Vector Calculus

Andrea Moiola
University of Reading, 19th September 2016

These notes are meant to be a support for the vector calculus module (MA2VC/MA3VC) taking place
at the University of Reading in the Autumn term 2016.
The present document does not substitute the notes taken in class, where more examples and
proofs are provided and where the content is discussed in greater detail.
These notes are self-contained and cover the material needed for the exam. The suggested textbook
is [1] by R.A. Adams and C. Essex, which you may already have from the first year; several copies are
available in the University library. We will cover only a part of the content of Chapters 10–16, more
precise references to single sections will be given in the text and in the table in Appendix G. This book
also contains numerous exercises (useful to prepare the exam) and applications of the theory to physical
and engineering problems. Note that there exists another book by the same authors containing the worked
solutions of the exercises in the textbook (however, you should better try hard to solve the exercises and, if
unsuccessful, discuss them with your classmates). Several other good books on vector calculus and vector
analysis are available and you are encouraged to find the book that suits you best. In particular, Serge
Lang’s book [5] is extremely clear. [7] is a collection of exercises, many of which with a worked solution
(and it costs less than 10£). You can find other useful books on shelf 515.63 (and those nearby) in the
University library. The lecture notes [2], the book [3] and the “Vector Calculus Primer” [6] are available
online; on the web page [4] of O. Knill you can find plenty of exercises, lecture notes and graphs. Note that
different books inevitably use different notation and conventions. In this notes we will take for granted
what you learned in the previous classes, so the first year notes might be useful from time to time (in
particular those for calculus, linear algebra and analysis).
Some of the figures in the text have been made with Matlab. The scripts used for generating these
plots are available on Blackboard and on the web page1 of the course; most of them can be run in Octave
as well. You can use and modify them: playing with the different graphical representations of scalar and
vector fields is a great way to familiarise with these concepts. On the same page you can find the file
VCplotter.m, which you can use to visualise fields, curves and changes of variables in Matlab or Octave.
Warning 1: The paragraphs and the proofs marked with a star “ ⋆” are addressed to students willing
to deepen the subject and learn about some closely related topics. Some of these remarks try to relate the
topic presented here to the content of other courses (e.g., analysis or physics); some others try to explain
some important details that were glossed over in the text. These parts are not requested for the exam
and can safely be skipped by students with modest ambitions.
Warning 2: These notes are not entirely mathematically rigorous, for example we usually assume
(sometimes tacitly) that fields and domains are “smooth enough” without specifying in detail what we
mean with this assumption; multidimensional analysis is treated in a rigorous fashion in other modules,
e.g. “analysis in several variables”. On the other hand, the (formal) proofs of vector identities and of some
theorems are a fundamental part of the lectures, and at the exam you will be asked to prove some
simple results. The purpose of this course is not only to learn how to compute integrals and divergences!
Suggestion: The content of this module can be seen as the extension to multiple variables and vector
quantities of the calculus you learned in the first year. Hence, many results you will encounter in this
course correspond to simpler similar facts, holding for real functions, you already know well from previous
classes. Comparing the vector results and formulas you will learn here with the scalar ones you already
know will greatly simplify the study and the understanding of the content of this course. (This also means
that you need to know and remember well what you learned in the first year.)

If you find typos or errors of any kind in these notes, please let me know at [email protected]
or in person during the office hours, before or after the lectures.

1 http://www.personal.reading.ac.uk/
~ st904897/VC2016/VC2016.html
A. Moiola, University of Reading 2 Vector calculus lecture notes, 2016–17

1 Fields and vector differential operators

For simplicity, in these notes we only consider the 3-dimensional Euclidean space R3 , and, from time to
time, the plane R2 . However, all the results not involving neither the vector product nor the curl operator
can be generalised to Euclidean spaces RN of any dimensions N ∈ N.

1.1 Review of vectors in 3-dimensional Euclidean space

We quickly recall some notions about vectors and vector operations known from previous modules; see
Sections 10.2–10.3 of [1] and the first year calculus and linear algebra notes. We use the word “scalar”
simply to denote any real number x ∈ R.
We denote by R3 the three-dimensional Euclidean vector space. A vector is an element of R3 .

Notation. In order to distinguish scalar from vector quantities, we denote vectors with boldface and a little
arrow: ~u ∈ R3 . Note that several books use underlined (u) symbols. We use the hat symbol (ˆ) to denote
unit vectors, i.e. vectors of length 1.
You are probably used to write a vector ~u ∈ R3 in “matrix notation” as a “column vector”
 
u1
~u = u2  ,
u3

where the scalars u1 , u2 and u3 ∈ R are the components, or coordinates, of ~u. We will always use the
equivalent notation
~u = u1 ı̂ + u2 ̂ + u3 k̂,

where ı̂, ̂ and k̂ are three fixed vectors that constitute the canonical basis of R3 . When we draw vectors,
we always assume that the canonical basis has a right-handed orientation, i.e. it is ordered according to
the right-hand rule: closing the fingers of the right hand from the ı̂ direction to the ̂ direction, the thumb
1
0
0

points towards the k̂ direction. You can think at ı̂, ̂ and k̂ as the vectors ı̂ = 0 , ̂ = 1 and k̂ = 0 ,
0 0 1
1
0
0
u1

so that the two notations above are consistent: ~u = u1 ı̂ + u2 ̂ + u3 k̂ = u1 0 + u2 1 + u3 0 = uu2 .
0 0 1 3

u3

~u
k̂
ı̂ ̂ u2
u1

Figure 1: The basis vectors ı̂, ̂, k̂ and the components of the vector ~u.

2
Definition 1.1. The magnitude, or length, or norm, of the vector ~u is the scalar defined as
q
u := |~u| := u21 + u22 + u23 .

The direction of a (non-zero) vector ~

u is the unit vector defined as

~u
û := .
|~u|

Note that often the magnitude of a vector ~u is written as k~uk (e.g. in your Linear Algebra lecture
notes). We use the same notation |~u| for the magnitude of a vector and |x| for the absolute value of a
2 The notation “A := B” means “the object A is defined to be equal to the object B”. A and B may be scalars, vectors,

matrices, sets. . .
A. Moiola, University of Reading 3 Vector calculus lecture notes, 2016–17

scalar, which can be thought as the magnitude of a one-dimensional vector; the meaning of the symbol
should always be clear from the argument.
Every vector satisfies ~u = |~u|û. Therefore length and direction uniquely identify a vector. The
vector of length 0 (i.e. ~0 := 0ı̂ + 0̂ + 0k̂) does not have a specified direction. Note that, if we want to
represent vectors with arrows, the point of application (“where” we draw the arrow) is not relevant: two
arrows with the same direction and length represent the same vector; we can imagine that all the arrows
have the application point in the origin ~0.
Physics provides many examples of vectors: e.g. velocity, acceleration, displacement, force, momentum.
Example 1.2 (The position vector). The position vector

~r = xı̂ + y̂ + z k̂

represents the position of a point in the three-dimensional Euclidean space relative to the origin. This is the
only vector for which we do not use the notation ~r = r1 ı̂ + r2 ̂ + r3 k̂. We will use the position vector mainly
for two purposes: (i) to describe subsets of R3 (for example {~r ∈ R3 , |~r| < 1} is the ball of radius 1 centred
at the origin), and (ii) as argument of fields, which will be defined in Section 1.2.
⋆ Remark 1.3 (Are vectors arrows, points, triples of numbers, or elements of a vector space?). There are
several different definitions of vectors, this fact may lead to some confusion. Vectors defined as geometric
entities fully described by magnitude and direction are sometimes called “Euclidean vectors” or “geometric
vectors”. Note that, even though a vector is represented as an arrow, in order to be able to sum any two
vectors the position of the arrow (the application point) has no importance.
One can use vectors to describe points, or positions, in the three-dimensional space, after an origin has
been fixed. These are sometimes called “bound vectors”. In this interpretation, the sum of two vectors does
not make sense, while their difference is an Euclidean vector as described above. We use this interpretation to
describe, for example, subsets of R3 ; the position vector ~r is always understood as a bound vector.
Often, three-dimensional vectors are intended as triples of real numbers (the components). This is equivalent
to the previous geometric definition once a canonical basis {ı̂, ̂, k̂} is fixed. This approach is particularly helpful
to manipulate vectors with a computer program (e.g. Matlab, Octave, Mathematica, Python. . . ). We typically
use this interpretation to write formulas and to do operations with vectors.
Vectors are usually rigorously defined as elements of an abstract “vector space” (who attended the linear
algebra class should be familiar with this concept). This is an extremely general and powerful definition which
immediately allows some algebraic manipulations (sums and multiplications with scalars) but in general does
not provide the notions of magnitude, unit vector and direction. If the considered vector space is real, finite-
dimensional and is provided with an inner product, then it is an Euclidean space (i.e., Rn for some natural
number n). If a basis is fixed, then elements of Rn can be represented as n-tuples of real numbers (i.e., ordered
sets of n real numbers).
See http://en.wikipedia.org/wiki/Euclidean vector for a comparison of different definitions.
Several operations are possible with vectors. The most basic operations are the addition ~u + w
~ and
the multiplication λ~u with a scalar λ ∈ R:

~u + w
~ = (u1 + w1 )ı̂ + (u2 + w2 )̂ + (u3 + w3 )k̂, λ~u = λu1 ı̂ + λu2 ̂ + λu3 k̂.

These operations are defined in any vector space. In the following we briefly recall the definitions and the
main properties of the scalar product, the vector product and the triple product. For more properties,
examples and exercises we refer to [1, Sections 10.2–10.3].

Warning: frequent error 1.4. There exist many operations involving scalar and vectors, but what is never
possible is the addition of a scalar and a vector. Recall: for λ ∈ R and ~u ∈ R3 you can never write anything
like“λ + ~u”! Writing a little arrow on each vector helps preventing mistakes.

⋆ Remark 1.5. The addition, the scalar multiplication and the scalar product are defined for Euclidean spaces
of any dimension, while the vector product (thus also the triple product) is defined only in three dimensions.

1.1.1 Scalar product

Given two vectors ~ ~ , the scalar product (also called dot product or inner product) gives in
u and w
output a scalar denoted by ~u · w
~:

~ ~ := u1 w1 + u2 w2 + u3 w3 = |~u||~
u·w w| cos θ, (1)
A. Moiola, University of Reading 4 Vector calculus lecture notes, 2016–17

~ , as in Figure 2. The scalar product is commutative (i.e. ~u · w

where θ is the angle between ~u and w ~ =w~ ·~u),
is distributive with respect to the addition (i.e. ~u · (~v + w
~ ) = ~u · ~v + ~u · w
~ ) and can be used to compute
the magnitude of a vector: √
|~u| = ~u · ~u.
The scalar product can be used to evaluate the length of the projection of ~u in the direction of w
~:

~u · w
~
projection = |~u| cos θ = = ~u · ŵ,
|~
w|

and the angle between two (non-zero) vectors:

 
~u · w
~

θ = arccos = arccos û · ŵ .
|~u||~w|

The vector components can be computed as scalar products:

u1 = ~u · ı̂, u2 = ~u · ̂, u3 = ~u · k̂.

Two vectors ~u and w ~ are orthogonal or perpendicular if their scalar product is zero: ~u · w ~ = 0; they
are parallel (or collinear) if ~ w for some scalar α 6= 0 (in physics, if α < 0 they are sometimes called
u = α~
“antiparallel”).

~
w

|~u| cos θ
θ
~u

Figure 2: The projection of ~u in the direction of w

~.

Exercise 1.6. ◮ Compute the scalar product between the elements of the canonical basis: ı̂ · ı̂, ı̂ · ̂, ı̂ · k̂ . . .
Exercise 1.7 (Orthogonal decomposition). ◮ Given two non-zero vectors ~u and w ~ , prove that there exists a
unique pair of vectors ~u⊥ and ~u|| , such that: (a) ~u⊥ is perpendicular to w ~ , (b) ~u|| is parallel to w
~ and (c)
~u = ~u⊥ + ~u|| .
Hint: in order to show the existence of ~u⊥ and ~u|| you can proceed in two ways. (i) You can use the
~ ” to represent ~u⊥ and ~u|| in dependence of a parameter α, and use the condition
condition “~u|| is parallel to w
“~u|| is perpendicular to w ~ ” to find an equation for α itself. (ii) You can use your geometric intuition to guess
the expression of the two desired vectors and then verify that they satisfy all the required conditions. (Do not
forget to prove uniqueness.)

1.1.2 Vector product

Given two vectors ~u and w
~ , the vector product (or cross product) gives in output a vector denoted
by ~u × w
~ (sometimes written ~u ∧ w
~ to avoid confusion with the letter x on the board):
 
 ı̂ ̂ k̂ 

~ : =  u1 u2 u3 
~ ×w
u
w1 w2 w3 
      (2)
u u3   u1 u3   u1 u2 
= ı̂  2  − ̂ 
w1  + k̂ 
w1
w2 w3 w3 w2 
= (u2 w3 − u3 w2 )ı̂ + (u3 w1 − u1 w3 )̂ + (u1 w2 − u2 w1 )k̂,

where | : : | denotes the matrix determinant. Note that the 3 × 3 determinant is a “formal determinant”, as
the matrix contains three vectors and six scalars: it is only a short form for the next expression containing
three “true” 2 × 2 determinants.
The magnitude of the vector product

|~u × w
~ | = |~u||~
w| sin θ
A. Moiola, University of Reading 5 Vector calculus lecture notes, 2016–17

~u × w
~

~
w
θ
~u

Figure 3: The vector product ~u × w ~ has magnitude equal to the area of the grey parallelogram and is
orthogonal to the plane that contains it.

is equal to the area of the parallelogram defined by ~u and w ~ . Its direction is orthogonal to both ~u and
~ and the triad ~u, w
w ~ , ~u × w
~ is right-handed. The vector product is distributive with respect to the sum
(i.e. ~u × (~v + w ~)=~ u × ~v + ~u×w ~ ) × ~u = ~v × ~u + w
~ and (~v + w ~ × ~u) but is not associative: in general
~u × (~v × w ~ ) 6= (~u × ~v) × w~.
~ ∈ R3 , their vector product ~u × w
Exercise 1.8. ◮ Prove what was claimed above: given ~u and w ~ is orthogonal
to both of them.
Exercise 1.9. ◮ Show that the elements of the standard basis satisfy the following identities:

ı̂ × ̂ = k̂, ̂ × k̂ = ı̂, k̂ × ı̂ = ̂,

̂ × ı̂ = −k̂, k̂ × ̂ = −ı̂, ı̂ × k̂ = −̂,
ı̂ × ı̂ = ̂ × ̂ = k̂ × k̂ = ~0.

Exercise 1.10. ◮ Show that the vector product is anticommutative, i.e.

~ × ~u = −~u × w
w ~ (3)

u × ~u = ~0.
~ in R3 , and satisfies ~
for all ~u and w
Exercise 1.11. ◮ Prove that the vector product is not associative by showing that (ı̂ × ̂) × ̂ 6= ı̂ × (̂ × ̂).
Exercise 1.12. ◮ Show that the following identity holds true:

~u × (~v × w
~ ) = ~v(~u · w
~)−w
~ (~u · ~v) ~ ∈ R3 .
∀~u, ~v, w (4)

(Sometimes ~u × (~v × w
~ ) is called “triple vector product”.)
~ , ~p ∈ R3 :
Exercise 1.13. ◮ Show that the following identities hold true for all ~u, ~v, w

~u × (~v × w ~ ) + ~v × (~
w × ~u) + w ~ × (~u × ~v) = ~0 (Jacobi identity),
(~u × ~v) · (~
w × ~p) = (~u · w
~ )(~v · p ~ )(~v · w
~ ) − (~u · p ~) (Binet–Cauchy identity),
2 2 2 2
|~u × ~v| + (~u · ~v) = |~u| |~v| (Lagrange identity).

Hint: to prove the first one you can either proceed componentwise (long and boring!) or use identity (4).
For the second identity you can expand both sides and collect some terms appropriately. The last identity will
follow easily from the previous one.
⋆ Remark 1.14. The vector product is used in physics to compute the angular momentum of a moving object,
the torque of a force, the Lorentz force acting on a charge moving in a magnetic field.
Given three vectors ~u and ~v and w
~ , their triple product is the scalar

~u · (~v × w
~ ).

Its absolute value is the volume of the parallelepiped P defined by the three vectors as in Figure 4. To
~ as n̂ := |~~vv×
see this, we define the unit vector orthogonal to the plane containing ~v and w ~
w
~ . Then
×w|
 
 ~ 
~v × w
Volume (P ) : = (area of base) × height = |~v × w ~ | ~u ·
~ | |~u · n̂| = |~v × w = |~u · (~v × w
~ )| .
~ |
|~v × w
A. Moiola, University of Reading 6 Vector calculus lecture notes, 2016–17

From the definition (2) of vector product, we see that the triple product can be computed as the deter-
minant of the matrix of the vector components:
   
 ı̂ ̂ k̂   u1 u2 u3 

~u · (~v × w u ·  v1 v2 v3  =  v1 v2 v3 
~)=~ (5)
w1 w2 w3  w1 w2 w3 
= u1 v2 w3 + u2 v3 w1 + u3 v1 w2 − u1 v3 w2 − u2 v1 w3 − u3 v2 w1 .

The triple product ~u · (~v × w

~ ) is zero if the three vectors are linearly dependent, is positive if the triad ~u,
~v and w~ is right-handed, and is negative otherwise.

Exercise 1.15. ◮ Show the following identities (you may use the determinant representation (5))

~u · (~v × w
~ ) = ~v · (~
w × ~u) = w
~ · (~u × ~v) ~ ∈ R3 .
∀~u, ~v, w

What is the relation between ~u · (~v × w ~ · (~v × ~u)? What is ~u · (~v × ~u)?
~ ) and w

P
~u

height = |~u · n̂| n̂

~
w

~v area = |~v × w
~|

Figure 4: The triple product ~u · (~v × w

~ ) as the volume of a parallelepiped.

Warning: frequent error 1.16 (Ambiguous expressions). It is extremely important that all the formulas we
write are unambiguous. For this reason, a correct use of brackets is fundamental, especially when dealing with
vectors. For example, the expression “~u × ~v × w ~ ” is ambiguous and not acceptable because it does not
make clear the order in which the products are performed: it might be interpreted either as (~u × ~v) × w ~ or as
~u × (~v × w ~ ), which are in general not equal to each other; cf. Exercise 1.11.
On the other hand ~ u · (~v × w~ ) might be written ~u · ~v × w
~ , which is unambiguous and acceptable because
there is only one possible way of bracketing this expression, i.e. ~u · (~v × w ~ ). The other option, “(~u · ~v) × w
~ ”,
makes no sense because it contains a vector product between a scalar and a vector. When in doubt, however,
it is safer to write some extra brackets (in the correct positions) rather than skip them.
Recall also that by convention all kinds of products (scalar ~u · w ~ , vector ~u × w
~ , scalar-times-vector λ~u)
have priorities over sums. E.g. ~u + ~v × w ~ means ~u + (~v × w ~ ) and is not equal to (~u + ~v) × w ~.
Both scalar and vector products are products: each term is made of sums of products between a coefficient
of the first vector and a coefficient of the second one. As such, they enjoy the distributive property with respect
to addition: ~u · (~v + w
~ ) = (~u · ~v) + (~u · w
~ ) and ~u × (~v + w
~ ) = (~u × ~v) + (~u × w
~ ). This property involves
addition, there is no such thing as distributive property between two products, e.g. λ(~u × w ~ ) 6= (λ~u) × (λ~
w)
(a correct identity is λ(~u × w ~ ) = (λ~u) × w ~ = ~u × (λ~w)).

1.1.3 ⋆ Open and closed subsets of R3

Comparison with scalar calculus 1.17. In one dimension we call an interval “open” if it does not include
its endpoints, e.g. (0, 1) = {t ∈ R, 0 < t < 1} and we call it “closed” if it includes them, e.g. [0, 1] = {t ∈
R, 0 ≤ t ≤ 1}. More generally, you should remember from last year’s Real Analysis lecture notes the definitions
of open and closed subsets of R (see Definition 7.1 and Proposition 7.10 therein). Open sets are also related
to the convergence of sequences. In this section we see how to extend these concepts to R3 (or Rn for any
n ∈ N).
Given a sequence of vectors {~uj }j∈N ⊂ R3 , i.e. a set of vectors indexed by a natural number j ∈ N,
we say that {~uj }j∈N has limit ~u ∈ R3 if limj→∞ |~uj − ~u| = 0. In this case, we also say that {~uj }j∈N
converges to ~u and we write ~u = limj→∞ ~uj and ~uj → ~u. (Note that, thanks to the use of the magnitude
of the difference ~uj − ~
u, we have defined the limit of a sequence of vectors using the limit of a sequence
of real numbers.)
A. Moiola, University of Reading 7 Vector calculus lecture notes, 2016–17

Exercise 1.18. ◮ (i) Given a sequence of vectors {~uj }j∈N ⊂ R3 and ~u ∈ R3 , prove that limj→∞ ~uj = ~u ∈ R3
if and only if each of the three real sequences of the three components of ~uj converges to the corresponding
component of ~u. In formulas, you have to prove that

lim ~
uj = ~u ⇐⇒ lim (~uj )1 = u1 , lim (~uj )2 = u2 , lim (~uj )3 = u3 .
j→∞ j→∞ j→∞ j→∞

(ii) Find a sequence of vectors {~uj }j∈N ⊂ R3 and a vector ~u ∈ R3 such that limj→∞ |~uj | − |~u| = 0 but ~uj
does not converge to ~u.
A set D ⊂ R3 is called an open set if for every point ~p ∈ D, there exists ǫ > 0 (depending on ~p) such
that all points ~q at distance smaller than ǫ from ~p belong to D. In formulas:

D is open if ∀~p ∈ D, ∃ǫ > 0 s.t. {~q ∈ R3 , |~q − ~p| < ǫ} ⊂ D.

The word domain is used as a synonym of open set (because the domain of definition of scalar and vector
fields, described in the following, is usually chosen to be an open set). A set C ⊂ R3 is called a closed set
if for all sequences contained in C and converging to a limit in R3 , the limit belongs to C. In formulas:

C is closed if ∀{~pj }j∈N ⊂ C s.t. lim ~pj = ~p ∈ R3 , we have ~p ∈ C.

j→∞

Examples of open sets are the open unit ball {~r ∈ R3 , |~r| < 1}, the open unit cube {~r ∈ R3 , 0 < x, y, z < 1},
the open half space {~r ∈ R3 , x > 0}. Examples of closed sets are the closed unit ball {~r ∈ R3 , |~r| ≤ 1},
the closed unit cube {~r ∈ R3 , 0 ≤ x, y, z ≤ 1}, the closed half space {~r ∈ R3 , x ≥ 0}, all the planes e.g.
{~r ∈ R3 , x = 0}, the lines e.g. {~r ∈ R3 , x = y = 0}, the sets made by a single points e.g. {~0}. Typically,
the sets defined using strict inequalities (i.e. > and <) are open, while those defined with non-strict
inequalities or equalities (i.e. ≥, ≤ and =) are closed. Closed sets may be “thin”, like planes and lines,
while open sets are always “fat”, as they contain little balls around each point. The empty set and R3 are
the only two sets that are simultaneously open and closed.
Analogous definitions can be given for two dimensional sets, i.e. subsets of R2 . Two-dimensional
domains are also called regions.
Exercise 1.19. ◮ Prove that the complement of an open set is closed and vice versa. (This is not easy!)
Notation. In the following we will use the letters D, E, . . . to denote three-dimensional open sets, and the
letters P, Q, R, S, . . . to denote two-dimensional open sets. S will also be used to name surfaces.

1.2 Scalar fields, vector fields and curves

The fundamental objects of first-year (scalar) calculus are “functions”, in particular real functions of
a real variable, i.e. rules that associate to every number t ∈ R a second number f (t) ∈ R. In this
section we begin the study of three different extensions of this concept to the vector case, i.e. we consider
functions whose domains, or codomains, or both, are the three-dimensional Euclidean space R3 (or the
plane R2 ), as opposed to the real line R. Fields3 are functions of position, described by the position
vector ~r = xı̂ + y̂ + z k̂, so their domain is R3 (or a subset of it). Depending on the kind of output,
they are called either scalar fields or vector fields. Curves are vector-valued functions of a real variable.
In Remark 1.27 we summarise the mapping properties of all these objects. Scalar fields, vector fields and
curves are described in [1] in Sections 12.1, 15.1 and 11.1 respectively.

1.2.1 Scalar fields

A scalar field is a function f : D → R, where the domain D is an open subset of R3 . The value of f at the
point ~r may be written as f (~r) or f (x, y, z). (This is because a scalar field can equivalently be interpreted
as a function of one vector variable or as a function of three scalar variables.) Scalar field may also be
called “multivariate functions” or “functions of several variables” (as in the first-year calculus modulus,
see handout 5). Some examples of scalar fields are

f (~r) = x2 − y 2 , g(~r) = xyez , h(~r) = |~r|4 .

We will often consider two-dimensional scalar fields, namely functions f : R → R, where now R is a
domain in R2 , i.e. a region of the plane. Two-dimensional fields may also be thought as three-dimensional
fields that do not depend on the third variable z (i.e. f (x, y, z) = f (x, y) for all z ∈ R).
3 Note that the fields that are object of this section have nothing to do with the algebraic definition of fields as abstract

sets provided with addition and multiplication (like R, Q or C). The word “field” is commonly used for both meaning and
can be a source of confusion; the use of “scalar field” and “vector field” is unambiguous.
A. Moiola, University of Reading 8 Vector calculus lecture notes, 2016–17

Smooth scalar fields can be graphically represented using level sets, i.e. sets defined by the equations
{f (~r)=constant}; see the left plot in Figure 5 for an example. The level sets of a two-dimensional field are
(planar) level curves and can be easily drawn, while the level sets of a three-dimensional vector field are
the level surfaces, which are harder to visualise. Level curves are also called contour lines or isolines,
and level surfaces are called isosurfaces. Note that two level surfaces (or curves) never intersect each
other, since at every point ~r the field f takes only one value, but they might “self-intersect”.

2
y
−2 −3
−3 −2 1
− 0
4
1.5
−1 −2
0 3

−1
1

1
1 2
0 1
3

3
0.5
z 0
2

2
0
−1
0 x
−2
0
1
1

−0.5 −3

0 −4
3
3

−1 2
−1
2
2

0 1 2
−1 −2 1
−1.5
1 1 y 0

−1 0 −1
0
x
−2 −3 −3 −2 −1
−2 −2
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2 −2

Figure 5: Left: the level set representation of the scalar field f (~r) = x2 − y 2 . Since f does not depend
on the z-component of the position, we can think at it as a two-dimensional field and represent it with the
level curves corresponding to a section in the xy-plane. Each colour represents the set of points ~r such that
f (~r) is equal to a certain constant, e.g. f (~r) = 0 along the green lines and f (~r) = 2 along the red curves
(see the .pdf file of these notes for the coloured version). This plot is obtained with Matlab’s command
contour.
Right: the same field f = x2 − y 2 represented as the surface Sf = {~r ∈ R3 s.t. z = f (x, y)}; see
Remark 1.21. Surfaces can be drawn in Matlab with the commands mesh and surf.

Example 1.20 (Different fields with the same level sets). Consider the level surfaces of the scalar field f (~r) =
|~r|2 . Every surface corresponds to the set of points that are solutions of the quadratic equation f (~r) = |~r|2 =
x2 + y 2 + z 2 = C for some C ∈ R (C ≥ 0), thus they are the spheres centred at the origin.
2
Now consider the level surfaces of the scalar field g(~r) = e−|~r| . They correspond to the solutions of the
2 2 2
equation e−x −y −z = C, or (x2 + y 2 + z 2 ) = − log C, for some 0 < C ≤ 1. Therefore also in this case they
are the spheres centred at the origin (indeed the two fields are related by the identity g(~r) = e−f (~r) ).
We conclude that two different scalar fields may have the same level surfaces, associated with different field
values (see Figure 6).

Comparison with scalar calculus 1.21 (Graph surfaces). A real function of one variable g : R → R is usually
represented with its graph Gg = {(x, y) ∈ R2 s.t. y = g(x)}, which is a subset of the plane. Exactly in the
same way, a two-dimensional scalar field f : R2 → R can be visualised using its graph

Sf = ~r = xı̂ + y̂ + z k̂ ∈ R3 , z = f (x, y) ,

which is a surface, i.e. a two-dimensional set lying in R3 ; see the example in the right plot of Figure 6. (We
study surfaces more in detail in Section 2.2.4.) The graph of a general three-dimensional scalar field is a
hypersurface and we cannot easily visualise as it is a three-dimensional set living in a four-dimensional space.
Exercise: represent the graphs of the two fields defined in Example 1.20 (or their planar section at z = 0).
Despite having the same level sets the surfaces representing their graphs are very different from each other.
⋆ Remark 1.22. A scalar field f : D → R, where D is an open set, is continuous at ~r0 ∈ D if for all ǫ > 0
there exists δ > 0 such that for all ~r ∈ D with |~r − ~r0 | < δ then |f (~r) − f (~r0 )| < ǫ. The field f is continuous
in D if it is continuous at each point of D. Equivalently, f is continuous at ~r0 ∈ D if for all sequences of
points {~rj }j∈N ⊂ D with limj→∞ ~rj = ~r0 , it follows that limj→∞ f (~rj ) = f (~r0 ). A continuous field is often
called “of class C 0 ”.
A. Moiola, University of Reading 9 Vector calculus lecture notes, 2016–17

y y
2 2
4

76 5
5 67
3
1.5 1.5
2

0.
0.2 0

1
1 1 .4

0.1

.5
.3
3

0.3
0.5 0.5

0.60
0.8
1

0.3 0

0.7
2

1
3
4

0.2
0.2
0 x 0 x

0.8
00.5

00.5.6
0.1
3

0
2

0.1
0.
.7
.4
4

0.4
9
−0.5 −0.5

1 0.3

3
−1 −1
0.2
5 67

76 5
4

−1.5 3 2 −1.5 0.1

−2
4 −2
−2 −1 0 1 2 −2 −1 0 1 2

Figure 6: The level sets of the two scalar fields described in Example 1.20; only the plane z = 0 is
represented here. In the left plot, the field f (~r) = |~r|2 (level sets for the values 1, 2,. . . , 7); in the right
2
plot, the field g(~r) = e−|~r| (level sets for the values 0.1, 0.2,. . . , 0.9). Each level set of f is also a level
set of g (do not forget that the plots represent only few level sets).

Note that studying the continuity of a field can be much more tricky than for a real function. For example,
it is clear that the field 1/|~r| is discontinuous at the origin, 1/xyz is discontinuous on the three coordinate
planes and sign(sin x)ey has jump discontinuities on the planes {x = πn}, n ∈ N. However, in other cases the
2
discontinuity is somewhat hidden: can you see why the two-dimensional fields f (~r) = x22xy r) = x2x
+y 2 and g(~
y
4 +y 2

are both discontinuous at the origin? (You may learn more on this in the “analysis in several variables” module;
if you cannot wait take a look at Chapter 12 of [1].)
⋆ Remark 1.23 (Admissible smoothness of fields). In this notes we always assume that scalar and vector
fields we use are “smooth enough” to be able to take all the derivatives we need. We will never be very precise
on this important point, and you can safely ignore it for this module (only!). When we require a field to be
“smooth”, we usually mean that it is described by a C ∞ function, i.e. that it is continuous and all its partial
derivatives (of any order) are continuous. However, in most cases C 2 regularity will be enough for our purposes
(i.e. we need the continuity of the field and its derivatives up to second order only). For example, whenever
we write an identity involving the derivatives of a scalar field f , we will be able to consider f (~r) = |~r|2 , which
enjoys C ∞ regularity, but not f (~r) = |~r| whose partial derivatives are not well-defined at the origin (it is not
continuously differentiable, i.e. it is not of class C 1 ). See also Remarks 1.22, 1.31 and 1.42 for the definition
of the regularity classes C k .

1.2.2 Vector fields

~ is a function of position whose output is a vector, namely it is a function
A vector field F
~ : D → R3 ,
F ~ r) = F1 (~r) ı̂ + F2 (~r) ̂ + F3 (~r) k̂,
F(~
~
where the domain D is a subset of R3 . The three scalar fields F1 , F2 and F3 are the components of F.
Some examples of vector fields are
~ r) = 2xı̂ − 2y̂,
F(~ ~ r) = yzı̂ + xz̂ + xy k̂,
G(~ ~ r) = |~r|2 ı̂ + cos y k̂.
H(~

Figure 7 shows two common graphical representations of vector fields.

Two-dimensional (or planar) vector fields are functions F ~ : R → R2 , with R a region in R2 . Two-
dimensional vector fields may also be considered as three-dimensional fields that do not depend on the
third variable z, and whose third component F ~ 3 is identically zero, i.e. F(~
~ r) = F1 (x, y)ı̂ + F2 (x, y)̂. The
field depicted in Figure 7 is two dimensional.

1.2.3 Curves
Curves are vector-valued functions of a real variable ~a : I → R3 , where I is either the real line I = R or
an interval I ⊂ R. Since the value ~a(t) at each t ∈ I is a vector, we expand a curve as:

~a(t) = a1 (t)ı̂ + a2 (t)̂ + a3 (t)k̂,

A. Moiola, University of Reading 10 Vector calculus lecture notes, 2016–17

y y
2 2

1.5 1.5

1 1

0.5 0.5

0 x 0 x
−0.5 −0.5

−1 −1

−1.5 −1.5

−2 −2
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2

Figure 7: Two visualisations of the two-dimensional vector field F(~~ r) = 2xı̂ − 2y̂. In the left plot, the
direction of the arrow positioned in ~r indicated the direction of the field in that point and its length is
~ r). In the right plot the lines are tangent to the field F(~
proportional to the magnitude of F(~ ~ r) in every
point; these curves are called field lines or streamlines. Note that the streamline representation does
not give any information about the magnitude of the vector field. These plots are obtained with Matlab’s
commands quiver and streamslice (unfortunately the latter command is not available in Octave).

where a1 , a2 and a3 : I → R (the components of ~a) are real functions.

We say that ~a is continuous if its three components are continuous real functions. If the interval I is
closed and bounded, namely I = [tI , tF ] ⊂ R, and ~a : I → R3 is a curve satisfying ~a(tI ) = ~a(tF ) (i.e. it
starts and ends in the same point), then ~a is called loop.
We denote by Γ := {~r ∈ R3 , s.t. ~r = ~a(t), for some t ∈ I} ⊂ R3 the image of the curve ~a, i.e. the
set of the points covered by ~a for the various values of t ∈ I; we call Γ the path (or trace, or support)
of ~a. We say that the path Γ is parametrised by the curve ~a, or vice versa that ~a parametrises Γ (i.e.
it associates to every “parameter” t ∈ I a point on the path Γ). Some examples of paths of curves are
shown in Figure 8.
2
1 1
y y
0.8 0.8
1.5

0.6 0.6

0.4 z 1 0.4

0.2 0.2

0
x 0.5
0
x
−0.2 −0.2
0
−0.4 1 −0.4

0.5 1
−0.6 −0.6

y 0
0.5
−0.8
−0.5
−0.5
0
x −0.8

−1 −1 −1
−1 −0.5 0 0.5 1 −1 −1 −0.5 0 0.5 1

Figure 8: Three examples of paths defined by curves. In the left plot the cubic ~a(t) = tı̂ + t3 ̂, plotted for
the interval [−1, 1]. In the centre plot the helix ~b(t) = cos t ı̂ + sin t ̂ + 0.1tk̂, for t ∈ [0, 6π]. In the right
plot the “clover” ~c(t) = (cos 3t)(cos t)ı̂ + (cos 3t)(sin t)̂ for t ∈ [0, π]. Note that ~c is not injective as it
intersects itself in the origin (~c(π/6) = ~c(π/2) = ~c(5π/6) = ~0). As ~c(0) = ~c(π), the curve ~c is a loop. The
curves are drawn with the Matlab commands plot (planar curves) and plot3 (three-dimensional curves).

If we interpret the variable t as time, a curve may represent the trajectory of a point moving in space.
Following this interpretation, curves are “oriented”, i.e. they possess a preferred direction of travelling,
while paths are not.
We stress that, as opposed to the common non-mathematical meaning, the word “curve” indicates a
function ~a, whose image (its path) is a subset of R3 , and not the image itself. Indeed, different curves
may define the same path. For example
~a(t) = cos t ı̂ + sin t ̂ t ∈ [0, 2π) and ~b(τ ) = cos 2τ ı̂ + sin 2τ ̂ τ ∈ [0, π)

are two different curves which have the same path: the unit circle in the xy-plane.
In the textbook [1] and in several other books, the general curves and their components are denoted
~r(t) = x(t)ı̂ + y(t)̂ + z(t)k̂.
A. Moiola, University of Reading 11 Vector calculus lecture notes, 2016–17

Remark 1.24 (How to parametrise a path). Given a path (or a curve), how can we find its curve (or its path)?
In one direction the procedure is straightforward. If we have the analytic expression of a curve ~a : I → R3 ,
its path can be drawn (approximately) simply by evaluating ~a(tj ) for many tj ∈ I and “connecting the points”.
This is exactly what is done automatically by Matlab or Octave to make e.g. Figure 8.
The opposite operation is sometimes very difficult: since every path corresponds to infinitely many curves,
there is no universal procedure to construct ~a from Γ. Here we only consider a few simple, but very important,
examples. Try to draw all the paths by computing some points via the parametrisation.
• Γ is the segment with vertices ~p and ~q. A parametrisation of Γ is

p + t(~q − ~p) = (1 − t)~p + t~q,

~a(t) = ~ for t ∈ [0, 1].

(The components are a1 (t) = (1 − t)p1 + tq1 , a2 (t) = (1 − t)p2 + tq2 and a3 (t) = (1 − t)p3 + tq3 .)
• Γ is part of the graph of a real function g : I → R laying in the xy-plane (where I ⊂ R is an interval).
Then we can take ~a(t) = tı̂ + g(t)̂, for t ∈ I. (The components are a1 (t) = t, a2 (t) = g(t) and a3 (t) = 0.)
• Γ is the circumference with centre ~p = p1 ı̂ + p2 ̂ and radius R lying in the xy-plane. Then we can take
~a(t) = ~p + R cos tı̂ + R sin t̂ for t ∈ [0, 2π). Indeed, this circumference has equation Γ = {~r, (x − p1 )2 +
(y − p2 )2 = R2 , z = 0}, which is satisfied by the curve above (verify this).
• What can we do if the path is defined by an equation? This might require some guesswork. For example,
consider the planar ellipse Γ = {~r, x2 /9 + y 2 /4 = 1, z = 0}. Since it lies in the plane z = 0, we fix
a3 (t) = 0 and we look for a1 (t) and a2 (t). This must be functions of t that satisfy a1 (t)2 /9 + a2 (t)2 /4 = 1.
Since the ellipse is an affine deformation of the unit circumference, we may expect to use trigonometric
functions, and indeed we see that ~a(t) = 3 cos tı̂ + 2 sin t̂ satisfies the desired equation.
Try to understand well these simple examples (segments, graphs, circumferences): they will be used very often
during the course and the solution of many exercises will require their use.
Remark 1.25 (How to change parametrisation). Sometimes we have a curve ~a : I → R3 , and we want to
find a different curve ~b : J → R3 with the same path Γ, but defined on a different interval J ⊂ R. To this
purpose, it is enough to find a function g : J → I that is bijective and continuous and define ~b(τ ) := ~a(g(τ ))
for τ ∈ J. Two parametrisations ~a and ~b of the same path are always related one another by a function g of
this kind. If g is increasing, then the two parametrisations have the same orientation (they run the path in the
same direction); if g is decreasing, then the two parametrisations have the opposite orientation (they run the
path in opposite directions). We see a few examples.

• In the example above, the unit circle is parametrised by ~a(t) = cos t ı̂ + sin t ̂ for t ∈ I = [0, 2π) and
~b(τ ) = cos 2τ ı̂ + sin 2τ for τ ∈ J = [0, π), so we have g(τ ) = 2τ .

• Consider the unit half circle centred at the origin and located in the half plane {~r = xı̂ + y̂, y ≥ 0}, which
can be defined by either of the two parametrisations
p
~a : [0, π] → R3 , ~a(t) = cos tı̂ + sin t̂, ~b : [−1, 1] → R3 , ~b(τ ) = τı̂ + 1 − τ 2 ̂.

Here we use g = arccos : [−1, 1] → [0, π] (which is decreasing: ~a runs anti-clockwise, ~b clockwise).
• We can use a change of parametrisation to invert the orientation of a curve. If ~a is defined on the interval
I = [0, tF ], choosing g(τ ) = tF − τ we obtain ~b(τ ) = ~a(tF − τ ) which maps I → Γ (as ~a does) but with
opposite orientation.
In the special case of a segment, from ~a(t) = ~p + t(~q − ~p) = (1 − t)~p + t~q for t ∈ [0, 1] we obtain
~b(τ ) = ~q + τ (~p − ~q) = τ ~
p + (1 − τ )~q for τ ∈ [0, 1]. ~a runs from ~p to ~q, while ~b runs from ~q to ~p.

⋆ Remark 1.26 (Fields in physics). Fields are important in all branches of science and model many physical
quantities. For example, consider a domain D that models a portion of Earth’s atmosphere. One can
associate to every point ~r ∈ D several numerical quantities representing temperature, density, air pressure,
concentration of water vapour or some pollutant (at a given instant): each of these physical quantities can
be mathematically represented by a scalar field (a scalar quantity is associated to each point in space). Other
physical quantities involve magnitude and direction (which can both vary in different points in space), thus they
can be represented as vector fields: for example the gravitational force, the wind velocity, the magnetic field
(pointing to the magnetic north pole). The plots you commonly see in weather forecast are representations of
some of these fields (e.g. level sets of pressure at a given altitude). (Note that all these fields may vary in time,
so they are actually functions of four scalar variables: three spacial and one temporal.) Also curves are used in
physics, for instance to describe trajectories of electrons, bullets, aircraft, planets and any other kind of body.
A. Moiola, University of Reading 12 Vector calculus lecture notes, 2016–17

Warning: frequent error 1.27. At this point it is extremely important not to mix up the different definitions
of scalar fields, vector fields and curves. Treating vectors as scalars or scalars as vectors is one of the
main sources of mistakes in vector calculus exams! We recall that scalar fields take as input a vector
~ r)),
and return a real number (~r 7→ f (~r)), vector fields take as input a vector and return a vector (~r 7→ F(~
curves take as input a real number and return a vector (t 7→ ~a(t)):
Real functions (of real variable) f :R→R t 7→ f (t),
Curves ~a : R → R3 t 7→ ~a(t),
Increasing complexity
Scalar fields f : R3 → R ~r 7→ f (~r),
Vector fields ~ : R3 → R3
F ~ r).
~r 7→ F(~

Vector fields might be thought as combinations of three scalar fields (the components) and curves as com-
binations of three real functions.

1.3 Vector differential operators

We learned in the calculus class what is the derivative of a smooth real function of one variable f : R → R.
We now study the derivatives of scalar and vector fields. It turns out that there are several important
“differential operators”, which generalise the concept of derivative. In this section we introduce them,
while in the next one we study some relations between them. The fundamental differential operators for
fields are the partial derivatives, described in Section 1.3.1; all the vector differential operators described
in the rest of this section are defined starting from them.
As before, for simplicity we consider here the three-dimensional case only; all the operators, with the
relevant exception of the curl operator, can immediately be defined in Rn for any dimension n ∈ N.
The textbook [1] describes partial derivatives in Sections 12.3–5, the Jacobian matrix in 12.6, the
gradient and directional derivatives in 12.7, divergence and curl operator in 16.1 and the Laplacian in 16.2.
⋆ Remark 1.28 (What is an operator?). An operator is anything that operates on functions or fields, i.e.
a “function of functions” or “function of fields”. For example, we can define the “doubling operator” T
that, given a real function f : R → R, returns its double, i.e. T f is the function T f : R → R, such that
T f (x) = 2f (x).
d
A differential operator is an operator that involves some differentiation. For example, the derivative dx
d ′
maps f : R → R in dx f = f : R → R.
Operators and differential operators are rigorously defined using the concept of “function space”, which is
an infinite-dimensional vector space whose elements are functions. This is well beyond the scope of this class
and is one of the topics studied by functional analysis.

1.3.1 Partial derivatives

Consider a smooth scalar field f : D → R. The partial derivatives of f in the point ~r = xı̂ + y̂+ z k̂ ∈ D
are defined as limits of difference quotients, when these limits exist:
∂f f (x + h, y, z) − f (x, y, z)
(~r) := lim ,
∂x h→0 h
∂f f (x, y + h, z) − f (x, y, z)
(~r) := lim , (6)
∂y h→0 h
∂f f (x, y, z + h) − f (x, y, z)
(~r) := lim .
∂z h→0 h
In other words, ∂f r) can be understood as the derivative of the real function x 7→ f (x, y, z) which “freezes”
∂x (~
the y- and z-variables. If the partial derivatives are defined for all points in D, then they constitute three
scalar fields, denoted by ∂f ∂f ∂f
∂x , ∂y and ∂z . We can also write

∂f f (~r + hı̂) − f (~r) ∂f f (~r + h̂) − f (~r) ∂f f (~r + hk̂) − f (~r)

(~r) = lim , (~r) = lim , (~r) = lim .
∂x h→0 h ∂y h→0 h ∂z h→0 h
~
∂F ∂F1 ∂F2 ∂F3
We also use the notation ∂x := ∂x ı̂ + +
to denote the vector field whose components are
∂x ̂ ∂x k̂
~
~ (and similarly ∂ F ~
∂F
the partial derivatives with respect to x of the components of a vector field F ∂y and ∂z ).

Exercise 1.29. ◮ Compute all the partial derivatives of the following scalar fields:
xy p |~r|2
f (~r) = xyez , g(~r) = , h(~r) = log(1 + z 2 eyz ), ℓ(~r) = x2 + y 4 + z 6 , m(~r) = xy , p(~r) = .
y+z x2
A. Moiola, University of Reading 13 Vector calculus lecture notes, 2016–17

Since partial derivative are nothing else than usual derivatives for the functions obtained by freezing
all variables except one, the usual rules for derivatives apply. For f and g scalar fields, λ and µ ∈ R, and
a real function G : R → R, we have the following identities: linearity

∂(λf + µg) ∂f ∂g ∂(λf + µg) ∂f ∂g ∂(λf + µg) ∂f ∂g

=λ +µ , =λ +µ , =λ +µ ; (7)
∂x ∂x ∂x ∂y ∂y ∂y ∂z ∂z ∂z

the product rule (extending the well-known formula (F G)′ = F ′ G + F G′ for real functions)

∂(f g) ∂f ∂g ∂(f g) ∂f ∂g ∂(f g) ∂f ∂g

=g +f , =g +f , =g +f ; (8)
∂x ∂x ∂x ∂y ∂y ∂y ∂z ∂z ∂z

and the chain rule

∂ G(f ) ∂f ∂ G(f ) ∂f ∂ G(f ) ∂f
= G′ (f ) , = G′ (f ) , = G′ (f ) . (9)
∂x ∂x ∂y ∂y ∂z ∂z

Note that the product rule (8) must be used when we compute a partial derivative of a product ; the chain
rule (9) when we compute a partial derivative of a composition.4 In other words, in (8) the two scalar
fields f and g are multiplied to each other, in (9) the function G is evaluated in f (~r), which is a scalar:
∂(G(f ))
∂x (~r) = G′ (f (~r)) ∂f
∂x (~
r).
⋆ Remark 1.30. Note that to be able to define the partial derivatives of a field f in a point ~r we need to be
able to take the limits in (6), so to evaluate f “nearby” ~r. This is possible if f is defined in a open set D, as
defined in Section 1.1.3, which ensures that all its points are completely surrounded by points of the same set.
For example, the “open half space” D = {x > 0} is open, while the “closed half space” E = {x ≥ 0} is not.
If a field f is defined in E, we can not evaluate f (h, y, z) for negative h, so the limit limh→0 f (h,y,z)−f
h
(0,y,z)

in (6) is not defined and we cannot compute ∂f ∂x in all the points y̂ + z k̂ ∈ E (those with x = 0).

⋆ Remark 1.31. A scalar field f : D → R is called differentiable or “differentiable of class C 1 ” if it is

continuous (see Remark 1.22) and all its first-order partial derivatives exist and are continuous. A vector field
is differentiable (of class C 1 ) if its three components are differentiable.

1.3.2 The gradient

Let f be a scalar field. The vector field whose three components are the three partial derivatives of f (in
the usual order) is called gradient of f and denoted by

~ (~r) := grad f (~r) := ı̂ ∂f (~r) + ̂ ∂f (~r) + k̂ ∂f (~r).

∇f (10)
∂x ∂y ∂z

The symbol
~ := ı̂ ∂ + ̂ ∂ + k̂ ∂
∇ (11)
∂x ∂y ∂z
is called “nabla” operator, or “del”, and is usually denoted simply by ∇.
Given a smooth scalar field f and a unit vector û, the directional derivative of f in direction û is
~ (~r). If n̂ is the unit vector orthogonal to a surface, ∂f := ∂f
defined as the scalar product ∂∂fû (~r) := û · ∇f ∂n ∂ n̂
is called normal derivative.
Example 1.32. The gradient of f (~r) = x2 − y 2 is

~ (~r) = ı̂ ∂ (x2 − y 2 ) + ̂ ∂ (x2 − y 2 ) + k̂ ∂ (x2 − y 2 ) = 2xı̂ − 2y̂;

∇f
∂x ∂y ∂z
thus the vector field in Figure 7 is the gradient of the scalar field depicted in Figure 5.

