Vector Analysis

—
32VAN Part II

Jan van Dijk

Version 2024/2025
2

Copyright 2019–2024 Eindhoven University of Technology.

All rights reserved. No parts of this work may be reproduced or distributed in any form or by any means,
or stored in a database or retrieval system, without written permission of the copyright holder.
Acknowledgments

The author would like to thank the following students for making valuable comments on the manuscript and
pointing out spelling mistakes and more substantial errors. In alphabetical order:
Haydar Almass, Edgar Ansems, Yngwie Baron, Stijn Borgers, Daan Brinkhof, Luc Brinkman, Stan van
der Burg, Mihály Calis, Mourad Doudouh, Harun Džafić, Laurenz Edelmann, Ben van Elderen, Ferdy Ellen,
Koen van Erve, Bart Goelema, Ruben van de Guchte, Nevin Heeman, Lennart Heijnen, Patrick Huisman,
Jurre de Jonge Mulock Houwer, Anouk Jurewicz, Jorn Kloosterman, Giovanni ’t Lam, Maikel Magnée,
Octavio Melendez, Danı̈el van der Merwe, Orson van der Molen, Maikel Morren, Marieke Mulders, Lucas
Norg, Yarno Peeters, Iris Podbevšek, Shaughn Prickarts, Elise Relyveld, Yannic Risters, Dennis van de
Sande, Jean-Luc van der Schaft, Bart Schepens, Remy Scheren, Jesse Smeets, Máire van der Tak, Robert-
Jan Thuis, Amaan Valiuddin, Dennis van Velthoven, Stoyan Vercruysse, Gijs Vermariën, Emma Vignoli,
Theo Witkamp, Koen Wuite, Fenn Zeelenberg.

3
4
Contents

1 Introduction 9
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2 The Geometry of our Universe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 Synthetic Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.4 Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.5 Coordinate Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.5.1 Cartesian Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.5.2 The Space Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.5.3 Polar and Spherical Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.6 Non-Euclidean Geometry, General Relativity . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2 Vectors Products and Reciprocal Vectors 21

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2 Vector Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.1 The dot product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.2 The cross product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2.3 The scalar triple product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2.4 The vector triple product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2.5 The scalar quadruple product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2.6 The vector quadruple product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3 Reciprocal Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.4 Points, Lines and Planes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.5 Summary and outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

3 Curves 31
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2 Vector-Valued Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2.2 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2.3 Differentiating vector-valued functions of one parameter . . . . . . . . . . . . . . . . . 32
3.3 Parameterized Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4 Velocity and Speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.5 The Arc Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.6 Re-parameterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.7 The Arc Length Parameterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

5
6 CONTENTS

4 Coordinate Systems 41
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.2 Coordinate Systems, Coordinate Lines and Surfaces . . . . . . . . . . . . . . . . . . . . . . . 43
4.3 Coordinate Transformations, the Jacobi Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.4 The Covariant Basis, Contravariant Vector Components . . . . . . . . . . . . . . . . . . . . . 44
4.5 Dot Products and the Metric Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.6 The Contravariant Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.7 Derivatives of Basis Vectors — the Christoffel Symbols . . . . . . . . . . . . . . . . . . . . . . 50
4.8 From Covariant to Contravariant and Vice Versa . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.9 Ortho-Curvilinear Systems; Physical Components and Basis Vectors . . . . . . . . . . . . . . 51
4.10 Volume and Surface Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.11 Summary and Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

5 Flux Densities & the Divergence of a Vector Field 55

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.2 Flux Densities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.3 Orientable and Non-Orientable Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.4 Examples: Balance Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.4.1 The Continuity Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.4.2 The Species mass Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.5 The Divergence — Coordinate-Free Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.6 The Divergence — Differential Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.7 The Divergence in Ortho-Curvilinear Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.8 The Divergence Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.9 Summary and Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

6 The Gradient of a Scalar Field, Line Integrals 67

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
6.2 The Gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
6.3 The Material Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
6.4 Results for Particular Coordinate Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
6.5 The Directional Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
6.6 Line Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
6.7 Application: a Particle in a Force Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
6.8 Summary and Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

7 The Curl of a Vector Field 77

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
7.2 The Maxwell-Faraday Law of Induction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
7.3 Derivation of the curl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
7.4 Results for Particular Coordinate Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
7.5 The Curl of Some Concrete Vector Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
7.6 Physical Interpretation of the Curl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
7.7 The Kelvin-Stokes Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
7.8 Summary and Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
CONTENTS 7

8 Combinations of Vector Operators, The Laplacian 87

8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
8.2 Properties of the Gradient, Divergence and Curl . . . . . . . . . . . . . . . . . . . . . . . . . 88
8.3 The Laplacian of a Scalar Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
8.4 The Laplacian of a Vector Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
8.5 Summary and Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

9 Electrostatics, the Poisson Equation 93

9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
9.2 Charge and Current Densities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
9.3 Point Sources, the Three-dimensional Delta Function . . . . . . . . . . . . . . . . . . . . . . . 94
9.4 Coulomb’s Law Revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
9.5 Gauß’ Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
9.6 The Displacement Current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
9.7 Summary and Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

10 Helmholtz’ Decomposition Theorem 99

10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
10.2 Solenoidal and Irrotational Fields, Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
10.3 Helmholtz Decomposition Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
10.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

11 The Dyadic Product, Dyads and Dyadics — Tensors 103

11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
11.2 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
11.3 Examples of Dyads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
11.3.1 Projections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
11.3.2 The Momentum Flux Density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
11.4 Dyadics, Coordinate Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
11.5 The Unit Dyad . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
11.6 The Gradient of a Vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
11.7 The Momentum Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
11.8 The Divergence of a Tensor Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
11.9 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
8 CONTENTS
Chapter 1

Introduction

Objectives
After studying this chapter, you will be able to
• Explain the essential properties of a Euclidean space (flatness, translation- and rotation-invariance);

• Explain the meaning of the term ‘geometrical object’;

• Understand vectors as displacements, defined between two points;
• Understand the relation between a column matrix such as v in R3 and the corresponding vector v;
• Can derive expressions for the basis vectors of coordinate systems in terms of those of a reference
coordinate system.

9
10 CHAPTER 1. INTRODUCTION

1.1 Introduction
What is a vector? In your studies so far you have had multiple encounters with vectors. In secondary school
you have accepted a vector as being an arrow, representing a quantity such as the velocity of a particle or the
force acting on it. Such a vector has a length, later also called the norm or the magnitude, and a direction.
You then learned how to scale and rotate vectors, and to add them using the ‘parallelogram law’. Later you
learned to think of vectors in terms of components, and something like a velocity then ‘became’, for example,
(vx , vy , vz ), assuming a Cartesian coordinate system.
In your more recent studies you have learned how the concept of vectors could be expanded to completely
different types of objects, such as functions of a particular class. You have seen that, by defining an inner
product, even concepts that seem to have a purely geometrical significance could be given a useful meaning
in such vector spaces, such as the norm of a function or the orthogonality of a pair of them.
In this course we will move our attention back to geometric vectors. More specifically, we will study
vectors in a flat three-dimensional Euclidean space. We continue the present chapter with a discussion of
Euclid’s postulates and the concept of ‘flatness’. Vectors will be introduced as a more convenient language
to express problems in geometry. Various representations will be discussed and the relation between the
geometric vectors and the space R3 of column matrices will be discussed. This discussion will use the
Cartesian, polar and spherical coordinate systems with which you ought to be familiar already. The chapter
concludes with a brief discussion on non-Euclidean geometries.
In chapter 2 we will lay the groundwork for our further studies of vectors and vector-valued functions. We
will define the dot, cross and scalar triple products of vectors and use these to calculate lengths, surfaces and
volumes. Three more products of three and four vectors will be shown as well, and applications to classical
geometrical problems such as projections and intersections of lines and planes will be demonstrated in the
form of exercises.
In chapter 3 we define a curve as a vector-valued continuous function of a single scalar parameter. Various
properties of curves will be introduced, such as a curve’s velocity, speed and arc length parameterization.
In your previous studies you have dealt (almost) exclusively with orthonormal coordinate systems. These
are systems with mutually orthogonal basis vectors of unit length. In chapter 4 we will provide a much
more general view on coordinate systems and discuss the important concepts of covariant and contravariant
basis vectors and components and the metric tensor. In addition, the physical component representation
will be presented for the special case of locally orthogonal coordinate systems. This chapter builds on your
knowledge of multivariable calculus, in particular transformations and the Jacobian matrix.
In chapters 5–7 we introduce three key concepts in vector calculus: the divergence of a vector field, the
gradient of a scalar field and the curl of a vector field. These are of the utmost importance in physics, and
we will motivate their definitions by starting from examples from physical reality. All expressions will be
developed in general coordinate systems, the expressions in Cartesian, cylindrical and spherical coordinates
that you should be familiar with are derived as special cases.
Chapter 8 presents the Laplacian of a scalar field. This particular combination of the divergence and the
gradient operators is omnipresent in physics, and a particular example of its application in electrostatics is
discussed in chapter 9.
Throughout the text the existence of potentials is observed a number of times. The text concludes with
an exposition of Helmholtz’ decomposition theorem, which states that every vector field can be written in
terms of a scalar and a vector potential. The subject of potential theory is important for physicists because
more often than not you will investigate the properties of a vector field by inspecting its potential(s), rather
than the fields themselves.
Chapter 11 concludes this text. It touches upon the subject of 2-tensors. The presentation is based on
Gibb’s concept of the dyadic product of two vectors and dyadics. The reason is that that presentation fits
in smoothly with the rest of the material in this lecture, and more importantly: that presentation is still
found in many books on applied physics, in particular fluid dynamics and electrodynamics. The concepts
1.2. THE GEOMETRY OF OUR UNIVERSE 11

are demonstrated by introducing the concepts of the momentum flux density and the momentum balance
equation for a fluid.

1.2 The Geometry of our Universe

The vectors that we will study in this course are the geometrical vectors that ‘live in our three-dimensional
space’. That is, of course, quite a vague statement: what is that three-dimensional space that we live in
exactly, and what are its properties? And what are these vectors anyway?
Geometry is a branch of mathematics that resembles physics, in the sense that its development is based
on measurable quantities. This is reflected by the Dutch translation of the word, meetkunde. In the case
of geometry, there are two such quantities, lengths and angles. The measurement of length is based on
comparison of a given length by comparing that with a given reference value. The same goes for angles: a
reference value can be defined as the right angle, which we take to be 90◦ or π/2 rad , or as the ratio of the
circumference and radius of a circle, which we take to be 2π rad .
After deciding about the quantities that are in the scope of our studies, we can start to build a model.
This is a set of basic assumptions and rules to combine those assumptions into more complex statements,
in this case about lengths and angles. Such a model, if it is any good, should be in accordance with our
observations of the physical world.
The model for the geometry of our world can be expressed in multiple ‘languages’. In later chapters we
will follow the approach with which you will probably feel most comfortable: we will adopt a coordinate
system, like a Cartesian system, and understand vectors in terms of their components with respect to such a
coordinate system. That will allow us to calculate such things as lengths of vectors, distances between two
points and angles between pairs of vectors.
In this chapter we will provide a conceptual foundation for such operations, which serves two purposes.
Firstly, this chapter will emphasize the fact that you do not need to introduce a coordinate system in order
to do such calculations. The concepts that are expressed by vectors, such as the local velocity of a fluid
element at a given time, or the force that acts on a particle, existed also before Descartes invented coordinate
systems in the 17th century [1].
The second purpose of this chapter is to make you realize the extent to which your view on physical
reality, and in particular on the structure and properties of the three-dimensional universe we live in, has
been shaped by an extrapolation of our limited experience as ‘earthlings’.
These two goals will be served by a brief exposition of the development of ‘Euclidean geometry’, which
models the properties of drawings in a flat piece of paper (in two dimensions), and a flat three-dimensional
space. We will demonstrate that ‘flatness’ is a property that must be defined and can then be measured, it
is no more objective than any other model we physicists and mathematicians have proposed to systematize
our experiences with the ‘real world’.
Geometry can be developed in various different ways. We will review these in this chapter in the order
in which they have been developed. This means that we will first look at Euclid’s method of Synthetic
Geometry. We will then discuss how geometric vectors can be introduced as displacements for given pairs of
points, and discuss the reason for the choice for an origin. In the last part we discuss Analytical Geometry, in
which vectors are described with reference to coordinate systems. We will discuss the concept of coordinates,
basis vectors and establish the relation between geometric vectors in an n-dimensional space and the space
Rn of column matrices with n elements. Cartesian and polar grids are discussed in this section.
We conclude the chapter with a brief discussion of non-Euclidean geometries: curved spaces in two and
three dimensions. That discussion will necessarily be very short, be hopefully just long enough to get you
interested and make you consider to enroll in the courses Tensor Calculus and Differential Geometry and
General Theory of Relativity.
12 CHAPTER 1. INTRODUCTION

1.3 Synthetic Geometry

We will first discuss the method of synthetic geometry, also called axiomatic geometry. This method was
established by Euclid of Alexandria, who lived around 300 BC. His book ‘Elements’ has dominated our
thinking of geometry for more than two millennia and it is enough to say, for example, Euclid’s I.32 to refer
to theorem 32 in the first book (chapter) of the Elements. It is the method that you used in high school
when you studied the geometry of points, lines, triangles et cetera.
The defining property of synthetic geometry is that it can be introduced without any reference to a
coordinate system. All you have available are definitions of basic objects, such as points, lines and circles,
and five postulates. The latter describe basic assumptions, that cannot be independently derived.1
Classical geometry is based on Euclid’s five postulates and to give you an idea of their basic, fundamental
nature we spell them out explicitly:

1. A straight line segment can be drawn between any pair of two points;

2. A straight line segment can be extended indefinitely into a straight line;

3. A line segment is the radius of a circle with one of the end points as its center;

4. All right angles are congruent.

The word ‘congruent’ means that one object (a right angle) can be transformed into the other by a trans-
lation, rotation and reflection. More generally, Euclidean geometry is based on the assumption that you
can translate, rotate and mirror objects without altering their nature. A triangle stays a triangle when it is
moved around, rotated or flipped, and all lengths and angles are conserved in the process.
There is one more postulate. Euclid’s fifth postulate, or parallel postulate has been formulated and re-
formulated in many equivalent ways, but perhaps the easiest statement is the following, due to Playfair [3,
p. 29, prop. XXXI]:

5. given, in a plane, a line and a point not on that line, there is only one line through that point that
does not intersect with the other line.

On the basis of these postulates, Euclid was able to prove many well-known results. We highlight two of
these, which we will also revisit when we start to talk about vectors a bit later in this text. The first is
Euclid’s theorem I.32, which states that the sum of the angles of a triangle is equal to 180◦ (‘two right
angles’). The second is Euclid’s I.47, which we all know as the theorem of Pythagoras.

Exercise 1.1 Make sure you realize what Euclid’s postulates say: please take a pen and a piece of paper and
draw points, lines and circles to visualize each of Euclid’s five postulates.

Exercise 1.2 Draw a square. Draw a second square with smaller sides that is rotated (30 degrees or so), such
that the corner points of the second square are located on the sides of the first square. Let c denote the length
of the sides of the small square. The corner points divide the sides of the big square in line segments with
lengths a and b, say. Calculate the size of the large square in terms of a and b. Equate the result to the sum
of the areas of the small square and the triangles that have appeared in the construction. Derive Pythagoras’
law.
1 For completeness sake, in Euclid’s work also a number of ‘common notions’ or axiomata appear. The interested student is

referred to one of the many good sources on Internet on the subject, or to reference [2, ch. 2]. Also note that the words axiom
and postulate are used interchangeable in some scientific communities.
1.4. VECTORS 13

a
a+b b b a+b
b a+b
a a
(a) Successive displacement: a + b (b) Successive displacement: b + a (c) The parallelogram law

Figure 1.1: Addition of two vectors can be achieved via the parallelogram law. The effect is the same as when the
two displacements a and b are applied successively in any order (figures c and d). This construction of vector
addition by successive applications makes the fact that translations of the vectors are required.

1.4 Vectors
Consider two points A and B in a Euclidean space. We can then draw a line segment that originates from
−−→
the point A and ends at point B. Such a line segment has a length and a direction and can be written as AB
or with a boldface small letter such as v. We call such a directed line segment a vector. 2 When we draw a
vector we use an arrow, rather than a line segment, to indicate its direction.
−−→
The vector v = AB has been explicitly constructed by drawing it from the point A. Such a vector, with
particular initial and end points is called a bound vector. But we see something interesting: v is an amount of
displacement in a particular direction that we could also apply to another point in space. When we interpret
the vector as a displacement, essentially ‘forgetting’ the point from which it originates, we speak of v as a
free vector. When we speak of vectors, we usually speak of free vectors, and we say that two vectors are equal
if they express the same displacement, that is: if they have the same magnitude and the same direction.
We can define what it means to add a (displacement) vector v to a point: this is the operation of applying
the displacement to a point A, say, resulting in a new point B. The operation of adding two vectors can also
be given a meaning: it is defined as the displacement vector that results from the successive application of
the two vectors. In a Euclidean space the order in which this operation is carried out does not matter, so
vector addition is commutative,
a + b = b + a. (1.1)
This is illustrated in figures 1.1(a) and 1.1(b). The result is equivalent to what you get from the parallelogram
law, illustrated in figure 1.1(c).
Let us now consider some other operations of vectors. Vectors can also be multiplied with a scalar,
subtracted and rotated. We will not dwell too much on these operations. Instead we refer to figure 1.2,
where the various operations have been summarized.
We have discussed the definitions of adding vectors to points, and vectors to vectors. But can we also
add points to points? In order to give that operation a meaning we need to take additional measures. We
2 A quick search on the Internet and in literature reveals the following, apparently contradictory explanations for the origin

of the word vector. The historians among you may want to please their beloved teacher by helping me to sort out this mess, so
I can improve the syllabus a little:
• Kreyszig [4, page 9]: “The concept of a vector was first used by W, Snellius (1581-1626) and L. Euler (1707-83).”
• [Link] “The word VECTOR (which, like the word vehicle, derives ultimately from the
Latin vehĕre to carry) was first a technical term in astronomical geometry. The OED’s earliest entry is from a technical
dictionary of 1704: J. Harris Lexicon Technicum I. s.v., “A Line supposed to be drawn from any Planet moving round a
Center, or the Focus of an Ellipsis, to that Center or Focus, is by some Writers of the New Astronomy, called the Vector;
because ’tis that Line by which the Planet seems to be carried round its Center.” (NOTE: OED is the Oxford English
Dictionary.)
• [Link] “The term vector was introduced by William Rowan Hamilton as
part of his system of quaternions...” (mid 19th century).
Same page: “It was first used by 18th century astronomers investigating planet rotation around the Sun. ref: The Oxford
English dictionary. (2nd. ed.). London: Claredon Press. 2001. ISBN 9780195219425.”.
14 CHAPTER 1. INTRODUCTION

a
2a
a a
−a a

(a) translation (b) rotation over 90◦ (c) Negation (d) Scaling

Figure 1.2: Various operations on vectors.

need to define a special point in space, which we will call the origin and write as O. Once we have chosen an
−−→
origin, we can construct, for every point P in space, the displacement vector OP . We call this special vector
the position vector x(P ) of point P . We can then define that the addition of two points P1 and P2 results
in the point that is obtained by adding both their position vectors to the point O.
−−−→
Exercise 1.3 Please draw two points P1 and P2 and the displacement vector P1 P2 . Next, draw a third point
−−→ −−→
O and the position vectors OP1 and OP2 . Also draw the sum of these position vectors. Observe that this
−−−→
vector depends on the choice for O, whereas the vector P1 P2 does not.

It is important to realize that the choice for an origin is arbitrary. It is therefore to be expected that our
laws of nature should be formulated in such a way that measurable quantities do not depend essentially on a
particular choice. Later we will equip space with a coordinate system, and describe vectors in terms of their
components and the basis vectors of that coordinate system. We will then make this same comment: the
universe is not affected by the choice for a particular type of coordinate system, its origin or its orientation,
and this fact (or belief?) should be reflected by the equations that constitute our model for the physical
world.

1.5 Coordinate Systems

Let us now see how coordinate systems can be used to express vectors in terms of components with respect
to a given basis. We will first review Cartesian systems, already well-known to the readers of this text.
The reader is also assumed to be familiar with the dot product of two vectors a and b, which is defined as
a · b = ||a||||b|| cos χab , where ||a|| denotes the length of a end χab is the angle between the two vectors.
Another important property of the dot product that we need in this discussion is its linearity in both
arguments. We will discuss the dot product in more detail in section 2.2.1.

1.5.1 Cartesian Coordinates

In a two-dimensional Euclidean space, we can define an origin O and set up two mutually perpendicular axes
(lines) xex and yey that pass through this origin. This is shown in figure 1.3(a). In a three-dimensional
space we introduce a third line, zez , perpendicular to both ez and ey (figure 1.3(b)). The vectors ex , ey
and ez are chosen to be normalized, i.e. they have unit length. With these definitions in place, the position
vector x of a point P can be realized by three successive displacements parallel to these axes,

x = xex + yey + zez . (1.2)

1.5. COORDINATE SYSTEMS 15

(x,y,z )

(0,0)
O z
x
y Y

(a) 2D (b) 3D

Figure 1.3: A Cartesian coordinate system for a two- and three-dimensional geometry. (Based on: https:
// en. wikipedia. org/ wiki/ Cartesian_ coordinate_ system — public domain.)

Given this expression for the position vector, the basis vectors of the coordinate system can be obtained by
taking the partial derivatives of x with respect to the coordinates,
∂x ∂x ∂x
ex = , ey = , ez = . (1.3)
∂x ∂y ∂z
Furthermore, the coordinates x, y and z can be obtained by taking the dot product of its position vector x
with the basis vectors: using the linearity of the dot product and the fact that the Cartesian basis vectors
are orthogonal, we get
x = x · ex , y = x · ey , z = x · ez . (1.4)
Also other vectors, such as the electric field E at some position P , or the flow velocity u in that point can
be expressed in terms of these basis vectors. For some vector a we can write

a = ax ex + ay ey + az ez , (1.5)

where ax , ay and az are the components of vector a, with reference to the coordinate system.

Exercise 1.4 Consider two vectors a and b and their component-expressions for some given Cartesian co-
ordinate system. Please show that the dot product of these vectors is given by a · b = ax bx + ay by + az bz .

Exercise 1.5 Consider a Cartesian coordinate system and draw two normalized position vectors a and b in
the xy-plane at angles α and β with respect to the x-axis. Express both vectors in terms of the basis vectors
ex and ey . Calculate the dot product and equate the result to the geometrical definition of the dot product.
Show that cos(β − α) = cos α cos β + sin α sin β.

1.5.2 The Space Rn

In section 1.5 we introduced coordinate systems, but did not abandon the geometric perspective on vectors
that was developed in earlier sections: the vector x and the basis vectors are still the directed line sigments,
16 CHAPTER 1. INTRODUCTION

the arrows with lengths and directions that we defined them to be. The description in terms of coordinates
also allows us to adopt a purely algebraic view on vectors and operations on vectors. That is achieved by
grouping the Cartesian components of the position vector and the basis vectors in column matrices that are
elements of R3 . We find that
       
x 1 0 0
[x]c = y  , [ex ]c = 0 , [ey ]c = 1 , [ez ]c = 0 . (1.6)
z 0 0 1

The rules of matrix addition allow us to write the coordinate vector [x]c as
       
x 1 0 0
y  = x 0 + y 1 + z 0 ⇐⇒ [x]c = x[ex ]c + y[ey ]c + z[ez ]c . (1.7)
z 0 0 1

When there is no risk of confusion about the coordinate system that was used, the column matrix is also
written as x. Note that a roman font is used to distinguish the coordinate vector x from the geometric vector
x that it describes. If we also write ex := [ex ]c etc, the expression for the coordinate vector becomes

x = xex + yey + zez . (1.8)

Please make sure that you appreciate the difference between the vector x and the coordinate vector [x]c
(or x). The former is a geometric object, which describes the displacement from the reference point O to the
point P . The latter is a collection of numbers that describe this displacement, it derives its meaning from
the choice for the coordinate system c. The same goes for the relation between the basis vector ex and the
coordinate vector ex = [ex ]c , et cetera.
We cannot say that x (a geometric vector) is equal to x (an element of R3 ), but obviously there is a close
relationship between the space of Euclidean geometric vectors and the space R3 . It is easy to show that the
mapping of the geometric vectors to the column matrices in R3 is linear and bijective, and mathematically
we say that the space of Euclidean vectors is isomorphic to R3 . The relation between the vectors is denoted
as x ≃ x (or x ≃ [x]c ).
What has been said about the position vector x is also true for more general vectors such as a: we have
a ≃ a, the column matrix that contains the Cartesian components of a.

Exercise 1.6 Please show that a · b = aT b. (Note: aT represents the transpose of a.)

A result of this state of affairs is that the words Euclidean space and the space R3 are sometimes used
interchangeably. That is a bit confusing, since the space R3 comes into existence only after a Cartesian
coordinate system has been introduced, and it is not necessary to do so. In chapter 2 we will see that
many interesting results about points, lines, planes and vectors can be discussed without ever referring to
a particular coordinate system. And even when coordinates are used, other coordinate systems than a
Cartesian one may be used. As an example, we discuss polar and spherical coordinates.

1.5.3 Polar and Spherical Coordinates

In chapter 4 we will discuss curvilinear coordinate systems. Those are the most general coordinate systems,
with basis vectors that, in general, are neither normalized or orthogonal, and depend on position. We
will now discuss two very special non-Cartesian coordinate systems, polar coordinates for a two-dimensional
geometry, and spherical coordinates for a three-dimensional geometry. We will already use these coordinate
systems in our discussion of curves in chapter 3. You are of course already familiar with these two systems,
but perhaps not with the way we introduce them:
1.5. COORDINATE SYSTEMS 17

1. We consider a set of equations that relate the Cartesian coordinates of a pre-existing coordinate system
to the new coordinates. These equations are in general not linear;
2. The partial derivatives of the position vector with respect to these new coordinates are calculated,
using the chain rule and the equations for the coordinates;
3. The basis vectors are obtained by normalizing the resulting vectors.
Also in chapter 4 we will define basis vectors as the derivatives of the position vector with respect to the
coordinates, but the resulting vectors are used as such, without normalization.

