefore be written radius pointing outwards; i1 is the tangent to
the local latitude circle and points east; i2 is
Dv
ρ = −∇ P − 2ρω × v + µ∇ 2 v. (4.22) the tangent to the local longitude circle and
Dt points north.
where
P = p − ρg · x − 21 ρ|ω × x|2
(P is called generalized pressure).
(2) A relatively large ocean, at latitude φ < π/2 is affected by a wind towards
the east, that is, according to figure 4.12, along i1 . Assume that the extension
of the ocean is small enough for us to consider its surface to be flat so that
we can use Cartesian coordinates with an origin on the ocean’s surface. The
surface itself defines the horizontal plane x 3 = 0 and the ocean defines the
area x 3 < 0.
Assume also that the pressure and the velocity field can be described by equa-
tions (4.17), (4.19), and (4.22) with the following additional assumptions.
• The generalized pressure P is independent of x 1 and x 2 .
• There is a time-independent solution.
• The velocity vector in the ocean is horizontal and depends only on
depth:
v1 = v1 (x 3 ), v2 = v2 (x 3 ), v3 = 0.
(a) Show that equation (4.17) is identically satisfied and that the compo-
nents along i1 and i2 of equation (4.22) give
v1′′ = −2αv2 , v2′′ = 2αv1 , (4.23)
where (′ ) denotes derivation with respect to x 3 and
α = ρω cos φ/µ.
(b) Show that the stress t on a horizontal surface, which the fluid above the
surface exerts on the fluid beneath the surface, has horizontal compo-
nents t1 and t2 given by the formulas
t1 = µv1′ , t2 = µv2′ .
56 4. FLUID MODELS

(c) As mentioned, a wind blows along i1 and it is assumed that the wind
produces a stress on the surface of the water, given by
t1 = µS, t2 = 0,
where S is a positive constant. Find the conditions which v1 and v2
must satisfy when x 3 = 0.
(d) Solve equations (4.23). Hints: (i) the boundary conditions from item 2c
are not sufficient to determine all the integration constants. Use an extra
√ on physical considerations; (ii) substitute V = v1 +i v2
condition based
where i = −1; (iii) the result is
S √ √ π
v1 = √ e αx3 cos( αx 3 − ),
2α 4
S √ √ π
v2 = √ e αx3 sin( αx 3 − ).
2α 4
(e) Draw, in the horizontal plane, the vector v1 i1 + v2 i2 . In which direction
is the velocity pointing on the surface? At which depth√ is the velocity
pointing southward? Westward? Hint: Substitute X 3 = αx 3 and give
depth in X 3 -values.
2
(The Ekman layer√ is the uppermost layer of water, with√ a thickness of
approximately π/ α. Note that, at depth x 3 = −π/ α, the velocity
is pointing in the opposite direction to the direction on the surface, and
the magnitude of the velocity vector is reduced in relation to its value
on the surface by a factor of exp(−π) ≈ 0.04. The velocity vector’s
rotation with increasing depth is called the Ekman spiral.)

2Named after Vagn Walfrid Ekman (1874–1954), a Swedish oceanographer.


Part 2

Dimensional analysis and


applications
Standard for fathom and foot.
Relief in marble from Greece, 460-450 BC
Ashmolean Museum, Oxford
CHAPTER 5

Measurements, units, dimensions

Many of the formulations in this chapter are from Bridgman’s book [8].
We begin by looking at units, those physical quantities that are necessary to
carry out measurements.

5.1. Introduction. The SI system of units


Units are used in measurements. Thus, in everyday life, we might use a car-
penter’s folding rule to measure the length of an object. The rule (or some fraction
of it, or some multiple of it) is then the unit of length. To measure the time spent
on a task we usually use a watch that provides us with a unit of time (the second,
minute, . . . ).
To obtain our average velocity after a walk, we usually perform two measure-
ments: one of length (for which we again use a unit of length), and one of time
(for which we use a unit of time). We divide length by time and state the velocity
as a number, to which we add some expression as “metres per minute”. We might,
finally, have to account for the fact that most people have no feeling for velocities
of everyday activities unless they are given in kilometers per hour. So that a con-
version from the meters per minute to kilometers per hour must also be done. Our
unit of velocity will then be “metres per minute” or “kilometers per hour”.
In the examples above, length was measured directly with a unit of length,
and time was measured directly with a unit of time. Velocity of the other hand,
was measured indirectly. We first measured two quantities directly, then used a
formula to obtain the result.
We shall now look at the SI system of units (SI standing for Système Inter-
national). A system of units is usually constructed inside a model or a theory, to
simplify the writing of formulas and to provide physical quantities with easily un-
derstandable numerical values. Thus if we are studying the orbits of planets around
the Sun, an interesting unit of length would be the average distance of the Earth to
the Sun. Obviously, such a unit of length will not be satisfactory to an electonics
engineer.
It is practical to give a number of definitions before presenting the system of
units itself.
D EFINITIONS OF PRIMARY AND SECONDARY QUANTITIES
(1) The term physical quantity (or quantity for short) will be used to denote
something having a definition in physics. Examples of physical quanti-
ties are: length, time, mass, velocity, momentum, angular momentum,
energy, electrical charge, . . . . We shall say that two physical quantities
are of the same kind if they are both lengths, or times, . . . . Otherwise,
we shall say that they of different kinds: thus a length and an energy are
quantities of different kinds.
(2) A system of units is a set of primary quantities, a corresponding set of
primary units, and a set of formulas for calculating the measures of the
59
60 5. MEASUREMENTS, UNITS, DIMENSIONS

secondary quantities in terms of the measures of the primary quantities.


The italicized terms are defined below.
(3) A quantity is a primary quantity by convention and in the framework of
a system of units. By definition, the measurement of a primary quantity
is a rule or procedure involving as unit a quantity of its own kind, and
resulting in a number. That number is the measure of the quantity. The
unit is called a primary unit. The rule or procedure must satisfy the
following requirement: if the unit is divided by a number x, then the
number which results from the measurement is multiplied by x.
In the examples at the beginning of this section, we have obviously
used a system of units where length and time are primary quantities.
(4) As with the case of the primary quantities above, a quantity is a sec-
ondary quantity by convention and in the framework of a system of units.
By definition, the measurement of a secondary quantity involves mea-
surements of certain primary quantities, the results of which are com-
bined according to a certain rule of calculation, or formula, yielding a
number. That number is the measure of the quantity. The formula is
characteristic of the physical quantity.
In the example at the beginning of this section, we have used a sys-
tem of units where velocity is a secondary quantity.
The measurement of a secondary quantity must satisfy the following
requirement: Assume a set of primary units has been chosen, let Q 1 and
Q 2 be two secondary quantities of the same kind (say two viscosities),
and let q1 and q2 be their measures. Let now the sizes of the primary
units be changed in an arbitrary manner, and let q1′ and q2′ be the results
of new measurements. It is required that q1′ /q2′ = q1 /q2 (with allowance
for experimental errors, of course). This is called the requirement of
the absolute significance of relative magnitude. It “is essential to all the
systems of measurement in scientific use” [8].
When we know that our system of measurement satisfies this re-
quirement, then we know that the statement “fluid A is twice as viscous
as fluid B” is independent of the units used to measure the viscosities.
We now turn to the SI system (SI standing for Système International). It is an all-
around system which can be used in models and theories with no special needs as
far as units are concerned. It has seven primary units. The primary quantities and
the names of their units are shown in table 5.1.
Quantity Unit
Length L Metre m The definitions of the units are as follows [20].
Time T Second s
Mass M Kilogram kg D EFINITIONS OF THE SI PRIMARY UNITS
El. current S Ampere A
• The metre was defined in 1791 by the French Academy of Sciences as
Temperature Te Kelvin K
the ten millionth part of the distance from the North Pole to the equator,
Quantity of matter Me Mole mole
along the meridian passing through Paris. The survey was completed in
Luminous intensity I Candela cd
1798 and platinum standards of the metre were constructed. The survey
was later shown to be wrong and, in 1889 the International Bureau of
Table 5.1: The SI-system of units. Left:
Weights and Measures established the metre as the distance between two
primary quantities and their abbreviations.
lines on a bar of 90 % platinum and 10 % iridium. In 1984, the metre was
Right: primary units and their abbreviations.
redefined at another international conference (the CGPM or Conférence
Générale des Poids et Mesures): it is the length of the path travelled by
light in vacuum during a time interval of 1/299 792 458 second. (The
speed of light is thus fixed by convention as c = 299 792 458 metres per
second.)
5.1. INTRODUCTION. THE SI SYSTEM OF UNITS 61