4 We recall the notion of composition of functions: for any three sets A, B, C, and two functions F : A → B and

G : B → C, the composition of G with F is the function (G ◦ F ) : A → C obtained by applying F and then G to the obtained
output. In formulas: (G ◦ F )(x) := G(F (x)) for all x ∈ A. If A = B = C = R (or they are appropriate subsets of R) and F
and G are differentiable, then from basic calculus we know the derivative of the composition: (G ◦ F )′ (x) = G′ (F (x))F ′ (x)
(this is the most basic example of chain rule).
A. Moiola, University of Reading 14 Vector calculus lecture notes, 2016–17

Proposition 1.33 (Properties of the gradient). Given two smooth scalar fields f, g : D → R, their gradients
satisfy the following properties:
1. for any constant λ, µ ∈ R, the following identity holds (linearity)
~
∇(λf ~ + µ∇g;
+ µg) = λ∇f ~ (12)

2. the following identity holds (product rule or Leibniz rule)

~ g) = g ∇f
∇(f ~ + f ∇g;
~ (13)

3. for any differentiable real function G : R → R the chain rule holds:

~
∇(G ◦ f )(~r) = ∇G ~ (~r),
~ f (~r) = G′ f (~r) ∇f (14)

where G′ (f (~r)) is the derivative of G evaluated in f (~r) and G ◦ f denotes the composition of G with f ;
~ (~r) is perpendicular to the level surface of f passing through ~r (i.e. to {~r′ ∈ D s.t. f (~r′ ) = f (~r)});
4. ∇f
~ (~r) points in the direction of maximal increase of f .
5. ∇f
Proof of 1., 2. and 3. We use the definition of the gradient (10), the linearity (7), the product rule (8)
and the chain rule for partial derivatives (9):

~ ∂(λf + µg) ∂(λf + µg) ∂(λf + µg)

∇(λf + µg) = ı̂ + ̂ + k̂
∂x ∂y ∂z
(7) ∂f ∂g ∂f ∂g ∂f ∂g ~ + µ∇g;
~
= ı̂λ + ı̂µ + ̂λ + ̂µ + k̂λ + k̂µ = λ∇f
∂x ∂x ∂y ∂y ∂z ∂z
~ g) = ı̂ ∂(f g) ∂(f g) ∂(f g)
∇(f + ̂ + k̂
∂x ∂y ∂z
     
(8) ∂f ∂g ∂f ∂g ∂f ∂g ~ + f ∇g,
~
= ı̂ g +f + ̂ g +f + k̂ g +f = g ∇f
∂x ∂x ∂y ∂y ∂z ∂z
~ ∂(G ◦ f ) ∂(G ◦ f ) ∂(G ◦ f )
∇(G ◦ f )(~r) = ı̂ (~r) + ̂ (~r) + k̂ (~r)
∂x ∂y ∂z
(9)
∂f
∂f
∂f

= ı̂G′ f (~r) (~r) + ̂G′ f (~r) (~r) + k̂G′ f (~r) (~r) ~ (~r).
= G′ f (~r) ∇f
∂x ∂y ∂z
Parts 4. and 5. will be proved in Remarks 1.88 and 1.89.

Exercise 1.34. ◮ Write the gradients of the scalar fields in Exercise 1.29.
Exercise 1.35. ◮ Verify that the gradients of the “magnitude scalar field” m(~r) = |~r| and of its square
s(~r) = |~r|2 satisfy the identities

~ r) = ∇(|~
∇s(~ ~ r|2 ) = 2~r, ∇m(~ ~ r|) = ~r
~ r) = ∇(|~ ∀~r ∈ R, ~r 6= ~0. (15)
|~r|
~ r|α ) for a general α ∈ R?
Represent the two gradients as in Figure 7. Can you compute ∇(|~
Comparison with scalar calculus 1.36. We know from first year calculus that we can infer information on
a function of real variable from its derivative. Many of the relations between functions and their derivatives
carry over to scalar fields and their gradients, we summarise some of them in the following table. (Recall that
“affine” is equivalent to “polynomial of degree at most one”.)
F : R → R is a function of real variable: f : R3 → R is a scalar field:
F ′ = 0 ⇐⇒ F is constant (F (t) = λ, ∀t ∈ R), ~ = ~0 ⇐⇒ f is constant (f (~r) = λ, ∀~r ∈ R3 ),
∇f
F ′ = λ ⇐⇒ F is affine (F (t) = tλ + µ), ~ = ~u ⇐⇒ f is affine (f (~r) = ~r · ~u + µ),
∇f
F ′ is a polynomial of degree p the components of ∇f ~ are polynomials of degree p
⇐⇒ F is a polynomial of degree p + 1, ⇐⇒ f is a polynomial of degree p + 1 in x, y, z,
F′ > 0 ~ > 0 ⇐⇒ f is increasing in direction n̂
n̂ · ∇f
⇐⇒ F is increasing (F (s) < F (t), s < t), (f (~r + hn̂) > f (~r), h > 0, here n̂ is a unit vector),
t0 is a local maximum/minumum for F ~r0 is a local maximum/minumum for f
⇒ F ′ (t0 ) = 0, ⇒ ∇f ~ (~r0 ) = ~0.
A. Moiola, University of Reading 15 Vector calculus lecture notes, 2016–17

y
2
−2 −3
−3 −2 1
− 0
1.5
−1 −2
0
−1

1
1

3
0.5

2
0
0 x
0

1
1
−0.5

3
3

−1 −1

2
2

0
−1 −2
−1.5 1
1
−3 −1 0
−2 −3 −2
−2
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2

Figure 9: A combined representation of the two-dimensional scalar field f (~r) = x2 − y 2 (coloured level
curves) and its gradient ∇f ~ (~r) = 2xı̂ − 2y̂ (thin black streamlines). Note that the streamlines are perpen-
dicular to the level lines in every point and they point from areas with lower values of f (blue) to areas with
higher values (red), compare with Proposition 1.33. (See the coloured plot in the .pdf file of this notes.)

Exercise 1.37. ◮ Consider the scalar field f = (2x + y)3 and the vector field G ~ = 2ez ı̂ + ez ̂. Verify that,
3 ~ ~
at every point ~r ∈ R , the level set of f is orthogonal to G, but G is not the gradient of f . Can you find a
formula relating G~ to ∇f
~ ?
Hint: figure out the exact shape of the level sets of f , compute two (linearly independent) tangent vectors
to the level set in each point, and verify that both these vectors are orthogonal to G. ~ Alternatively, use the
relation between G ~ and ∇f
~ .

Example 1.38 (Finding the unit normal to a surface). Consider the surface
n p o
S = ~r ∈ R3 s.t. y = 1 + x2

and compute the field n̂ : S → R3 composed of the unit vectors perpendicular to S and pointing upwards as
in the left plot of Figure 10.
From Proposition 1.33(4), we know that the gradient of a field that admits S as level set indicates the
direction perpendicular to S itself. First, we find a scalar field f such that S is given as f (~r) =constant: we
can choose f (~r) = x2 − y 2 , which satisfies f (~r) = −1 for all ~r ∈ S. Taking the partial derivatives of f we
immediately see that its gradient is
p
~ (~r) = 2xı̂ − 2y̂,
∇f ⇒ ~ | = 2 x2 + y 2 .
|∇f
~ (~r), i.e.
Thus for every ~r ∈ S, n̂(~r) is one of the two unit vectors parallel to ∇f
~ (~r)
∇f xı̂ − y̂
n̂(~r) = ± = ±p .
~ (~r)|
|∇f x2 + y 2

From the left plot in Figure 10 we see that we want to choose the sign such that ~n(0, 1, 0) = ̂, thus we choose
the minus sign in the last equation above and conclude
−xı̂ + y̂
n̂(~r) = p .
x2 + y 2

~ r) = xı̂ + ̂ be a scalar and a vector field defined in R3 .

Exercise 1.39. ◮ Let f (~r) = ex − sin y + xz 2 and F(~
~ (This is a directional derivative whose direction
Compute the directional derivative of f in the direction of F.
may be different at every point).
A. Moiola, University of Reading 16 Vector calculus lecture notes, 2016–17

y 5

n̂ 3
√
y = 1 + x2 2

n̂ 1

0
n̂
-1
-2

x 0 -2
-1
0
2 1
2

Figure 10: Left: The section in the xy-plane (z = 0) of the surface S of Example 1.38 and its unit normal
vectors.
Right: The surface graph of the planar scalar field f = x2 +y 2 and its tangent plane at ~r0 +f (~r0 )k̂ = ̂+ k̂
~ (~r0 ) · (~r −~r0 ) = 1 + (2̂) · (xı̂ + y̂ − ̂) = 2y − 1.
for ~r0 = ̂. The tangent plane has equation z = f (~r0 ) + ∇f

⋆ Remark 1.40. The gradient of a scalar field f at a given position ~r0 can be defined as the best linear
approximation of f near ~r0 . More precisely, for a differentiable scalar field f : D → R and a point ~r0 ∈ D, the
~ (~r0 ) is the unique vector L
gradient ∇f ~ such that

~ · (~r − ~r0 ) + |~r − ~r0 |g(~r)

f (~r) = f (~r0 ) + L where lim g(~r) = 0 and ~r ∈ D.
~r→~r0

We know from first-year calculus that, given a differentiable real function F : R → R, the tangent line to
its graph at a point (t0 , F (t0 )) is t 7→ F (t0 ) + F ′ (t0 )(t − t0 ) and the derivative F ′ (t0 ) represents its slope.
The tangent line can be defined as the line that best approximates F near t0 . Similarly, the affine (i.e. “flat”)
field that best approximates f near ~r0 is ~r 7→ f (~r0 ) + ∇f ~ (~r0 ) · (~r − ~r0 ). If the field f is two-dimensional,
~ (~r0 ) · (~r −~r0 )} is the tangent plane to the surface graph of f (see Remark 1.21), an example
{z = f (~r0 ) + ∇f
is in the right plot in Figure 10. If the field is three-dimensional, this is the tangent space to the graph of f ,
which now lives in a four-dimensional space, so it is hard to visualise.

1.3.3 The Jacobian matrix

We have seen that a scalar field has three first-order partial derivatives that can be collected in a vector
field (the gradient). A vector field has nine partial derivatives, three for each component (which is a scalar
field). To represent them compactly we collect them in a matrix.
Consider a smooth vector field F, ~ the Jacobian matrix J F ~ (or simply the Jacobian, named after
Carl Gustav Jacob Jacobi 1804–1851) of F is ~
 
∂F1 ∂F1 ∂F1
 ∂x ∂y ∂z 
 
 ∂F ∂F2 ∂F2 
~ := 
JF 
2 
. (16)
 ∂x ∂y ∂z 
 
 ∂F3 ∂F3 ∂F3 
∂x ∂y ∂z

This is a field whose values at each point are 3 × 3 matrices. Note that sometimes the noun “Jacobian”
is used to indicate the determinant of the Jacobian matrix.
Exercise 1.41. ◮ Compute the Jacobian of the following vector fields
~ r) = 2xı̂ − 2y̂,
F(~ ~ r) = yzı̂ + xz̂ + xy k̂,
G(~ ~ r) = |~r|2 ı̂ + cos y k̂.
H(~

1.3.4 Second-order partial derivatives, the Laplacian and the Hessian

The second-order partial derivatives of a scalar field f are the partial derivatives of the partial derivatives
of f . When partial derivatives are taken twice with respect to the same component of the position, they
A. Moiola, University of Reading 17 Vector calculus lecture notes, 2016–17

are denoted by
∂2f ∂ ∂f ∂2f ∂ ∂f ∂2f ∂ ∂f
= , = and = .
∂x2 ∂x ∂x ∂y 2 ∂y ∂y ∂z 2 ∂z ∂z
When they are taken with respect to two different components, they are called “mixed derivatives” and
denoted by

∂2f ∂ ∂f ∂2f ∂ ∂f ∂2f ∂ ∂f

= , = , = ,
∂x∂y ∂x ∂y ∂x∂z ∂x ∂z ∂y∂x ∂y ∂x
∂2f ∂ ∂f ∂2f ∂ ∂f ∂2f ∂ ∂f
= , = , and = .
∂y∂z ∂y ∂z ∂z∂x ∂z ∂x ∂z∂y ∂z ∂y

If f is smooth, the order of differentiation is not relevant (Clairault’s or Schwarz theorem):

∂2f ∂2f ∂2f ∂2f ∂2f ∂2f

= , = and = . (17)
∂x∂y ∂y∂x ∂x∂z ∂z∂x ∂y∂z ∂z∂y
3
⋆ Remark 1.42. It is clear that we can define partial derivatives of higher order, e.g. ∂x∂ 2 f∂y = ∂x
∂ ∂ ∂f
∂x ∂y . A
k
scalar field f : D → R is called “differentiable of class C ” with k ∈ N, if it is continuous (see Remark 1.22),
all its partial derivatives of order up to k exist and are continuous. It is called “differentiable of class C ∞ ” if
all its partial derivatives of any order exist and are continuous.
The Laplacian ∆f (or Laplace operator, sometimes denoted by ∇2 f , named after Pierre-Simon
Laplace 1749–1827)5 of a smooth scalar field f is the scalar field obtained as sum of the pure second
partial derivatives of f :
∂2f ∂2f ∂2f
∆f := ∇2 f := 2
+ 2 + 2. (18)
∂x ∂y ∂z
A scalar field whose Laplacian vanishes everywhere is called harmonic function.
⋆ Remark 1.43 (Applications of the Laplacian). The Laplacian is used in some of the most studied partial
differential equations (PDEs) such as the Poisson equation ∆u = f and the Helmholtz equation −∆u−k 2 u = f .
Therefore, it is ubiquitous and extremely important in physics and engineering, e.g. in the models of diffusion
of heat or fluids, electromagnetism, acoustics, wave propagation, quantum mechanics. In the words of T. Tao6 :
“The Laplacian of a function u at a point x measures the average extent to which the value of u at x deviates
from the value of u at nearby points to x (cf. the mean value theorem for harmonic functions). As such, it
naturally occurs in any system in which some quantity at a point is influenced by the value of the same quantity
at nearby points.”
The Hessian Hf (or Hessian matrix, named after Ludwig Otto Hesse 1811–1874) is the 3 × 3 matrix
field of the second derivatives of the scalar field f :
 
∂2f ∂2f ∂2f
 ∂x2 ∂y∂x ∂z∂x 
 
 2 
 ∂ f ∂2f ∂2f 
~
Hf := J(∇f ) =  . (19)
 ∂x∂y ∂y 2 ∂z∂y 
 
 2 
 ∂ f ∂2f ∂ f 
2

∂x∂z ∂y∂z ∂z 2

Note that, thanks to Clairault’s theorem (17), (if the field f is smooth enough) the Hessian is a symmetric
matrix in every point. The Laplacian equals the trace of the Hessian, i.e. the sum of its diagonal terms:

∆f = Tr(Hf ). (20)

~ is the Laplace operator applied componentwise to vector fields:

The vector Laplacian ∆

~F
∆ ~ := (∆F1 )ı̂ + (∆F2 )̂ + (∆F3 )k̂ (21)
 ∂2F ∂ 2 F1 ∂ 2 F1   ∂2F ∂ 2 F2 ∂ 2 F2   ∂2F ∂ 2 F3 ∂ 2 F3 
1 2 3
= + + ı̂ + + + ̂ + + + k̂.
∂x2 ∂y 2 ∂z 2 ∂x2 ∂y 2 ∂z 2 ∂x2 ∂y 2 ∂z 2
5 The reason for the use of the notation ∇2 will be slightly more clear after equation (24); this symbol is mainly used by

physicists and engineers.

6 http://mathoverflow.net/questions/54986
A. Moiola, University of Reading 18 Vector calculus lecture notes, 2016–17

Exercise 1.44. ◮ Compute the Laplacian and the Hessian of the following scalar fields:

f (~r) = x2 − y 2 , g(~r) = xyez , h(~r) = |~r|4 .

Warning: frequent error 1.45. Once again, we remind that symbols cannot be moved around freely in
∂g ∂f ∂g ∂ 2 (f g) ∂2 f
mathematical expressions. For example, ∂f ∂x ∂y , ∂y ∂x , ∂x∂y and g ∂x∂y , are all different fields and must not
be confused with each other. Equation (17) only says that we can swap the order of derivation for the same
∂2f ∂2f
field, not between two fields, i.e. ∂x∂y = ∂y∂x but ∂f ∂g ∂f ∂g ∂f g
∂x ∂y 6= ∂y ∂x . Moreover ∂x does not mean anything: is
∂(f g)
it supposed to be ∂x or ( ∂f
∂x )g? All these errors appears too often in exams, watch out!

1.3.5 The divergence operator

~ its divergence is defined as the the trace of its Jacobian matrix:
Given a differentiable vector field F,

~ ·F
∇ ~ := div F ~ = ∂F1 + ∂F2 + ∂F3 .
~ := Tr(J F) (22)
∂x ∂y ∂z

~ ·F
The notation ∇ ~ is justified by the fact that the divergence can be written as the formal scalar product
between the nabla symbol ∇ ~
~ and the field F:
   
~ = ı̂ ∂ + ̂ ∂ + k̂ ∂ · F1 ı̂ + F2 ̂ + F3 k̂ .
~ ·F
∇
∂x ∂y ∂z
Exercise 1.46. ◮ Compute the divergence of the following vector fields:
~ r) = 2xı̂ − 2y̂,
F(~ ~ r) = yzı̂ + xz̂ + xy k̂,
G(~ ~ r) = |~r|2 ı̂ + cos y k̂.
H(~

What is the intuitive meaning of the divergence of a vector field F? ~ Very roughly speaking, the
~
divergence measures the “spreading” of F around a point ~r, i.e. the difference between the amount of F ~
~ ~
exiting from an infinitesimally small ball centred at ~r and the amount entering it. If ∇ · F is positive
in ~r we may expect F ~ to spread or expand near ~r, while if it is negative F
~ will shrink. Of course this
is explanation is extremely hand-waving: we will prove a precise characterisation of the meaning of the
divergence in Remark 3.33.
~ = xı̂ + y̂ has positive divergence ∇
Example 1.47 (Positive and negative divergence). The field F ~ ·F
~ =
1 + 1 + 0 = 2 > 0, thus we see in the left plot of Figure 11 that it is somewhat spreading. The field
~ = (−x − 2y)ı̂ + (2x − y)̂ has negative divergence ∇
G ~ ·G
~ = −1 − 1 + 0 = −2 < 0, and we see from the
right plot that the field is converging.

2
2

1.5
1.5

1
1

0.5 0.5

0 0

−0.5 −0.5

−1
−1

−1.5
−1.5

−2
−2
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 −2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5

~ = xı̂ + y̂ (left) and G

Figure 11: A representation of the fields F ~ = (−x − 2y)ı̂ + (2x − y)̂ (right) from
~ ~
Example 1.47. F has positive divergence and G negative.

⋆ Remark 1.48. From the previous example it seems to be possible to deduce the sign of the divergence
of a vector field from its plot, observing whether the arrows “converge” or “diverge”. However, this can be
misleading, as also the magnitude (the length of the arrows) matters. For example, the fields F ~ a = |~r|−a~r
defined in R \ {~0}, where a is a real positive parameter, have similar plots but they have positive divergence
3

if a < 3 and negative if a > 3. (Can you show this fact? It is not easy!)
A. Moiola, University of Reading 19 Vector calculus lecture notes, 2016–17

1.6

1.4

1.2

0.8

0.6

0.4

0.2

−0.2

−0.4

−0.5 0 0.5 1 1.5

Figure 12: The field F~ = (x2 − x)ı̂ + (y 2 − y)̂ has divergence ∇ ~ = 2(x + y − 1) which is positive above
~ ·F
~
the straight line x + y = 1 and negative below. If F represents the velocity of a fluid, the upper part of the
space acts like a source and the lower part as a sink.

Warning: frequent error 1.49. Since strictly speaking ∇~ is not a vector (11), the symbol ∇
~ ·F
~ does not
represent a scalar product and7
~ ·F
∇ ~ 6= F
~ · ∇.
~
~ in a mathematical expression!
Recall: you can not move around the nabla symbol ∇

1.3.6 The curl operator

~ (often denoted “curl” and more rarely
The last differential operator we define is the curl operator ∇×
“rot” and called rotor or rotational), which maps vector fields to vector fields:
 
 ı̂ ̂ k̂  
~ := curl F
~ ×F

~ := ∂ ∂ ∂  ∂F3 ∂F2   ∂F
1 ∂F3   ∂F
2 ∂F1 
∇  ∂x ∂y ∂z  = − ı̂ + − ̂ + − k̂. (23)
F ∂y ∂z ∂z ∂x ∂x ∂y
1 F2 F 
3

As in the definition of the vector product (2), the matrix determinant is purely formal, since it contains
~ ×F
vectors, differential operators and scalar fields. Again, ∇ ~ 6= −F~ × ∇,
~ as the left-hand side is a vector
field while the right-hand side is a differential operator.
Among all the differential operators introduced so far, the curl operator is the only one which can be
defined only in three dimensions, since it is related to the vector product.
Exercise 1.50. ◮ Compute the curl of the following vector fields:
~ r) = 2xı̂ − 2y̂,
F(~ ~ r) = yzı̂ + xz̂ + xy k̂,
G(~ ~ r) = |~r|2 ı̂ + cos y k̂.
H(~

How can we interpret the curl of a field? The curl is in some way a measure of the “rotation” of the
field. If we imagine to place a microscopic paddle-wheel at a point ~r in a fluid moving with velocity F, ~
it will rotate with angular velocity proportional to |∇~ × F(~
~ r)| and rotation axis parallel to ∇ ~ × F(~
~ r). If
the fluid motion is planar, i.e. F3 = 0, and anti-clockwise, the curl will point out of page, if the motion is
clockwise it will point into the page (as the motion of a usual screw). See also Figure 13.
~ ·F
7 Indeed ∇ ~ and F ~ ·∇~ are two very different mathematical objects: ∇
~ ·F
~ is a scalar field, while F
~ ·∇
~ is an operator
that maps scalar fields to scalar fields
 ∂ ∂ ∂  ∂f ∂f ∂f
~ · ∇)f
(F ~ = F1 + F2 + F3 f = F1 + F2 + F3 .
∂x ∂y ∂z ∂x ∂y ∂z
~ = yzı̂ + xz̂ + xy k̂ and g = xyez , then
For example if G
 ∂ ∂ ∂ 
~ · ∇)g
(G ~ = yz + xz + xy g = yz yez + xz xez + xy xyez = (x2 z + y 2 z + x2 y 2 )ez .
∂x ∂y ∂z
The operator F~ ·∇~ can also be applied componentwise to a vector field (and in this case it returns a vector field, see (35)).
~ ·∇
F ~ is often called advection operator; when |F| ~ = 1, then it corresponds to the directional derivative we have already
encountered in Section 1.3.2. We will use the advection operator only in Remark 1.56; in any other situation in this module
you should not use the symbol (F~ · ∇).
~
A. Moiola, University of Reading 20 Vector calculus lecture notes, 2016–17

2
2

1.5
1.5

1
1

0.5
0.5

0
0

−0.5
−0.5

−1
−1

−1.5
−1.5

−2
−2
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 −2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5

Figure 13: The planar field F ~ = 2yı̂ − x̂ depicted in the left plot has curl equal to ∇
~ ×F ~ = −3k̂ which
~ ~
“points into the page”, in agreement with the clockwise rotation of F. The planar field G = x2 ̂ depicted
~ ×G
in the right plot has curl equal to ∇ ~ = 2xk̂ which points “into the page” for x < 0 and “out of the
page” for x > 0. Indeed, if we imagine to immerse some paddle-wheels in the field, due to the differences
in the magnitude of the field on their two sides, they will rotate clockwise in the negative-x half-plane and
anti-clockwise in the positive-x half-plane.

⋆ Remark 1.51 (Linearity of differential operators). All operators introduced in this section (partial derivatives,
gradient, Jacobian, Laplacian, Hessian, divergence, curl) are linear operators. This means that, denoting D any
of these operators, for all λ, µ ∈ R and for any two suitable fields F and G (either both scalar or both vector
fields, depending on which operator is considered), we have the identity D(λF + µG) = λD(F ) + µD(G).

1.4 Vector differential identities

In the following two propositions we prove several identities involving scalar and vector fields and the
differential operators defined so far. It is fundamental to apply each operator to a field of the appropriate
kind, for example we can not compute the divergence of a scalar field because we have not given any
meaning to this object. For this reason, we summarise in Table 1 which types of fields are taken as input
and returned as output by the various operators.

Operator name symbol field taken as input field returned as output order
∂ ∂ ∂
Partial derivative ∂x , ∂y or ∂z scalar scalar 1st
Gradient ~
∇ scalar vector 1st
Jacobian J vector matrix 1st
Laplacian ∆ scalar scalar 2nd
Hessian H scalar matrix 2nd
Vector Laplacian ~
∆ vector vector 2nd
Divergence ~ (or div)
∇· vector scalar 1st
Curl (only 3D) ~ (or curl)
∇× vector vector 1st

Table 1: A summary of the differential operators defined in Section 1.3. The last column shows the order
of the partial derivatives involved in each operator.

1.4.1 Second-order vector differential identities

Next proposition shows the result of taking the divergence or the curl of gradients and curls.
~ be a vector
Proposition 1.52 (Vector differential identities for a single field). Let f be a scalar field and F
3
field, both defined in a domain D ⊂ R . Then the following identities hold true:

∇~ · (∇f
~ ) = ∆f, (24)
~ · (∇
∇ ~ × F)
~ = 0, (25)
~ × (∇f
∇ ~ ) = ~0, (26)
~ × (∇
∇ ~ × F)
~ = ∇(
~ ∇~ · F)
~ −∆
~ F.
~ (27)
A. Moiola, University of Reading 21 Vector calculus lecture notes, 2016–17

Proof. The two scalar identities (24) and (25) can be proved using only the definitions of the differential
~ )1 and (∇×
operators involved (note that we write (∇f ~ F)
~ 1 to denote the x-components of the corresponding
vector fields, and similarly for y and z):




~ )1
(22) ∂ (∇f ~ )2
∂ (∇f ~ )3
∂ (∇f
~ ~
∇ · (∇f ) = + +
∂x ∂y ∂z
(10) ∂ ∂f ∂ ∂f ∂ ∂f
= + +
∂x ∂x ∂y ∂y ∂z ∂z
(18)
= ∆f,




(22) ∂ (∇ ~ × F)
~ 1 ∂ (∇ ~ × F)
~ 2 ∂ (∇~ × F)~ 3
~ ~ ~
∇ · (∇ × F) = + +
∂x ∂y ∂z
     
∂F3 ∂F2 ∂F1 ∂F3 ∂F2 ∂F1
(23)
∂ ∂y − ∂z ∂ ∂z − ∂x ∂ ∂x − ∂y
= + +
∂x ∂y ∂z
∂ 2 F1 ∂ 2 F1 ∂ 2 F2 ∂ 2 F2 ∂ 2 F3 ∂ 2 F3
= − + − + −
∂y∂z ∂z∂y ∂z∂x ∂x∂z ∂x∂y ∂y∂x
(17)
= 0.

An alternative proof of (24) uses the facts that the divergence is the trace of the Jacobian matrix (22),
the Jacobian matrix of a gradient is the Hessian (19), and the trace of the Hessian is the Laplacian (20):

~ ) (22)
~ · (∇f
∇ ~ ) (19)
= Tr(J ∇f
(20)
= Tr(Hf ) = ∆f.

The two remaining vector identities can be proved componentwise, i.e. by showing that each of the
three components of the left-hand side is equal to the corresponding component of the right-hand side:



~ )3

(23) ∂ (∇f ~ )2
∂ (∇f
~ ~
∇ × (∇f ) 1 = −
∂y ∂z
(10) ∂ ∂f ∂ ∂f
= −
∂y ∂z ∂z ∂y
(17)
= 0,




(23) ∂ (∇ ~ × F)
~ 3 ~ × F)
∂ (∇ ~ 2
~ ~ ~
∇ × (∇ × F) 1 = −
∂y ∂z
   
(23) ∂ ∂F2 ∂F1 ∂ ∂F1 ∂F3
= − − −
∂y ∂x ∂y ∂z ∂z ∂x
  2 2
(17) ∂ ∂F2 ∂F3 ∂ F1 ∂ F1
= + − −
∂x ∂y ∂z ∂y 2 ∂z 2
  2
∂ ∂F1 ∂F2 ∂F3 ∂ F1 ∂ 2 F1 ∂ 2 F1
= + + − − −
∂x ∂x ∂y ∂z ∂x2 ∂y 2 ∂z 2
(22),(18) ∂
= (∇~ · F)
~ − ∆F1
∂x
(10),(21)
 
= ~ ∇
∇( ~ · F)
~ −∆ ~F~ ,
1

and similarly for the y- and z-components.

Note that, when we prove a certain vector identity, even if we consider only the first component of the
equality, this may involve all components of the fields involved and partial derivatives with respect to all
variables; see for instance the proofs of (26) and (27).
⋆ Remark 1.53. Since in all the identities of Proposition 1.52 we take second derivatives, the precise assump-
~ are differentiable of class C 2 .
tions for this proposition are that f and (the components of) F

1.4.2 Vector product rules

Comparison with scalar calculus 1.54. When we compute the derivative of the product of two functions of real
variable we apply the well-known product rule (f g)′ = f ′ g + g ′ f . When we calculate partial derivatives or the
gradient of the product of two scalar fields, similar rules apply, see Equations (8) and (13). But scalar and vector
fields may be multiplied in several different ways (scalar product, vector product and scalar-vector product)
A. Moiola, University of Reading 22 Vector calculus lecture notes, 2016–17

and other differential operators may be applied to these products (divergence, curl, Laplacian,. . . ). In these
cases, the product rules obtained are slightly more complicated; they are described in next Proposition 1.55
(and Remark 1.56). Note the common structure of all these identities: a differential operator applied to a
product of two fields (at the left-hand side) is equal to the sum of some terms containing the derivative of the
first field only and some terms containing derivatives of the second field only (at the right-hand side).
A special identity is (32): since the Laplacian contains second-order partial derivatives, all terms at the
right-hand side contain two derivatives, exactly as in the 1D product rule for second derivatives (f g)′′ =
f ′′ g + 2f ′ g ′ + f g ′′ .
Proposition 1.55 (Vector differential identities for two fields, product rules for differential operators).
~ and G
Let f and g be scalar fields, F ~ be vector fields, all of them defined in the same domain D ⊂ R3 . Then
8
the following identities hold true :
~ g) = f ∇g
∇(f ~ + g ∇f,
~ (28)
∇ ~ = (∇f
~ · (f G) ~ + f∇
~ )·G ~
~ · G, (29)
~ · (F
∇ ~ × G)
~ = (∇
~ × F)
~ ·G~ −F
~ · (∇
~ × G),
~ (30)
~ × (f G)
∇ ~ = (∇f
~ )×G ~ + f∇
~ × G,
~ (31)
~ · ∇g
∆(f g) = (∆f )g + 2∇f ~ + f (∆g). (32)

Proof. Identity (28) has already been proven in Proposition 1.33. We show here the proof of identities
(30) and (32) only; identities (29) and (31) can be proven with a similar technique. All the proofs use only
the definitions of the differential operators and of the vector operations (scalar and vector product), the
product rule (8) for partial derivatives, together with some smart rearrangements of the terms. In some
cases, it is convenient to start the proof from the expression at the right-hand side, or even to expand
both sides and match the results obtained. The vector identity (31) is proven componentwise (see the
similar proof of (33) below).
Identity (30) can be proved as follows:




(22) ∂ (F ~ × G)
~ 1 ~ × G)
∂ (F ~ 2 ~ × G)
∂ (F ~ 3
~ ~ ~
∇ · (F × G) = + +
∂x ∂y ∂z




(2) ∂ F G
2 3 − F G
3 2 ∂ F G
3 1 − F G
1 3 ∂ F1 G2 − F2 G1
= + +
∂x ∂y ∂z
(8) ∂F2 ∂G3 ∂F3 ∂G2 ∂F3 ∂G1 ∂F1 ∂G3
= G3 + F2 − G2 − F3 + G1 + F3 − G3 − F1
∂x ∂x ∂x ∂x ∂y ∂y ∂y ∂y
∂F1 ∂G2 ∂F2 ∂G1
+ G2 + F1 − G1 − F2 (now collect non-derivative terms)
 ∂z  ∂z  ∂z  ∂z  
∂F3 ∂F2 ∂F1 ∂F3 ∂F2 ∂F1
= − G1 + − G2 + − G3
∂y ∂z ∂z ∂x ∂x ∂y
     
∂G3 ∂G2 ∂G1 ∂G3 ∂G2 ∂G1
− F1 − − F2 − − F3 −
∂y ∂z ∂z ∂x ∂x ∂y
(23)
~ 1 G1 + (∇
~ × F)
= (∇ ~ 2 G2 + (∇
~ × F) ~ 3 G3 − F1 (∇
~ × F) ~ 1 − F2 (∇
~ × G) ~ 2 − F3 (∇
~ × G) ~ 3
~ × G)
(1)
~ × F)
= (∇ ~ ·G
~ −F
~ · (∇
~ × G).
~

Identity (32) follows from a repeated application of the product rule (8):

(18) ∂ 2 (f g) ∂ 2 (f g) ∂ 2 (f g)
∆(f g) = + +
∂x2 ∂y 2 ∂z 2
     
(8) ∂ ∂f ∂g ∂ ∂f ∂g ∂ ∂f ∂g
= g+f + g+f + g+f
∂x ∂x ∂x ∂y ∂y ∂y ∂z ∂z ∂z
 2 2
  2 2
  2 
(8) ∂ f ∂f ∂g ∂ g ∂ f ∂f ∂g ∂ g ∂ f ∂f ∂g ∂2g
= g+2 +f 2 + g+2 +f 2 + g+2 +f 2
∂x2 ∂x ∂x ∂x ∂y 2 ∂y ∂y ∂y ∂z 2 ∂z ∂z ∂z
(18),(10),(1)
= ~ · ∇g
(∆f )g + 2∇f ~ + f (∆g).
8 You do not need to remember these identities by hearth, even though are simple, but you need to be able to use them

and to derive them.

⋆ Here the precise assumptions for the proposition are that all fields are differentiable of class C 1 , except for identity (32)
where class C 2 is required.
A. Moiola, University of Reading 23 Vector calculus lecture notes, 2016–17

The following is an alternative proof of identity (32) that uses the vector identities previously proved:
(24)

∆(f g) = ∇·~ ∇(f
~ g)
(28)
~
= ∇·(f ~ + g ∇f
∇g ~ )
(12)
~
= ∇·(f ~ + ∇·(g
∇g) ~ ∇f ~ )
(29)
~ ) · (∇g)
= (∇f ~ + f∇
~ · (∇g)
~ + (∇g)
~ · (∇f
~ ) + g∇
~ · (∇f
~ )
(24)
~ · ∇g
= (∆f )g + 2∇f ~ + f (∆g).

⋆ Remark 1.56. In Proposition 1.55 we found expression (28) for the gradient of the product of two scalar
fields. What about the gradient of the scalar product of two vector fields? We also computed the divergence
(29) and the curl (31) of the product of a vector and a scalar field, and the divergence of the vector product
between two vector fields (30). What about the curl of a vector product? The answer two these two questions
is given by the following two vector product rules:
~ F
∇( ~ · G)
~ = (F
~ · ∇)
~ G~ + (G
~ · ∇)
~ F~ +G~ × (∇~ × F)
~ +F ~ × (∇
~ × G),
~ (33)
~ × G)
~ × (F
∇ ~ = (∇ ~ F
~ · G) ~ − (∇ ~ G
~ · F) ~ + (G
~ · ∇)
~ F~ − (F
~ · ∇) ~
~ G. (34)

In identities (33) and (34) we use the advection operator (which was mentioned in Remark 1.49) applied to
a vector field:
~ · ∇)
(G ~ F~ = (G~ · ∇)F
~ 1 ı̂ + (G
~ · ∇)F
~ 2 ̂ + (G
~ · ∇)F
~ 3 k̂ (35)
 ∂F ∂F ∂F   ∂F ∂F ∂F   ∂F ∂F ∂F 
1 1 1 2 2 2 3 3 3
= G1 + G2 + G3 ı̂ + G1 + G2 + G3 ̂ + G1 + G2 + G3 k̂.
∂x ∂y ∂z ∂x ∂y ∂z ∂x ∂y ∂z
Proof of (33). We prove (33) componentwise, starting from the first component of the expression at the
right-hand side:
 
~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
(F · ∇)G + (G · ∇)F + G × (∇ × F) + F × (∇ × G)~
1
(2)
= (F ~ · ∇)G
~ 1 + (G ~ · ∇)F
~ 1 + G2 (∇~ × F)
~ 3 − G3 (∇
~ × F)
~ 2 + F2 (∇
~ × G)
~ 3 − F3 (∇
~ × G)
~ 2
   
(23) ∂G1 ∂G1 ∂G1 ∂F1 ∂F1 ∂F1
= F1 + F2 + F3 + G1 + G2 + G3
∂x ∂y ∂z ∂x ∂y ∂z
       
∂F2 ∂F1 ∂F1 ∂F3 ∂G2 ∂G1 ∂G1 ∂G3
+ G2 − − G3 − + F2 − − F3 −
∂x ∂y ∂z ∂x ∂x ∂y ∂z ∂x
(collect the six non-derivative terms F1 , F2 ,. . . )
∂G1 ∂G2 ∂G3 ∂F1 ∂F2 ∂F3
= F1 + F2 + F3 + G1 + G2 + G3 (collect terms pairwise)
 ∂x ∂x   ∂x ∂x ∂x
  ∂x 
∂G1 ∂F1 ∂G2 ∂F2 ∂G3 ∂F3
= F1 + G1 + F2 + G2 + F3 + G3
∂x ∂x ∂x ∂x ∂x ∂x
(8) ∂(F1 G1 ) ∂(F2 G2 ) ∂(F3 G3 )
= + +
∂x ∂x ∂x
~ · G)
(1) ∂(F ~
=
∂x
(10)

= ∇( ~ F~ · G)
~ ,
1

and similarly for the y- and z-components.

Exercise 1.57. Prove identities (29), (31) and (34) .
Remark 1.58 (Why use vector identities?). Despite looking complicated at a first glance, the identities pre-
sented in this section are very useful tools to simplify computations involving differential operators. If you learn
how to use them—rather than reinventing the wheel each time expanding all expressions in components and
the operators in partial derivatives—many exercises and theoretical results in the rest of this module and in
future courses will definitely turn out easier.
Consider, for example the following exercise, which was part of the exam in May 2014.
~ × (f ∇g)
Let f and g be two smooth scalar fields. Prove the following identity: ∇ ~ +∇~ × (g ∇f
~ ) = ~0.
A. Moiola, University of Reading 24 Vector calculus lecture notes, 2016–17

Using two vector identities and the linearity of curl, the solution takes no more than one line9 :

(26)
∇~ × (f ∇g)
~ +∇ ~ × (g ∇f
~ ) = ∇ ~ × (f ∇g
~ + g ∇f~ ) (28)= ∇~ × ∇(f
~ g) = ~0.

If you do not use the vector identities as above, but prove the result componentwise, the solution is longer and
definitely much more subject to mistakes: for the first component we have

∂
~ × (f ∇g)
∇ ~ +∇ ~ ) (23)
~ × (g ∇f = ~ 3 − ∂ (f ∇g)
(f ∇g) ~ 2 + ∂ (g ∇f
~ )3 − ∂ (g ∇f
~ )2
1 ∂y ∂z ∂y ∂z
(10) ∂
 ∂g  ∂  ∂g  ∂  ∂f  ∂  ∂f 
= f − f + g − g
∂y ∂z ∂z ∂y ∂y ∂z ∂z ∂y
(8) ∂f ∂g ∂2g ∂f ∂g ∂2g ∂g ∂f ∂2f ∂g ∂f ∂2f
= +f − +f + +g − +g
∂y ∂z ∂y∂z ∂z ∂y ∂z∂y ∂y ∂z ∂y∂z ∂z ∂y ∂z∂y
(17)
= 0,
and similarly for the second and third components.
⋆ Remark 1.59. As a further example of the convenience of the use of vector notation consider the following
example. The so-called “Maxwell equations in time-harmonic regime” describe the propagation of an electric
~ (a vector field) in a homogeneous materials under certain physical conditions. These partial differential
field E
equations in vector form simply read as:
~ × (∇
∇ ~ − k2 E
~ × E) ~ = ~0,

where k is a positive scalar called wavenumber. The same equations written in coordinates have a more
complicated and obscure expression (compare with the proof of (27)):
 2

 ∂ E2 ∂ 2 E3 ∂ 2 E1 ∂ 2 E1

 + − − − k 2 E1 = 0,

 ∂x∂y ∂x∂z ∂y 2 ∂z 2

 2
∂ E1 ∂ 2 E3 ∂ 2 E2 ∂ 2 E2
+ − − − k 2 E2 = 0,

 ∂x∂y ∂y∂z ∂x2 ∂z 2



 ∂ 2 E1 ∂ 2 E2 ∂ 2 E3 ∂ 2 E3

 + − − − k 2 E3 = 0.
∂x∂z ∂y∂z ∂x2 ∂y 2
Example 1.60. We demonstrate some of the results of Proposition 1.52 for the vector field
~ = xyı̂ + ey cos x ̂ + z 2 k̂:
F
~ ·F
∇ ~ = y + ey cos x + 2z,
~ ×F
∇ ~ = (0 − 0)ı̂ + (0 − 0)̂ + (ey sin x − x)k̂ = (ey sin x − x)k̂,
~ ∇
∇( ~ · F)
~ = −ey sin x ı̂ + (1 + ey cos x)̂ + 2k̂,
~ ×∇
∇ ~ ×F~ = ey sin xı̂ − (ey cos x − 1)̂,
~F
∆ ~ = ∆(xy)ı̂ + ∆(ey cos x)̂ + ∆(z 2 )k̂ = 0 + (−ey cos x + ey cos x)̂ + 2k̂ = 2k̂

~ F
and from the last three formulas we immediately demonstrate (27). Since ∇× ~ is parallel to k̂ and independent
of z, its divergence vanishes, thus also the identity (25) is verified.
~ = zı̂ + x2 ̂ + y 3 k̂ and G
Exercise 1.61. ◮ Demonstrate identity (30) for the fields F ~ = exy k̂.
Exercise 1.62. ◮ Let α ∈ R be a constant. Prove that (Hint: recall Exercise 1.35.)
~ · (|~r|α~r) = (3 + α)|~r|α
∇ and ~ × (|~r|α~r) = ~0.
∇
Exercise 1.63. ◮ (i) Prove that ∇ ~ × (f ∇f
~ ) = ~0 for all smooth scalar fields f .
(ii) Prove that, for all natural numbers n, ℓ ∈ N, the more general identity ∇ ~ × (f n ∇(f
~ ℓ )) = ~0 holds true.
Here f n and f ℓ simply denote the nth and ℓth powers of f .
Hint: use an appropriate version of the chain rule to compute the gradients of f ℓ and f n .
⋆ Remark 1.64. The following identity involving the Jacobian matrix holds true:


(∇~ × F)
~ ×G ~ = JF ~ − (J F) ~
~ T G,
T
where the symbol denotes matrix transposition. Try to prove it for exercise.
9 Another possible solution using two vector identities and the anti-commutativity of the vector product is the following:
~ × (f ∇g)
∇ ~ +∇ ~ ) (31)
~ × (g ∇f ~ ) × (∇g)
= (∇f ~ +f∇
~ × (∇g) ~ × (∇f
~ +(∇g) ~ )+g∇ ~ ) (3)
~ × (∇f ~ ) × (∇g)
= (∇f ~ − (∇f
~ ) × (∇g)
~ =~
0.
| {z } | {z }
=~
0, (26) =~
0, (26)
A. Moiola, University of Reading 25 Vector calculus lecture notes, 2016–17

1.5 Special vector fields and potentials

Definition 1.65. Consider a vector field F ~ defined on a domain D ⊂ R3 .

~ ×F
If ∇ ~ = ~0, then F~ is called irrotational (or curl-free).
~ ~ ~
If ∇ · F = 0, then F is called solenoidal (or divergence-free, or incompressible).
~ = ∇ϕ
If F ~ for some scalar field ϕ, then F ~ is called conservative and ϕ is called scalar potential of F.
~
~ ~ ~ ~ ~
If F = ∇ × A for some vector field A, then A is called vector potential of F. ~

10 ~ is irrotational precisely when

From the definition of the curl operator, we note that F
∂F2 ∂F3 ∂F3 ∂F1 ∂F1 ∂F2
= , = and = ,
∂z ∂y ∂x ∂z ∂y ∂x
~ is symmetric. If F
i.e. when the Jacobian matrix J F ~ admits a scalar or a vector potential, then the
potential is not unique: for all constant scalars λ ∈ R and for all scalar fields g (recall (26))

if ~ = ∇ϕ
F ~ ~ = ∇(ϕ
then F ~ + λ); if ~ =∇
G ~ ×A
~ then ~ =∇
G ~ × (A
~ + ∇g);
~

therefore ϕ ~e := A
e := ϕ + λ and A ~ + ∇g ~ and G.
~ are alternative potentials for F ~

Example 1.66. Consider the following vector fields:

~ r) = 2xı̂ − 2y̂,
F(~ ~ r) = yzı̂ + xz̂ + xy k̂,
G(~ ~ r) = |~r|2 ı̂ + cos y k̂.
H(~

In Exercises 1.46 and 1.50 we showed that F ~ and G

~ are both irrotational and solenoidal while H ~ is neither
~ ~
irrotational nor solenoidal. Moreover, F and G admit both scalar and vector potentials (thus in particular they
are conservative):
 2 
~ ~ 2 2 ~ ~ ~ ~ ~ z xı̂ + x2 y̂ + y 2 z k̂
F = ∇(x − y ) = ∇ × (2xy k), G = ∇(xyz) = ∇ × .
2
(For the technique used to compute these potentials see Example 1.67 below.) Of course there exist fields
which are irrotational but not solenoidal and the other way round: verify this fact for the following fields

~ A = 2xı̂ + 3y 2 ̂ + ez k̂, ~ B = e−x2 ı̂ + 1 ~ C = x3 y 4 ı̂ + x4 y 3 ̂ + z 2 k̂,

F F k̂, F
1 + z2
~ D = (x + y)ı̂ − y̂,
F ~ E = yzı̂ + xz̂ − xy k̂,
F ~ F = (x2 + cos z)ı̂ − 2xy̂.
F
Example 1.67 (Computation of potentials). Given a field, how can we compute a scalar or vector potential?
We consider as example the irrotational and solenoidal field
~ = zı̂ + xk̂.
F
~ it must satisfy F
If ϕ is a scalar potential of F, ~ = ∇ϕ,
~ thus, acting componentwise,
∂ϕ
=z ⇒ ϕ = xz + f (y, z) for some two-dimensional scalar field f,
∂x

∂ϕ ∂ xz + f (y, z) ∂f (y, z)
=0 ⇒ = =0 ⇒ ϕ = xz + g(z) for some real function g,
∂y ∂y ∂y


∂ϕ ∂ xz + g(z) ∂g(z)
=x ⇒ =x+ =x ⇒ ϕ = xz + λ.
∂z ∂z ∂z
~
Thus, for every real constant λ, the fields xz + λ are scalar potentials of F.
~ ~ ~ ~
A vector potential A must satisfy ∇ × A = F thus
∂A3 ∂A2 ∂A1 ∂A3 ∂A2 ∂A1
− = z, − = 0, − = x.
∂y ∂z ∂z ∂x ∂x ∂y

Many choices for A~ are possible, the simplest one is to set A1 = A3 = 0 and A2 (~r) = a(x, z) for some
two-dimensional scalar field a(x, z) which now must satisfy
∂a ∂a
= x, = −z.
∂x ∂z
10 In ~ = −∇ϕ,
physics, the scalar potential is often defined to satisfy F ~ this can be a big source of confusion!
A. Moiola, University of Reading 26 Vector calculus lecture notes, 2016–17

We integrate and obtain

1 2


x2 ∂ 2x + b(z) ∂b(z) −z 2
a= + b(z) ⇒ = = −z ⇒ b(z) = ,
2 ∂z ∂z 2
~ = 1 (x2 − z 2 )̂ is a vector potential of F
and finally A ~ (and many others are possible).
2
~ A and F
As exercise consider the fields defined in Example 1.66 and compute a scalar potential of F ~ C and
~
a vector potential of FD .
You can practice more in the computation of potentials with Exercises 1–10 of [1, Section 15.2, p. 857].
Note that what we are actually doing in Example 1.67 is computing indefinite integrals, so in most
practical cases a closed form for the potentials is not available.
Exercise 1.68. ◮ Verify that the position field ~r is irrotational, conservative and not solenoidal. Compute a
scalar potential for ~r.
~ and G
Exercise 1.69. ◮ Use two irrotational vector fields F ~ to construct a solenoidal vector field H.
~ Hint:
use the identities proved in the previous section.
~ × (∇ϕ)
Because of identities (26) and (25) (∇ ~ ~ · (∇
= ~0, ∇ ~ × A)
~ = 0) all conservative fields are
irrotational and all fields admitting a vector potential are solenoidal.