Polar Coordinates
Let us consider polar coordinates r ∈ [0, ∞) and ϕ ∈ [0, 2π) for a two-dimensional geometry. These are
related to the Cartesian coordinates by the relations

x(r, ϕ) = r cos ϕ, (1.9)

y(r, ϕ) = r sin ϕ. (1.10)

The position vector is given by x = xex + yey . Let us now develop the expressions for basis vectors er and
eϕ in terms of the Cartesian basis vectors. As in equation (1.3), these are obtained as the normalized partial
derivatives of x with the coordinates. Since r and ϕ are both functions of x and y, we get
∂x ∂x ∂x ∂x ∂y
= + = +ex cos ϕ + ey sin ϕ, (1.11)
∂r ∂x ∂r ∂y ∂r
∂x ∂x ∂x ∂x ∂y
= + = −rex sin ϕ + rey cos ϕ. (1.12)
∂ϕ ∂x ∂ϕ ∂y ∂ϕ
The first vector already has length 1, the second has length r. The normalized basis vectors are then

er = +ex cos ϕ + ey sin ϕ, (1.13)

eϕ = −ex sin ϕ + ey cos ϕ, (1.14)

and for the position vector we find

x = ex r cos ϕ + ey r sin ϕ = rer (ϕ). (1.15)

In terms of the Rn language, the previous results can be expressed as follows:

     
cos ϕ − sin ϕ r cos ϕ
er = , eϕ = , x= . (1.16)
sin ϕ cos ϕ r sin ϕ

Exercise 1.7 Plaease verify that the basis is orthonormal: that er · er = eTr er = 1, eϕ · eϕ = eTϕ eϕ = 1 and
er · eϕ = eTr eϕ = 0

Spherical Coordinates
We consider spherical coordinates r ∈ [0, ∞), θ ∈ [0, π] and ϕ ∈ [0, 2π) for a three-dimensional geometry.
These are related to the Cartesian coordinates by the relations

x(r, θ, ϕ) = r sin θ cos ϕ, (1.17)

y(r, θ, ϕ) = r sin θ sin ϕ, (1.18)
z(r, θ, ϕ) = r cos θ. (1.19)
18 CHAPTER 1. INTRODUCTION

Exercise 1.8 Basis vectors for spherical coordinate systems.

1. Please derive expressions for the partial derivatives of x(r, θ, ϕ) with respect to the coordinates r, θ and
ϕ (in terms of the Cartesian basis);

2. Please show that ||∂x/∂r|| = 1, ||∂x/∂θ|| = r and ||∂x/∂ϕ|| = r sin θ.

3. Please show that the normalized basis vectors are given by

er = +ex sin θ cos ϕ + ey sin θ sin ϕ + ez cos θ, (1.20)

eθ = +ex cos θ cos ϕ + ey cos θ sin ϕ − ez sin θ, (1.21)
eϕ = −ex sin ϕ + ey cos ϕ. (1.22)

4. Just like for polar coordinates, the position vector in spherical coordinates only has a radial component.
Please show that
x(r, θ, ϕ) = rer (θ, ϕ). (1.23)

Exercise 1.9 Please provide the expressions for the coordinate vectors of er , eθ and eϕ . Use these expressions
to show that the spherical basis vectors are orthonormal.

Exercise 1.10 Show that the angle α between two position vectors on a sphere with spherical angles (θ1 , ϕ1 )
and (θ2 , ϕ2 ) is given by
cos α = sin θ1 sin θ2 cos(ϕ1 − ϕ2 ) + cos θ1 cos θ2 . (1.24)

This angle is called the central angle or the angular distance between two points on a sphere.

Exercise 1.11 The latitude and longitude of Eindhoven are 51◦ 26′ N, 5◦ 29′ E, Tokyo is located at 35◦ 41′ N,
139◦ 41′ E.

1. Please show that the angular distance between the two cities is equal to 84.1◦ , or 1.47 rad.

2. Show that the distance across the surface of the earth is approximately 9.4 × 103 km. Assume that the
earth is a sphere with radius 6371 km.

Exercise 1.12 Consider a particle with time-dependent coordinates r(t), θ(t) and ϕ(t) in a spherical coordinate
system. The position vector depends only on t (via the coordinates), it is given by x(r(t), θ(t), ϕ(t)).

1. Use the chain rule and the rules of partial derivation to demonstrate that

dx dr dθ dϕ
= er + eθ r + eϕ r sin θ . (1.25)
dt dt dt dt

2. Approximate the earth as a sphere and introduce a fixed spherical coordinate system with the center
of the earth at r = 0 and the North Pole at θ = 0, so the earth is rotating in this system (in the
ϕ-direction). What is the angular velocity dϕ/dt? What are the speeds of objects at rest on earth on
the equator and on the North Pole?
1.6. NON-EUCLIDEAN GEOMETRY, GENERAL RELATIVITY 19

1.6 Non-Euclidean Geometry, General Relativity

For Euclid himself, his postulates were self-evident, probably simply because they were in such perfect
agreement with his observations of physical reality. This situation persisted for two thousand years. In this
period various details of Euclid’s postulates and theorems were fixed, and some implicit assumptions were
spelled out more explicitly, but no essential changes were made. Noteworthy are the dozens of attempts to get
rid of the fifth postulate: many mathematicians thought that it should be a theorem instead, that is would
be possible to derive it from the other four. Various ‘proofs’ were developed, but all turned out to be flawed.
It took until the nineteenth century before this situation changed. It was then observed independently by
Schweikart, Gauß, Bolyai and Lobachevsky that consistent geometries could be defined by replacing the fifth
postulate with a different one [2, ch. 3]. A different choice for that postulate can then be interpreted as a
different model for the space we live in.

Exercise 1.13 Consider two meridians. Are these parallel? Do these intersect in any point?

The Euclidean geometry is that of a ‘flat space’. In two dimensions it is easy to imagine what that means:
Euclid’s fifth postulate is consistent with our observations of the behavior of lines on a plane — a flat piece
of paper, if you wish. It is also easy to visualize a curved two-dimensional surface, such as that of a sphere.
And on such a surface, Euclid’s fifth postulate cannot be expected to be valid. But how do we find out
whether the three-dimensional space in which we live is flat (Euclidean), or not? It is hard to imagine what
a curved three-dimensional space looks like. But once you have developed the mathematics of geometry, like
Euclid did, you do not need to. Instead, you can define a flat space as a space that is described by Euclid’s’
postulates (all five) and in which, as a consequence, his theorems can be verified by experiments. In other
words: we can construct triangles, big or small, and do experiments to test if their angles add up to 180◦ ,
and such things as Pythagoras’ theorem are valid.
As an example, imagine that you are an earthling, living in the times of Euclid, confined to the surface
of the earth, and skeptical about the new insight that the earth is a sphere, rather than a flat surface with
boundaries unknown and better avoided. How could you use Euclid’s theorems to find out whether the earth
is flat or not? This is the subject of the next exercises.

Exercise 1.14 The earth is flat. Or not?

1. Walk 1 meter forward. Then step aside one meter, to the right, and walk backwards one more meter.
Where do you end up, and what direction are you facing?
2. Repeat this exercise. But imagine you are initially on the equator, facing north, and walk a distance of
Rπ/2 each time, where R is the radius of the earth. Do you end up in the same place?
3. What is the sum of the three angles of the triangle that you have walked in the previous exercise?
4. Do the two previous results differ if you swap the second and third parts of the journeys? That is: are
these commutative?
5. Is the surface of the earth Euclidean (flat)?

When a space is non-Euclidean, some things start failing. We have already seen that finite-sized displace-
ment vectors are problematic (non-commutative), and since the position vector is merely such displacement
from an arbitrarily chosen ‘origin’, also the introduction of such special point is no longer useful. Also, on a
curved geometry it is not possible to set up a Cartesian coordinate system. If you try to imagine the issues
at hand, make very sure that you are aware of the difference between a curved space, and a curved structure
that is embedded in a higher-dimensional flat space. As an example, consider the position vectors x(θ, ϕ) of
points in a three-dimensional Euclidean space that lie on the (two-dimensional) surface of a sphere), with
20 CHAPTER 1. INTRODUCTION

the origin in the center of the sphere. Adding such three-dimensional position vectors is perfectly possible,
but will result in a point that is not on the spherical surface.
If we consider only the spherical surface (not embedded in a three-dimensional space), we run into
problems: we cannot meaningfully define an origin, position vectors and vector addition. But on the other
hand, the infinitesimal displacement that is associated with variations of the spherical angles will result in a
displacement that brings you to another point on the curved (spherical) surface,

∂x ∂x
dx = dθ + dϕ := aθ dθ + aϕ dϕ, (1.26)
∂θ ∂ϕ
and note that this addition, of vectors in the ‘tangent plane’ that is spanned by the vectors aθ and aϕ , is
is linear and commutative. This observation of ‘local flatness’ is one of the cornerstones of the treatment of
non-Euclidean spaces in differential geometry.
Until the beginning of the 20th century it was universally accepted that the Euclidean model was an
adequate description of our three-dimensional universe. This changed when Einstein formulated his theory
of general relativity. The common formulation of that theory is that our three-dimensional space does not
have a flat, Euclidean structure, but is curved by the presence of mass. Since this curvature manifests itself
only on cosmological scales, ‘physics on earth’ is described accurately by a Euclidean model of our space. Of
course, if this were not the case, Euclid himself would have discovered that his model was flawed.
This course deals with vectors and vector calculus in three-dimensional Euclidean space. But before you
feel cheated: much of the nomenclature, concepts, rules, and manipulations that you will familiarize yourself
with apply equally well to non-Euclidean space. As it turns out, the property of space that tells us whether
it is Euclidean or not is a very central and important part of the exposition that will follow, but rather small
at the same time. It is the metric of the space, as encoded by the metric tensor. We will get back to that
concept in a later chapter. If you want to learn more about curved space-time, we recommend that you
follow the TU/e elective course General Relativity — after finishing this course with good marks.
We conclude this section with two exercises that deal with measurements of distances. That is something
that you are so familiar with, that it is easy forget that even this concept is much more subjective than you
think. The following exercises will fix that.

Exercise 1.15 Let R denote the radius of the earth, which we model as a sphere. What is the distance between
the North Pole and the equator?

Exercise 1.16 The Euclidean and Manhattan distances between two points.
1. Roads in many American cities are organized as a rectangular grid of avenues in one direction and
streets in the other. If your destination lies 300 m to the east, and 400 m to the north, what distance
do you need to travel? Assume that you are a bird.

2. Idem. But this time assume that you are human.

Exercise 1.17 Look at the list of objectives at the beginning of this chapter and ensure that you have acquired
the mentioned skills. If not, please study the relevant parts of this chapter again and repeat this exercise.
Chapter 2

Vectors Products and Reciprocal Vectors

Objectives
After studying this chapter, you will be able to
• Use the dot product of two vectors to calculate a vector’s length and the angle between two vectors
and to express orthogonal and oblique projections in terms of dot products;

• Use the cross product of two vectors to calculate the area of the parallelogram that is spanned by those
vectors;
• Use the triple scalar product to calculate the volume of the parallelepiped that is defined by three vec-
tors, to establish the linear independence of three vectors and to express the product as a determinant;

• Use the triple vector product and two quadruple vector products.
• To explain the Kronecker symbol δji ;
• Set up the reciprocal vectors of a set of three vectors, and use these to solve vector-valued equations;
• Apply the theory of this chapter to solve basic geometric problems involving points, lines and planes.

21
22 CHAPTER 2. VECTORS PRODUCTS AND RECIPROCAL VECTORS

2.1 Introduction
In chapter 1 we have seen that vectors can be scaled, rotated and that they can be combined by addition
or subtraction. In section 2.2 we discuss various definitions of products of vectors. You are already familiar
with the dot and cross products of two vectors, but the ‘scalar triple product’ will be new. We will also
discuss the ‘vector triple product‘ and two quadruple products. The latter three will be used less in this
course, except in some proofs, but is is good that you are aware of their existence.
In section 2.3 we discuss the reciprocal vectors of a set of three independent vectors. These are useful
for solving systems of vector-valued equations, and generalize, formulated sloppily, the concept of division of
numbers to vectors.
In section 2.4 we demonstrate briefly how these concepts can be used to solve classical problems in
geometry involving points, lines and planes. The products, as well as the concept of reciprocal vectors will
be used extensively in the remainder of this document, so study that material with care.

2.2 Vector Products

In this section we will discuss six vector products of two, three and four vectors:
• The dot product a · b;
• The cross product a × b;
• The scalar triple product (a b c);
• The vector triple product a × (b × c);
• The scalar quadruple product (a × b) · (c × d);
• The vector quadruple product (a × b) × (c × d).
Please notice how important the first three products are. We will show how these are related to lengths,
areas and volumes of objects that are defined by vectors. We will also show that the dot product can be
used to describe projections of vectors. The latter three are less widely used, but have interesting properties
that can be used to prove important relationships among vectors, as will be shown in the form of a number
of exercises

2.2.1 The dot product

The Euclidean dot product a · b of two vectors a and b is defined as
a · b = ||a|| ||b|| cos χab , (2.1)
where ||a|| is the length of vector a, while χab is the angle between the two vectors. By choosing b = a it
follows that √
||a|| = a · a, (2.2)
From the definition of the dot product it is clear that it is commutative,
a · b = b · a. (2.3)
It can be proven from the definition (2.1) that the dot product is linear in the first argument. This means
that
(λ1 a1 + λ2 a2 ) · b = λ1 (a1 · b) + λ2 (a2 · b). (2.4)
Because the dot product is also commutative, it is easy to prove that it is also linear in the second argument:
the dot product is said to be bilinear.
2.2. VECTOR PRODUCTS 23

Exercise 2.1 Please prove that a · (µ1 b1 + µ2 b2 ) = µ1 (a · b1 ) + µ2 (a · b2 ).

Exercise 2.2 Consider figure 1.1(a) and define c = a+b. Give an expression for ||c||2 in terms of the lengths
of the vectors a and b and the angle χab between them. Prove Pythagoras’ theorem as a special result.
Projections are of major importance in vector algebra. In the form of two exercises, we develop the
mathematics of projecting a vector on a line, and show that this can be expressed in terms of dot products.
A defining property of a projection is that it is an idempotent operation: applying the projection a second
time will no longer change the vector.
We will first consider orthogonal projections, in which case the projection is effectuated in a direction
that is perpendicular to the target line.
Exercise 2.3 Orthogonal projections on a line. Draw, from a given point, a vector b and extend the line that
is defined by this vector. Also draw a second vector u.
1. Decompose the vector u into vectors u∥ and u⊥ that are parallel and perpendicular to vector b, draw
these vectors in your figure. The vector u∥ is called the projection of u on b, the vector u⊥ is the
rejection (it expresses the part of u that gets removed by the projection).
2. Show that
b·u
u∥ = b . (2.5)
b·b
Hint: u∥ = αb for some α and u⊥ is perpendicular to b.
3. Show that u∥ = u when u = λb.
4. Show that u∥ = 0 when u ⊥ b.
5. Show that orthogonal projections are idempotent by considering the expression for u∥ as input vector.

6. Introduce b̂ = b/||b|| and show that

u∥ = b̂(b̂ · u). (2.6)
This highlights the fact that the projection depends on the direction of b, not its length.
In an oblique projection, the vector b on which to project and the vector c perpendicular to which the
projection is effectuated are chosen independently. It is required that these vectors are not orthogonal. We
discuss such more general projections in the form of another exercise.
Exercise 2.4 Oblique projections on a line.
Draw, from a given point, two vectors b and c and extend the lines that are defined by these vectors. Also
draw a third vector u, with arbitrary length.
1. Decompose the vector u into a vector u∥ that is parallel to b and a vector ur that is orthogonal to c.
Draw these vectors in your figure.
2. Show that
c·u
u∥ = b . (2.7)
b·c
3. Show that u∥ = u when u = λb.
4. Show that u∥ = 0 when u ⊥ c.
5. Show that oblique projections are idempotent.
6. Derive an expression for the ratio of the lengths of the vectors u∥ and u. Make a sketch of the two
cases that the lengths are equal. (Hint: express the dot products in terms of lengths and angles.)
24 CHAPTER 2. VECTORS PRODUCTS AND RECIPROCAL VECTORS

χAa
a
h

base
c
χbc b

Figure 2.1: The volume of a parallelepiped with sides a, b and c is given by the scalar triple product (a b c),
see equation (2.10). (Based on: https: // en. wikipedia. org/ wiki/ File: Parallelepiped_ volume. svg —
public domain.)

2.2.2 The cross product

Consider the parallelogram with sides a and b that is shown in figure 1.1(c). The area of this parallelogram
is given by the expression
A = ||a|| ||b|| sin χab . (2.8)
If we construct this parallelogram in a three-dimension space, we must also consider its orientation. That
can be described by a vector that is perpendicular to the plane that is spanned by the two vectors a and b.
We define the cross product a × b as the vector:

• perpendicular to the vectors a and b, where the sense is defined by the right-hand rule, applied to the
vectors a and b in that order;

• with a length that is equal to the surface of the parallelogram that is spanned by the vectors a and b,
as given by equation (2.8).

We call this type of vector a directed surface element and will encounter it later on in our discussions of
surface integrals over vector fields.
It can be shown that the cross product is bilinear, just like the dot product. But unlike the dot product
it is anti commutative,
b × a = −a × b. (2.9)
Just like the dot product, the cross product yields a geometric property; it does not depend on the choice
for any coordinate system.

2.2.3 The scalar triple product

We have seen that the length of a vector a can be calculated with the help of the dot product. We have seen
that the area of the parallelogram that is spanned by two vectors a and b can be calculated with the help of
the cross product. It will not come as a surprise that we will now consider the volume of the parallelepiped
with adjacent sides a, b and c. Such parallelepiped is shown in figure 2.1.
The volume of a parallelepiped is equal to the product of the surface of its base plane and the height,
measured perpendicularly to that base plane. If the vectors a, b and c are sorted according to the right-hand
2.2. VECTOR PRODUCTS 25

rule, as in the figure, this volume is given by

(a b c) = a · (b × c). (2.10)

This particular combination of a, b and c is known as the scalar triple product.

Exercise 2.5 Prove that the scalar triple product produces the volume of the parallelepiped that is defined by
its vector arguments. Use the symbols that are also used in figure 2.1.

Note that the scalar triple product can be negative. This happens if its three vector arguments are not
‘sorted’ according to the right-hand rule. We therefore say more precisely that the scalar triple product is
equal to the signed volume of the parallelepiped that is spanned by its three vector arguments. If the usual
(positive) volume is needed, one must either take the absolute value of the triple scalar product, or ensure
that its arguments are sorted according to the right-hand rule.
The scalar triple product has various properties, most can easily be derived from those of the dot and
cross products on which it is based. As an example, it is linear in all three of its arguments. Secondly, the
product is unaltered by a circular shift of the arguments,

(a b c) = (c a b) = (b c a). (2.11)

This property can be understood geometrically by calculating the parallelepiped’s volume using different
choices for base plane.
The scalar triple product changes sign if any of the two arguments are interchanged: it is anticommutative.
As an example,
(a b c) = −(b a c). (2.12)

Exercise 2.6 Please verify the properties (2.11) and (2.12) by imagining (using the right-hand rule) how the
volume calculation would be affected (or not). Start with vectors a, b and c as defined in the picture. In the
second part of the exercise it may help to consider the vector −a × c as an intermediate step.

The scalar triple has another important application: it can be used to decide if a set of three vectors a,
b and c is linearly independent. That is the case if

αa + βb + γc = 0 =⇒ α = β = γ = 0. (2.13)

If that is the case, the scalar triple product is non-zero.

Otherwise, at least one of the vectors can be expressed as a linear combination of the other ones. Then
the three vectors lie in one plane (they are coplanar), or are even parallel (colinear). In this case the triple
scalar product vanishes or, equivalently, the volume of the parallelepiped that is spanned by these three
vectors will vanish. This is left as an exercise.
Exercise 2.7 Linear dependence of vectors.
1. Show that the scalar triple product is zero if any of the arguments are the same.
2. Show that the scalar triple product is zero if one argument is a linear combination of the others.
If the components of the three vectors are known with respect to a Cartesian coordinate system, the
scalar triple product can be calculated as the determinant of the matrix that contains the components as
columns,  
ax bx cx
(a b c) = det A, with A := ay by cy  . (2.14)
az bz cz
26 CHAPTER 2. VECTORS PRODUCTS AND RECIPROCAL VECTORS

This expression can be verified by simply carrying out the cross and dot products, and comparing the result
with the expression for the determinant of a 3 × 3 matrix. But there is more: linear algebra tells us that
det(AT ) = det(A), and that the determinant of the product of two matrices is equal to the product of their
determinants. Then  
a·a a·b a·c
(a b c)2 = det(AT A) = det  b · a b · b b · c  . (2.15)
c·a c·b c·c
The scalar product itself can be obtained by taking the square root of this relation, up to a minus sign. We
leave the proofs of these relations as an exercise.

Exercise 2.8 The scalar triple product as a determinant.

1. Check equation (2.14) by a direct calculation.

2. Prove relation (2.15).

3. Show that (a b c) = ||a|| ||b|| ||c|| for three orthogonal vectors (ordered such that the volume is positive).

4. Show that (ex ey ez ) = 1.

2.2.4 The vector triple product

The vector triple product is defined as a × (b × c). Note that the parentheses are essential and are placed
around the last two arguments. The resulting vector is obviously perpendicular to a and also to the product
b × c. But since the vector b × c is perpendicular to both b and c, the result must be a linear combination of
b and c. What remains is the determination of the coefficients of this combination. The resulting expression
turns out to be1 .
a × (b × c) = b(a · c) − c(a · b). (2.16)
This expression allows a decomposition of a vector in two meaningful parts: assume that a · b ̸= 0, then it
follows that
a · c a × (b × c)
c=b − . (2.17)
a·b a·b
We leave the interpretation of this equation as an exercise.

Exercise 2.9 Consider equation (2.17).

1. What is the meaning of the first term on the right-hand side?

2. What is (apparently) the meaning of the latter?

Exercise 2.10 Test equation (2.16) by taking the dot product with vectors a and b × c.

2.2.5 The scalar quadruple product

The scalar quadruple product is defined as (a × b) · (c × d). The following Binet-Cauchy identity allows its
evaluation in terms of dot products2 .:

(a × b) · (c × d) = (a · c)(b · d) − (a · d)(b · c). (2.18)

Exercise 2.11 Please prove the special case of this equation for a = c and b = d.
1 See for example [Link]
2 For a proof, we refer to [Link]
2.3. RECIPROCAL VECTORS 27

2.2.6 The vector quadruple product

Consider the vector triple product that has been introduced in section 2.2.4. If the vector a in this expression
is itself a cross product of two vectors, the result is a cross product of two cross products. By renaming the
four arguments it can be written as
(a × b) × (c × d).
This four-argument product has a name on its own and is known as the vector quadruple product. It derives its
value mostly from the following identity, which immediately follows from (2.16) (just do proper bookkeeping
of the argument names). This relation is at the basis of various proofs later on in this text. Of interest is also
the special case that two of the arguments are the same. We leave the proofs of these relations as exercises.
Exercise 2.12 Please show that

(a × b) × (c × d) = (a b d)c − (a b c)d. (2.19)

Exercise 2.13 Please show that

(a × b) × (b × c) = (a b c)b. (2.20)

2.3 Reciprocal Vectors

Consider the equation v = au for given real values v and u. This can be solved for a by dividing this
expression by u or, equivalently, by multiplying with its reciprocal value u−1 , since vu−1 = auu−1 = a. Of
course this requires that u ̸= 0. Can we somehow generalize this idea of reciprocal values to vectors?
To that end we will investigate solutions for ai of the vector equation v = a1 u1 + a2 u2 + a3 u3 in a
three-dimensional space. Here the superscripts 1, 2 and 3 should not be confused with exponents.3 This
equation will have solutions for all v only if the three vectors ui are independent. In section 2.2.3 we have
discussed that this means that their scalar triple product (u1 u2 u3 ) is not zero. The question is: can we
construct vectors ui such that v · ui = ai ? We will let you develop these reciprocal vectors in the form of an
exercise.

Exercise 2.14 Consider v = a1 u1 + a2 u2 + a3 u3 and produce a vector u1 such that v · u1 = a1 .

1. How should we choose u1 such that terms involving a2 and a3 disappear from v · u1 ?
2. Choose the most general expression for such vector.
3. Choose the vector such that v · u1 = a1 .

The vector u1 has been derived as part of the previous exercise, the vectors u2 and u3 can be obtained along
the same lines. The resulting definitions of the reciprocal vectors can be summarized as follows:
u2 × u3 u3 × u1 u1 × u2
u1 = , u2 = , u3 = . (2.21)
(u1 u2 u3 ) (u1 u2 u3 ) (u1 u2 u3 )
From the definition it immediately follows that
(
1 i = j;
ui · uj = (2.22)
0 i ̸= j.

This is how close as we can get to the equivalent of u−1 u = 1 in a vectorial setting, hence the name ‘reciprocal
vectors’.
3 In chapter 4 we will use this notation extensively, and make a clear difference between quantities that are written with a

subscript, and those written with a superscript.

28 CHAPTER 2. VECTORS PRODUCTS AND RECIPROCAL VECTORS

Exercise 2.15 Please verify equation (2.22) for i = 1 and j = 1, 2. The other combinations are more-of-the-
same. Remember that (u1 u2 u3 ) = (u2 u3 u1 ) = (u3 u1 u2 ).

The right-hand side of equation (2.22) appears frequently in the remainder of this text. We therefore define
the Kronecker symbol δji as
(
i 1 i = j;
δj = (2.23)
0 i ̸= j.

We also define variants δij and δ ij with the same dependence on the indices i and j. In terms of the Krocker
symbol equation (2.22) takes the form
ui · uj = δji . (2.24)

Exercise 2.16 Let us come back to the vector-valued equation that we used in the introduction of this section,
but this time we express v in terms of the reciprocal vectors. In a three-dimensional geometry, we consider
three linearly (non-coplanar) vectors u1 , u2 and u3 and the vector equation v = α1 u1 + α2 u2 + α3 u3 .

1. Please provide expressions for the unkowns αi .

2. Show that the resulting expression for v is the sum of three oblique projections, each along ui along a
line perpendicular to ui .

For a non-zero real number u we have u = (u−1 )−1 : the reciprocal of the reciprocal is the original number.
Is this also true for the reciprocal vectors? The answer is affirmative, because indeed

u2 × u3 u3 × u1 u1 × u2
u1 = , u2 = , u3 = . (2.25)
(u1 u2 u3 ) (u1 u2 u3 ) (u1 u2 u3 )

We also find that

(u1 u2 u3 ) = (u1 u2 u3 )−1 . (2.26)

We leave the proofs of these relations as an exercise.

Exercise 2.17 Please prove equations (2.25) (for u3 , that suffices) and (2.26). Hint: apply equation (2.20)
to the vectors u2 , u3 and u1 . Also consider the dot product of the result and the vector u3 .

Exercise 2.18 TODO: make this an exercise. In crystallography, a Bravais lattice is a set of points defined
by
R(n1 , n2 , n3 ) = n1 a1 + n2 a2 + n3 a3 . (2.27)

The reciprocal lattice (crystallographer’s definition, without factor 2π) is given by

G(m1 , m2 , m3 ) = m1 a1 + m2 a2 + m3 a3 . (2.28)

We find that
G · R = m1 n1 + m2 n2 + m3 n3 . (2.29)

Consider a planar wave of the form

ei2π(G(m1 ,m2 ,m3 )·r−νt) .