• The second was formerly defined as 1/86 400 of the mean solar day
(average period of rotation of the Earth around its axis). In 1956 it was
redefined as 1/315 569 259 747 of the length of the tropical year 1900.
It was further redefined in 1968 as the duration of 9 192 631 770 periods
of the radiation corresponding to the transition between two hyperfine
levels of the ground state of the cesium 133 atom.
• The kilogram, originally defined in 1799, was supposed to be the mass
of a litre of water at the temperature of its maximum density. The mass
of the present standard was intended to be equal but turned out to be
slightly less than the old standard. It is the mass of a platinum-iridium
cylinder. The cylinder is kept at the International Bureau of Weights and
Measures laboratory at Sèvres, near Paris.
• The ampere is the constant current which, if maintained in two straight
parallel conductors of infinite length and negligible circular cross section
and placed one metre apart in a vacuum, would produce between these
conductors a force equal to 2 × 10−7 newton per metre of length. See
also section 5.3.2.1.
• The kelvin is the fraction 1/273.16 of the thermodynamic temperature of
the triple point of water. The definition was adopted in 1968, and it was
decided at the same time that the name of the unit is kelvin, symbol K
(not degree kelvin, symbol ◦ K): thus the temperature of the triple point
of water is 273.16 K. See also section 5.3.2.2.
• The mole is the amount of mater of a system which contains as many
elementary entities as there are atoms in 0.012 kg of carbon 12. The el-
ementary entities must be specified and may be molecules, atoms, ions,
electrons, or other particles or groups of particles. One mole of any
substance is the same for all substances, about 6.0221367 × 1023 (Avo-
gadro’s number).
• The candela is the intensity of light in a given direction of a source which
emits monochromatic radiation of frequency 540 × 1012 hertz (corre-
sponding to a wavelength of about 550 × 10−9 m) and whose radiant en-
ergy in that direction is 1/683 watt per steradian. Note that the word light
above is used in its technical sense of visible electromagnetic radiation,
i.e., electromagnetic radiation with a wavelength between 360 × 10−9 m
and 830 × 10−9 m. See also section 5.3.2.3.
Any physical quantity not appearing in table 5.1 is, in the SI system, a secondary
quantity in the sense that it is measured in secondary units. A few of the most
usual secondary units are listed below.
The unit of surface is the surface of a flat square region whose side is 1 m: it
is called the square metre and written m2 .
The unit of velocity is the velocity of a point moving on a straight line a
distance of 1 m in 1 s: it is denoted m/s. Correspondingly, the unit of acceleration
is the acceleration of a point whose velocity increases by 1 m/s per second: it
is denoted m/s2 . These definitions are in accordance with the usual formulas for
velocity and acceleration found in physics textbooks:
dx dv
v= , a= . (5.1)
dt dt
The velocity v and the acceleration a are written here in terms of the displacement
x along a straight line, t being time. For a constant velocity the equation on the
left is equivalent to
x2 − x1
v= , (5.2)
t2 − t1
62 5. MEASUREMENTS, UNITS, DIMENSIONS

and for a constant acceleration the equation on the right is equivalent to


v2 − v1
a= , (5.3)
t2 − t1
The unit of force is the force required to give an acceleration of 1 m/s2 to a
mass of 1 kg: it has been given a special name, the newton (N): 1 N = 1 kg m/s2 .
We note here also that the unit of force so defined is in accordance with the usual
formula linking the force F to the mass m and the acceleration a:
F = ma. (5.4)
See the appendix at the end of this chapter for a list of derived SI units with
special names.

5.2. Concerning secondary quantities


The units of the secondary quantities appearing in the table of the appendix at
the end of this chapter are expressed as products of powers of primary units. The
question is whether this is true of all secondary units of any systems of units. We
answer this question in section 5.2.2.1.
We begin, with section 5.2.1, by looking at a very important general property
concerning the formulas for the calculation of secondary quantities. We shall then,
in section 5.2.3, look at the concept of dimensionality, and finally at the units of
secondary quantities in section 5.2.2.
5.2.1. A theorem on the calculation secondary quantities. We shall show
that the requirements on the measurements of primary and secondary quantities
imply that any secondary quantity must be expressible as a product of powers of
primary quantities. To properly emphasize the importance of these requirements
we begin by rephrasing them in the form of postulates.
P OSTULATES ON MEASUREMENTS
(1) If a primary unit is divided by some number, then the measure of a pri-
mary quantity of the same kind is multiplied by that number.
(2) The ratio of the measures of any two quantities of the same kind does
not change when the sizes of the primary units are changed. (This is the
postulate of the absolute significance of relative magnitude.)
We see that these requirements involve changing the sizes of the primary units.
Since a system of units has a set of specified primary units, changing the sizes of
the units actually means changing the system of units. Let us then first establish
the notation.
Let S be a system of units. It consists of a set of r primary quantities, abbre-
viated as P1 , . . . , Pr . There is also a set of formulas for calculating the measures
of the secondary quantities in terms of the measures of the primary quantities. Let
pu1 , . . . , pur be the abbreviated names of the r primary units which correspond to
the r primary quantities. (If we wanted S to be identical to SI, then we would put
r = 7, P1 = L, . . . , Pr = I, as well as pu1 = m, . . . , pur = cd.)
Let S be another system of units, identical to S except for the sizes of the
primary units. Thus its set of primary quantities and its set of formulas are identical
to the corresponding sets in S. The primary units however have different names
and different abbreviations from those in S. They are abbreviated as pu1 , . . . , pur
and are such that
Size of pui = Size of pui divided by αi , αi > 0 (i = 1, . . . , r ). (5.5)
(If S was the SI system, then examples of these new units would be: pu1 = ft
(foot), pu2 = lb (pound), and so on.)
5.2. CONCERNING SECONDARY QUANTITIES 63

In S, a measure of the primary quantity Pi is denoted by pi and given in terms


of the unit pui .
We shall use the overbar to indicate values expressed with the new (over-
barred) units. Thus according to postulate 1 above, equations (5.5) imply
p̄1 = α1 p1 , ... p̄r = αr pr . (5.6)
We shall now prove the following theorem.
T HEOREM 5.1. Let s be the measure of a secondary quantity in some system
of units with r primary quantities. Then
s = C p1a1 . . . prar , (5.7)
where p1, . . . , pr are measures of primary quantities, and where C, a1 , . . . , ar are
some real numbers.
The proof is as follows.
Let S be a secondary quantity in S. As we know (see definition 4 in D EFINI -
TIONS OF PRIMARY AND SECONDARY QUANTITIES , in section 5), its measure-
ment involves the measurements of primary quantities and the use of a formula.
Let p1 , . . . , pr be the measures of P1 , . . . , Pr that were found when measuring S.
Then the measure s of S is obtained as
s = f ( p1 , . . . , pr ), (5.8)
where f is the “formula”. We want to prove that f ( p1, . . . , pr ) can only be of the
a
type C p11 . . . prar , where C, a1 , . . . , ar are real numbers.
To be able to use postulate 2 above, we must consider two secondary quantities
of the same kind. Let then S ′ and S ′′ be two such secondary quantities. According
to equation (5.8) above we can write their measures as
s ′ = f ( p1′ , . . . , pr′ ), s ′′ = f ( p1′′ , . . . , pr′′ ). (5.9)
Note that the measures of the primary quantities have one set of values when mea-
suring S ′ and another set of values when measuring S ′′ . These two sets of values
are indicated by a prime and a double prime above. Note also that, since S ′ and
S ′′ are of the same kind, their measures are calculated by using the same formula,
symbolized by the function f in (5.9).
Now suppose we go from system S to system S. We thus change the sizes of
the primary units in the manner shown by equations (5.5). Then the measures of
the primary quantities change according to equations (5.6) so that
s̄ ′ = f (α1 p1′ , . . . , αr pr′ ), s̄ ′′ = f (α1 p1′′ , . . . , αr pr′′ ). (5.10)
The formulas for calculating secondary measures are same in the two systems, so
that the function f in (5.10) is the same as in (5.9).) According to postulate 2 we
must have s̄ ′ /s̄ ′′ = s ′ /s ′′ , which implies that the function f must be such that
f (α1 p1′ , . . . , αr pr′ ) f ( p1′ , . . . , pr′ )
= .
f (α1 p1′′ , . . . , αr pr′′ ) f ( p1′′ , . . . , pr′′ )
This is equivalent to
f ( p1′ , . . . , pr′ )
f (α1 p1′ , . . . , αr pr′ ) = f (α1 p1′′ , . . . , αr pr′′ ) . (5.11)
f ( p1′′ , . . . , pr′′ )
We now assume that f is differentiable. We take the derivative of both sides of
equation (5.11) with respect to α1 , then put α1 = . . . = αr = 1 and rearrange a
little, getting
p1′ ∂ ′ ′ p1′′ ∂
′ ′ f ( p 1 , . . . , pr ) = f ( p1′′ , . . . , pr′′ )
f ( p1 , . . . , pr′ ) ∂ p1 f ( p1 , . . . , pr′′ ) ∂ p1′′
′′
64 5. MEASUREMENTS, UNITS, DIMENSIONS