⋆ Remark 1.70 (Existence of potentials). Under some assumptions on the domain D of definition of F~ the
converse of the statements in the box is true. If D is R3 or a ball or a parallelepiped, for example, then all
irrotational fields are conservatives and all solenoidal fields admit a vector potential.
More general assumptions on the shape of D can be made. If the domain D is “simply connected”, which
means, informally, that it has no holes or tunnels passing through it, then the converse of the first statement
above is true: every irrotational field on D is conservative. If the domain D “has no holes in its interior”,
the converse of the second statement is true: every solenoidal field on D admits a vector potential A.~ E.g. a
doughnut is not simply connected and the complement of a ball does not satisfies the second condition. We
will prove the first statement for a simpler class of domains in Theorem 2.18.
⋆ Remark 1.71 (Relation between potentials and domains). The fact that a vector field F~ defined in D is
conservative may depend on the choice of the domain D. For instance, consider the irrotational (planar) field

~ = −y x
F ı̂ + 2 ̂
x2+y 2 x + y2

(exercise: show that F~ is irrotational); see also [5, VII Section 3]. In the half space D = {~r ∈ R3 s.t. x > 0},
which is simply connected, it admits the scalar potential ϕ = arctan xy (exercise: show that ∇ϕ ~ =F ~ in D,
using ∂t∂ 1
arctan t = 1+t ~
2 ). However, F may be defined on the larger domain E = {~ r ∈ R3 s.t. x2 + y 2 > 0},
i.e. the whole space without the z-axis; E is not simply connected. The potential ϕ can not be extended to
the whole E and it is not possible to define any other scalar potential for F ~ in the whole of E; this will follow
from some properties of conservative fields we will see in Section 2.1.3.
To summarise: F~ is irrotational in E, conservative in the subset D but not conservative in the whole of E.

⋆ Remark 1.72 (Helmholtz decomposition). ~ defined in a bounded domain D can

Any smooth vector field F
~ = ∇ϕ
be expressed as the sum of a conservative field and a field admitting a vector potential: F ~ +∇ ~ ×A ~
~ respectively. The potentials are not unique. This decomposition is
for some scalar and vector field ϕ and A,
termed Helmholtz decomposition after Hermann von Helmholtz (1821–1894).
Exercise 1.73. ◮ Show that the divergence of a conservative field equals the Laplacian of its scalar potential.
Exercise 1.74. ◮ Find all smooth scalar fields in R3 whose gradient admits a conservative vector potential.
Exercise 1.75. ◮ Assume F ~ and G~ are conservative fields with scalar potentials ϕ and ψ, respectively. Use
~ ×G
the identities of Section 1.4 to show that the product F ~ is solenoidal and find a vector potential for it.

1.6 Total derivatives of a curve

So far we have applied various differential operators to scalar and vector fields but we have not considered
the derivatives of the curves introduced in Section 1.2.3. These are indeed easier to deal with than the
derivative of fields, since they can be computed as the “usual derivatives” of the components (which are
real-valued functions of a single variable) as learned in the first calculus class:

d~a da1 da2 da3

(t) := ~a ′ (t) := (t)ı̂ + (t)̂ + (t)k̂.
dt dt dt dt
A. Moiola, University of Reading 27 Vector calculus lecture notes, 2016–17

d
No partial derivatives are involved here. Note that we used the total derivative symbol dt instead of
∂
the partial derivative one ∂t because ~a and its three components depend on one variable only (i.e. on t).
The derivative of a curve is a curve.
Remark 1.76 (Difference between partial and total derivatives). For f (~r) = f (x, y, z), the partial derivative
∂f
∂x is the derivative of f with respect to x when all the other variables are kept fixed. The total derivative
df
dx takes into account also the possible dependence of the other variables (y and z) on x. We will see some
examples later on, e.g. in Section 1.7.
Given two curves ~a and ~
b and two scalar constants λ, µ, linearity and product rule hold:

d(λ~a + µ~b) d~a d~b

=λ +µ ,
dt dt dt
~
d(~a · b) d~a ~ d~b
(~a · ~b)′ = = · b + ~a · , (36)
dt dt dt
d(~a × ~b) d~a ~ d~b
= × b + ~a × .
dt dt dt

Note that t 7→ (~a(t) · ~b(t)) is simply a real function of a real variable, thus we write its derivative as (~a · ~b)′ .
⋆ Remark 1.77 (Arc length). A curve ~a : I → R3 is called smooth if its three components are smooth
functions and | d~a
a : I → R3 with path Γ satisfies | d~
dt | 6= 0 for all t ∈ I. If a curve ~
a
dt | = 1 for all t ∈ I, then
t is called arc length of Γ and ~a is called intrinsic parametrisation of Γ. Every smooth path has exactly two
intrinsic parametrisations, proceeding in opposite directions from each other.
If you wonder why a curve to be smooth needs to satisfy | d~ a 3 2
dt | 6= 0, draw the curve a(t) = t ı̂ + t ̂, whose
~
components are smooth functions (polynomials). Recall that the path of a curve is its image, not its graph.
Exercise 1.78. ◮ Compute the total derivatives of the two parametrisations of the unit circle seen in Sec-
tion 1.2.3: ~a(t) = cos t ı̂ + sin t ̂ with t ∈ [0, 2π) and ~b(t) = cos 2τ ı̂ + sin 2τ ̂ with τ ∈ [0, π).
d~
The paths of ~a and ~ b coincide, is the same true for the paths of d~ a b
dt and dt ?
Can you find a curve ~c with path different from ~a whose derivative d~ c d~
dt has the same path of dt ?
a

Exercise 1.79. ◮ Demonstrate the two product rules in equation (36) (for scalar and vector product) for the
two parametrisations of the unit circle in Exercise 1.78: ~a(t) = cos t ı̂ + sin t ̂ and ~b(t) = cos 2t ı̂ + sin 2t ̂.
Exercise 1.80. ◮ Prove the product rules for scalar and vector products of curves in equation (36).
Exercise 1.81. ◮ Given a vector ~ u, find a curve ~a(t) defined for t ≥ 0 that is equal to its own total derivative
and such that ~a(0) = ~u. Is this curve unique? What is the path of ~a?

1.7 Chain rule for fields and curves

Comparison with scalar calculus 1.82. The chain rule is the formula that allows to compute derivatives of
the composition of functions, i.e. the application of a function to the output of a second one. Given two
differentiable real functions F, G : R → R, you know well the corresponding chain rule:

′

(G ◦ F )′ (t) = G(F (t)) = G′ F (t) F ′ (t),

(recall footnote 4). See also page 11 of Handout 2 of last year’s Calculus lecture notes. In this section we study
some extensions of this formula to curves and fields. We have seen in (9) and (14) how to compute partial
derivatives and the gradient of the composition of a real function with a scalar field (~r 7→ f (~r) 7→ G(f (~r))).
Note the special structure of the chain rule formula above. The derivative of the composition equals the
derivative of the “external” function G evaluated in the output of F times the derivative of the “internal”
function F . All versions of the chain rule share the same structure (see (9), (14), (37) and (38)).
Imagine we have a scalar field f evaluated on a smooth curve ~a, i.e. f (~a). This is a real function of
real variable t 7→ (f ◦ ~a)(t) = f (~a(t)). Its derivative in t can be computed with the chain rule:


d f (~a) ∂f
da1 ∂f
da2 ∂f
da3
(t) = ~a(t) (t) + ~a(t) (t) + ~a(t) (t)
dt ∂x dt ∂y dt ∂z dt (37)


~ ~a(t) · d~a (t).
= ∇f
dt
df d(f (~
a)) ∂f ∂f
When no ambiguity can arise, we will write dt for dt (t) and ∂x for ∂x (~
a(t)) omitting the argument.
A. Moiola, University of Reading 28 Vector calculus lecture notes, 2016–17

Exercise 1.83. ◮ Consider the scalar field f = xyez and the curve ~a = tı̂ + t3 ̂ (see Figure 8). Compute the
total derivative of f (~a) using either the chain rule or explicitly computing the composition f (~a).

We have seen different chain rules for the composition of real functions with scalar fields in (9) and (14)
(~r 7→ f (~r) 7→ G(f (~r))) and for the composition of scalar fields with curves in (37) (t 7→ ~a(t) 7→ f (~a(t))).
What about the partial derivatives of the composition of a scalar field with a vector field? Let g be a
scalar field and F~ a vector field, and define the scalar field φ(~r) := g(F(~
~ r)) = (g ◦ F)(~
~ r). Then the partial
derivatives and the gradient of φ can be computed as

∂φ ~
~  ∂F3 
~ · ∂ F = (∇g)
= ∇g ~ ~ r) · ∂ F (~r) = ∇g
F(~ ~ · ∂F1 ı̂ + ∂F2
̂ + k̂ ,
∂x ∂x ∂x ∂x ∂x ∂x
∂φ ~
∂F ~  ∂F ∂F3 
~ · ∂ F = (∇g)
= ∇g ~ ~ r) ·
F(~ ~ ·
(~r) = ∇g
1
ı̂ +
∂F2
̂ + k̂ , ~ = (J F)
∇φ ~ T (∇g).
~ (38)
∂y ∂y ∂y ∂y ∂y ∂y
∂φ ~
~  ∂F3 
~ · ∂ F = (∇g)
= ∇g ~ ~ r) · ∂ F (~r) = ∇g
F(~ ~ · ∂F1 ı̂ + ∂F2
̂ + k̂ ,
∂x ∂z ∂z ∂z ∂z ∂z
~ T (∇g)
In (38), (J F) ~ denotes the matrix–vector multiplication between the transpose of the Jacobian matrix
~ and the gradient of g. Note that in this formula ∇g
of F ~ is always evaluated in F(~
~ r), even when we do
not write it explicitly. Similar formulas hold for two-dimensional fields.
⋆ Remark 1.84. All instances of the chain rule (9), (14), (37) and (38) are special cases of a more general result
for functions between Euclidean spaces of arbitrary dimensions, see e.g. [1, pages 708–709] or [5, XVI, Section 3].
Example 1.85 (Derivative of a field constrained to a surface). In some problems, the coordinates of ~r depend
on each other. In this case the partial derivatives of a field g(~r) do not give the desired rate of change of g.
For example, imagine that the point ~r is constrained to the graph surface of h : R2 → R

Sh = ~r ∈ R3 , s.t. ~r = xı̂ + y̂ + h(x, y)k̂ .

In this case, the component z depends on x and y through the two-dimensional field h. Thus also the value of
g on Sh only depends on x and y. We define e g(x, y) = g(x, y, h(x, y)) to be the evaluation of g on Sh ; ge is a
field depending on two variables only. We can compute the derivative of g along x or y using the chain rule:


∂ g x, y, h(x, y)   
∂eg ∂g
∂g
∂h
(x, y) = = (x, y, h(x, y) + (x, y, h(x, y) (x, y) ,
∂x ∂x ∂x ∂z ∂x


∂ g x, y, h(x, y)   
∂eg ∂g
∂g
∂h
(x, y) = = (x, y, h(x, y) + (x, y, h(x, y) (x, y) .
∂y ∂y ∂y ∂z ∂y
∂g ∂g
Here ∂x (x, y, h(x, y)) (similarly, ∂y (x, y, h(x, y)) and ∂g
∂z (x, y, h(x, y))) indicates the derivative of g with re-
spect to its first (second and third, respectively) scalar variable; the field is evaluated in (x, y, h(x, y)) after
the derivative has been taken. This formula can be seen as a special case of the chain rule (38) for the vector
~ = xı̂ + y̂ + h(x, y)k̂ (independent of z), whose Jacobian reads
field F
 
1 0 0
JF~ =0 1 0 .
∂h ∂h
∂x ∂y 0

Let us consider a concrete example. Let h(x, y) = x2 − y 2 , so that the surface Sh = {~r = xı̂ + y̂ + (x2 −
2
y )k̂} is that depicted in Figure 14 (compare also with the right plot in Figure 5). Let g = x sin z. Then, from
the formulas above, we compute the derivatives of g with respect to x and y:


∂ g x, y, h(x, y)
= sin z + (x cos z)(2x) = sin z + 2x2 cos z = sin(x2 − y 2 ) + 2x2 cos(x2 − y 2 ),
 ∂x


∂ g x, y, h(x, y)
= 0 + (x cos z)(−2y) = −2xy cos z = −2xy cos(x2 − y 2 ).
∂y

Note that even if g is independent of the y-coordinate, its derivative along the surface defined by h (which
does depend on y) is non-zero. We could compute these derivatives also by first computing e g (x, y) =
g(x, y, h(x, y)) = x sin(x2 − y 2 ) and then deriving.
A. Moiola, University of Reading 29 Vector calculus lecture notes, 2016–17

1.5

3 1

2
0.5
1

z 0
0
−1

−2
−0.5
−3

−4
−1
2

1 2
1 −1.5
y 0

−1
0
x
−1
−2 −2

Figure 14: The surface Sh = {~r = xı̂ + y̂ + (x2 − y 2 )k̂} described in Example 1.85. The colours represent
the values of the field g = x sin z, which can be written as g = x sin(x2 − y 2 ) when restricted to Sh . Note
that g varies in the y coordinate if we move along Sh keeping x fixed, while it is constant in that direction
if we move in free space (see the .pdf file of these notes for the coloured version).

Example 1.86 (Derivative of a field constrained to a curve). Compute the total derivative
d

g a(z), b(z), z
dz
where g = exz y 4 , a(z) = cos z, b(z) = sin z.
This is again a derivative along the curve a(t)ı̂ + b(t)̂ + tk̂ (compare with the centre plot of Figure 8)
with t and z identified with each other, so formula (37) applies. We combine the partial derivatives of g and
the total derivatives of a and b:
d
∂g da ∂g db ∂g
g a(z), b(z), z = + + = −zexz y 4 sin z + 4exz y 3 cos z + xexz y 4
dz ∂x dz ∂y dz ∂z


= ez cos z − z sin5 z + 4 sin3 z cos z + cos z sin4 z .

Exercise 1.87. ◮ Given the curve ~a(t) = cos tı̂ + sin t̂, the scalar field g = x2 − y 2 and the vector field
~ = xı̂ + z̂ − y k̂, compute the partial derivatives of g ◦ F
F ~ and the total derivatives of g ◦ ~a and of g ◦ F
~ ◦ ~a.
Compute them twice: (1) by calculating the compositions and then deriving them, and (2) by using the vector
formulas introduced in this section.
Remark 1.88. In Section 1.3.2 we defined the directional derivative ∂∂fû of a scalar field f in the direction of
~ · û. Using the chain rule (37) it is immediate to verify that it corresponds to
the unit vector û as ∂∂fû = ∇f
the rate of increase of f in the direction û:


∂f ~ d f (~r + tû) 
(~r) = ∇f (~r) · û =  .
∂ û dt t=0

From Cauchy–Schwarz inequality (see Proposition 3.5 in last year’s Linear Algebra lecture notes, handout 2,
or Proposition 13.4 in two years ago’s notes, handout 4), we have | ∂∂fû | = |∇f ~ · û| ≤ |∇f
~ | |û| = |∇f
~ |. This
means that the magnitude of the gradient is at least as large as the rate of increase of f in any other direction.
Since |û · ŵ| = |û||ŵ| if and only if û and ŵ are parallel, the direction of the gradient is the unique direction
of maximal increase for f at ~r (when ∇f ~ (~r) 6= ~0). We have proved part 5 of Proposition 1.33.

⋆ Remark 1.89. We can use the chain rule to prove part 4 of Proposition 1.33. Assume f is a smooth scalar
field defined in an open set D, fix ~r0 ∈ D and denote the corresponding level set of f as L = {~r ∈ D such
that f (~r) = f (~r0 )}. Consider a smooth curve ~a : I → L with ~a(t0 ) = ~r0 for some t0 ∈ I, where I is an
interval. Then, by the definition of the level set, the real function t 7→ f (~a(t)) is constant, so its derivative
d(f (~
a)) ~ (~r0 ) · d~a (t0 ) = d(f (~a)) (t0 ) = 0. This means that the gradient of
dt vanishes. By the chain rule (37), ∇f dt dt
f in ~r0 is perpendicular to the tangent vectors d~ a
~ ~
dt to all smooth curves a passing through r0 and lying in the
level set L. This is the same as saying that the gradient is perpendicular to the level set itself (note that so far
we have not defined perpendicularity between vectors and sets).
A. Moiola, University of Reading 30 Vector calculus lecture notes, 2016–17

⋆ Remark 1.90. (Differentials) As in one-dimensional calculus, we can consider “differentials”, infinitesimal

changes in the considered quantities. The differential
∂f ∂f ∂f
df = dx + dy + dz
∂x ∂y ∂z
of the scalar field f (~r) represents the change in f (~r) due to an infinitesimal change in ~r. If ~r depends on t as
~r(t) = x(t)ı̂ + y(t)̂ + z(t)k̂, we can express its vector differential as
 dx   dy   dz   dx dy dz 
d~r = dxı̂ + dy̂ + dz k̂ = dt ı̂ + dt ̂ + dt k̂ = ı̂ + ̂ + k̂ dt.
dt dt dt dt dt dt
Thus, if f is evaluated in ~r(t), its differential df can be expressed in terms of dt:

~ · d~r = ∂f dx dt + ∂f dy dt + ∂f dz dt.
df = ∇f
∂x dt ∂y dt ∂z dt
Differentials
p can be used to estimate (compute approximately) the values of a field. Consider for example
the field f = x2 + y 3 . We want to estimate its value at the point ~r = 1.03ı̂ + 1.99̂ knowing that f and its
gradient in the “nearby” point ~r0 = ı̂ + 2̂ take values
√ 2
f (~r0 ) = 1 + 8 = 3, ~ (~r0 ) = 2xp
∇f
0 ı̂ + 3y0 ̂ 1
= ı̂ + 2̂.
2
2 x0 + y0 3 3
We approximate the desired value as
 
~ (~r0 ) = 3 + (0.03ı̂ − 0.01̂) · 1 ı̂ + 2̂ = 2.99,
f (~r) ≈ f (~r0 ) + ~δ · ∇f
3
where ~δ = ~r − ~r0 is the discrete increment. (The exact value is f (~r) = 2.9902339 . . .)

1.8 Review exercises for Section 1

Consider the following scalar and vector fields:
f = xy 2 z 3 , ~ = (y + z)ı̂ + (x + z)̂ + (x + y)k̂,
F
g = cos x + sin(2y + x), ~ = e−z2 ı̂ + y 2 ̂,
G
h = ex cos y, ~ = ~r × ı̂,
H (39)
ℓ = x|~r|2 − y 3 , ~ = |~r|2 ~r,
L
m = xy (x > 0), ~ = yz 2 (yzı̂ + 2xz̂ + 3xy k̂).
M
Consider also the three following curves, defined in the interval −1 < t < 1:
~a = (t3 − t)ı̂ + (1 − t2 )k̂, ~
b = t3 ı̂ + t2 ̂ + k̂, ~c = et cos(2πt)ı̂ + et sin(2πt)̂.
Answer the questions, trying to be smart and to avoid brute force computations whenever possible.
1. Compute gradient, Hessian and Laplacian of the five scalar fields.
2. Compute Jacobian, divergence, curl and vector Laplacian of the five vector fields.
3. Which of the fields are solenoidal and which are irrotational? Which are harmonic?
~ a vector potential for H
4. Compute a scalar and a vector potential for F, ~ (can (31) help you?) and a
~ Can you guess a scalar potential for L?
scalar potential for M. ~
~ does not admit neither a scalar nor a vector potential.
5. Show that G
~ r) and L(~
6. Show that H(~ ~ r) are orthogonal to each other at every point ~r ∈ R3 .
7. Try to graphically represent the fields in (39). E.g. you can draw a qualitative plot like those of
Section 1.2.2 for G~ on the plane y = 0, for F,
~ H~ and L~ on the plane z = 0.
~ (26) for f ; (28) for f and h; (33) for G
8. Demonstrate identities (25) and (27) for G; ~ and H;
~ (29) and
(31) for h and H.~ (You can also demonstrate other identities of Propositions 1.52 and 1.55 for the
various fields in (39).)
a d~
d~ b d~c
9. Compute the (total) derivatives of the curves (i.e. dt , dt and dt ) and try to draw them.
a) df (~
dh(~ b) dℓ(~c)
10. Compute the following total derivatives of the scalar fields along the curves: dt , dt and dt .
(You can either use the chain rule (37) or first compute the composition.)
A. Moiola, University of Reading 31 Vector calculus lecture notes, 2016–17

2 Vector integration
So far we have extended the concept of differentiation from real functions to vectors and fields (differential
vector calculus); in this section we consider the extension of the idea of integration for the same objects
(integral vector calculus).
In vector calculus we do not consider “indefinite integrals”, so we usually do not understand integration
as “inverse of derivation”.11 On the contrary, every integral defined in this section must be understood as
a cumulative measure of a quantity (named integrand, which is a scalar or vector field) distributed on a
geometric object, the domain of integration. This can be a one-, two- or three-dimensional subset of
R3 , namely the path of a curve, a surface or a domain, respectively. This is reflected in the notation: the
number of integral signs used represents the dimension R of the domain of integration, which is written at
subscript. Integrals RR along a path Γ use the symbol Γ
and are called “line integrals”, those over a surface
RRR
S are denoted by S and called “surface integrals”, those on a domain D are written as D and called
“volume integrals”.
A good way to have a physical understanding of the integrals of (positive) scalar fields, is to think
at the domain as a physical object (in the cases of paths or surfaces, as wires or plates of infinitesimal
diameter or thickness, respectively) and at the scalar field as the density of mass of the object, which can
vary in different points if the material is heterogeneous. The integral is then the total mass of the object.
If the field changes sign, it can be thought as electric charge density, and its integral as the total charge.
The integrals of vector fields, representing e.g. the work of a force or the flux of a substance through an
aperture, are slightly more complicated and involve concepts such as the orientation of a path or a surface,
and will be described in the following sections.
All integrals of scalar fields will have to satisfy some fundamental properties12 :

1. Linearity: the integral of f + g equals the sum of the integral of f and the integral of g  

on the same domain; the integral of λf equals λ times the integral of f (here f and g are  

scalar fields and λ ∈ R). 





2. Additivity: the integral over the union of two disjoint (i.e. with empty intersection) (40)
domains of integration (may be two paths, two surfaces or two domains) equals the sum  

of the two integrals of the same field over each domain. 





3. The integral of the constant field f (~r) = 1 equals the measure of the domain of inte-  

gration: the length of a path, the area of a surface, the volume of a domain.
Items 1.–2.–3. above extend the following well-known facts about integrals of functions of a real variable:
Z b Z b Z b
1. (λf + µg) dt = λ f dt + µ g dt,
a a a
Z c Z b Z c
2. f dt = f dt + f dt,
a a b
Z b
3. 1 dt = b − a for all a < b < c, λ, µ ∈ R, f, g : [a, c] → R.
a

Other natural properties of the integrals follow: the integral of positive scalar field is a positive number;
a dilation of the domain by a factor λ > 0 corresponds to a multiplication of the integral by a factor λ if
the domain is a curve, λ2 if it is a surface, and λ3 if it is a volume.

2.1 Line integrals

Here we study how to integrate scalar and vector fields along curves. See also Sections 15.3 and 15.4 of [1],
which contain several useful exercises.

2.1.1 Line integrals of scalar fields

Consider a curve ~a : [tI , tF ] → R3 (where [tI , tF ] ⊂ R is an interval) and a scalar field f : R3 → R. We
recall that Γ = ~a([tI , tF ])R⊂ R3 is the path of the curve ~a. We want to define the integral of f over Γ,
which will be denoted by Γ f ds, satisfying the conditions 1.–2.–3. stipulated in (40).
11 In Sections 2.1.3 and 3, however, we study some integrals of differential operators, generalising the fundamental theorem

of calculus, but we do not use this idea to define the integrals themselves.
12 We stress again a fundamental concept. All integrals requires two inputs: an integrand (a function or field), and a

domain (a set). Linearity concerns integrals of different integrand on the same domain, while additivity concerns integrals
of an integrand over different domains.
A. Moiola, University of Reading 32 Vector calculus lecture notes, 2016–17

Intuitively, what does this integral represent? In the special case of a planar curve, i.e. ~a(t) =
a1 (t)ı̂ + a2 (t)̂, and a two-dimensional positive scalar field f (x, y) > 0 we can have a simple geomet-
ric representation of the integral. We know well that the one-dimensional integral of a function of a real
variable represents the (signed) area between its graph and its projection on the x-axis, which is the
domain of integration (see the left plot in Figure 15). Similarly, the line integral will represent the area
of the (curvilinear!) surface between the path of the curve in the xy-plane (the domain of integration)
and the graph of f , represented as a surface as we have seen in Remark 1.21. The example depicted in
Figure 15 is probably more helpful than the description. Note that two surfaces are involved here: the
graph of the scalar field f (the upper mesh in the plot), and the vertical surface between the graph and
xy-plane (the grey one in the plot).

g(t)

12

10

8
f (x, y)

0
2

1 2
1
y 0
Γ = {~a(t)} 0
−1 x
−1
−2 2 t −2 −2

R2 2
Figure 15: Left: the integral −2 g(t) dt of the real function g(t) = 6 + t2 − 49 (e−x )2 is the area of the
shaded region between the graph of the function and the x-axis.
2
Right: the path Γ of the planar curve ~a(t) = tı̂ − 23 e−t ̂ (lower Rthick plot) and the surface plot of the
scalar field f = 6 + x2 − y 2 (the mesh above). The line integral Γ f ds of f over Γ is the area of the
vertical surface lying between Γ itself and its projection ~a(t) + f (~a(t))k̂ on the graph of f , i.e. the upper
thick curve. (The shaded area in the left plot is equal to the projection on the xz-plane of the shaded area
in the right plot, since g(t) = f (~a(t)) and a1 (t) = t, but the integrals are different since the “stretching”
given by the shape of Γ is not taken into account by g.)

If the field is not always positive but changes sign in its domain, as in the scalar case, the integral is
the difference between the area of the part of the surface above the xy-plane and the part below it.
In three dimensions (i.e. if all three components of ~a(t) vary in t) it is not possible to have an easy
geometric picture of the integral, since it would require a fourth dimension. In this case it is preferable to
think at f as some pointwise mass (or charge) density along a wire and the line integral as its total mass
(or charge).
How can we write this integral in a computable formula? We first write the formula and we justify it
in Remark 2.1 below. The line integral of f along Γ is
Z Z tF   Z tF r

 d~a 
 da1 2  da2 2  da3 2
f ds =  
f ~a(t)   dt = f ~a(t) + + dt, (41)
Γ tI dt tI dt dt dt

where ds denotes the infinitesimal element of the path Γ. The right-hand side of (41) is an ordinary
one-dimensional integral over the interval [tI , tF ], involving the magnitude of the total derivative d~ a
dt of
the curve ~a. If ~a is a loop (i.e.
H ~a(tI ) = ~a(tF )), then the line integral along Γ is called contour integral
and denoted by the symbol Γ f ds.
A. Moiola, University of Reading 33 Vector calculus lecture notes, 2016–17

⋆ Remark 2.1 (Justification of the line integral formula (41)). To justify formula (41) we recall Riemann’s
sums for one-dimensional integrals (in a very simple case), which you might have already seen in other courses.
Consider a “well-behaved” function g : [0, 1] → R. Given a natural number N ∈ N, we can split the
domain [0, 1] in N subintervals [0, N1 ], [ N1 , N2 ], . . . , [ NN−1 , N
N
] with equal lengths (see Figure 16). In each small
subinterval we approximate g(t) with the constant function taking the value of g at the right extreme of the
subinterval, i.e. g(j/N ). If the intervals are small enough, or equivalently if N is large enough, the integral
of g on the subinterval is approximated by the integral of this constant function (which is simply g(j/N )/N ,
since the length of the interval is j/N − (j − 1)/N = 1/N ). Summing over the N subintervals, we obtain an
approximation of the integral of g over [0, 1], (see Figure 16):
Z N Z  j  j j − 1  1 X j 
1 X j/N XN N

g(t) dt = g(t) dt ≈ g  − = g , (42)
0 j=1 (j−1)/N j=1
N N N N j=1 N

(here the symbol “≈” means “is approximately equal to”). If the integrand g is sufficiently “well-behaved” (for
example it is continuous in the closed interval [0, 1]), the approximation will actually converge to the desired
integral, i.e.
Z
1 X j
1 N
g(t) dt = lim g .
0 N →∞ N N
j=1

This is the basic idea behind Riemann sums (although more general partitions of the interval and points of
evaluation may be taken).

g(x)

1 2
x
N −1
0 N N N 1

Figure 16: The approximation of the integral of a real variable function g : [0, 1] → R by Riemann sums.
The area of the grey part bounded by the plot of the integrand represents the integral of g, the area covered
with diagonal lines is its Riemann sum approximation as in equation (42).

How can we write Riemann sums for the line integral of a scalar field? Consider a path Γ parametrised by
the curve ~a : [0, 1] → R3 and a scalar field f (both sufficiently smooth). Fix N ∈ N; for all integers 1 ≤ j ≤ N
call Γj the sub-path parametrised by the curve ~a restricted to the subinterval [(j − 1)/N, j/N ]. We repeat
the same procedure used for scalar functions: we use the additivity of the integral to split it in a sum over
the sub-paths (step (i) in the equation below) and we approximate the integrals with those of the constant
f (~a(j/n)), (ii). To compute the integrals of these constants we need to know the lengths of each sub-path Γj
(iii) (compare with the assumptions made in (40)). We approximate again the length of Γj with the distance
between its endpoints ~a((j − 1)/N ) and ~a(j/N ), (iv). We multiply and divide by 1/N (v) and we note that
the ratio obtained is a difference quotient approximating the total derivative of ~a(t) at t = j/N (vi). Using
the Riemann sum (42) for real functions, we see that the formula obtained is an approximation of the integral
in (41) and we conclude (vii)–(viii). In formulas:
Z N Z
X
(i)
f ds = f ds
Γ j=1 Γj

N Z
X   j 
(ii)
≈ f ~a ds
j=1 Γj N

(iii) XN   j 
= f ~a Length(Γj )
j=1
N
A. Moiola, University of Reading 34 Vector calculus lecture notes, 2016–17

(iv)XN   j    j   j − 1 
 
≈ f ~a ~a − ~a 
j=1
N N N
    
(v) 1 X
N   j  ~a Nj − ~a j−1N


= f ~a j j−1
N j=1 N N − N

1 X   j   d~a  j 
N
(vi)
≈ f ~a  
N j=1 N dt N
N Z
(vii) X j/N
 d~a 
≈ f ~a(t)  (t) dt
j=1 (j−1)/N dt
Z
(viii)
1
 d~a 
= f ~a(t)  (t) dt.
0 dt
We have derived (in a hand-waving way) formula (41) for the line integral of a scalar field, using only the
Riemann sums for functions of real variable and the fundamental properties of the integrals stated in (40).
In this class we take the line integral formula (41) (and all the similar formulas for double, triple, surface
and flux integrals in the rest of Section 2) as a definition; however, as we have seen, these can be (rigorously)
derived from the general theory of Riemann integration.
Example 2.2 (Length of a path). Integrating the constant field f = 1 (cf. (40)), we see that the length of a
path Γ is
Z Z tF  
 d~a 
length(Γ) = ds =   dt. (43)
 dt 
Γ tI

The line integral in (41) measures a quantity f distributed along Γ, thus

 it must be independent of
the special parametrisation ~a. This is guaranteed by the factor  d~ a
dt , which takes into account the
“speed” of travel along Γ (if we think at the parameter t as time). Similarly, the travel direction along
Γ (the curve orientation) does not affect the value of the line integral. In other words, f and Γ uniquely
R of the integral in (41), independently of the choice of ~a we use to compute it. This makes
define the value
the notation Γ f ds in (41) well-posed, without the need of specifying ~a. Example 2.3 demonstrates the
invariance of the line integral with respect to the parametrisation.
Example 2.3 (Independence of parametrisation). Consider the scalar field f = y and the unit half circle C
centred at the origin and located in the half plane {~r = xı̂ + y̂, y ≥ 0}, which can be defined by either of the
two parametrisations
p
~a : [0, π] → R3 , ~a(t) = cos tı̂ + sin t̂, ~b : [−1, 1] → R3 , ~b(τ ) = τı̂ + 1 − τ 2 ̂;

(see Figure 17). Verify that

Z   Z 1  
π
 d~a 
 d~b 
 
f ~a(t)   dt = ~
f b(τ )   dτ.
0 dt −1 dτ
We start by computing the total derivatives of the two parametrisations
d~a d~b −τ
= − sin tı̂ + cos t̂, = ı̂ + √ ̂,
dt dτ 1 − τ2
and their magnitudes
  p   r
 d~a   d~b  τ2 1
  = sin2 t + cos2 t = 1,  
 dt   = 1+ = √ .
 dτ  1 − τ2 1 − τ2
√
The values of the field f along the two curves are f (~a(t)) = sin t and f (~b(τ )) = 1 − τ 2 . We compute the
two integrals and verify they give the same value:
Z π   Z π π

 d~a  
f ~a(t)   dt = sin t 1 dt = − cos t = 1 − (−1) = 2,
0 dt 0 0
Z 1   Z Z 1


 d~b  1 p 1
f ~
b(τ )   dτ = 1 − τ2 √ dτ = dτ = 2.
−1  dτ  −1 1 − τ2 −1

Note  that, not only the two parametrisations travel along C with different speeds (~a has constant speed
 d~a  = 1, while ~
b “accelerates” at the endpoints), but they also travel in opposite directions: ~a from right to
dt
left and ~b from left to right.
A. Moiola, University of Reading 35 Vector calculus lecture notes, 2016–17

̂
C

~a ~b

ı̂

Figure 17: The half circle parametrised by two curves ~a and ~b as in Exercise 2.3.

Exercise 2.4. ◮ Draw the following curves (helix, cubic, Archimedean spiral; compare with Figure 8)

ΓA defined by ~aA (t) = cos t ı̂ + sin t ̂ + tk̂, 0 ≤ t ≤ 2π,

3
t
ΓB defined by ~aB (t) =
ı̂ + tk̂, 0 ≤ t ≤ 1,
3
ΓC defined by ~aC (t) = t cos t ı̂ + t sin t ̂, 0 ≤ t ≤ 10,
R R R
and compute ΓA fA ds, ΓB fB ds and ΓC fC ds for the following fields:
s
p x2 + y 2
fA (~r) = x + y + z, fB (~r) = 9x2 + z 2 + y 3 , fC = .
1 + x2 + y 2

(Despite looking nasty, the third integral is quite easy to compute, find the trick!)
Example 2.5 (Length of a graph). Formula (43) can be used to compute the length of the graph of a real
function. Given a function g : [tI , tF ] → R, its graph can be represented by the curve ~a(t) = tı̂ + g(t)̂, whose
total derivative is d~
a ′
dt (t) = ı̂ + g (t)̂. Therefore, its length is
Z tF   Z tF Z tF q
 d~a 
2
length of the graph of g =   dt = |ı̂ + g ′
(t)̂| dt = 1 + g ′ (t) dt.
 dt 
tI tI tI

⋆ Remark 2.6 (Line integrals of densities in physics). If fR represents the density of a wire with shape Γ, line
integrals can be used to compute not only its total mass Γ f ds, but also its centre of mass (or barycentre)
R R R
( Γ xf ds)ı̂ + ( Γ yf ds)̂ + ( Γ zf ds)k̂ and its moment of inertia about a certain axis. You can find some
examples in Section 15.3 of the textbook [1].
Exercise 2.7. ◮ Consider the following two planar curves corresponding to the logarithmic and the algebraic
spirals shown in Figure 18:

~a(t) = e−t cos t ı̂ + e−t sin t ̂, t ∈ [0, ∞), ~b(τ ) = 1 cos τ ı̂ + 1 sin τ ̂, τ ∈ [1, ∞).
τ τ
1. Compute the length of the logarithmic spiral traced by ~a.
√
2. Show that the algebraic spiral traced by ~b has infinite length. (You might use that arsinh′ τ = 1/ 1 + τ 2 ,
but an easier way is to bound from below the integral obtained in a smart way.)
3. Show that both paths spin infinitely many times around the origin.
4. Given a positive number ℓ can you construct a spiral ~c that starts from ı̂, swirls infinitely many times
around the origin, eventually reaches ~0 and has length ℓ? What is the minimal ℓ possible?
You can find more exercises on path lengths and line integrals in Section 11.3 (exercises 13–20, p. 641)
and Section 15.3 (exercises 1–9, p.861) of the textbook [1].

2.1.2 Line integrals of vector fields

~ : R3 → R3 (we denote again by
Consider a curve ~a : I → R3 as before, now together with a vector field F
3 ~
Γ := ~a(I) ⊂ R the path of the curve ~a). The line integral of F along the path Γ is
Z Z Z tF

~ · d~r =
F ~ · d~a =
F ~ ~a(t) · d~a (t) dt.
F (44)
Γ Γ tI dt
A. Moiola, University of Reading 36 Vector calculus lecture notes, 2016–17

1
̂
0.8

0.6
~b(τ )
0.4 ~a(t)

0.2

ı̂
-0.2

-0.4
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

Figure 18: The logarithmic and the algebraic spirals of Exercise 2.7. We see that both swirls infinitely
many times around the origin but ~a approaches the origin itself much more quickly than ~b. Indeed the
path of ~a has finite length, while the length of the path of ~b is infinite.

The right-hand side of (44) is an ordinary one-dimensional integral (note that ~a, d~ a ~
dt and F are vector-
valued but the integral is scalar since it contains a scalar product). If ~a is a loop, then the line integral of
H
~ along Γ is called contour integral or circulation and denoted by the symbol
F ~ · d~r.
F
Γ
Similarly to scalar fields, different parametrisations of Γ give the same integral, if the travel direction
is the same. If the curve orientation is reversed, the sign of the integral changes (recall that the
sign does not change for line integrals of scalar fields!):
Z Z
~ ~
if b(t) = ~a(1 − t) then ~a([0, 1]) = b([0, 1]) = Γ and ~ ~ ~
F(b) · db = − F(~ ~ a) · d~a.
Γ Γ

~ ~
The sign change follows from ddtb (t) = − d~ a
dt (1 − t) (when b(t) = a(1 − t)). Example 2.10 demonstrates
~
the invariance of the line integral and the sign change. Since, up to orientation, all the R parametrisations
of Γ give the same integral for a given field, we can denote the line integral itself by Γ F ~ · d~r, forgetting
the dependence on the parametrisation ~a. In this notation, Γ is an oriented path, namely a path with
a given direction
R of travelling;
R if we want to consider the integral taken in the reverse direction we can
simply write −Γ F ~ · d~r = − F ~ · d~r. (In this notation, Γ and −Γ are two different oriented paths sharing
Γ
the same support.)
The integral in (44) represents the integral along Γ of the component of F ~ tangential to Γ
itself. To see this, for a smooth curve ~a : [tI , tF ] → R (under the assumption dt (t) 6= ~0 for all t), we
3 d~
a

define the unit tangent vector τ̂ :

da1 da2 da3
1 d~a dt ı̂ + dt ̂ + dt k̂
τ̂ = τ1 ı̂ + τ2 ̂ + τ3 k̂ =  d~a  = q


.
  dt da1 2 da2 2 da3 2
dt
dt + dt + dt

This is a vector field of unit length defined on Γ and tangent to it at every point. As mentioned, the line
~ is equal to the line integral of the scalar field (F
integral of the vector field F ~ · τ̂ ), the projection of F
~ on
the tangent vector of Γ:
Z Z tF Z tF  d~a   Z tF   Z
  
~ · d~r (44)
F = ~ · d~a dt =
F F~ ·  dt   d~a  dt = (F~ · τ̂ ) d~a  dt (41)
= ~ · τ̂ ) ds.
(F (45)
Γ tI dt tI
 d~a   dt  tI
 dt 
Γ
dt

This an important relation between the main definitions of Sections 2.1.1 and 2.1.2.
Remark 2.8. We introduce a notation that will be used in Section 3 and is often found in books. The line
integrals of a scalar field f with respect to the elements dx, dy and dz (over the oriented path Γ = ~a([tI , tF ]))
are defined as
Z Z Z Z tF
(45) (44)
da1
f dx := f ı̂ · d~r = f (ı̂ · τ̂ ) ds = f ~a(t) (t) dt,
Γ Γ Γ tI dt
A. Moiola, University of Reading 37 Vector calculus lecture notes, 2016–17
Z Z Z Z tF
(45)
da2
(44)
f dy := f ̂ · d~r = f (̂ · τ̂ ) ds =
f ~a(t) (t) dt, (46)
Γ Γ Γ tI dt
Z Z Z Z tF
(45) (44)
da3
f dz := f k̂ · d~r = f (k̂ · τ̂ ) ds = f ~a(t) (t) dt.
Γ Γ Γ tI dt

~ = F1 ı̂ + F2 ̂ + F3 k̂ as the sum of the integrals

Therefore, we can expand the line integral of a vector field F
of its components:
Z Z Z Z Z


~ · d~r =
F F1 dx + F2 dy + F3 dz = F1 dx + F2 dy + F3 dz . (47)
Γ Γ Γ Γ Γ

da1 (t)
One can also use the formal identity dx = dx dt dt, identify x = a1 (t) and obtain dx = dt dt, which gives
again the first equation in (46) (and similarly for the remaining two). R
If the path Γ can be parametrised by the coordinate x, then it is possible to write the integral Γ f dx more
explicitly. Assume Γ to be parametrised by ~a(x) = xı̂ + y(x)̂, for xL < x < xR and for a continuous function
yR : (xL , xR ) → R (here it is important that the parametrisation satisfies da dx = 1). Then, the line integral
1

Γ
f dx can be written as the one-dimensional integral in x of the field evaluated along the graph of y(x):
Z Z xR Z xR Z xR
(46)
d~a
da1

f dx = f ~a(x) ı̂ · (x) dx = f x, y(x) (x) dx = f x, y(x) dx. (48)
Γ xL dx xL dx xL
R
In other words, the integral Γ f dx is a standard integral in x if the path Γ is parametrised by a curve whose
first component is a1 (x) = x + λ for some λ ∈ R. This fact also extends to curves in three dimensions. If
the path is run in the opposite direction (right to left, or from higher to lower values of x) with a similar
parametrisation, then the sign of the integral is reversed, as expected.
⋆ ~ along a curve ~a, representing the
Remark 2.9 (Work). In physics, the line integral of a force field F
~
trajectory of a body, is the work done by F.
Example 2.10 (Integrals of a vector field along curves with common endpoints). Consider the following five
curves, all connecting the points ~u = ~0 and w
~ = ı̂ + ̂:

~aA (t) = tı̂ + t̂, 0 ≤ t ≤ 1,

π
~aB (t) = sin t ı̂ + sin t ̂, 0≤t≤ ,
2
~aC (t) = e−t ı̂ + e−t ̂, 0 ≤ t < ∞,
~aD (t) = t2 ı̂ + t4 ̂, 0 ≤ t ≤ 1,
(
tı̂, 0 ≤ t ≤ 1,
~aE (t) =
ı̂ + (t − 1)̂, 1 < t ≤ 2.

The curves ~aA and ~aB run along the diagonal of the square S = {~r s.t. 0 ≤ x ≤ 1, 0 ≤ y ≤ 1} at different
speeds; ~aC travels along the same diagonal but in the opposite direction and
R in infinite time; ~aD and ~aE trace
different paths (see Figure 19). Note that ~aE is not smooth. Compute Γn F ~ · d~an for F
~ = 2yı̂ − x̂ and
n = A, B, C, D, E, where Γn is the trajectory described by ~an (thus ΓA = ΓB = ΓC 6= ΓD 6= ΓE ).
We begin by computing the total derivatives of the curves
d~aA d~aB d~aC d~aD
= ı̂ + ̂, = cos t ı̂ + cos t ̂, = −e−t ı̂ − e−t ̂, = 2tı̂ + 4t3 ̂,
dt ( dt dt dt
d~aE ı̂ 0 ≤ t ≤ 1,
=
dt ̂ 1 < t ≤ 2.

Now we insert them in formula (44):

Z Z 1 Z 1 Z 1 1
~ · d~aA =


t2  1
F 2y(t)ı̂ − x(t)̂ · (ı̂ + ̂) dt = 2tı̂ − t̂ · (ı̂ + ̂) dt = t dt =  = ,
ΓA 0 0 0 2 0 2
Z Z π
2

~ · d~aB =
F 2 sin t ı̂ − sin t ̂ · (cos t ı̂ + cos t ̂) dt
ΓB 0
Z π Z π π
2 1 2 − cos 2t  2 1 1 1
= sin t cos t dt = sin 2t dt =  = 4+4= ,
0 0 2 4 0 2
Z Z ∞ Z ∞ ∞
~ · d~aC =

e−2t  0−1 1
F 2e−t ı̂ − e−t ̂ · (−e−t ı̂ − e−t ̂) dt = (−e−2t ) dt =  = =− ,
ΓC 0 0 2 0 2 2
A. Moiola, University of Reading 38 Vector calculus lecture notes, 2016–17
Z Z 1 Z 1
~ · d~aD =
F (2t4 ı̂ − t2 ̂) · (2tı̂ + 4t3 ̂) dt = (4t5 − 4t5 ) dt = 0,
ΓD 0 0
Z Z 1 Z 2 Z 2


~ · d~aE =
F (−t̂) · ı̂ dt + 2(t − 1)ı̂ − ̂ · ̂ dt = 0 + (−1) dt = −1,
ΓE 0 1 1

As expected, we have
Z Z Z Z Z
~ · d~aA =
F ~ · d~aB = −
F ~ · d~aC = 1 =
F 6 ~ · d~aD = 0 6=
F ~ · d~aE = −1,
F
ΓA ΓB ΓC 2 ΓD ΓE

since the first and the second are different parametrisations of the same path, the third is a parametrisation of
the same path with reverse direction, while the last two correspond to different paths.

~
F
1
~
w

0.8

0.6
ΓA

0.4 ΓD

0.2

~u ΓE
0

−0.2 0 0.2 0.4 0.6 0.8 1 1.2

Figure 19: The vector field F~ = 2yı̂ − x̂ and the curves described in Example 2.10. Note that F ~ is
R R
perpendicular to ΓD in each point (thus ΓD F~ · d~aD = 0 ds = 0) and to Γ E in its horizontal segment.
ΓD

~ = xı̂ + y 2 ̂ along the five curves described in Example 2.10.

Exercise 2.11. ◮ Compute the line integral of G
3
(Hint: recall the derivative of sin t.)
~ r) = ~r along the two spirals defined in
Exercise 2.12. ◮ Compute the line integrals of the position vector F(~
Exercise 2.7. Can you guess the signs of the integrals before computing them?
To practise more on line integrals of vector fields, see e.g. Problems 1–9 of [1, Section 15.4, p. 869].
⋆ Remark 2.13 (Line integrals of vector fields as limits of Riemann sums). An intuitive explanation of the
relation between formula (44) and its interpretation as integral of the tangential component of F, ~ can be
obtained by considering a discrete approximation of the line integral with Riemann sums (similarly to what we
have already done in Remark 2.1 for integrals of scalar fields). Consider the times 0 = t0 < t1 < · · · < tN −1 <
tN = 1, for i = 0, . . . , N , and the corresponding points ~ai = ~a(ti ) ∈ Γ. Define the increments δ~ai := ~ai −~ai−1
and δti := ti − ti−1 . Then, the integral of the component of F ~ tangential to Γ in the interval (ti−1 , ti ) is
~
approximated by F(~ai ) · δ~ai (recall the projection seen in Section 1.1.1), as depicted in Figure 20, and the
integral on Γ is approximated by the sum of these terms:
Z N
X N
X
~ a) · d~r ≈
F(~ ~ ai ) · δ~ai =
F(~ ~ ai ) · δ~ai δti .
F(~
Γ i=1 i=1
δti

In the limit N → ∞ (if the times ti are “well spaced” and ~a is smooth), δ~ai
~
δti tends to the total derivative of a
and the sum of δti gives the integral in dt, similarly to (42). Thus we recover formula (44).

2.1.3 Independence of path and line integrals for conservative fields

We consider the line integral of the gradient of a smooth scalar field ϕ along a curve ~a : [tI , tF ] → Γ ⊂ R3 .
Using the chain rule, since the function t 7→ ϕ(~a(t)) is a real function of real variable and the usual
fundamental theorem of calculus applies, we obtain
Z Z tF Z tF

(44)
d~a (37) d ϕ(~a)


~
(∇ϕ) · d~r = ~
∇ϕ ~a(t) · (t) dt = (t) dt = ϕ ~a(tF ) − ϕ ~a(tI ) .
Γ tI dt tI dt
A. Moiola, University of Reading 39 Vector calculus lecture notes, 2016–17

z
~a0 ~ a1 )
F(~
~a1 ~ ai )
F(~

Γ
δ~ai ~ai ~aN
~ai−1

~ along
Figure 20: A schematic representation of the approximation of the line integral of the vector field F
Γ using discrete increments (see Remark 2.13).