Show that this is periodic in r = ai , for i = 1, 2, 3.

2.4. POINTS, LINES AND PLANES 29

2.4 Points, Lines and Planes

We can use vectors to describe positions on lines and planes in terms of real-valued parameters:
• A line in an Euclidean space that passes through point c and has direction d is the set of points

xl (s) = c + sd, s ∈ R. (2.30)

• A plane in an Euclidean space that passes through point c and is parallel to u and v is the collection
of points
xp (s, t) = c + su + tv, s, t ∈ R. (2.31)
The vector n = u × v is perpendicular to the plane. Taking the dot product of the equation for the
points in the plane with n yields an alternative, parameterless characterisation of the points xp on the
plane:
(xp − c) · n = 0. (2.32)

The concepts (products, reciprocal vectors) that have been discussed in this chapter are convenient tools
for the analysis of basic geometric problems involving points, lines and planes. We exercise these tools in
the form of a number of exercises.

Exercise 2.19 Consider a plane xp (s, t) = c + su + tv and a point p.

1. The point is on the plane if xp (s, t) = p for some s and t, or

(c − p) + su + tv = 0. (2.33)

What do we know about the three vectors c − p, u and v, if a linear combination yields 0?
2. Show that the point is on the plane if and only if ((c − p) u v) = 0.

Let us elaborate on the previous exercise. If we want to calculate the distance of a point p to the plane, we
proceed as follows. We know that the shortest path from p to the plane is along a vector perpendicular to
the plane. Such vector is given by n = u × v. The orthogonal projection of the vector c − p on n will be in
the plane and on the shortest path. This projection is given by
n · (c − p)
n , (2.34)
n·n
and its length is the desired distance. It is given by ||c − p|| cos θ, where θ is the angle between c − p and n.
This distance is zero if c = p or if this vector is perpendicular to n (in this case it is in the plane).
The point on a line that is closest to a given point can be found in a similar way. We leave this as an
exercise.
Exercise 2.20 Consider a line xl (s) = c + sd and a point p.
1. Construct a vector that, when added to p, brings you to a point on the line;
2. Calculate the orthogonal projection of this vector on the line and the rejection vector. The latter is the
smallest vector that brings you from the point p to the line. Give the expression for the point on the
line that is closest to p;
3. This vector can also be obtained in another way. Consider the length of the vector xl (s) − p (or its
square), and calculate the value of s that minimizes this distance. Substitute this vector in the expression
for xl (s) − p.
30 CHAPTER 2. VECTORS PRODUCTS AND RECIPROCAL VECTORS

Example 2.1 A line xl (s1 )c1 + s1 u1 intersects a plane xp (s2 , s3 ) = c2 + s2 u2 + s3 u3 if there is a solution
for s1 , s2 and s3 of the equation xl (s1 ) = xp (s2 , s3 ), or

c1 − c2 = −s1 u1 + s2 u2 + s3 u3 . (2.35)

There is a unique solution if the three vectors ui are not coplanar. In that case the parameters can be found by
dot-multiplying with the reciprocal vectors ui . This yields, for example, s1 = (c2 − c1 ) · u1 . Backsubstitution
in the equation for the line yields the point of intersection

x = c1 + u1 (c2 − c1 ) · u1 (2.36)

Using the definitions of the reciprocal vectors, this can also be written in terms of n = u2 × u3 ,
n · (c2 − c1 )
x = c1 + u1 . (2.37)
u1 · n
(Please verify this expression.)
Let us discuss the meaning of these expressions in words: starting from c1 , the vector c2 − c1 will bring
you to a point on the plane — which is no longer on the line, in general. Instead we project the vector c2 − c1
along a vector that is perpendicular to n (so it will stay in the plane) on the vector u1 (so it will also be on
the line). Adding this modified vector to c1 yields the point of intersection.

Example 2.2 Consider two lines c1 +s1 d1 and c2 +s2 d2 in a three-dimensional space. Do these lines intersect
each other in a single point?
If there are shared points, these are solutions of the equation

c1 + s1 d1 = c2 + s2 d2 ⇐⇒ c1 − c2 = s2 d2 − s1 d1 .

• If d1 and d2 are parallel, so d2 = αd1 , the equation can be written as

c2 − c1 = (s1 − s2 α)d1 ,

which requires that c2 − c1 is also parallel to d1 . In this case the two lines coincide, there are infinitely
many intersection points.
• Otherwise, this has a solution only if c1 − c2 = 0, or is a linear combination of d1 and d2 (the
three vectors are coplanar). Both cases are covered by the requirement that their scalar triple product
vanishes, ((c2 − c1 ) d1 d2 ) = 0.

2.5 Summary and outlook

In this chapter we have introduced a number of scalar- and vector-valued products of two, three and four
vectors, and have shown how these can be used to calculate lengths, surfaces, volumes, express projections
of vectors and establish linear independence of sets of vectors. Furthermore, the reciprocal vectors of a set of
three vectors have been introduced, and it has been shown how these can be used to solve sets of vector-valued
equations. Applications of vector products and reciprocal vectors in solving classical geometrical problems
such as finding the intersection point of a line and a plane have been presented.
In the upcoming section we will present curves. At the end of that chapter we will be equipped with all
the tool that we need to introduce curvilinear coordinate systems, which will be the subject of chapter 4.

Exercise 2.21 Look at the list of objectives at the beginning of this chapter and ensure that you have acquired
the mentioned skills. If not, please study the relevant parts of this chapter again and repeat this exercise.
Chapter 3

Curves

Objectives
After studying this chapter, you will be able to
• Explain the concepts of curves, closed curves and simple curves in Rn ;

• Calculate the velocity and speed of curves;

• Calculate the arc lengths of rectifiable curves;
• Re-parameterize curves;
• Create an arc length parameterization of a given curve and explain the key feature of such parameter-
ization.

31
32 CHAPTER 3. CURVES

3.1 Introduction
Consider the position of a particle in a Euclidean space. For a given choice for an origin (see section 1.4)
this can be described by a position vector x. If the particle is moving, the position depends on time t and
we write x(t). This is an example of a vector-valued function. It maps a value from the set of real numbers
R to a vector.
Once we have defined functions that take and/or produce vectors, it is only natural to develop the basic
tools for analyzing such functions. These are the concepts of continuity and differentiation. As we will see in
the next sections, the definitions are very close to those for scalar functions. The essential difference is the
measurement of distances between argument and result values. Whereas in the scalar case the ‘absolute value’
of a difference between scalar values was used, we will use the (Euclidean) norm when vectorial distances
must be measured.
In this chapter we will discuss curves in two- and three-dimensional Euclidean space. We will resort mostly
to analytical methods to describe curves. We will consider vector-valued functions of one parameter, and
define continuity and differentation of such functions (section 3.2. We will then define a curve as continuous
vector-valued functions of one parameter. Some of these inquiries can be done without the introduction of a
coordinate system. Others are facilitated by equipping space with one, so vectors can be studied in terms of
their coordinate representation with respect to that system. It is important to bear in mind though that the
choice for a coordinate system is a matter of convenience: curves are geometric objects that exist independent
of a coordinate system
There is nothing that keeps us from using coordinate systems other than a Cartesian one. In the chapter
that follows this one, we will make an in-depth study of more general coordinate systems, based on the
properties of curves that we will outline in the next sections.

3.2 Vector-Valued Functions

3.2.1 Definition
3.2.2 Continuity
Consider a time-dependent vector x(t). This is an example of a vector-valued function, a function that
produces a vector. The function x(t) is said to be continuous in t = a if

lim ||x(t) − x(a)|| = 0. (3.1)

t→a

This is completely analogous to the definition of continuity for non-vectorial functions. The only difference
is that we use the (Euclidean) distance to measure the distance between to vectors, for scalar functions we
would have used the absolute value of the difference between the two values.

3.2.3 Differentiating vector-valued functions of one parameter

Consider a time-dependent vector u(t). This is an example of a vector-valued function, a function that
produces a vector. The derivative of this function is defined in the same way it is for a ‘normal’ function; it
is given by
du ∆u
= lim , where ∆u = u(t + ∆t) − u(t). (3.2)
dt ∆t→0 ∆t
Alternative notations for the derivative of a function of one variable are u′ and, for time-dependent functions,
u̇. The derivative is itself a vector, which exists independently of the choice for a particular coordinate system.
The derivative has mostly the same properties as that of a real-valued function. In particular, it im-
mediately follows from the definition that the derivative of a constant vector is the vector 0, and that the
3.2. VECTOR-VALUED FUNCTIONS 33

derivative of a linear combination of vectors is equal to the linear combination of the derivatives: let v(t) be
a second vector, and α and β constant scalar values, then
′
(αu + βv) = αu′ + βv ′ . (3.3)

Now let f (t) be a real-valued function. Then it is easy to show that

′
(f u) = f ′ u + f u′ . (3.4)

This is the product rule for differentiation. Other variants of the product rule state that:
′
(u · v) = u′ · v + u · v ′ ; (3.5)
′
(u × v) = u′ × v + u × v ′ . (3.6)

Finally we mention the chain rule. For a vector function u(f (t)) this states that
d du df
u(f (t)) = . (3.7)
dt df dt
Exercise 3.1 The derivative of a product of scalar and vector-valued functions. Consider f (t) and u(t) as
above, and define ∆f and ∆u such that f (t + ∆t) = f (t) + ∆f and u(t + ∆t) = u(t) + ∆u.
1. Show that
(f u)(t + ∆t) − (f u)(t) ∆f ∆u ∆f ∆u
= u(t) + f (t) + ∆t .
∆t ∆t ∆t ∆t ∆t
2. Argue that, if the derivatives of f and u exist, in the limit ∆t → 0 this gives1

(f u)′ (t) = f ′ (t)u(t) + f (t)u′ (t).

3. Prove, along the same lines, relation (3.5). State precisely which property of the dot product you rely
on.

Exercise 3.2 Let x(t) be the position of a particle.

1. Show that x′ ⊥ x′′ if the particle has a constant speed.
Assume that the particle stays at a constant distance from the origin (the velocity may vary).
2. Show that x ⊥ x′ .
3. Show that x · x′′ = −||x′ ||2 .

Exercise 3.3 Let x(t) be the position of a particle with mass m. Let p, L, F and ϵ denote the linear and
angular momentum of the species, the force acting on the species and the kinetic energy of the species. Use
that Newton’s laws of motion.
1. Give expressions for these four quantities in terms of m, x and its derivatives.
2. Express the rate of change of ϵ in terms of the particles velocity and force.
3. A magnetic flux density B exerts a force qx′ × B on a particle with charge q. Show that this does not
affect the kinetic energy of that particle.
4. Show that the rate of change of the angular momentum is equal to the torque r × F .
1 You may want to look up the ‘product rule for limits’.
34 CHAPTER 3. CURVES

3.3 Parameterized Curves

In section 3.2 we have discussed vector-valued functions, and a typical example is the function x(t), which
maps real-valued numbers from an interval I ⊂ R to positions in Rn . If the function is continuous (see
section 3.2.2), such a function defines a curve. Note that in the two-dimensional case, continuity implies
that you can draw x(t) without lifting your pen from the paper. In higher dimensions, continuity can be
understood in a similar fashion.

A curve is a continuous vector-valued function of one real-valued parameter.

If the interval is left-closed, the curve has a begin point A, say. Similarly, if the interval is right-closed,
it has an end point B, say. If the begin and end points coincide, we speak of a closed curve. In this text we
will be concerned mostly with simple curves. Such curves do not intersect themselves anywhere, except in
the begin and end point (if the curve is closed).
Exercise 3.4 We consider a two-dimensional space that has been equipped with a Cartesian coordinate system.
We define ω = 2π/T , R is a fixed parameter and n an integer number, n ≥ 2. Please sketch the following
curves and discuss their properties (left/right-closed, simple).
1. x(t) = ex R cos ωt + ey R sin ωt on t ∈ [0, T ];
2. x(t) = ex R cos ωt + ey R sin nωt on t ∈ [0, T ].
The graphs of these figures are examples of Lissajous figures, which you have probably seen before on an
oscilloscope, as in figure 3.1. To which n does that figure correspond?
Exercise 3.5 We consider a three-dimensional space that has been equipped with a Cartesian coordinate sys-
tem. Please describe the following curve (again we have defined ω = 2π/T ; R and H are fixed parameters):
x(t) = ex R cos ωt + ey R sin ωt + ez Ht/(2T ),
where t ∈ [0, 2T ].
Although we will not formally introduce other than Cartesian systems until later in this text, we assume
that you are familiar with polar coordinates (r, ϕ), where r is the distance of a point P to the origin, and
−−→
ϕ the angle between the x-axis of a Cartesian system and the line OP . The relation with the Cartesian
coordinates is given by x = r cos ϕ and y = r sin ϕ.
Example 3.1 Consider the Archimedean spiral, which is given by [5, p. 64]:
r(ϕ) = a + bϕ (3.8)
with ϕ ∈ [0, ∞). In terms of a Cartesian coordinate system this can be written in parametric form as
x(ϕ) = (a + bϕ)(ex cos ϕ + ey sin ϕ).
This curve is shown in figure 3.2 for a = 0, b = 1/(2π) and ϕ ∈ [0, 4π].
Example 3.2 A curve that is somewhat similar to the Archimedean spiral is the Logarithmic spiral, or Spira
mirabilis, the marvelous spiral. It is defined by
r(ϕ) = aebϕ , (3.9)
where ϕ ∈ (−∞, ∞). In a Cartesian coordinate system this can be written in parametric form as
x(ϕ) = aebϕ (ex cos ϕ + ey sin ϕ).
This curve is omnipresent in nature, some examples are shown in figure 3.3. We will investigate this curve
further in exercise 3.8
3.4. VELOCITY AND SPEED 35

Figure 3.1: A scope reading, showing a Lissajous figure. See exercise 3.4 (CC BY-SA 3.0, referred to by
[Link]

3.4 Velocity and Speed

Consider a curve x(t). The derivative of x is a vector that is tangential to the curve and is called the velocity
of the curve. The length of the velocity vector in a given point is called the speed in that point.
Because we have used t as the parameter name it is difficult not to interpret x′ (t) as the velocity. But
please bear in mind that the concepts of the velocity and speed of a curve are more general than that. The
velocity of a parameterized curve x(u) is given by x′ (u), the derivative of x with respect to the parameter
u, and u does not, in general, need to represent time.

Exercise 3.6 Consider the 2-dimensional curves that were presented in exercise 3.4. Calculate the velocities
and speeds of those curves as a function of the parameter t.

Exercise 3.7 Consider the 3-dimensional curve that was presented in exercise 3.5. Calculate the velocity and
speed of that curve as a function of the parameter t.

Exercise 3.8 Consider the Spira Mirabilis that was presented in example 3.2:

1. Show that for this curve ||x|| = aebϕ ;

2. Show that the velocity of the curve as a function of the parameter ϕ can be written as

x′ (ϕ) = bx + aebϕ (−ex sin ϕ + ey cos ϕ)

and observe that x · x′ = b||x||2 ;

3. Show that the speed of the curve is given by

p
||x′ || = ||x|| 1 + b2 ;
√
4. Show that the angle α between x and x′ is given by cos α = b/ 1 + b2 , or α = arctan(1/b). This angle
is constant, which is a remarkable property of the logarithmic spiral. Make sure that you understand
the geometric significance of this property.
36 CHAPTER 3. CURVES

(a) An Archimedean spiral (b) A vinyl LP

Figure 3.2: (a): an Archimedean Spiral (Guillaume Jacquenot — Own work, CC BY-SA 3.0, [Link]
[Link]/w/[Link]?curid=25156165). (b): The groove of old vinyl LPs forms an Archimedean spiral
(CFCF — Own work, CC BY-SA 4.0, [Link]

3.5 The Arc Length

Consider a parameterized curve x(t), with t ∈ [ta , tb ]. An infinitesimal change dt of the parameter t results
in an infinitesimal change dx of the position that is given by

dx = x′ dt. (3.10)

The magnitude of this vector represents the infinitesimal arc length of the curve and is given by

ds = ||dx|| = ||x′ (t)||dt. (3.11)

Integrating this expression over the interval [ta , t] gives us the arc length of the part of the curve that is
described by this parameter interval,
Zt
s(t) = ||x′ (τ )||dτ (3.12)
ta

and in particular, for t = tb we obtain the length of the entire curve. We have of course not proven that
every curve has a finite length, and that the function s(t) exists. A curve that does have a finite length is
called rectifiable, for a more detailed discussion we refer to an external text on the subject2

Exercise 3.9 Consider the parameterized circle x(t) that was introduced in the first half of exercise 3.4. Use
the expression for the arc length to calculate the circumference of the circle and to verify the result.
2 See for example [Link] and the references on that page.
3.5. THE ARC LENGTH 37

(a) A nautilus shell. (b) A Lithuanian gold coin.

(c) A cyclone over Iceland. (d) The Whirlpool Galaxy.

Figure 3.3: Examples of logarithmic spirals in society and nature. (a): Cutaway of a nautilus shell showing the
chambers arranged in an approximately logarithmic spiral. The plotted spiral (dashed blue curve) is based on
growth rate parameter b = 0.1759 (Dicklyon - Own work, CC BY-SA 4.0, [Link]
w/[Link]?curid=35543222). (b): A Lithuanian 10 litas gold coin (Lithuanian Bank — public domain, see
[Link] (c): An extratropical cyclone over Iceland (NASA — public
domain, [Link] (d): The Whirlpool Galaxy (NASA/ESA
— public domain, [Link]
38 CHAPTER 3. CURVES

Exercise 3.10 Calculate the arc length of the parameterized helix x(t) that was introduced in exercise 3.5.

Exercise 3.11 Calculate the arc length s(ϕ) of the Logarithmic spiral x(ϕ) that was introduced in example
3.2. Also find an expression in terms of r and α. (Hint: use the results of exercise 3.8.)

3.6 Re-parameterization
Consider a parameterized curve x(t), with t ∈ [ta , tb ]. Let us now assume that t is a monotonically increasing
function of u, with ua = u(ta ) and ub = u(tb ). Then, as u changes from ua to ub , t will change from ta to tb
and the same curve is described by this parameter u.
What we have obtained by transformation is a re-parameterization of the curve. The question we would
like to answer is, of course, why one would want to do that. To that end, let us consider the velocity and
speed of the curve with respect to both parameters. With the help of the chain rule (3.7) we find that

dx(t(u)) dt dx(t)
= . (3.13)
du du dt
It is obvious that the velocity changes when we choose a new parameter, and so will the speed.

Exercise 3.12 Consider thepcurve x(t) = R(ex cos ωt + ey sin ωt), with t ∈ [0, T ]. Here ω = 2π/T . Next,
consider the function u = t/T . Calculate the speed of the curves x(t) and x(t(u)).

3.7 The Arc Length Parameterization

Let us take another look at expression (3.13), which tells us how the velocity changes as a result of using
a new parameter u(t) instead of the original t. We can ask ourselves the question: is there a function u(t)
such that the speed of the new curve is unity?

Exercise 3.13 Derivation of the arc length parameterization.

1. Show that the requirement that the speed of a curve is unity implies that
−1
dt dx(t)
= (= ||x′ (t)||) ;
du dt

2. Choose t(u) to be monotonically increasing, so the modulus signs on the left hand side can be omitted.
Also use du/dt = 1/(dt/du). Show that integration over [ta , t) yields that u(t) can be chosen to be
equal to s(t), the arc length.

The preceeding exercise suggests the following procedure for obtaining the arc length parameterization of
a curve:

1. For a given curve x(t) with t ∈ [ta , tb ], give an expression for the arc length s(t);

2. Find the inverse t(s) of the function s(t);

3. Adopt the arc length s as parameter of the (modified) curve xs (s) ≡ x(t(s)), with s ∈ [0, s(tb )]. This
is the arc length parameterization of the curve. We repeat its most important property, that its speed
has the value 1 on the entire curve.
3.7. THE ARC LENGTH PARAMETERIZATION 39

An analytical establishment of the arc length parameterization is usually not possible. In the following
exercise we discuss a somewhat trivial exception, which allows you to practice all the steps involved in
obtaining an arc length parameterization and appreciating its key feature: evenly spread samples of the (arc
length) parameter result in an evenly distributed set of positions on the curve.
Exercise 3.14 Consider the curve x(t) = Dt2 (ex + ey ) for t ∈ [0, 1]. (Indeed, this is just a line segment with
an unusual ‘quadratic’ parameterization.) Here D is the horizontal and vertical extent of the line, it is a
given constant number.
1. Draw the curve and mark with dots the points for t = 0, t = 21 and t = 1. Notice that these points are
not evenly spread, the second lies much closer to x(0) than to x(1);
√
2. Calculate the velocity and speed of the curve x(t); show that the arc length is given by s(t) = 2Dt2 ;

3. What are the boundary parameter values sa = s(0) and sb = s(1) that correspond to t = 0 and t = 1?
Show that the latter is equal to the length of the curve, as it should be by the definition of s(t);
4. Obtain the function t(s) and show that the arc length (re-)parameterization of the curve is given by
ex + ey
xs (s) ≡ x(t(s)) = s √ ;
2

5. Calculate the positions on the curve for sa , (sa +sb )/2 and sb and show that these are evenly distributed
on the curve.

Exercise 3.15 Look at the list of objectives at the beginning of this chapter and ensure that you have acquired
the mentioned skills. If not, please study the relevant parts of this chapter again and repeat this exercise.
40 CHAPTER 3. CURVES
Chapter 4

Coordinate Systems

Objectives
After studying this chapter, you will be able to
• Describe coordinate lines and planes of common coordinate systems;

• Calculate and explain the Jacobi matrix of a coordinate transformation;

• Use the Einstein summation convention as a notational aid;
• Explain the origin of covariant and contravariant sets of basis vectors;
• Express vectors in terms of such bases;

• Express line, surface and volume elements in terms of curvilinear coordinate systems;
• To explain the Levi-Civita symbol ϵijk ;
• Explain the meaning and relevance of the metric tensor;

• Convert covariant quantities in contravariant quantities and vice versa;

• Explain the implications of choosing an orthogonal curvilinear coordinate system;
• Explain and apply the concept of ‘physical’ components and basis vectors.

41
42 CHAPTER 4. COORDINATE SYSTEMS

4.1 Introduction
In chapter (2) we have studied vectors, vector products and reciprocal vectors and demonstrated that the
dot, cross and triple scalar products are related to lengths, areas and volumes. We have demonstrated various
applications, such as projections and the analysis of classical problems involving points, lines and planes.
We have argued that all these geometric objects do not depend on the choice for a coordinate system, and
indeed in that entire chapter, we have not assumed any specific system. In chapter 1 we have already argued
that such coordinate-free synthetic geometry can be traced back to Euclid’s treatment of the subject. The
vectors, vectors products et cetera that we have seen so far, are a different language for describing Euclid’s
line segments, areas, volumes, projections, et cetera.
But most of the time that you as a physicist will work with quantities that depend on space (and maybe
time) coordinates, you will do so with reference to a particular coordinate system. In this chapter we will
present the general case of curvilinear coordinate systems. These are coordinate systems that are based on
a set of basis vectors that are, in general, neither orthogonal, nor normalized. Moreover, the basis vectors
are in general functions of the (curvilinear) coordinates — they vary from point to point.
We start this section with the definitions of coordinate curves and coordinate surfaces. The chapter
continues with a discussion about transformations, answering the question how the components and basis
vectors change if we switch from one system to another. Of special interest will be transformations from
Cartesian systems to other ones. The role of the Jacobi matrix and its determinant in this process will be
addressed. In this discussion, you will start to use the knowledge that you have acquired in your classes on
Linear Algebra and Multi-Variable Calculus, and you will not stop needing that background knowledge until
the end of this course.
Subsequently, we will introduce the covariant basis vectors of a coordinate system. We will demonstrate
that these basis vectors naturally arise in the calculation of the velocity of a particle at position x(t). The
components of a vector, relative to the covariant basis, are called the covariant components.
In section 2.3 we have considered the reciprocal vectors of a set of vectors in a three-dimensional space.
We will define the contravariant basis vectors to be the reciprocal vectors of the covariant set. This allows
an alternative representation of vectors, using contravariant vector components.
You may wonder why we need all these concepts. After all, once you have access to a basis, by definition
every vector can be expressed in terms of its components with respect to that basis. Why would you need
more than one? The answer is that some physical quantities are most naturally expressed in terms of the
covariant basis, other in terms of the contravariant basis. But it is possible to convert one representation
into the other, and we will show you how, using the so-called metric tensor.
In the second part of this chapter we will discuss the special case of ortho-curvilinear (OCL) coordinate
systems, which are characterized by basis vectors that are orthogonal (but not necessarily normalized).
Examples of such coordinate systems are the Cartesian, cylindrical and spherical coordinate systems with
which you are already familiar. Such vectors can be normalized, resulting in a third representation of vectors
in terms of its physical components. That is the representation that you have been dealing with so far in
your career.
The chapter concludes with expressions for volumes and surfaces that are defined by infinitesimal vari-
ations of the coordinates. You will use these expressions extensively in later chapters.
The language that you will learn to speak is compatible with the language in advanced texts on fluid
dynamics, like Ref. [6] or the theory of general relativity, like Ref. [7]. We therefore believe that some
familiarity with this subject matter has great added value for students of applied physics. And even if you
do not aspire to master such books, you will be dealing with coordinate representations of vectors in numerous
courses in the remainder of your curriculum, so a better understanding of coordinates and coordinate systems
will be of great help.
4.2. COORDINATE SYSTEMS, COORDINATE LINES AND SURFACES 43

4.2 Coordinate Systems, Coordinate Lines and Surfaces

Our starting point is the idea is that every point in a three-dimensional space can be uniquely identified by
a set of three coordinates (x1 , x2 , x3 ). Note that we use superscripts to identify the coordinates, the reason
for that will be made clear later on. In a Euclidean space we can introduce the position vector of a point,
and the introduction of the coordinate system allows us to interpret this vector as a function x(x1 , x2 , x3 )
of the three coordinates. We will assume that this function is continuous; informally this means that points
that have almost the same coordinates will also be located close to each other in space. We will also assume
that the coordinate chart is ‘smooth’, which is to say that the first and second partial derivatives of x with
respect to the coordinates exist.
Let us also consider a particular reference position x(x10 , x20 , x30 ), and investigate the function x(x1 , x20 , x30 )
for fixed x20 and x30 . The result is a vector-valued function of a single parameter, a parameterized curve, that
depends on the coordinate x1 . We call such curve an x1 coordinate curve.
Next, let us fix (only) the coordinate x1 to the value x10 . The result is the function x(x10 , x2 , x3 ) that
depends on the parameters x2 and x3 . This describes a surface that passes through the reference point and
is called an x1 coordinate surface.