Since p1′ , . . . , pr′ , p1′′ , . . . , pr′′ are arbitrary this equation implies that
p1 ∂
f ( p1 , . . . , pr ) = a1 ,
f ( p1 , . . . , pr ) ∂ p1
where a1 is some constant. This equation can easily be integrated, and one finds
f ( p1 , . . . , pr ) = p1a1 g( p2 , . . . , pr ). (5.12)
When we use this expression for f in equation (5.11) we get:
g( p2′ , . . . , pr′ )
g(α2 p2′ , . . . , αr pr′ ) = g(α2 p2′′ , . . . , αr pr′′ ) .
g( p2′′ , . . . , pr′′ )
Thus g must satisfy the same kind of equation as f . It follows that
g( p2 , . . . , pr ) = p2a2 h( p3 , . . . , pr ),
where a2 is some constant, and that
f ( p1 , . . . , pr ) = p1a1 p2a2 h( p3 , . . . , pr ).
It is now obvious that
f ( p1 , . . . , pr ) = C p1a1 . . . prar ,
so that, referring to equation (5.8), we have proved that the formula for calculating
a
the measure of a secondary quantity can only be of the type C p11 . . . prar , and we
have proved the theorem.
The constant C in equation (5.7) is usually set equal to 1, except in the cases
where we need to introduce so-called “practical” secondary units. Some examples
of the flexibility allowed by this constant are given in section 5.2.2.1.
The set of exponents {a1 , . . . , ar } characterizes a secondary quantity of a
given kind: quantities of the same kind have identical sets, while quantities of
different kinds have sets that differ by at least one member. We shall return to
these sets of exponents in section 5.2.3.
5.2.2. The units of secondary quantities.
5.2.2.1. Coherent and non-coherent systems. We have seen in section 5.2.1
that the formula for the calculation of a secondary quantity must be of the form
(5.7). We rewrite this formula below for the sake of convenience:
s = C p1a1 . . . prar , (5.13)
and ask the question “what is the unit of the secondary quantity s?”
We shall look at the two cases C = 1 and C 6= 1 separately.
We begin with the case C = 1.
Let us begin with the example of velocity. It is defined in mechanics as
1x
v= , (5.14)
1t
where 1x is an interval of length and 1t an interval of time. We restrict ourselves
to one-dimensional motion to avoid the complication of having to deal with vector
components. In addition, we assume that we have a system of units where length
and time are primary quantities, while velocity is a secondary quantity.
Note that formula (5.14), is a particular case of (5.13) where, say,
p1 = 1x, p2 = 1t, a1 = 1, a2 = −1, a3 = . . . = ar = 0, C = 1.
To answer our question about the unit of velocity we find the condition for v =
1. Obviously, we must have 1x = 1t. Taking the simple case of 1x and 1t
respectively being equal to one unit of length and one unit of time, we see that the
unit of velocity is the velocity of an object moving one unit of length in one unit
of time. By convention, we express this unit as a fraction where the numerator is
5.2. CONCERNING SECONDARY QUANTITIES 65

the name of the unit of length, and the denominator is the name of the unit of time.
Calling the unit of length ul and the unit of time ut we write the unit of velocity
ul/ut or ul ut−1 . In SI units for example, we write the unit of velocity as m/s or
m s−1 .
We can obviously generalize the procedure: if the formula for calculating the
measure s a secondary quantity S in terms of the r primary quantities is
s = p1a1 . . . prar , (5.15)

then

Unit of S = pua11 . . . purar . (5.16)


where we have used the notation of the beginning of section 5.2.1 for the primary
units.
We now turn to the case C 6= 1.
According to theorem 5.1, it should be possible to define velocity by including
a proportionality constant different from one on the right-hand side of formula
(5.14). So let us now define a velocity as
1x
ṽ = C, (5.17)
1t
and let us assume that C > 0. The rule for calculating velocity is quite clear, if
somewhat unfamiliar. What is the unit of velocity in this case? To answer this
question we write C in terms of two other constant, C = µ1 /µ2 , and rewrite
equation (5.17) as
µ1 1x
ṽ = .
µ2 1t
Introducing
1x̃ = µ1 1x, 1t˜ = µ2 1t,
we can write
1x̃
ṽ =
1t˜
We have now the same expression as in equation (5.14). We remember that the
unit of velocity for that case is ul/ut, where ul is the unit in which 1x is expressed
and ut is the unit in which 1t is expressed. Equivalently, if ul e is the unit in which
1x̃ is expressed and uet is the unit in which 1t˜ is expressed, then the unit in which
e uet. The sizes of ul
ṽ is expressed is ul/ e and uet are determined by postulate 1 at the
begining of section 5.2.1 (or, equivalently, by equations (5.5) and (5.6)), giving
e = Size of ul divided by µ1 ,
Size of ul
(5.18)
Size of uet = Size of ut divided by µ2 .
Now we see that, given C, there are infinitely many ways of choosing µ1 and µ2 .
However, we should look at the equations we have obtained in reversed order. To
be more specific, let us assume that we work in a system of units where length
and time are primary quantities, and velocity is a secondary quantity. Let us also
assume that we are performing calculations on a model where it is useful to think
of the metre and the second as units of length and time, but where the unit of
velocity that then follows from formula (5.14), the metre/second, is not practical.
We would prefer kilometre/hour. We can easily satisfy our wish by replacing, in
the model, equation (5.14) by equation (5.17) where C is determined as follows.
In equations (5.18), set
ul = m, e = 1000 m
ul
ut = s, uet = 3600 s.
66 5. MEASUREMENTS, UNITS, DIMENSIONS

We then find that


µ1 = 1/1000, µ2 = 1/3600,
and it follows that C = 3.6. The formula for velocity is thus
1x
ṽ = 3.6 , (5.19)
1t
The constant C is always used as illustrated above, i.e., for the purpose of in-
troducing “practical” units for secondary quantities. It is called a dimensionless
conversion factor. (We look at the concept of dimension and at the characteristic
of being dimensionless in section 5.2.3.) There are numerous examples of dimen-
sionless conversion factors. See problem 5.4.
Note that, in the example, we do not change the sizes of the primary units
since we are still using the metre and the second for mesuring lengths and times.
We have just expressed our secondary quantity in units which are multiples of the
existing primary units. This is the reason for using a tilde instead of an overbar to
denote the units introduced for expressing the secondary quantity.
Let us now generalize from the example above. We use the notation of the
beginning of section 5.2.1. Assume that we have a system of units with r primary
quantities, abbreviated P1 , . . . , Pr , and let pu1 , . . . , pur be the abbreviated names
of the corresponding r primary units. Assume also that we have a secondary quan-
tity S whose rule of calculation is given by formula (5.13):
s̃ = C p1a1 . . . prar , (5.20)
where p1 , . . . , pr are the measures of P1 , . . . , Pr that were found when measuring
S, expressed in the units pu1 , . . . , pur . C is now to be adjusted in such a way that
s̃ is expressed in some preferred units. The sizes of these preferred units are given
multiples of the primary units:
e i = Size of pui divided by µi ,
Size of pu µi > 0 (i = 1, . . . , r ),
where µ1 , . . . , µr are known positive numbers since we known which multiples to
take to obtain the preferred units. We introduce the following set of numbers:
p̃1 = µ1 p1 , ... p̃r = µr pr .
These are the measures of P1 , . . . , Pr that were found when measuring S, ex-
e 1 , . . . , pu
pressed in the units pu e r . Using these expressions in formula (5.20) we
get
s̃ = C ( p̃1 /µ1 )a1 . . . ( p̃r /µr )ar ,
!
C
= p̃1a1 . . . p̃rar .
µa11 . . . µrar
We choose now C such that
C = µa11 . . . µrar ,
so that
s̃ = p̃1a1 . . . p̃rar ,
and
e a11 . . . pu
Unit of S = pu e rar . (5.21)
It is important to realize that formulas such as (5.17) and (5.20) are correct in
all systems of units. However, they are practically useful in the one system for
which C has been so adjusted to the primary units as to lead to a practical unit for
the secondary quantity. Thus formula (5.19), when applied in a system where the
5.2. CONCERNING SECONDARY QUANTITIES 67

units of length and time are the metre and the second, gives velocities in kilometres
per hour, which is practical in some situations. The same formula, used in a system
where the units of length and time are the foot and the second, gives velocities in
kilofeet per hour. This is a legitimate unit, but it is doubtful whether somebody
will find it practical.
It must be mentioned that there is a definite disadvantage in the use of dimen-
sionless conversion factors. This becomes evident when manipulating equations,
since we then have to think about the proper conversion factors, and write them
when necessary. This can be illustrated by the elementary example of a particle in
free fall near the surface of the earth. Let us assume that we are using the system
of units mentioned above, where metre and second are the units of length and time,
and where kilometer/hour is the unit of velocity. In addition we must choose the
unit of acceleration. We shall assume here that metre/(second)2 is a convenient
unit. Let now x(t) be the position of the particle at time t, with the x-axis pointing
downwards, and let x 0 and v0 be the particle position and velocity at t = 0. If g is
the acceleration due to gravity, then
d2x
= g.
dt 2
Integrating once we get
dx
= gt + C, (5.22)
dt
where C is a constant. Putting t = 0 we get