This is a simple but important result and we state it as a theorem. It is often called “fundamental theorem
of vector calculus” (or “gradient theorem”), in analogy to the similar theorem known from one-dimensional
calculus. This is the first relation between differential operators and integration we encounter; we will see
several others in Section 3 (see also the comparisons in Table 3 and in Appendix A).
~ with scalar
Theorem 2.14 (Fundamental theorem of vector calculus). Consider a conservative vector field F
potential ϕ and a (piecewise smooth) curve ~a : [tI , tF ] → Γ ⊂ R3 . Then
Z Z



~ · d~r =
F ~
(∇ϕ) · d~r = ϕ ~a(tF ) − ϕ ~a(tI ) . (49)
Γ Γ

The only information on the path Γ contained in the right-hand side of equation (49) are its endpoints
~ on it (once we know the
~a(tI ) and ~a(tF ): the actual path Γ is not needed to compute the line integral of F
scalar potential ϕ). This means that line integrals of conservative fields are independent of the
particular integration path, but depend only on the initial and the final points of integration. Thus,
given a conservative field F ~ and two points ~p and ~q in its domain, we can use the notation
Z ~
q Z
~ · d~r :=
F ~ · d~r,
F
~
p Γ

R ~q
where Γ is any path connecting ~p to ~q. (Note that ~
p is not well-defined for non-conservative fields.)
Rb
Comparison with scalar calculus 2.15. The fundamental theorem of calculus states that a g(t) dt = G(b) −
G(a) for all continuous real functions g : [a, b] → R, where G is a primitive of g, namely G′ = g. Similarly, the
R ~q
fundamental theorem of vector calculus gives ~p F ~ · d~r = ϕ(~a(tF )) − ϕ(~a(tI )), where ϕ is a scalar potential
~ So we can consider scalar potentials as the line-integral equivalent of primitives. Recall
of the vector field F.
also that both primitives and scalar potentials are defined up to constants.
Example 2.16. From Example 2.10 and formula (49), we immediately deduce that F ~ = 2yı̂ − x̂ is not
conservative, since integrals on different paths connecting the same endpoints give different values. The field
~ = xı̂ + y 2 ̂ of Exercise 2.11 might be conservative as its integral has the same value along three different
G
curves; however, in order to prove it is conservative we should proceed as we did in Example 1.67.
Exercise 2.17. ◮ Use formula (49), together with the results of Example 1.67 and Exercise 1.68, to compute
Z Z
~r · d~r and (zı̂ + xk̂) · d~r,
Γ Γ

p = ̂ + 2k̂ to the point ~q = −ı̂ − ̂ − k̂.

where Γ is any path connecting the point ~
In the next theorem we use path integrals to prove what was mentioned in Remark 1.70: while
the implication “conservative ⇒ irrotational” is true for all vector fields, its converse “irrotational ⇒
conservative” holds for fields defined on suitable domains. We define a star-shaped domain (or star-
like) to be a domain D such that for all ~r ∈ D the straight segment connecting the origin to ~r lies entirely
inside D. All convex shapes, such as parallelepipeds and balls, containing the origin are star-shaped.
A. Moiola, University of Reading 40 Vector calculus lecture notes, 2016–17

Theorem 2.18. Let D be a star-shaped domain, and F ~ be an irrotational vector field defined on D. Then, F
~
R
~
is conservative and a scalar potential is given by Φ(~r) = Γ~r F · d~r, where Γ~r is the segment connecting the
origin to ~r.
Proof. Given ~r ∈ D, ~r 6= 0, a parametrisation of the segment Γ~r is ~a(t) = t~r for 0 ≤ t ≤ 1 (recall Remark
1.24). We note that ∂F ∂F1 ∂F3 ∂F1 ~
∂x = ∂y and ∂x = ∂z because F is irrotational. Using this, together with product
2

and chain rule, we can compute the partial derivative of Φ in x:

Z
∂Φ ∂ ~ · d~r
(~r) = F (from the definition of Φ)
∂x ∂x Γ~r
Z 1
∂ ~ r) · ~r dt
= F(t~ (from the parametrisation of Γ~r )
∂x 0
Z 1
∂

= xF1 (t~r) + yF2 (t~r) + zF3 (t~r) dt
0 ∂x
Z 1



∂ F1 (t~r) ∂ F2 (t~r) ∂ F3 (t~r) 
= F1 (t~r) + x +y +z dt (from product rule)
0 ∂x ∂x ∂x
Z 1 
∂F1 ∂F2 ∂F3
= F1 (t~r) + tx (t~r) + ty (t~r) + tz (t~r) dt (from chain rule)
0 ∂x ∂x ∂x
Z 1 
∂F1 ∂F1 ∂F1 ~ ×F~ = ~0)
= F1 (t~r) + tx (t~r) + ty (t~r) + tz (t~r) dt (from ∇
0 ∂x ∂y ∂z
Z 1 
= ~ 1 (t~r) dt
F1 (t~r) + t~r · ∇F
0
Z  1 
d d~a
= F1 (t~r) + t F1 (t~r) dt (chain rule (37) and = ~r)
0 dt dt
Z 1  
d d
= t F1 (t~r) dt (from the product rule for )
0 dt dt
= 1F1 (~r) − 0F1 (~0) = F1 (~r).
~ = F,
Similar identities can easily be derived for the derivatives in y and z, so we obtain ∇Φ ~ i.e. Φ is a
~
scalar potential of F and this is conservative.
In next Theorem 2.19, we use the fundamental theorem of vector calculus 2.14 to characterise conser-
vative fields as those fields whose line integrals only depend on the endpoints. The proof of this theorem
also provides a construction of the scalar potential.
~ The following three conditions are equivalent:
Theorem 2.19. Consider a continuous vector field F.
~ is conservative;
(i) F
(ii) for all loops Γ I
~ · d~r = 0;
F
Γ

(iii) for all pairs of paths ΓA and ΓB with identical endpoints

Z Z
~ · d~r =
F ~ · d~r.
F
ΓA ΓB

⋆ Proof. [(i)⇒(ii)] If F ~ is conservative, then F ~ = ∇ϕ ~ for some scalar potential ϕ. If the loop Γ is
3
parametrised by ~a : [tI , tF ] → Γ ⊂ R with ~a(tI ) = ~a(tF ), then by the fundamental theorem of vector
H
calculus (49) we have Γ F ~ · d~r = ϕ(~a(tF )) − ϕ(~a(tI )) = 0.
[(ii)⇒(iii)] Assume condition (ii) is satisfied. The two paths ΓA and ΓB can be combined in a loop13
Γ = ΓA − ΓB (see Figure 21). Using the additivity property (40), the difference between the two path
integrals is then
Z Z Z Z I
~
F · d~r − ~
F · d~r = ~
F · d~r + ~
F · d~r = ~ · d~r = 0,
F
ΓA ΓB ΓA −ΓB Γ
13 As we have done in Section 2.1.2, here we consider all paths Γ to be “oriented” paths, i.e. paths equipped with a preferred
direction of travel (orientation). We denote by −Γ the same path as Γ with the opposite orientation. If Γα and Γβ are paths
such that the final point of Γα coincides with the initial point of Γβ , we denote by Γα + Γβ their “concatenation”, i.e., the
path corresponding to their union (obtained travelling first along Γα and then along Γβ ). Thus ΓA − ΓB is the concatenation
obtained travelling first along ΓA and then along ΓB , the latter run in the opposite direction,
A. Moiola, University of Reading 41 Vector calculus lecture notes, 2016–17

which is condition (iii).

R
[(iii)⇒(i)] Fix a point ~r0 in the domain of F.~ Then, for all points ~r, the scalar field Φ(~r) := ~r F~ · d~r
~r0
is well-defined (does not depend on the integration path). Consider the partial derivative
 Z ~r+hı̂ Z ~r 
∂Φ 1
1
(~r) = lim Φ(~r + hı̂) − Φ(~r) = lim ~
F · d~r − ~
F · d~r .
∂x h→0 h h→0 h ~r0 ~r0

Since in the integrals we can choose any path, in the first integral we consider the path passing through ~r
and moving in the x direction along the segment with endpoints ~r and ~r +hı̂. This segment is parametrised
by ~a(t) = ~r + tı̂ for 0 ≤ t ≤ h (which has total derivative d~
a
dt (t) = ı̂), thus
Z ~r Z ~r+hı̂ Z ~r 
∂Φ 1 ~ · d~r + ~ · d~r − ~ · d~r
(~r) = lim F F F
∂x h→0 h ~r0 ~r ~r0
Z ~r+hı̂
1 ~ · d~r
= lim F
h→0 h ~r
Z h
(44) 1 ~ r + tı̂) · ı̂ dt
= lim F(~
h→0 h 0
Z h
1

= lim F1 (x + t)ı̂ + y̂ + z k̂ dt
h→0 h
0


= lim F1 (x + τ )ı̂ + y̂ + z k̂ for some τ ∈ [0, h]
h→0
= F1 (~r)

(using, in the last two equalities, the integral version of the mean value theorem for real continuous
1
Rb
functions b−a a
f (t) dt = f (τ ) for a τ ∈ [a, b] and the continuity of F1 ). Similarly, ∂Φ ∂Φ
∂y = F2 and ∂z = F3 ,
~ = ∇Φ.
thus F ~ The vector field F ~ is conservative and Φ is a scalar potential.

z
ΓB

~p
ΓA ~q
y
x

Figure 21: Two paths ΓA and ΓB connecting the points ~p and ~q and defining a loop Γ = ΓA − ΓB , as in
Theorem 2.19.

We summarise what we learned here and in Section 1.5 about conservative fields in the following scheme:
Z ~
q I
~ = ∇ϕ
F ~ ~ · d~r
F ~ · d~r = 0
F ∇~ ×F ~ = ~0
⇐⇒ ~
p ⇐⇒ Γ =⇒ (50)
(conservative) ∀ loops Γ (irrotational)
is path independent

~ is star-shaped, by Theorem 2.18 the converse of the last implication holds.

If the domain of definition of F
Exercise 2.20. ◮ Consider the two oriented paths from the point ı̂ to the point ̂ (see Figure 22):
 
Γa := xı̂ + y̂, x + y = 1, x, y ≥ 0 , Γb := xı̂ + y̂, x2 + y 2 = 1, x, y ≥ 0 ,
R R
~ r) = (x2 + y 2 )(ı̂ + 2̂). Compute the line integrals
and the vector field F(~ ~ · d~r,
F ~ · d~r, and use the
F
Γa Γb
~ is not conservative.
values of these integrals to prove that F
⋆ Remark 2.21 (An irrotational, non-conservative field). In Remark 1.71, we have seen that the field

~ = −y x
F ı̂ + 2 ̂
x2 + y 2 x + y2

is conservative in the half-space D = {~r ∈ R3 s.t. x > 0} and we have claimed (without justification) that no
scalar potential can be defined in its larger domain of definition E = {~r ∈ R3 s.t. x2 + y 2 > 0}. We can easily
A. Moiola, University of Reading 42 Vector calculus lecture notes, 2016–17

y
̂
Γb

Γa

x
ı̂

Figure 22: The paths Γa and Γb in Exercise (2.20).

compute the circulation of F ~ along the unit circle C = {~r ∈ R3 , s.t. x2 + y 2 = 1, z = 0} (parametrised by
~a(t) = cos t ı̂ + sin t̂ with 0 ≤ t ≤ 2π):
Z Z 2π Z 2π
~ · d~r =
F (− sin tı̂ + cos t̂) · (− sin tı̂ + cos t̂) dt = 1 dt = 2π.
C 0 0

~ is not conservative in the domain E. (We

Since this integral is not equal to zero, Theorem 2.19 proves that F
have completed the proof of the claim made in Remark 1.71.)
⋆ Remark 2.22. From the results studied in this section we can learn the following idea. We already know
that any continuous real functions f : (tI , tF ) → R can be written as the derivative of another function
Rt
F : (tI , tF ) → R (the primitive of f , F (t) = t∗ f (s) ds for some t∗ ), even if sometimes computing F may be
very difficult. Thus, the integral of f = F ′ can be evaluated as the difference of two values of F . Similarly,
a scalar field g : D → R, D ⊂ R3 , can be written as a partial derivative (with respect to a chosen direction)
of another scalar field. On the other hand, only some vector fields can be written as gradients of a scalar field
(the potential) and their line integral computed as differences of two evaluations of the potential. These are
precisely the conservative fields, characterised by the conditions in box (50).

2.2 Multiple integrals

We have learned how to compute integrals of scalar and vector fields along curves. We now focus on
integrals on (flat and curved) surfaces and volumes.

2.2.1 Double integrals

For a two-dimensional region R ⊂ R2 and a scalar field f : R → R, we want to define and compute
ZZ
f (x, y) dA, (51)
R

namely the integral of f over R. The differential dA = dx dy represents the infinitesimal area element,
which is the product of the infinitesimal length elements in the x and the y directions. Since the domain
is two-dimensional, this RR
is commonly called double integral.
The double integral R f dA over a region R represents the (signed) volume of the portion of
space delimited by the graph surface of the two-dimensional field f and the xy-plane; see
Figures 23 and 24. “Signed” means that the contribution given by the part of R in which f is negative is
subtracted from the contribution from the part where f is positive.
In the same way as integrals of real functions are approximated by sums of areas of rectangles (recall
Figure 16), double integrals are approximated by sums of volumes of parallelepipeds, see Figure 24.
For simplicity, we first consider only two-dimensional domains which can be written as
n o
R = xı̂ + y̂ ∈ R2 , s.t. xL < x < xR , a(x) < y < b(x) , (52)

for two functions a(x) < b(x) defined in the interval (xL , xR ). The domains that can be expressed in this
way are called y-simple. These are the domains such that their intersection with any vertical line is either
a segment or the empty set. A prototypical example is shown in Figure 25. Examples of non-y-simple
domains are the C-shaped domain {~r s.t. 2y 2 − 1 < x < y 2 }, the “doughnut” {~r s.t. 1 < |~r| < 2} (draw
them!) and any other domain containing a hole. These and other more complicated non-y-simple domains
can be decomposed in the disjoint union of two or more y-simple or x-simple (similarly defined) domains.
A. Moiola, University of Reading 43 Vector calculus lecture notes, 2016–17

f (x, y)

R
x
RR
Figure 23: The double integral R f dA over a domain R (a triangle in the picture) represents the (signed)
volume delimited by the graph of the two-dimensional field f (the upper surface) and the xy-plane.

1.5 1.5

1 1

0.5 1 0.5 1

0.8 0.8

0 0.6 0 0.6
0 0
0.4 0.4
0.2 0.2
0.4 0.4
0.2 0.2
0.6 0.6
0.8 0.8
0 0
1 1

p
Figure 24: TheRRsurface plot of the scalar field f = x + y/2 over the square Q = (0, 1) × (0, 1). The
double integral Q f dA measures the volume shown in the left figure. If Q is partitioned in smaller squares
S1 , . . . , Sn , the double integral can be approximated by the sum over j = 1, . . . , n of the volumes
RR of the
parallelepipeds with basis the small squares Sj and height the value of f at one point ~x ∈ Sj : Q f dA ≈
Pn
j=1 f (~ xj )Area(Sj ). In this example the integral is approximated by n = 25 parallelepipeds. In the limit
n → ∞ (i.e. more smaller squares), the sum converges to the desired integral. This is how double integrals
are actually rigorously defined. RR P25 2.1 for the case of line
Recall√the similar argument made in Remark
integrals. (In this example, Q f dA = 2(35/2 −1−25/2)/15 ≈ 0.842, and j=1 f (~xj )Area(Sj ) ≈ 0.734.)

y = b(x)

dy dA = dx dy
dx

y = a(x)
x
xL xR

Figure 25: A y-simple two-dimensional domain R, bounded by the functions a and b in the interval
(xL , xR ), and the infinitesimal area element dA.
A. Moiola, University of Reading 44 Vector calculus lecture notes, 2016–17

In order to compute the integral (51), we view it as an iterated or nested integral.14 In other words,
for each value of x ∈ (xL , xR ), we first consider the one-dimensional integral of f along the segment parallel
to the y-axis contained in R, i.e.
Z b(x)
If (x) = f (x, y) dy.
a(x)
Note that both the integrand and the integration endpoints depend on x. The integral If is a function of
x, but is independent of y; we integrate it along the direction x:
ZZ Z xR Z xR Z b(x) 
f (x, y) dA = If (x) dx = f (x, y) dy dx. (53)
R xL xL a(x)

Of course, the role of x and y are exchanged in x-simple domains.

We consider as concrete example the isosceles triangle

Q := ~r ∈ R2 , s.t. 0 < y < 1 − |x| ; (54)
see Figure 26. In this case, xL = −1, xR = 1, a(x) = 0, b(x) = 1 − |x|. If we first integrate a field f along
the y-axis and then along the x-axis, we obtain:
ZZ Z 1  Z 1−|x|  Z 0  Z 1+x  Z 1  Z 1−x 
f (x, y) dA = f (x, y) dy dx = f (x, y) dy dx + f (x, y) dy dx.
Q −1 0 −1 0 0 0

We can also consider the same triangle as an x-simple domain and first integrate along the x-axis and
then along the y-axis (right plot in Figure 26):
ZZ Z 1  Z 1−y 
f (x, y) dA = f (x, y) dx dy.
Q 0 y−1

Of course, the two integrals will give the same value.

yR = 1

y =1+x y =1−x x=y−1 x=1−y

Q
Q

xL = −1 y=0 xR = 1 yL = 0
RR
Figure 26: The computation of the double integral Q f (x, y) dA on the triangle Q. The left plot represents
R 1 R 1−|x|
(
−1 0
f dy) dx, where the inner integral is taken along the shaded band (of infinitesimal width) and
the outer integral corresponds to the movement of the band alongR 1 the x axis. The right plot represent the
R 1−y
integration where the variables are taken in the opposite order 0 ( y−1 f dx) dy.

To have a more concrete example, we compute the integral of f = x − y 2 on the triangle Q in both
the ways described above:
ZZ Z 0  Z 1+x  Z 1  Z 1−x 
2 2 2
(x − y ) dA = (x − y ) dy dx + (x − y ) dy dx
Q −1 0 0 0
Z 1+x Z 1 1−x
0
1  1 
= xy − y 3  dx + xy − y 3  dx
−1 3 y=0 0 3 y=0
Z 0 Z 1
1
1 3 1
= − x3 + 1 dx + x − 2x2 + 2x − dx
−1 3 0 3 3
 0 1
1 4 1  1
4 2 3 2 1  1 1 1 2 1 1
= − x − x  + x − x +x − x  = − + − +1− =− ;
12 3 −1 12 3 3 0 12 3 12 3 3 6
ZZ Z 1  Z 1−y 
(x − y 2 ) dA = (x − y 2 ) dx dy
Q 0 y−1
Z 1 1 1−y Z 1 1
1 4 2 3  1
= 2
x − xy  2
dy = 3 2
2(y − y ) dy = y − y  = − .
0 2 x=y−1 0 2 3 0 6
14 You have already seen iterated integrals during the first year: see handout 5, from page 23 on, of the Calculus module.
A. Moiola, University of Reading 45 Vector calculus lecture notes, 2016–17

Exercise 2.23. ◮ Compute the following double integrals (drawing a sketch of the domain might be helpful):
ZZ

e3y sin x dx dy where R = (0, π) × (1, 2) = xı̂ + y̂ s.t. 0 < x < π, 1 < y < 2 ,
R
ZZ
y dx dy where Q is the triangle with vertices ~0, ̂, 5ı̂ + ̂,
Q
ZZ n o
π
cos x dx dy where S = xı̂ + y̂ s.t. 0 < x < , 0 < y < sin x .
S 2
You can find plenty of exercises on double integrals in Section 14.2 (p. 802, exercises 1–28) of [1].

2.2.2 Change of variables

Comparison with scalar calculus 2.24. We learned in first-year calculus how to compute one-dimensional
integrals by substitution of the integration variable, for example
Z 1 ξ(x)=1+ex Z ξ(1) Z 1+e p 1+e √
√ dξ=ex dx p dx 2 3 2 3 4 2
x
e 1 + ex dx = (ξ − 1) ξ dξ = ξ dξ = ξ 2  = (1 + e) 2 − ≈ 2.894.
0 ξ(0) dξ 2 3 2 3 3
How do we extend this technique to double integrals? In this one-dimensional example, the interval of inte-
gration x ∈ (0, 1) is mapped15 into a new interval (2, 1 + e) (whose variable is denoted by ξ) by the change
of variable T : x 7→ ξ = 1 + ex , which is designed to make “simpler” (more precisely: easier to integrate) the
function to be integrated. The factor dx/ dξ takes care of the “stretching” of the variable. In two dimensions,
we can use a change of variables not only to make the integrand simpler, but also to obtain a domain of
integration with a simpler shape (e.g. a rectangle instead of a curvilinear domain).
Consider a region R ⊂ R2 on which a two-dimensional, injective vector field T ~ : R → R2 is defined.
16
We can interpret this field as a “change of variables”, i.e. a deformation of the domain R into a new
~
domain T(R). We will informally call T ~ a “transformation” or “mapping” of the region R. Departing
from the usual notation, we denote by ξ(x, y) and η(x, y) (“xi” and “eta”) the components of T. ~ We
understand ξ and η as the Cartesian coordinates describing the transformed domain: R is a subset of the
~
plane described by x and y, T(R) is a subset of the plane described by ξ and η. We denote by x(ξ, η) and
y(ξ, η) the components of the inverse transformation T ~ −1 from the ξη-plane to the xy-plane17 :
~ : R → T(R)
T ~ ~ −1 : T(R)
T ~ →R
xı̂ + y̂ 7→ ξ(x, y)ξ̂ + η(x, y)η̂ ξ ξ̂ + η η̂ 7→ x(ξ, η)ı̂ + y(ξ, η)̂,

or, with a different notation,

(x, y) 7→ ξ(x, y), η(x, y) (ξ, η) 7→ x(ξ, η), y(ξ, η) .
~
For example, we consider the following (affine) change of variables18 T:
−x − y + 1 x−y+1
ξ(x, y) = , η(x, y) = ,
2 2 (55)
~ −1
whose inverse T corresponds to x(ξ, η) = −ξ + η, y(ξ, η) = −ξ − η + 1.

(The transformation T ~ associated to ξ and η is a combination of a translation of length 1 in the negative

√
y direction, an anti-clockwise rotation of angle 3π/4, and a shrinking of factor 2 in every direction.)
The triangle Q in (54) in the xy-plane is mapped by T ~ into the triangle with vertices ~0, ξ̂ and η̂ in the
ξη-plane, as shown in Figure 27. Other examples of changes of coordinates are displayed in Figure 28.
15 In case you are not familiar with the mathematical use of the verb “to map”, it is worth recalling it. Given a function

T defined in the set A and taking values in the set B (which we can simply write as T : A → B), we say that T “maps”
an element a ∈ A into b ∈ B if T (a) = b, and “maps” the subset C ⊂ A into T (C) ⊂ B. Thus, a function is often called a
“map” or a “mapping”; we will use this words to denote the functions related to changes of variables or operators. (If you
think about it, you see that a map in the non-mathematical meaning is nothing else than a function that associates to every
point in a piece of paper a point on the surface of the Earth. We will also define “charts” to be some special maps from a
flat surface to a curvilinear one.)
16 Another simple and useful interpretation of a change of variables is the following. Each point ~ r = xı̂ + y̂ ∈ R ⊂ R2 is
identified by two numbers, the coordinates x and y. A change of variables is simply a rule to associate to each point in R
other two numbers, the coordinates ξ and η.
17 Here and in the following, we use the vectors ξ̂ and η̂. They can be understood either as the unit vectors of the canonical

basis of the ξη-plane (exactly in the same role of ı̂ and ̂ in the xy-plane) or as vector fields with unit length, defined in R,
pointing in the direction of increase of ξ and η, respectively. The first interpretation is more helpful at this stage, while the
second will be useful in Section 2.3.
18 This change of variables is “affine”, meaning that its components ξ and η are polynomials of degree one in x and y.

Affine change of variables translate, rotate, dilate and stretch the coordinates but do not introduce curvature: straight lines
are mapped into straight lines.
A. Moiola, University of Reading 46 Vector calculus lecture notes, 2016–17

y η

−2ξ = x + y − 1 = 0

1
x
0

−
+
=

ξ
y
1
~

−
T

−
+

η
1
y
~

=
T(Q)

=
−
Q

y
x

=
0
x ξ
y=0 2η = x − y + 1 = 0

Figure 27: The triangle Q in the xy-plane and in the ξη-plane. Along the edges are shown the equations
of the corresponding lines.

1 η
1
0.8
0.8
0.6 0.6

0.4 0.4

0.2
0.2
0

0 ξ
0 0.5 1 0 0.5 1

η η
1.5 1

0.5

1 0

−0.5

−1
0.5
−1.5

−2
0 ξ ξ
0.5 1 1.5 2 −1 0 1 2

Figure 28: The unit square 0 < x, y < 1 (upper left) under three different mappings: ξ = x, η = y +
1 xy
5 sin(2πx) (upper right); ξ = (1 + x) cos y, η = (1 + x) sin y (lower left); ξ = e cos(2πy), η = exy sin(2πy)
(lower right).

A transformation T ~ : (x, y) 7→ (ξ, η) warps and stretches the plane, if we want to compute an integral
in the transformed variables we need to take this into account. The infinitesimal area element dA = dx dy
is modified by the transformation in the ξη-plane.
We recall that the Jacobian matrices of T ~ and T~ −1 are 2 × 2 matrix fields containing their first order
partial derivatives (see Section 1.3.3). We denote their determinants by

∂(ξ, η)

~ = ∂ξ ∂η − ∂ξ ∂η ,
:= det(J T) and
∂(x, y) ~ −1 ) = ∂x ∂y − ∂y ∂x .
:= det J(T (56)
∂(x, y) ∂x ∂y ∂y ∂x ∂(ξ, η) ∂ξ ∂η ∂ξ ∂η

These are called Jacobian determinants19 ; their absolute values are exactly the factors needed to
~
compute double integrals under the change of coordinates T:
ZZ ZZ  

 ∂(x, y) 
f (x, y) dx dy = 
f x(ξ, η), y(ξ, η)   dξ dη, (57)
R ~
T(R) ∂(ξ, η) 

for any scalar field f : R → R. This is the fundamental formula for the change of variables in a double
integral. In other words, we can say that the infinitesimal surface elements in the xy-plane and in the
19 In some books, when used as noun, the word “Jacobian” stands for Jacobian determinant, as opposed to Jacobian

matrix. Again, this can be a source of ambiguity.

A. Moiola, University of Reading 47 Vector calculus lecture notes, 2016–17

ξη-plane are related to each other by the formulas:

   
 ∂(x, y)   ∂(ξ, η) 
dx dy =   dξ dη, dξ dη =   dx dy.
∂(ξ, η)  ∂(x, y) 

We also have
∂(ξ, η) 1
= ∂(x,y)
. (58)
∂(x, y)
∂(ξ,η)

Formula (58) is often useful when the partial derivatives of only one of the transformations T~ and T ~ −1
∂(x,y)
are easy to compute. In other words, if you need ∂(ξ,η) for computing an integral via a change of variable
~ −1 , you can compute ∂(ξ,η) and then apply (58).
~ is easier to obtain than J(T)
as in (57), but J T ∂(x,y)

Comparison with scalar calculus 2.25. Recall that in the one-dimensional integration by substitution in 2.24
dξ d(ξ,η)
we use the factor dx . The corresponding factor for double integrals is the Jacobian determinant d(x,y) (56).

⋆ Remark 2.26. Note that here we are implicitly assuming that the Jacobian matrices of T ~ and T ~ −1 are
never singular, i.e. their determinants do not vanish in any point of the domain, otherwise equation (58) would
make no sense. Under this assumption, it is possible to prove that J(T ~ −1 ) = (J T)
~ −1 , namely the Jacobian
matrix of the inverse transform is equal to the inverse matrix of the Jacobian of T ~ itself; this is part of the
Inverse Function Theorem you might study in a future class.
We return to the affine transformations (55). Their Jacobian matrices are (recall the definition in (16))
 1 1
  
JT~ = −12 − 12 , J(T~ −1 ) = −1 1 .
2 −2 −1 −1

Note that this case is quite special: since the transformations are affine (namely polynomials of degree
one), their Jacobian are constant in the whole plane R2 . From this, we compute the Jacobian determinants
∂(ξ, η)  1  1  1  1  1 ∂(x, y)
= − − − − = , = (−1)(−1) − 1(−1) = 2. (59)
∂(x, y) 2 2 2 2 2 ∂(ξ, η)

Note that, as expected, ∂(x,y) ~

∂(ξ,η) = 2 is the ratio between the areas of Q and T(Q) (since in this simple case
the Jacobian of T~ and its determinant are constant in Q). Now we can easily integrate on the triangle Q
in (54) the scalar field f = (1 − x − y)5 , for example, using the change of variables (55) which maps Q
~
into T(Q) as in Figure 27:
ZZ ZZ  
(55),(57)  ∂(x, y) 
f (x, y) dx dy = f (−ξ + η, −ξ − η + 1)  dξ dη
Q ~
T(Q) ∂(ξ, η) 
ZZ
(59)
= (1 + ξ − η + ξ + η − 1)5 2 dξ dη
~
T(Q)
ZZ
= (2ξ)5 2 dξ dη
~
T(Q)
Z 1 Z 1−ξ 
(51) 5
= 64 ξ 1 dη dξ
0 0
Z 1
= 64 (ξ 5 − ξ 6 ) dξ
0
1 1
1  64
= 64 ξ 6 − ξ 7  = ≈ 1.524.
6 7 0 42

Remark 2.27. Note that in all the formulas related to change of variables the domain of integration (appearing
at subscript of the integral sign) must be a subset of the plane spanned RRby the integration variables (appearing in
the differential term d . . . d . . .). In the example above, the integral Q is always associated to the differential
RR
dx dy, meaning that Q is a set in the xy-plane; the integral T(Q) ~ always appears with the differential dξ dη,
~
as T(Q) is a set in the ξη-plane.
Example 2.28 (Area of regions). As we required inRR(40), the area of a domain R ⊂ R2 can be computed
by integrating on it the constant one: Area(R) = R 1 dx dy. If R is y-simple, one can use the iterated
integral (53). Otherwise, if R is complicated but can be mapped into a simpler shape, the change of variable
formula (57) can be used to compute its area.
A. Moiola, University of Reading 48 Vector calculus lecture notes, 2016–17

We now want to use formula (57) to compute the areas of the curvilinear domains plotted in Figure 28. Each
of them is obtained from the unit square S = {0 < x < 1, 0 < y < 1} (upper left plot) from the transformations
listed in the figure caption. We denote the domains by RUR , RLL , RLR (as upper right, lower left and lower
right plot) to distinguish one another, and use a similar notation for the change of variable transformations
and coordinates (e.g. T ~ UR : S → RUR ). In the first case (upper right plot), the transformation, its Jacobian
matrix and determinant are
 
1 ~ UR = 2 1 0 ∂(ξUR , ηUR )
ξUR = x, ηUR = y + sin(2πx), JT , = 1.
5 5 π cos(2πx) 1 ∂(x, y)
From this we obtain
ZZ ZZ Z 1 Z 1
∂(ξUR , ηUR )
Area(RUR ) = dξUR dηUR = dx dy = 1 dx dy = 1,
RU R ~ −1 (RU R )
T UR
∂(x, y) 0 0

(which we could have guessed from the picture!). For the second picture we have

ξLL = (1 + x) cos y, ηLL = (1 + x) sin y,

 
~ LL cos y −(1 + x) sin y ∂(ξLL , ηLL )
JT = , = 1 + x,
sin y (1 + x) cos y ∂(x, y)
which leads to
ZZ ZZ Z 1 Z 1 Z 1
∂(ξLL , ηLL ) 3
Area(RLL ) = dξLL dηLL = dx dy = (1 + x) dy dx = (1 + x) dx = .
RLL S ∂(x, y) 0 0 0 2

Exercise 2.29. ◮ Compute the area of the domain RLR in the lower-right plot of Figure 28, obtained from
the unit square S = {0 < x < 1, 0 < y < 1} via the change of variables

ξLR = exy cos(2πy), ηLR = exy sin(2πy).

(If you do it in the right way, i.e. with no mistakes, despite looking nasty the computation is actually easy).
Exercise 2.30. ◮ (i) Draw the quadrilateral Q ⊂ R2 and compute its area:

Q = xı̂ + y̂, such that x = (4 − η)ξ, y = (π − ξ)η, for 0 < ξ < 1, 0 < η < 1 .

(ii) More generally, given two constants a, b ≥ 2 compute the area of


Qab = xı̂ + y̂, such that x = (a − η)ξ, y = (b − ξ)η, for 0 < ξ < 1, 0 < η < 1 .

Exercise 2.31. ◮ Use the change of variables

(
ξ = ξ(x, y) = x4 + y − 1,
(60)
η = η(x, y) = y − x4 ,
RR
to compute the double integral P f (ξ, η) dξ dη, where f (ξ, η) = e−ξ−η , and where the two-dimensional
domain P in the ξη-plane is the square with vertices ξ̂, η̂, −ξ̂, −η̂.
⋆ Remark 2.32 (Intuitive justification of the change of variable formula). Where does the change of variable
formula (57) come from? Why are the infinitesimal surface elements related to the Jacobian determinant?
Consider the square S with vertices

~pSW = xı̂ + y̂, ~pSE = (x + h)ı̂ + y̂, ~pN W = xı̂ + (y + h)̂, ~pN E = (x + h)ı̂ + (y + h)̂,

where h > 0 is “very small”. Of course, it has area equal to |~pSE − ~pSW ||~pN W − ~pSW | = h2 . Under a
transformation (x, y) 7→ (ξ, η), the first three vertices are mapped into

~qSW = ξ(x, y)ξ̂ + η(x, y)η̂,

 ∂ξ   ∂η 
~qSE = ξ(x + h, y)ξ̂ + η(x + h, y)η̂ ≈ ξ(x, y) + h (x, y) ξ̂ + η(x, y) + h (x, y) η̂,
∂x ∂x
 ∂ξ   ∂η 
~qN W = ξ(x, y + h)ξ̂ + η(x, y + h)η̂ ≈ ξ(x, y) + h (x, y) ξˆ + η(x, y) + h (x, y) η̂,
∂y ∂y
where we have approximated the values of the fields ξ and η with their Taylor’s expansions centred a xı̂ + y̂
(which is a good approximation if h is small).
A. Moiola, University of Reading 49 Vector calculus lecture notes, 2016–17

y η
~pN W h ~pN E
~qN W
~qN E
S h
~qSW
~pSW ~pSE
= xı̂ + y̂ ~qSE

x ξ

Figure 29: As described in Remark 2.32, the square S of area h2 in the xy-plane is mapped into a
curvilinear shape in the ξη-plane. The Jacobian determinant (multiplied by h2 ) measures the area of the
dashed parallelogram in the ξη-plane, which approximates the area of the image of S.

We then approximate the image of S in the ξη-plane with the parallelogram with edges [~qSE , ~qSW ] and
[~qN W , ~qSW ], see Figure 29. Using the geometric characterisation of the vector product as in Section 1.1.2, we
see that this parallelogram has area equal to the magnitude of the vector product of two edges:
 
~





 ∂ξ
 ∂η   ∂ξ ∂η 
Area T(S) ≈ (~qSE − ~qSW ) × (~qN W − ~qSW ) =  h ξ̂ + h η̂ × h ξ̂ + h η̂ 
∂x ∂x ∂y ∂y
 
 ∂ξ ∂η ∂η ∂ξ 
= h2  −  (61)
∂x ∂y ∂x ∂y 
 
(56) 2  ∂(ξ, η) 
= h  .
∂(x, y) 
Therefore, the Jacobian determinant is the ratio between the area of the parallelogram approximating the image
of S, and the area of the square S itself. In the limit h → 0, the approximation error vanishes and the Jacobian
determinant is thus equal to the ratio between the “infinitesimal areas” in the xy-plane and in the ξη-plane.
When we perform the change of variables (x, y) 7→ (ξ, η) inside an integral, in order to preserve the value of
the integral itself, we need to replace the contribution given by every “infinitesimally small” square S with the
∂(ξ,η)
contribution given by its transform, which is the same as multiplying with ∂(x,y) .
To pass from an infinitesimal element to a finite domain, we use two-dimensional Riemann sums. Consider
a domain R partitioned in N disjoint small squares {Sj }j=1,...,N with area h2 ; for example R can be the
unit square in the upper left plot of Figure 28, where N = 25. The domain T(R) ~ obtained by a change of
~
variables (x, y) 7→ (ξ, η) is composed of N curvilinear subdomains {T(Sj )}j=1,...,N , see e.g. one of the other
three plots in Figure 28. Denoting by fj := f (~pj ) the value of the integrand f (a continuous scalar field) in
a point ~pj ∈ Sj , for all 1 ≤ j ≤ N , we can decompose the integral at the right-hand side of (57) over all the
subdomains and approximate each of them with the integral of the constant fj . Then, approximating the area
of each subdomain with (61), we obtain an approximation of the change of variables formula (57):
ZZ   XN ZZ  
 ∂(x, y)   ∂(x, y) 
f   dξ dη = f  
 ∂(ξ, η)  dξ dη
~
T(R) ∂(ξ, η)  j=1
~ j)
T(S

N  
X  ∂(x, y) 

≈ fj  ~ j)
(~pj ) Area T(S
j=1
∂(ξ, η)
N
X    
(61)  ∂(x, y)   ∂(ξ, η) 
≈ fj  (~pj ) h2  (~pj )
j=1
∂(ξ, η) ∂(x, y)
N
X ZZ
(58)
= f j h2 ≈ f dx dy.
j=1 R

If the partition of R in subdomains is sufficiently regular, this approximate identity can be made rigorous and
is an equality in the limit N → ∞.
Example 2.33 (Change of variables for y-simple domains). The change of variables formula (57) can be
immediately applied for computing integrals in y-simple domains, as an alternative to (53). Indeed, the
transformation with components
y − a(x)
ξ(x, y) = x, η(x, y) =
b(x) − a(x)
A. Moiola, University of Reading 50 Vector calculus lecture notes, 2016–17

maps the domain R of (52) into the rectangle


(xL , xR ) × (0, 1) = ξ ξ̂ + η η̂ s. t. xL < ξ < xR , 0 < η < 1 .
The inverse transformation is


x(ξ, η) = ξ, y(ξ, η) = b(ξ) − a(ξ) η + a(ξ),
whose Jacobian determinant reads
 
∂(x, y)  1 0 
 = b(ξ) − a(ξ).
= ′
∂(ξ, η) ηb (ξ) + (1 − η)a′ (ξ) b(ξ) − a(ξ)

For instance, we compute the integral of f (x, y) = y1 in the “smile” domain (see left plot in Figure 30):
n o
R = xı̂ + y̂ ∈ R2 , s.t. − 1 < x < 1, 2x2 − 2 < y < x2 − 1 ;

(note that a(x) = 2x2 − 2, b(x) = x2 − 1, xL = −1 and xR = 1, in the notation of (52)). Using the change
of variables formula (57), we have
ZZ ZZ   ∂(x, y) 
f (x, y) dx dy = f x(ξ, η), y(ξ, η)   dξ dη
R (−1,1)×(0,1) ∂(ξ, η) 
ZZ  


= f ξ, b(ξ) − a(ξ) η + a(ξ) b(ξ) − a(ξ) dξ dη
(−1,1)×(0,1)
ZZ  
= f ξ, (1 − ξ 2 )η + 2ξ 2 − 2 (1 − ξ 2 ) dξ dη
(−1,1)×(0,1)
ZZ
1
= 2 )(η − 2)
(1 − ξ 2 ) dξ dη
(−1,1)×(0,1) (1 − ξ
Z 1 Z 1 1
−1 
= dξ dη = 2 log(2 − η) = −2 log 2 ≈ −1.386.
−1 0 2−η 0

y0
−0.2
−0.4
−0.6
−1
−0.8
−1.5
−1
−2
−1.2
−2.5
−1.4
−3
−1.6 1
−3.5
−1.8
−4 0.5
−1 −0.5 0 0.5 1
x
−4.5
0
−5
0
−0.5
−0.5
−1
−1.5
−1

Figure 30: The “smile domain” described in Example 2.33 (left plot) and the field f = y1 (right plot).
(Note that the valuesR attained by the field approach −∞ at the two “tips” of the domain, however the
value of the integral D f dx dy is bounded since the tips are thin.)

Example 2.34 (Integrals in domains bounded by level sets). In order to apply the change of variable formula
to the computation of areas of domains delimited by four curvilinear paths, it is sometimes useful to express
the paths as level sets of some field. For instance, we want to compute the area of the domain R bounded by
the parabolas
y = x2 , y = 2x2 , x = y2, x = 3y 2
(see sketch in Figure 31). We note that these parabolas can be written as level curves for the two scalar fields
x2 y2
ξ(x, y) = and η(x, y) = .
y x
A. Moiola, University of Reading 51 Vector calculus lecture notes, 2016–17

1
In other words, ξ and η constitute a change of variables that transforms R in the rectangle 2 < ξ < 1 and
1
3 < η < 1 in the ξη-plane. The corresponding Jacobian determinants are
 
∂(ξ, η)  2 x − x2  ∂(x, y) 1
 y y 2
=  y2 y
 = 3, ⇒ = ,
∂(x, y) − 2 2  ∂(ξ, η) 3
x x

from which we compute the area of R

ZZ Z 1 Z 1
1 1
dx dy = dη dξ = .
R 1
2
1
3
3 9

1.4

1.2

1
y = 2x2

0.8 y = x2
x = y2
0.6
R
x = 3y 2
0.4

0.2

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4

Figure 31: The curvilinear quadrilateral R bounded by four parabolas as described in Example 2.34.

Remark 2.35 (How to find a change of variables?). So far we have seen several examples where a change of
variables T~ between two domains R and T(R) ~ ~ is not known,
was given. Often the explicit expression of T
and different exercises might be posed. Given two regions R and Q, how can we construct a suitable map
~ : R → Q? Or, given a single domain R, how can we find a change of variable T
T ~ such that the image T(R)
~
is particularly simple to be used as domain of integration (typically a rectangle)?
The possible changes of variables are never unique, so it is not possible to give a general answer to these
questions. (Note that this situation is very similar to that encountered for the parametrisation of a path, see
Remark 1.24.) The only requirements on T ~ are that it must be bijective from R to Q and that the Jacobian
~ ~ −1
matrices J T and J(T ) are invertible in all points. Under these conditions, we usually require the change of
variables to be “as simple as possible”, in particular to have Jacobian determinants that (when multiplied with
the given integrand, if this is given) are easy to integrate. Here we consider only some examples; to understand
them you need to draw pictures of the domains involved.

1. A typical technique to construct changes of variables consists in matching the edges and the vertices
of the two desired domains. For example, consider the triangles Q and T(Q) ~ in Figure 27. The line
{y = 1 − x} (or equivalently {x + y − 1 = 0}) is part of the boundary of Q and must be mapped to the
line {ξ = 0}. A way to ensure this is to choose a change of coordinates such that ξ = (x + y − 1)A, with
A to be chosen (in this case A will be a constant as the transformation is affine, more in general it might
be a function A(x, y)). Similarly, since the line {y = 1 + x} is mapped to the line {η = 0}, we have the
condition η = (x − y + 1)B. To fix A and B we look at the vertices of the triangles: ı̂ must be mapped
to η̂, thus plugging its coordinates x = 1 and y = 0 into η = (x − y + 1)B must give η = 1, which is
possible if B = 1/2. Similarly, the vertex −ı̂ must be mapped to ξ̂, giving A = −1/2. Substituting the
values of A and B found, we obtain the change of variables (55).
2. We consider another similar example. Let P = {0 < x < 1, 0 < y < 2 − x} ⊂ R2 be the trapezoid with
vertices ~0, ı̂, ı̂ + ̂, 2̂ (draw it). We look for a change of variables mapping P to the unit square (0, 1)2 .
The vertical lines {x = 0} and {x = 1} are mapped into {ξ = 0} and {ξ = 1}, so we can simply choose
ξ = x. The horizontal line {y = 0} is also preserved (i.e. mapped into {η = 0}) so we choose η = yA,
where now A will depend on x. The edge with equation {y = 2 − x} is mapped into {η = 1}, so we
A. Moiola, University of Reading 52 Vector calculus lecture notes, 2016–17

have 1 = η = yA = (2 − x)A so A(x) = 1/(2 − x) and the transformation, with its inverse, reads20
( (
ξ = x, x = ξ,
η = y/(2 − x), y = (2 − ξ)η.

3. If R has four edges, denoted ΓS , ΓE , ΓN and ΓW , we can look for two two-dimensional continuous fields
ξ, η : R → R2 such that each edge corresponds to a level set:
ΓS = {η = cS }, ΓE = {ξ = cE }, ΓN = {η = cN }, ΓW = {ξ = cW },
where cS < cN and cW < cE are real numbers. Then, if the map ξ(x, y)ξ̂ + η(x, y)η̂ is injective on R,
it is a change of variables with image the rectangle (cW , cE ) × (cS , cN ). A prototypical case is described
in Example 2.34.
4. If a domain is x- or y-simple (see (52)), it can be mapped to a rectangle with the procedure described
in Example 2.33
5. If the domain has some circular symmetry, polar coordinates can be used, as described in Section 2.3.1.
Another instructive example is in Exercise 2.36.
Exercise 2.36 (Assignment 3, MA2VC 2013). ◮ Consider the unit square S = (0, 1)2 = {xı̂ + y̂, 0 < x <
1, 0 < y < 1} and the triangle T = {ξ ξ̂ + η η̂, 0 < ξ < η < 1} with vertices ~0, η̂ and ξ̂ + η̂.
(i) Find a simple (bijective) change of variables (x, y) 7→ (ξ, η) that maps S into T .
Hint: consider a polynomial transformation that deforms RR the x variable only.
(ii) Use the change of variables you found to compute T ηξ dξ dη.

2.2.3 Triple integrals

All what we have said about double integrals immediately extends to triple integrals, namely integrals on
three-dimensional domains. It is not possible to have an easy geometrical interpretation of triple integrals
as that in Figure 25 for double integrals. The volume integral of a positive field f on a domain D may be
understood as total mass of a body of shape D and density f .
As before, triple integrals can be computed as iterated integrals. For instance, consider a “z-simple”
domain D ⊂ R3 , defined as
n o
D = xı̂ + y̂ + z k̂ s. t. xL < x < xR , a(x) < y < b(x), α(x, y) < z < β(x, y)

for two real numbers xL and xR , two real functions a and b, and two scalar fields (in two variables) α and
β. Then, the integral in D of a scalar field f can be written as
ZZZ Z xR Z b(x)  Z β(x,y)  !
f (x, y, z) dx dy dz = f (x, y, z) dz dy dx. (62)
D xL a(x) α(x,y)

The infinitesimal volume element is defined as dV = dx dy dz.

k̂

~0
̂

ı̂

Figure 32: The tetrahedron with vertices ~0, ı̂, ̂ and k̂.

For example, the tetrahedron B with vertices ~0, ı̂, ̂ and k̂ depicted in Figure 32 can be written as
n o
B = xı̂ + y̂ + z k̂ s. t. 0 < x < 1, 0 < y < 1 − x, 0 < z < 1 − x − y .
20 You can visualise this change of coordinates in Matlab/Octave using the function VCplotter.m (available on the course

web page), with the command: VCplotter(6, @(x,y) x, @(x,y) y*(2-x),0,1,0,1);

A. Moiola, University of Reading 53 Vector calculus lecture notes, 2016–17

In this example, xL = 0, xR = 1, a(x) = 0, b(x) = 1 − x, α(x, y) = 0 and β(x, y) = 1 − x − y.

The change of variable formula (57) extends to three dimensions in a straightforward fashion. If the
~ : D → T(D)
transformation T ~ has components
ξ(x, y, z), η(x, y, z), σ(x, y, z),
and its Jacobian determinant is
 
 ∂ξ ∂ξ ∂ξ 
 
 ∂x ∂y ∂z 

∂(ξ, η, σ)  ∂η 
~ =  ∂η
= det(J T)
∂η
,
∂(x, y, z)  ∂x ∂y ∂z 

 ∂σ ∂σ ∂σ 

 
∂x ∂y ∂z
then the change of variable formula reads
ZZZ Z  

 ∂(x, y, z) 
f (x, y, z) dx dy dz = 
f x(ξ, η, σ), y(ξ, η, σ), z(ξ, η, σ)   dξ dη dσ, (63)
D ~
T(D) ∂(ξ, η, σ) 

for any scalar field f defined in D.

Returning to the previous example, the tetrahedron B is the image of the unit cube 0 < x, y, z < 1
under the transformation T ~ with components

ξ = (1 − z)(1 − y)x, η = (1 − z)y, σ = z. (64)

The corresponding Jacobian is
 
(1 − z)(1 − y) −(1 − z)x −(1 − y)x
∂(ξ, η, σ)  
~ =
= det(J T) 0 (1 − z) −y  = (1 − y)(1 − z)2 .
∂(x, y, z)  
 0 0 1 

In order to demonstrate the formulas of this section, we compute the volume of the tetrahedron B
integrating the constant 1 first with the iterated integral (62) and then with the change of variables (63):
ZZZ
Vol(B) = 1 dx dy dz
B
Z Z 1 Z  !
1−x 1−x−y
(62)
= 1 dz dy dx
0 0 0
Z Z ! Z 1
1 1−x
1
1 1 1  1
= 1 − x − y dy dx = (1 − x)2 dx = x − x2 + x3  = ,
0 0 0 2 2 3 0 6
ZZZ
Vol(B) = 1 dξ dη dσ
Z ZB
Z
(63) ∂(ξ, η, σ)
= dx dy dz
~ −1 (B)
T ∂(x, y, z)
Z 1 Z 1Z 1 Z 1 Z 1 Z 1 
1 1 1
= (1 − y)(1 − z)2 dx dy dz = dx (1 − y) dy (1 − z)2 dz = 1 = .
0 0 0 0 0 0 2 3 6

2.2.4 Surface integrals

There is another kind of integrals we have not encountered yet: we have seen integrals along curves, on flat
surfaces, and on volumes, what is still missing are integrals on curved surfaces. This section introduces
surface integrals of scalar fields; Section 2.2.5 considers surface integrals of vector fields.
Comparison with scalar calculus 2.37. Surfaces might be thought as the two-dimensional generalisation of
paths: as paths are “curvilinear segments”, surfaces are “curvilinear planar regions”. Intuitively, paths and
surfaces are one- and two-dimensional subsets of R3 , respectively. As paths can be defined by two equations
in x, y and z, surfaces can be written with a single equation, e.g.