Example 4.1 Consider a Cartesian system, for which (x1 , x2 , x3 ) = (x, y, z) and x(x, y, z) = xex +yey +zez .
The z coordinate lines are given by x(x0 , y0 , z) = x0 ex + y0 ey + zez . The z coordinate surfaces are given by
x(x, y, z0 ) = xex + yey + z0 ez .

Exercise 4.1 Consider a cylindrical system with (x1 , x2 , x3 ) = (z, r, ϕ); the symbols have their usual meaning.
Describe the coordinate lines and surfaces for all three coordinates.

Exercise 4.2 Consider a spherical system with (x1 , x2 , x3 ) = (r, θ, ϕ); the symbols have their usual meaning.
Describe the coordinate lines and surfaces for all three coordinates.

4.3 Coordinate Transformations, the Jacobi Matrix

So far we have assumed the existence of some coordinate system xi . In many cases, such coordinate system
is constructed from some other coordinate system xi by specifying functions of the form xi = xi (x1 , x2 , x3 ).
In general, these relations are not linear. But we expect that these transformations are invertible, at least in
‘most’ points. From your classes on multivariable calculus you will recall that this requires that the Jacobi
matrix J has a non-vanishing determinant. You will recall that J is defined as
 1
∂x1

∂x
 i   ∂x1 · · · ∂xn 
∂x  . .. 
J= j
= .. . .
 (4.1)
∂x 
 ∂xn n
∂x
···
∂x1 ∂xn

The Jacobian of the inverse transformation is then given by J = [∂xj /∂xk ]. Using the chain rule and
definition (2.23) of the Kronecker symbol δji we find that

3
X ∂xi ∂xj ∂xi
j k
= k
= δki . (4.2)
j=1
∂x ∂x ∂x

In matrix form, this can be written as JJ = I. In other words, the Jacobian of the inverse transformation is
the matrix inverse of the Jacobian, J = J−1 .
44 CHAPTER 4. COORDINATE SYSTEMS

a3

a2

a1

Figure 4.1: Covariant basis vectors are defined as the partial derivatives of the position vector with respect to
the coordinates of the coordinate system. These basis vectors are tangential to the corresponding coordinate
lines.

Exercise 4.3 Please verify that (4.2) represents element (i, k) of the matrix equation JJ = I.

Before continuing, we will introduce a clever and very useful notational trick, the Einstein summation
convention:

1. When an index appears twice in a term, once as a subscript and once as a superscript, summation
over that index is implied (the summation sign can be omitted).
2. A superscript that appears in a denominator counts as a subscript.

The summation convention results in nice, compact syntax. As an example, equation (4.2) can be written
as
∂xi ∂xj
= δki . (4.3)
∂xj ∂xk
An index over which a summation is implied (here j) is called a dummy index. Such an index can be freely
changed into a different one that is not already in use, say m. In the remainder of this text, this Einstein
convention is used, unless stated otherwise, or the summation signs are written explicitly.

4.4 The Covariant Basis, Contravariant Vector Components

In chapter 3 we have seen that the derivative of a parameterized curve yields a vector that is tangential to
the curve. But the derivative of the function x(x1 , x20 , x30 ) for x1 = x10 is of course nothing but the partial
derivative of the function x(x1 , x2 , x3 ) in the reference point. The same goes for the x2 and x3 coordinate
lines. We will call these tangential vectors ai , so

∂x(x1 , x2 , x3 )
ai (x1 , x2 , x3 ) := . (4.4)
∂xi
In figure 4.1 we have visualized the coordinate lines and tangential vectors for a (non-orthogonal) curvilinear
coordinate system. The figure also shows the parallelepiped that is spanned by the vectors a1 , a2 and a3 .
4.4. THE COVARIANT BASIS, CONTRAVARIANT VECTOR COMPONENTS 45

The vectors a2 and a3 are tangential to this surface, the vector a2 × a3 is perpendicular to it.
For an admissible coordinate system, the scalar triple product is positive, (a1 a2 a3 ) > 0. It then
represents the volume of the parallelepiped that is sketched in figure 4.1. In section 2.2.3 we have seen that
this means that the vectors (4.4) are linearly independent and, consequently, they constitute a right-handed
basis for the three-dimensional space. The vectors ai are called the covariant basis vectors. They are more
precisely referred to as the local covariant basis vectors, to highlight the fact that in general the directions
and the lengths of these vectors vary from point to point.
The covariant basis appears naturally in expressions for displacements and related quantities. As an
example, let us consider a particle at a position x as a function of time. After the introduction of a
coordinate system, positions can be described in terms of coordinates, and the expression for the position
of a particle as a function of time becomes x(x1 (t), x2 (t), x3 (t)). The velocity is obtained by differentiating
this function with respect to time, and with the chain rule and Einstein’s summation convention the result
is
dx dxi ∂x dxi
v(t) = = i
= ai = v i ai . (4.5)
dt dt ∂x dt
The components v i of a vector v with respect to the covariant basis are called the contravariant components.
Remember that
• covariant quantities are indicated with a subscript,
• contravariant quantities are indicated with a superscript.
These names are explained by the behavior of these quantities under a change of coordinates, as we will
now see. Let us first see how the covariant basis vectors change on the change of coordinates. Using the
chain rule we find that
∂x ∂xi ∂x ∂xi
aj = = = ai . (4.6)
∂xj ∂xj ∂xi ∂xj
This means that the transformation is described by the elements of the Jacobian. This transformation law
is a property of the coordinate system, not of any particular vector.
Next, we look at the components of the velocity vector. For this particular vector we find that

dxj ∂xj dxk

= , (4.7)
dt ∂xk dt
in other words: the transformation is described by the elements of the inverse of the Jacobian. We call these
components contravariant.
Let us now see what this means for the velocity vector as a whole. We use the previous results for the
transformation of the components and of the basis vectors, but change some dummy indices to avoid index
clashes (we use k for the components’ transformation, l for that of the basis vectors). The result is that

dxj ∂xj dxk ∂xl ∂xl ∂xj dxk dxk dxk

v= aj = k j
al = j k
al = δkl al = ak . (4.8)
dt ∂x dt ∂x ∂x ∂x dt dt dt
This result proves that v is indeed a vector, or stated differently: that dxi /dt are the (contravariant)
components of a vector. The combined transformation of the contravariant components and the covariant
basis vectors is such that the vector as a whole is invariant under a coordinate transformation. In general,
v i are contravariant components of a vector if and only if

∂xi k ∂xi k
vi = v , or v i = v . (4.9)
∂xk ∂xk
Please take some time to digest this essential concept.
46 CHAPTER 4. COORDINATE SYSTEMS

The results above have been written in index notation. We will now demonstrate what these expressions
look like in ‘matrix language’ and derive an expression for the Jacobian determinant. Firstly, by taking the
dot product of equation (4.6) with Cartesian basis vectors, and grouping the results in a column matrix, we
get the Rn -representation of the equation for that Cartesian basis. The equation is the same as (4.6), but
with boldface symbols for the column matrices, as discussed in section 1.5.2. By introducing matrices that
host the columns aj and ai , the result can be written as
   
a1 a2 a3 = J a1 a2 a3 . (4.10)
 
In exercise 2.8 we have proven that det a1 a2 a3 = (a1 a2 a3 ), see equation 2.14. By taking the
determinant of the previous equation and using the product rule for determinants we then find

(a1 a2 a3 ) = det J (a1 a2 a3 ) (4.11)

In the special case that the original basis is Cartesian, we have (a1 a2 a3 ) = 1 (see again exercise 2.8), in
which case
(a1 a2 a3 ) = det(J). (4.12)
In section 4.10 we will see that this expression appears in the expression for volume integrals in terms of
non-Cartesian coordinate systems. (You should also recall this from your classes on Multi-Variable Calculus.)

4.5 Dot Products and the Metric Tensor

Changing the symbol of a dummy index may be necessary to avoid index clashes when multiple vectors are
combined, for example when the dot product of the two vectors is calculated. That is no different when we
use explicit summations and want to bring the summation sign to the front. If we use j instead of i for
vector v we can write !  3 
X 3 X 3 X
X 3
u·v = ui ai ·  v j aj  = ui v j ai · aj . (4.13)
i=1 j=1 i=1 j=1

In this case two indices appear in subscript-superscript pairs. Thanks to the Einstein convention, the sum-
mation signs can be omitted and the dot product can be written very compactly as

u · v = ui v j ai · aj . (4.14)

Let us take another look at this expression for the dot product. We see that its evaluation requires the
the components of the vectors, and the dot products of the basis vectors, not the basis vectors themselves.
These coefficients
gij := ai · aj (4.15)
are the elements of the metric tensor. This can be written in matrix form as
 
a1 · a1 a1 · a2 a1 · a3
g =  a2 · a1 a2 · a2 a2 · a3  . (4.16)
a3 · a1 a3 · a2 a3 · a3

In terms of the elements of the metric tensor, the dot product can be written as

u · v = ui v j gij . (4.17)

Exercise 4.4 The metric tensor.

4.5. DOT PRODUCTS AND THE METRIC TENSOR 47

1. Please show that the metric tensor is symmetric, which means that gij = gji . Also state this property
in terms of the matrix g, in ‘matrix language’.
2. Consider a coordinate system for which the basis vectors are orthogonal in each point. Write down g
for this special case and describe the meaning of the values of the non-zero elements. (Such coordinate
systems will be elaborated on in section 4.9.)
3. Show that for a Cartesian coordinate system

g = I, (4.18)

the metric tensor is the identity matrix.

4. Show that for a spherical coordinate system (r, θ, ϕ) we have
 
1 0 0
g =  0 r2 0 . (4.19)
0 0 (r sin θ)2

You may re-use your answers to exercise 1.8.

In equation (4.15) we have defined the components of the metric tensor. How do these elements change
under a coordinate transformation? We use the transformation law for the covariant basis vectors (twice),
again ensuring that no index collisions occur by choosing two different dummy indices k and l. The result is
that
∂xk ∂xl ∂xk ∂xl ∂xk ∂xl
gij = ai · aj = ak · al = ak · al = g . (4.20)
∂xi ∂xj ∂xi ∂xj ∂xi kl ∂xj
Exercise 4.5 Consider equation (4.17) for the dot product. Show that the dot product is a scalar, a number
that is invariant under a coordinate transformation.
The partial derivatives that appear in equation (4.20) can be recognized as the elements of the Jacobian
matrix (4.1), which allows us to write
gij = Jik g kl Jjl . (4.21)
Remember that Jik is the element of J at row k and column i. But that value is also equal to the element
at row i and column k of JT . The result is that the equation can be written in matrix form as

g = JT gJ. (4.22)

Note that the order in which the factors at the right-hand side of equation (4.21) are written is immaterial,
whereas the order of the matrices in equation (4.22) does matter. That is a peculiarity of the language of
matrices. The advantage of the matrix notation is that we have access to many linear-algebraic tools that we
have gotten to know in that language. As an example, taking the determinants of the left- and right-hand
sides, using (again) the product rule for determinants and the identity det(AT ) = det(A), results in

det(g) = det(J)2 det(g). (4.23)

In the special case that the original basis is Cartesian, we have g = I, cf. equation (4.18). Combination
of equations (4.23) and (4.11) then yields
p √
(a1 a2 a3 ) = det(J) = det(g) := g. (4.24)

Please remember this definition of g as the determinant of the metric tensor. It wil appear numerous times
in the text that follows.
48 CHAPTER 4. COORDINATE SYSTEMS

4.6 The Contravariant Basis

In section 2.3 we have considered the reciprocal vectors of a set of three independent vectors in a three-
dimensional space. For the reciprocal vectors of the covariant basis vectors ai we get, using equations (2.21)
and (4.24),
a2 × a3 a3 × a1 a1 × a2
a1 = √ , a2 = √ , a3 = √ . (4.25)
g g g
The vectors ai are called the contravariant basis vectors and constitute the (local) contravariant basis for
the coordinate system. Equation (2.24) gives us the important result that

ai · aj = δji . (4.26)

Exercise 4.6 Please prove relation (4.26).

The expressions (4.25) can be conveniently summarized with the help of the Levi-Civita permutation
symbol, see for example [6, page 723]. It is defined as

 0 if any of the i, j, k are identical;
eijk = +1 if ijk is an even permutation of 123; (4.27)
−1 if ijk is an odd permutation of 123.


A permutation is even/odd if the number of index swaps that is required to obtain 123 is even/odd. As an
example e132 = −1 (one exchange operation is needed, of 2 and 3), whereas e312 = 1 (two such operations
are needed). We will also use the symbol eijk , which is defined in exactly the same way. In terms of the
Levi-Civita symbol, equations (4.25) can be summarized as
√
ai × aj = geijk ak . (4.28)

This can be verified by evaluating this expression for specific i and j values and comparing the result with
equation (4.25). We leave this as an exercise.
Exercise 4.7 Please test equation (4.28) by considering the following representative cases:
1. i = 1 and j = 2;
2. i = 2 and j = 1;
3. i = 1 and j = 1.

Exercise 4.8 Please show that

1 eijk
ak = √ ai × aj . (4.29)
2 g
It suffices to show this for k = 3.

Since the vectors ai form a basis, any vector u can be written in terms of its components with respect
to this basis as
u = ui ai . (4.30)
The components ui with respect to the contravariant basis are called the covariant components. Note that
here again subscripts are used for covariant quantities, superscripts for contravariant ones.
At this point you probably ask yourself the question: why would I need two sets of basis vectors? Cannot
I just stick to the covariant ones, for example? The rationale for having both will be presented later, but in
short the answer is as follows: some concepts are defined much more naturally in terms of a covariant set of
4.6. THE CONTRAVARIANT BASIS 49

basis vectors, others more conveniently using the contravariant ones. As an example, we have seen that the
calculation of the velocity vector and the lengths of line segments suggest the usage of a covariant basis. On
the other hand, the gradient, to be discussed in a later chapter, is defined more conveniently with reference
to a contravariant basis.
Let us now consider two vectors that are expressed in covariant coordinates. For the dot product we find
that
u · v = ui vj g ij , (4.31)
where we have introduced
g ij := ai · aj . (4.32)
ij
We will later see that the matrix with elements g is the inverse of the metric tensor (equation 4.47). Finally,
for the mixed case we find that
u · v = ui vi = ui v i . (4.33)
Exercise 4.9 Please derive equations (4.31) and (4.33).
Let us now take a look at the cross product of two vectors u and v. As before, we use distinct indices
i and j to write the vectors in terms of components and basis vectors, and with the help of equation (4.28)
we find
√
u × v = ui ai × v j aj = ui v j ai × aj = geijk ui v j ak . (4.34)
Next we consider the triple scalar product that was defined in equation (2.10). This can also be expressed
in a compact form using the Levi-Civita symbol. The result is
√
(u v w) = geijk ui v j wk . (4.35)

Exercise 4.10 The scalar triple product.

1. Please prove equation (4.35), using (4.34) and (4.26).
2. How many terms appear on the right-hand side of equation (4.35)? How many of these are non-zero
in general?
Please recall equations (2.25) and (2.26), which state that the ‘reciprocal vectors of the reciprocal vectors’
are the vectors themselves. Also recall the definition of eijk , just below equation (4.27), which is defined the
same way as eijk . Those equations can be combined to yield the following result for the cross product of two
contravariant basis vectors, and the cross product and triple scalar product of vectors that are expressed in
terms of covariant components,

eijk
ai × aj = √ ak , (4.36)
g
1√
ak = geijk ai × aj , (4.37)
2
eijk
u × v = √ ui vj ak , (4.38)
g
eijk
(u v w) = √ ui vj wk . (4.39)
g

We leave the proofs of these relations as exercises. They are very similar to the previous proofs, and you
should be able to complete these tasks without looking back in the text too much.
Exercise 4.11 Please prove equations (4.36)–(4.39).
50 CHAPTER 4. COORDINATE SYSTEMS

4.7 Derivatives of Basis Vectors — the Christoffel Symbols

In many calculations we require not only the basis vectors, but also their derivatives with respect to the
coordinates. These can be expressed in terms of the basis vectors themselves, and the result can be written
as
∂ai
= Γkij ak , (4.40)
∂xj
where we have introduced the Christoffel symbols of the second kind Γijk . For the contravariant basis vector
we have
∂ai
= −Γijk ak . (4.41)
∂xj
As an example, in equation (4.5) we have established the expression for the velocity of a particle in terms
of contravariant coordinates. The acceleration of the particle can be expressed in terms of the Christoffel
symbols,  k 
dv dv k i j
= + Γij v v ak . (4.42)
dt dt
We leave the derivation as an exercise.
Exercise 4.12 Derive equation (4.42). (Remember that the basis vectors depend on time through the time-
varying coordinates.)
Exercise 4.13 The Christoffel symbols of the second kind.
• Please show from the definitions of the covariant basis vectors and the Christoffel symbols that Γikj =
Γijk .
Exercise 4.14 What are the values of the Christoffel symbols for a Cartesian coordinate system?

4.8 From Covariant to Contravariant and Vice Versa

Let us now address the question what the relation is between the covariant and contravariant basis vectors
and components of a vector, and how we can convert one into the other. We will show that:
vi = gij v j , (4.43)
i ij
v = g vj , (4.44)
j
ai = gij a , (4.45)
i ij
a = g aj . (4.46)
ij −1
[g ] = [gij ] . (4.47)
The last expression states that the matrix [g ij ] is the inverse of the matrix g = [gij ]. Apparently, all
conversions of components and base vectors can be achieved with the metric tensor g and its inverse. We
leave the proofs of these relations as a set of exercise. These are relatively straightforward, and allow you to
practice a bit more with index index notation and manipulation.
Exercise 4.15 Consider a vector v and its representations in terms of contravariant and covariant compon-
ents,
v = v i ai = vi ai .
1. Prove equations (4.43) and (4.44) by taking the dot product of this equation with suitable vectors.
2. Prove equations (4.45) and (4.46).
3. Prove equation (4.47) from equations (4.43) and (4.44) (beware of index collisions).
4.9. ORTHO-CURVILINEAR SYSTEMS; PHYSICAL COMPONENTS AND BASIS VECTORS 51

4.9 Ortho-Curvilinear Systems; Physical Components and Basis Vectors

Let us take another look at equation (4.16). In the special case of an ortho-curvilinear (OCL) coordinate
system, by definition the local basis vectors are orthogonal. This means that all off-diagonal elements are
zero,    2 
a1 · a1 0 0 h1 0 0
[gij ] =  0 a2 · a2 0  =  0 h22 0  . (4.48)
0 0 a3 · a3 0 0 h23
In the last expression we have introduced the scale factors or Lamé coefficients hi of the coordinate system,
which represent the lengths of the covariant basis vectors,
√
hi := ||ai || = ai · ai . (4.49)

Exercise 4.16 Consider an orthogonal coordinate system. Show that in this case the matrix [g ij ] is given by
  −2
a1 · a1
 
0 0 h1 0 0
ij
[g ] =  0 a2 · a2 0 = 0 h−2
2 0 . (4.50)
0 0 a3 · a3 0 0 h−2
3

Examples of ortho-curvilinear coordinate systems are the Cartesian, cylindrical and spherical coordinate
systems. Let us summarize the definitions of these systems and their scale factors:

• Cartesian coordinates (x, y, z):

hx = 1; (4.51)
hy = 1; (4.52)
hz = 1. (4.53)

• Cylindrical coordinates (z, r, ϕ):

hz = 1; (4.54)
hr = 1; (4.55)
hϕ = r. (4.56)

• Spherical coordinates (r, θ, ϕ):

hr = 1; (4.57)
hθ = r; (4.58)
hϕ = r sin θ. (4.59)

Remember that the scale factors are the lengths of the covariant basis vectors. The cylindrical coordinate
system is an extension of the two-dimensional polar coordinate system with a z-axis (for which hz = 1). The
numbers that appear in this table have been computed in section 1.5.3.
An interesting consequence of the orthogonality of the basis vectors, and the resulting expression for the
metric tensor and its inverse, is that the covariant and contravariant are proportional: from equation (4.45)
or (4.46) we find that

ai = h−2 i −1
i ai =⇒ e(i) ≡ hi a = hi ai (no summation) . (4.60)
52 CHAPTER 4. COORDINATE SYSTEMS

For the components we find, using equation (4.43) or (4.44), that

ui = h−2
i ui =⇒ u
(i)
≡ hi ui = h−1
i ui (no summation) . (4.61)
Here we have introduced the normalized basis vectors e(i) and the components u(i) . These allow yet another
representation of a vector u, since combining these relations yields
u = ui ai = ui ai = u(i) e(i) . (4.62)

The latter is the physical component representation. The numbers u(i) are called the physical components,
the basis vectors e(i) are termed the physical basis vectors. For physical components and basis vectors, the
placement of the indices is arbitrary, since these are neither covariant, nor contravariant. We will therefore
also write u(i) = u(i) and e(i) = e(i) . We will use the placement of the indices only to support the Einstein
summation convention. As before, if (and only if) the same index occurs once as a subscript and once as a
superscript, summation over that index is implied.
Exercise 4.17 Please show that ui ai = u(i) e(i) and that ui ai = u(i) e(i) . (Hint: use explicit summation signs
where necessary.)
The physical component representation is interesting because the associated basis vectors are not only
orthogonal, but also normalized: from definition (4.60) it immediately follows that
e(i) · e(i) = 1. (4.63)
This means that the components have the same unit as the vector itself. Moreover, the dot product of two
vectors is given by
u · v = u(i) v(i) (4.64)
and, in particular, the length of the vector can be calculated as
q
||u|| = u(i) u(i) . (4.65)

Note that in these last two expressions we have used the freedom in the placement of the physical indices to
enable the Einstein summation.
Exercise 4.18 This exercise generalizes exercise (1.12), using the newly acquired language. Please show that
dxi
 
dx X
= hi e(i) . (4.66)
dt i
dt

Do we know what the dimension of ∂xi /∂t is? And of hi ? What is the dimension of hi ∂xi /∂t? Also show
that the speed is given by ‘Pythagoras’,
v
u 2
dxi

dx uX
= t hi . (4.67)
dt i
dt

Note that here we have used an explicit summation to make clear that this is a sum of squares, not the square
of a sum.
In later chapters we will discuss such concepts as the gradient, the divergence and the curl. You can
find the expressions for these operations in many textbooks and on many websites. In many of these
presentations the physical components of the vectors are used, although that is rarely stated explicitly. It
is therefore important that you have a thorough understanding of the three representations of vectors, and
recognize which one is assumed. We will get back to this later.
4.10. VOLUME AND SURFACE ELEMENTS 53

4.10 Volume and Surface Elements

Consider a covariant basis ai and consider an infinitesimal change dx1 of coordinate x1 . This corresponds
to a displacement (∂x/∂x1 )dx1 = a1 dx1 . In the same way, changes in the coordinates x2 and x3 result in
displacements a2 dx2 and a3 dx3 , and the infinitesimal volume of the parallelepiped that is spanned by these
displacements is given by the scalar triple product

dV = (a1 dx1 a2 dx2 a3 dx3 ) = (a1 a2 a3 )dx1 dx2 dx3 . (4.68)

If the transformation is from Cartesian coordinates xi to the coordinates xi , equation (4.24) tells us that
this can be written as
√
dV = det(J) dx1 dx2 dx3 = g dx1 dx2 dx3 . (4.69)
√
This shows that the factor g, is the ratio of the physical volume dV and the ‘coordinate volume’ dx1 dx2 dx3 .
This physical volume is what you need if you evaluate volume integrals. The integral of a field f (x1 , x2 , x3 )
over a volume that is defined by the coordinate intervals [x1− , x1+ ] et cetera is given by
3 2 1
Zx+ Zx+ Zx+
√
f (x1 , x2 , x3 ) g(x1 , x2 , x3 )dx1 dx2 dx3 . (4.70)
x3− x2− x1−

Equation (4.70) takes a familiar form for the usual coordinate systems. The calculation of the volumes
is left as an exercise.

Exercise 4.19 Calculation of volume elements in orthogonal coordinate systems.

√
1. Consider an orthogonal coordinate system. Show that in this case g = h1 h2 h3 , the product of the
scale factors, and that dV = (h1 dx1 )(h2 dx2 )(h3 dx3 ). Interpret the latter result.

2. Give an expression for the volume element dV in Cartesian coordinates;

3. Give an expression for the volume element dV in cylindrical coordinates;

4. Give an expression for the volume element dV in spherical coordinates;

Next, let us consider the directed surface elements that is obtained by taking the cross product of the
vectors a1 dx1 and a2 dx2 . The result is given by
√
dA3 := a1 dx1 × a2 dx2 = a1 × a2 dx1 dx2 = g a3 dx1 dx2 . (4.71)

Similar relations can be derived for the oriented surfaces perpendicular to the x2 and to the x3 coordinate
curves. The results can be summarized as
√
dA1 = g a1 dx2 dx3 ,
√
dA2 = g a2 dx1 dx3 , (4.72)
√
dA3 = g a3 dx1 dx2 .

We see that the contravariant basis vectors are directly related to elements of coordinate surfaces, just like
covariant basis vectors are related to coordinate lines.
A further explanation of these expressions is similar to the one that was provided for the volume element,
equation (4.69). As an example we consider the expression for dA1 . This is a directed surface element, which
54 CHAPTER 4. COORDINATE SYSTEMS

is directed perpendicularly to the surface and has a norm that is equal to the size of the surface element.
This absolute value is given by
√
dA1 = ||dA1 || = ||a2 × a3 || dx2 dx3 = g ||a1 || dx2 dx3 .
√
The factor g ||a1 || translates the ‘coordinate surface element’ dx2 dx3 into a ‘physical’ surface element dA1 .
Again, these expressions will come to live once you specialize them for the usual coordinate systems. This is
left as an exercise:
Exercise 4.20 Calculation of surface elements in orthogonal coordinate systems.

1. For an orthogonal coordinate system we have ||a2 × a3 || = ||a2 || ||a3 ||. Show that in this case dA1 =
(h2 dx2 )(h3 dx3 ) and interpret this expression;
2. Give an expression for the surface element dAx in Cartesian coordinates;
3. Give an expression for the surface element dAr in cylindrical coordinates;
4. Give an expression for the surface element dAr in spherical coordinates;

Please make a sketch of these coordinate systems and visualize the surface elements.
These ‘physical’ surface elements are what you need if you evaluate surface integrals. We will discuss such
integrals in a later chapter.