d x v0
= = C. (5.23)
dt t =0 3.6
Note the use of equation (5.19). Equation (5.22) is now
dx v0
= gt + . (5.24)
dt 3.6
Integrating, we get
1 2 v0
x= gt + t + x0. (5.25)
2 3.6
Note that, due to equation (5.19), we do not obtain the velocity of the particle as
the derivative of x with respect to time. We have already used this in equation
(5.23) for the velocity at time zero. To get the velocity at time t we use equation
(5.24) and get
v = 3.6 gt + v0 . (5.26)
Most people working with theories and models try to avoid dimensionless
conversion factors, even when that implies using inconvenient units. One reason
is that one often has to make up one’s mind as to how many decimals to write:
in equation (5.25), for example, one is tempted to divide 1 by 3.6, but that gives
0.277 . . .. Another reason is that writing numerical constants is particularly boring
and a constant source of errors. Still another reason is that numerical constants
can be characterized as “visual noise” or “clutter”, which attract the attention of
the reader away from the important quantities in the equations where they appear.
A system of units is called coherent if all the secondary units of the system are
expressed as products of powers of primary units, with the proportionality factor
equal to 1. Referring to equations (5.15) and (5.16), this means that, in a coherent
system, C = 1 for any secondary quantity S.
68 5. MEASUREMENTS, UNITS, DIMENSIONS

5.2.2.2. Changing units. We here return to the formula for the calculation of
a secondary quantity, established in theorem 5.1. Recall that we found there that
s = C p1a1 . . . prar . (5.27)
In this formula, p1 , . . . , pr are the measures of the primary quantities that are found
while measuring the secondary quantity. They are expressed in the corresponding
primary units.
Assume that that we change the sizes of the primary units according to the
rule (5.5). The measures of the primary quantities change according to equations
(5.6). The measure of the secondary quantity S changes also. We denote it s̄ and
look for a fomula linking s̄ to s. We can write, according to equation (5.27),
s̄ = C p̄1a1 . . . p̄rar .

Using equation (5.6), we then get

s̄ = C(α1 p1 )a1 . . . (αr pr )ar

which gives, by equation (5.27),

s̄ = α1a1 . . . αrar s. (5.28)


This is the rule by which the measure of a secondary quantity changes under a
change of the sizes of the primary units. Note that the formula linking s̄ to s does
not depend on the value of C in formula (5.27).
Let us now look at an example where this equation might be used. Let a
force be given as F = 10 : m kg s−2 , and let us assume that we want the new
numerical value of F resulting from changing the unit of length from the metre to
the centimetre, the unit of mass from the kilogram to the gram, and the unit of time
from the second to the minute. To use equations (5.5), (5.6), and (5.28), we first
note that (see the table of the appendix at the end of this chapter)
original unit of F = N = m kg s−2 . (5.29)
Then we identify pu1 , pu2 , pu3 with m, kg, s, so that pu1 , pu2 , pu3 are cm, g, mn.
Then equations (5.5) show that α1 = 100, α2 = 1000, and α3 = 1/60. In addition,
equations (5.16) and (5.29) give a1 = 1, a2 = 1, and a3 = −2. So that the new
numerical value of F is
 −2
1
10 × 1001 × 10001 × = 3.6 × 109
60
the unit being cm g mn.
It must be noted that although formulas (5.6), (5.7), (5.16), and (5.28) are
useful to our understanding of units, their application to the practical problem of
finding new numerical values following from a change of unit size is somewhat
awkward and indirect. In fact, the calculations in the example above can be done
as follows
F = 10 : m kg s−2
1 m × 1 kg
= 10
1 s2
100 cm × 1000 g
= 10
(1/60)2 mn2
= 3.6 × 109 cm g mn−2 ,
which is direct and intuitively clear.
5.2. CONCERNING SECONDARY QUANTITIES 69

5.2.3. The dimension of a physical quantity. Returning to equation (5.7)


we see that, given a system of units, the formula for the calculation of a secondary
quantity is determined by the constant C, and the set of exponents {a1 , . . . , ar }.
These constants also determine the unit of the secondary quantity as explained in
section 5.2.2.1. We can then say that the constants characterize the quantity in a
certain sense. We shall see in what sense presently.
It turns out that there is a conceptual difference between the proportionality
constant C and the exponents. The role played by C is discussed in section 5.2.2.1.
We concentrate here on the exponents.
Let us consider coherent systems first. (See section 5.2.2.1.) The manner in
which the exponents characterize a secondary quantity can be illustrated by the
example of velocity. Inside the SI system, the unit of velocity is m1 s−1 . Changing
to other systems, identical to SI except for the sizes of the primary units, the unit
of velocity becomes km1 h−1 , or ft1 mn−1 . The exponents, 1 of the unit of length,
and -1 of the unit of time, remain the same when we change the size of the units
of length and time. To express this fact in a manner that does not refer to a partic-
ular unit size, we say that “velocity has dimension 1 in length and dimension -1 in
time”. We write this symbolically in the form of a formula called the dimensional
formula: LT−1 , where Land Tare the abbreviations we use for the primary quan-
tities length and time (see table 5.1). These abbreviations are written in a special
font, to remind us that they represent primary quantities.
Let us now briefly look at non-coherent systems. We have seen (section
5.2.2.1) that the inclusion of a number C in the formula giving the measure of
a secondary quantity can be seen as a way of expressing the secondary quantity
in units involving multiples of primary units, the exponents remaining the same.
Thus the factor 3.6 in equation (5.19) allows us to express velocity in km1 h−1
when lengths are expressed in metres and times in seconds. Thus C is unimpor-
tant from the point of view of dimension: the exponents are the only numbers that
characterize a quantity.
It has become common practice to abbreviate the “dimensional formula of
a secondary quantity Q” by the notation [Q]. Thus we write the dimensional
formula of velocity as
[v] = LT−1 .
As a further example, we see in the table of the appendix at the end of this chapter
that the SI unit of force is m kg s−2 . If we imagine that we change to another
system, identical to SI except for the sizes of the primary units, then force will be
expressed as a unit of length to the power 1, times a unit of mass at the power 1,
times a unit of time at the power -2. We can then write the dimensional formula of
force in terms of the abbreviations used for primary quantities of the SI system, as
[F] = LMT−2 . (5.30)
It is sometimes useful to be able to deduce the dimensional formula of a quantity
without knowing the units it is expressed in. This is done by using a formula where
the quantity in question plays an important role, for example a definition. Thus the
central equation concerning force is that it is equal to mass times acceleration. We
can thus write
[F] = M[ A], (5.31)
with A for acceleration. Thus we have an expression giving the dimensional for-
mula of force in terms of a primary quantity, namely M, and of the dimensional
formula of acceleration. Using now the definition of acceleration, we get
[ A] = LT−2 , (5.32)
70 5. MEASUREMENTS, UNITS, DIMENSIONS

and, combining formulas (5.31) and (5.32), we get the formula that we found by
looking up the SI units of force, namely (5.30).
Note that the dimensional formula of a quantity is, in a sense, a rewriting of
its unit, but in a manner that is slightly more general since it is independent of the
sizes of the primary units.
We say that a quantity is dimensionless if the exponents a1 , . . . , ar are all
equal to zero. The dimensional formula of a dimensionless quantity Q is thus
[Q] = 1.
We meet the very important concept of dimensionlessness many times in what
follows.