~r ∈ R3 , x2 + y 2 = 1, z = 0 is a path (the unit circle),
 3 2 2 2

~r ∈ R , x + y + z = 1 is a surface (the unit sphere).
However, we have seen in Section 2.1 that to compute integrals along a path Γ we need a parametrisation ~a,
i.e. a bijective function mapping an interval to the path itself. Similarly, computing surface integrals will require
a parametrisation, which in this case will be a map (denoted X) ~ from a planar region to the surface itself.
A. Moiola, University of Reading 54 Vector calculus lecture notes, 2016–17

We will consider surfaces satisfying the following definition.

Definition 2.38. We say that a set S ⊂ R3 is a parametric surface if there exist a two-dimensional open
~ : R → R3 such that S is the image of R under X,
region R ⊂ R2 and a continuous, injective vector field X ~ i.e.
3 ~ ~
S = {~r ∈ R such that ~r = X(~p), for some ~p ∈ R}. The field F is called parametrisation, or chart, of S.

We always tacitly assume that the field X ~ is smooth. For a given (non-empty) surface S there exist
infinitely many different parametrisations X. ~ The assumption that X ~ be injective (i.e. if ~p, ~q ∈ R are
~ ~
different ~p 6= ~q, then they have different images X(~p) 6= X(~q)) is needed to avoid self-intersections of S.

Remark 2.39. The three components X1 , X2 and X3 of the field X ~ are the components of the points ~r ∈ S,
~ with the letters
thus to avoid confusion we will usually denote the two input variables (living in R ⊂ R2 ) of X
~
u and w instead of x and y. So we will write X(u, ~
w) = X(uû + wŵ) = X1 (u, w)ı̂ + X2 (u, w)̂ + X3 (u, w)k̂.
The scalars u and w play a similar role to t for curves, but being two we can not interpret them as time.
Example 2.40. The graph of the scalar field in two variables f = x2 − y 2 in Figure 5 is a surface and is defined
~
by the equation S = {~r ∈ R3 , z = x2 − y 2 }. Thus a chart for it is X(u, w) = uı̂ + w̂ + (u2 − w2 )k̂. If
we want to consider the whole graph of f as an unbounded surface we take the region R as the whole of R2 .
If we want to define S to be only the bounded portion shown in Figure 5, we choose the region R to be the
square (−2, 2) × (−2, 2) = {|u|, |w| < 2}. If we want S to look like a Pringles crisp we choose a circular region
R = {u2 + w2 < 1}.
Example 2.41. The unit sphere {x2 + y 2 + z 2 = 1} is clearly a surface, but can not be written in the simple
framework of Definition 2.38. However the “punctured unit sphere” {x2 + y 2 + z 2 = 1, z 6= 1} i.e. the unit
sphere with the north pole point removed is a parametric surface. The chart X ~ : R2 → S is the famous
“stereographic projection” (see Figure 33):
u2 w2


~ uı̂ + w̂ + 4 + 4 − 1 k̂
X(u, w) = u2 w2
. (65)
1+ 4 + 4

Exercise 2.42. ◮ Show that the stereographic projection (65) maps R = R2 into the punctured unit sphere.
~ −1 : S → R2 .
Compute the inverse map X

k̂

~0
x
~
X(u, w)

(u, w)

Figure 33: A representation of the stereographic projection X~ : R2 → S in (65), S being the punctured unit
sphere. The figure shows the vertical section {y = 0}. We identify the uw-plane with the horizontal plane
{z = −1} (the lower horizontal line in the figure). A point ~p = uû + wŵ = uı̂ + w̂ − k̂ in this plane is
mapped by X ~ into the intersection between the unit sphere and the segment connecting ~p itself to the north
pole k̂. The point u = 0, w = 0 is mapped to the south pole −k̂, the points in the circle {u2 + w2 = 2} are
mapped to the equator {x2 + y 2 = 1, z = 0}, and the point “at infinity” is mapped to the north pole ~k.

Exercise 2.43. ◮ Fix two numbers 0 < b < a. Try to draw the surface defined by the chart
~
X(u, w) = (a + b cos w) cos uı̂ + (a + b cos w) sin u̂ + b sin wk̂ − π < u < π, −π < w < π.

From Example 2.41 and Exercise 2.43 we see that some simple surfaces, such as the sphere, cannot be
written as parametric surfaces according to Definition 2.38. A more general definition could be given to
include these surfaces but it is beyond the scope of this course. However, Example 2.40 suggests that this
definition includes a very important class of surfaces, those defined as graphs of fields in two variables.
We describe them in the next very important remark.
A. Moiola, University of Reading 55 Vector calculus lecture notes, 2016–17

Remark 2.44. Given a region R ⊂ R2 and a planar scalar field g : R → R, the graph of g is the surface

Sg = ~r ∈ R3 s.t. xı̂ + y̂ ∈ R, z = g(x, y) ⊂ R3 . (66)

A chart for Sg is given by the vector field X~ g (u, w) = uı̂ + w̂ + g(u, w)k̂, which is defined on R. In this case,
we can identify u with x and w with y. The main property of a graph surface is that for each point xı̂ + y̂
in the planar region R there is exactly one point in S that lies in the vertical line through xı̂ + y̂, namely the
point xı̂ + y̂ + h(x, y)k̂.
A graph surface can be written in brief as Sg = {z = g(x, y)}, meaning that Sg is the set of the points
~r whose coordinates x, y and z are solutions of the equation in braces. Some examples of graph surfaces are
the paraboloid of revolution {z = x2 + y 2 } (e.g. a satellitep dish is a portion of it), the hyperbolic paraboloid
{z = x − y } of Example 2.40, the upper half sphere {z = 1 − x2 − y 2 , x2 + y 2 < 1}. The entire sphere is
2 2

not a graph, since to each point xı̂ + y̂ in the plane, with x2 + y 2 < 1, correspond two different points on the
sphere; on the other hand the sphere can be written as union of two graphs, the north and south hemispheres.
Given a surface S defined by the chart X ~ and the region R, and given a point ~r0 = X(u ~ 0 , w0 ) ∈ S, the
~ i.e. ∂ ~
F ∂ ~
F
partial derivatives of the vector field F, ∂u (u0 , w0 ) and ∂w (u0 , w0 ) represent two tangent vectors
~
∂F ~
∂F
to S. Assuming that the tangent vectors ∂u and ∂w are linearly independent, it follows that their vector
~
∂X ~
∂X
product ∂u × ∂w is perpendicular to S.
Exercise 2.45. ◮ Consider the surface graph Sg for the field g(x, y) = 16xy(1 − x)(1 − y) defined in
R = (0, 1)2 = {xı̂ + y̂, 0 < x, y < 1}. Draw Sg and compute its tangent vector fields.

⋆ Remark 2.46 (Tangent plane to a surface). ~ 0 , w0 ) ∈ S,

Given a parametric surface S and a point ~r0 = X(u
the plane
n ~ ~ o
~ 0 , w0 ) + u ∂ X (u0 , w0 ) + w ∂ X (u0 , w0 ), for all u, w ∈ R
T S~r0 = X(u
∂u ∂w
is called the tangent plane to S in ~r0 . When we say that a vector is perpendicular to a surface S in a point
~r0 , what we actually mean is that it is perpendicular to T S~r0 . If ~r0 is placed on a vertex or an edge of S, then
~ is not differentiable at u0 û + w0 ŵ and no tangent plane nor perpendicular vectors can be defined.
X
For a graph surface Sg as in (66), using the chart X~ g (u, w) = uı̂ + w̂ + g(u, w)k̂, we see that the tangent
~
∂X ∂g ~
∂X ∂g
vectors are ∂ug = ı̂ + ∂u k̂ and ∂wg = ̂ + ∂w k̂. Thus, renaming x = u + u0 , y = w + w0 , x0 = u0 , y0 = w0 ,
the tangent plane can be written as
n 
 o
~
xı̂ + y̂ + g(x0 , y0 ) + ∇g(x 0 , y0 ) · (x − x0 )ı̂ + (y − y0 )̂ k̂, for all x, y ∈ R .

This coincides with the formula already obtained in Remark 1.40.

We now have all the tools needed to compute integrals over surfaces. Given a scalar field f defined on
~ and the region R, the integral of f over S is
a parametric surface S, given by the chart X

ZZ ZZ   ZZ  
 ∂X ~ 
 ∂ X 
~  ~ ∂X ~
~ ~
∂X 
f dS = f (X) ×  dA = f X(u, w)  (u, w) × (u, w) du dw. (67)
S R  ∂u ∂w  R  ∂u ∂w 

where the integral at the right-hand side is aRRdouble integral on the (flat) region R as those studied in
Section 2.2.1. Note that we used the symbol , since the domain of integration is two-dimensional.
The symbol dS denotes the infinitesimal area element on S; it is the “curvilinear analogue” of dA or
the “two-dimensional analogue” of ds. Formula (67) states that the ratio between the area element on
S and that on the parametric region R equals the magnitude of the vector product between the partial
derivatives of the chart.
~ : R → S and Y
The surface integral (67) is independent of the chart: if X ~ : Q → S are different charts
RR ~ ~ RR ~ ~
21 ~ ∂X ∂X
of the same surface S, then it is possible to prove that R f (X)| ∂u × ∂w | dA = Q f (Y)| ~ ∂Y ∂Y
RR ∂u × ∂w | dA,
thus the notation S f dS is well-defined (recall Example 2.3 for path integrals).
We can easily specialise formula (68) to the case of a graph surface. If Sg is the graph of a field g as
~
in (66), the chart X(u, w) = uı̂ + w̂ + g(u, w)k̂ has derivatives

~
∂X ∂g ~
∂X ∂g ~
∂X ~
∂X ∂g ∂g
= ı̂ + k̂, = ̂ + k̂, ⇒ × = − ı̂ − ̂ + k̂.
∂u ∂u ∂w ∂w ∂u ∂w ∂u ∂w
21 This ~ −1 ◦ X
follows from the fact that Y ~ : R → Q is a change of variables.
A. Moiola, University of Reading 56 Vector calculus lecture notes, 2016–17

Plugging this into the surface integral formula (67) and identifying u = x and w = y, we have that the
integral on Sg of a scalar field f can be computed as

ZZ ZZ q ZZ s

 ∂g 2  ∂g 2
f dS = ~ 2 dA =
f 1 + |∇g| f x, y, g(x, y) 1 + (x, y) + (x, y) dA, (68)
Sg R R ∂x ∂y

The area of a parametric surface (or of a graph) are computed by integrating the constant scalar field
f = 1 (recall the basic properties of integrals in (40) and the path length formula (43)):

Z Z  ~ 
~  ZZ q
 ∂X ∂X ~ 2 dx dy.
Area(S) =  ×  du dw, Area(Sg ) = 1 + |∇g| (69)
R  ∂u ∂w  R

q p
Remark 2.47. The measure factor 1 + |∇g| ~ 2 in (68) should recall you the similar coefficient 1 + (g ′ (t))2
we found in the computation of integrals along the graph of a real function in Example 2.5.
⋆ Remark 2.48 (Relation between change of variables and surface integrals). A two-dimensional change of
~ : R → T(R)
variables T ~ can be considered the chart of a “flat parametric surface” by identifying X ~ = T,~
~ ~ ∂(x,y)
u = ξ, w = ξ. Since X3 = 0, we have ∂∂u X
× ∂∂w
X
= ( ∂T 1 ∂T2 ∂T2 ∂T1
∂ξ ∂η − ∂ξ ∂η )k̂ = ∂(ξ,η) , so the change of variables
formula (57) is a special case of the surface integral formula (67). Thus a justification of the latter follows along
the lines of Remark 2.32. (In particular formula (61) shows how the vector product enters the area element.)
~
Exercise 2.49. ◮ Compute the area of the parametric surface defined by the chart X(u, w) = (u + w)ı̂ +
(u − w)̂ + k̂ on the region R = (0, 1)2 = 0 < u, w < 1.
Exercise 2.50. ◮ Compute the area of the surface
n 2 3 3

o
S = ~r ∈ R3 , z = x 2 + y 2 , 0 < x < 1, 0 < y < 1 .
3
Exercise 2.51. ◮ (i) Compute the area of the parametric surface defined in Exercise 2.43. You can choose
a = 2 and b = 1 for simplicity.
(ii) Compute the integral on S of the field f (~r) = (x2 + y 2 )−1/2 .
We will examine other examples of surface integrals in the next sections.

2.2.5 Unit normal fields, orientations and fluxes (surface integrals of vector fields)
Remark 2.52 (Physical motivation for the definition of flux). Imagine a pipe containing a fluid which moves
~ r) at each point ~r. Denote by S a section of the pipe,
with stationary (i.e. independent of time) velocity F(~
~
which we model as a surface. Given F and S, how can we compute the flux through the pipe, i.e. the volume
of fluid passing through the section S in a unit of time? This must be an integral, summing the contributions
from all portions of the section. Moreover, it must be independent of the tangential component of F ~ on S,
as particles moving tangentially do not cross the surface. We also need to be able to distinguish the amount
of flux crossing S in one direction from that crossing it in the opposite direction, thus to distinguish the two
“sides” of S. This situation motivates the following definitions.
(We consider for a moment general surfaces, even those not included in Definition 2.38.) For every
point ~r on a smooth surface S, it is possible to define exactly two unit vectors n̂A and n̂B that are
orthogonal to S in ~r. These two vectors are opposite to each other, i.e. n̂A = −n̂B , and they are called
unit normal vectors. If for every point ~r ∈ S we fix a unit normal vector n̂(~r) in a continuous fashion
(i.e. n̂ is a continuous unit normal vector field defined on S), then the pair (S, n̂) is called oriented
surface. Note that the pairs (S, n̂) and (S, −n̂) are two different oriented surfaces, even though they
occupy the same portion of the space.22
Not all surfaces admit a continuous unit normal vector field, the most famous example of non-orientable
surface is the Möbius strip shown in Figure 34. On the other hand, the surfaces in the following three
important families admit an orientation.
22 Oriented surfaces share many properties with the oriented paths seen in Section 2.1. Also in that case, a path Γ supports

two different orientations, corresponding to travel directions, leading to two different oriented paths. In both cases, the
orientation is important to integrate vector fields, while is not relevant for the integration of scalar fields; can you guess why?
A. Moiola, University of Reading 57 Vector calculus lecture notes, 2016–17

0.4
0.2
0
-0.2
-0.4

-1
1

-0.5
0.5

0
0

0.5
-0.5

1
-1

1.5

Figure 34: The Möbius strip. This surface is not orientable: if we continuously move a unit normal vector
along the surface, after one turn it will point in the direction opposite to the direction it started from. You
can easily build a Möbius strip with a strip of paper.
The chart X ~ = (1 + w cos(u/2)) cos u ı̂ + (1 + w cos(u/2)) sin u ̂ + w sin(u/2)k̂ for 0 < u < 2π and
−0.5 < w < 0.5 gives the Möbius strip without the the segment {0.5 < x < 1.5, y = z = 0} (in red in the
figure), corresponding to u = 0. Indeed, if we cut the surface along this segment, we cannot complete a
turn along the surface itself and the surface obtained is orientable.

~ : R → S, then it is orientable with

(i) If S is a parametric surface as in Definition 2.38 with chart X
unit normal vector
~
∂X ~
∂X
×
n̂(~r) =  ∂u ∂w  . (70)
~
 ∂X ~ 
∂X
 × 
 ∂u ∂w 

A special case is when Sg is the graph of a two-dimensional scalar field g as in (66), then (70) reads

∂g ∂g
− ı̂ − ̂ + k̂
∂x ∂y
n̂(~r) = s  2  2 . (71)
∂g ∂g
1+ +
∂x ∂y

Since the z-component of n̂ in this formula is positive, this is the unit normal that points upwards.
(ii) Another family of surfaces admitting a unit normal vector fields are the boundaries of three-
dimensional domains. The boundary of a domain D ⊂ R3 is commonly denoted by ∂D and is a
surface (if D is “smooth enough”). In this case, we usually fix the orientation on ∂D by choosing
the outward-pointing unit normal vector field.
~ 6= ~0 near S (see
(iii) If the surface S is the level surface of a smooth scalar field f satisfying ∇f
~ ~
Section 1.2.1), then n̂ = ∇f /|∇f | is an admissible unit normal field and (S, n̂) is an oriented
surface23 . This is a consequence of Part 4 of Proposition 1.33. We have seen a special case of this
situation in Example 1.38. These surfaces are defined by the equation S = {f (~r) = λ}.
Note that if S is the graph of g as in item (i), then it is also the level set of the field f (x, y, z) = z − g(x, y),
~ = − ∂g ı̂ − ∂g ̂ + k̂. So, the unit normal vector fields in (71) is a special case of both
whose gradient is ∇f ∂x ∂y
(70) and of item (iii) of the list above.
23 Can ~ = ~0?
you imagine what happens if ∇f
A. Moiola, University of Reading 58 Vector calculus lecture notes, 2016–17

⋆ Remark 2.53 (Normal unit vectors on piecewise smooth surfaces). The situation is more complicated when
the surface S is not smooth but only “piecewise smooth”, for instance is the boundary of a polyhedron. In
this case it is not possible to define a continuous unit normal vector field. For instance, on the boundary
∂C = {max{|x|, |y|, |z|} = 1} of the unit cube C = {max{|x|, |y|, |z|} ≤ 1}, the outward-pointing unit
normal vectors on two faces meeting at an edge are orthogonal to each other, so when crossing the edge they
suddenly “jump”, i.e. they are not continuous. However, it is possible to give a precise definition of orientation
also in this case, formalising the idea that n̂ stays on “the same side” of S (see [1, page 881]). In all practical
cases, with a bit of geometric intuition it should be clear how to define a normal field in such a way that it
stays on the same side of the surface (and so it determines the surface orientation).

Given an oriented surface (S, n̂) and a vector field F~ defined on S we define the flux of F
~ through
~
(S, n̂) as the value of the integral over S of the normal component of F:
ZZ ZZ
~ through (S, n̂) :=
flux of F ~ · d~S :=
F ~ · n̂ dS.
F (72)
S S

Note that the integrand F ~ · n̂ is nothing else than a scalar field defined on S, so the flux can be computed
as a surface integral studied in Section 2.2.4.
Combining the integral formula (67) and the expression (70) of the unit normal vector, we immediately
obtain the formula for the flux of a vector field through a parametric surface S with chart X ~ : R → S:

ZZ ZZ ~ ~
~ · d~S =
F F~ · ∂ X × ∂ X dA (73)
S R ∂u ∂w
ZZ

 ∂X
~ ∂X~ 
= F~ X(u,
~ w) · (u, w) × (u, w) du dw.
R ∂u ∂w
q
∂g 2
Similarly, on the graph surface (66) of g, the area element 1 + ( ∂x ) + ( ∂g 2
∂y ) in (68) and the denominator
in the unit normal (71) cancel each other and the integral in the flux (72) simplifies to24
ZZ ZZ  
~ · d~S = ∂g ∂g
F − F1 − F2 + F3 dA (74)
Sg R ∂x ∂y
ZZ  

∂g
∂g

= − F1 x, y, g(x, y) (x, y) − F2 x, y, g(x, y) (x, y) + F3 x, y, g(x, y) dx dy.
R ∂x ∂y
If the surface is a domain boundary ∂D, the surface integral and the flux are often denoted by the symbols
ZZ ZZ
f dS and F~ · d~S.
∂D ∂D
RR
Remark 2.54. The flux S F ~ · d~S measures “how much” of the vector field F ~ crosses the surface S in the
~ is tangent to the surface in every point then the flux is zero.
direction given by n̂. If F
~ = ~r × ı̂ = z̂ − y k̂ through the hyperbolic paraboloid
Exercise 2.55. ◮ Compute the flux of the vector field F
S = {z = x2 − y 2 , 0 < x, y < 1} (see Figure 5).
Exercise 2.56. ◮ Consider the graph surface Sg of Exercise 2.45 (i.e. g(x, y) = 16x(1 − x)y(1 − y) and
~ = ı̂ and G
R = (0, 1)2 ). Let F ~ = xı̂ be two vector fields. Equip Sg with the unit normal vector of (71) (you
do not need to compute n̂!). Compute the fluxes of F ~ and G ~ through Sg .
Can you guess the value of one of this fluxes and the sign of the other before computing them?
Exercise 2.57. ◮ Compute the flux of the position vector F(~~ r) = ~r through the parametric surface of
Exercise 2.43. You can assume again a = 2 and b = 1. Hint: the computations already done in Exercise 2.51
might be helpful.
A surface S may have a boundary ∂S, which is the path of a curve. For instance, the boundary of
the upper half sphere S = {|~r| = 1, z ≥ 0} is the unit circle ∂S = {xı̂ + y̂, x2 + y 2 = 1} (the equator),
while the complete sphere {|~r| = 1} has no boundary. The boundary of the open cylindrical surface
C = {x2 + y 2 = 1, |z| < 1} is composed of the two circles ∂C = {x2 + y 2 = 1, z = ±1}.
24 Note
RR ∂g
that when we write, for example, R F1 ∂x dA what we actually mean is the integral over the region R of

∂g
F1 x, y, g(x, y) ∂x (x, y): the two dimensional field g (giving the shape of the surface S) is defined on R ⊂ R2 , while
the three-dimensional field F1 is defined only on the surface S ⊂ R3 , whose points can be written as xı̂ + y̂ + g(x, y)k̂.
A. Moiola, University of Reading 59 Vector calculus lecture notes, 2016–17

We have defined orientations for both paths and surfaces, we now relate the orientation of a surface
with the that of its boundary. An oriented surface (S, n̂) “induces an orientation” on its boundary,
i.e. there is a standard way of choosing one of the two possible path orientations of its boundary. Given
S and n̂, the induced path orientation on ∂S is the travel direction such that walking along
∂S, on the side of S defined by n̂, we leave S itself on our left. This definition may be very
confusing, try to visualise it with a concrete example, e.g. using a sheet of paper as model for a surface;
see also Figure 35.

z n̂

S ∂S

Figure 35: The relation between the orientation of an oriented surface, given by the continuous unit
normal field n̂, and the orientation of the boundary ∂S (the direction of travel along the path). Imagine
to walk along the path ∂S, according to the arrows, and staying on the same side of the vector n̂ (i.e.
above the surface). Then S is always at our left. So the path-orientation of ∂S is that one induced by the
surface-orientation of S.

⋆ Remark 2.58. In Section 2 we have defined several different kinds of integrals. Single, double and triple
integrals, denoted by the differentials dt, dA, dV , are standard or iterated integrals computed on “flat” (1-,
2-, or 3-dimensional) domains. Line and surface integrals denoted by the differentials ds, d~r, dS, d~S require
a parametrisation describing the shape of the domain, which is a curve or a surface and might be not flat.
⋆ Remark 2.59 (The boundary of a boundary). Note that if the surface S is the boundary of a volume
D ⊂ R3 , i.e. S = ∂D, then S has empty boundary ∂S = ∅. You may find this fact stated as ∂ 2 = ∅, where
the symbol “∂” stands for the action “taking the boundary”. Try to think at some geometric examples.
Warning: do not mix up the different objects, the boundary ∂D of a volume D is a surface (the skin of an
apple), the boundary ∂S of a surface S is a path (the edge of a sheet, the border of a country).
⋆ Remark 2.60 (Boundaries and derivatives). Why are boundaries and derivatives denoted by the same
symbol “∂”? This notation comes from the profound theory of differential forms, however, we can see that
these two apparently unrelated objects share some properties. A fundamental property of derivatives is the
~ g) = (∇f
product rule: (f g)′ = f ′ g + f g ′ for scalar functions, or ∇(f ~ )g + f ∇g~ for scalar fields, (28).
Now consider the Cartesian product of the two segments [a, b] ⊂ R and [c, d] ⊂ R, namely the rectangle
R = [a, b] × [c, d] = {xı̂ + y̂, a ≤ x ≤ b, c ≤ y ≤ d}. Its boundary is the union of four edges [a, b] × {c},
[a, b]×{d}, {a}×[c, d], and {b}×[c, d]. The boundaries of the segments are ∂[a, b] = {a, b} and ∂[c, d] = {c, d}.
Therefore we can write






∂ [a, b] × [c, d] = [a, b] × {c} ∪ [a, b] × {d} ∪ {a} × [c, d] ∪ {b} × [c, d]



= [a, b] × {c, d} ∪ {a, b} × [c, d]



= [a, b] × ∂[c, d] ∪ ∂[a, b] × [c, d]

which closely resembles the product rule for derivatives. Here the Cartesian product (×) plays the role of multi-
plication, the set union (∪) that of addition and the action “take the boundary” plays the role of differentiation.
A similar formula holds true for much more general domains; try to see what is the boundary of the Cartesian
product of a circle and a segments, two circles, or a planar domain and a segment. If you are curious about
this analogy take a look at the post http://mathoverflow.net/questions/46252.
A. Moiola, University of Reading 60 Vector calculus lecture notes, 2016–17

2.3 Special coordinate systems

Among the possible changes of coordinates described in Section 2.2.2, some are particularly important.
Polar coordinates are used for two-dimensional problems with some circular symmetry around a centre,
while cylindrical and spherical coordinates are used in three dimensions. Several other special (more com-
plicated) coordinate systems exist, we do not consider them here. You should remember polar coordinates
from page 31 of handout 5 of last year calculus class. In Table 2 at the end of this section we summarise
the most important facts to keep in mind for the three system of coordinates studied here25 .
These systems of coordinates are meant to simplify computations in problems with some geometrical
symmetry. For example, cylindrical coordinates are related to axial symmetry and spherical coordinates
to central symmetry. So the former system will be useful to compute triple integrals on cylinders and
shapes like barrels or funnels; the latter for shapes like spheres or portions of them. On the other hand,
if you try to use these systems on different geometries, like tetrahedra or parallelepipeds, you will end up
with horrible computations, as they are not meant to be used in these cases.
In table 4 in Section G at the end of these notes, you can find a list of exercises from the textbook
where you can combine special coordinate systems with the techniques learned in the previous sections,
such as line integrals, multiple integrals, potentials.

2.3.1 Polar coordinates

The polar coordinates r and θ are defined by the relations26
( ( p
x = r cos θ, r = x2 + y 2 ,
y (75)
y = r sin θ, θ = tan−1 .
x
The xy-plane R2 is mapped by this transformation into the strip r ≥ 0, −π < θ ≤ π. Given a point
xı̂ + y̂, the value of r is the distance from the origin, while the value of θ is the (signed) angular distance
between the direction ı̂ of the x axis and the direction of the point itself. If r = 0, i.e. for the origin of
the Cartesian axes, the angular coordinate θ is not defined.

y θ = π/2
̂
θ̂ r̂
θ = 43 π

r sin θ ı̂
r ~r θ
θ π
x
r cos θ θ=0

~0 r
r=2 r=1
θ = − 14 π
−π

Figure 36: Left: the polar coordinate system.

Right: the change of variables (x, y) 7→ (r, θ) maps the punctured plane R2 \ {~0} to the semi-infinite strip
(0, ∞) × (−π, π] in the rθ-plane.

We compute the Jacobian of the direct and the inverse transformations:

!  
∂x ∂x
∂r ∂θ cos θ −r sin θ
∂y ∂y
= ,
sin θ r cos θ
∂r ∂θ
25 The page http://en.wikipedia.org/wiki/Del_in_cylindrical_and_spherical_coordinates contains a useful sum-

mary of these and many other identities and formulas (in 3D only). However, the notation used there is completely different
from that we have chosen in this notes: to compare those formulas with ours, r and ρ must be exchanged with each other
and similarly θ and φ! (See also footnote 29.) Our notation is chosen to be consistent with that of [1, Section 10.6].
26 Here we are being quite sloppy. The “inverse” of the tangent function is usually called arctan and takes values in (− π π ).
2 2
y y
Here tan−1 x is meant to be equal to arctan x only when x > 0 and to take values in (−π, π2 ] ∪ [ π2 , π] otherwise. For the
exact definition, see http://en.wikipedia.org/wiki/Atan2. However, θ(x, y) can be easily computed “visually” as the angle
determined by the point xı̂ + y̂. E.g. if x > 0, y > 0, then xı̂ + y̂ is in the first quadrant so θ belongs to the interval (0, π2 );
if x
e = −x < 0, ye = −y < 0 then x eı̂ + ye̂ belongs to the third quadrant and θe must be in the interval (−π, − π ), even if ye = y
2 e
x x
e
y y
and arctan e
x
= arctan x
.
A. Moiola, University of Reading 61 Vector calculus lecture notes, 2016–17
! !−1   !
∂r ∂r ∂x ∂x x y
∂x ∂y ∂r ∂θ 1 r cos θ r sin θ r r
= = = . (76)
∂θ
∂x
∂θ
∂y
∂y
∂r
∂y
∂θ
r − sin θ cos θ − ry2 x
r2

∂(tan−1 t) 1
(The second matrix can also be calculated directly using ∂t = 1+t2 .) The Jacobian determinants
are immediately computed as

∂(x, y) ∂(r, θ) 1 1
= r(cos2 θ + sin2 θ) = r, = = p . (77)
∂(r, θ) ∂(x, y) r x + y2
2

Therefore, the infinitesimal area element in polar coordinates is

dA = dx dy = r dr dθ.

Example 2.61 (Area of the disc). As a simple application of the polar coordinate system, we compute the
area of the disc of radius a > 0 centred at the origin, i.e. Q = {xı̂ + y̂ s.t. x2 + y 2 < a}:
ZZ Z aZ π Z a a
r2 
Area(Q) = dx dy = r dθ dr = 2πr dr = 2π  = πa2 ,
Q 0 −π 0 2 0

which agrees with the formula we know from elementary geometry.

In the “punctured plane” R2 \ {~0}, we define two vector fields r̂ and θ̂ of unit magnitude:

x y ~r
r̂(x, y) = cos θı̂ + sin θ̂ = p ı̂ + p ̂ = , ⇒ ı̂(r, θ) = cos θr̂ − sin θθ̂,
2
x +y 2 2
x +y 2 |~r|
(78)
y x
θ̂(x, y) = − sin θı̂ + cos θ̂ = − p ı̂ + p ̂, ̂(r, θ) = sin θr̂ + cos θθ̂.
x2 + y 2 x2 + y 2

In every point ~r 6= ~0 the two unit vectors r̂ and θ̂ are orthogonal one another and point in the direction of
increase of the radial coordinate r and of the angular one θ, respectively; see Figure 36. In other words, r̂
points away from the origin and θ̂ points in the anti-clockwise direction. Using (78) every planar vector
field defined in R2 \ {~0} can be expanded in r̂ and θ̂:

~ = F1 ı̂ + F2 ̂ = (cos θF1 + sin θF2 )r̂ + (− sin θF1 + cos θF2 )θ̂ = (F
F ~ · r̂)r̂ + (F
~ · θ̂)θ̂.

Example 2.62 (Curves in polar coordinates and area of domains delimited by polar graphs). If the path of a
curve ~a can be expressed in the form r = g(θ) for some positive function g, it is sometimes called a “polar
graph”. In this kind of curves, the magnitude is function of the angle and the path can be seen as the graph
of the function g in the rθ-plane. In this case we can choose t = θ and write ~a(t) = g(t)(cos tı̂ + sin t̂) for
−π < t ≤ π. For example, a circle of radius a can be written as r = a, i.e. the function g has constant value
a, thus we find the usual parametrisation ~a(t) = a(cos tı̂ + sin t̂).27
The domain R delimited by the curve ~a is “r-simple” in the rθ-plane (compare with the definition of
y-simple domain in (52)):
n p
o
R = ~r = xı̂ + y̂ ∈ R2 , s.t. x2 + y 2 < g θ(x, y)
n o
= ~r = r cos θı̂ + r sin θ̂ ∈ R2 , s.t. − π < θ ≤ π, 0 ≤ r < g(θ) .

R is called “star-shaped region”28 and its area can be computed as

ZZ Z π Z g(θ)  Z π g(θ) Z
1 2  1 π 2
Area(R) = dx dy = r dr dθ = r  dθ = g (θ) dθ. (79)
R −π 0 −π 2 0 2 −π

For example, we compute the area of the domain delimited by the “propeller” curve r = (2 + cos 3θ)
depicted in left plot of Figure 37: using cos2 t = 21 (t + sin t cos t)′ ,

1
Z π
2  θ sin 3θ cos 3θ 2 π 9
Area(R) = 2 + cos 3θ dθ = 2θ + + + sin 3θ  = π.
2 −π 4 12 3 −π 2
27 The “clover” curve ~c(t) = (cos 3t)(cos t)ı̂ + (cos 3t)(sin t)̂ in Figure 8 might be written in polar form as r = cos 3θ. Note
that here we are allowing negative values for the radial component! What does it mean? Compare with the curve plot.
28 We have already defined star-shaped domains in Section 2.1.3, can you see the relation between the two definitions?
A. Moiola, University of Reading 62 Vector calculus lecture notes, 2016–17

(We can deduce from this computation that the area delimited by r = (2 + cos nθ) is independent of n ∈ N,
i.e. all the “flower” domains in the centre plot of Figure 37 have the same area.) The path of the curve
r = (2 + cos 3θ) in the rθ-plane is the wave in the right plot of Figure 36.
We can also compute the area of more complicated shapes, for instance the domain between the two spirals
r = θ and r = 2θ (with 0 < θ < 2π) in the right plot of Figure 37:
Z 2π  Z 2θ  Z 2π
1
(2π)3
r dr dθ = 4θ2 − θ2 dθ = = 4π 3 ≈ 124.
0 θ 0 2 2
90 4
120 60
150 2 30

180 0

210 330
240 300
270
90 4 90 20
120 60 120 60
90 4
150 2 30 120 60 150 10 30
150 2 30

180 0 180 0 180 0

210 330
210 330 240 300 210 330
270
240 300 240 300
270 270
90 4
120 60
150 2 30

180 0

210 330
240 300
270

Figure 37: The polar graph r = (2 + cos nθ) for n = 3 (left), n = 2 (upper centre), n = 7 (centre centre),
n = 12 (lower centre). As seen in Example 2.62, they all have the same area. In the right plot, the two
spirals r = θ and r = 2θ (with 0 < θ < 2π). These curves can be plotted with Matlab’s command polar.

⋆ Remark 2.63. Any scalar field in two variables f (x, y) can be expressed in polar coordinates as F (r, θ) =
f (r cos θ, r sin θ). Here F is a function of two variables (as f ) which represents the field f but has a different
functional expression (we have already used this fact several times without spelling it out). The gradient and
the Laplacian of f can be computed in polar coordinates using the chain rule:

~ (x, y) = ∂f (x, y) ı̂ + ∂f (x, y) ̂

∇f
∂x ∂y



∂F r(x, y), θ(x, y) ∂F r(x, y), θ(x, y)
= ı̂ + ̂
∂x ∂y
   
∂F ∂r ∂F ∂θ ∂F ∂r ∂F ∂θ
= + ı̂ + + ̂ (chain rule (38))
∂r ∂x ∂θ ∂x ∂r ∂y ∂θ ∂y
   
∂F ∂r ∂r ∂F ∂θ ∂θ
= ı̂ + ̂ + ı̂ + ̂
∂r ∂x ∂y ∂θ ∂x ∂y
   
(76) ∂F x y ∂F −y x (78) ∂F 1 ∂F ~ (r, θ).
= ı̂ + ̂ + ı̂ + ̂ = r̂ + θ̂ = ∇F (80)
∂r r r ∂θ r2 r2 ∂r r ∂θ
Thus, whenever we have a scalar field F expressed in polar coordinates, we can compute its gradient using
formula (80) (do not forget the 1r factor!) without the need of transforming the field in Cartesian coordinates.

2.3.2 Cylindrical coordinates

There are two possible extensions of polar coordinates to three dimensions. The first one is used to treat
problems with some axial, or cylindrical, symmetry. By convention, the axis of symmetry is fixed to be
the axis z. The cylindrical coordinate r, θ and z corresponds to the polar coordinates r, θ in the xy-plane
and the height z above this plane:
  p
x = r cos θ, 
r = x2 + y 2 ,
  y r ≥ 0,
y = r sin θ, θ = tan−1 , −π < θ ≤ π, (81)

 
 x
z = z,  z = z, z ∈ R.
A. Moiola, University of Reading 63 Vector calculus lecture notes, 2016–17

Proceeding as in (77), it is immediate to verify that the Jacobian determinants and the infinitesimal
volume element are
∂(x, y, z) ∂(r, θ, z) 1
= r, = , dV = dx dy dz = r dr dθ dz. (82)
∂(r, θ, z) ∂(x, y, z) r
This can be used to compute the volume of the solids of revolution

Dg = ~r ∈ R3 s.t. zbot < z < ztop , r < g(z) , (83)
where g : (zbot , ztop ) → R is a non-negative function, as
ZZZ Z ztop  Z π  Z g(z)   Z ztop Z π  Z ztop
g 2 (z)
Vol(Dg ) = dV = r dr dθ dz = dθ dz = π g 2 (z) dz.
Dg zbot −π 0 zbot −π 2 zbot
(84)
Example
 2.64 (Volume of solids of revolution). Compute the volume of the “pint” domain E in Figure 38:
E = ~r ∈ R3 s.t. 0 < z < 1, r < 13 + 21 z 2 − 31 z 3 .
Z 1 2 Z 1 
(84) 1 1 2 1 3 1 1 4 1 6 1 2 2 3 1 5
Vol(E) = π + z − z dz = π + z + z + z − z − z dz
0 3 2 3 9 4 9 3 9 3
0 
1 1 1 1 1 1 223
=π + + + − − = ≈ 0.177.
9 20 63 9 18 18 1260

r
~r

θ
x

Figure 38: The domains described in Examples 2.64, 2.65 and Exercise 2.66: the pint E (left), the ellipsoid
x2 + y 2 + 4z 2 < 1 (upper centre), the rugby ball B (lower centre) and the funnel F (right).

Exercise 2.65 (Volume of the ellipsoid).◮ Compute the volume of the “oblate ellipsoid of revolution” bounded
2
by the surface x2 + y 2 + zc2 = 1, for a real number c > 0.
Exercise 2.66. ◮ Calculate the volume of the rugby ball B and the (infinite) funnel F in Figure 38:
 
B = ~r ∈ R3 s.t. − 1 < z < 1, r < 1 − z 2 , F = ~r ∈ R3 s.t. − ∞ < z < 0, r < ez .
Exercise 2.67. ◮ Cylindrical coordinates can  also be used to deal with domains which are not solids of
revolution. Can you draw the shape of D = ~r ∈ R3 s.t. − π2 < z < π2 , r < (cos z)(2 + sin 3θ) ? Can you
compute its volume? You need to derive a slightly more general formula than (84) (compare with formula (79)
for the area of polar graphs).

Exercise 2.68 (Triple integrals on a cone). ◮ Consider the cone C = ~r ∈ R3 s.t. 0 < z < 1, r < z . Verify
the following computations:
ZZZ ZZZ ZZZ
π 3π
z dV = , x dV = 0, (x2 + y 2 + z 2 ) dV = .
C 4 C C 10
(The domain is easily defined in cylindrical coordinates, while the last two integrands are defined in Cartesian
coordinates, figure out how to deal with this fact.) Prove that, for any function g : (0, 1) → R,
ZZZ Z 1
g(z) dV = π g(z)z 2 dz.
C 0
A. Moiola, University of Reading 64 Vector calculus lecture notes, 2016–17

⋆ Remark 2.69. As we did in (78) for the polar coordinates, also for the cylindrical system we can compute
the vector fields with unit length lying in the direction of increase of the three coordinates:
x y
r̂(x, y, z) = p ı̂ + p ̂, ı̂ = cos θr̂ − sin θθ̂,
2
x +y 2 x + y2
2
y x (85)
θ̂(x, y, z) = − p ı̂ + p ̂, ̂ = sin θr̂ + cos θθ̂,
2
x +y 2 x + y2
2

ẑ(x, y, z) = k̂, k̂ = ẑ.

⋆ Remark 2.70 (Vector differential operators in cylindrical coordinates). As we did in equation (80) for
polar coordinates, we can compute the gradient of a scalar field f expressed in cylindrical coordinates. Since
we are now considering a three-dimensional space, we also compute divergence and curl of a vector field
~ = Gr r̂ + Gθ θ̂ + Gz ẑ. We only write the final results (they can be derived from (81), (85) and the chain
G
rule, but the computation are quite complicated):

~ (r, θ, z) = ∂f r̂ + 1 ∂f θ̂ + ∂f ẑ,
∇f
∂r r ∂θ ∂z
~ · G(r,
~ 1 ∂(rG r ) 1 ∂F θ ∂Fz ∂Gr 1 1 ∂Fθ ∂Fz
∇ θ, z) = + + = + Gr + + , (86)
r ∂r r ∂θ ∂z ∂r r r ∂θ ∂z
 1 ∂F ∂Fθ   ∂Fr ∂Fz   ∂Fθ 1 1 ∂Fr 
~ × G(r,
~ z
∇ θ, z) = − r̂ + − θ̂ + + Fθ − ẑ.
r ∂θ ∂z ∂z ∂r ∂r r r ∂θ
Exercise 2.71. ◮ Consider the vector field F ~ = −yı̂ + x̂ + k̂. Show that in cylindrical coordinates it reads
~ ~
F = rθ̂ + ẑ. Use (86) to prove that F is solenoidal. This might be the field that gave the name to all
solenoidal fields: the streamlines of F~ are the helices (see Figure 8), which have the shape of a “solenoid” used
in electromagnetism (although Wikipedia suggests a different origin of the name).
Example 2.72 (Surface integrals in cylindrical coordinates). Polar and cylindrical coordinates can be used to
compute surface integrals. Consider the conic surface of unit radius and unit height (written in Cartesian and
cylindrical coordinates):
n p o n o
S = ~r ∈ R3 , z = x2 + y 2 , x2 + y 2 < 1 = ~r ∈ R3 , z = r, r < 1 .

(Note that S is part of boundary of the cone C in Exercise 2.68.) The surface S is the graph of the two-
dimensional field g(r, θ) = r defined on the unit disc Q = {~r ∈ R3 , r < 1}. Thus, its area can be computed
usingpthe graph surface integral formula (68), choosing f = 1. We also exploit the fact that the gradient of
g = x2 + y 2 = r satisfies ∇g~ = ∂r r̂ = r̂, when written in polar coordinates, see (80), so it has length one.
∂r
We obtain:
ZZ q Z π Z 1 √ Z 1 √
(68) √
Area(S) = ~ 2 dA =
1 1 + |∇g| 1 + 1 r dr dθ = 2π 2 r dr = π 2 ≈ 4.443,
Q −π 0 0

~ = |r̂| = 1.
where in the second equality we used |∇g|

2.3.3 Spherical coordinates

In order to deal with three-dimensional situations involving a centre of symmetry, we introduce the
spherical coordinate system. The three coordinates, denoted by ρ, φ and θ, are called radius, colatitude
and longitude (or azimuth), respectively29 . (The latitude commonly used in geography corresponds to
π
2 − φ, hence the name colatitude.) For a point r, the radius ρ ≥ 0 is its distance from the origin (the
~
magnitude of ~r); the colatitude 0 ≤ φ ≤ π is the angle between the direction of ~r and the unit vector k̂
on the axis z; the longitude −π < θ ≤ π is the polar angle of the projection of ~r on the xy-plane. In
formulas:30
 p 

 ρ = |~
r | = x2 + y 2 + z 2 , 
 x = ρ sin φ cos θ, ρ ≥ 0,
φ = arccos zρ = arccos √ 2 z 2 2 , y = ρ sin φ sin θ, 0 ≤ φ ≤ π, (87)
 x +y +z 

θ = tan−1 y ,  −π < θ ≤ π.
z = ρ cos φ,
x

29 We assign to colatitude and longitude the same symbols used in the textbook (see [1, Page 598]). However, in many

other references, the symbols θ and φ are swapped, e.g. in Calvin Smith’s “vector calculus primer” you might have used.
According to http://en.wikipedia.org/wiki/Spherical_coordinates, the first notation is more common in mathematics
and the second in physics. Different books, pages of Wikipedia and websites use different conventions: this might be a
continuous source of mistakes, watch out!
Also the naming of the radial variable can be an issue: we use ρ for spherical coordinates and r for cylindrical as
in [1, Section 10.6], but, for instance, most pages of Wikipedia swap the two letters.
30 You do not need to remember by hearth the formulas in (87). On the other hand, if you know their geometrical meaning

it is easy to derive them using some basic trigonometry. You need to remember the formula of the volume element (88).
A. Moiola, University of Reading 65 Vector calculus lecture notes, 2016–17

They are depicted in Figure 39. Spherical and cylindrical coordinate systems are related one another by
the following formulas (note that the variable θ plays exactly the same role in the two cases):
 √ 
 2 2 
ρ = r + z , r = ρ sin φ,
φ = arctan zr , θ = θ,

 

θ = θ, z = ρ cos φ.

The corresponding Jacobian matrix and the two determinants are:

 ∂x ∂x ∂x   
∂ρ ∂φ ∂θ sin φ cos θ ρ cos φ cos θ −ρ sin φ sin θ
 ∂y ∂y ∂y  
 ∂ρ ∂φ ∂θ  = sin φ sin θ ρ cos φ sin θ ρ sin φ cos θ  ,
∂z ∂z ∂z cos φ −ρ sin φ 0
∂ρ ∂φ ∂θ
∂(x, y, z) ∂(ρ, φ, θ) 1
⇒ = ρ2 sin φ, = 2 , dV = dx dy dz = ρ2 sin φ dρ dφ dθ. (88)
∂(ρ, φ, θ) ∂(x, y, z) ρ sin φ

Exercise 2.73. Verify the derivation of the Jacobian determinant in (88).

z
ρ̂

θ̂
~r

φ̂
φ ρ
ρ cos φ
y
ρ sin φ
θ ρ sin φ cos θ

ρ sin φ sin θ

Figure 39: A representation of the spherical coordinates ρ, φ and θ for the position vector ~r ∈ R3 . Note the
angles φ and θ, it is important to understand well their definition. The big sphere is the set of points with
equal radius ρ. The dashed horizontal circle is the set of points with equal radius ρ and equal colatitude
φ. The dashed vertical half-circle is the set of points with equal radius ρ and equal longitude θ. The set
of points with equal colatitude φ and equal longitude θ is the half line starting at the origin and passing
through ~r. The unit vectors ρ̂, φ̂ and θ̂ point in the directions of increase of the corresponding coordinates
(see formulas (89)).

⋆ Remark 2.74. The unit vector fields in the direction of the spherical coordinates, pictured in Figure 39, are
1

ρ̂(x, y, z) = p xı̂ + y̂ + z k̂ ,
2 2 2
(x + y + z )
1

φ̂(x, y, z) = p p xzı̂ + yz̂ − (x2 + y 2 )k̂ , (89)
2 2 2
(x + y + z ) x + y2 2

1

θ̂(x, y, z) = p − yı̂ + x̂ .
2
x +y 2

⋆ Remark 2.75 (The gradient in spherical coordinates). The gradient of a scalar field f expressed in spherical
coordinates reads:

~ (ρ, φ, θ) = ∂f ρ̂ + 1 ∂f φ̂ + 1 ∂f θ̂.
∇f
∂ρ ρ ∂φ ρ sin θ ∂θ
A. Moiola, University of Reading 66 Vector calculus lecture notes, 2016–17

~ in
Since their expression is quite complicated, for the formulas of divergence and the curl of a vector field F
spherical coordinates we refer to Section 16.7 of [1] (compare with those in cylindrical coordinates shown in
Remark 2.70).
Example 2.76 (Domain volume in spherical coordinates). Consider the star-shaped domain

D = ~r ∈ R3 , s.t. ρ < r(φ, θ) ,

where r : [0, π] × (−π, π] → R is a positive two-dimensional field. Its volume is

ZZZ Z π Z π Z r(φ,θ)   Z Z π 
1 π
Vol(D) = dV = sin φ ρ2 dρ dφ dθ = r3 (φ, θ) sin φ dφ dθ.
D −π 0 0 3 −π 0

For r(φ, θ) = 1, we recover the volume of the unit sphere 34 π (verify the computation as exercise).
For example, we compute the volume of the pumpkin-shaped domain B of Figure 40 (right)

B = ~r ∈ R3 , s.t. ρ < (sin φ)(5 + cos 7θ) .

From the formula above, we have

Z Z π 
1 π
Vol(B) = sin3 φ (5 + cos 7θ)3 sin φ dφ dθ
3 −π 0
Z π  Z π 
1
= sin4 φ dφ (5 + cos 7θ)3 dθ
3 0 −π
Z π   Z π 
1 1 − cos 2φ 2
= dφ (125 + 25 cos 7θ + 5 cos2 7θ + cos3 7θ) dθ
3 0 2 −π
Z π  
1 1 cos 2φ cos2 2φ 
= − + dφ
3 4 2 4
 0Z π   
1 + cos 14θ 2
· 125 + 25 cos 7θ + 5 + cos 7θ(1 − sin 7θ) dθ
−π 2
Z π   π
1 1 cos 2φ cos 4φ + 1  255 26 5 1 
= − + dφ θ+ sin 7θ + sin 14θ − sin 7θ 
3
3 0 4 2 8 2 7 28 21 −π
  π  
1 1 sin 2φ sin 4φ 1  
1 3
255 2
= φ− + + φ  255π = π 255π = π ≈ 315,
3 4 4 32 8 3 8 0 8

where we used twice the double-angle formula cos 2t = 1 − 2 sin2 t = 2 cos2 t − 1 to expand the square of a
sine or a cosine, while the cubic power is integrated using cos3 t = cos t − cos t sin2 t = (sin t − 13 sin3 t)′ .