4.11 Summary and Outlook

In this chapter we have discussed curvilinear coordinate systems. We have discussed that a set of basis vectors
can be introduced by considering, in each point, the partial derivatives of the position vector x(x1 , x2 , x3 )
with respect to the coordinates of the system. This basis is called the (local) covariant basis. If a vector is
expressed in terms of this basis, the components are called the contravariant components.
A second basis could be obtained by considering vectors perpendicular to coordinate surfaces. This
contravariant basis is accompanied by covariant components. Covariant quantities are written with subscript
indices, contravariant quantities with superscripts.
It has been shown that the calculation of lengths, areas and volumes requires knowledge of the dot
products of the basis vectors. These can be assembled in the metric tensor. The metric tensor is also needed
to convert covariant representations into contravariant ones, and vice versa. For orthogonal coordinate
systems the metric tensor only has diagonal elements, the square roots of these elements are called the scale
factors of the coordinate system. These scale factors relate displacements of the coordinates into ‘physical’
displacements (as in δϕ → rδϕ, for cylindrical coordinates).
A third representation of vectors uses physical components and normalized basis vectors. A practical
advantage of this representation is that Pythagoras’ theorem can be applied directly to these components in
order to obtain the size of the vector. (We have not addressed the conceptual disadvantages of this ‘physical
representation’, that is outside the scope of this text.)
In the next chapters we will start to use the tools that we have developed in this chapter to study vector
fields. In particular, we will discuss the gradient, divergence and curl, which are omnipresent in problems in
mechanics, transport physics, electro-magnetism and other branches of physics.

Exercise 4.21 Look at the list of objectives at the beginning of this chapter and ensure that you have acquired
the mentioned skills. If not, please study the relevant parts of this chapter again and repeat this exercise.
Chapter 5

Flux Densities & the Divergence of a

Vector Field

Objectives
After studying this chapter, you will be able to
• Explain the concept of a flux variable by considering transport through a fixed directed surface element;
• Express fluxes through a finite-sized oriented surface as a surface integral over the normal component
of the flux density;
• Explain the meaning of the orientability of a surface and provide an example of a surface that is not
orientable;
• Construct the integral form of the continuity equation that describes mass flow in transport physics;

• Construct the integral form of the species mass balance that describes particle flow and transport in
transport physics;
• Reproduce the coordinate-free definition of the divergence and explain it by considering the steady-state
species mass balance;

• Derive the differential form of the divergence of a vector field in terms of that field’s contravariant
components;
• Specialize that expression for the usual Cartesian, cylindrical and spherical coordinate systems in terms
of physical vector components;
• Calculate the divergence for a given field;

• Explain the meaning of the divergence (Gauß, Ostrogradsky) theorem;

• Use the latter theorem to simplify the calculation of some volume or flux integrals.

55
56 CHAPTER 5. FLUX DENSITIES & THE DIVERGENCE OF A VECTOR FIELD

dx = vdt

||dA||

Figure 5.1: Consider a fixed surface element dA = n ||dA|| = n dA. If the medium is moving with velocity v at
the location of the surface, its displacement in a time dt is given by dx = vdt. The parallelepiped with volume
dV = dx · dA = v · n dAdt contains the medium that passed through the surface dA in time dt.

5.1 Introduction
In the preceeding chapters we have discussed various vector-valued functions of space and/or time coordin-
ates. An important class of such vector fields represent flux densities. Such variables may express the rate
and direction of the transport of an associated material quantity, such as mass or charge. Fluxes also play
an important role in electromagnetics, as an example we can think of the magnetic flux through a surface,
whose changes induce an electromotive force according to Faraday’s law.
In this chapter we will present the concept of a flux density, and how that can be constructed from an
analysis of transport through a fixed directed surface element. We will then consider transport through a
finite-sized surface and express that as a surface integral. The result will be used to derive two important
relations from transport physics: the continuity equation and the mass balance for a given component
(species) of a multi-component system. We will use the latter equation to derive the concept of the divergence
of a field and demonstrate that it can be interpreted as the source of that vector field. In the last part of
this chapter we will derive an alternative expression for the divergence and discuss the divergence theorem
and its applications.

5.2 Flux Densities

Let us consider a medium with a mass density ρ(x) and a velocity v(x). Also consider a fixed directed
surface element dA somewhere in space. We can ask ourselves the question: how much mass is transported
through that surface element per unit of time?
To answer that question we will first investigate the size of the volume that moved through the surface
element in a time interval dt. Picture 5.1 visualizes that this volume dV is defined by the parallelepiped that
is defined by dA and the vector dx = v dt.

Exercise 5.1 The mass flow rate through an infinitesimal surface element.

1. Express the volume dV in terms of dA, v and dt;

2. What is the mass that is contained in this volume?

3. Show that the mass flux dΦ per unit of time through the surface element dA is given by

dΦ = ρv · dA. (5.1)
5.2. FLUX DENSITIES 57

Once we have calculated the flux through a surface element dA, we can obtain the flux through a finite-sized
surface A by integrating over that surface,
Z
ΦA = ρv · dA. (5.2)
A

This requires that the surface is orientable, which means that a consistent choice can be made for the
orientation of the surface, which is usually the case for problems of practical interest. A brief discussion of
the subject is provided in section 5.3.
The quantity ρv is called the mass flux density. It is an example of a flux density, a vector that describes
how much of a given property (here mass) is transported per unit of time and per unit of cross-sectional area
in the direction in which it points.
Flux densities can be defined for many quantities. In transport physics they typically take the form
‘density of a quantity × the velocity with which that quantity is transported’. There are also flux density
variables that do not have an obvious associated flowing material property: examples are the electric field
E and the magnetic flux density B. In order to highlight the general nature of the text that follows, let us
use the symbol Γ for the flux density, rather than ρv, so the previous equation becomes
Z
ΦA = Γ · dA. (5.3)
A

Before discussing some applications of this expression we will present two other common ways of writing the
expression for the flux ΦA through a surface.
Firstly, in the expressions above we have used the directed surface element dA, which encodes both the
orientation and the size of the surface element. It is straightforward to rewrite dA as the product of a
normalized vector and a scalar that represents the size dA = ||dA|| of the element:
dA
dA = ||dA|| = n dA. (5.4)
||dA||
The vector n has unit length and points perpendicularly to the surface, and of course you know this vector
very well under the name normal vector. In terms of the normal vector the previous expression for the flux
can be cast as Z
ΦA = Γ · n dA. (5.5)
A
There is a third expression that you will routinely find in literature. That is based on the notion that
Γ · n is the normal component of the flux Γ. It is not unusual to see this written as Γn , which results in the
expression Z
ΦA = Γn dA. (5.6)
A

A flux density Γ(x, t) is a vector-valued function of position and time that expresses the rate of
transport of a quantity per unit of time and per unit of area perpendicular to the direction in which it
points. For a given oriented surface A, the transport rate can be calculated by integrating the normal
component of the flux. This can be written in three equivalent ways as
Z Z Z
ΦA = Γ · dA = Γ · n dA = Γn dA. (5.7)
A A A

Note that in much literature the word flux is used instead of flux density. Such is life.
58 CHAPTER 5. FLUX DENSITIES & THE DIVERGENCE OF A VECTOR FIELD

(a) Schematic view. (b) A Möbius strip at TU/e

Figure 5.2: The Möbius strip is the canonical example of a non-orientable surface. The picture on the left shows
a schematic view, the picture on the right depicts the Möbius strip at Eindhoven University of Technology by
Hans Kalkhoven after a design by Dom Hans van der Laan. It was a gift from Philips at the occasion of the 30’th
anniversary of TU/e in 1986. The text shows the university’s motto Mens Agitat Molem, Het menselijke vernuft
brengt de dode materie tot leven. (left: [Link] (public domain), right:
[Link] (unknown).)

5.3 Orientable and Non-Orientable Surfaces

In our discussion of equation (5.2) we have mentioned that the surface A must be orientable for the surface
integral to exist. Essentially this means that we must be able to move along the surface continuously
without ever ending up on the other side. All surfaces of practical interest in physics and engineering have
this property, but it is important to know that non-orientable surfaces exist. The singly best-known example
is the Möbius strip (Dutch: Möbiusring). This has been visualized in figure 5.2. Apart from a schematic
view it also shows the Möbius strip that is located on the campus of TU/e.
Such surfaces have been the topic of extensive study by mathematicians, all of which is —unfortunately—
outside the scope of this text. They have many interesting properties that you may already have seen before.
If not, please take a look at the following exercise.

Exercise 5.2 Create Möbius strip from a piece of paper and some Scotch tape. Start cutting it along the
length of the strip until you have reached the initial point. What is the result? Repeat this exercise one more
time. What happens then?

Exercise 5.3 In old-fashioned mechanical type writers and dot-matrix printers, the ink ribbon was folded to
form a Möbius strip. Why do you think that was done?

5.4 Examples: Balance Equations

5.4.1 The Continuity Equation
Let us now revisit the expression for the mass flux, with Γ = ρv. This time we consider a fixed volume
V with a boundary surface A that we will write as ∂V . For closed surfaces we always choose the surface
5.5. THE DIVERGENCE — COORDINATE-FREE DEFINITION 59

elements to be pointing in the outward direction. Then the total outward mass flux is given by integrating
the mass flux rate over this (closed) surface, the result is
Z
Φ = (ρv · n)dA. (5.8)
∂V

Next we calculate the total mass that is contained in the volume. This can be obtained by integrating the
mass density over the volume, Z
m = ρ(x)dV. (5.9)
V

As physicists we know that mass is conserved — it cannot be produced or destroyed. This means that
changes in m can only be due to transport. Equating the rate of change of m to the inward transport, and
realizing that Φin = −Φ, we get
Z Z Z Z
d d
ρ(x)dV = − (ρv · n)dA, or ρ(x)dV + (ρv · n)dA = 0. (5.10)
dt dt
V ∂V V ∂V

This is the continuity equation, one of the cornerstones of fluid dynamics.

Exercise 5.4 Please simplify equation (5.10) for the case of an incompressible medium, which has a uniform
and constant mass density ρ.

5.4.2 The Species mass Balance

The concept of a flux density can be introduced for many variables. As an example, we can consider particles
of type s (from ‘species’) in a multi-component gas mixture. This time we are interested in the particle flux
density, assuming that locally the particle density is given by ns , while the mean molecular speed is given by
vs . Furthermore we assume that the particles of type s are produced at a rate Ss (x, t) per unit of volume
and time in chemical reactions.

Exercise 5.5 The species mass balance.

1. Derive an expression for the particle flux density Γs . Try to derive this expression by repeating the
derivation of the mass flux density earlier in this chapter, ideally without looking back at the explanation
given there;
2. Derive the species mass balance,
Z Z Z
d
ns (x)dV + (Γs · n)dA = Ss (x)dV (5.11)
dt
V ∂V V

and explain the meaning of the individual terms.

5.5 The Divergence — Coordinate-Free Definition

In the last section we have seen the mass balance. Let us now assume that the system is stationary, which
means that the time-derivative on the left hand vanishes,
Z Z
(Γs · n)dA = Ss (x)dV. (5.12)
∂V V
60 CHAPTER 5. FLUX DENSITIES & THE DIVERGENCE OF A VECTOR FIELD

(a) div Γ > 0 (b) div Γ = 0 (c) div Γ = 0 (d) div Γ < 0

Figure 5.3: A graphical visualization of the meaning of the divergence of a vector field, as defined by equation
(5.15). When the flux vectors in some neighbourhood tend to diverge in a point, the divergence is positive
(a), when they converge the divergence is negative (d). The divergence can be zero for non-uniform fields; this
happens if the variations occur in directions perpendicular to the direction of the flux (b) or otherwise there is
a local balance between ingoing and outgoing flux contributions (c). In all cases the divergence is related to the
total net flux that leaves the virtual dashed volume.

Let us divide this equation by the volume V . Then the right-hand side can be interpreted as the mean value
Ss of Ss in the volume V , the result is
Z Z
1 1
Ss = Ss (x)dV = (Γs · n)dA. (5.13)
V V
V ∂V

It is important to realize that this relation holds for an arbitrary volume. Let us consider what happens if
we consider a volume around a point x and consider the limit that the volume is shrunk to zero size. In that
case the average value Ss will approach the value Ss (x) in the point under consideration. This demonstrates,
informally, that the limit of the right-hand side exists as well,
Z
1
Ss (x) = lim (Γs · n)dA. (5.14)
V →0 V
∂V

We define the quantity on the right-hand side, which depends solely on the (flux density) vector Γ, as the
divergence of the flux density: Z
1
∇ · Γ = lim (Γ · n)dA. (5.15)
V →0 V
∂V

An alternative notation for the divergence of a flux field Γ is ‘div Γ’. You will recognize ∇ as the nabla-
operator and may have noticed that the notation ∇ · Γ suggests that taking the divergence somehow amounts
to taking the dot product of ∇ and that field. Later in this section we will see that this is essentially true.
For now, please accept ∇ · Γ simply as a notation of the divergence defined by equation (5.15). The definition
of the divergence shows that its local value is related to the net flux out of a ‘small’ test volume. Figure 5.3
makes this more clear by showing typical vector field patterns and the resulting divergence.
Note that we have used the mass balance as a vehicle to make it plausible that the divergence, as defined
above, exists, but that the definition is completely independent of that context. The divergence is defined
for any vector field, be it a flux or a quantity with a completely unrelated meaning. Please realize that the
definition (5.15) is also independent of the coordinate system that is used.
In this section we have motivated the definition of the divergence by considering the steady-state case
of a transport problem. The following exercises elaborate on this particular derivation and present a sneak
preview of two important results: the differential form of the (steady-state) transport equation and the
5.6. THE DIVERGENCE — DIFFERENTIAL FORMULATION 61

y + ∆y/2
x2 + ∆x2 /2

y − ∆y/2 x2 − ∆x2 /2
x1 + ∆x1 /2
1 1
x − ∆x/2 x + ∆x/2 x − ∆x /2
(a) Cartesian grid. (b) Curvilinear grid

Figure 5.4: A volume is bounded by three sets of nearby coordinate surfaces. The picture shows a two-dimensional
cross section, in reality the volume also has a vertical extent. In general the coordinate lines and surfaces are
curved (right), for a Cartesian system (left) these are straight lines and flat surfaces, respectively.

divergence theorem. The latter is of the utmost importance, more generally valid, and the topic of section
5.8.

Exercise 5.6 Show that in the construction that resulted in definition (5.15) we have established a relation
between Ss and the divergence of the flux density in a given point. This relation is the differential form of
the steady-state mass balance. Please state this equation.

Exercise 5.7 Following up on the previous exercise: substitute the expression for Ss in terms of the divergence
of the flux in equation (5.12). Take some time to digest the result, which is a manifestation of the Divergence
Theorem, also named after Gauß and Ostrogradsky. (It turns out to be generally valid: we have merely shown
that it applies to our transport equation — we have by no means provided a general proof.)

5.6 The Divergence — Differential Formulation

Expression (5.15) for the divergence has the advantage that it is a coordinate-free formulation. It is, however,
not very practical and fortunately we can derive a more convenient expression for the divergence of a vector
field. To that end we will apply equation (5.15) to a small box-shaped volume that is aligned with the
coordinate surfaces of a coordinate system. We will first demonstrate the calculation for the special case of a
Cartesian coordinate system, then repeat the exercise for the general case of a curvilinear coordinate system.
In order to derive the results for a Cartesian coordinate system, please consider a ‘coordinate volume’
V = ∆x × ∆y × ∆z centered around the poins (x, y, z). Figure 5.4(a) shows a two-dimensional cross section
of this box, in reality it also has a height ∆z. We will calculate the divergence using equation (5.15) and to
that end we split the surface integral in six parts, one for every side of the box. Let us first look at the flux
that leaves the box at the ‘eastern’ side at location x + 21 ∆x. The area is equal to the product ∆y∆z and
the integral flux can be approximated by Γx (x + 12 ∆x, y, z)∆y∆z. To obtain the net flux in the x-direction
we subtract the flux that enters at the western side, Γx (x − 21 ∆x, y, z)∆y∆z. The result is

Γx (x + 12 ∆x, y, z) − Γx (x − 21 ∆x, y, z) ∂Γx

Γx (x + 21 ∆x, y, z) − Γx (x − 12 ∆x, y, z) ∆y∆z =
 
∆x∆y∆z ≈ V.
∆x ∂x
62 CHAPTER 5. FLUX DENSITIES & THE DIVERGENCE OF A VECTOR FIELD

The net flux that flows through the y and z coordinate surfaces is obtained in an identical way, and for the
total surface flux integral we find
Z   Z
∂Γx ∂Γy ∂Γz 1 ∂Γx ∂Γy ∂Γz
(Γs · n)dA ≈ + + V ⇐⇒ (Γs · n)dA ≈ + + . (5.16)
∂x ∂y ∂z V ∂x ∂y ∂z
∂V ∂V

As the intervals ∆x, ∆y and ∆z tend to zero, the approximation becomes better and better, and combination
with (5.15) yields, for the limit that V → 0,
∂Γx ∂Γy ∂Γz
∇·Γ= + + . (5.17)
∂x ∂y ∂z
We will now derive the expresion for the divergence of a vector for a general curvilinear coordinate system
(x1 , x2 , x3 ). The result is
√

1 ∂ g Γ · ai
∇·Γ= √ , (5.18)
g ∂xi
or, using Γ · ai = Γi ,
√

1 ∂ g Γi
∇·Γ= √ . (5.19)
g ∂xi
The derivation follows the same steps as before. We choose a small coordinate box with sides ∆xi around a
point xi , as depicted in figure 5.4(b). Note that in general the coordinate lines and surfaces are curved, and
that, just like before, the x3 -direction is not drawn. We leave the details of the derivation for an exercise.

Exercise 5.8 The divergence in curvilinear coordinates.

1. Provide an expression for the outward surface element ∆A(e) at the ‘eastern’ side x1 + 21 ∆x1 , x2 , x3 .

Show that the contribution to the integral of the eastern side is given by
√
Z
gΓ · a1 e ∆x2 ∆x3 .

Γ · dA ≈ [Γ · ∆A]e =
east

2. Combine this expression with that for the outward flux at the western side, and show that the result
can be written as
√
√ 1
√ 1 2 3 ∂( gΓ · a1 )
∆x1 ∆x2 ∆x3 .
 

gΓ · a e − gΓ · a w ∆x ∆x ≈
∂x1

3. Provide an (approximate) expression for the physical volume V under consideration and show that the
previous expression can be approximated as
√
1 ∂( gΓ · a1 )
√ V
g ∂x1

4. Also consider the other two directions, not just east and west; argue that the complete surface integral
can be approximated as √
1 ∂( gΓ · ai )
Z
Γ · dA = √ V.
g ∂xi
∂V

Dividing by V , taking the limit V → 0 and using definition (5.15) gives the requested result (5.19).
(Note that all approximations become exact as the coordinate intervals tens to zero.)
5.7. THE DIVERGENCE IN ORTHO-CURVILINEAR COORDINATES 63

5.7 The Divergence in Ortho-Curvilinear Coordinates

Let us take another look at the expression for the divergence in general curvilinear coordinates that is given
by equation (5.19),
√

1 ∂ g Γi
∇·Γ= √ .
g ∂xi
We will now investigate what this expression looks like in orthogonal curvilinear coordinates. Please note that
√
under such circumstances the volume g can be expressed in terms of the scale factors hi of the coordinate
system. Secondly, in orthogonal coordinate systems it is sometimes useful to use ‘physical’ components Γ(i) ,
rather than the contravariant Γi or covariant Γi . This has been discussed in section 4.9. The expressions
that you find in most books assume such physical coordinates, even when this is usually not stated. Care
must therefore be taken.
We leave it as an exercise to demonstrate that for an ortho-curvilinear coordinate system the divergence
of a vector Γ = Γ(i) e(i) is given by

∂(h2 h3 Γ(1) ) ∂(h1 h3 Γ(2) ) ∂(h1 h2 Γ(3) )

 
1
∇·Γ= + + . (5.20)
h1 h2 h3 ∂x1 ∂x2 ∂x3
Exercise 5.9 Please derive equation (5.20). Consult exercise 4.19 and section 4.9, if necessary.
For Cartesian coordinate systems the lengths of the basis vectors are unity, so hx = hy = hz = 1. For cyl-
indrical and spherical coordinate systems the scale factors have been discussed in section 4.9. The expressions
for the divergence in such coordinate systems are then given by:
 (x)
∂Γ ∂Γ(y) ∂Γ(z)
+ + Cartesian;





 ∂x ∂y ∂z
1 ∂rΓ(r) 1 ∂Γ(ϕ) ∂Γ(z)

∇·Γ= + + Cylindrical; (5.21)

 r ∂r r ∂ϕ ∂z
1 ∂r2 Γ(r) 1 ∂sin θΓ(θ) 1 ∂Γ(ϕ)




 2 + + Spherical.
r ∂r r sin θ ∂θ r sin θ ∂ϕ
The first of these expressions was of course constructed already in section 5.6. Please note that we have used
parentheses to indicate that the components are the physical components. In literature you will typically
find, for example, ϕ instead of (ϕ) and the nature of the components is made clear in the context — if you
are lucky.
Exercise 5.10 Please prove the expressions in equation (5.21).
Exercise 5.11 Please calculate the divergence of each of the following vector fields. (The arguments indicate
whether the coordinate system is Cartesian, cylindrical or spherical.
1. v(x(x, y, z)) = x3 ex + y 3 ey + z 3 ez ;
2. v(x(r, ϕ, z)) = r cos2 ϕ er + 2z sin2 ϕ ez , for r > 0;
3. v(x(r, θ, ϕ)) = r−n er , for r > 0.
Exercise 5.12 Please calculate the divergence of the position vector x. Do this in Cartesian, cylindrical and
spherical coordinate systems and verify that the results are equal.
1 q
Exercise 5.13 Please consider the vector field E(r, θ, ϕ) = 4πϵ 2 er for r ≥ 0. What is the physical signific-
0 r
ance of this expression? In the preceeding sections we have seen that the divergence of a field is related to
the ‘source’ of that field. What is that source here, and where is it located? Discuss that your result for the
divergence is consistent with this fact.
64 CHAPTER 5. FLUX DENSITIES & THE DIVERGENCE OF A VECTOR FIELD

5.8 The Divergence Theorem

In section 5.5 we have introduced the coordinate-free definition of the divergence. This definition was inspired
by a manipulation of the mass balance for a species, in which we considered the limit that the test volume
was shrunk to zero size. In the process we made a remarkable discovery, which was the subject of exercise
5.7. The result of that exercise is that, for that particular case, the volume integral over the divergence
of a vector field is equal to the integral of the vector field itself over the surface that bounds the volume.
This result turns out to be generally valid, and is known as the Divergence Theorem. Depending on your
geographic location it is also known as the Gauß Theorem or the Ostrogradsky Theorem. It states that:

For a volume V with (closed) boundary surface ∂V , and for a continuously differentiable vector field
Γ, Z Z
(Γ · n)dA = ∇ · Γ dV. (5.22)
∂V V

The divergence theorem has numerous applications in mathematics, physics and engineering. As an
example, it can be used to transform integral equations into differential equations, as shown by the folowing
exercise.
Exercise 5.14 In equation (5.11) we have stated the integral mass balance for a species s. We will now
consider this equation for a fixed finite-sized volume V .
1. Show that this equation can be rewritten as
Z  
∂ns
+ ∇ · Γs − Ss dV = 0;
∂t
V

Bringing the d/dt inside the integral over the fixed volume is an application of the more general Leibniz
integral rule. Please look up yourself how that works if necessary;
2. Suppose that the integrand is non-zero in some neighborhood of a point x in V . Then we could construct
a small volume around that point where the relation above would fail. But that relation is valid for any
volume: conclude that the integrand must vanish identically. This results in the differential form of the
mass balance, which is usually written as
∂ns
+ ∇ · Γs = Ss . (5.23)
∂t
3. Make sure you understand the argument why the integrand must vanish identically. Starting from the
integral mass balance, you should be able to complete this entire exercise without looking at the text.
Please try this.
In the remainder of this section we will demonstrate that the divergence theorem can be used to simplify
the calculation of some nasty-looking volume and surface integrals. For example, if the bounding surface is
divided into ∂V1 and ∂V2 , the divergence theorem yields
Z Z Z
(Γ · n)dA = ∇ · Γ dV − (Γ · n)dA. (5.24)
∂V1 V ∂V2

There are situations where both terms on the right-hand side can be easily calculated, while the surface
integral on the left-hand side cannot. Then the expression above can be used for the indirect evaluation of
the flux integral on the left-hand side.
5.9. SUMMARY AND OUTLOOK 65
√
Example 5.1 Consider the vector field F (x) = (x + 1)ex + sin(2πz)ey + 1 − x2 ez and a volume V that is
the half-sphere x2 + y 2 + z 2 ≤ 1, x ≥ 0. Calculate the total flux through the curved part of the boundary
surface, C, which is defined by x2 + y 2 + z 2 = 1, x ≥ 0.
We start with the observation that the divergence is given by ∇ · F = 1 (please verify this). Then the
right-hand side of equation (5.22) is equal to 1 × V = 12 34 π13 , half of the volume of a sphere with radius 1.
This must be equal to the total flux through the surface. We divide that flux in two parts: the unknown flux
ΦC and the flux ΦP through the planar part of the boundary surface, which is located at x = 0. In the planar
part we have n = −ex , so F · n = −(x + 1) = −1, the surface of the planar part is equal to π12 , that of a
circle with radius 1. Then ΦP = −1 × π12 = −π. Wrapping up, we find that
2 5
ΦC − π = π =⇒ ΦC = π.
3 3
Exercise 5.15 Consider a cone that is described by the relation r ≤ R − z for z ∈ [0, R]. Consider the flux
field v(z) = zez . We use cylindrical coordinates. What is the flux through the surface of the cone, excluding
the base plane at z = 0? (Note: as you know, the volume of a cone with radius R and height H is given by
1 2
3 πR H.)

5.9 Summary and Outlook

In this chapter we have introduced two concepts that are omnipresent in mathematics, transport physics,
electromagnetism and elsewhere. These are the flux density, fluxes, and the divergence of a vector field. It
was shown that commonly the divergence can be interpreted as the ‘source’ of a vector field. The divergence
is defined in a coordinate-free way, but expressions in terms of general curvilinear coordinates were presented
as well, then specialized for the usual Cartesian, cylindrical and spherical coordinate systems. The divergence
theorem and some of its basic applications have been presented. In the next chapters we will present two
more differential operators on vectors, the gradient and the curl.

Exercise 5.16 Look at the list of objectives at the beginning of this chapter and ensure that you have acquired
the mentioned skills. If not, please study the relevant parts of this chapter again and repeat this exercise.
66 CHAPTER 5. FLUX DENSITIES & THE DIVERGENCE OF A VECTOR FIELD
Chapter 6

The Gradient of a Scalar Field, Line

Integrals

Objectives
After studying this chapter, you will be able to
• Obtain the exact differential of a scalar function f (x);
• Express that in terms of the gradient of f and a displacement vector dx;

• Derive the expression for the gradient of f with respect to a contravariant basis;
• Derive the expression for the gradient of f in terms of physical components and normalized basis
vectors;
• Define and explain the concept of a directional derivative;

• Calculate line integrals over vector fields;

• Define the meaning of a conservative vector field;
• Simplify line integrals over conservative vector fields;

• Calculate the work and power that is exerted by a force field on a moving particle;
• Derive the principle of conservation of mechanical energy.