5.3. On the arbitrariness of primary quantities


5.3.1. Dimensional conversion factors. In the SI system of units, length,
time, and mass are chosen as primary quantities. It then follows from the classical
formula linking force to mass and acceleration, F = ma, that force is a secondary
quantity. The choice of length, mass, and time as primary quantities should not be
given any importance other than practical. There is, in particular, no compelling
physical or logical argument leading to that choice. One could for example choose
length, time, and force as primary quantities, thus making mass a secondary quan-
tity. Letting F be the abbreviation for force, we easily see that the dimensional
formula for mass is then given by
[m] = FL−1 T2 .
It is of course also possible to choose length, mass, and force as primary quantities,
thus making time a secondary quantity. It is left as an exercice to the reader to show
that in this case
[t] = F−1/2 M1/2 L1/2 .
More surprising, perhaps, is the fact that one can choose length, mass, time, and
force to be primary units. Formula F = ma seems to exclude this possibility.
However, we have seen in section 5.2.2.1 that it is possible to modify a formula by
introducing a a dimensionless factor k, when we want to introduce a “practical”
unit for F. We can now generalize this possiblity by introducing a dimensional
factor k.
Let us do this explicitly. We assume that we want to build a system of units for
mechanics where length, time, mass, and force are primary quantities. We assume
that the units of length, time, and mass are the same as in the SI system. We need
a unit for force, and this we define as follows: the unit of force is the weight of a
mass of 100 kg, measured at a place on Earth where the acceleration due to gravity
is 9.81 m s−2 . We shall call this unit the Hercules, abbreviated H .
We now rewrite the classical relation between force, mass, and accelaration as
F = κma.
The value of κ is easily found by looking at the one-dimensional version of this
equation and setting F = 1 H, m = 100 kg, and a = 9.81 ms−2 :
1
κ= H kg−1 m−1 s2 .
981
The dimensional formula for κ is obviously
[κ] = FM−1 L−1 T2 .
We need not go any further with this system of units. It is presented here as an
example of what one can do, but it is not of any real practical use.
5.3. ON THE ARBITRARINESS OF PRIMARY QUANTITIES 71

We shall now look at examples of the use of dimensional constants in the SI


system.
5.3.2. Examples from the SI system.
5.3.2.1. Electric charge. Length, time, and mass, are chosen as primary quan-
tities in the SI system and in a number of other systems, the Gaussian system in
particular. In most of those systems force is taken to be a secondary quantity, with
the dimensional formula MLT−2 . Now Coulomb’s law states that there is a force
between two electrical charges q1 and q2 at rest relative to each other, which can
be written
q1 q2
F =k 2 , (5.33)
r
where r is the distance between the charges and k is a factor of proportionality.
In the Gaussian system, k is dimensionless and equal to 1, and it follows that
charge is a secondary quantity. It is easy to see that its dimensional formula is
[q] = M1/2 L3/2 T−1 . It follows that electric current (charge per unit time) is also a
secondary quantity.
The authors of the SI system decided that, for practical reasons, electric cur-
rent was to be a primary quantity, with unit the Ampere (A). Then the unit of
charge is the Ampere second (A s), also called the Coulomb (C). They thus needed
to “disconnect” the units of force and charge, and introduced a dimensional k in
equation (5.33). To simplify the writing of other equations they put
1
k= ,
4πǫ0
where ǫ0 is “the permittivity of free space”. It is usually given in C2 N−1 m−2 , and
its dimensional formula is easily found to be M−1 L−3 T4 S2 .
5.3.2.2. Temperature. Temperature is related to our sensation of hot or cold.
Experiments show that many objects expand when they get hotter, and one has
measured temperature by measuring the length of a metallic rod or the volume
of a gas or a liquid. One must of course define a scale, and the Celsius scale
for example defines the coldness of freezing water as having temperature zero,
and the hotness of boiling water as having temperature 100. Then it was realized
that the temperature of an object is a measure of the average kinetic energy of
the molecules of the object. It would then have been compatible with physical
facts to express temperature in units of energy. This however, has turned out to be
inconvenient, at least for general uses, and a dimensional factor k (the Boltzmann
constant) is used so that it is 32 kT which is identified with the average kinetic
energy per molecule.
5.3.2.3. Intensity of light. The intensity of electromagnetic radiation emitted
by a source, and the intensity of light in particular, is the power (energy per unit
time) emitted. The SI unit for intensity of light (the candela) tries to account for
the efficiency with which the eye transforms light energy received into impulses
to the brain. This eficiency strongly depends on the wavelength λ of the light, and
varies from person to person. To eliminate individual differences an average (or
standard) eye has been defined1. Its efficiency as a function of λ is a bell-shaped
curve rising from a negligibly small value at 380 nm to2 a maximum (equal to 1)
at 555 nm, then decreasing to a negligible value again at 780 nm. Sources emitting
outside the nanometer interval [380, 780] are not seen. The varying efficiency of
the eye is the reason for considering a source emitting at the one wavelength of
1There are actually two standard eyes, one for day and one for night vision. The one described in
the text is for day vision.
2The nanometer, abbreviated nm, is equal to 10−9 m.
72 5. MEASUREMENTS, UNITS, DIMENSIONS

550 nm in the definition of the candella. Further, since the power emitted by a
source can vary with the direction of emission, one imagines a device measuring
the power emitted in a given direction. The device is placed at a distance r from the
source and admits light through an opening with area S. The power measured, P,
is thus the power emitted inside the solid angle  equal to S/r 2 , approximately3.
To eliminate the dependence on the device size and distance one then defines the
power per solid angle P/. The unit for this quantity is W/sr. To obtain the
power in candela (cd) one multiplies by a constant k = 683. This numerical value
is chosen so that the candela is as close as possible to the unit of light intensity
introduced in the nineteenth century, the flame of a candle.
Note that the SI-unit candela is introduced as a primary unit so that k has units
cd sr W−1 .

5.4. Some specialized systems of units


It has been mentioned above that people working in applications of theories
and models try to avoid dimensionless conversion factors. This is often also true
of dimensional conversion factors. Thus in thermodynamics or statistical mechan-
ics one often sets the Boltzmann constant k equal to 1, simply because it feels
unreasonable to repeatedly write kT instead of T . That the temperature is then
measured in units of energy is not perceived as a problem. Avoiding to write a
symbol not only saves the effort, but generally produces formulas that are unclut-
tered and consequently easier to read.
We shall now see some examples where this unwillingness to write constants
is extended to many of the fundamental constants of nature like the gravitational
constant G, the speed of light c, and the Planck constant h̄.
5.4.1. A system of units for planetary astronomy. The orbital period of a
planet whose mass is negligibly small compared to the mass of the Sun, and whose
elliptical orbit has semi-major a, is
s
2π a3
T = √ , (5.34)
G M⊙
(known as Kepler’s third law) where
M⊙ = 1.9891 × 1030 kg, (5.35)
is the mass of the sun.
For the Earth (symbol ⊕ in astronomy)
a = a⊕ = 1.495 978 70 × 1011 m, (5.36)
and
T = T⊕ = 365.2564 days = 1 yr. (5.37)
Assume now that we make a system of units with three primary quantities, length,
time, and mass. The related units are called: (i) the astronomical unit of length, ab-
breviated AU (internationally accepted notation), (ii) the astronomical unit of time,
abbreviated AUT, and (iii) the astronomical unit of mass, abbreviated AUM. (AUT
and AUM are not international notations.) These units are defined as follows:
• The AU is the semi-major axis of the orbit of the Earth a⊕ .
• The AUT is the orbital period of the Earth T⊕ .
• The AUM is the mass of the Sun M⊙ .
3To be exact,  is equal to the surface cut out on the sphere of unit radius, centered at the source,
by the cone whose apex is at the source and whose base is the opening of the device.
5.4. SOME SPECIALIZED SYSTEMS OF UNITS 73

When we now write equation (5.34) for the Earth and with the new units above we
get r
2π 1
1= √ ,
G 1
so that the gravitational constant is, in these units,
G = 4π 2 AUM−1 AU3 AUT−2 .
Kepler’s third law, equation (5.34), takes on the very simple form
T = a 3/2. (5.38)
This formula, and others written in astronomical units, give access to properties of
the solar system through elememtary calculations, where the numerical values of
G and M⊕ are not needed. It is, for example, well known that Halley’s comet has
a period of about 76 years. (This period has irregularly varied from about 74 years
to about 80 years, due to perturbations of the comet’s orbit by the “giant planets”
Jupiter and Saturn). Equation (5.38) gives a ≈ 18 AU. An upper bound for the
maximum distance of the comet from the sun is then 2a ≈ 36 AU. (This assumes
that the eccentricity is close to 1, which is not unreasonable since the comet is only
seen throughout a short interval of time.) The planet farthest from the sun, Pluto,
has an orbital semi-major axis of about 40 AU, so that Comet Halley moves inside
the solar system. (The values for Comet Halley published on the web in 2005 are
as folows: the eccentricity is 0.967, the farthest distance from the sun is 35 AU,
the inclination of the orbit relative to the ecliptic is about 18◦ .)
For some more applications of astronomical units, see problem 5.2.
5.4.2. A system of units for relativistic quantum mechanics. It was men-
tioned at the end of section 5.2.2.1 that people doing calculations in models and
theories avoid using systems of units where dimensionless conversion factors are
necessary. Perhaps the most important reason for doing so is the wish to keep
equations free from clutter. We shall now look at an example where a system of
units is introduced with the specific purpose of eliminating universal constants, the
presence of which is perceived as distracting clutter.
There are two universal constants in relativistic quantum mechanics, the ve-
locity of light c and Planck’s constant h (or h̄ = h/(2π)) which is an angular
momentum. Almost all equations written in this theory contain these two con-
stants, each raised at some power, in almost every term. To simplify the writing
of such equations one then introduces a system of units where h̄ and c are both
dimensionless and equal to 1.
We shall now construct such a system. We do not know, at this stage, what
the primary quantities are, so that we symbolize length, mass, and time by L, M,
and T , and assume that they are secondary quantities. Their dimensional formulas,
denoted by [L], [M], and [T ] (see section 5.2.3) are unknown. Now, since c is a
velocity, and h̄ is an angular momentum, we can write
[c] = [L] [T ]−1 ,
[h̄] = [M] [L]2 [T ]−1 .