0.5
2

0 0

−1

−2
−0.5

5
−1
5
0.5 0
0.5 0
0
0
−0.5
−0.5 −5
−5

Figure 40: Left: the curve {θ = 10φ, ρ = 1} lying on the unit sphere.
Right: the pumpkin domain whose volume is computed in Example 2.76.

Exercise 2.77. ◮ Compute the integral of the field f = ρ4 (sin2 φ + cos 12θ) on the unit ball B (i.e. the ball
centred at the origin with radius 1).
A. Moiola, University of Reading 67 Vector calculus lecture notes, 2016–17

Example 2.78 (Surface integrals in spherical coordinates). Not surprisingly, spherical coordinates can be useful
to compute surface integrals. The field
~
X(φ, θ) = a sin φ cos θı̂ + a sin φ sin θ̂ + a cos φk̂ defined on the region R = (0, π) × (−π, π)

is a chart (in the sense of Definition 2.38) for the sphere of radius a > 0 centred at the origin. To be more
~ is the unit sphere without the meridian in the half plane {y = 0, x ≤ 0} (recall that a
precise, the image of X
chart must be defined in a open region R). Here φ and θ play the roles of u and w. Note that this chart is
~ we obtain
different from the stereographic projection of Example 2.41. Deriving the expression of X,

∂X~ ∂X~
= a cos φ cos θı̂ + a cos φ sin θ̂ − a sin φk̂, = −a sin φ sin θı̂ + a sin φ cos θ̂,
∂φ ∂θ
∂X~ ∂X~
× = a2 (sin2 φ cos θı̂ − sin2 φ sin θ̂ + sin φ cos φk̂),
∂φ ∂θ
 2
 ∂X ~ 
 ~ ∂X
 ×  = a4 (sin4 φ cos2 θ + sin4 φ sin2 θ + sin2 φ cos2 φ) = a4 sin2 φ.
 ∂φ ∂θ 

From this, together with the surface integral formula (67), we have that the curvilinear area element is

dS = a2 sin φ dφ dθ

(recall that 0 ≤ φ ≤ π, so sin φ ≥ 0). For example, we compute the area of the sphere S of radius a (the
absence of a meridian does not affect the integral, as the meridian has zero area):
ZZ Z π Z π 
Area(S) = dS = a sin φ dφ dθ = 4πa2 .
2
S −π 0

~ = z k̂ = ρ cos φk̂ through S, equipped with the outward pointing

We compute the flux (72) of the vector field F
unit normal vector n̂(~r) = ρ̂ = ~ρr = ~ar :
ZZ ZZ  
(89) z (87)
~ · d~S =
F z k̂ · ρ̂ dS using k̂ · ρ̂ = = cos φ
S ρ
Z πS  Z π 
= (a cos φ)(cos φ)a2 sin φ dφ dθ
−π 0
Z π  1 π 4
= 2πa 3
cos φ sin φ dφ = 2πa − cos φ 
2 3 3
= π 2 a3 .
0 3 φ=0 3
RR
~ and the sphere S, can you deduce the sign of
(Draw the field F ~ · d~S from the plot?)
F
S

Exercise 2.79. ◮ The borders of the state of Colorado are defined by the parallels with latitude 37◦ north and
41◦ north and by the meridians with longitude 102◦03′ west and 109◦ 03′ west. Assuming that the Earth is a
sphere with radius 6371 km, compute the area of the state.

Coordinate system Coordinates Measure Domain

Cartesian x, y dA = dx dy x, y ∈ R
R2 Polar x = r cos θ dA = dx dy = r dr dθ r≥0
y = r sin θ θ ∈ (−π, π]
Cartesian x, y, z dV = dx dy dz x, y, z ∈ R
x = r cos θ r≥0
Cylindrical y = r sin θ dV = dx dy dz = r dr dθ dz θ ∈ (−π, π]
R3 z=z z∈R
x = ρ sin φ cos θ ρ≥0
Spherical y = ρ sin φ sin θ dV = dx dy dz = ρ2 sin φ dr dφ dθ φ ∈ [0, π]
z = ρ cos φ θ ∈ (−π, π]

Table 2: Summary of the coordinate systems described in Section 2.3

A. Moiola, University of Reading 68 Vector calculus lecture notes, 2016–17

3 Green’s, divergence and Stokes’ theorems

The first two sections of these notes were mainly devoted to the study of “derivatives” (better: differential
operators) and integrals of scalar and vector fields. This third section will focus on some important
relations between differential operators and integrals.
We recall that the fundamental theorem of calculus gives the basic relation between differentiation
and integration of real functions. It states that, for any continuous function f : R → R with continuous
derivative, and for any two real numbers a < b, the formula
Z b
df
(t) dt = f (b) − f (a) (90)
a dt

holds. This is often understood as “integration reverses differentiation”. The fundamental theorem of
vector calculus (49) of Section 2.1.3 extends this result to the integration of gradients along curves:
Z ~
q
~ r) · d~r = ϕ(~q) − ϕ(~p).
∇ϕ(~ (91)
~
p

Here the integral at the left-hand side is the line integral along any path going from p ~ to ~q. How can
we extend this to multiple integrals and partial derivatives? We will see several different extensions. The
integral at the left-hand side of equation (90) will become a double or a triple integral and the derivative
will become a vector differential operator involving partial derivatives. How is the evaluations of f at
the domain endpoints (i.e. f (b) − f (a)) generalised to higher dimensions? The set of the two values
{a, b} ⊂ R can be thought as the boundary of the interval (a, b), and similarly the points {~p, ~q} ⊂ R3 are
the boundary of a path Γ. In the same way, when integrating a differential operator applied to a field,
over a two- or three-dimensional domain D, we will obtain a certain integral of the same field over the
boundary ∂D of D.31
In the fundamental theorem of calculus (either the scalar one (90) or the vector version (91)), the
values at the endpoints are summed according to a precise choice of the signs: the value at the “initial
point” (f (a) or ϕ(~q)) is subtracted from the value at the “final point” (f (b) or ϕ(~p)). This suggests
that the boundary integrals will involve oriented paths (for the boundaries of two-dimensional domains)
and oriented surfaces (for the boundaries of three-dimensional domains); see Section 2.2.5. The unit
tangent field τ̂ and the unit normal field n̂ will play an important role in assigning a “sign” to the
integrals of vector fields on the boundaries of two- and three-dimensional domains, respectively.
The most important results of this sections are collected in three main theorems. Green’s theorem 3.4
allows to compute the double integral of a component of the curl of a vector field as a path integral. The
divergence theorem 3.14 (see also Theorem 3.10) states that the volume integral of the divergence of
a vector field equals the flux of the same fields through the domain boundary. This key theorem holds
in any dimensions, is probably the most used in applications, and is the most direct generalisation of the
fundamental theorem of calculus to multiple integrals. Finally, Stokes’ theorem 3.28 generalises Green’s
theorem to oriented surfaces, equalling the flux of the curl of a field to the boundary circulation of the
same field32 . All their proof are quite similar to each other and rely on the use of the fundamental theorem
of calculus. We will also prove several other important identities. Table 3 at the end of the section collects
the main formulas obtained.
To fix the notation, we will use the letter R to denote two-dimensional domains (or regions) and the
letter D to denote three-dimensional domains. We will always assume that they are piecewise smooth,
namely their boundaries are unions of smooth parts (∂R is union of smooth paths and ∂D is union of
smooth surfaces). Green’s, divergence and Stokes theorems concern planar regions R, three-dimensional
domains D and surfaces S, respectively.
Green’s, divergence and Stokes’ theorems are treated, together with several examples and exercises, in
Sections 16.3, 16.4 and 16.5 of the textbook [1].
⋆ Remark 3.1 (Smoothness and integrability). As in the previous sections, we will never attempt to be precise
with the assumptions on the regularity of the fields. Even assuming “smooth fields”, meaning C ∞ , may not
be enough. For instance, the function f (t) = 1t is perfectly smooth in the open interval (0, 1), but its integral
R1
0
f (t) dt is not bounded (i.e. is infinite). Roughly speaking, possible assumptions for the theorems we will
prove are that all the derivatives involved in the formulas are continuous in the closure of the considered domains
(even if we usually consider the domains to be open sets). However, we have already seen in Example 2.33
31 You may think at the difference f (b) − f (a) as the signed integral of f over the zero-dimensional set {a, b}, or as the

integral of f over the “oriented set” {a, b}.

32 These theorems receive their names from George Green (1793–1841) and Sir George Gabriel Stokes (1819–1903). The

divergence theorem is sometimes called Gauss’ theorem from Johann Carl Friedrich Gauss (1777–1855). Stokes’ theorem is
also known as Kelvin–Stokes’ theorem (from William Thomson, 1st Baron Kelvin, 1824–1907) or curl theorem.
A. Moiola, University of Reading 69 Vector calculus lecture notes, 2016–17

(the “smile domain”, see also Figure 30) that we can easily integrate fields tending to infinity on the domain
boundary. An even more complicated issue is related to the regularity of domains, curves and surfaces. We
will ignore this problem and always implicitly assume that all the geometric object considered are sufficiently
“well-behaved”.

3.1 Green’s theorem

In this section, we fix a two-dimensional, piecewise smooth, connected, bounded region R ⊂ R2 . “Con-
nected” means that for any two points ~p, ~q ∈ R there exists a path Γ lying entirely in R with endpoints
~p and ~q (R “is made of only one piece”). “Bounded” means that sup{|~r|, ~r ∈ R} < ∞, i.e. the region is
contained in some ball of finite radius (R “does not go to infinity”). The boundary ∂R of R is composed
by one or more loops (closed paths); in the first case, the domain has no “holes” and is called simply
connected33 .
We consider the boundary as an oriented path, with the orientation induced by R after setting the
unit normal field on R to be equal to k̂ (see the last paragraph of Section 2.2.5). In other words, if we
draw R on the ground and walk on the path ∂R according to its orientation, we will see the region R at
our left. The external part of ∂R is run anti-clockwise, the boundary of every hole (if any is present) is
run clockwise; see Figure 41.

∂R

τ̂
R
n̂ ∂R

τ̂
n̂

Figure 41: The shaded area represents a two-dimensional region R. R is connected (composed of only
one piece), piecewise smooth (its boundary is the union of four smooth paths), but not simply connected
since it contains a hole (its boundary is composed by two loops). Its boundary ∂R is composed by two
oriented paths which inherit the orientation from R: the external path is run anti-clockwise and the inner
one clockwise. In both cases, if we proceed along the path according to its orientation we see the region R
on our left. τ̂ is the unit tangent vector to ∂R and n̂ is the outward-pointing unit normal vector.

⋆ Remark 3.2. The line integrals we are using here are slightly more general than those seen in the previous
sections. If R is not simply connected, its boundary isR composed
R of two orR more loopsR Γ1 , . . . , Γn . In this case
the integrals on ∂R are meant as sums of integrals: ∂R = Γ1 ∪...∪Γn = Γ1 + · · · + Γn .
Since all boundaries are loops (i.e. paths starting and ending at the same points), the initial point of
integration is irrelevant.
We prove
R an important
R lemma, which contains the essence of Green’s theorem. We recall that the
notation Γ f dx and Γ f dy for an oriented path Γ and a scalar field f was defined in equation (46).
Lemma 3.3. Consider a smooth scalar field f defined on a two-dimensional region R ⊂ R2 . Then
ZZ I
∂f
dA = − f dx, (92)
R ∂y ∂R
ZZ I
∂f
dA = f dy. (93)
R ∂x ∂R

Note that in (92) the double integral of the derivative in y is associated to the line integral in dx, as
defined in (46); the roles of x and y are swapped in (93). The asymmetry in the sign is due to our choice
of the anti-clockwise orientation of ∂R.
33 Note that the definition of simply-connected domains in three dimensions is quite different from the two-dimensional

definition we use here.

A. Moiola, University of Reading 70 Vector calculus lecture notes, 2016–17

Proof of Lemma 3.3. We prove only the first identity (92); the second (93) follows in a similar way. The
main ingredient of the proof is the use of the fundamental theorem of vector calculus in the y-direction to
reduce the double integral to a line integral. We split the proof in three main steps.
Part 1. We first consider a y-simple domain R, as in formula (52):
n o
R = xı̂ + y̂ ∈ R2 , s.t. xL < x < xR , a(x) < y < b(x) ,

where xL < xR , a, b : (xL , xR ) → R, a(x) < b(x) for all x ∈ (xL , xR ). We compute the double integral at
the left-hand side of (92) using the formula for the iterated integral (53) and the fundamental theorem of
calculus (90) applied to the partial derivative of f in the y-direction:
ZZ Z Z Z b(x) Z 
∂f (53)
xR b(x)
∂f (90)
xR  xR


dA = (x, y) dy dx = f (x, y) dx = f x, b(x) − f x, a(x) dx.
R ∂y xL a(x) ∂y xL y=a(x) xL
(94)

We now consider the boundary integral at the right-hand side of the assertion (92). We split it in four
components, corresponding to the four oriented paths34 in which the boundary ∂R is divided, as in the
left plot of Figure 42:
I Z Z Z Z
f dx = f dx + f dx + f dx + f dx.
∂R ΓS ΓE ΓN ΓW

We first note that the two lateral paths ΓRE = {xR ı̂+ y̂, a(xR ) ≤ y ≤ b(xR )} and ΓW = {xL ı̂+ y̂, a(xL ) ≤
y ≤ b(xL )} are vertical. The integrals · · · dx take into account only the horizontal component of the
curve total derivative (see the definition in (46)) which is zero on these two segments, thus the two vertical
paths do not give any contribution to the integral at the right-hand side of (92).35 The x components of the
parametrisations of the two remaining curvilinear sides can be chosen as affine functions (ı̂ ·~cS (t) = xL + t
and ı̂ · ~cN (t) = xR − t), while the y components depend on x through R the boundary functions a and b
(y = b(x) on ΓN and y = a(x) on ΓS ), thus we can write the integral · · · dx as in formula (48):
I Z Z Z Z
f dx = f dx + f dx + f dx + f dx
∂R ΓS ΓE ΓN ΓW
| {z } | {z }
=0 =0
Z xR 
(48)


= f x, a(x) − f x, b(x) dx
xL
ZZ
(94) ∂f
= − dA,
R ∂y

which is the assertion (92). We have concluded the proof for y-simple domains.
Part 2. We now consider a region R that is the union of two non-overlapping subregions R1 and R2
(i.e. R1 ∪ R2 = R and R1 ∩ R2 = ∅, being quite imprecise with open and closed sets) such that on both
R1 and R2 equation (92) holds true. Then we prove that the same equation holds also on the whole of R.
We denote by Γ = ∂R1 ∩ ∂R2 the intersection of the boundaries of the two subregions. This is a path
that cuts R in two; if the domain R is not simply-connected, i.e. it contains some holes, then the interface
Γ might be composed of two or more disconnected paths. We fix on Γ the same orientation of ∂R1 and
we note that this is opposite to the orientation of ∂R2 ; see the right plot of Figure 42. Combining the
34 The path ΓW collapses to a point if a(xL ) = b(xL ) and similarly the path ΓE if a(xR ) = b(xR ); see e.g. Figure 25.
35 To verify this in formulas, we can write explicitly the parametrisations of the four sides:
d~cS
ΓS : ~cS (t) = (xL + t)ı̂ + a(xL + t)̂, (t) = ı̂ + a′ (xL + t)̂, 0 < t < xR − xL ,
dt

d~cE
ΓE : ~cE (t) = xR ı̂ + a(xR ) + t ̂, (t) = ̂, 0 < t < b(xR ) − a(xR ),
dt
d~cN
ΓN : ~cN (t) = (xR − t)ı̂ + b(xR − t)̂, (t) = −ı̂ − b′ (xR − t)̂, 0 < t < xR − xL ,
dt

d~cW
ΓW : ~cW (t) = xL ı̂ + b(xL ) − t ̂, (t) = −̂, 0 < t < b(xL ) − a(xL ).
dt
We see that the two vertical paths ΓE and ΓW do not give any contribution to the integral at the right-hand side of (92):
Z Z Z b(xR )−a(xR ) Z b(xR )−a(xR )
(46) (44)
d~cE

f dx = fı̂ · d~r = f ~cE (t) ı̂ · (t) dt = f ~cE (t) ı̂ · ̂ dt = 0;
ΓE ΓE 0 dt 0 |{z}
=0

and similarly on ΓW .
A. Moiola, University of Reading 71 Vector calculus lecture notes, 2016–17

ΓN

R1
R ΓE
Γ
ΓW −Γ

ΓS R2
x

xL xR

Figure 42: Proof of Lemma 3.3: the boundary of a y-simple domain is decomposed in four paths (left plot,
part 1 of the proof ), and the region R is split in two subregions R1 and R2 (right plot, part 2 of the proof ).
The interface Γ = ∂R1 ∩ ∂R2 has the same orientation of ∂R1 and opposite to ∂R2 .

integrals over different parts of the boundaries involved and using the relations between the sign of line
integrals and path orientations we conclude:
ZZ ZZ ZZ
∂f ∂f ∂f
dA = dA + dA from R = R1 ∪ R2 and (40)
R ∂y R ∂y R2 ∂y
I1 I
=− f dx − f dx since (92) holds in R1 , R2
∂R ∂R2
Z 1 Z  Z Z 
=− f dx + f dx − f dx + f dx
∂R1 \Γ Γ ∂R2 \Γ −Γ
Z Z  Z Z 
=− f dx + f dx − f dx + f dx
∂R1 \Γ ∂R2 \Γ
Z |Γ {z −Γ }
=0
=− f dx
(∂R \Γ)∪(∂R2 \Γ)
Z 1
=− f dx since ∂R = (∂R1 \ Γ) ∪ (∂R2 \ Γ).
∂R

(To understand this part of the proof observe carefully in the right plot of Figure 42 how the different
geometric objects are related to each other.)
Part 3. Every bounded, piecewise smooth region can be split in a finite number of y-simple subregions
R1 , . . . , RN (this needs to be proved, we take it for granted here). We proved in part 1 that in each of
these regions equation (92) holds true. From part 2 we see that this equation holds true in Q2 := R1 ∪ R2 .
Applying repeatedly part 2 of the proof we see that the result is true for Qj := Qj−1 ∪ Rj = R1 ∪ · · · ∪ Rj ,
for all 3 ≤ j ≤ n and this concludes the proof because QN = R.
Green’s theorem immediately follows from Lemma 3.3.
~
Theorem 3.4 (Green’s theorem). Consider a smooth two-dimensional vector field F(x, y) = F1 (x, y)ı̂ +
2
F2 (x, y)̂ defined on a region R ⊂ R . Then
ZZ   I
∂F2 ∂F1

− dA = F1 dx + F2 dy , (95)
R ∂x ∂y ∂R

which can also be written as

ZZ I


~ ×F
∇ ~ dA = ~ · d~r.
F (96)
3
R ∂R

Proof. Equation (95) follows from choosing f = F1 in (92), f = F2 in (93) and summing the two results.
Equation (96) is obtained from the definition of the curl (23) and the expansion of the line integral (47).
A. Moiola, University of Reading 72 Vector calculus lecture notes, 2016–17

Remark 3.5 (Green’s theorem for three-dimensional fields in two-dimensional domains). We have stated Green’s
~
theorem for a two-dimensional vector field F(x, ~ is any three-dimensional, smooth vector
y) = F1 ı̂ + F2 ̂. If F
~
field, i.e. F(x, y, z) = F1 ı̂ + F2 ̂ + F3 k̂ may depend also on the z coordinate and have three non-zero
components, formula (96) still holds true. In this case, we think at R ⊂ R2 as lying in the xy-plane of R3 (the
plane {z = 0}). Indeed, the left-hand side of the equation is not affected by this modification as (from the
definition of the curl (23)) the third component of the curl (∇ ~ × F) ~ 3 = ∂F2 − ∂F1 does not involve neither
∂x ∂y
F3 nor the partial derivative in z. Similarly, the right-hand side of (96) is not modified because the circulation
of F~ along ∂R is the integral of the scalar product of F ~ and the total derivative of the curve defining ∂R (see
(44)), whose z-component vanish, thus the component F3 does not contribute to the line integral.
Example 3.6. Use Green’s theorem to compute the circulation of F ~ = (2xy + y 2 )ı̂ + x2 ̂ along the boundary
of the bounded domain R delimited by the line y = x and the parabola y = x2 .
The domain R can be written as R = {~r ∈ R2 , 0 < x < 1, x2 < y < x}; see the left plot in Figure 43.
~ ×F
The curl of the given field is ∇ ~ = (2x − 2x − 2y)k̂ = −2y k̂. Then, Green’s theorem gives
I ZZ Z 1 Z x  Z 1
~ (96)
~ ~ (53) 1 1 2
F · d~r = (∇ × F)3 dA = (−2y) dy dx = (−x2 + x4 ) dy = − = − .
∂R R 0 x2 0 5 3 15
Of course, it is possible to directly compute the circulation, but the calculation is a bit longer, as we have
to split the boundary in the straight segment ΓS (e.g. parametrised by ~a(t) = (1 − t)(ı̂ + ̂), 0 < t < 1)
and the parabolic arc ΓP (e.g. parametrised by ~b(t) = tı̂ + t2 ̂, 0 < t < 1). (Recall the parametrisations of
Remark 1.24 and that the boundary must be oriented anti-clockwise.) Then we have
I Z Z
~
F · d~r = ~
F · d~r + ~ · d~r
F
∂R ΓS ΓP
Z1 Z 1



= 3(1 − t)2 ı̂ + (1 − t)2 ̂ · (−ı̂ − ̂) dt + (2t3 + t4 )ı̂ + t2 ̂ · (ı̂ + 2t̂) dt
0 0
Z 1 Z 1
4 1 2
= −4 (1 − t)2 dt + (4t3 + t4 ) dt = − + 1 + = − .
0 0 3 5 15

y
y
̂ ̂

R R

x x
ı̂ −2ı̂ ı̂ 2ı̂

Figure 43: Left plot: the domain R = {~r ∈ R2 , 0 < x < 1, x2 < y < x} in Example 3.6.
Right plot: the half ellipse R = {xı̂ + y̂ ∈ R2 , 14 x2 + y 2 < 1, y > 0} in Exercise 3.8.

~ = (x + 2y)(ı̂ + 3̂) and the disc

Exercise 3.7. ◮ Demonstrate Green’s theorem for the vector field F
{xı̂ + y̂, such that x2 + y 2 < 4}.
~ = y 2 ı̂ on the half-ellipse R = {xı̂ + y̂ ∈ R2 , 1 x2 + y 2 <
Exercise 3.8. ◮ Demonstrate Green’s theorem for F 4
1, y > 0} (right plot in Figure 43).
Example 3.9 (Measuring areas with line integrals). A (perhaps) surprising application of Green’s theorem is
the possibility of measuring the area of a planar region R by computing only line integrals on its boundary36 .
To do this, we need a vector field F ~ such that the integrand on R in (95) is everywhere equal to one, namely
∂F2 ∂F1 ~ ~ ~ 1
∂x − ∂y = 1. Several choices are possible, for instance F = −yı̂, F = x̂, or F = 2 (−yı̂ + x̂). Each of
these gives rise to a line integral that computes the area of R:
ZZ I I I
1
Area(R) = dA = − y dx = x dy = (−y dx + dy).
R ∂R ∂R 2 ∂R
36 The possibility of computing areas with line integrals should not be a surprise: we already know that the area of a

y-simple domain (52), for example, can be computed as

ZZ Z xR  Z b(x)  Z xR


Area(R) = dA = dy dx = b(x) − a(x) dx
R xL a(x) xL
H
which is nothing else than − ∂R y dx. This is also the formula for the area of the region comprised between two graphs you
already know from first year.
A. Moiola, University of Reading 73 Vector calculus lecture notes, 2016–17
 a
a  d~
In Section 2.1.2, we defined the unit tangent vector field τ̂ = d~ 
dt / dt of an oriented path Γ defined by
the curve ~a. If Γ = ∂R is the boundary of R (with the orientation convention described above), then the
outward pointing unit normal vector n̂ on ∂R is related to the unit tangent τ̂ by the formulas

n̂ = τ̂ × k̂ = t2 ı̂ − t1 ̂, τ̂ = k̂ × n̂ = −n2 ı̂ + n1 ̂, (n1 = t2 , n2 = −t1 ); (97)

see Figure 41. (Note that, if R contains a hole, then on the the corresponding part of the boundary the
vector field n̂ points out of R and into the hole.) For a vector field F~ = F1 ı̂ + F2 ̂, choosing f = F1 in
(93), f = F2 in (92), and summing the resulting equations we obtain
ZZ   I
∂F1 ∂F2 (92),(93)

+ dA = F1 dy − F2 dx
R ∂x ∂y
I∂R
(46)

= F1 ̂ − F2 ı̂ · d~r
I∂R
(2)

= k̂ × F~ · d~r
I∂R
(45)

= ~ · τ̂ ds
(k̂ × F)
I∂R
Ex. 1.15

= ~ ds
(τ̂ × k̂) · F
I∂R
(97)

= ~ · n̂ ds.
F
∂R

This is the proof of the divergence theorem in two dimensions.

Theorem 3.10 (Divergence theorem in two-dimensions). Let F ~ = F1 ı̂ + F2 ̂ be a smooth vector field defined
in a region R ⊂ R2 with outward pointing unit normal vector field n̂. Then
ZZ   I
∂F1 ∂F2 ~ · n̂ ds.
+ dA = F (98)
R ∂x ∂y ∂R

The integrand in the double integral at the left-hand side of (98) is the two-dimensional divergence
~ ·F
∇ ~ of F.
~
~ = (2xy + y 2 )ı̂ + x2 ̂ and the bounded domain
Example 3.11. As in Example 3.6, consider the vector field F
H
R delimited by the line y = x and the parabola y = x . Compute the “flux”37 ∂R F
2 ~ · ~n ds of F
~ through ∂R.
We simply apply the divergence theorem (98):
I ZZ   ZZ Z 1 Z x 
~ ·~ ∂F1 ∂F2 2
F n ds = + dA = 2y dA = 2y dy dx = ,
∂R R ∂x ∂y R 0 x2 15

with the same computation as in the previous example. The direct computation of the contour integral is slightly
more complicated in this case than in Example 3.6, in that it requires the calculation of the outward-pointing
unit normal vector field.
Exercise 3.12. ◮ Prove the second formula in (97) without using the expansion of the vectors in components.
You can find a helpful formula in the exercises in Section 1.1.

3.2 The divergence theorem

In this section we consider a three-dimensional domain D ⊂ R3 and we assume it to be piecewise smooth,
bounded and connected. As in the previous section, we start with a preliminary lemma.
Lemma 3.13. Let f be a smooth scalar field on D. Then we have
ZZZ ZZ
∂f
dV = f k̂ · d~S. (99)
D ∂z ∂D

Here the surface ∂D is equipped with outward-pointing unit normal vector field n̂.
37 Note that here the word “flux” is used to denote the contour integral of the normal component of a two-dimensional

field, as opposed to the surface integral of the normal component in three dimensions.
A. Moiola, University of Reading 74 Vector calculus lecture notes, 2016–17

Proof. The proof is very similar to that of Lemma 3.3 (in a sense, it is its extension to three dimensions).
As in that case, we divide the proof in three parts: the first is devoted to simple domains, the second
extends the result to the unions of two such simple domains, the third completes the extension to general
domains by partitioning these in simpler parts. Once more, the basic instrument in the proof is the
fundamental theorem of calculus.
Part 1. We consider a z-simple domain as that introduced at the beginning of Section 2.2.3:
n o
D = xı̂ + y̂ + z k̂ s. t. xL < x < xR , a(x) < y < b(x), α(x, y) < z < β(x, y) ,

where xL < xR are two real numbers, a, b : [xL , xR ] → R two real functions (with a < b), and α, β : R → R
two scalar fields in two variables (with α < β). We call R the planar region R = {xı̂ + y̂ s. t. xL < x <
xR , a(x) < y < b(x)}, the projection of D on the xy-plane. See a representation of D and R in Figure 44.
The boundary ∂D is composed of six (curvilinear) faces38 . Four of these faces are vertical, as they lie
straight above the four (curvilinear) sides of R. Thus on these faces the outward-pointing unit normal
RR
vector field n̂ has zero vertical component k̂ · n̂, and the corresponding terms f k̂ · d~S in the surface
integral at the right-hand side of (99) vanish.39
On the other hand, the top face ST is the graph of the two-dimensional field β : R → R. The
contribution to the surface integral at the right-hand side of (99) given by ST is the flux of the vector field
f k̂ (whose first two components are constantly zero) through ST itself, and can be reduced to a double
integral on R using formula (74):
ZZ ZZ


f (x, y, z) k̂ · d~S = f x, y, β(x, y) dA.
ST R

Similarly, the bottom face SB is the graph of the two-dimensional field α : R → R. However, on this face,
the outward-pointing unit normal vector field n̂ points downward, opposite to the convention stipulated
in Section 2.2.5 for graph surfaces. Thus the sign in (74) needs to be reversed and the contribution of SB
to the right-hand side of (99) reads
ZZ ZZ


~
f (x, y, z) k̂ · dS = − f x, y, α(x, y) dA.
ST R

To conclude the proof of the first part we expand the right-hand side of the identity in the assertion in a
sum over the faces of D and use the fundamental theorem of calculus to transform it into a triple integral:
ZZ ZZ ZZ
f k̂ · d~S = f k̂ · d~S + f k̂ · d~S
∂D ST SB
ZZ ZZ ZZ ZZ
+ f k̂ · d~
S+ f k̂ · d~S + f k̂ · d~S + f k̂ · d~S
SW SS SE SN
| {z }
=0 (k̂·n̂=0)
ZZ ZZ



= f x, y, β(x, y) dA − f x, y, α(x, y) dA
Z ZR  R



= f x, y, β(x, y) − f x, y, α(x, y) dA
R
ZZ  Z β(x,y) 
(90) ∂f
= (x, y, z) dz dA (fundamental theorem of calculus)
R α(x,y) ∂z
ZZZ
(62) ∂f
= (x, y, z) dV (triple integral as an iterated integral).
D ∂z

38 Actually, if a(x ) = b(x ) or a(x ) = b(x ), then the left and right faces are collapsed to a segment and the total
L L R R
number of faces p is reduced to 5 orp4. The number p of non-empty faces
p can be even lower, consider the following example:
{−1 < x < 1, − x2 + y 2 < y < x2 + y 2 , 1 − x2 − y 2 < z < 2 1 − x2 − y 2 }. How many non-empty faces does this
z-simple domain have?
39 More explicitly, the faces and the corresponding unit normals can be written as:

 

 n̂W = −ı̂ on x = xL , a(xL ) < y < b(xL ), α(xL , y) < z < β(xL , y) ,

 ′
(x)ı̂−̂ 



n̂S = √ a
on xL < x < xR , y = a(x), α x, a(x) < z < β x, a(x) ,
1+(a′ (x))2
n̂(~r) = 

 n̂E = ı̂ on x = xR , a(xR ) < y < b(xR ), α(xR , y) < z < β(xR , y) ,

 


 ′
n̂N = √−b (x)ı̂+̂ on xL < x < xR , y = b(x), α x, b(x) < z < β x, b(x) .
1+(b (x))
′ 2

You can verify these expressions for n̂ by finding two tangent vectors and checking that they are orthogonal to n̂.
A. Moiola, University of Reading 75 Vector calculus lecture notes, 2016–17

n̂T
z

ST = β(R)
SN
D n̂N
SE

n̂E
y
xL

a(x) R b(x)

xR W
S N
x E

Figure 44: A representation of the the z-simple domain in the first part of the proof of Lemma 3.13. The
bottom face SB is that underneath, the west face SW is behind D and the south face SS is on the left of
the figure. The outward-pointing unit vector field n̂ on SW , SS , SE and SN , has zero vertical component
n̂ · k̂. The fundamental theorem of calculus is applied along all vertical segments connecting the bottom
face with the top face.

Part 2. We now show the following: if a domain D is disjoint union of two subdomains D1 and D2 ,
and if identity (99) holds in both of them, then the same identity holds in D.
We denote by Σ = ∂D1 ∩ ∂D2 the interface between the subdomains and with n̂1 , n̂2 the outward-
pointing unit normal vector fields on ∂D1 and ∂D2 , respectively. Then ∂D1 = Σ ∪ (∂D1 ∩ ∂D) (and
∂D2 = Σ ∪ (∂D2 ∩ ∂D)), n̂1 = −n̂2 on Σ and n̂1 = n̂ on ∂D1 ∩ ∂D (similarly n̂2 = n̂ on ∂D2 ∩ ∂D).
Combining everything we have the assertion:
ZZZ ZZZ ZZZ
∂f ∂f ∂f
dV = dV + dV
D ∂z D ∂z D ∂z
ZZ 1 Z Z2
= f k̂ · n̂1 dS + f k̂ · n̂2 dS
Z Z∂D1 ∂D2
ZZ ZZ ZZ
= f k̂ · n̂1 dS + f k̂ · n̂1 dS + f k̂ · n̂2 dS + f k̂ · n̂2 dS
∂D1 ∩∂D Σ ∂D2 ∩∂D Σ
ZZ ZZ
= f k̂ · n̂ dS + (f k̂ · n̂1 + f k̂ · n̂2 ) dS
(∂D1 ∩∂D)∪(∂D2 ∩∂D) Σ | {z }
=0
ZZ
= f k̂ d~S.
∂D

Part 3. Any “regular” domain can be decomposed in a finite union of disjoint z-simple domains (of
course, this should be proved rigorously, here we only assume it since we have not even described in detail
the regularity of the domains). Thus, to conclude the proof of the lemma, we can proceed by induction,
exactly as we did in part 3 of Lemma 3.3.
Proceeding as in Lemma 3.13, we can prove that
ZZZ ZZ ZZZ ZZ
∂f ∂f
dV = f ı̂ · d~S, dV = f ̂ · d~S. (100)
D ∂x ∂D D ∂y ∂D

From identities (99) and (100) we obtain the following fundamental theorem.
~ be a smooth vector field defined on D. Then
Theorem 3.14 (Divergence theorem in three-dimensions). Let F
ZZZ ZZ
~ ·F
∇ ~ dV = ~ · d~S.
F (101)
D ∂D

~ equals the flux of F

In words: the triple integral over D of the divergence of F ~ through the boundary of D.
A. Moiola, University of Reading 76 Vector calculus lecture notes, 2016–17

Proof. We simply sum (99) and (100) applied to the three components of F: ~
ZZZ ZZZ  
~ ~ ∂F1 ∂F2 ∂F3
∇ · F dV = + + dV
D ∂x ∂y ∂z
Z Z ZD ZZZ ZZZ
∂F1 ∂F2 ∂F3
= dV + dV + dV
D ∂x D ∂y D ∂z
ZZ ZZ ZZ
= F1 ı̂ · d~S + F2 ̂ · d~S + F3 k̂ · d~S
∂D ∂D ∂D
ZZ


= F1 ı̂ + F2 ̂ + F3 k̂ · d~S
Z Z∂D
= ~ · d~S.
F
∂D

The divergence theorem, together with its corollaries, is one of the most fundamental results in analysis
and in calculus. It holds in any dimensions, appears in every sort of theoretical and applied setting (PDEs,
differential geometry, fluid dynamics, electromagnetism, . . . ), has a deep physical meaning, is the starting
point for the design of many numerical methods, and its extension to extremely irregular fields and
domains is very challenging (and indeed it is currently an object of research after two centuries from the
first discovery of the theorem!).

Remark 3.15 (Intuitive physical interpretation of the divergence theorem). If the vector field F ~ represents the
velocity of a fluid, its divergence might be interpreted as a measure of the sources and sinks in the fluid. Thus
the divergence theorem may be interpreted as follows: the total net production of a quantity in a domain equals
the amount of that quantity that exits the domain through its boundary (per unit of time). For this reason,
the divergence theorem is used to formulate many “conservation laws” or “continuity equations” describing the
evolution and the conservation of certain physical quantities (e.g. mass, energy, charge,. . . ).

Exercise 3.16. ◮ Use the divergence theorem to compute the flux of the field F ~ = ex ı̂ + cos2 x̂ + (x + y + z)k̂
through the boundary of the box D = (0, 1) × (−1, 1) × (0, 10).
RR
Exercise 3.17. ◮ (Difficult!) Use the divergence theorem to compute S (x2 + y 2 ) dS, where S = {|~r| = R}
is the sphere of radius R centred at the origin. Hint: you need to define a suitable vector field.
Exercise 3.18. ◮ Consider the cone C = {0 < z < 1, x2 + y 2 < z 2 }. Use the divergence theorem to compute
RRR
C
|~r|2 dV .
Exercise 3.19. ◮ (i) Use the divergence theorem to compute the flux of the vector field

~ = 1 (x3 + y 3 + z 3 )ı̂ + yz(z̂ + y k̂)

F
3
through the boundary of the unit ball B = {~r ∈ R3 , |~r| < 1}. Hint: use spherical coordinates.
~ through the boundary of the cube CL = (0, L)3 = {xı̂+y̂+z k̂, 0 <
(ii) Find L > 0 such that the flux of F
x, y, z < L} is equal to the flux of the same field through the boundary of the unit ball B.
Exercise 3.20. ◮ [⋆ Very difficult!] Let D ⊂ R3 be a bounded domain. Prove that
ZZ (
~r ~ 4π if the origin lies in D,
3
· dS =
∂D |~r| 0 if the origin lies outside D.

In the exercises above we used the divergence theorem for different purposes: to compute a surface
integral over a boundary (by finding a vector field whose normal component on the boundary is given, as
in 3.17) and to compute a volume integral (by finding a vector field whose divergence is given, as in 3.18).
⋆ Remark 3.21 (The divergence theorem in one dimension is the fundamental theorem of calculus).
The divergence theorems 3.10 and 3.14 directly generalise the fundamental theorem of calculus to higher
dimensions. Indeed, a bounded, connected (open) domain in R must be an interval I = (a, b), whose boundary
is composed by two points ∂I = {a, b} with outward-pointing normal vector n(a) = −1 and n(b) = 1 (a
one-dimensional vector is simply a scalar). The “zero-dimensional integral” of a function f on ∂I reduces to
the sum of the two values of f in b and a, and the flux to their difference. A “one-dimensional vector field” is
a real function and its divergence is its derivative. So we have
Z Z Z
f ′ (x) dx = ∇ ~ · f (x) dx = f (x)n(x) dx = f (b)n(b) + f (a)n(a) = f (b) − f (a).
I I ∂I
A. Moiola, University of Reading 77 Vector calculus lecture notes, 2016–17

In the next corollary we show several other identities which follow from equations (99) and (100). On
the boundary ∂D (or more generally on any oriented surface), we call normal derivative the scalar
product of the gradient of a scalar field f and the unit normal field n̂ of f , and we denote it by

∂f ~
:= n̂ · ∇f.
∂n

Corollary 3.22. Let f and g be scalar fields and F ~ be a vector field, all defined on a domain D. Then
ZZZ ZZ
~ dV =
∇f f n̂ dS, (102)
D ∂D
ZZZ ZZ
~ ×F
∇ ~ dV = n̂ × F~ dS, (103)
D ∂D
ZZZ ZZ
~ · ∇f
~ ) dV = ∂g
(f ∆g + ∇g f dS (Green’s 1st identity), (104)
∂n
ZDZ Z Z Z∂D  
∂g ∂f
(f ∆g − g∆f ) dV = f −g dS (Green’s 2nd identity). (105)
D ∂D ∂n ∂n
Note that identities (102) and (103) are vectorial (the values at the two sides of the equal sign are vectors).
Proof. The proof of identity (102) is very easy:
ZZZ ZZZ  
~ dV ∂f ∂f ∂f
∇f = ı̂ + ̂ + k̂ dV
D ∂x ∂y ∂z
Z Z D  ZZ
(99),(100)
= f (ı̂ · n̂)ı̂ + f (̂ · n̂)̂ + f (k̂ · n̂)k̂ dS = f n̂ dS.
∂D ∂D

We prove identity (103) for the first component:

ZZZ ZZZ  
~ ~ (23) ∂F3 ∂F2
(∇ × F)1 dV = − dV
D ∂y ∂z
ZZ D
(99),(100)

= F3 ̂ · n̂ − F2 k̂ · n̂ dS
Z Z∂D ZZ

(2)
= F3 n2 − F2 n3 dS = ~ 1 dS;
(n̂ × F)
∂D ∂D

and the same holds for the second and third components.
Using the vector identities of Section 1.4, we have

∇ ~ (29)
~ · (f ∇g) ~ · ∇g
= ∇f ~ + f∇ ~ (24)
~ · (∇g) ~ · ∇g
= ∇f ~ + f ∆g.

Then, Green’s first identity (104) follows immediately from the application of the divergence theorem 3.14
~ = f ∇g.
to the field F ~ Green’s second identity (105) follows from subtracting from (104) the same identity
with f and g interchanged.
Exercise 3.23. ◮ Demonstrate identity (102) for the field f = xyz and the unit cube D = (0, 1)3 .
Exercise 3.24. ◮ Show that if a smooth scalar field f is harmonic in a domain D and its normal derivative
vanishes everywhere on the boundary ∂D, then f is constant.
⋆ Remark 3.25. The two Green’s identities (104) and (105) are particularly important in the theory of partial
differential equations (PDEs). In particular, several second order PDEs involving the Laplacian operator can
be rewritten using (104) in a “variational form” or “weak form”, which involves a volume integral and a
boundary one. This is the starting point for the definition of the most common algorithms for the numerical
approximation of the solutions on a computer, in particular for the finite element method (FEM, you might see
it in a future class). Another common numerical scheme, the boundary element method (BEM), arises from
the second Green identity (105).
Comparison with scalar calculus 3.26 (Integration by parts in more dimensions). In the first calculus class
you learned how to “integrate by parts”:
Z b Z b
′
f (t)g(t) dt + f (t)g ′ (t) dt = f (b)g(b) − f (a)g(a),
a a

for real functions f, g : [a, b] → R. This is a straightforward consequence of the fundamental theorem of
calculus (90) and the product rule (f g)′ = f ′ g + f g ′ . How does this extend to higher dimensions? The
A. Moiola, University of Reading 78 Vector calculus lecture notes, 2016–17

product rule was extended to partial derivatives in (8) and to differential operators in Proposition 1.55; we saw
in Remark 3.21 that the divergence theorem extends the fundamental theorem of calculus. Several formulas
arising from different combinations of these ingredients are usually termed “multidimensional integration by
parts”, for example
ZZZ ZZZ ZZ
~ ·G
∇f ~ dV + f∇~ ·G~ dV = fG~ · d~S,
ZDZ Z Z ZDZ Z Z∂D
~ )g dV +
(∇f ~ dV =
f ∇g f g n̂ dS,
D D ∂D

~ is a vector field on D. For exercise, prove these two identities.

where f , g are scalar fields and G
⋆ Remark 3.27 (Differential and integral form of physical laws). The main reason of the importance and
ubiquity of the divergence theorem in physics and engineering is that it constitutes the relation between the two
possible formulation of many physical laws: the “differential form”, expressed by a partial differential equation,
and the “integral form”, expressed by an integral equation. The integral form usually better describes the main
physical concepts and contains the quantities that can be measured experimentally, while the differential form
allows an easier mathematical manipulation. The two forms may lead to different numerical algorithms for the
approximation of the solutions of the equation.
For example, we consider Gauss’ law of electrostatics, you might have already encountered elsewhere. Its
~ (which is a vector field) through a closed surface ∂D
integral form states that the flux of an electric field E
(in vacuum) is proportional to the total electrical charge in the volume D bounded by that surface:
ZZ ZZZ
~ ~ ρ
E · dS = dV,
∂D ε
D 0

where ρ is the charge density (whose integral gives the total charge) and ε0 is a constant of proportionality (the
“vacuum permittivity”). Its differential form reads ∇~ ·E~ = ρ/ε0 . The divergence theorem immediately allows
to deduce the integral form of Gauss law from the differential one; since the former holds for any domain D,
also the converse implication can be proved. Gauss’ law is one of the four celebrated Maxwell’s equations, the
fundamental laws of electromagnetism: all of them have a differential and an integral form, related to each
other either by the divergence or Stokes’ theorem.