67
68 CHAPTER 6. THE GRADIENT OF A SCALAR FIELD, LINE INTEGRALS

6.1 Introduction
In this chapter we will discuss the gradient of a scalar function and some of its applications. You have already
encountered this operator in your previous studies of mechanics and electrostatics, where you have learned
that a force that can be expressed as the gradient of a scalar potential represents a conservative force field.
That means, among other things, that the work that is exerted by that force on a moving particle depends
only on that particle’s initial and actual positions, not on the path that was followed by that particle.
In this chapter we will recapitulate those results, but there is news as well. Firstly, we will present the
relation between the exact differential of a scalar field and the gradient, and present a definition of the
gradient in general curvilinear coordinates. We will then discuss the concept of the directional, normal and
material derivatives of a scalar field. In the second part we will discuss the gradient in the context of line
integrals over vector fields.

6.2 The Gradient

Consider a scalar field f (x) that depends (only) on position x. How does a small change dx of the position
influence the field value f ? Let us assume that a coordinate system (x1 , x2 , x3 ) has been chosen, so the
position vector itself is a function x(x1 , x2 , x3 ). If the field is continuously differentiable in a neighborhood
of the point, one has
∂f
df = dxi . (6.1)
∂xi
This differential form is called an exact differential. A disadvantage of this formulation is that it expresses
the change in f in terms of changes of particular coordinates. Is it also possible to express it in terms of the
displacement vector dx itself? Let us see how this can be achieved.
We recall that the displacement vector can be expressed in terms of the coordinates (x1 , x2 , x3 ) using the
covariant basis vectors that were introduced in equation (4.4),
∂x
dx = dxj = aj dxj , (6.2)
∂xj
Combining equation (6.1), the duality relation (4.26) and this expression for the displacement vector, it is
easy to show that
∂f i
df = a · dx. (6.3)
∂xi
We leave it as an exercise to verify this result.
Exercise 6.1 Please prove the validity of relation (6.3).
Exercise 6.2 The above derivations reveal some important manipulations involving (curvilinear) coordinates:
the definition of ai , the duality relation ai · aj = δji and the calculation of dx and df in terms of their
arguments. Typical exam stuff, one might say. For this reason: please consider an f (x(x1 , x2 , x3 )) and
derive equation (6.1) yourself, without looking back at the text. If you cannot do this, please study the subject
matter one more time and repeat this exercise.
In equation (6.3) we have introduced a vector that is one of the main subjects of this chapter. It is called
the gradient of a scalar field and is defined as
∂f i
∇f (x(x1 , x2 , x3 )) =
a. (6.4)
∂xi
We see that the derivation of the gradient naturally resulted in an expression in terms of the contravariant
basis vectors ai — and covariant components ∂f /∂xi . Combining the previous results we can write
df = ∇f · dx. (6.5)
6.3. THE MATERIAL DERIVATIVE 69

Vectorial Nature
While the substitution of definition (6.4) into expression (6.3) is a trivial operation, the resulting equation
(6.5) emphasizes an important feature of the gradient. The gradient of a field f is a vector; it is independent
of the choice for a particular coordinate system. This is, of course, to be expected: the change df depends
only on the field f (x) and the displacement dx, and these concepts themselves exist independent of the
choice for a particular coordinate system. While equation (6.4) tells us how to calculate the gradient for a
given coordinate system, the final result will not depend on that particular choice.

The Direction of the Gradient

Let us take another look at expression (6.5). This is the dot product of a vector ∇f and a (displacement)
vector dx. We remember that the dot product attains its highest value if its two arguments are aligned. This
means that df will be highest if dx points in the same direction as ∇f . This demonstrates the important
fact that the gradient points in the direction in which f increases.
A level surface of f is a surface on which f has a constant value. The level surface that contains the point
x0 is defined by the solutions of the equation f (x) = f (x0 ). This implies that df (x(x1 , x2 , x3 )) = ∇f ·dx = 0.
This shows that displacements dx that are orthogonal to ∇f are tangent to a level surface.

6.3 The Material Derivative

We have discussed that the gradient is a measure for the variations of a scalar field with position. The exact
differential (6.3) shows how the local field value changes for a given displacement vector dx. The exact
differential itself is, however, rarely the final result of a calculation. In this and the following sections we
will discuss the relation between the gradient and the differential of a field in more detail, and show more
complete applications of the gradient in the form of examples.
Let us first consider a moving particle at position x(t) in a temperature field T (x). What is the rate of
change of the temperature that is experienced by the particle?
It is important to realize that we are not interested in the time-derivative of the temperature field T (x) as
such: t is not even a parameter of the function T (x), so dT (x)/dt is meaningless. Instead, we are interested
in the rate of change of the temperature at the location of the particle. Since that is given by x(t), we are
interested in the derivative of the one-parameter function T (x(t)). In order to calculate that derivative from
the expression for the differential we choose x = x(t) and invoke the chain rule, the result is

dT (x(t)) = ∇T · dx = ∇T · x′ (t) dt. (6.6)

It follows that the requested rate of change of the local temperature is given by
dT (x(t))
= ∇T · x′ (t). (6.7)
dt
What can we learn from this derivation? It is important to be very precise about the arguments of all the
functions. Whereas T (x) is a very general ‘recipe’ for calculating the temperature in an arbitrary point
x in space, the function T (x(t)) represented the temperature at the particular, time-dependent position of
a particle. Where necessary we have therefore explicitly written T (x(t)) instead of just T . Last but not
least: what can we learn from the resulting expression? We see that when the motion is perpendicular to
the temperature gradient, the rate of change of the local temperature is zero; when it is parallel to it, it is
highest.
Note that the results that were discussed in the last example are correct only for a time-independent
temperature field. In this case the temperature variations that are experienced by the particle are only due
to the variations in the position of the particle, not because the temperature field itself changes. Let us
70 CHAPTER 6. THE GRADIENT OF A SCALAR FIELD, LINE INTEGRALS

now assume that the temperature field also depends on time, so we have T (t, x). How does that change the
results so far?
Essentially we start just like we did in the beginning of this chapter, but this time we consider a function
f (t, x). The differential will now have an additional term due to the time-dependence and is given by
∂f
df = dt + ∇f · dx. (6.8)
∂t
Exercise 6.3 Please derive equation (6.8). You should be able to do this without consulting the text by now.
If you have problems with this, take another very close look at the theory and the exercises in section 6.2.

Let us now use this result to investigate the rate of change of the temperature at the location x(t) of a
particle that moves in a time-dependent temperature field T (t, x).

Exercise 6.4 Consider a moving particle with position x(t) in a temperature field T (t, x). Show that the rate
of change of the temperature that is experienced by the particle is given by
dT (t, x(t)) ∂T
= + ∇T · v, (6.9)
dt ∂t
where all terms in the last expression are evaluated at the actual position of the particle, x(t).

This expression shows that the rate of change has two contributions: the first is also present if the particle
is standing still and represents the change of the temperature itself. The second represents, as before, the
change due to the motion of the particle in a non-uniform field.
This particular derivative is very common, especially in fluid dynamics and is called the material derivative
and is usually denoted with a capital D. Usually the symbol v is used for the velocity and the gradient term
is written at the end of the expression (that is OK, since the dot product is commutative). The expression
then looks like
DT ∂T
= + v · ∇T. (6.10)
Dt ∂t

The material derivative is an operator that is defined by

D ∂
= + v · ∇.
Dt ∂t
When applied to a scalar field, it describes the variations of the value of that field at the location of a
particle that is moving with velocity v.

6.4 Results for Particular Coordinate Systems

Equation (6.4) defines the gradient of a scalar field for a general curvilinear coordinate system. Let us now
specialize this expression for some particular systems.

Example 6.1 In a spherical coordinate system (r, θ, ϕ) the gradient is given by

∂f r ∂f θ ∂f ϕ
∇f (x(r, θ, ϕ)) = a + a + a .
∂r ∂θ ∂ϕ
That’s it. We really only need to differentiate with respect to each coordinate, multiply with the corresponding
contravariant basis vectors, and take the sum of the results.
6.4. RESULTS FOR PARTICULAR COORDINATE SYSTEMS 71

If the coordinate system is orthogonal, folklore has it that one tends to use normalized basis vectors and
expresses vectors in terms of the resulting ‘physical components’, as discussed in section 4.9. By rewriting
ai in terms of e(i) , the result is
3
X 1 ∂f
∇f (x(x1 , x2 , x3 )) = e(i) . (6.11)
h
i=1 i
∂xi
Here we have used equation (4.60). Note that the transition to the physical basis results in a third index i:
in such cases the Einstein convection does not work well, so we have written the summation sign explicitly.
Exercise 6.5 Let us look at the expressions for the gradient in various coordinate systems. In the questions
2–4 you simply substitute the expressions for the scale factors that have been discussed in section 4.9. Those
are common enough that you may want to learn these by heart by now. The exercise is trivial, but it is good
to have seen these common expressions.
1. Please verify the derivation of (6.11) from (6.4);
2. Show that in Cartesian coordinates:
∂f ∂f ∂f
∇f (x(x, y, z)) = ex + ey + ez . (6.12)
∂x ∂y ∂z

3. Show that in cylindrical coordinates:

∂f 1 ∂f ∂f
∇f (x(r, ϕ, z)) = er + eϕ + ez . (6.13)
∂r r ∂ϕ ∂z

4. Show that in spherical coordinates:

∂f 1 ∂f 1 ∂f
∇f (x(r, θ, ϕ)) = er + eθ + eϕ . (6.14)
∂r r ∂θ r sin θ ∂ϕ

Exercise 6.6 Consider a spherically symmetric temperature field T (r). With reference to the discussion of a
moving particle in such field that was provided in section 6.3, what is the rate of change of the temperature
that is experienced by a particle that is:
1. Moving radially outward, r(t) = vr t, where vr is constant?
2. Moving azimuthally, ϕ(t) = ωϕ t, where ωϕ is constant?

Exercise 6.7 The Gradient Theorem. (Exam 3BMX0, 24 January 2017.)

The divergence theorem (5.22) is at the basis of a variety of other interesting vector relations. An example
of such derived result is the gradient theorem, which is the subject of this exercise.
1. Consider a vector field in R3 of the form v(x) = p(x)c, where p(x) is a continuously differentiable
scalar field and c an arbitrary constant vector. Prove that

∇ · (pc) = (∇p) · c.

2. Consider an arbitrary volume V in R3 that is bounded by ∂V . Prove that

Z Z
pndA = (∇p)dV. (6.15)
∂V V

(Hint: first derive an integral relation of the form c · (. . .) = 0 and notice that this must hold for any
vector c, so...)
72 CHAPTER 6. THE GRADIENT OF A SCALAR FIELD, LINE INTEGRALS

Exercise 6.8 Archimedes’ Law. (Exam 3BMX0, 24 January 2017.)

We once more consider the gradient theorem that was derived in the previous exercise. If p represents
pressure, the left-hand side of this equation is equal to the total pressure force that the volume excerts on
its surroundings. Then, according to Newton’s third law, the pressure that the environment excerts on the
volume is given by Z
F =− (∇p)dV.
V

1. We introduce a Cartesian coordinate system (x, y, z). Show that the expression for the gradient in such
system reduces to
∂p ∂p ∂p
∇p = ex + ey + ez . (6.16)
∂x ∂y ∂z
2. Space is filled with a fluid at rest with massa density ρ(z); the gravitational force per unit of volume is
given by fg (z) = −ρ(z)gez . The hydrostatic pressure is given by
Zz
p(z) = p(0) − ρ(z ′ )gdz ′ . (6.17)
0

Show that this pressure is the scalar potential of the field fg .

3. Prove Archimedes’ Law, which states that the total (pressure) force that is excerted on an object by a
medium is equal to the weight mg of the fluid that is replaced by that object,
F = mgez . (6.18)
Here m is the total mass of the fluid that is replaced by the object.

6.5 The Directional Derivative

The gradient is the key to the definition of another important concept, that of the directional derivative. Let
us start from equation (6.3) and write the vector dx as the product of its length ds = ||dx|| and a normalized
vector u, such that dx = u ds, then
df = ∇f · u ds.
By dividing by ds we obtain the rate of change of f per unit of displacement in the direction of u. The
result is called the directional derivative of f in the direction of u. There is a special symbol Du f for this
quantity, but you will also find the notation ∂f /∂u:
∂f
Du (f ) = ∇f · u, = ∇f · u.
or (6.19)
∂u
Note that for a normalized field u, the term ∇f · u represents the component of ∇f in the direction of u.
Exercise 6.9 Let T be a temperature field. What is the (SI) unit of the directional derivative ∂T /∂u?
In a particularly important application of the directional derivative, one chooses u = n, a normalized
vector that is locally perpendicular to a surface and points in the outward direction (see equation (5.4)),
∂f
= ∇f · n
∂n
In this particular case the directional derivative of f is called the normal derivative of f . This normal
derivative is equal to the normal component of ∇f at the boundary point.
6.6. LINE INTEGRALS 73

In our definition (6.19) of the directional derivative we have insisted that u is a unit vector, just like
is done in your copy of Adams [8, p. 717]. But the expression ∇f · u is equally well defined for non-
normalized vectors u, and in many, if not most treatments of the subject the restriction that ||u|| = 1
is lifted. Of course this will change the value of the directional derivative by a factor ||u||, and its
interpretation changes accordingly. If you read a text about the subject, please always look how local
folklore has it.

Exercise 6.10 Give expressions for ∂f /∂n for the case that n is equal to er , eθ and eϕ , the normalized basis
vectors of a spherical coordinate system.

Exercise 6.11 The conductive heat flux density q ( W/m2 ) can be modeled by Fourier’s law from 1822 [9, 10,
section VIII, item 98],
q = −λ∇T, (6.20)
where λ ( Wm−1 K−1 ) is the thermal heat conductivity.
1. Please show that the conductive heat losses Q through a surface A can be written as
Z
∂T
Q=− λ dA. (6.21)
∂n
A

(You may want to review what you have learned about flux integrals in chapter 5.)
2. Consider a spherically symmetric temperature distribution T (r) and assume that λ is uniform. Show
that the conductive heat loss through the boundary of a sphere with radius R is given by

∂T
Q(R) = −4πR2 λ . (6.22)
∂r R

3. Assume that T (r) = TR + (T0 − TR )(1 − r2 /R2 ). Describe this temperature profile and specialize Q(R)
for this case.

6.6 Line Integrals

Consider a vector field u(x). The line integral of u along a curve C is defined as
Z
u(x) · dx. (6.23)
C

Practically this integral can be carried out by parameterizing the curve, as discussed in chapter 3. Let x(t),
with t ∈ [ta , tb ] be such a parameterization, then the infinitesimal displacement dx is related to the velocity
x′ (t) of the curve via dx = x′ (t) dt and equation (6.23) can be written in the form

Ztb
u(x(τ )) · x′ (τ )dτ. (6.24)
ta

In the expressions above we have assumed that the curve is smooth, so the velocity x′ (t) is well-defined in
each point. We will also allow piecewise smooth curves, which feature a finite number of locations where the
74 CHAPTER 6. THE GRADIENT OF A SCALAR FIELD, LINE INTEGRALS

curve is non-smooth. In that case the integral is calculated for each smooth section of the curve and the sum
of the results is taken.
In general, the result of the integration will depend on the curve C. However, if u is the gradient of a
scalar field f , one can derive that Z
u(x) · dx = f (xb ) − f (xa ), (6.25)
C
where xa and xb are the begin and end points of the curve, respectively. This is a result that you know well
from your previous studies of mechanical and electrostatic problems. There you have already learned that a
vector field u that can be expressed as the gradient of some scalar field is called a conservative force field.
Two consequences of equation (6.25) are that
• The line integral is path-independent; only the begin and end points matter for the result;
• The line integral vanishes if the path is closed, so the begin and end points coincide.
Exercise 6.12 Assume that u(x) = ∇f (x). Please prove equation (6.25). (Hint: choose a parameterization
x(t) and show that the integrand is equal to df (x(t))/dt.)

6.7 Application: a Particle in a Force Field

In this section we apply the theory of the previous section to a problem that you ought to be familiar
with: the motion of a particle in a force field. Please make sure that you are comfortable with the
ideas discussed here — you have seen them before, possible in a slightly different language. Some of the
concepts will be referred to in later chapters.

Consider a particle with mass m at position x(t) in a force field F (x). Its motion is governed by Newton’s
laws. We recall that:
• If the particle moves, the force does work on the particle; the work done when the particle is displaced
an amount dx is given by
dW = F (x) · dx = F (x) · x′ (t)dt; (6.26)
• The power is defined as the amount of work that is accomplished by the force field per unit of time,
dW
P (t) = = F (x) · x′ (t); (6.27)
dt
• The kinetic energy of the particle is given by
1
ϵ(x′ (t)) = m(x′ (t))2 . (6.28)
2
Exercise 6.13 Show that the rate of change of the kinetic energy of the particle is equal to the power that is
delivered by the force field:
dϵ
= P (t). (6.29)
dt
Assume that the field F (x) is conservative. This means that for some potential energy function U (x) we
have
F = −∇U. (6.30)
(Note that the addition of a minus sign is the usual convention in physics, in mathematics one usually does
without.)
6.8. SUMMARY AND OUTLOOK 75

Exercise 6.14 Recall some elements of exercise (6.12) and show that

dU (x(t))
P (t) = − . (6.31)
dt
Exercise 6.15 Prove the principle of conservation of mechanical energy,
d
(ϵ(x′ (t)) + U (x(t))) = 0 ⇐⇒ ϵ(x′ (t)) + U (x(t)) = constant. (6.32)
dt
(Note that this relies on the force field being conservative.)

6.8 Summary and Outlook

In this chapter we have introduced the gradient of a scalar field and various related concepts, such as the
material, directional and normal derivatives. We have shown that line integrals over conservative fields,
which can be written as the gradient of a field, can be evaluated by taking the difference between the values
of those fields at the begin and end points. A few questions remain at this point. Is there an easy way to find
out if a given vector field can be written as the gradient of a function? And if there is, how do we calculate
that? There is an important third differential operator in vector calculus that will help us to answer these
and other questions, the curl of a vector field. Since that is the topic of the next chapter, we will postpone
our discussion of these questions until a later moment.

Exercise 6.16 Look at the list of objectives at the beginning of this chapter and ensure that you have acquired
the mentioned skills. If not, please study the relevant parts of this chapter again and repeat this exercise.
76 CHAPTER 6. THE GRADIENT OF A SCALAR FIELD, LINE INTEGRALS
Chapter 7

The Curl of a Vector Field

Objectives
After studying this chapter, you will be able to
• Explain the Maxwell-Faraday law for the induced electromotive force;

• Derive an expression for the curl of a vector field by considering line integrals over infinitely small
closed curves;
• Derive the differential form of the curl in terms of curvilinear coordinates and specialize those for the
usual Cartesian, cylindrical and spherical coordinate systems;
• Calculate the curl of given vector fields;

• Explain how the curl of a velocity field is related to the tendency of the flow field to rotate small objects
that are embedded in that flow;
• Explain and apply the Kelvin-Stokes theorem.

77
78 CHAPTER 7. THE CURL OF A VECTOR FIELD

Figure 7.1: The relation between the sense of a closed curve and the orientation of the enclosed surface is
chosen in accordance with the right-hand rule. (From: https: // en. wikipedia. org/ wiki/ Right-hand_ rule
— public domain.)

7.1 Introduction
In the previous chapter we have introduced the gradient and some of its properties. In particular we have
seen that the line integral of a gradient field over a closed curve vanishes. In the present chapter we will take
a look at non-vanishing integrals of the form
I
v · dx.
C

In section 7.2 we start with a small history lesson and speak briefly about the 19th-century development of
electromagnetism. We will present Faraday’s law and use that in section 7.3 to derive the concept of the curl
of a vector field1 . The approach that we will follow is similar to the one that we adopted when we developed
the coordinate-free definition of the divergence (section 5.5): we start from a relation between integrals, see
how that behaves when applied to infinitely small parts of space and use the result to define the curl of a
vector field v. We will then derive a differential expression for the curl in curvilinear coordinates, just like we
did for the divergence in chapter 5. In later sections we will discuss the curl for concrete vector fields, give a
physical interpretation of the curl of a velocity field and conclude with a presentation of the integral theorem
of Kelvin-Stokes, which is the ‘curl-equivalent’ of the divergence theorem due to Gauß and Ostrogradsky.
In this chapter we will be concerned a lot with fluxes through surfaces. For surfaces we will again use
the symbol A, for a closed curve that is the boundary of A we will use the symbol ∂A but also C. It is
important to notice that there is an ambiguity in such integrals, since there are two possible choices for the
orientation of the surface — one can be chosen at will. Note that there is no natural choice for an outward
normal vector if the surface is not closed.
If we integrate a vector field over the loop ∂A we have again two choices, this time for the sense of the
curve, the direction in which it is traversed. Here we adopt the right-hand rule: if you would draw the curve
on a piece of paper, it is traversed in anti-clockwise direction if the surface element is directed towards the
reader. We refer to picture 7.1 for a visualization of this convention.

7.2 The Maxwell-Faraday Law of Induction

On 29 August 1831, Faraday demonstrated the law of electromagnetic induction [11, p. 71]: if we form a
closed loop from a conducting wire, a change of the magnetic flux that passes through that loop generates an
electromotive force that is equal to the negative of the time-derivative of the enclosed flux. This electromotive
force is a quite tricky concept, but essentially it is just the voltage difference that one would measure between
1 Dutch: ‘De rotatie van een vectorveld.’
7.2. THE MAXWELL-FARADAY LAW OF INDUCTION 79

n = nc

A1

A2 C

nc

n = −nc

Figure 7.2: Two surfaces A1 and A2 share the boundary contour C. The normal vectors nc at both surfaces
are directed in accordance with the right-hand rule, based on the sense of C. The two surfaces form a closed
surface, but in the evaluation of a surface integral over this surface we must negate nc for A2 to make it point
outward. The normal vectors for the combined closed surface are written as n.

the two ends of the open loop that can be created by making a small cut in the closed loop. It can be expressed
in terms of the electric field as I
E(t) = E(t, x) · dx. (7.1)
∂A(t)

The magnetic flux through a surface A is equal to the surface integral of the normal component of the
magnetic field B over that surface. Such surface integrals have been studied in chapter 5, the result is
Z
Φm (t) = B(t, x) · dA. (7.2)
A(t)

The expression makes clear that there can be two reasons why the enclosed flux changes: the magnetic field
B(t, x) itself is in general a function of time, but also the size and shape of the surface A(t) can change.
Faraday’s law can now be stated as:
I Z
d
E(t, x) · dx = − B(t, x) · dA. (7.3)
dt
∂A(t) A(t)

For a fixed surface A, this equation can be written as

I Z
∂
E(t, x) · dx = − B(t, x) · dA. (7.4)
∂t
∂A A

Exercise 7.1 Please argue that the electric field E in equation (7.4) cannot be written as the gradient of a
scalar field.

There are, of course, many surfaces A that have C as boundary curve, and in general one would expect
that the flux integral in the right-hand side depends on the particular choice for A. But the Maxwell-Faraday
law states that result is always the same, depending only on the curve C. How can we understand this?
80 CHAPTER 7. THE CURL OF A VECTOR FIELD

n
e
x2 + δx2 /2

w δA3

x2 − δx2 /2 s x1 + δx1 /2
x1 − δx1 /2

Figure 7.3: In order to derive a differential expression for the curl of a vector field, we calculate the line integral
that is indicated in the text. The four segments of the closed curve are labeled s (south), e (east), n (north) and
w (west) and coincide with coordinate lines. The surface element δA3 is directed perpendicular to this curve
and is pointing towards the reader.

The reason is that the magnetic field has the following important property: according to one of the
fundamental laws of electromagnetics its divergence is zero,

∇ · B = 0. (7.5)

In the following exercises we will demonstrate that the result of this fact is that the Maxwell-Faraday law is
indeed independent of the choice for A — as long as it has the boundary curve C.

Exercise 7.2 Prove that the magnetic flux through a closed surface is zero.

Consider two non-crossing surfaces A1 and A2 that share the same boundary curve C, as visualized
in figure 7.2. The orientation of the surfaces is chosen in accordance with the right-hand rule, consistent
with the sense of the curve C. The normal vectors that correspond to this choice are written as nc , the
corresponding magnetic fluxes through the surfaces are written as Φ1 and Φ2 .

Exercise 7.3 On the consistency of the Maxwell-Faraday equation.

Show that Φ1 − Φ2 = Φ = 0. (Hint: you may want to rewrite dA in terms of a suitable normal vector and
dA.)

The result of the exercise is that the fluxes are the same, Φ1 = Φ2 . This proves that Faraday’s law gives the
same result for E, independent of the choice for A, for a given boundary curve.

7.3 Derivation of the curl

Our next step is to develop a differential form of the Maxwell-Faraday equation (7.4). In preparation we ask
you to remember our derivation of the divergence operator, which was presented in section 5.5. There we
considered the stationary species mass balance and investigated its behavior when the test volume V was
shrunk to an infinitely small size around a point x. The concept of the divergence of a vector field emerged
naturally.
In the present section we will proceed in a similar way: we will consider what happens with the integral
(7.4) when we consider infinitely small loops; this time we will investigate a limit in which the area of a
planar surface is reduced to zero. There is a complication, however: when we investigate a surface element
we must still make a choice for its orientation. And once we have done that, the surface integral will tell us
something about the component of the vector field −∂B/∂t in the direction perpendicular to the surface.
7.3. DERIVATION OF THE CURL 81

Let us now make this a bit more concrete. Firstly, for notational convenience, let us introduce the vector
S = −∂B/∂t. Then the Maxwell-Faraday equation takes the form
I Z
E · dx = S · dA. (7.6)
∂A A

We write the vector S in terms of the covariant set of basis vectors of a curvilinear coordinate system
(x1 , x2 , x3 ), S = S i ai , then I Z
E · dx = S i ai · dA. (7.7)
∂A A

We must now make a choice for the surface. First we choose for A a ‘small’ surface element in the direction
perpendicular to the x3 coordinate plane. Such surface element is given by
√
δA3 = ga3 δx1 δx2 . (7.8)

This surface area and the resulting boundary curve have been visualized in figure 7.3, please have a look at
that figure first. Because of this choice, the integral on the right-hand side can be simplified considerably.
We find that
√ √
I
E · dx ≈ S i ai · ga3 δx1 δx2 = S 3 gδx1 δx2 . (7.9)
∂A

Let us now look at the left-hand side of the equation. This time we express E in covariant coordinates,
E = Ei ai . The reason is that this facilitates the calculation of the line integral over the rectangular curve
that is defined by the surface element. We will calculate that in four parts, starting at the south side s and
traveling anti-clockwise. The contribution from the southern segment is given by

(E · dx)s ≈ Ei (x1 , x2 , x3 )ai · a1 δx1 = E1 (x1 , x2 , x3 )δx1 ,

The northern segment yields a contribution (note that now dx1 = −δx1 ):

(E · dx)n ≈ Ei (x1 , x2 + δx2 , x3 )ai · a1 dx1 = −E1 (x1 , x2 + δx2 , x3 )δx1 .