Since we want c and h̄ to be dimensionless, the above formulas tell us that the
dimensional formulas of M, L, and T must be related by
[L] [T ]−1 = 1, and [M] [L]2 [T ]−1 = 1,
which implies that
[L] = [T ] = [M]−1 . (5.39)
74 5. MEASUREMENTS, UNITS, DIMENSIONS

We can now chose one of the three quantities, L, M, T , to be primary. The other
two will then be secondary, their dimensional formulas being given by equations
(5.39).
The most frequent choice, especially in experimental work, is to use mass
(or, equivalently4, energy) as the primary quantity. According to the notational
convention of section 5.2.3, we write
[M] = M,
[L] = M−1 ,
[T ] = M−1 .
We mention for the sake of completeness, that the unit of energy (or mass) used
in recent years is the electron volt (abbreviated eV) and its multiples, the mega
electron volt (MeV = 106 eV) and the giga electro volt (GeV = 109 eV). The def-
inition of the eV is as follows: it is the energy given to an electron by accelerating
it through a potential difference of 1 volt. By using the formula E = q V giving
the energy E in terms of the charge q and the potential difference V , one finds
1 ev = 1.60218 × 10−19 J. (5.40)
See problem 5.3 to clarify the meaning of a length given as 1 eV−1 , or of a time
interval given in the same way.

Problems
Problem 5.1 Use equation (4.10) to find the dimensions of the viscosities µ and
ζ . For water, µ = 0.001 SI units. Find the value of µ in CGS units (cm, g, s).
Problem 5.2 Find the unit of velocity in astronomical units, its equivalent in m/s
and km/h. Look at Halley’s comet.
Problem 5.3 Consider the system of units introduced in section 5.4.2. What is a
velocity of 0.62 in m/s. What is a mass of 1 eV in kg. What is a length of 1 eV−1
in m. What is a time of 1 eV−1 in s.
Problem 5.4 The volume V of a cube whose side has length l is defined in text-
books on calculus as
V = l 3.
It is implied (but unfortunately very seldom explicitly stated) that the unit of vol-
ume is a cube whose side is equal to one unit of length. For example, if the metre
is the unit of length, then the unit of volume is a cube whose side is 1 m. Now
there are areas of application where one wants the unit of length to be the metre
but the unit of volume to be a cube whose side is one foot, the foot (abbreviated ft)
being 0.3048 m. In that case, a dimensionless conversion factor is introduced, and
the formula for volume becomes
Ṽ = C l 3 .
Find C.

4Since c is dimensionless and equal to 1 in this system of units, the relation E = mc2 reduces to
E = m, so that mass and energy are expressed with the same unit.
APPENDIX: DERIVED SI UNITS WITH SPECIAL NAMES 75

Appendix: Derived SI units with special names


The table below is from reference [39].

Derived SI unit
Quantity Name Symbol Expressed in terms Expressed
of primary in terms of
units other
SI units
angle radian rad m m−1
solid angle steradian sr m2 m−2
frequency hertz Hz s−1
force newton N m kg s−2 J/m
pressure pascal Pa m−1 kg s−2 N/m2
energy joule J m2 kg s−2 Nm
power watt W m2 kg s−3 J/s
el. charge coulomb C sA As
el. potential volt V m2 kg s−3 A−1 W/A
capacitance farad F m−2 kg−1 s4 A2 C/V
el. resistance ohm  m2 kg s−3 A−2 V/A
conductivity siemens S m−2 kg−1 s3 A2 A/V
magn. flux weber Wb m2 kg s−2 A−1 Vs
magn. flux density tesla T kg s−2 A−1 Wb/m2
inductance henry H m2 kg s−2 A−2 Wb/A
luminous flux lumen lm cd sr
illumination lux lx m−2 cd sr
activity becquerel Bq s−1
absorbed dose gray Gy m2 s−2 J/kg
CHAPTER 6

The Pi-theorem

We shall now look at some applications of the concepts presented in chapter 5

6.1. Unit-free equations


D EFINITION OF A UNIT- FREE EQUATION
• An equation between physical quantities is called unit-free if it remains
invariant when the sizes of the primary units are changed in an arbitrary
manner.

First remark: the property of being unit-free means roughly that the form of the
equation does not change (for the exact meaning of this see below) when we
change units of measurement. It is important to keep in mind that the unit changes
we are considering are restricted to changes in sizes: we are not changing the set
of primary quantities.

Second remark: the invariance of an equation under unit change has the following
meaning. Assume that the equation is
f (Q 1 , . . . , Q n ) = 0, (6.1)
where f is some known expression that could, incidentally, involve derivatives
or partial derivatives of some quantities with respect to some others. With the
notation introduced at the beginning of section 5.2.1 we use overbars to indicate
values expressed in the new units. The equation is invariant under the considered
change of units if it can be written
f (Q 1 , . . . , Q n ) = 0, (6.2)
where the expression f is the same in all details as in (6.2), so that the only differ-
ence between (6.1) and (6.2) is the presence of the overbars in the second.

Let us look at some examples of unit-free and not unit-free equations.


E XAMPLE 6.1: We look here at the equation describing free fall near the
surface of the earth. The calculations are done in some detail in section
5.2.2.1 so that we can use the results obtained there. Let x(t) be the
position of the particle at time t, with the x-axis pointing downwards,
and let x 0 and v0 be the particle position and velocity at t = 0. If g is
the acceleration due to gravity, then
1 2
x= gt + v0 t + x 0 , (6.3)
2
if we assume that we are using a coherent system of units. Specifically,
we assume that velocity is equal to d x/dt, without the factor 3.6 that
was included in the free fall example of section 5.2.2.1.
We shall now show that equation (6.3) is unit-free. It relates the five
physical quantities x, x 0 , t, v0 , g. Let us assume that we are using a
77
78 6. THE PI-THEOREM

system of units where length and time are primary quantities and where
velocity and acceleration are secondary quantities. Then
[x] = [x 0 ] = L, [t] = T, [v0 ] = LT−1 , [g] = LT−2 . (6.4)
We use the notation introduced at the beginning of section 5.2.1. We
assume that we introduce a new unit of length and a new unit of time
with sizes equal to, respectively, 1/α L and 1/αT times the sizes of the
old units. Using overbars to indicate values expressed with the new units,
we obtain
x̄ = α L x,
x̄ 0 = α L x 0 ,
t¯ = αT t,
v̄0 = α L αT−1 v0 , (6.5)
ḡ = α L αT−2 ,
where we have used equations (5.6) and (5.28), together with equations
(6.4). The form of equation (6.3) in the new units now follows:
1 −1 2
α −1 −1 ¯ 2 −1 −1 ¯ −1
L x̄ = (α L αT ḡ)(αT t ) + (α L αT v̄0 )(αT t ) + α L x̄ 0 ,
2
which simplifies to
1 2
x̄ = ḡ t¯ + v̄0 t¯ + x̄ 0 .
2
Comparing with equation (6.3) we see that the form is conserved under
the change of unit sizes so that the equation is unit-free according to the
definition.
E XAMPLE 6.2: Bridgman [8] considers the following example. Suppose
we have observed the tides in a given port, using a foot rule to measures
heights and a clock graduated in hours to measure times. After many
measurements we conclude that the height of water may be represented
by
h = 5 sin(0.5066 t). (6.6)
Obviously, this equation is to be used with t given in hours and h in
feet, so that it is not unit-free. We can check that this is indeed the case
by changing units as was done in example 6.1. We assume that we use
a system of units where length and time are primary quantities, then
introduce a new unit of length, equal to 1/α L times the foot, and a new
unit of time, equal to 1/αT times the hour. Using overbars to indicate
values expressed with the new units, we get
h̄ = α L h,
t¯ = αT t.
It follows that equation (6.6) becomes, in the new units,
 
h̄ t¯
= 5 sin 0.5066 .
αL αT
The constants α L and αT do not cancel, and we do not end up with the
same form as equation (6.6), which means, as expected, that it is not
unit-free.
The reader can easily check, by the method used in this example and
in example 6.1, that equation (5.38) is not unit-free.
6.1. UNIT-FREE EQUATIONS 79