3.3 Stokes’ theorem

Stokes’ theorem extends Green’s theorem to general oriented surfaces. It states that the flux of the curl of
a vector field through an oriented surface equals the circulation along the boundary of the same surface.
Theorem 3.28 (Stokes’ theorem). Let S ⊂ R3 be a piecewise smooth, bounded, oriented surface with unit
~ be a smooth vector field defined on S. Then
normal field n̂, and let F
ZZ I
(∇ ~ · d~S =
~ × F) ~ · d~r,
F (106)
S ∂S

where the boundary ∂S is understood as the oriented path with the orientation inherited from (S, n̂).
Proof. We prove the theorem only for the special class of surfaces describe in Remark 2.44: we assume S
to be the graph of a two-dimensional field g, i.e. S = {~r = xı̂ + y̂ + z k̂ ∈ R3 s.t. xı̂ + y̂ ∈ R, z = g(x, y)}
where R ⊂ R2 is a planar region as in Section 3.1. The proof of the theorem in its generality requires
pasting together two or more simpler surfaces.
The main idea of the proof is to reduce the two integrals in (106) to similar integrals on the flat region
R and its boundary, using the fact that S is the graph of g, so the variables on S are related to each other
by the equation z = g(x, y). Then one applies Green’s theorem on R.
The surface integral (flux) at the left-hand side of the assertion (106) can easily be transformed into a
double integral over the planar region R by using formula (74):
ZZ ZZ  
(∇ ~ · d~S (74)
~ × F) = ~ × F)
− (∇ ~ 1 ∂g − (∇ ~ 2 ∂g + (∇
~ × F) ~ × F) ~ 3 dA
S R ∂x ∂y
ZZ      (107)
(23) ∂F3 ∂F2 ∂g ∂F1 ∂F3  ∂g  ∂F2 ∂F1 
= − − − − + − dA.
R ∂y ∂z ∂x ∂z ∂x ∂y ∂x ∂y

We denote by ~a : [tI , tF ] → ∂S a curve that parametrises ∂S. The surface S is the graph of g, thus the
components of ~a are related one another by the relation a3 (t) = g(a1 (t), a2 (t)). From the chain rule (37)
∂g da1 ∂g da2
we have da
dt = ∂x dt + ∂y dt . Moreover, the planar curve a1 (t)ı̂ + a2 (t)̂ is a parametrisation of ∂R,
3
A. Moiola, University of Reading 79 Vector calculus lecture notes, 2016–17
H H
thus ∂S f (x, y, z) dx = ∂R f (x, y, g(x, y)) dx for any scalar field f (and similarly for the integral in dy).
Putting together all the pieces:
I Z tF  
(44)
da1 (t)
da2 (t)
da3 (t)
~
F · d~r = F1 ~a(t) + F2 ~a(t) + F3 ~a(t) dt
∂S tI dt dt dt
Z tF   ∂g da (t) ∂g da (t)  

da1 (t)
da2 (t)
1 2
= F1 ~a(t) + F2 ~a(t) + F3 ~a(t) + dt
tI dt dt ∂x dt ∂y dt
| {z }
chain rule (37) on a3 (t)=g(a1 (t),a2 (t))
Z  
tF

∂g  da1 (t) 

∂g  da2 (t)
= F1 ~a(t) + F3 ~a(t) + F2 ~a(t) + F3 ~a(t) dt
t ∂x dt ∂y dt
I I     
(46) ∂g ∂g
= F1 (x, y, z) + F3 (x, y, z) (x, y) dx + F2 (x, y, z) + F3 (x, y, z) (x, y) dy
∂S ∂x ∂y
I    


∂g

∂g 
= F1 x, y, g(x, y) + F3 x, y, g(x, y) dx + F2 x, y, g(x, y) + F3 x, y, g(x, y) dy
∂R ∂x ∂y
ZZ 
(92),(93) ∂ 

∂g 
= − F1 x, y, g(x, y) + F3 x, y, g(x, y) (x, y)
R ∂y ∂x
∂ 

∂g 
+ F2 x, y, g(x, y) + F3 x, y, g(x, y) (x, y) dA (Lemma 3.3 / Green’s theorem)
∂x ∂y
ZZ 
∂F1 ∂F1 ∂g ∂F3 ∂g ∂F3 ∂g ∂g ∂ 2g
= − − − − − F3
R ∂y ∂z ∂y ∂y ∂x ∂z ∂y ∂x ∂x∂y

∂F2 ∂F2 ∂g ∂F3 ∂g ∂F3 ∂g ∂g ∂2g
+ + + + + F3 dA (chain and product rules)
∂x ∂z ∂x ∂x ∂y ∂z ∂x ∂y ∂x∂y
Z Z  
∂F2 ∂F3  ∂g  ∂F3 ∂F1  ∂g ∂F2 ∂F1
= − + − + − dA
∂z ∂y ∂x ∂x ∂z ∂y ∂x ∂y
Z ZR
(107)
= ~ × F)
(∇ ~ · d~ S.
S

Therefore, the right-hand side of (106) equals the left-hand side and we have proved the assertion.
⋆ Remark 3.29 (Only for brave students!). We give also a more general proof of Stokes’ theorem for a
parametric surface S (with connected boundary) parametrised by a chart X ~ : R → S as in Definition 2.38.
The main tools are the same as in the proof above, namely repeated use of the chain rule to reduce to
Green’s theorem on the region R, but this proof is a bit more technical. We denote by ~b : [tI , tF ] → ∂R
a parametrisation of ∂R and fix ~a(t) = X(~ ~b(t)), the corresponding parametrisation of ∂S. In the following
chain of equalities we do not write all the obvious dependences of the various integrands, e.g. F ~ stands for
~ a(t)) = F(
F(~ ~ ~b(t))) in the integrals in t and stands for F(
~ X( ~ X(u,
~ w)) in the integrals in u and w. Then
I Z tF Z tF ~ ~
~ · dr = ~ d~a ~ · dX(b) (t) dt
F F· (t) dt = F
∂S tI dt tI dt
Z tF 
(37) ∂X1 ~
db1 ∂X1 ~
db2 ∂X2 ~
db1
= F1 b(t) (t) + F1 b(t) (t) + F2 b(t) (t)
tI ∂u dt ∂w dt ∂u dt
∂X2 ~
db2 ∂X3 ~
db1 ∂X3 ~
db2 
+ F2 b(t) (t) + F3 b(t) (t) + F3 b(t) (t) dt
∂w dt ∂u dt ∂w dt
Z tF  ~ ~ 
= ~ · ∂ X db1 + F
F ~ · ∂ X db2 dt
tI ∂u dt ∂w dt
I  ~ ~ 
(46)
= ~ · ∂ X du + F
F ~ · ∂ X dw (using notation of Rem. 2.8 in uw-plane)
∂R ∂u ∂w
ZZ  
(96) ∂ ~ ∂X  ∂  ~ ∂X 
= − F· + F· dA (Green’s theorem)
R ∂w ∂u ∂u ∂w
ZZ  ~ ∂X ~ 2~ ~ ~ 2~ 
(8) ∂F ~ · ∂ X + ∂F · ∂X + F ~ · ∂ X dA
= − · −F (product rule for partial der.)
R ∂w ∂u ∂w∂u ∂u ∂w ∂u∂w
ZZ  ~ ~ ~ ~ ~
(38)
~ 1 · ∂ X ∂X1 − ∇F ~ 2 · ∂ X ∂X2 − ∇F ~ 3 · ∂ X ∂X3 ∂2X ∂2X
= − ∇F (Clairault theorem =
R ∂w ∂u ∂w ∂u ∂w ∂u ∂w∂u ∂u∂w
~ ~ ~ 
~ 1 · ∂ X ∂X1 + ∇F
+ ∇F ~ 2 · ∂ X ∂X2 + ∇F ~ 3 · ∂ X ∂X3 dA ~ X(u,
and chain rule for F( ~ w)))
∂u ∂w ∂u ∂w ∂u ∂w
A. Moiola, University of Reading 80 Vector calculus lecture notes, 2016–17
ZZ 

~ ~
(16)
= JF ~ T ∂ X · ∂ X dA
~ − (J F) (collecting terms, tricky!)
R ∂u ∂w
ZZ  ~  ~
= (∇ ~ × ∂ X · ∂ X dA
~ × F) (Remark 1.64)
R ∂u ∂w
ZZ  ∂X
~ ·
~ × F)
~ ∂X 
~
= (∇ × dA (Exercise 1.15)
R ∂u ∂w
ZZ
(67)
= ~ × F)
(∇ ~ · d~S (surface integral formula).
S

We note an immediate consequence of Stokes’ theorem. If the surface S is the boundary of a three-
dimensional domain D, then ∂S is empty (see Remark 2.59). Thus, (106) implies that, for every bounded,
connected, piecewise smooth domain D and for every smooth vector field F~ defined on ∂D, the flux of
~ ~
∇ × F on ∂D vanishes: ZZ
~ × F)
(∇ ~ · d~S = 0.
∂D

Note that using the divergence theorem 3.14 and the vector identity ∇~ · (∇
~ × F)
~ = 0 (25) we could have
~ was defined (and smooth) in the whole domain D,
derived the same identity, but only in the case when F
and not only on its boundary.
Exercise 3.30. ◮ Demonstrate Stokes’ theorem for the upper half sphere S = {|~r| = 1, z > 0} and the field
~ = (2x − y)ı̂ − yz 2 ̂ − y 2 z k̂.
F
H
Exercise 3.31. ◮ Compute ∂S |~r|2 dx, with S = {z = x2 − y 2 , 0 < x, y < 1} (cf. Exercise 2.55).
~ along the
Exercise 3.32. ◮ Let f : R3 → R be a scalar field. Prove that the line integral of the product f ∇f
boundary of an orientable surface S is zero.
Remark 3.33 (An alternative definition of divergence and curl). The divergence and the curl of a vector field
~ are often defined as the following limits (whenever the limits exist):
F
ZZ
~ · F(~
~ r0 ) = lim 1 ~ · d~S,
∇ F
R→0 Vol(BR ) ∂BR
I

1
~ r0 ) = lim
~ × F(~
â · ∇ ~ · d~r,
F
R→0 Area(QR (â)) ∂Q (â)
R

where BR = {~r ∈ R3 , |~r −~r0 | < R} is the ball of centre ~r0 and radius R, â is a unit vector, and QR (â) is the
disc of centre ~r0 , radius R and perpendicular to â. (Note that the curl is defined by its scalar product with
all unit vectors.) Then one proves40 that, if F ~ is differentiable, these definitions coincide with those given in
Section 1.3. This also provides rigorous, geometric, coordinate-free interpretations of divergence and curl as
“spreading” and “rotation” of a vector field, respectively:
The divergence of a vector field F~ is the average of the flux of F ~ through an infinitesimal sphere.
The curl of F~ is the average of the circulation of F
~ around an infinitesimal circle.
Note that also the usual derivative of a real function G : R → R can be defined in a completely analogous
1
way as G′ (t) = limR→0 Length([t−R,t+R]) (G(t + R) − G(t − R)). Here, the interpretation is that the derivative
of G is the average of the increment of G in an infinitesimal interval.
⋆ Remark 3.34. With the fundamental theorem of vector calculus we have seen that theRline integral from a
~ · d~r depends
point ~p to a point ~q of a gradient is independent of the integration path. In other words, Γ ∇ϕ
only on the scalar field ϕ and the endpoints of Γ, i.e. its boundary ∂Γ (as opposed to the entire path Γ).
Stokes’ theorem may be interpreted similarly: the flux of the curl of a vector field through a surface S only
depends on the field itself and the boundary of S (as opposed to the entire surface S). If two surfaces share
the boundary (e.g. the north and the south hemispheres of the same sphere), the flux of a curl through them
will give the same value. The equivalences of (50) can be translated to this setting:
40 If ∇ ~ or ∇
~ ·F ~ ×F~ are constant in a neighbourhood of ~r0 , these proofs are immediate from divergence and Stokes’ theorem:
RR RRR
~ · d~
F S (101) ∇~ ·F~ dV ~ · F(~
~ r0 )
∂BR BR (40) Vol(BR )∇ ~ · F(~
~ r0 ),
lim = lim = lim =∇
R→0 Vol(BR ) R→0 Vol(BR ) R→0 Vol(BR )
H RR
~ · d~r
F ~ × F)
(∇ ~ · d~ S (40) ~ × F(~
~ r0 )) · n̂

∂QR (â) (106) QR (â) Area(QR (â))(∇ ~ × F(~
~ r0 ) · â,
lim = lim = lim = ∇
R→0 Area(QR (â)) R→0 Area(QR (â)) R→0 Area(QR (â))
(since n̂ = â on QR (â)); in the general case the proofs require a suitable version of the mean value theorem (attempt to
write the proofs!).
A. Moiola, University of Reading 81 Vector calculus lecture notes, 2016–17
ZZ ZZ
~
G=∇×A~ ~ ~ ~
G · dS ~ · d~S = 0
G ~ ·G
∇ ~ =0
~ ⇐⇒ S ⇐⇒ ∂D =⇒ ~
(∃ vector potential for G) is independent of S ∀ closed surfaces ∂D (G solenoidal)

RR RR
(where, with “independent of S”, we mean that G ~ · d~S if SA and SB are two surfaces
~ · d~S =
G
SA SB
with ∂SA = ∂SB ). The converse of the last implication is true if the domain of G ~ does not contain “holes”.

⋆ Remark 3.35. Often the name “Stokes’ theorem” is referred to a much more general version of it, coming
from the branch of mathematics known as “differential geometry”. This involves the integration of a differential
operator called “exterior derivative”, simply denoted by “ d”. This acts on “differential forms”, which are
generalisations of scalar and vector fields defined on a “manifold” Ω, which in turn is an object that generalises
domains, paths and surfaces
R to any
R dimension. This extremely general and deep theorem is stated with a simple
and elegant formula: Ω dω = ∂Ω ω. The fundamental theorem of (vector) calculus, Green’s, divergence and
Stokes’ theorems are special instances of this result. If you are interested in deepening this topic, take a look
at Chapter 10 of W. Rudin’s book “Principles of mathematical analysis” (this is not an easy read!).

Z b
Fundamental theorem dφ
1D (90) dt = φ(b) − φ(a)
of calculus dt
Za
(49) Fundamental theorem ~ · d~r = f (~q) − f (~p)
3D* of vector calculus ∇f
(91)
ZΓZ I
∂f
2D (92) Lemma 3.3 dA = − f dx
∂y
Z ZR I ∂R
∂f
2D (93) Lemma 3.3 dA = f dy
∂x
Z R ∂R
 I
∂F2 ∂F1

2D (95) Green’s theorem − dA = F1 dx + F2 dy
∂x ∂y
ZRZ Z Z ∂R I
2D (96) Green’s theorem ~ × F)
(∇ ~ 3 dA = ~ × F)
(∇ ~ · k̂ dA = ~ · d~r
F
Z ZR I R I ∂R
2D (98) Divergence theorem ~ · F)
(∇ ~ dA = (F1 dy − F2 dx) = ~ · n̂ ds
F
ZZZR ZZ∂R ∂R
(99) ∂f
3D Lemma 3.13 dV = f k̂ · d~S
(100) D ∂z ∂D
ZZZ ZZ
3D (101) Divergence theorem ~ dV =
~ · F)
(∇ F~ · d~S
ZZZ D ZZ ∂D

3D (102) Corollary 3.22 ~ dV =

∇f f n̂ dS
ZZZ D ∂D
ZZ
3D (103) Corollary 3.22 ∇~ ×F ~ dV = ~ × n̂ dS
F
ZZZ D ∂D ZZ
~ · ∇f
~ ) dV = ∂g
3D (104) Green’s 1st identity (f ∆g + ∇g f dS
D ∂D ∂n
ZZZ ZZ 
∂g ∂f 
3D (105) Green’s 2nd identity (f ∆g − g∆f ) dV = f −g dS
∂n ∂n
ZZ D Z ∂D

3D* (106) Stokes’ theorem ~ ×F

∇ ~ · d~S = ~ · d~r
F
S ∂S

Table 3: A summary of the important integro-differential identities proved in Section 3. Here:

Γ is a path with starting point ~p and end point ~q;
R ⊂ R2 is a two-dimensional, bounded, connected, piecewise smooth region;
D ⊂ R3 is a three-dimensional, bounded, connected, piecewise smooth domain;
(S, n̂) is a piecewise smooth, bounded, oriented surface;
φ is a real function;
f and g are smooth scalar fields;
~ is a smooth vector field.
F
(* Note that the first column of the table denotes the dimension of the space in which the domain of
integration is defined; the intrinsic dimension of the domain of integration for the fundamental theorem
of vector calculus is one (oriented path), and for Stokes’ theorem is two (oriented surface).)
A. Moiola, University of Reading 82 Vector calculus lecture notes, 2016–17

A General overview of the notes

The three main sections of these notes discuss the following general topics:
1. differential vector calculus (differentiation of vector quantities);
2. integral vector calculus (integration of vector quantities);
3. generalisations of the fundamental theorem of calculus (relation between differentiation and integra-
tion of vector quantities).
~ ∈ R3 and scalars λ ∈ R:
Section 1.1 recalls some operations involving vectors ~u, ~v, w
inputs output
Vector addition: 2 vectors ~u, w~ 7 → vector ~u + w ~
Scalar-vector multiplication: a vector ~u and a scalar λ 7→ vector λ~u
Scalar product: 2 vectors ~u, w~ 7→ scalar ~u · w
~
Vector product: 2 vectors ~u, w~ 7→ vector ~u × w ~
Triple product: 3 vectors ~u, ~v, w
~ 7 → scalar ~u · (~v × w
~)
Sections 1.2 introduces different kinds of functions (or fields), taking as input and returning as output
either scalar values (in R) or vectors (in R3 ):
Real functions (of real variable) f :R→R t 7→ f (t)
Curves ~a : R → R3 t 7→ ~a(t)
Scalar fields f : R3 → R ~r 7→ f (~r)
Vector fields ~ : R3 → R3
F ~ r)
~r 7→ F(~
∂ ∂
The fundamental differential operators for scalar and vector fields are the partial derivatives ∂x , ∂y and
∂
∂z .They can be combined to construct several vector differential operators, as described in Section 1.3.
The most important are:
∂ ∂ ∂
Partial derivatives ∂x , ∂y or ∂z scalar 7 → scalar
Gradient ~
∇ scalar 7→ vector
Divergence ~ (or div)
∇· vector 7 → scalar
Curl ~ (or curl)
∇× vector 7 → vector
Laplacian ∆ scalar 7→ scalar
Proposition 1.52 describes the result of the composition of differential operators.
Proposition 1.55 describes the result of the application of differential operators to different products of
scalar and vector fields (generalising the product rule).
Section 1.5 studies some relations between vector fields F ~ lying in the kernel (or nullspace) of the
divergence and the curl operators (i.e. solenoidal fields ∇~ ·F~ = 0 and irrotational fields ∇ ~ ×F~ = ~0) and
~ = ∇ϕ
those in the image of the gradient and the curl (i.e. fields admitting a scalar potential F ~ or a vector
~ =∇
potential F ~ × A).
~
Section 1.6 describes the total derivatives (in t) of curves, denoted d~a
dt .
Section 1.7 investigates the differentiation of composition of fields and curves.
Sections 2.1 and 2.2 extends the definition of integrals of real functions to:
Type of integral domain of integration integrand notation
Rb
Integrals of real functions interval (a, b) real function Ra f dt
Line integrals of scalar fields path Γ scalar field f ds
RΓ
~
Line integrals of vector fields oriented path Γ vector field RRΓ F · d~r
Double integrals planar domain R ⊂ R2 scalar field RRRR
f dA
Triple integrals domain D ⊂ R3 scalar field RR f dV
D
Surface integrals surface S scalar field f dS
RRS
Fluxes oriented surface (S, n̂) vector field ~ · d~S
F
R
Several theorems establish important relations between integration and differentiation, see Table 3.
The most relevant are:
The integral of (the) of a is equal of at/on the
on a(n) to the
FTC interval (a, b) derivative function f difference f endpoints
FTVC oriented path Γ τ̂ · gradient scalar field f difference f endpoints
Green 2D region R k̂· curl ~ ~
vector field F circulation F boundary ∂R
Stokes oriented surface (S, n̂) n̂· curl ~ circulation F
vector field F ~ boundary ∂S
Divergence 3D domain D divergence vector field F ~ flux ~ boundary ∂D
F
1st domain differential function type of 2nd domain
of integration operator or field integral of integration
A. Moiola, University of Reading 83 Vector calculus lecture notes, 2016–17

B Solutions of the exercises

B.1 Exercises of Section 1
Solution of Exercise 1.6. ı̂ · ı̂ = ̂ · ̂ = k̂ · k̂ = 1, ı̂ · ̂ = ̂ · ı̂ = ı̂ · k̂ = k̂ · ı̂ = ̂ · k̂ = k̂ · ̂ = 0.
Solution of Exercise 1.7. Existence. Since ~u|| must be parallel to w ~ and w~ 6= ~0, it must have the
form ~u|| = α~w for some α ∈ R. Thus ~ u⊥ = ~u − ~u|| = ~u − α~ w. The orthogonality condition gives
0=w ~ · ~u⊥ = w
~ · (~u − α~ w|2 , i.e. α = w
~ · ~u − α|~
w) = w w|2 . Thus the pair
~ · ~u/|~

~ · ~u
w ~ · ~u
w
~
u|| = ~,
w ~u⊥ = ~u − ~u|| = ~u − ~
w
w |2
|~ w |2
|~

satisfies all the requests.

Uniqueness. Assume by contradiction that two pairs ~u|| , ~u⊥ and ~v|| , ~v⊥ give an orthogonal decompo-
sition of ~u with respect to w
~ . Since ~u|| and ~v|| are both parallel to w~ , they can be written as ~u|| = α~
w
and ~v|| = β w~ . Reasoning as in the existence part of the exercise, we find a unique possible value of α and
β, so ~u|| = ~v|| and consequently ~u⊥ = ~u − ~u|| = ~u − ~v|| = ~v⊥ and there exists only one admissible pair.
Solution of Exercise 1.8. We only need to verify that the two scalar products between ~u × w
~ and the
two vectors are zero:
(2)

(~u × w
~ ) · ~u = (u2 w3 − u3 w2 )ı̂ + (u3 w1 − u1 w3 )̂ + (u1 w2 − u2 w1 )k̂ · ~u
= u2 w3 u1 − u3 w2 u1 + u3 w1 u2 − u1 w3 u2 + u1 w2 u3 − u2 w1 u3
= w1 (u3 u2 − u2 u3 ) + w2 (u1 u3 − u3 u1 ) + w3 (u2 u1 − u1 u2 ) = 0,


(~u × w
~)·w
~ = (u2 w3 − u3 w2 )ı̂ + (u3 w1 − u1 w3 )̂ + (u1 w2 − u2 w1 )k̂ · w~
= u2 w3 w1 − u3 w2 w1 + u3 w1 w2 − u1 w3 w2 + u1 w2 w3 − u2 w1 w3 = 0.

Solution of Exercise 1.9. We write in detail only ı̂ × ̂ and ı̂ × ı̂, all the other cases are similar:
 
 ı̂ ̂ k̂      
 0 0 1 0  1 0
ı̂ × ̂ = 1 0 0  = ı̂ 
 
− ̂   + k̂  = 0ı̂ + 0̂ + 1k̂ = k̂,
0 1 0 0 0 0 1
1 0
 
 ı̂ ̂ k̂      
 0 0 1 0  1 0
ı̂ × ı̂ = 1 0 0  = ı̂ 
 
− ̂   + k̂  = 0ı̂ + 0̂ + 0k̂ = ~0.
1 0 0 1 0 1 0
0 0

Solution of Exercise 1.10. The anticommutativity w ~ ×~u = −~u × w

~ follows immediately by computing
~ × ~u with Formula (2) and swapping the uj ’s with the wj ’s. From this it follows that ~u × ~u = −~u × ~u,
w
so 2~u × ~u = ~0.
Solution of Exercise 1.11. We use Exercises 1.9 and 1.10: (ı̂ × ̂) × ̂ = k̂ × ̂ = −ı̂ 6= ~0 = ı̂ × ~0 =
ı̂ × (̂ × ̂).
Solution of Exercise 1.12. We prove identity (4) componentwise, namely we verify that the three
components on the left-hand side and the right-hand side of the identity agree with each other. Consider
the first component:
  (2)
~ ) 1 = u2 [~v × w
~u × (~v × w ~ ]3 − u3 [~v × w
~ ]2
(2)
= u2 (v1 w2 − v2 w1 ) − u3 (v3 w1 − v1 w3 )
= (u2 w2 + u3 w3 )v1 − (u2 v2 + u3 v3 )w1
= (~u · w
~ − u1 w1 )v1 − (~u · ~v − u1 v1 )w1
= (~u · w
~ )v1 − (~u · ~v)w1
 
= (~u · w~ )~v − (~u · ~v)~
w 1.

A similar computation holds for the y- and the z-components, thus the assertion (4) follows.
A. Moiola, University of Reading 84 Vector calculus lecture notes, 2016–17

Solution of Exercise 1.13. Jacobi identity follows from applying three times the identity (4), using
the commutativity of the scalar product, and cancelling all terms:

~u × (~v × w
~ ) + ~v × (~ w×~ ~ × (~u × ~v)
u) + w
(4)



= ~v(~u · w~)−w ~ (~u · ~v) + w ~ (~v · ~u) − ~u(~v · w
~ ) + ~u(~ w · ~v) − ~v(~w · ~u)




= ~u (~ ~ ) + ~v(~u · w
w · ~v) − (~v · w ~ ) − (~w · ~u) + w ~ (~v · ~u) − (~u · ~v) = ~0.

To prove Binet–Cauchy identity we expand the left-hand side, collect the positive terms, add and subtract
the term u1 w1 v1 p1 + u2 w2 v2 p2 + u3 w3 v3 p3 :
(~
u × ~v) · (w
~ × ~p)
   
= (u2 v3 − u3 v2 )ı̂ + (u3 v1 − u1 v3 )̂ + (u1 v2 − u2 v1 )k̂ · (w2 p3 − w3 p2 )ı̂ + (w3 p1 − w1 p3 )̂ + (w1 p2 − w2 p1 )k̂
= u2 v3 w2 p3 + u3 v2 w3 p2 + u3 v1 w3 p1 + u1 v3 w1 p3 + u1 v2 w1 p2 + u2 v1 w2 p1
− u3 v2 w2 p3 − u2 v3 w3 p2 − u1 v3 w3 p1 − u3 v1 w1 p3 − u2 v1 w1 p2 − u1 v2 w2 p1
= u2 v3 w2 p3 + u3 v2 w3 p2 + u3 v1 w3 p1 + u1 v3 w1 p3 + u1 v2 w1 p2 + u2 v1 w2 p1 + (u1 w1 v1 p1 + u2 w2 v2 p2 + u3 w3 v3 p3 )
− u3 v2 w2 p3 − u2 v3 w3 p2 − u1 v3 w3 p1 − u3 v1 w1 p3 − u2 v1 w1 p2 − u1 v2 w2 p1 − (u1 w1 v1 p1 + u2 w2 v2 p2 + u3 w3 v3 p3 )
= (u1 w1 + u2 w2 + u3 w3 )(v1 p1 + v2 p2 + v3 p3 ) − (u1 p1 + u2 p2 + u3 p3 )(w1 v1 + w2 v2 + w3 v3 )
= (~ ~ )(~v · ~
u·w p) − (~ p)(~v · w).
u·~ ~

~ = ~u and ~p = ~v in the Binet–Cauchy identity and moving

Lagrange identity is obtained by choosing w
the last term to the left-hand side.
Solution of Exercise 1.15. The formula u ~ · (~v × w
~ ) = ~v · (~
w × ~u) = w
~ · (~u × ~v) follows immediately
from (5). From this and the anticommutativity of the vector product (Exercise 1.10) we have

~ · (~v × ~u) = −w
w ~ · (~u × ~v) = −~u · (~v × w
~ ), ~u · (~v × ~u) = ~v · (~u × ~u) = 0.

Solution of Exercise 1.18. ⋆ (i) If limj→∞ ~uj = u ~ then limj→∞ |~uj − ~u| = 0 by definition. Since
0 ≤ |(~uj )1 − u1 | ≤ |~uj − ~u| from the definition of vector magnitude, by the sandwich theorem (see
Theorem 4.18 in the Real Analysis lecture notes) we have limj→∞ |(~uj )1 − u1 | = 0, which is equivalent to
limj→∞ (~uj )1 = u1 . Similarly, we can prove limj→∞ (~uj )2 = u2 and limj→∞ (~uj )3 = u3 .
To prove the converse, assume limj→∞ (~uj )1 = u1 , limj→∞ (~uj )2 = u2 and limj→∞ (~uj )3 = u3 . Then,
we obtain the desired convergence simply using the properties of limits and the definition of magnitude:
q
2
2
2
lim |~uj − ~
u| = lim (~uj )1 − u1 + (~uj )2 − u2 + (~uj )3 − u3
j→∞ j→∞
r

2
2
2
= lim (~uj )1 − u1 + lim (~uj )2 − u2 + lim (~uj )3 − u3 = 0.
j→∞ j→∞ j→∞

~ 6= ~u with |~
(ii) It suffices to take a sequence converging to some w w| = |~u|. For example, ~uj = ı̂ for
all j ∈ N and ~ u = k̂.
Solution of Exercise 1.19. ⋆ Let D ⊂ R3 be open and denote C := R3 \ D its complement. We want
to prove that C is closed. Consider a sequence {~pj }j∈N ⊂ C such that limj→∞ ~pj = ~p. To prove that C
is closed we need to prove that ~p ∈ C, which is the same as ~p ∈ / D, by definition of set complement. We
proceed by contradiction: if ~p ∈ D was true, then there would exist ǫ > 0 such that all points at distance
smaller than ǫ from ~p were in D. But from limj→∞ |~pj −~p| = 0, there exists a ~pj in the sequence (actually,
infinitely many) at distance smaller than ǫ from ~p but belonging to C. So we have a contradiction: ~p
must belong to C, thus C is closed.
Now we prove the second statement (the vice versa), which is very similar. Let C ⊂ R3 be closed and
denote D := R3 \C its complement. We want to prove that D is open. Again, we proceed by contradiction:
if D were not open, there would be a point ~p ∈ D such that for all ǫ > 0 we have a point ~pǫ ∈ / D (i.e.
~pǫ ∈ C) such that |~p − ~pǫ | < ǫ. We fix a real positive sequence ǫj converging to 0 (e.g. ǫj = 1/j). Then by
the contradiction assumption there exists ~pj ∈ C such that |~pj − ~p| < ǫj → 0. Thus limj→∞ ~pj = ~p. As
C is closed, this means that ~ p ∈ C, but we assumed ~p ∈ D, so we have a contradiction and we conclude.
A. Moiola, University of Reading 85 Vector calculus lecture notes, 2016–17

|~r|2 x2 +y 2 +z 2 y 2 +z 2
Solution of Exercise 1.29. (We expand p(~r) = x2 = x2 =1+ x2 .)

∂f ∂f ∂f
f (~r) = xyez , = yez , = xez , = xyez = f ;
∂x ∂y ∂z
xy ∂g y ∂g xz ∂g −xy
g(~r) = , = , = , = ;
y+z ∂x y+z ∂y (y + z)2 ∂z (y + z)2
∂h ∂h z 3 eyz ∂h (2z + yz 2 )eyz
h(~r) = log(1 + z 2 eyz ), = 0, = , = ;
∂x ∂y 1 + z 2 eyz ∂z 1 + z 2 eyz
p ∂ℓ x ∂ℓ 2y 3 ∂ℓ 3z 5
ℓ(~r) = x2 + y 4 + z 6 , = p , = p , = p ;
∂x x + y4 + z 6
2 ∂y x2 + y 4 + z 6 ∂z x2 + y 4 + z 6
∂m y ∂y ∂m
m(~r) = xy = e(log x)y , = e(log x)y = e(log x)y log x = 0;
∂x x ∂y ∂z
|~r|2 ∂m −2y 2 − 2z 2 ∂y 2y ∂m 2z
p(~r) = = , = 2, = 2.
x2 ∂x x3 ∂y x ∂z x

Solution of Exercise 1.34.

~ = yez ı̂ + xez ̂ + xyez k̂, ~ = y xz xy

∇f ∇g ı̂ + 2
̂ − k̂,
y+z (y + z) (y + z)2
3 yz 2 yz 3 5
~ = z e ̂ + (2z + yz )e k̂ ,
∇h ~ = xı̂p+ 2y ̂ + 3z k̂ ,
∇ℓ
1+z e2 yz
x2 + y 4 + z 6
2 2
~ = y e(log x)y ı̂ + e(log x)y log x̂ = y xy−1 ı̂ + xy log x̂,
∇m ~ = −2y − 2z ı̂ + 2y̂ + 2z k̂ .
∇p
x x 3 x2

Solution of Exercise 1.35. The field s is a polynomial in three variables, so its gradient is easy to
compute:
~ r) = ∇(|~
∇s(~ ~ r|2 ) = ∇(x
~ 2 + y 2 + z 2 ) = 2xı̂ + 2y̂ + 2z k̂ = 2~r.

To compute ∇m ~ we use a simple trick (otherwise we can compute it directly). For α ∈ R, define the
real function Gα (t) := tα/2 , whose derivative is Gα ′ (t) = (α/2)tα/2−1 . Using the chain rule (14) for the
gradient we obtain the gradient of |~r|α :






~ r |α ) = ∇
∇(|~ ~ s(~r)α/2 = ∇ ~ Gα s(~r) = Gα ′ s(~r) ∇s(~~ r) = α s(~r) α/2−1 2~r = α |~r|2 α/2−1~r = α|~r|α−2~r.
2
~
The field m(~r) = |~r| corresponds to the case α = 1, so ∇m(~
r) = ~r/|~r|, as required.
Solution of Exercise 1.37. The level surfaces of f = (2x + y)3 are the sets {~r, (2x + y)3 = C}, or
equivalently {~r, 2x + y = C}. These are parallel planes. They are all perpendicular to the xy-plane, so k̂
is tangent to each one of these planes. They intersect the xy-plane in the lines {2x + y = C, z = 0}, so
also the vector ı̂ − 2̂ is tangent to all planes in all their points.
We have G ~ = 2ez ı̂ + ez ̂, so G
~ · k̂ = 0 and G
~ · (ı̂ − 2̂) = 2ez − 2ez = 0, thus G
~ is perpendicular to the
~
level sets of G in all points.
Alternatively one can check that ∇f ~ = 3(2x + y)2 (2ı̂ + ̂) = 3(2x + y)2 e−z G,
~ so ∇f
~ and G ~ are parallel
and we conclude by Proposition 1.33(4).
~
Solution of Exercise 1.39. We compute the gradient of f and the unit vector in the direction of F:
~ r)
F(~ xı̂ + ̂
~ (~r) = (ex + z 2 )ı̂ − cos y̂ + 2xz k̂,
∇f = √ .
~ r)|
|F(~ x2 + 1
The directional derivative is
∂f ~ r)
F(~
x
+ z 2 ) − cos y
(~r) = ~ (~r) = √xı̂ + ̂ · (ex + z 2 )ı̂ − cos y̂ + 2xz k̂ = x(e √
· ∇f .
∂ F̂ ~ r)|
|F(~ x2 + 1 x2 + 1

Solution of Exercise 1.41. We simply compute all the partial derivatives:

     
2 0 0 0 z y 2x 2y 2z
~ r) = 0 −2 0 ,
J F(~ ~ r ) = z
J G(~ 0 x , ~ r) =  0
J H(~ 0 0 .
0 0 0 y x 0 0 − sin y 0
A. Moiola, University of Reading 86 Vector calculus lecture notes, 2016–17

Solution of Exercise 1.44. We first compute the gradients and then all the second derivatives (recall
the identity h(~r) = (x2 + y 2 + z 2 )2 = x4 + y 4 + z 4 + 2x2 y 2 + 2x2 z 2 + 2y 2 z 2 ):

~ = 2xı̂ − 2y̂,
∇f ~ = yez ı̂ + xez ̂ + xyez k̂,
∇g
~ = ∇(x
∇h ~ 2 + y 2 + z 2 )2 = 2(x2 + y 2 + z 2 )(2xı̂ + 2y̂ + 2z k̂) = 4|~r|2~r,
   
2 0 0 0 ez yez
Hf = 0 −2 0 , Hg =  ez 0 xez  ,
0 0 0 yez xez xyez
 
12x2 + 4y 2 + 4z 2 8xy 8xz
Hh =  8xy 4x2 + 12y 2 + 4z 2 8yz ,
8xz 8yz 4x2 + 4y 2 + 12z 2
∆f = 0, ∆g = xyez = g, ∆h = 20|~r|2 .

The field f is a “harmonic function”, since its Laplacian vanishes everywhere.

Solution of Exercise 1.46. We only have to compute three derivatives for each field and sum them:

~ ·F
∇ ~ = ∂(2x) + ∂(−2y) + ∂0 = 2 − 2 + 0 = 0,
∂x ∂y ∂z
~ ~
∇ · G = 0 + 0 + 0 = 0,
~ ·H
∇ ~ = 2x + 0 + 0 = 2x.

Solution of Exercise 1.50. For each field we have to evaluate six partial derivatives (the non-diagonal
~ and sum them appropriately:
terms of the Jacobian matrix J F)

~ ×F
∇ ~ = ~0,
~ ×G
∇ ~ = (x − x)ı̂ + (y − y)̂ + (z − z)k̂ = ~0,
~ ×H
∇ ~ = (− sin y − 0)ı̂ + (2z − 0)̂ + (0 − 2y)k̂ = − sin y ı̂ + 2z̂ − 2y k̂.

Solution of Exercise 1.61. We compute the left-hand side and the right-hand side, starting from the
computation of the terms appearing in them, and we verify that they give the same result:

F~ ×G~ = x2 exy ı̂ − zexy ̂,

~ · (F
left-hand side = ∇ ~ × G)~ =∇ ~ · (x2 exy ı̂ − zexy ̂) = 2xexy + x2 yexy − zxexy = (2x + x2 y − zx)exy ;
~ ×F
∇ ~ = 3y 2 ı̂ + ̂ + 2xk̂, ∇~ ×G ~ = xexy ı̂ − yexy ̂,
~ × F)
right-hand side = (∇ ~ ·G
~ −F
~ · (∇
~ × G)
~
= (3y 2 ı̂ + ̂ + 2xk̂) · (exy k̂) − (zı̂ + x2 ̂ + y 3 k̂) · (xexy ı̂ − yexy ̂)
= 2xexy − zxexy + x2 yexy = (2x − zx + x2 y)exy = left-hand side.

Solution of Exercise 1.62. We first compute the divergence and the curl of the position vector:

~ · (xı̂ + y̂ + z k̂) = ∂x + ∂y + ∂z = 3,

~ · (~r) = ∇
∇
∂x ∂y ∂z
     
~ × (~r) = ∇
∇ ~ × (xı̂ + y̂ + z k̂) = ∂z − ∂y ı̂ + ∂x − ∂z ̂ + ∂y − ∂x k̂ = ~0.
∂y ∂z ∂z ∂x ∂x ∂y

(This solves the exercise for α = 0.) To compute divergence and curl of the vector field F ~ = |~r|α~r we
α
decompose it as the product between the scalar field |~r| (for which we have already computed the gradient
in Exercise 1.35) and the vector field ~r, so that we can apply the product rules (29) and (31):

~ · (|~r|α~r) (29)
∇ = (∇ ~ · ~r)|~r|α + ~r · ∇(|~
~ r |α )


= 3|~r|α + ~r · α|~r|α−2~r from solution to Exercise 1.35
A. Moiola, University of Reading 87 Vector calculus lecture notes, 2016–17

= (3 + α)|~r|α from ~r · ~r = |~r|2 ,

~ × (|~r|α~r) (31)
∇ ~ r|α ) × ~r + |~r|α ∇
= ∇(|~ ~ × ~r = α|~r|α−2 ~r × ~r +~0 = ~0.
| {z } | {z }
=~
0 =~
0

Solution of Exercise 1.63. (i) Using that for all vectors ~u × u ~ = ~0, we have:
(31) (26)
~ × (f ∇f
∇ ~ ) = (∇f~ ) × (∇f
~ ) + f∇ ~ × (∇f
~ ) = ~0.
(ii) Using vector identities, the chain rule (14) (with G(u) = uℓ ) and again ~u × ~u = ~0 for all ~u ∈ R3 :



∇ ~ ℓ ) (31)
~ × f n ∇(f ~ n ) × ∇(f
= ∇(f ~ ℓ) + f n ∇
~ × ∇~ (f ℓ )
| {z }
=~
0,(26)
(14)
~ ) × (ℓf ℓ−1 ∇f
= (nf n−1 ∇f ~ ) = nℓf n+ℓ−2 ∇f
~ × ∇f
~ = ~0.

Solution of Exercise 1.67. Proceeding as in the example, we obtain the potentials

x4 y 4 z3  2
ϕA = x2 + y 3 + ez , ϕC = + , ~ D = xy + y k̂.
A
4 3 2

Solution of Exercise 1.68. By direct computation we obtain ∇~ ~ r = ∂x + ∂y + ∂z = 3, so the position

∂x ∂y ∂z
vector is not solenoidal. Since the jth component of ~r depends only on the jth component of its argu-
ment (which is itself!), by definition its curl vanishes, so the position vector field is irrotational. From
~ r|2 ) = 2~r, so a scalar potential of ~r is 1 |~r|2 .
equation (15) we have ∇(|~ 2

~ =F
Solution of Exercise 1.69. Simply take H ~ × G:
~ by (30) we have
~ H
∇· ~ = ∇·(
~ F ~ × G)
~ = (∇×
~ F)
~ ·G~ −F
~ · (∇×
~ G)
~ = ~0 · G
~ −F
~ · ~0 = 0, so H
~ is solenoidal.

Solution of Exercise 1.73. If the field F~ is conservative then its scalar potential ϕ satisfies F
~ = ∇ϕ,
~
~ ·F
so by identity (24) we have ∇ ~ =∇
~ · (∇ϕ)
~ = ∆ϕ.
Solution of Exercise 1.74. If the gradient of a scalar field f admits a vector potential A,~ we have that
~ ~ ~ ~ ~
∇f = ∇ × A. If the vector potential itself is conservative, it means that A = ∇ψ for some scalar field ψ.
~ =∇
Thus ∇f ~ ×A ~ =∇ ~ × ∇ψ
~ = ~0 by the identity (26). A scalar field whose gradient vanishes everywhere
is constant, so the fields sought are simply the constant ones f (~r) = λ for all ~r.
Solution of Exercise 1.75. Using that the curl of a gradient vanishes, we immediately see that the
product is solenoidal:

~ · (F
∇ ~ (30)
~ × G) = (∇~ × F)
~ ·G
~ −F
~ · (∇
~ × G)
~ = (∇
~ × ∇ϕ ~ −F
~ )·G ~ · (∇
~ × ∇ψ
~ ) = 0.
| {z } | {z }
=~
0, (26) =~
0, (26)

Two easy choices of the potential are possible:

~ := ϕ∇ψ
A ~ ⇒ ∇ ~ (31)
~ ×A ~
= (∇ϕ) ~
× (∇ψ) ~ × (∇ψ)
+ ϕ∇ ~ ~ × G;
=F ~
| {z }
=0, (26)
(31)
~ := −ψ ∇ϕ
B ~ ⇒ ∇ ~ = −(∇ψ)
~ ×B ~ ~
× (∇ϕ) ~ × (∇ϕ)
−ψ∇ ~ = −G ~ (3)
~ ×F = F ~ × G.
~
| {z }
=0, (26)

~
Solution of Exercise 1.78. We have d~ a db
dt (t) = − sin t ı̂ + cos t ̂ and dt (t) = −2 sin 2τ ı̂ + 2 cos 2τ ̂.
d~a
The path of dt is again the unit circle, but the initial point is placed in ̂ (as opposed to ı̂ as for ~a).
~
The path of ddtb is the circle centred at the origin with radius 2. Since ~b moves more quickly than ~a (it
covers the unit circle in “time” π as opposed to 2π) the total derivative, representing its velocity, has
greater magnitude.
To find a curve whose derivative equals that of ~a it suffices to add a constant vector, e.g. ~c(t) =
(3 + cos t)ı̂ + sin t ̂.
A. Moiola, University of Reading 88 Vector calculus lecture notes, 2016–17

Solution of Exercise 1.79. We simply compute all the terms:

d~a d~b
= − sin t ı̂ + cos t ̂, = −2 sin 2t ı̂ + 2 cos 2t ̂,
dt dt
~ (1)
(~a · b) = cos t cos 2t + sin t sin 2t,
LHS of (36)ii (~a · ~b) = − sin t cos 2t − 2 cos t sin 2t + cos t sin 2t + 2 sin t cos 2t
′

= sin t cos 2t − cos t sin 2t,

d~a ~ d~b
RHS of (36)ii · b + ~a · = (− sin t ı̂ + cos t ̂) · (cos 2t ı̂ + sin 2t ̂)
dt dt
+ (cos t ı̂ + sin t ̂) · (−2 sin 2t ı̂ + 2 cos 2t ̂)
(1)
= sin t cos 2t − cos t sin 2t,
(2)
(~a × ~b) = (cos t sin 2t − sin t cos 2t)k̂,
d(~a × ~ b)
LHS of (36)iii = (− sin t sin 2t + 2 cos t cos 2t − cos t cos 2t + 2 sin t sin 2t)k̂
dt
= (cos t cos 2t + sin t sin 2t)k̂ = cos(2t − t)k̂ = cos t k̂,
d~a ~ d~b
RHS of (36)iii × b + ~a × = (− sin t ı̂ + cos t ̂) × (cos 2t ı̂ + sin 2t ̂)
dt dt
+ (cos t ı̂ + sin t ̂) × (−2 sin 2t ı̂ + 2 cos 2t ̂)
(2)
= (− sin t sin 2t − cos t cos 2t)k̂ + (2 cos t cos 2t + 2 sin t sin 2t)k̂
= (cos t cos 2t + sin t sin 2t)k̂ = cos t k̂.

Solution of Exercise 1.80. We simply use the definition of scalar and vector product, the product rule
for real functions (F G)′ = F ′ G + F G′ , and rearrange some terms. Since the second equation is vectorial,
we prove the identity only for its first component.
(1)
′
(~a · ~b)′ = a1 b1 + a2 b2 + a3 b3


= a′1 b1 + a1 b′1 + a′2 b2 + a2 b′2 + a′3 b3 + a3 b′3
(1)




= a′1 ı̂ + a′2 ̂ + a′3 k̂ · b1 ı̂ + b2 ̂ + b3 k̂ + a1 ı̂ + a2 ̂ + a3 k̂ · b′1 ı̂ + b′2 ̂ + b′3 k̂
d~a ~ d~b
= · b + ~a · ;
dt dt
 d(~a × ~b)  d(~a × ~b)1 (2)
= = (a2 b3 − a3 b2 )′ = (a′2 b3 + a2 b′3 − a′3 b2 − a3 b′2 ) = (a′2 b3 − a′3 b2 ) + (a2 b′3 − a3 b′2 )
dt 1 dt
 a
(2) d~
  d~b 
= × ~b + ~a × .
dt 1 dt 1

Solution of Exercise 1.81. Saying that ~a equal its own derivative means that d~ a
dt (t) = ~
a(t), which is
a vector linear ordinary differential equation (ODE). This can be written in components as da dt = a1 ,
1

da2 da3
dt = a2 and dt = a3 , so we know how to solve it from first-year calculus. Using the initial condition
~a(0) = ~u we have a1 (t) = u1 et , a2 (t) = u2 et and a3 (t) = u3 et , which can be written in vector form as
~a(t) = ~uet . We know from ODE theory that this is the unique solution of the differential equation. If
~u = ~0 the path of ~a collapses to a point. If ~u 6= ~0 the path of ~a is the straight half-line starting at ~u and
pointing away from the origin. Be careful: despite the expression ~a(t) = ~uet contains an exponential, the
corresponding path is a straight line, recall the difference between path and graph!
Solution of Exercise 1.83. (i) We compute the total derivative of ~a, the gradient of f evaluated in ~a,
and their scalar product according to the chain rule (37):

d~a ~ = yez ı̂ + xez ̂ + xyez k̂ = t3 ı̂ + t̂ + t4 k̂

= ı̂ + 3t2 ̂, ∇f
dt
df ~ · d~a = t3 + 3t3 = 4t3 .
⇒ = ∇f
dt dt
A. Moiola, University of Reading 89 Vector calculus lecture notes, 2016–17

(ii) Alternatively, we can compute the composition f (~a(t)) and then derive it with respect to t:



3 0 4 df df ~a(t)
f ~a(t) = t t e = t , = = 4t3 .
dt dt

Solution of Exercise 1.87. We first compute all the compositions and then take the desired derivatives:
~ r) = g(xı̂ + z̂ − y k̂) = x2 − z 2 ,
(g ◦ F)(~
(g ◦ ~a)(t) = g(cos tı̂ + sin t̂) = cos2 t − sin2 t,


~ ◦ ~a)(t) = g F(cos
(g ◦ F ~ tı̂ + sin t̂) = g(cos tı̂ − sin tk̂) = cos2 t,
~
∂(g ◦ F) ∂(x2 − z 2 ) ~
∂(g ◦ F) ∂(x2 − z 2 ) ~
∂(g ◦ F) ∂(x2 − z 2 )
= = 2x, = = 0, = = −2z,
∂x ∂x ∂y ∂y ∂z ∂z
d(g ◦ ~a) d(cos2 t − sin2 t)
= = −4 sin t cos t,
dt dt
~ ◦ ~a)
d(g ◦ F d(cos2 t)
= = −2 sin t cos t.
dt dt
Using the vector versions of the chain rule, the computations turn out to be slightly more complicated, in
particular we have to be careful to evaluate the fields at the right values:
~
∂F ∂F~ ∂F~
~ = 2xı̂ − 2y̂,
∇g = ı̂, = −k̂, = ̂,
∂x ∂y ∂z
~
∂(g ◦ F) ~

(38)
= ∇g~ · ∂ F = 2F1 (~r)ı̂ − 2F2 (~r)̂ · (ı̂) = 2F1 (~r) = 2x,
∂x ∂x
~
∂(g ◦ F) ~

(38)
= ∇g~ · ∂ F = 2F1 (~r)ı̂ − 2F2 (~r)̂ · (−k̂) = 0,
∂y ∂y
~
∂(g ◦ F) ~

(38)
= ∇g~ · ∂ F = 2F1 (~r)ı̂ − 2F2 (~r)̂ · (̂) = −2F2 (~r) = −2z,
∂z ∂z
d(g ◦ ~a)

(37)
~ a) · d(~a) = 2a1 (t)ı̂ − 2a2 (t)̂ · (− sin tı̂ + cos t̂)
= ∇g(~
dt dt
= (2 cos tı̂ − 2 sin t̂) · (− sin tı̂ + cos t̂) = −4 sin t cos t,
~ ◦ ~a)
d(g ◦ F ~  
(37)
= ∇g~ · d(F ◦ ~a) = ∇g ~ · d(F1 ◦ ~a) ı̂ + d(F2 ◦ ~a) ̂ + d(F3 ◦ ~a) k̂
dt dt dt dt dt
(37)
 d~a d~ a d~
a 
= ∇g~ · ∇F ~ 1· ~ 2·
ı̂ + ∇F ~ 3·
̂ + ∇F k̂
 dt dt dt




~ · ı̂ · (− sin tı̂ + cos t̂) ı̂ + − k̂ · (− sin tı̂ + cos t̂) ̂ + ̂ · (− sin tı̂ + cos t̂) k̂
= ∇g


~ · (− sin tı̂ + cos tk̂) = 2F1 (~a)ı̂ − 2F2 (~a) · (− sin tı̂ + cos tk̂)
= ∇g
= (2a1 ı̂ + a3 ̂) · (− sin tı̂ + cos tk̂) = −2a1 (t) sin t = −2 sin2 t.

Solution of review exercises in Section 1.8

1. Compute gradient, Hessian and Laplacian of the five scalar fields.
We only need to compute several partial derivatives and combine them appropriately.
~
∇f = y 2 z 3 ı̂ + 2xyz 3 ̂ + 3xy 2 z 2 k̂ = M,
 
0 2yz 3 3y 2 z 2
Hf = 2yz 3
2xz 3 6xyz 2  , ∆f = 2xz(z 2 + 3y 2 ),
2 2 2 2
3y z 6xyz 6xy z


∇g = − sin x + cos(2y + x) ı̂ + 2 cos(2y + x)̂,
 
− cos x − sin(2y + x) −2 sin(2y + x) 0
Hg =  −2 sin(2y + x) −4 sin(2y + x) 0 , ∆g = − cos x − 5 sin(2y + x),
0 0 0
 
h −ex sin y 0
∇h = ex cos yı̂ − ex sin y̂, Hh = −ex sin y −h 0 , ∆h = 0,
0 0 0
A. Moiola, University of Reading 90 Vector calculus lecture notes, 2016–17

∇ℓ = (3x2 + y 2 + z 2 )ı̂ + (2xy − 3y 2 )̂ + 2xz k̂,

 
6x 2y 2z
Hℓ = 2y −6y 0  , ∆ℓ = 6(x − y),
2z 0 2x
∇m = yxy−1 ı̂ + xy log x̂,
 
y(y − 1)xy−2 xy−1 (1 + y log x) 0
Hm = x y−1
(1 + y log x) xy log2 x 0 , ∆m = y(y − 1)xy−2 + xy log2 x.
0 0 0

2. Compute Jacobian, divergence, curl and vector Laplacian of the five vector fields.
~ = ~r × ı̂ = z̂ − y k̂ ,
We use H ~ = ∇f
M ~ , ~ = 1 ∇(|~
L ~ r4 |) (see also Exercise 1.44).
4
 
0 1 1
~ = 1 0 1 ,
JF ∇~ ·F
~ = 0, ~ ×F
∇ ~ = ~0, ~F
∆ ~ = ~0,
1 1 0
 2
0 0 −2ze−z
~ = 0 2y
JG 0 , ~ ·G
∇ ~ = 2y,
0 0 0
~ ×G
∇ ~ = −2ze−z2 ̂, ~G
∆ ~ = 2(2z 2 − 1)e−z2 ı̂ + 2̂,
 
0 0 0
~ = 0 0 1 ,
JH ∇~ ·H~ = 0, ~ ×H
∇ ~ = −2ı̂, ~H
∆ ~ = ~0,
0 −1 0
 2 
3x + y 2 + z 2 2xy 2xz
~ =
JL 2xy x2 + 3y 2 + z 2 2yz , ~ ·L
∇ ~ = 5|~r|2 ,
2xz 2yz x2 + y 2 + z 2
~ ×L
∇ ~ = ~0, ~L
∆ ~ = 10~r,
~ = Hf,
JM ~ ·M
∇ ~ = ∆f, ~ ×M
∇ ~ =∇ ~ × ∇f
~ = ~0,
∆~M
~ =∆~ ∇f
~ = ∇∆f
~ = (2z 3 + 6y 2 z)ı̂ + 12xyz̂ + (6xz 2 + 6xy 2 )k̂.