The sum of both contributions is given by

∂E1 1 2
(E · dx)s + (E · dx)n ≈ (E1 (x1 , x2 , x3 ) − E1 (x1 , x2 + δx2 , x3 ))δx1 ≈ − δx δx .
∂x2
The contributions of the western and eastern parts of the curve can be handled similarly, the over-all result
is
I  
∂E2 ∂E1
E · dx ≈ − δx1 δx2 (7.10)
∂x1 ∂x2
∂A

Combining this result with equation (7.9) and considering the limit that δx1 and δx2 tend to zero yields
 
3 1 ∂E2 ∂E1
S =√ − . (7.11)
g ∂x1 ∂x2

In order to obtain expressions for S 1 and S 2 we can repeat this exercise for small planar surface elements
in these directions. The result can also be retrieved by applying a cyclic permutation to the indices in the
result above, where we substitute 1 → 2, 2 → 3 and 3 → 1.
82 CHAPTER 7. THE CURL OF A VECTOR FIELD

The result for the vector S depends only on the field E. We call this field the curl of the field E and
write it as ∇ × E,
     
1 ∂E3 ∂E2 1 ∂E1 ∂E3 1 ∂E2 ∂E1
∇×E = √ − a1 + √ − a2 + √ − a3 . (7.12)
g ∂x2 ∂x3 g ∂x3 ∂x1 g ∂x1 ∂x2

In the preceeding text we have presented the derivation of the expression for the curl in general curvilinear
coordinates. In the derivation we have seen, again, some typical operations involving curvilinear coordinates
and basis vectors that you ought to feel comfortable with. Apart from that, the derivation really involves
not much more than the evaluation of integrals over a simple surface element and its boundary curve. You
should definitely be able to do all that. For that reason we recommend that you try to reproduce the steps
above. If you find this a bit spicy, it will be mostly because you will need to work systematically and do
proper bookkeeping of coordinates and lengths. That is one more reason to do this exercise, since we also
expect these virtues from you when you do a written exam.

Exercise 7.4 The definition of the curl. ( )

Starting from (7.6), please derive expression (7.11) for the third (contravariant) component of the vector
∇ × E. Try to complete this exercise without consulting the text.

There is an interesting compact expression for the curl that makes use of the permutation symbol eijk , see
equation (4.27).

Exercise 7.5 Show that equation (7.12) can be rewritten as

1 ∂Ek
∇ × E = √ eijk j ai . (7.13)
g ∂x

In our derivation we have used the vector field S as a short name for the negative of the partial time derivative
of the magnetic field. In the course of our development of the definition of the curl we have called this field
∇ × E. From the definition of the curl it then follows that
∂
∇ × E(t, x) = − B(t, x). (7.14)
∂t
This is the differential form of the Maxwell-Faraday equation.

Exercise 7.6 Please verify that we have indeed derived equation (7.14) as a by-product of our development of
the definition of the curl.

Exercise 7.7 Please substitute the differential form of the Maxwell-Faraday equation back into the integral
form. The result is a relation between a surface integral of the curl of the electric field over a surface and the
line integral of the field over the boundary curve of that surface. Please state this relation. The result is an
expression of the theorem of Kelvin-Stokes, which turns out to be more generally valid. It will be discussed
in section 7.7.

7.4 Results for Particular Coordinate Systems

Equation (7.12) defines the curl of a vector field for a general curvilinear coordinate system. Let us now
specialize this expression for some particular systems. We will use these results in the next sections, where
we will give a physical interpretation of the curl and present some of its properties and applications.
Let us first look at the expression for the curl in ‘physical components’, defined relative to normalized
orthogonal basis vectors. You may want to consult section 4.9 for details. We start by observing that this
7.5. THE CURL OF SOME CONCRETE VECTOR FIELDS 83

expression expects the ‘incoming’ vector (the argument) to be expressed in terms of covariant components,
while the result (the curl) is expressed in contravariant components with respect to a covariant basis. This is
no big deal: just express all these quantities in their physical counterparts using equations (4.60) and (4.61)
√
and use the relation g = h1 h2 h3 , which was the subject of exercise 4.19. The result is
 
1 ∂h3 E(3) ∂h2 E(2)
∇×E = − e(1)
h2 h3 ∂x2 ∂x3
 
1 ∂h1 E(1) ∂h3 E(3)
+ − e(2)
h1 h3 ∂x3 ∂x1
 
1 ∂h2 E(2) ∂h1 E(1)
+ − e(3) . (7.15)
h1 h2 ∂x1 ∂x2
Exercise 7.8 We will now present expressions for the usual three coordinate systems. All you need are the
scale factors for those systems, and if you (still) do not know these by heart we refer to section 4.9.
1. Please verify the derivation of (7.15) from (7.12);
2. Show that in Cartesian coordinates:
     
∂vz ∂vy ∂vx ∂vz ∂vy ∂vx
∇ × v(x(x, y, z)) = − e(x) + − e(y) + − e(z) . (7.16)
∂y ∂z ∂z ∂x ∂x ∂y

3. Show that in cylindrical coordinates:

     
1 ∂vz ∂vϕ ∂vr ∂vz 1 ∂rvϕ ∂vr
∇ × v(x(r, ϕ, z)) = − e(r) + − e(ϕ) + − e(z) . (7.17)
r ∂ϕ ∂z ∂z ∂r r ∂r ∂ϕ

4. Show that in spherical coordinates:

 
1 ∂vϕ sin θ ∂vθ
∇ × v(x(r, θ, ϕ)) = − e(r)
r sin θ ∂θ ∂ϕ
 
1 1 ∂vr ∂rvϕ
+ − e(θ)
r sin θ ∂ϕ ∂r
 
1 ∂rvθ ∂vr
+ − e(ϕ) . (7.18)
r ∂r ∂θ

7.5 The Curl of Some Concrete Vector Fields

Exercise 7.9 In fluid dynamics, the curl of the velocity field is called the vorticity of the flow. Please calculate
the vorticity of the following (velocity) vector fields. The arguments indicate the nature of the coordinate
system.
1. a shear flow with v(x(x, y, z)) = (z/Z)V ex ;
2. a Hagen-Poiseuille flow in a cylinder with radius R; the axial speed is given by v(x(r, ϕ, z)) = V (1 −
r2 /R2 )ez ;
3. a rotating fluid with v(x(r, ϕ, z)) = rωeϕ , where ω is constant;
4. a rotating fluid with v(x(r, ϕ, z)) = V eϕ , where V is constant and r > 0;
5. a rotating fluid with v(x(r, ϕ, z)) = (A/r)eϕ , where A is constant and r > 0.
84 CHAPTER 7. THE CURL OF A VECTOR FIELD

Figure 7.4: A non-uniform flow field. The arrows indicate the direction of the velocity field, their size the speed
of the flow for each height. A rod that is initially placed vertically in the transversal direction will tend to rotate
as it moves with the flow field. The position and orientation of the rod are shown at different moments in time.
The magnitude and direction of the angular velocity are closely related to the curl of the velocity field (see text).

Exercise 7.10 Prove from the definitions of the divergence and curl in curvilinear coordinates that for any
vector field v the divergence of its curl is zero,
∇ · (∇ × v) = 0. (7.19)
Exercise 7.11 In equation (7.5) we have stated the important fact that ∇ · B = 0. Show that this ‘almost’
follows from the Maxwell-Faraday equation (7.14); that it can be interpreted as an ‘initial condition’.
Exercise 7.12 Prove from the definitions of the gradient and curl in curvilinear coordinates that for any scalar
field f the curl of the gradient of the scalar field is the zero vector,
∇ × ∇f = 0. (7.20)

7.6 Physical Interpretation of the Curl

Let us look back at some of the exercises above. As an example, one question called for the curl of a flow
field of a shear flow with v(x(x, y, z)) = (z/Z)V ex , and hopefully you got the answer
∇ × v(x(x, y, z) = (V /Z)ey .
Imagine that you place a small rod L in this flow, and that it is initially placed vertically at position x0 . Let
us assume that the rod is so light that it does not influence the flow field whatsoever. What will the position
of the rod be after a small time dt has passed? This situation is sketched in figure 7.4. The difference in flow
velocity at the top and at the bottom of the rod is given by ((z + L)V /Z − zV /Z)ex = LV /Zex , so the top
of the rod will be moved to the right faster than the bottom is. The difference in horizontal displacement
is then given by (LV /Z)dt and this will result in a rotation of the rod that is equal to dϕ = (V /Z)dt. The
initial angular speed of the rod is then given by dϕ/dt = V /Z. This is equal to the magnitude of the vorticity.
The direction of the rotation is consistent with applying the right-hand rule (figure 7.1): when your thumb
is pointing in the direction of the vorticity vector, your fingers will point in the direction of the rotation.
This interpretation of the curl of a velocity field is more generally valid: the vorticity is a measure for the
tendency of the flow to rotate a ‘small object’. The induced angular velocity is of the order of the magnitude
of the vorticity, though generally not equal to it.
Exercise 7.13 Consider the flow field v(x(r, ϕ, z)) = rωeϕ , where ω is constant. In exercise 7.9 you have
already established that the vorticity (the curl) of this flow field is given by 2ωez . Please consider a volume
that is defined by the coordinate values r ∈ [R, R + δR] and ϕ = [0, δϕ] for some value of R and small values
of δR and δϕ. Sketch this situation in an r, ϕ plane for an arbitrary value of z. Draw the positions of the
corner points of the object after a time t = π/(2ω) has passed. Has the object been deformed? Has it rotated?
If so, how much? What is the rotational speed?
7.7. THE KELVIN-STOKES THEOREM 85

7.7 The Kelvin-Stokes Theorem

In exercise 7.7 we found a remarkable relation between the integral of the electric field along a closed curve
and the integral of the normal component of the curl over the surface that has that curve as boundary. By
no means that constituted a general proof of such relation, but it turns out to be generally valid. If a vector
field v is smooth and the curve is piecewise smooth, the following relation is valid:
I Z
v · dx = (∇ × v) · dA. (7.21)
∂A A

This is the Kelvin-Stokes theorem. It is related to the curl like the theorem of Gauß-Ostrogradsky is related
to the divergence. We will see applications of this theorem later on in this text.

7.8 Summary and Outlook

In this chapter we have introduced the third major differential operator in vector calculus, the curl of a
vector field. Starting from a physical example, the electromagnetic law of Maxwell-Faraday, we have shown
that the curl arises naturally when one considers line integrals over infinitely small loops. We have developed
the expression for the curl in curvilinear coordinates, then specialized this expression for the usual Cartesian,
cylindrical and spherical coordinate systems in terms of ‘physical components’ and normalized basis vectors.
Examples of the curl that are relevant to fluid dynamics have been presented.
The divergence, gradient and curl are among the key tools in vector calculus, and many important
relations involving these concepts exist. As an example, we have already seen that ∇ · (∇ × v) = 0 and that
∇ × ∇f = 0 for any vector field v and scalar field f . In the next chapter we will spell out many other rules
that apply when working with these operators. We will then discuss how vectors can be decomposed into
two parts that represent the gradient and the curl of a scalar and vector potential, respectively.

Exercise 7.14 Look at the list of objectives at the beginning of this chapter and ensure that you have acquired
the mentioned skills. If not, please study the relevant parts of this chapter again and repeat this exercise.
86 CHAPTER 7. THE CURL OF A VECTOR FIELD
Chapter 8

Combinations of Vector Operators, The

Laplacian

Objectives
After studying this chapter, you will be able to
• Apply the product rules to expressions involving the divergence, gradient and curl of fields;
• Explain the definition of the Laplacian of a scalar field;

• Derive expressions for the Laplacian of scalar fields in the usual (Cartesian, cylindrical and spherical)
coordinate systems from the expression in general curvilinear coordinates;
• Reproduce by heart the expression for the Laplacian in Cartesian coordinates;
• Reproduce by heart the fact that the curl of the gradient of a scalar field vanishes;

• Reproduce by heart the fact that the divergence of the curl of a vector field vanishes;
• Explain the definition of the Laplacian of a vector field and the care that must be taken in its evaluation
in non-Cartesian coordinate systems.

87
88 CHAPTER 8. COMBINATIONS OF VECTOR OPERATORS, THE LAPLACIAN

8.1 Introduction
In the preceeding chapters we have defined three basic elements of vector calculus: the gradient of a scalar
field and the divergence and curl of a vector field. We have chosen to introduce these concepts by starting
from physical problems (such as the mass balance and the law of Maxwell-Faraday). While this approach
should be appealing for a student of physics, it has the disadvantage that the information is served piecemeal,
as it presents itself in the development of the subject matter, and as a consequence the big picture may get
lost on the reader. We will fix that by starting this chapter with a summary of our findings so far and an
exposition of properties of the divergence, gradient and curl that have not been exposed yet.
We will then study a particularly relevant combination of these elements, the Laplacian, which is defined
as the divergence of the gradient of its argument. We will soon see that it has many interesting mathematical
and physical applications.

8.2 Properties of the Gradient, Divergence and Curl

Let α and β be constant values, f (x) and g(x) scalar fields and a(x) and b(x) vector fields. If the fields are
smooth, a great number of identities involving the fields and the nabla-operator apply. We list those that are
relevant for our discussion, more can be found in any textbook on the subject or a website like Wikipedia 1 .

Linearity
The gradient, divergence and curl are linear functions of their arguments; a linear combination of arguments
results in the linear combination of results. In equation form this can be stated as

∇(αf + βg) = α∇f + β∇g; (8.1)

∇ · (αa + βb) = α∇ · a + β∇ · b; (8.2)
∇ × (αa + βb) = α∇ × a + β∇ × b. (8.3)

Multiplication Rules
For the derivative of the product of two functions the Leibniz product rule tells us that (f g)′ = f g ′ + f ′ g.
Since the gradient, divergence and curl are related to first-order spatial derivatives of their arguments we
expect that they are also governed by multiplication rules of some sorts. Indeed, such rules exist and below
we list the ones that are relevant for our present studies:

∇(f g) = f (∇g) + (∇f )g; (8.4)

∇ · (f a) = f (∇ · a) + (∇f ) · a; (8.5)
∇ · (a × b) = (∇ × a) · b − a · (∇ × b); (8.6)
∇ × (f a) = f (∇ × a) + (∇f ) × a. (8.7)

Second-order Derivatives
In exercises 7.10 and 7.12 you have already proven that the divergence of a curl is the number 0, and that
the curl of a gradient is the zero-valued vector:

∇ · (∇ × a) = 0; (8.8)
∇ × (∇f ) = 0. (8.9)
1 See for example [Link]
8.3. THE LAPLACIAN OF A SCALAR FIELD 89

These are particularly important properties that we will refer to extensively in the remainder of this course.
You are expected to know these two properties by heart.

Exercise 8.1 Please prove that

∇ · (f ∇ × v) = (∇f ) · (∇ × v).

8.3 The Laplacian of a Scalar Field

The Laplacian of a scalar field is defined as the divergence of the gradient of its argument and is usually
written as ∇2 or as ∆:
∇2 f ≡ ∆f ≡ ∇ · ∇f. (8.10)
Obviously we can calculate the Laplacian of a scalar field by first calculating its gradient, then the divergence
of the result, but the Laplacian is sufficiently important that in practice these operations are combined.
Let us now derive an expression for the Laplacian in terms of the coordinates of a given coordinate
system. In the case of general curvilinear coordinates this means that equation (6.4) for the gradient must
be combined with equation (5.19) for the divergence. The only complication is that the expression for the
divergence requires the contravariant components of its vector argument, whereas the expression for the
gradient produces covariant components, so a conversion is due, as explained in section 4.8. The derivation
is left as an exercise.

Exercise 8.2 The Laplacian in curvilinear coordinates.

1. Please show that the gradient can be expressed in terms of contravariant coordinates as

∂f ji
∇f = g ai . (8.11)
∂xj

2. Use the result to find the expression for the Laplacian in terms of curvilinear coordinates,
 
2 1 ∂ √ ji ∂f
∇ f=√ gg . (8.12)
g ∂xi ∂xj

Note that these expressions involve two sums, over the dummy indices i and j.

Exercise 8.3 Show that for an orthogonal curvilinear coordinate system equation (8.12) reduces to
      
2 1 ∂ h2 h3 ∂f ∂ h1 h3 ∂f ∂ h2 h1 ∂f
∇ f= + + . (8.13)
h1 h2 h3 ∂x1 h1 ∂x1 ∂x2 h2 ∂x2 ∂x3 h3 ∂x3

The expressions for the usual coordinate systems are given below. As before, we use the arguments to
indicate whether a Cartesian, cylindrical or spherical coordinate system is assumed.

• Cartesian coordinates:
∂2f ∂2f ∂2f
∇2 f (x(x, y, z)) = + + ; (8.14)
∂x2 ∂y 2 ∂z 2

• Cylindrical coordinates:

1 ∂2f ∂2f
 
1∂ ∂f
∇2 f (x(r, ϕ, z)) = r + + ; (8.15)
r ∂r ∂r r2 ∂ϕ2 ∂z 2
90 CHAPTER 8. COMBINATIONS OF VECTOR OPERATORS, THE LAPLACIAN

• Spherical coordinates:

∂2f
   
1 ∂ ∂f 1 ∂ ∂f 1
∇2 f (x(r, θ, ϕ)) = r2 + sin θ + 2 . (8.16)
r2 ∂r ∂r r2 sin θ ∂θ ∂θ r2 sin θ ∂ϕ2

Exercise 8.4 Please derive the expressions (8.14)–(8.16) from equation (8.13).

Exercise 8.5 Consider the field f (x) = ||x||2 . Calculate the Laplacian using Cartesian, cylindrical and
spherical coordinate systems and show that the result is ∇2 f = 6 in either case.

In the previous exercise you may have neglected the singularity at r = 0 when the calculation was done in
cylindrical or spherical coordinates. Let us, in the form of an exercise, take a look at the limit that r ↓ 0 for
a spherically symmetric field f and use l’Hôpital’s rule to find an expression for this limit if the field is ‘flat’
at r = 0.
Exercise 8.6 Consider a spherically symmetric field f (x) = g(||x||). Using spherical coordinates we can write
r = ||x||.
1. Please show that in this case
2
∇2 f = g ′′ (r) + g ′ (r); (8.17)
r
2. Show that, if dg/dr = 0 at r = 0 and the second derivative of g is finite at this point, we have

lim(∇2 f ) = 3g ′′ (0) (8.18)

r↓0

and verify that this result is consistent with the result from the previous exercise.
Another important case is a field of the form f (x(r, θ, ϕ)) = αr−1 , where α is a constant. This models, for
example, the electrostatic potential due to a point charge in the origin, and will be discussed extensively
later on in this chapter. It is spherically symmetric, but obviously not flat in the origin (the function is not
even defined in that point).
Exercise 8.7 Show that for this field ∇2 f = 0 for r ̸= 0.
Let us briefly dwell on another important equation that involves the Laplacian. A popular model for the
diffusive transport of a substance in its environment is Fick’s law. For a diluted component s with density
ns of a multi-component gas mixture it states that the diffusive particle flux density Γs is related to the
non-uniformity of its density by the relation

Γs = −Ds ∇ns , (8.19)

where Ds is the diffusion coefficient.

Exercise 8.8 Assume that the volumetric production rate of species s is given by Ss , and that the species’
transport is governed by Fick’s law. Show that under these circumstances the density ns is governed by the
equation
∂ns
− ∇ · Ds ∇ns = Ss . (8.20)
∂t
Please simplify this equation for the special case that Ds is uniform and show that the steady-state solutions
of the resulting equation are given by
∇2 ns = −Ss /Ds . (8.21)
8.4. THE LAPLACIAN OF A VECTOR FIELD 91

In the last part of the previous exercise we have produced an equation of the form

∇2 f = ω.

This equation is called the Poisson equation. In the special case that the source term ω is zero, it reduces
to the Laplace equation
∇2 f = 0.
These equations and their solutions will be studied in more detail in the following chapters.

8.4 The Laplacian of a Vector Field

In the previous section we have introduced the Laplacian of a scalar field. The Laplacian is also defined
for vector fields, and both notation and definition are the same; it is the divergence of the gradient of its
argument,
∇2 a = ∇ · ∇a. (8.22)
But here we seem to run into troubles: we have not yet discussed the concept of the gradient of a vector
field, which is a so-called 2-tensor (see chapter 11). We are therefore lucky that the vector-Laplacian can be
expressed in terms of vector operations only: without proof we state that

∇2 a = ∇(∇ · a) − ∇ × ∇ × a. (8.23)

If one wants to avoid the usage of tensors, this relation can be adopted as an alternative definition of the
vector Laplacian.

In a Cartesian coordinate system a = ax ex + ay ey + az ez we have

∇2 (ax ex + ay ey + az ez ) = (∇2 ax )ex + (∇2 ay )ey + (∇2 az )ez , (8.24)

in other words: the components of the Laplacian are equal to the Laplacians of the components. The
reason that this works is that the Cartesian basis vectors are uniform (‘constant’).
In any other coordinate system the derivatives of the basis vectors with respect to
the coordinates


must be taken into account and we cannot use the results above: in general ∇2 v (i) e(i) ̸= ∇2 v (i) e(i) .

Exercise 8.9 Please verify identity (8.24) using the alternative ‘double-curl’ definition of the vector Laplacian
(it suffices to do this for one component).

8.5 Summary and Outlook

With the exposition of the properties of the gradient, divergence and curl and the introduction of the
Laplacian we have finished building our differential vector calculus toolkit. In this and the preceeding chapters
we have already seen many applications of vector calculus (transport, electrostatics, magnetodynamics), but
you may have noticed that we have not discussed any general procedure for solving the relevent equations.
This will be the subject of the two remaining chapters of this syllabus. We will start with an analysis of the
Poisson equation and present a general solution, then generalize our discussion to arrive at the Helmholtz
decomposition of a vector field and its formal solution.
92 CHAPTER 8. COMBINATIONS OF VECTOR OPERATORS, THE LAPLACIAN

Exercise 8.10 Look at the list of objectives at the beginning of this chapter and ensure that you have acquired
the mentioned skills. If not, please study the relevant parts of this chapter again and repeat this exercise.
Chapter 9

Electrostatics, the Poisson Equation

Objectives
After studying this chapter, you will be able to
• State and derive the charge continuity equation;

• State the properties of the three-dimensional delta function and evaluate volume integrals involving
this function;
• Derive the potential for an arbitrary charge distribution;
• Provide an informal derivation of Gauß’ law for the electrostatic field;

• Motivate and explain Maxwell’s correction of the equation of Ampère;

93
94 CHAPTER 9. ELECTROSTATICS, THE POISSON EQUATION

9.1 Introduction
In the previous chapters we have developed a set of tools for describing vectors in integral and differential
form. In this chapter and the next we will complete our discussions by looking at techniques for solving some
of these equations. In this chapter, in particular, we will rewrite the Coulomb equation for the electrostatic
field in a differential form that is known as Gauß’ equation, and develop a solution for this field for an
arbitrary charge distribution. In the course of this discussion we will present the three-dimensional delta
function, the three-dimensional analog of the delta function that you have encountered in your studies of
signals and systems. We will demonstrate that this delta function can be used as a model for a point source.
In the last part of this chapter we will demonstrate that solutions of field equations for arbitrary source
distributions can easily be obtained once the solution for the point source has been established. But in order
to facilitate the discussion that follows we will first need to look into the concept of charge and current
densities.

9.2 Charge and Current Densities

We consider a charge density ρc (x, t) such that the charge that is contained in a volume dV around x is
given by dQ = ρc (x, t)dV . Then the total charge Q in a volume V is given by the integral
Z
Q(t) = ρc (x, t) dV. (9.1)
V

Note that this charge is, in general, a function of time, since charged particles may enter or leave the volume
V through its boundary surface ∂V . It is therefore interesting to consider a current density field J , which
is defined such that the total charge flux out of the volume is given by the surface integral over the outward
normal component of J . Such flux densities were introduced in chapter 5. Since (net) charge cannot be
produced or destroyed,1 the rate of change of the charge Q(t) must be balanced by the inward transport of
charges. From this integral relation one can easily derive the relation

∂ρc
+ ∇ · J = 0. (9.2)
∂t
This is the differential form of the charge continuity equation. We will soon see that this equation has played
a vital role in the development of the Maxwell equations that govern electromagnetic fields.

Exercise 9.1 Please derive equation (9.2).

9.3 Point Sources, the Three-dimensional Delta Function

In the previous section we have introduced the charge density field ρc (x, t). We will now answer the question
how we can model a point charge qi at position xi . If this is the only charge that is present in a given domain,
the charge density is zero, except in the point xi itself. If we consider a volume that contains the point xi ,
we expect that the integral of the charge density is equal to the charge qi , even if the volume is arbitrarily
small. How do we describe such situation mathematically?
From your study of Signals and Systems you will remember that the delta function can be used in
such situations. Back then it was defined as a function of one or two parameters; we will now define a
1 Note that this is not a mathematical truth: it is a model of physical reality, based on experience and backed up by

experiments.
9.4. COULOMB’S LAW REVISITED 95

three-dimensional version in an analogous way by its properties that:

δ(x − x0 ) = 0 if x ̸= x0 ; (9.3)
Z
f (x)δ(x − x0 ) dV = f (x0 ), (9.4)

where the latter integration is over any volume that contains the point x0 .
Using the delta function, we can write the charge distribution of a point charge at position x0 as

ρc (x) = qδ(x − x0 ). (9.5)

We leave it as an exercise to show that this definition has the desired properties.
Exercise 9.2 Please verify that this definition of the charge density of a point charge q yields the correct total
charge contained in volumes that contain the charge, and volumes that do not.