E XAMPLE 6.3: We look here at the case of the non-coherent system of


units examplified by the equation of free fall, where a dimensionless
conversion factor is used in the definition of velocity. This equation has
been derived in section 5.2.2.1 and is
1 1
x = gt 2 + v0 t + x 0 . (6.7)
2 3.6
For the definitions of the symbols, see example 6.1. The number 3.6
originates in the definition of velocity as
dx
v = 3.6 , (6.8)
dt
and makes sure that, when lengths are measured in metres and times in
seconds, then velocities are given in kilometers per hour.
We show here that equations (6.7) and (6.8) are unit-free. Actually,
the calculations that are necessary to do so are exactly the same as the
ones carried out in example 6.1, and need not be repeated here: they lead
to
1 1
x̄ = ḡ t¯2 + v̄0 t¯ + x̄ 0 , (6.9)
2 3.6
and
d x̄
v̄ = 3.6 . (6.10)
d t¯
It should be noted, in particular, that equation (6.5) is still valid: see
equation (5.28) and the remark following that equation.
The reader might still have some doubts as to the unit-free character
of equations (6.7) and (6.8), because of the number 3.6 and its justifi-
cation: it ensures that, when lengths are given in metres and times in
seconds, then velocity is given in kilometres per hour. Assuming that
the metre and the second are primary units for equation (6.7), they are
no longer so for equation (6.9). What is then the use of the number 3.6
in equation (6.9)?
The answer to this question is that it is as useful in equation (6.9) as
it is in equation (6.7). It is not necessary to assume that equation (6.7)
is only correct when the primary units of length and time are the metre
and the second. It is correct for whatever other units of length and time
we care to choose. It is for example correct if we choose the foot and the
minute. The unit of velocity which is then implied by equation (6.10)
is the velocity of an object travelling 1000 feet in 60 hours. This is a
valid unit, even if it is not practical. (See also the discussion following
equation (5.21).)
Reminders and extensions
CHAPTER 13

Reminders and extensions

13.1. Differentiation under the integral sign


We have repeatedly applied the following theorem:
T HEOREM 13.1. Given
Z x 2 (t )
I (t) = f (x, t) d x (13.1)
x 1 (t )
where x 1 (t) and x 2 (t) are differentiable for t1 ≤ t ≤ t2 , and where f and ∂ f /∂t
are continuous for t1 ≤ t ≤ t2 and x 1 (t) ≤ x ≤ x 2 (t), then:
Z x2 (t )
dI ∂f d x2 d x1
= d x + f (x 2 , t) − f (x 1 , t) . (13.2)
dt x 1 (t ) ∂t dt dt
For a proof, see [34], section 39 (see also problem 1.7). The proof makes use
of the fact that ∂ f /∂t is uniformly continuous. Uniform continuity is explained
in [34], section 13 and in more detail in [10], section 4.6.
There are also theorems for differentiation under the integral sign for multiple
integrals. Let for example f (x, y, α) be a function of the variables x, y, and the
parameter α, and let R be an area in the (x, y)-plane. The integral
ZZ
F(α) = f (x, y, α) d x d y
R
defines a function of α and d F/dα can, under certain conditions, be written
ZZ
dF ∂f
= d x d y. (13.3)
dα R ∂α
See [34], section 98.

13.2. Problems on extrema with constraints


It is often necessary to identify the points at which a function of several vari-
ables has extremal values, given that the variables satisfy a constraint. An example
of such a problem consists in finding the shortest distance from the origin to the
area x 2 − z 2 − 1 = 0. Since the square of the distance from the origin to an arbi-
trary point (x, y, z) is x 2 + y 2 + z 2 the problem consists in finding the minimum
value of
f (x, y, z) = x 2 + y 2 + z 2 (13.4)
given that
g(x, y, z) ≡ x 2 − z 2 − 1 = 0. (13.5)
Such problems can be solved by applying the following theorem.
T HEOREM 13.2. Given that
(1) f (x 1 , x 2 , x 3 ) and g(x 1, x 2 , x 3 ) have continuous partial derivatives of the
first order,
(2) f is extremal at a point P,
(3) x 1 , x 2 , x 2 obey g = k,
151
152 13. REMINDERS AND EXTENSIONS

(4) the gradient of g does not vanish at P,


then there exists a constant λ such that it and the coordinates of P satisfy
∂F
= 0, (i = 1, 2, 3)
∂ xi (13.6)
g = k,
where
F = f − λg.
For a proof, see [36, section 17.3]. The parameter λ in the system of equa-
tions (13.6) is called a Lagrange multiplier.
With f and g given by the equations (13.4) and (13.5), system (13.6) becomes
2x + λ 2x = 0
2y = 0
2z − λ 2z = 0
x 2 − z 2 − 1 = 0.
which has two real solutions:
x = ±1
y=0
z=0
λ = −1.
The theorem above can be extended to accommodate more variables and more
constraints, as follows.
T HEOREM 13.3. Given that
(1) f (x 1 , x 2 , . . . , x n ) and g j (x 1 , x 2 , . . . , x n ), where j = 1, . . ., m < n,
have continuous partial derivatives of the first order,
(2) f is extremal at a point P,
(3) x 1 , . . ., x n obey g j = k j ( j = 1, . . . , m),
(4) the gradient of g j does not vanish at P, for any j ,
(5) the gradient of gi at P is different from the gradient of g j at P, for any
i and j ,
then there exist constants λ1 , . . . , λm such that they and the coordinates of P sat-
isfy
∂F
= 0, (i = 1, . . . , n)
∂ xi (13.7)
g j = k j , ( j = 1, . . . , m)
where
m
X
F= f − λjgj.
j =1

For a proof, see [34, section 91].

13.3. The divergence theorem and some extensions


The divergence theorem is familiar from elementary analysis. It is given below
for the sake of completeness.
T HEOREM 13.4. In a three-dimensional space, let D be the region within a
closed and piecewise smooth surface S, and let n be the unit normal at a point of
13.4. PROOF STRATEGIES 153

S, pointing out of D. Let the Cartesian coordinates (F1 , F2 , F3 ) of a vector field


F, and ∂ F1 /∂ x 1 , ∂ F2 /∂ x 2 , ∂ F3 /∂ x 3 , be continuous in D and on S. Then
ZZZ ZZ
∇ · F dV = F · n dσ. (13.8)
D S
For a proof see [36] or [34]. The theorem can be extended from 3 to n dimen-
sions. With 2 dimensions, for example, it becomes (see [36, section 17.5])
ZZ I
∇ · F dσ = F · n ds, (13.9)
D S
where F now has the two components F1 (x 1 , x 2 ) and F2 (x 1 , x 2 ), S is a simple
closed curve in the plane and the integration along S is performed in the positive
direction. D is the region inside the curve, and n is the unit normal at a point of S,
pointing out of D.
The divergence theorem can also be used to extend the formula for partial
integration. For an example of this, start with equation (13.9) which can be written
ZZ   I
∂ F1 ∂ F2
+ d x 1 d x 2 = (F1 n 1 + F2 n 2 ) ds, (13.10)
D ∂ x1 ∂ x2 S
where n 1 and n 2 are the components of n. Assume
F1 = η A1 , F2 = η A2
where the functions η(x 1 , x 2 ), A1 (x 1 , x 2 ), and A2 (x 1 , x 2 ) have the same properties
of continuity and differentiability as F1 and F2 . Equation (13.10) then becomes
ZZ   ZZ  
∂η ∂η ∂ A1 ∂ A2
A1 + A2 d x1 d x2 = − η + d x1 d x2
D ∂ x1 ∂ x2 ∂ x1 ∂ x2
I D
+ η(A1 n 1 + A2 n 2 ) ds, (13.11)
S
which can be seen as a generalization of the formula for partial integration.

13.4. Proof strategies


Some of the notations used in this section are taken from logic. These nota-
tions are defined as follows:
A ∧ B : A and B,
A ∨ B : A or B,
¬A : not A,
A ⇒ B : A implies B,
A ⇐⇒ B : A is equivalent to B,
∀x : For all x,
∃x : There is a value of x.
All theorems are of the form P ⇒ C where P is the premise (or a set of premises)
and C the conclusion. Formally then, a theorem is in general a statement with n
premises and a conclusion,
Theorem T : P1 ∧ P2 ∧ . . . ∧ Pn ⇒ C,
and there is a proof that the conclusion follows from the premises.
There are several proof strategies [38]. We will here look at two of these in
the hope that the proofs which have been explicitly carried out in previous chapters
will become easier to understand. To clarify the strategies we will apply them to
the following theorem, using Theorem T as a template:
T HEOREM 13.5. The premises (P1 and P2 ) and the conclusion (C) are:
P1 : f (x) is defined and continuous for all real values of x.
154 13. REMINDERS AND EXTENSIONS