9.4 Coulomb’s Law Revisited

In exercise 5.13 we have considered the electric field due to a point charge q in the origin; for r ̸= 0 that is
given by
1 q 1 x
E(x(r, θ, ϕ)) = er = q . (9.6)
4πϵ0 r2 4πϵ0 ||x||3
It is well-known that the electric field due to a point charge can be described with a potential2 such that
E = −∇Φ. It is given by
1 q
Φ(x) = , (9.7)
4πϵ0 ||x||
or, in terms of the spherical coordinates:
1 q
Φ(x(r, θ, ϕ)) = . (9.8)
4πϵ0 r
Exercise 9.3 Please verify that the potential Φ that is given by equation (9.8) indeed results in the electric
field (9.6).
Let us next consider a continuous charge distribution ρc (x). The charge in a small volume dV ′ at position
x′ is given by dq = ρc (x′ )dV ′ and the contribution to the potential is given by

1 ρc (x′ )
dΦ = dV ′ , (9.9)
4πϵ0 ||x − x′ ||
The potential is obtained by considering all charges, which amounts to integrating over the entire volume,
ρc (x′ )
Z
1
Φ(x) = dV ′ . (9.10)
4πϵ0 ||x − x′ ||
In principle we can use this equation for calculating the potential for a given charge distribution. But that
is not as easy as it seems: in practice one is dealing with problems in which the charge density is specified in
some region of space, but the solution must be subjected to boundary conditions. It is then more convenient
to formulate the problem for the electrostatic field as a differential equation, which we will study in the next
section.
2 Here we use the symbol Φ for the potential, rather than V , to avoid a conflict with volumes, which are also written with

the letter V .
96 CHAPTER 9. ELECTROSTATICS, THE POISSON EQUATION

9.5 Gauß’ Law

We will now develop a differential form of the equations for the electric field and potential. We warn you
in advance that the presentation below lacks mathematical rigor, mostly because of the vagaries that are
associated with the usage of the delta function. If you are interested in a more strict presentation of the
subject matter, we advise you to study the subject of distributions and their applications in mathematical
physics. As was mentioned in the Signals & Systems part of this course, distributions are a modern alternative
to ‘generalized functions’ such as the delta function.
In exercise (5.13) it has been established that the divergence of the electric field due to a point source
vanishes in any point that does not coincide with the point source, so

∇·E =0 for r ̸= 0.

The question remains how to handle the singularity at r = 0. We know that there is a point charge present
at that location, but what can we say about the divergence of the electric field in that point? We propose
that
∇ · E = αδ(x). (9.11)

We will now justify this expression and obtain a value for the constant α.
The first observation is that this expression is consistent with the fact that the divergence is zero for
r ̸= 0: indeed the delta function vanishes at those points. In order to find the value of α we consider a sphere
S with radius R that is centered around the point charge at r = 0 and calculate the flux through the surface
of this sphere using equation (9.6). This integration is straightforward and yields
Z
q
E · dA = . (9.12)
ϵ0
∂S

Exercise 9.4 Please verify this result.

On the other hand, if we integrate both sides of equation (9.11) over the sphere and apply the divergence
theorem to the left-hand side, we obtain
Z Z
E · dA = αδ(x)dV = α. (9.13)
∂S S

Comparing results yields α = q/ϵ0 , so for the divergence of the electric field we find the expression

1 ρc (x)
∇·E = qδ(x) = , (9.14)
ϵ0 ϵ0

where we have used expression (9.5) for the charge density due to a point charge. This is Gauß’ law of
electrostatics. It can be shown that this expression is also valid for any other electric charge density field.
In terms of the electrostatic potential it can be stated as

ρc (x)
∇2 Φ(x) = − . (9.15)
ϵ0

This is the differential form of the equation for the electrostatic field. Remember that its solution is given
by equation (9.10).
9.6. THE DISPLACEMENT CURRENT 97

9.6 The Displacement Current

Gauß’ law for the electrostatic field is one of the four Maxwell equations, the corner stones of electromagnet-
ism. We have already seen two more of these equations in previous chapters: the Maxwell-Faraday law and
the statement that the magnetic flux density B is divergence-free. In this section we discuss the observation
that has resulted in Maxwell’s formulation of the fourth equation. Or even better: we let you re-discover
Maxwell’s findings.
In 1826, Ampère proposed that the magnetic flux density is governed by the relation

µ−1
0 ∇ × B = J, (9.16)

where µ0 is the permeability (of vacuum), and J the current density, which was introduced in equation (9.2).
In 1861, Maxwell realized that this equation is inconsistent in the presence of a varying charge density
field, and proposed a correction. Let us look at this story in the form of a few small exercises.
Exercise 9.5 The Maxwell-Ampère law.
1. Show that Ampère’s law is inconsistent with equation (9.2);
2. Show that this inconsistency can be fixed by adding an additional term to the equation,
∂E
µ−1
0 ∇ × B = J + ϵ0 .
∂t

We see that in addition to the ‘material current’ due to the moving charges there is an additional term in the
‘Maxwell-Ampère’ equation, which is related to a time-varying field. It is called the displacement current.

9.7 Summary and Outlook

In this chapter we have looked at charge distributions and established the current continuity equation. This
equation turned out to have played a vital role in the development of Maxwell’s electrodynamic theory. In
order to model the charge distribution of a point charge, a three-dimensional variant of the delta function
was introduced. We then used this delta function to derive integral and differential forms of Coulomb’s law,
generalized to arbitrary charge distributions.
One problem remains. We have mentioned on various occasions that not every field has a scalar potential.
In the following —and last— chapter of this course we will present the concept of a vector potential, state
that every vector field can be written in terms of the combination of a scalar and a vector potential, and
show how these can be calculated.

Exercise 9.6 Look at the list of objectives at the beginning of this chapter and ensure that you have acquired
the mentioned skills. If not, please study the relevant parts of this chapter again and repeat this exercise.
98 CHAPTER 9. ELECTROSTATICS, THE POISSON EQUATION
Chapter 10

Scalar and Vector Potentials, Helmholtz’

Decomposition Theorem

Objectives
After studying this chapter, you will be able to
• State the definitions of solenoidal and irrotational vector fields;
• Explain the concept of a vector potential;

• Explain the conditions for the existence of a scalar and a vector potential;
• State Helmholtz’ Decomposition Theorem...
• ... and explain why it is so important in mathematical physics.

99
100 CHAPTER 10. HELMHOLTZ’ DECOMPOSITION THEOREM

10.1 Introduction
We are almost ready. We have seen scalar and vector fields, numerous operations on such fields, such as the
gradient, divergence, curl and the Laplacian, and we have provided various applications on the way. Yet our
discussion is incomplete: we have paid relatively little attention to methods for solving equations involving
vector fields. We will address that issue in the present chapter. We will start by defining the vector potential,
a concept that co-exists with that of a scalar potential, then discuss under what conditions a scalar and a
vector potential exist. The discussion continues with Helmholtz’ decomposition theorem, which allows the
calculation of vector fields which have a specified divergence and curl and are further defined by a particular
set of boundary conditions. This is a very interesting subject, since many problems in physics have exactly
this structure.

10.2 Solenoidal and Irrotational Fields, Potentials

We have seen that some vector fields v can be written as the gradient of a scalar field. This scalar field is
then called the scalar potential of the vector field (possibly with a minus sign added). Some fields allow the
introduction of a vector potential, whose curl produces the field v.
It can be shown that a vector field has a scalar potential if its curl is zero. Such a field is called curl-free
or irrotational. Secondly, a field has a vector potential when its divergence is zero. Such fields are called
divergence-free, or solenoidal. These definitions and statements have been summarized in two-column format
below, to highlight the ‘symmetry’ of these facts:

A field v is irrotational, or curl-free, if A field v is solenoidal, or divergence-free, if

∇ × v = 0. (10.1) ∇ · v = 0; (10.3)

If v is the gradient of some scalar field Φ it is ir- If v is the curl of some vector field A, it is solen-
rotational, since for all Φ oidal, since for all A

∇ × (∇Φ) = 0. ∇ · (∇ × A) = 0.

An irrotational vector field v in R3 has a scalar A solenoidal vector field v in R3 has a vector po-
potential Φ such that tential A such that

v = −∇Φ. (10.2) v = ∇ × A. (10.4)

10.3 Helmholtz Decomposition Theorem

We have seen that solenoidal fields allow a vector potential, and that irrotational fields allow a scalar potential.
As it turns out, every vector field can be written in terms of the two potentials as

v = −∇Φ + ∇ × A. (10.5)

This is the Helmholtz Decomposition Theorem.

Exercise 10.1 Show that the vector potential A of a field is not unique. (Think of a vector expression that
can be added to A without altering v.)
10.4. SUMMARY 101

Exercise 10.2 Consider two scalar fields f and g. Show that the vector field v = ∇f × ∇g is solenoidal and
that its vector potential is given by A = f ∇g.1

Exercise 10.3 Consider the vector field v(x) = a × x, where a is a constant vector. Please show that this
field is solenoidal and determine its vector potential.2 Hint: this is of the form A = f (x)x. You may use
the fact that the curl of the position vector vanishes,

∇ × x = 0. (10.6)

The question remains how we can calculate the potentials. It can be proven with a little bit of effort that
the potentials are given by

∇′ · v(x′ ) ′ n′ · v(x′ ) ′
Z I
1 1
Φ(x) = dV − dS (10.7)
4π ||x − x′ || 4π ||x − x′ ||
V ∂V

and
∇′ × v(x′ ) ′ n′ × v(x′ ) ′
Z I
1 1
A(x) = ′
dV − dS . (10.8)
4π ||x − x || 4π ||x − x′ ||
V ∂V
′ ′ ′ ′
Here the primes on ∇ , n , dV and dS indicate that the variables and the derivatives are with reference to
the coordinates x′ . This may look somewhat clumsy, but avoids a name conflict with the variable name x.
Do not be misled by this complexity: ∇′ · v(x′ ) is nothing but the divergence of v, evaluated in the point x′ .
Helmholtz’ decomposition theorem is of the utmost importance: it allows the calculation of the field
potentials in terms of the divergence and curl of the vector field, while once those potentials have been
evaluated, the field itself can be obtained. The situation that the divergence and curl of a field are known,
and one is interested in the field itself, is a common situation in physics. The Maxwell equations are a good
example.
If the region of interest is unbounded and the potentials approach zero at infinity, the surface integral
contributions to the potentials vanish. Please observe that the expression for the scalar potential then reduces
to an expression that is similar to the one that we have obtained in the previous chapter for the electrostatic
potential.

10.4 Summary
We have discussed the concepts of solenoidal and irrotational fields, and discussed that these allow the
introduction of vector and scalar potentials, respectively. We have then discussed that every field can be
written in terms of a scalar and a vector potential, and provided recipes for calculating these potentials.

Exercise 10.4 Look at the list of objectives at the beginning of this chapter and ensure that you have acquired
the mentioned skills. If not, please study the relevant parts of this chapter again and repeat this exercise.

1 based on: dr J.C. van der Meer, TU/e — Vectoranalyse Aanvulling 7, 2WA15.
2 ibid.
102 CHAPTER 10. HELMHOLTZ’ DECOMPOSITION THEOREM
Chapter 11

The Dyadic Product, Dyads and Dyadics

— Tensors

Objectives
After studying this chapter, you will be able to
• State the properties of dyads, dyadics and dyadic products (products, dot products with vectors);
• Explain how projections and the mass flux density can be formulated as dyadic products;

• Set up a basis for dyadics using covariant and/or contravariant vectors, and express a dyadic in terms
of this basis; (products, dot products with vectors);
• Understand dyadics as ‘2-tensors’;
• Express the gradient of a vector as a dyadic (2-tensor);

• State the momentum balance equation for a perfect fluid;

• Derive the divergence of a dyadic (2-tensor) and state the differential form of the momentum balance.

103
104 CHAPTER 11. THE DYADIC PRODUCT, DYADS AND DYADICS — TENSORS

11.1 Introduction
In the previous chapters we have encountered two binary vector products: the dot product u · v, which yields
a scalar, and the cross product u × v, which yields a vector. We will now introduce a third product, the
dyadic product of two vectors, which is written as uv. The result of this operation is neither a scalar, nor a
vector. It is a so-called dyad. Furthermore, a dyadic is defined as the sum of 1 or more dyads.
The language of dyadics has been introduced by Gibbs [12] in the late nineteenth century and is still used
in electrodynamics and fluid dynamics (see for example [6]). That fact alone is a good reason to discuss the
subject. But there is more: the language of dyadics is an excellent stepping stone to the subject of general
tensors. We will later identify scalars, vectors and dyadics as tensors of orders 0, 1 and 2, respectively. A
defining property of tensors is that they do not change when a change of coordinate system is made. (This
was discussed for the velocity vector in equation (4.9).)
We will proceed as follows. in section 11.2 we will define dyads, dyadics and the dyadic product in terms
of their properties. In the subsequent section we will demonstrate some of their applications: in section 11.3
we will express projections (see exercises 2.3 and 2.4) in terms of dyads and introduce another dyad: the
momentum flux density. Section 11.4 explains how dyadics can be expressed in terms of basis vectors and
components. In section 11.6, we will use this representation when we define the gradient of a vector field. In
section (11.7) we discuss the momentum balance equation; we will use that in section (11.8) to motivate the
divergence of a tensor. We conclude with a number of general remarks

11.2 Definitions
In the introduction we have mentioned that the dyadic product of two vectors u and v is written as uv
(juxtaposed, without any multiplication symbol inbetween). The relevance of this dyadic product will be
clarified in the next section. But let us first define the properties of dyads and the dyadic product, that give
it a meaning:

• The dyadic product is linear in both arguments:

(α1 u1 + α2 u2 )v = α1 u1 v + α2 u2 v, u(α1 v1 + α2 v2 ) = α1 uv1 + α2 uv2 . (11.1)

• It is not commutative; in general

uv ̸= vu. (11.2)

• Taking the dot product of a vector and a dyadic product amounts to taking the dot product with the
adjacent vector of the dyadic product: for all vectors t and w we have

t · (uv) = (t · u)v, (uv) · w = u(v · w), (11.3)

and the result is a vector in both cases. Note that, as a result of this property, the parentheses in the
expressions above can be omitted.

• The dot product of two dyadic products is given by

(tu) · (vw) = (u · v)tw. (11.4)

In words: You can form dot products of adjacent vectors (and move the resulting scalar around).

Exercise 11.1 Consider vectors a, b, ..., f and simplify the following expression, so it no longer contains
dyads. Also describe the result in words:
a · (bc) · (de) · f . (11.5)
11.3. EXAMPLES OF DYADS 105

11.3 Examples of Dyads

Let us now look at two real-world exampled of dyads.

11.3.1 Projections
In exercise 2.4 of chapter 2 we have discussed the projection of a vector u on a line that is defined by a
vector b. The result was stated in equation (2.7). Using the dyadic product, this can be written as
c·u bc
u∥ = b = · u.
b·c b·c
In other words: the projection is the dot product of a (scaled) dyadic product and the input vector u,
bc
u∥ := Pb,⊥c · u, with Pb,⊥c = . (11.6)
b·c
Exercise 11.2 Oblique projections as dyadic products.
1. Show that Pb,⊥c · Pb,⊥c = Pb,⊥c . This states that an oblique projection is idempotent: applying it
more than once to an arbitrary vector no longer changes that vector.
2. Please show that Pb,⊥c · u ∝ b;
3. Please show that the rejection vector (u − Pb,⊥c · u) is perpendicular to c, but not necessarily to b (and
the projection vector).

11.3.2 The Momentum Flux Density

Remember the discussion in chapter 5. Then we considered an infinitesimal directed surface element dA =
ndA, and derived that the amount of mass that is transported through that surface element over a time dt
is given by
dΦ = ρv · ndAdt.
This resulted in the definition of the mass flux density,

Γ = ρv.

We could say that the flux density vector derives its meaning from the result of taking the dot product
with the vector dA: that gives the amount of mass flowing through the surface element per unit of time.
Interestingly, the flux density depends only on the fluid properties ρ and v, and is therefore a property of
the fluid itself.
We will now repeat this exercise, but this time we consider the transport of momentum, rather than
mass. Our starting point is figure 5.1. The amount of momentum that is carried by the mass inside the
infinitesimal volume dV = vdt · dA is given by

vdm = ρvdV = ρv(v · dA) dt.

Using the definition of the dyadic product and its dot product property, this can be written as

ρv(v · dA) dt = ρvv · dA dt.

The dyadic product ρvv represents the momentum flux density. Note that this is dyad depends only on the
fluid properties ρ and v. By taking the dot product with a directed surface element, this dyad produces a
vector that describes the amount of momentum that flows through that surface per unit of time (a vectorial
quantity). For example, see [13, p. 52] or [6, p. 78].
106 CHAPTER 11. THE DYADIC PRODUCT, DYADS AND DYADICS — TENSORS

As pointed out earlier, in literature on hydrodynamics you will also encounter the tensor product
notation for the dyadic product. The momentum flux density tensor ρvv is then written as ρv ⊗ v
instead.

11.4 Dyadics, Coordinate Representations

A dyadic is a sum of one or more dyads. As an example, such dyadic is formed when we express the
multiplicands of a dyadic product in terms of components and basis vectors, and use the linearity of the
dyadic product, as in
uv = (ui ai )(v j aj ) = ui v j ai aj . (11.7)
We could also have used a contravariant basis for u and/or v, resulting in the alternative expressions:

uv = ui v j ai aj ; (11.8)

uv = ui vj ai aj ; (11.9)
uv = ui v j ai aj . (11.10)
Remember that these are different representations of the same tensor uv. Any dyadic can be written as

T = T ij ai aj .

A dyad is ‘just’ a special case.

11.5 The Unit Dyad

We are looking for a dyad I with properties v · I = v. But v = v i ai = (v · ai )ai = v · (ai ai ), so

I = ai ai . (11.11)

Exercise 11.3 Show that ai ai = ai ai .

11.6 The Gradient of a Vector

Remember how we derived the gradient of a scalar function in chapter 6. We now repeat this recipe to
measure variations of a vector (the velocity field, in our example), and use that to arrive at the definition of
the gradient of a vector field. The result is a dyadic:
   
1 2 3 ∂v j ∂v j i ∂v j i ∂v j
dv(x , x , x ) = dx = (a · ai dx ) = a · (ai dx ) = a · dx. (11.12)
∂xj ∂xj ∂xj ∂xj
This suggests the definition of the gradient of the vector field v,
∂v j
∇v := a . (11.13)
∂xj
In principle we could stop here. But it is interesting to express the gradient in terms of components of v
with respect to the same coordinate system. In terms of contravariant components (and a covariant basis)
we get
∂v ∂v k ak ∂v k ∂ak
= = ak + v k j . (11.14)
∂xj ∂xj ∂xj ∂x
11.7. THE MOMENTUM BALANCE 107

Using equation (4.40) to express the partial derivatives of ak in terms of the covariant basis vectors and the
Christoffel symbols Γijk , we get
∂v k
∇v = ak aj + v k Γijk ai aj . (11.15)
∂xj
After changing dummy index k into i in the first term this gives:

∂v i
 
k i j
∇v = + v Γ jk ai a . (11.16)
∂xj

The first term in parentheses is due to the change of v i ; the second is due to the change of the local basis
vectors when a particle moves around in space.

Exercise 11.4 Starting with expression dv(x1 , x2 , x3 ) = ∂v j

∂xj dx , derive expression (11.13), then develop the
expression in terms of contravariant components of v. You should be able to do this without looking at the
text above, but may use the definitions for the Christoffel syymbols Γijk .

11.7 The Momentum Balance

Like we did in chapter 5, we consider a fixed volume V that contains a fluid with mass density ρ(x) and
velocity v(x). This time we develop a balance for the momentum that is contained in the volume. The
momentum per unit of volume is given by ρv, and the total amount of momentum is given by integrating
over the volume. The rate of flow of momentum out of the volume is described by the momentum flux
density and is given by the integral of ρvv · n over the surface that bounds the volume V . The rate of change
of momentum is equal to the net inflow of momentum, plus the increase of momentum due to the forces that
act on the mass inside the volume. We will consider two types of forces: volume and surface forces. This
results in the following statement of the momentum balance, which is essentially Newton’s law of motion,
applied to the mass in the volume:
Z Z Z Z
d
ρ(x)vdV = − (ρvv · n)dA + fvol dV + fsur dA. (11.17)
dt
V ∂V V ∂V

Gravity results in a volume force f = ρg per unit of volume. For a perfect fluid, there are no viscous effects,
and the only surface force that acts on the volume, per unit of area, is given by fsur = −pn, where p is the
pressure. The result is
Z Z Z Z
d
ρ(x)vdV = − (ρvv · n)dA + ρgdV − pndA. (11.18)
dt
V ∂V V ∂V

Using (6.15), the last term can be rewritten as a volume integral and be combined with the second term on
the right-hand side,
Z Z Z
d
ρ(x)vdV = − (ρvv · n)dA + (ρg − ∇p)dV. (11.19)
dt
V ∂V V

Exercise 11.5 Study the material above carefully, and reproduce the integral form of the momentum balance,
equation (11.19) without looking back at the text. (You may use the ‘gradient theorem’, though.)
108 CHAPTER 11. THE DYADIC PRODUCT, DYADS AND DYADICS — TENSORS

11.8 The Divergence of a Tensor Field

Like we did in chapter 5, we consider the stationary momentum balance, so the time-derivative vanishes,
divide the equation by V , and let the V shrink around some point x. This gives
 
Z
1
lim  (ρvv · n)dA = ρg − ∇p. (11.20)
V ↓0 V
∂V

As before, the existence of the right-hand side suggests that the left-hand side exists as well, and we again
call this the divergence, this time of a tensor. For some arbitrary tensor T we define
 
Z
1
∇ · T = lim  (T · n)dA (11.21)
V ↓0 V
∂V

This allows us to state the differential form of the momentum balance for a perfect stationary fluid,
∇ · ρvv = ρg − ∇p. (11.22)
Like we did in 5, we will now develop a more practicle means of evaluating the divergence for a given
curvilinear coordinate system. The first part of this task is exactly the same as the derivation of the divergence
of a vector field; all we need to do is replacing the vector field Γ in exercise (5.8) with our tensor field T.
The result is √
1 ∂( gT · ai )
∇·T= √ , (11.23)
g ∂xi
which is identical to equation (5.18). For the vectorial divergence, we could simplify this expression using
Γ · ai = Γi , arriving at (5.19). In the present case, T is a tensor, the product T · ai is a vector, and some
more work needs to be done to arrive at an expression in terms of components of T. We proceed as follows.
The product rule for differentiation gives
∂(T · ak )
∇·T= + (T · ak )Γm
km . (11.24)
∂xk
Where we have used [6, p. 744, eq. 91]1 : √
1 ∂ g
√ = Γm
km (11.25)
g ∂xk
and the Γijk are the Christoffel symbols of the second kind, given by (4.40).
In order to find an expression in terms of the components of T, we need to choose a representation for
that tensor, for example
T = T ij ai aj . (11.26)
Using the rule for taking the dot product of a dyad, and the duality relation (4.26), we find that T·ak = T ik ai ,
and substitution in the previous expression yields
∂T ik ai ∂ai ik ∂T ik ∂T ik
∇·T = k
+ Γm ik
km T ai = k
T + k ai + Γm ik
km T ai = ai + Γm ik m ik
ik T am + Γkm T ai . (11.27)
∂x ∂x ∂x ∂xk
After swapping the dummy indices i and m in the second term we get the final result,
 ik 
∂T i mk m ik
∇·T= + Γ mk T + Γ km T ai . (11.28)
∂xk
Expressions for other representations of T can be obtained along the same lines.
1 See also: [Link]
11.9. CONCLUSIONS 109

Exercise 11.6 Please verify all steps in the derivation of the tensor-divergence; try to look at the text above
as little as possible. You should be able to derive the expression for the divergence without any reference to
the text above, with the exception of equation (11.25).

NOTE that another definition of ∇ · T exists, which is based on the product n · T. In many relevant
applications, the tensors are symmetric, and the result will be the same. The bad news is that this fact
allows this annoying confusion to linger on...

11.9 Conclusions
We have now arrived at the end of this text. In a few weeks time, you have learned a new language to
describe physics that you sometimes already knew in a way that does just to the fact that we believe that
our laws of nature do not depend on any particular coordinate system. Later we did introduce coordinate
systems in the most general way, but when we developed the key concepts of the divergence, gradient and
divergence of fields, we always started with definitions that do not depend on a particular choice.
In the final chapter, we investigated a new type of object: dyadics, and we also used the more modern
terminology ‘2-tensors’. We then identified scalars and vectors as 0- and 1-tensors. What makes a number
(or a collection of them, in the case of vectors and dyadics) a tensor? A tensor has the property that it does
not change when we switch to another coordinate system. That is the reason that the language of tensors
can so succesfully be applied to physical phenomena.
We have, of course, only be able to scratch the surface of the subject of tensors, and restricted ourselves
to a Euclidean space. For those of you who have gotten interested in the subject, we recommend once more
that you consider the courses on Tensor Calculus and Differential Geometry and General Relativity. Also
if you don’t, we are confident that Vector Analysis will help you with your studies of Physics of Transport
Phenomena and other courses in your further studies.

Exercise 11.7 Look at the list of objectives at the beginning of this chapter and ensure that you have acquired
the mentioned skills. If not, please study the relevant parts of this chapter again and repeat this exercise.
110 CHAPTER 11. THE DYADIC PRODUCT, DYADS AND DYADICS — TENSORS
Bibliography

[1] Carl B. Boyer. The invention of analytic geometry. Scientific American, 180(1):40–45, 1949.

[2] Howard Eves. Foundations and Fundamental Concepts of Mathematics. Dover, 3rd edition, 1997.
[3] John Playfair. Elements of Geometry. W.E. Dean, New York, 1846.
[4] Erwin Kreyszig. Differential Geometry. Dover, 1st edition, 1991.

[5] Thomas Little Heath. A History of Greek Mathematics. Adamant Media Corporation, Boston, 1921.
[6] Z.U.A. Warsi. Fluid Dynamics — Theoretical and Computational Approaches. CRC Press, 2nd edition,
1998.
[7] Charles W. Misner, Kip S. Thorne, and John Archibald Wheeler. Gravity. W.H. Freeman, 1973.

[8] Robert A. Adams. Calculus — A Complete Course. Pearson, 8th edition, 2013.
[9] Joseph Fourier. Théorie analytique de la chaleur. F. Didot, Paris, 1822.
[10] Joseph Fourier. The analytical theory of heat. Dover, New York, 1955.
[11] Peter Day. The philosopher’s tree: a selection of Michael Faraday’s writings. CRC Press, 1999.

[12] Edwin Bidwell Wilson. Vector Analysis — A Text-Book for the Use of Students of Mathematics and
Physics. Yale University, New Haven, Connecticut, 1901. Founded upon the lectures of J. Willard
Gibbs. Reprint 2015 by Forgotten Books, FB&c Ltd, London, isbn 978-1-4400-8977-0.
[13] F. Moukalled, L. Mangani, and M. Darwish, editors. The finite volume method in computational fluid
dynamics. Springer International Publishing AG Switzerland, 2016.

111