Rb
P2 : a f (x) d x = 0 for all a and b.
C: f (x) = 0 for all x.
We will now look at two completely different methods of proving this theorem,
namely direct proof and proof by contradiction.
13.4.1. Direct proof. A direct proof of Theorem T is based on the belief that
a good strategy consists in starting with the premises of the theorem and to work
towards a goal which is the conclusion of the theorem:
Start: P1 , P2 ,. . . , Pn . Goal: K .
Let us prove Theorem 13.5 by applying the direct proof method. The P1 -premise
enables us to define F(x) by
dF
= f (x).
dx
The set of functions satisfying this equation is infinite, and it is known that the
difference between two such functions is a constant [36]. Let F be an arbitrary
member of the set. There is a theorem which guarantees that F is continuous [36].
Another known theorem enables us to write
Z b
f (x) d x = F(b) − F(a).
a
The P2 -premise now implies that F(b) = F(a) for all a and b. This means that
F is constant, giving d F/d x = 0 for all x. But d F/d x = f (x), which means that
we have proved that f (x) = 0 for all x.
13.4.2. Proof by contradiction. A proof by contradiction is based on the
following logical statement:
A⇒B ⇐⇒ A ∧ ¬B ⇒ σ,
where σ is a contradiction (a statement which is always false). This gives an
alternative strategy for proving Theorem T: assume that K is false and then proceed
to demonstrate that this results in a contradiction. This method of proof can be
expressed as follows:
Start: P1 , P2 ,. . . , Pn , and ¬K . Goal: Contradiction.
Thus ¬K is now one of the assumptions. When our goal is reached, that is, when
we have arrived at a contradiction, we conclude our argumentation as follows:
Since we have a contradiction, and since we assume that all P1 , . . . , Pn are true,
then ¬K must be false. Consequently, K must be true.
Note that, in this strategy, an expression for ¬K must be found, given K . This
is not always straightforward 1 and reference to literature on Discrete Mathematics,
for example[38], is recommended. In theorem 13.5 K is the statement " f (x) = 0
for all x", which is of the type "∀x P(x)". It can be shown that
¬∀x P(x) ⇐⇒ ∃x ¬P(x),
so that ¬K in theorem 13.5 is the statement "There exists a value of x where
f (x) 6= 0"2.
Let us now prove theorem 13.5 by the method of proof by contradiction. We
introduce the premise P3 : there is a value of x, let us call it x 0 , such that f (x 0 ) 6= 0.
Then f (x 0 ) is either > 0 or < 0. Premise P1 implies that there is an interval

1For example the negation of “it is raining” is obvious. For most people, however, the negation
of “I hate all cats” is not easy to find.
2With reference to the previous footnote, the negation of “I hate all cats” is: “There exists a cat
which I do not hate”.
13.4. PROOF STRATEGIES 155

[a0 , b0 ] such that f (x) has the same sign as f (x 0 ) for all values of x in the interval.
This again implies that
Z b0
f (x) d x 6= 0,
a0
which contradicts the premise P2 . Since we assume premises P1 and P2 to be true,
P3 must be false. Consequently, f (x) must equal 0 for all x.
We can see that theorem 13.5 is just about as easy to prove by direct proof
as by proof by contradiction. However, the proof by contradiction is much to be
preferred because it can easily be applied to a generalization of the theorem to
more than one dimension. An example of such a generalization is as follows:
T HEOREM 13.6. The premises (P1 and P2 ) and the conclusion (K ) are:
P1 : Rf (x, y) is defined and continuous for all real values of x and y.
P2 : D f (x, y) d x d y = 0 where D is an arbitrary region in the plane.
K : f (x, y) = 0 for all x and y.
(We recognize the Du Bois-Reymond lemma for two dimensions, see section 1.6.)
An attempt at direct proof runs into major complications while the proof by con-
tradiction works without problems.
Bibliography

[1] M. Abramowitz and I. Stegun: Handbook of Mathematical Functions, Dover, 1970.


[2] R. Aris: Vectors, Tensors, and the Basic Equations of Fluid Mechanics, Prentice-Hall, 1962.
[3] D. K. Arrowsmith and C. M. Pace: Ordinary Differential Equations, Chapman and Hall, 1982.
[4] H. A. Barnes, J. F. Hutton, and K. Walters: An introduction to rheology, Elsevier, 1989.
[5] G. K. Batchelor: An introduction to fluid dynamics, Cambridge University Press, 1967.
[6] R. B. Bird, R. A, Armstrong, and O. Hassager: Dynamics of polymeric liquids, Vol. 1, Wiley,
1987.
[7] M. Braun: Differential Equations and their Applications, Springer Verlag, 1984.
[8] P. W. Bridgman: Dimensional analysis, Yale Univesity Press, 1931.
[9] H. S. Carslaw and J. C. Jaeger: Conduction of Heat in Solids, 2nd. ed., Clarendon, 1959.
[10] C. Clark: Elementary Mathematical Analysis, Wadsworth, 1972.
[11] C. H. Edwards, Jr., and D. E. Penney: Calculus and Analytic Geometry, Prentice-Hall, 1982.
[12] N. H. Fletcher and T. D. Rossing: The physics of musical instruments, 2nd edition, Springer,
1998.
[13] I. M. Gelfand and S. V. Fomin: Calculus of variations, Prentice Hall, 1963.
[14] P. Glendinning: Stability, instability, and chaos: an introduction to the theory of non-linear dif-
ferential equations, Cambridge University Press, 1994.
[15] H. Goldstein: Classical Mechanics, Addison-Wesley, 1964.
[16] I. S. Gradshteyn and I. M. Ryzhik: Tables of Integrals, Series, and Products, Academic Press,
1965.
[17] R. B. Guenther and J. W. Lee: Partial Differential Equations of Mathematical Physics and Inte-
gral Equations, Prentice-Hall, 1988.
[18] J. O. Hirschfelder, C. F. Curtiss, and R. B. Bird: Molecular theory of gases and liquids, Wiley,
1954.
[19] W. F. Hughes and J. A. Brighton: Theory and Problems of Fluid Dynamics, Schaum’s Outline
Series, McGraw-Hill, 1967.
[20] H. G. Jerrard and D. B. McNeill: A Dictionary of Scientific Units, Including Dimensionless num-
bers and Scales, 5th. ed., Chapman and Hall, 1986.
[21] D. W. Jordan and P. Smith: Nonlinear Ordinary Differential Equations, Clarendon Press, 1977.
[22] M. S. Klamkin (ed.): Mathematical Modelling: Classroom Notes in Applied Mathematics, SIAM,
Philadelphia, 1987.
[23] H. Krogstad: Forelesningsnotater i Matematisk Modellering, Universitetet i Trondheim, 1985.
[24] H. L. Langhaar: Dimensional Analysis and the Theory of Modelling, John Wiley, 1951.
[25] L. D. Landau and E. M. Lifshitz: Fluid Mechanics, Pergamon Press, 1987.
[26] C. C. Lin and L. A. Segel: Mathematics Applied to Deterministic Problems in the Natural Sci-
ences, Macmillan, 1974.
[27] A. R. Mitchell and R. Wait: The Finite Element Method in Partial Differential Equations, Wiley,
1977.
[28] P. M. Morse and H. Feshbach: Methods of Theoretical Physics, McGraw-Hill, 1953.
[29] N. de Nevers: Fluid Mechanics, Addison-Wesley, 1970.
[30] J. R. Newman: The World of Mathematics, Simon and Scuster, 1956.
[31] F. Oberhettinger and L. Badii: Tables of Laplace Transforms, Springer-Verlag, 1973.
[32] F. Reif: Fundamentals of statistical and thermal physics, McGraw-Hill international edition,
1985.
[33] E. B. Saff and A. D. Snider: Fundamentals of complex analysis for mathematics, science and
engineering, Prentice-Hall, 1993.
[34] I. S. Sokolnikoff: Advanced Calculus, McGraw-Hill, 1939.
[35] I. Stakgold: Green’s Functions and Boundary Value Problems, Wiley, 1979.
[36] G. B. Thomas, Jr. and R. L. Finney: Calculus and Analytic Geometry, 7th ed., Addison-Wesley,
1988.
[37] A. N. Tychonov and A. A. Samarski: Partial Differential Equations of Mathematical Physics,
Holden-Day, 1964.

157
158 BIBLIOGRAPHY

[38] D. J. Velleman: How to prove it, Cambridge University Press, 1994.


[39] R. C. Weast, ed.: Handbook of Chemistry and Physics, Chemical Rubber publishing Company,
1986-1987.
[40] R. Weinstock: Calculus of Variations, with Applications to Physics and Engineering, Dover,
1974.
[41] J. R. Welty, C. E. Wicks, and R. E. Wilson: Fundamentals of Momentum, Heat and Mass Transfer,
John Wiley, 1969.
[42] G. B. Whitham: Linear and Nonlinear Waves, John Wiley, 1974.
[43] E. T. Whittaker and G. N. Watson: A Course of Modern Analysis, Cambridge University Press,
1965.
[44] W. E. Williams: Partial Differential Equations, Clarendon, 1980.
[45] E. C. Zachmanoglou and D. W. Thoe: Introduction to partial differential equations with applica-
tions, Dover, 1986.

You might also like