3. Which of the fields are solenoidal and which are irrotational? Which are harmonic?
The only scalar harmonic field is h.
~ is solenoidal and irrotational.
F
~ is neither solenoidal nor irrotational.
G
~
H is solenoidal only.
~ is irrotational only.
L
M~ is irrotational only.
(Moreover F ~ and H~ are harmonic in the sense that all their components are harmonic.)

~ a vector potential for H

4. Compute a scalar and a vector potential for F, ~ (can (31) help you?) and a
~ Can you guess a scalar potential for L?
scalar potential for M. ~
Proceeding as in Example 1.67, we obtain the following potentials:
 
~ = ∇(xy
F ~ + xz + yz + λ) = ∇~ × 1 (z 2 − y 2 )ı̂ + (x2 − z 2 )̂ + (y 2 − x2 )k̂ ,
2
 2   
~ =∇
H ~ × |~r| ı̂ , ~ =∇
L ~ 1 |~r|4 + λ , M~ = ∇(f~ + λ) for any scalar λ.
2 4

5. Show that G~ does not admit neither scalar nor vector potential.
~ is neither irrotational nor solenoidal, therefore the existence of a scalar or a vector potential would
G
entail a contradiction with the box in Section 1.5 (conservative ⇒ irrotational, vector potential ⇒
solenoidal) or with identities (24) and (25).
~ r) and L(~
6. Show that H(~ ~ r) are orthogonal to each other at every point ~r ∈ R3 .

~ r) · L(~
H(~ ~ r) = (z̂ − y k̂) · |~r|2 ~r = |~r|2 (0x + zy − yz) = 0.
A. Moiola, University of Reading 91 Vector calculus lecture notes, 2016–17

~ (z = 0)
F ~ (y = 0)
G
z
1.5

0.5 1

0.5
z0 0

−0.5
−0.5
2
2 −1
0
y 0
x −1.5 x
−2 −2 −1 0 1 2

~ (z = 0)
H ~ (z = 0)
L
y y
1.5 1.5

1 1

0.5 0.5

0 0

−0.5 −0.5

−1 −1

−1.5 x −1.5 x
−2 −1 0 1 2 −2 −1 0 1 2

Figure 45: Representation of some sections of the vector fields

7. Try to graphically represent the fields in (39). E.g. you can draw a qualitative plot like those of
~ on the plane y = 0, for F,
Section 1.2.2 for G ~ H~ and L~ on the plane z = 0.
The suggested sections are plotted in Figure 45. Note that only for G,~ H ~ and L
~ the plots are in the
same form as those seen in the notes, since in these cases the fields are tangential to the considered
~ is not tangent to the plane {z = 0}, thus it can not be represented as a
planes. On the contrary, F
two-dimensional field. You might prove that F ~ is tangential to the plane {x + y + z = 0}.

~ (26) for f ; (28) for f and h; (33) for G

8. Demonstrate identities (25) and (27) for G; ~ and H;
~ (29)
~
and (31) for h and H. (You can also demonstrate other identities of Propositions 1.52 and 1.55 for
the various fields in (39).)
We demonstrate the suggested identities. In some cases, the best strategy is to first expand the
terms on both sides of the differential identities and then manipulate them, aiming at a common
expression.

(25) ∇~ · (∇
~ × G) ~ · (−2ze−z2 ̂) = ∂ (−2ze−z2 )0,
~ =∇
∂y
2
(27) ∇~ × (∇~ × G)
~ =∇
~ × (−2ze ̂)
−z

2
= −2(2z 2 − 1)e−z ı̂
 2

= 2̂ − 2̂ + 2(2z 2 − 1)e−z ı̂
~
= ∇(2y) −∆~G
~
~ ∇
= ∇( ~ · G)
~ −∆~ G,
~
~ × (∇f
(26) ∇ ~ )=∇
~ × (y 2 z 3 ı̂ + 2xyz 3 ̂ + 3xy 2 z 2 k̂)
= (6xyz 2 − 6xyz 2 )ı̂ + (3y 2 z 2 − 3y 2 z 2 )̂ + (2yz 3 − 2yz 3 )k̂ = ~0,
~ h) = ∇(xy
(28) ∇(f ~ 2 3 x
z e cos y)
= (1 + x)ex y 2 z 3 cos yı̂ + xex (2y cos y − y 2 sin y)z 3 ̂ + 3xex y 2 cos yz 2 k̂
A. Moiola, University of Reading 92 Vector calculus lecture notes, 2016–17

= xy 2 z 3 (ex cos yı̂ − ex sin y̂) + ex cos y(y 2 z 3 ı̂ + 2xyz 3 ̂ + 3xy 2 z 2 k̂)
~ + h∇f,
= f ∇h ~
2

~ G
(33) ∇( ~ · H)
~ =∇ ~ (e−z ı̂ + y 2 ̂) · (z̂ − y k̂)
~ 2 z)
= ∇(y
= 2yz̂ + y 2 k̂
2 2
= −y 2~k + (2yz̂ + 2yze−z ı̂) − 2yze−z ı̂ + 2y 2 k̂
 2 ∂ ∂   ∂ ∂  −z2
= e−z + y2 (z̂ − y k̂) + z −y (e ı̂ + y 2 ̂)
∂x ∂y ∂y ∂z
2 2
+ (z̂ − y k̂) × (−2ze−z ̂) + (e−z ı̂ + y 2 ̂) × (−2ı̂)
~ · ∇)
= (G ~ H ~ + (H ~ · ∇)
~ G~ +H~ × (∇ ~ × G)
~ +G ~ × (∇ ~ × H),
~


~ · (hH)
(29) ∇ ~ =∇ ~ · (ex cos y)z̂ − (ex cos y)y k̂
= 0 − ex sin yz + 0
= (ex cos yı̂ − ex sin y̂) · (z̂ − y k̂) + 0
~ ·H
= (∇h) ~ + h∇ ~ · H,
~

(31) ∇ ~ =∇
~ × (hH) ~ × (ex cos y)z̂ − (ex cos y)y k̂


= − ex (cos y − y sin y) − ex cos y ı̂ + ex cos yy̂ + ex cos yz k̂
= yex sin yı̂ + yex cos y̂ + zex cos y k̂ − 2ex cos yı̂
= (ex cos yı̂ − ex sin y̂) × (z̂ − y k̂) − 2ex cos yı̂
~ ×H
= (∇h) ~ + h∇ ~ × H.~

a d~
d~ b d~c
9. Compute the (total) derivatives of the curves (i.e. dt , dt and dt ) and try to draw them.

d~a d(t3 − t)ı̂ + (1 − t2 )k̂

= = (3t2 − 1)ı̂ − 2tk̂,
dt dt
d~b d(t3 ı̂ + t2 ̂ + k̂)
= = 3t2 ı̂ + 2t̂,
dt dt

d~c d et cos(2πt)ı̂ + et sin(2πt)̂
=
dt  dt
t



=e cos(2πt) − 2π sin(2πt) ı̂ + sin(2πt) + 2π cos(2πt) ̂
 
= et (~i + 2π̂) cos(2πt) + (̂ − 2πı̂) sin(2πt) .

The curves and their derivatives are shown in Figure 46. From the plots of the curves we note that
~a is a loop, ~b is singular, ~c is a (part of a) logarithmic spiral. The total derivatives of ~a and ~b
(which are also curves) have almost the same plot, they can be transformed into each other by a
rigid motion. Despite this fact, ~a and ~b have quite different shapes.
a) df (~
dh(~ b) dℓ(~c)
10. Compute the following total derivatives of the scalar fields along the curves: dt , dt and dt .
(You can either use the chain rule (37) or first compute the composition.)
Deriving the composition of the field with the curve:

3
d h(~a) d(et −t ) 3
= = et −t (3t2 − 1) (note that ~a lies in plane y = 0),
dt dt


d f (~b) d(t7 )
= = 7t6 ,
dt
dt

d ℓ(~c) d e3t cos(2πt) − e3t sin3 (2πt)
=
dt  dt 
= e3t 3 cos(2πt) − 3 sin3 (2πt) − 2π sin(2πt) − 6π sin2 (2πt) cos(2πt) .

Alternatively, using the vector chain rule (37):

 
d h(~a) ~ a) · d~a = (ex cos yı̂ − ex sin y̂) · (3t2 − 1)ı̂ − 2tk̂ = et3 −t (3t2 − 1),
= ∇h(~
dt dt
A. Moiola, University of Reading 93 Vector calculus lecture notes, 2016–17

z ~a(t) y ~b(t) ~c(t)

1 1 2
y

0.8 0.8 1

0.6 0.6 0

0.4 0.4 −1

0.2 0.2 −2

0 x 0 x −3 x
−0.5 0 0.5 −1 0 1 −2 0 2 4

z d~a(t)/dt y d~b(t)/dt y d~c(t)/dt

2 2 20

1 1 10

0 0 0

−1 −1 −10

−2 x −2 x −20 x
−1 0 1 2 0 1 2 3 −10 0 10 20

Figure 46: The images of the three curves and of their total derivatives.

d f (~b) ~ (~b) · db
~
= ∇f
dt  dt 
= y z ı̂ + 2xyz 3 ̂ + 3xy 2 z 2 k̂ · (3t2 ı̂ + 2t̂)
2 3

 
= t4 ı̂ + 2t5 ̂ + 3t7 k̂ · (3t2 ı̂ + 2t̂)
= 7t6 ,


d ℓ(~c) ~ c) · d~c
= ∇ℓ(~
dt  dt 
= (3x + y 2 + z 2 )ı̂ + (2xy − 3y 2 )̂ + 2xz k̂
2




· et cos(2πt) − 2π sin(2πt) ı̂ + sin(2πt) + 2π cos(2πt) ̂



= e2t 3 cos2 (2πt) + sin2 (2πt) ı̂ + 2 cos(2πt) sin(2πt) − 3 sin2 (2πt) ̂



· et cos(2πt) − 2π sin(2πt) ı̂ + sin(2πt) + 2π cos(2πt) ̂
 
= e3t 3 cos(2πt) − 3 sin3 (2πt) − 2π sin(2πt) − 6π sin2 (2πt) cos(2πt) .

B.2 Exercises of Section 2

Solution of Exercise 2.4. We start by computing the total derivative of the curve and its magnitude:
 
d~aA  d~aA  p √
= − sin tı̂ + cos tı̂ + k̂,   = (− sin t)2 + (cos t)2 + 12 = 2.
dt  dt 

The value of the field fA (~r) = x + y + z along the curve is

fA ~aA (t) = cos t + sin t + t.

To conclude, we simply apply formula (36) for the line integral of a scalar field:
Z Z 2π   Z 2π 

 d~aA 
 
√ (2π)2 √ √ 2
fA ds = fA ~aA (t)  dt = (cos t + sin t + t) 2 dt = 0 + 0 + 2 = 2 2π .
ΓA 0 dt  0 2
Similarly, for the two remaining integrals we have:
  Z Z 1p Z 1
 d~aB  p 2
  = 1 + t4 , f B ds = (1 + t 4 )(t6 + t2 ) dt = (t5 + t) dt = ,
 dt  3
ΓB 0 0
A. Moiola, University of Reading 94 Vector calculus lecture notes, 2016–17
  Z Z r Z
 d~aC  p p 10
t2 p 10
  2 2 2 1 + t2 dt =
 dt  = 1 + t = 1 + x + y , fC ds =
1 + t2
t dt = 50.
ΓC 0 0

Solution of Exercise 2.7. (1) We compute the length of the logarithmic spiral using the formula (43):
d~a
= e−t (− sin t − cos t)ı̂ + e−t (cos t − sin t)̂,
 dt
 d~a  p √
  = e−t sin2 t + cos2 t + 2 sin t cos t + sin2 t + cos2 t − 2 sin t cos t = 2e−t ,
 dt 
Z Z ∞  Z ∞√ √ −t ∞ √
(43)  d~a 
length(Γa ) = ds =   dt = 2e −t
dt = − 2e  = 2.
 dt  0
Γa 0 0

(2) To compute the length of the second spiral we can proceed directly:

d~b −τ sin τ − cos τ τ cos τ − sin τ

= 2
ı̂ + ̂,
 dt τ τ2
 d~b  √
  1p 2 2 2 2 2 2 τ2 + 1
  = 2 τ sin τ + cos τ + 2τ sin τ cos τ + τ cos τ + sin τ − 2τ sin τ cos τ = ,
 dt  τ τ2
Z ∞  ~  Z ∞√ 2  √
(43)  db  τ +1 τ 2 + 1 ∞ √
length(Γb ) =   dτ = 2
dτ = arsinhτ −  = ∞ − 1 − arsinh1 + 2 = ∞.
1  dτ  1 τ τ 1

√
2
However the integral of ττ 2+1 , which we computed by parts, is not easy. A simpler way is to note that
√
2
the integrand satisfies the lower bound ττ 2+1 ≥ ττ2 = τ1 , thus
Z ∞√ 2 Z ∞ ∞
τ +1 1 
length(Γb ) = 2
dτ ≥ dτ = log τ  = ∞.
1 τ 1 τ 1

Thus the length of the path cannot be shorter than ∞.

(3) Both curves are in the form F (t)(cos tı̂ + sin t̂) for some positive function F , so they rotate around
the origin in anticlockwise direction with period 2π. Since both curves are defined for an unbounded
parametrisation interval (i.e. t ≥ 0 and τ ≥ 1), they rotate infinitely many times: for all values t, τ = 2nπ
with n ∈ N we have a2 (t) = b2 (τ ) = 0 and a1 (t), b1 (τ ) > 0 so the curves cross the positive x axis; for all
values t, τ = (2n + 1)π with n ∈ N we have a2 (t) = b2 (τ ) = 0 and a1 (t), b1 (τ ) < 0 so the curves cross the
negative x axis. This happens for all n ∈ N, so infinitely many times.
(4) We first note that ℓ must be at least 1, since no curve connecting ı̂ and ~0 can be shorter than the
straight segment, which has length 1. The simplest way to construct ~c is to slightly modify the curve ~a.
To guarantee infinitely many turns around the origin we keep the cos t and sin t terms. To modify the
speed of convergence to the origin, thus the length of the path, we modify the exponential term as e−λt ,
using a positive parameter λ. So we define a new curve ~c, depending on λ, and we compute its length:
~c(t) = e−λt cos tı̂ + e−λt sin t̂, t ∈ [0, ∞),
d~c
= e−λt (− sin t − λ cos t)ı̂ + e−λt (cos t − λ sin t)̂,
 dt p p
 d~c 
  = e−λt sin2 t + λ2 cos2 t + 2λ sin t cos t + λ2 sin2 t + cos2 t − 2λ sin t cos t = 1 + λ2 e−λt ,
 dt 
Z ∞  Z ∞p √ √
(43)  d~c 
  dt = 2 e−λt dt = −
1 + λ2 −λt ∞ 1 + λ2 p
length(Γc ) =  dt  1 + λ e  = = 1 + 1/λ2 .
0 0 λ 0 λ
p √
Thus to obtain a length ℓ we need to take 1 + √
1/λ2 = ℓ, which gives
√
λ = 1/ ℓ2 − 1, which is acceptable
2 2
for any ℓ > 1. The desired curve is ~c(t) = e−t/ ℓ −1 cos t ı̂ + e−t/ ℓ −1 sin t ̂ for t ∈ [0, ∞).
Solution of Exercise 2.11. Proceeding as in Example 2.10 we obtain
Z Z Z Z Z
~
G · d~aA = ~
G · d~aB = − ~
G · d~aC = ~
G · d~aD = ~ · d~aE = 5 .
G
ΓA ΓB ΓC ΓD ΓE 6

Solution of Exercise 2.12. We first note that the position Rvector field evaluated
R on any curve equals
~ a(t)) = ~a(t), thus
the parametrisation of the curve itself: F(~ ~ · d~r =
F ~
a · d~
a
Γa Γa dt ds. We report the
parametrisation of the curves and their total derivatives, compute their scalar products and integrate:

~a(t) = e−t cos t ı̂ + e−t sin t ̂, t ∈ [0, ∞), ~b(τ ) = 1 cos τ ı̂ + 1 sin τ ̂, τ ∈ [1, ∞),
τ τ
A. Moiola, University of Reading 95 Vector calculus lecture notes, 2016–17

d~a d~b −τ sin τ − cos τ τ cos τ − sin τ

= e−t (− sin t − cos t)ı̂ + e−t (cos t − sin t)̂, = ı̂ + ̂,
dt dt τ2 τ2
d~a
~a · = e−2t (− sin t cos t − cos2 t + cos t sin t − sin t sin t) = −e−2t ,
dt
~
~b · db = 1 (−τ sin τ cos τ − cos2 τ + τ cos τ sin τ − sin2 τ ) = − 1 ,
dt τ3 τ3
Z Z Z ∞ 
−2t ∞
~ · d~r = d~a e  1
F ~a · ds = −e−2t dt =  = −2 ,
Γa Γa dt 0 2 0
Z Z Z ∞ ∞
d ~b 1 1  1
F~ · d~r = ~
b· ds = − 3 dτ = 2  = − .
Γb Γa dτ 1 τ 2τ 2 1

Here we see a fact that might be surprising: the second spiral Γb has infinite length, as we have verified
in Exercise (2.7), but the line integral of ~r along it is finite. This is because, when we approach the origin
proceeding along Γb , the field ~r becomes smaller and “more orthogonal” to the curve itself.
We could have guessed from the beginning the signs of the integrals. The position vector points in the
radial direction away from the origin. The total derivatives of each curve instead point roughly towards
the origin, as the magnitude of the curve points decreases monotonically with t. Thus the scalar products
~ ~b) · d~b are negative, and the integrals of negative integrands are negative.
~ a) · d~a and F(
F(~ dt dτ

Solution of Exercise 2.17. We know the scalar potentials of the two vector fields considered: ~r =
∇( ~
~ 1 |~r|2 ) and zı̂ + xk̂ = ∇(xz). Applying the fundamental theorem of vector calculus (49) we have:
2
Z Z
1 2 1 2 1
~r · d~r = |~q| − |~p| = (3 − 5) = −1, and (zı̂ + xk̂) · d~r = (−1)(−1) − (0)(2) = 1.
Γ 2 2 2 Γ

Solution of Exercise 2.20. We first need to find a parametrisation of the two paths41 :

~a(t) = (1 − t)ı̂ + t̂, 0 ≤ t ≤ 1, ~b(τ ) = cos τ ı̂ + sin τ ̂, 0 ≤ τ ≤ π/2.

The corresponding total derivatives are

d~a(t) d~b(τ )
= −ı̂ + ̂, = − sin τ ı̂ + cos τ ̂.
dt dτ
The line integrals are computed using the definition (44):
Z Z 1 Z 1



~ · d~r =
F ~ a1 (t), a2 (t) · d~a(t) dt =
F a21 (t) + a22 (t) (ı̂ + 2̂) · (−ı̂ + ̂) dt
Γa 0 dt 0
Z 1

1 1 2
= 1 + t2 − 2t + t2 (−1 + 2) dt = 1 + − 1 + = ,
0 3 3 3
Z Z π/2 Z

~ π/2
~ · d~r =
F ~ b1 (τ ), b2 (τ ) · db(τ ) dτ =
F (cos2 τ + sin2 τ )(ı̂ + 2̂) · (− sin τ ı̂ + cos τ ̂) dτ
Γb 0 dτ 0
Z π/2
= (− sin τ + 2 cos τ ) dτ
0
π π
= cos − cos 0 + 2 sin − 2 sin 0 = 0 − 1 + 2 − 0 = 1.
2 2
~ over two paths connecting the same endpoints are different from
Since the values of the line integrals of F
~ is not conservative.
one another ( 23 6= 1), by Theorem 2.19 (or by the equivalence in box (50)) the field F
Solution of Exercise 2.23.
ZZ Z π  Z 2   π  3y 2 
 e  2
3y
e sin x dx dy = sin x dx e 3y
dy = − cos x  = (e6 − e3 ),
R 0 1 0 3 1 3
41 How did we find these parametrisations? Many choices are possible, we look for the simplest one; see also Remark 1.24.
We first look for the first component a1 (t) of ~a(t). If we choose to parametrise Γa with the unit interval [0, 1], a1 must
be a scalar function such that a1 (0) = 1 and a1 (1) = 0: the easiest possible choice is a1 (t) = 1 − t. Since by definition of Γa
we have a1 (t) + a2 (t) = 1, we have 1 − t + a2 (t) = 1 and thus a2 (t) = t.
For the second path, we can proceed similarly (see Example 2.3), but we would obtain a complicated parametrisation.
Instead, we note that we need two scalar functions b1 and b2 such that b21 (τ ) + b22 (τ ) = 1, so the most natural choice is
b1 (τ ) = cos τ and b2 (τ ) = sin τ . For this choice, we need to fix the parametrisation endpoints to be τI = 0 and τF = π/2,
so that ~b(τI ) = ı̂ and ~
b(τF ) = ̂.
A. Moiola, University of Reading 96 Vector calculus lecture notes, 2016–17
ZZ Z 5 Z 1  Z 5  
1 x2 5 125 5
y dx dy = y dy dx = − dx = − = ,
Q 0 x
5 0 2 50 2 150 3
ZZ Z π Z sin x  Z π2 Z π2 π
2 1 − cos(2x)  2 1
cos x dx dy = cos x dy dx = cos x sin x dx = sin(2x) dx =  = 2.
S 0 0 0 0 2 4 0

Solution of Exercise 2.29. Proceeding as in Example 2.28, we have to compute the Jacobian deter-
minant ∂(ξ∂(x,y)
LR ,ηLR )
and integrate it over the unit square S = {0 < x < 1, 0 < y < 1}:
 xy xy xy

~ LR = ye xy cos(2πy) xexy cos(2πy) − 2πexy sin(2πy) , ∂(ξLR , ηLR )
JT = 2πye2xy ,
ye sin(2πy) xe sin(2πy) + 2πe cos(2πy) ∂(x, y)
Z 1Z 1 Z 1 1 Z 1  e2y 1

2xy  π(e2 − 3)
Area(RLR ) = 2π 2xy
ye dx dy = π e  dy = π 2y
(e −1) dy = π −y  = .
0 0 0 x=0 0 2 y=0 2

Solution of Exercise 2.30. (i) The domain is defined implicitly by the change of coordinates Q ∋
xı̂ + y̂ 7→ ξ ξ̂ + η η̂ ∈ (0, 1)2 . We compute the Jacobian determinant and the integral:
 
∂(x, y) 4−η −ξ
= det = (4 − η)(π − ξ) − ξη = 4π − 4ξ − πη;
∂(ξ, η) −η π−ξ
ZZ ZZ   Z 1Z 1
 ∂(x, y)  4 π 7
Area(Q) = 1 dA =   dξ dη = |4π − 4ξ − πη| dξ dη = 4π − − = π − 2.
 
Q (0,1)2 ∂(ξ, η) 0 0 2 2 2
You can visualise this domain in Matlab/Octave using the function VCplotter.m (available on the course
web page), with the command:
VCplotter(6, @(x,y) x*(4-y), @(x,y) y*(pi-x),0,1,0,1);
(ii) We repeat the same computations done in part (i) with the general parameters:
 
∂(x, y) a − η −ξ
= det = (a − η)(b − ξ) − ξη = ab − aξ − bη;
∂(ξ, η) −η b−ξ
ZZ ZZ   Z 1Z 1
 ∂(x, y)  a+b
Area(Qab ) = 1 dA =   dξ dη = |ab − aξ − bη| dξ dη = ab − .
 
Qab (0,1)2 ∂(ξ, η) 0 0 2

Alternative solution: Qab is the quadrilateral with vertices ~0, aı̂, (a − 1)ı̂ + (b − 1)̂, b̂ and straight edges,
thus computing its area by decomposing it in the rectangle ~0, (a − 1)ı̂, (a − 1)ı̂ + (b − 1)̂, (b − 1)̂, the
triangle (a − 1)ı̂, aı̂, (a − 1)ı̂ + (b − 1)̂ and the triangle (b − 1)̂, b̂, (a − 1)ı̂ + (b − 1)̂ (so that the area is
(a − 1)(b − 1) + 21 (a − 1 + b − 1) = ab − 21 (a + b)) is also correct.
Solution of Exercise 2.31. We first compute the inverse change of variables, solving (60) for x and y:
taking the sum and the difference of the two equations we have
( (
ξ + η = 2y − 1, x = 2(ξ − η + 1),
⇒ (108)
1
ξ − η = 2 x − 1, y = ξ+η+1
2 .
The Jacobian determinant we will need in the integral is
∂(ξ, η) (56) ∂ξ ∂η ∂ξ ∂η 1  1 1
= − = 1−1 − = .
∂(x, y) ∂x ∂y ∂y ∂x 4 4 2
~ the change of variables and R the counterimage of P , i.e.:
We call T
~ : R → P,
T ~
T(xı̂ + y̂) = ξ(x, y)ξ̂ + η(x, y)η̂.
The domain P is a square, so we want to identify the four parts of the boundary of R corresponding
to the four edges42 of P . For this purpose, it is enough to insert the equations of the edges of P either in
(60) or in (108):
line connecting ξ̂, η̂ : ξ + η = 1, y = 1,
line connecting η̂, −ξ̂ : ξ − η = −1, x = 0,
ˆ −η̂ :
line connecting − ξ, ξ + η = −1, y = 0,
line connecting − η̂, ξ̂ : ξ − η = 1, x = 4,
42 If you don’t remember how to compute the equations of the straight lines defining the edges of P , draw the domain and
η−η ξ−ξ
use four times the formula for the line through two points ~ q = ξq ξ̂ + ηq η̂: η −ηp = ξ −ξp . You have to
p = ξp ξ̂ + ηp η̂ and ~
q p q p
repeat this for each of the four edges. In this exercise all the coefficients are either 1, 0 or −1. See also Figure 47.
A. Moiola, University of Reading 97 Vector calculus lecture notes, 2016–17

so all the four lines are parallel to the axes in the xy-plane and the domain R is the rectangle R =
(0, 4) × (0, 1) = {xı̂ + y̂ ∈ R2 , 0 < x < 4, 0 < y < 1}. Now we can compute the desired integral using the
change of variables formula (57):
ZZ ZZ   
(57) x x  ∂(ξ, η) 
f (ξ, η) dξ dη = f + y − 1, y − dx dy
P 4 4  ∂(x, y) 
ZZ R
x x 1
= e− 4 −y+1−y+ 4 dx dy
R 2
Z 1Z 4  Z 1 1
−2y+1 1 −2y+1 e−2y+1 
= e dx dy = 2e dy = 2 = e − e−1 ≈ 2.35.
0 0 2 0 −2  0

−ξ + η = 1 ξ+η =1
y y=1 P

ξ
x=0 R x=4
−ξ − η = 1 ξ−η =1
y=0 x

Figure 47: The rectangle R and the square P in Exercise 2.31.

Solution of Exercise 2.36. (i) We use the following change of variables (we write also the inverse):43
( !   (
∂ξ ∂ξ
ξ = xy, ∂x ∂y y x ∂(ξ, η) x = ηξ , ∂(x, y) 1
(I) ∂η ∂η = , = y, = .
η = y, ∂x ∂y
0 1 ∂(x, y) y = η, ∂(ξ, η) η

(ii) The integral is computed as:

ZZ ZZ   ZZ Z 1 Z 1
ξ (57) ξ(x, y)  ∂(ξ, η)  xy 11 1
dξ dη = dx dy = y dx dy = x dx y dy = = .
T η S η(x, y)  ∂(x, y)  S y 0 0 2 2 4
43 How did we find this transformation? First we decide how to change the domain geometrically: since we are transforming
a square into a triangle, have to “get rid of” one of the sides of S, we choose to maintain the upper and the left side, collapse
the lower one and transform the right one into the diagonal (see below for other options). We then note that we can
maintain the vertical coordinate of every point and modify only the horizontal one, so we fix η = y. We want to modify x
by multiplying with a suitable function, so ξ = xg(x, y). The function g must have value 1 on the horizontal line y = 1 (so
the upper side is preserved), and value 0 on the horizontal line y = 0 (so the lower side is collapsed). The choice g(x, y) = y
is the easiest that satisfies these two conditions (g(x, 1) = 1, g(x, 0) = 0), so the candidate transformation is ξ = xy, η = y.
Then, we can immediately verify that, under this transformation, the upper side of S is preserved (y = 1 ⇒ η = 1), the
left side is preserved (x = 0 ⇒ ξ = 0), the right side is mapped into the diagonal side of T (x = 1 ⇒ ξ = η), the lower side
is collapsed to the origin (y = 0 ⇒ ξ = η = 0). Thus the transformation maps correctly the boundaries: ∂S → ∂T . It is
surjective (for all ξ ξ̂ + ηη̂ ∈ T , ξ/ηı̂ + η̂ belongs to S), and it is injective in the interior of S:

~ A , yA ) = xA yA ξ̂ + yA η̂ = xB yB ξ̂ + yB η̂ = T(x
~ B , yB ) x A yA = x B yB , yA ,yB >0 xA = xB ,
T(x ⇒ ⇒
yA = yB , yA = yB .
(The verification of the mapping of the boundary and the bijectivity was not required in the exercise, however it is what you
should do to check if your transformation is correct.)
Another smart way to obtain and verify the transformation, is to rewrite the domains as
T = {0 < ξ < η < 1} = {0 < ξ/η < 1, 0 < η < 1}, S = {0 < x < 1, 0 < y < 1},

from which the inverse transformation x = ηξ , y = η is immediate (this corresponds to the setting of Example 2.34).
Many other changes of variables are possible, some simple ones are
(

 ξ = 12 x,
( 
 for x + y ≤ 1,
ξ = x,  η = y + 21 x,
(II) (III) (
η = x + y − xy, 
 ξ = x + 12 (y − 1),

 for x + y > 1.

η = 21 (y + 1),

The transformation (II) corresponds to that described in Example 2.33 for general y-simple domain; (III) is “piecewise
affine” (and continuous). See a representation of the three transformations in Figure 48; you can play around with these
transformations using VCplotter.m. Note that the transformation of the unit cube into a tetrahedron in equation (64) is
very similar to this exercise.
A. Moiola, University of Reading 98 Vector calculus lecture notes, 2016–17

We can double check the computation above using the iterated integral on T :
ZZ Z 1Z η Z 1 2
ξ ξ η 1
dξ dη = dξ dη = dη = .
T η 0 0 η 0 2η 4

y η η η

(I) (II) (III)

x ξ ξ ξ

Figure 48: The square S and its image T under the three transformation described in exercise (1); many
other options are possible. In each case, the point ~0, ̂ and ı̂ + ̂ are mapped into ~0, η̂ and ξˆ + η̂,
respectively, while ı̂ is mapped into the point denoted with the little circle. The continuous, dashed and
grey lines represent the images of the vertical, horizontal and diagonal segments in the first plot. Note that
the first two transformations (which are polynomials of degree two) map some straight lines into parabolas,
while the third one (which is piecewise affine) maps straight lines into polygonal lines.

Solution of Exercise 2.42. We only need to show that for all u, w ∈ R we have ~
X(u, w) ∈ S, i.e.
X12 (u, w) + X22 (u, w) + X32 (u, w) = 1 and X3 (u, w) 6= 1. We compute the magnitude of ~
X(u, w):
2 2
2 u4 w4 u2 w2 u2 w 2
~ u2 + w2 + u4 + w4 − 1 u2 + w 2 + + +1− − 2 +
|X(u, w)|2 = 2 2
2
= u4
16
w4
16
u2
2
w2 u2 w 2
8
= 1,
1 + u4 + w4 1+ 16 + 16 + 2 + 2 + 8

2 2 2 2
~
so X(u, w) belongs to the unit sphere. Since X3 (u, w) = 1 reads u4 + w4 − 1 = 1 + u4 + w4 , which has
~ maps into S.
no solutions, no point is mapped to the north pole and we conclude that X
~ ~
To compute the inverse of X, we have to solve X(u, w) = ~r for u and w. Using the relation x2 + y 2 +
2
z = 1 between the components of a point in S, after some computations we find u = 2x/(1 − z) and
w = 2y/(1 − z).
Solution of Exercise 2.43. The surface is a torus, i.e. a “mathematical doughnut”. It can be described
as the set of points with distance b from a circle of radius a centred at the origin and lying in the xy-plane.
You can see it in the left plot of Figure 49 for a = 3 and b = 1. To be more precise, S is the torus after
the inner circle (corresponding to w = ±π) and the circle in the {y = 0, x < 0} half plane (corresponding
to u = ±π) have been deleted. The two deleted circles are highlighted in the figure.

0.8

0.6

0.4

0.2
0.5
0
-0.5 0

1
3
2 0.8
1 0.6 1
2
0 0.4
-1 0 0.5
0.2
-2
-2 0
-3
0

Figure 49: Left: The torus described by the parametrisation in Exercise 2.43 with a = 3 and b = 1.
Right: The “hump” graph surface of Exercise 2.45. The little coloured arrows represent the (scaled)
tangent vectors. The black arrow is the (scaled) unit normal vector at 1/2ı̂ + 1/4̂ + 3/4k̂.
A. Moiola, University of Reading 99 Vector calculus lecture notes, 2016–17

Solution of Exercise 2.45. We have 0 ≤ g ≤ 1 in R with g(1/2, 1/2) = 1 and g = 0 on the boundary
of R, i.e. where either x or y are equal to 0 or 1. So its graph Sg looks like a “hump”, see the right plot
of Figure 49.
~ = 16(1 − 2x)y(1 − y)ı̂ + 16x(1 − x)(1 − 2y)̂. We can see Sg as a parametric
The gradient of g is ∇g
~g
curve with chart X~ g = xı̂ + y̂ + g(x, y)k̂ defined on R, so ∂ X = ı̂ + ∂g k̂ = ı̂ + 16(1 − 2x)y(1 − y)k̂ and
∂x ∂x
~g
∂X ∂g
∂y = ̂ + ∂y k̂ = ı̂ + 16x(1 − x)(1 − 2y)k̂. These are tangent vector fields to Sg .
Solution of Exercise 2.49.
 
~ ~ ~ ~  ∂X ~  √
∂X ∂X ∂X ∂X  ~ ∂X
= ı̂ + ̂ + k̂, = ı̂ − ̂, × = −ı̂ + ̂ − 2k̂,  ×  = 6,
∂u ∂w ∂u ∂w  ∂u ∂w 
ZZ ZZ   Z 1Z 1√
 ∂X ~ 
 ~ ∂X √
Area(S) = dS =  ×  dA = 6 du dw = 6.
S R  ∂u ∂w  0 0

Since the chart is affine, S is the flat parallelogram with vertices X(0, ~ ~
0) = ~0, X(1, 0) = ı̂ + ̂ + k̂,
~ ~ ~
X(1, 1) = 2ı̂+ k̂, X(0, 1) = ı̂−̂. Moreover, since two sides of this are perpendicular (i.e. (X(1, ~
0)− X(0, 0))·
√
~
(X(0, ~
1)− X(0, 0)) = 0), this is a rectangle. The sides of this rectangle have length | ~
X(1, 0)− ~
X(0, 0)| = 3
√ √
~ ~
and |X(0, 1) − X(0, 0)| = 2, thus we have verified geometrically that the area of S is 6.
Solution of Exercise 2.50. We simply integrate the field f = 1 over S using formula (69). Here the
region R is the unit square (0, 1) × (0, 1).
s
ZZ ZZ  ∂( 2 x 32 + 2 y 23 ) 2  ∂( 2 x 32 + 2 y 23 ) 2
3 3 3 3
Area(S) = 1 dS = 1+ + dA
S (0,1)×(0,1) ∂x ∂y
Z 1Z 1p
= 1 + x + y dy dx
0 0
Z 1 1
2 3
= (1 + x + y) 2  dx
0 3 y=0
Z
2 1 3 3

= (2 + x) 2 − (1 + x) 2 dx
3 0
2 2 5 5
1
= (2 + x) − (1 + x) 
2 2
35 x=0
4 5 5 5
 4 √ √ 
= 32 − 22 − 22 + 1 = 243 − 2 32 + 1 ≈ 1.407.
15 15

Solution of Exercise 2.51. (i) This surface is not a graph, so we cannot use formula (68) but we need
to use the more general (67). We compute the area element, by taking the partial derivatives of the
~ using the definition of the vector product ×, the definition of vector magnitude | · |, and the
chart X,
trigonometric identity sin2 u + cos2 = 1:
~
X(u, w) = (a + b cos w) cos uı̂ + (a + b cos w) sin u̂ + b sin wk̂ − π < u < π, −π < w < π,
∂X~
= −(a + b cos w) sin uı̂ + (a + b cos w) cos u̂,
∂u
∂X~
= −b sin w cos uı̂ − b sin w sin u̂ + b cos wk̂,
∂w
~
∂X ∂X~
× = b(a + b cos w) cos u cos wı̂ + b(a + b cos w) sin u cos w̂ + b(a + b cos w) sin wk̂,
 ∂u ∂w
 ∂X ~ 
 ~ ∂X
 ×  = b|a + b cos w|(cos2 u cos2 w + sin2 u cosw + sin2 w)1/2 = b|a + b cos w|.
 ∂u ∂w 
 
 X ~ ~ 
Since 0 < b < a, we have a + b cos w > 0 for all values of w, so we can write  ∂∂u × ∂X
∂w  = b(a + b cos w).
The area of S is simply the surface integral of the constant field 1:
ZZ Z Z  ~ 
~ 
 ∂X ∂X
Area(S) = dS =  ×  dA
S R  ∂u ∂w 
A. Moiola, University of Reading 100 Vector calculus lecture notes, 2016–17
Z π Z π Z π Z π
= b(a + b cos w) du dw = du b(a + b cos w) dw = (2π)(2πab) = 4π 2 ab.
−π −π −π −π

Choosing a = 2 and b = 1 we have that the area is 8π 2 .

(ii) To compute the integral of the scalar field f (~r) = (x2 + y 2 )−1/2 we need to evaluate f on S and
write it in terms of u and w. We use the chart X:~


−1/2
~
f X(u, w) = (a + b cos w)2 cos2 u + (a + b cos w)2 sin2 u = |a + b cos w|−1 = (a + b cos w)−1 ,
ZZ ZZ   Z π Z π
 ∂X ~ 
(67)
~  ~ ∂X
f dS = f (X) ×  dA = (a + b cos w)−1 b(a + b cos w) du dw = 4π 2 b.
S R  ∂u ∂w  −π −π

Solution of Exercise 2.55. We use formula (74) together with the expression of F ~ and g(x, y) = x2 −y 2 :
ZZ ZZ  
~ ~ ∂(x2 − y 2 ) ∂(x2 − y 2 )
F · dS = − |{z}
0 − z(x, y) + (−y) dx dy
S (0,1)×(0,1) ∂x | {z } ∂y | {z }
=F1 =F2 =F3
Z 1 Z 1 Z 1  

2 2 1 1 2
= (y 2 − x2 )(−2y) − y dx dy = 3
− 2y + y − y dy = − + − = − .
0 0 0 3 4 3 2 3

Solution of Exercise 2.56. Recall that in Exercise 2.45 we computed ∇g ~ = 16(1 − 2x)y(1 − y)ı̂ +
16x(1 − x)(1 − 2y)̂. The fluxes are
ZZ ZZ   Z 1Z 1
~ ~ (74) ∂g
F · dS = − + 0 + 0 dA = 16(2x − 1)y(1 − y) dx dy
Sg R ∂x 0 0
Z 1 Z 1
1
= 16 (2x − 1) dx (y − y 2 ) dx dy = 16(0) = 0,
0 0 6
ZZ ZZ   Z 1Z 1
~ · d~S (74) ∂g
G = −x + 0 + 0 dA = 16x(2x − 1)y(1 − y) dx dy
Sg R ∂x 0 0
Z 1 Z 1
1 1 4
= 16 (2x2 − x) dx (y − y 2 ) dx dy = 16 = .
0 0 6 6 9
To guess the signs of the fluxes we observe the shape of Sg in the right plot of Figure 49. The flux
~ can be understood as the “amount of vector field” passing through the surface and is defined as the
of F
integral of the scalar product between F ~ and the unit normal vector n̂, which points upwards as in (71).
The first field F~ = ı̂ is constant (i.e. has the same value everywhere) and points in the x direction,
from the plot we see that n̂ has positive x component only for x > 1/2 so F ~ · n̂ is positive for x > 1/2
and negative for x < 1/2. Since Sg is symmetric with respect to the vertical plane {x = 1/2}, so it is n̂,
and the contributions to the flux from {~r ∈ Sg , x < 1/2} and {~r ∈ Sg , x > 1/2} have the same absolute
value and cancel each other. So the total flux is zero.
The second field G ~ = xı̂ points again in the x direction but is not constant. So the contribution to the
flux from {~r ∈ Sg , x < 1/2} is again negative (because here F ~ · n̂ < 0) and that from {~r ∈ Sg , x > 1/2} is
~
positive. But now F is larger in this second portion of Sg , so the positive term is larger in absolute value
than the negative one and the sum of the two terms is then positive.
What do you obtain if you make a similar reasoning for H ~ = yı̂?

Solution of Exercise 2.57. We recall that we have:

~ = (a + b cos w) cos uı̂ + (a + b cos w) sin u̂ + b sin wk̂
X − π < u < π, −π < w < π,
~
∂X ~
∂X
× = b(a + b cos w) cos u cos wı̂ + b(a + b cos w) sin u cos w̂ + b(a + b cos w) sin wk̂.
∂u ∂w
~
The position vector has value equal to its argument, so ~r(X(u, ~
w)) = X(u, w). The flux is
ZZ ZZ ~ ~ ZZ ~ ~
(73) ∂X ∂X ~ · ∂ X × ∂ X dA
~r · d~S = ~r · × dA = X
S ∂u ∂w ∂u ∂w
Z πR Z π  R

= b(a + b cos w)2 (cos2 u cos w + sin2 u cos w) + b2 (a + b cos w) sin2 w du dw
−π −π
Z π Z π  
= b(a + b cos w) (a + b cos w) cos w + b sin2 w du dw
−π −π
A. Moiola, University of Reading 101 Vector calculus lecture notes, 2016–17
Z π Z π
= du b(a + b cos w)(a cos w + b) dw
−π −π
Z π    
= 2π ab2 + (a2 b + b3 ) cos w + ab2 cos2 w dw = 2π 2πab2 + πab2 = 6π 2 ab2 .
−π

2
Solution of Exercise 2.65. The ellipsoid E = {~r ∈ R3 s.t. x2 + y 2 + zc2 < 1}, in cylindrical coordinates
reads    r 
3 2 z2 3 z2
E = ~r ∈ R s.t. r < 1 − 2 = ~r ∈ R s.t. r < 1 − 2 .
c c
Thus, it is a solid of revolution as in equation (83), with −c < z < c (the admissible interval for z
2
corresponds to the largest interval which guarantees 1 − zc2 ≥ 0). From formula (84),
ZZZ Z c   c
z2 z3  4
Vol(E) = dV = π 1 − 2 dz = π z − 2  = πc.
E −c c 3c −c 3

Note that this result agrees with the formula for the volume of the sphere when c = 1.
16 π
Solution of Exercise 2.66. Using formula (84) it is easy to find Vol(B) = 15 π and Vol(F ) = 2.

Solution of Exercise 2.67. Using the computations already done in Example 2.62 we have:
ZZZ Z π/2 Z π Z (cos z)(2+sin 3θ)
Vol(D) = dV = r dr dθ dz
D −π/2 −π 0
Z π/2 Z π
1 1 π 9π 2
= (cos z)2 dz (2 + sin 3θ)2 dθ = 9π = .
2 −π/2 −π 2 2 4

You can plot the domain D in Matlab/Octave using the function VCplotter.m with the command:
VCplotter(7, @(theta,z) (cos(z))*(2+sin(3*theta)), -pi/2, pi/2);
Solution of Exercise 2.68. We simply write the triple integrals in cylindrical coordinates using the
volume element (82). The Cartesian coordinates are written in the cylindrical system using (81).
ZZZ Z 1 Z π Z z Z π Z 1  Z z  Z 1
z2 π
z dV = zr dr dθ dz = dθ z r dr dz = 2π z dz = ;
C 0 −π 0 −π 0 0 0 2 4
ZZZ Z 1 Z π Z z Z π Z 1 Z z 
(81) 1
x dV = (r cos θ)r dr dθ dz = cos θ dθ r2 dr dz = 0 = 0;
C 0 −π 0 −π 0 0 12
ZZZ Z 1 Z π Z z
(81)
(x2 + y 2 + z 2 ) dV = (r2 cos2 θ + r2 sin2 θ + z 2 )r dr dθ dz
C 0 −π 0
Z 1Z π Z z
= (r2 + z 2 )r dr dθ dz
0 −π 0
Z π Z 1 Z z  Z 1
3 4 3π
= dθ (r3 + z 2 r) dr dz = 2π z dz = ;
−π 0 0 0 4 10
ZZZ Z 1Z π Z z
g(z) dV = g(z)r dr dθ dz
C 0 −π 0
Z π Z 1  Z z  Z 1
z2
= dθ g(z) r dr dz = 2π g(z) dz.
−π 0 0 0 2

~ using (81), (85) and the fundamental trigonometry equation:

Solution of Exercise 2.71. We expand F

~ = −yı̂+x̂+k̂ = −(r sin θ)(cos θr̂−sin θθ̂)+(r cos θ)(sin θr̂+cos θθ̂)+ ẑ = r(sin2 θ+cos2 θ)θ̂+ ẑ = rθ̂+ ẑ.
F

From (86), its divergence is ∇ ~ =

~ ·F ∂0
+ r1 0 + 1 ∂r
+ ∂1 ~ is solenoidal.
= 0, thus F
∂r r ∂θ ∂z
A. Moiola, University of Reading 102 Vector calculus lecture notes, 2016–17

Solution of Exercise 2.77.

ZZZ Z π Z π Z 1
f dV = ρ4 (sin2 φ + cos 12θ) ρ2 sin φ dρ dφ dθ
B −π 0 0
Z 1 Z π Z π
= ρ dρ(sin2 φ + cos 12θ) sin φ dθ dφ
6
0 0 −π
Z π
1 π 3 2π cos 3t − 9 cos t  8π
= 2π sin φ dθ dφ =  = .
7 0 7 12 0 21

Solution of Exercise 2.79. We first convert the coordinates from degrees to radians. Denoting ϕ the
latitude, the colatitude is φ = π2 − ϕ.
2π 2π
ϕS = 37◦ = 37 ≈ 0.646, ϕN = 41◦ = 41 ≈ 0.716,
360 360
π π
φS = − ϕS = 0.925, φN = − ϕN = 0.855,
2 2
2π  3 2π  3
θE = −102◦03′ = − 102 + ≈ −1.781, θW = −109◦ 03′ = − 109 + ≈ −1.903,
360 60 360 60
(where the minus signs come from the fact that θ measures the longitude in the East direction).
Placing the origin of the axis in the Earth’s centre, Colorado can be represented as the surface S =
{~r, ρ = R = 6371, θW ≤ θ ≤ θE , φS ≤ φ ≤ φN }, so its area is
ZZ Z θ E Z φS


Area(S) = 1 dS = R2 sin φ dφ dθ = (θE − θW )R2 − cos(φS ) + cos(φN ) ≈ 269,305 km2 ,
S θW φN

which roughly agree with the value 269,837 km2 found on Wikipedia.

B.3 Exercises of Section 3

Solution of Exercise 3.7. We compute left- and right-hand sides of (96), and verify they coincide:
ZZ ZZ
~ ~
∇ × F = k̂, ⇒ left-hand side = ~ ~
(∇ × F)3 dA = 1 dA = Area(R) = 4π,
R R
d~a
~a(t) = 2 cos tı̂ + 2 sin t̂, (t) = −2 sin tı̂ + 2 cos t̂, 0 ≤ t ≤ 2π,
dt


~ ~a(t) = (2 cos t + 4 sin t)(ı̂ + 3̂),
F


~ ~a(t) · d~a (t) = 4(cos t + 2 sin t)(− sin t + 3 cos t) = 4(3 cos2 t − 2 sin2 t + 5 cos t sin t),
F
dt  



= 2 3(cos 2t + 1) − 2(1 − cos 2t) + 5 sin 2t = 2 5 cos 2t + 1 + 5 sin 2t ,
I Z 2π


F~ · d~r = ~ ~a(t) · d~a (t) dt
F
∂R 0 dt
Z 2π   2π

5 5 
= 2 5 cos 2t + 1 + 5 sin 2t dt = 2 sin 2t − cos 2t + t  = 4π.
0 2 2 0

Solution of Exercise 3.8. Since the domain R is y-simple, we compute the double integral at the
left-hand side of (96) as an iterated integral:
q
ZZ ZZ Z 2 Z 1− x4
2 


∇ ~ dA (23)
~ ×F = (−2y) dA =
(53)
(−2y) dy dx
3
R R −2 0
Z 2 
x2   8 8
= −1+ dx = 2 − 2 + =− .
−2 4 12 3
The boundary of R is composed of two parts. The horizontal segment {−2 < x < 2, y = 0} does not give
any contribution to the line integral at the right-hand side of (96), since F ~ vanishes on the x axis. The
upper arc is parametrised by the curve ~a(t) = 2 cos tı̂ + sin t̂ for 0 < t < π (recall that it needs to be
oriented anti-clockwise), whose total derivative is d~
a
dt (t) = −2 sin tı̂ + cos t̂. So the line integral reads:
I Z
~
F · d~r = ~ · d~r
F
∂R ∂R∩{y>0}
A. Moiola, University of Reading 103 Vector calculus lecture notes, 2016–17
Z π
(44)
= y 2 ı̂ · (−2 sin tı̂ + cos t̂) dt
0
Z π
= −2y 2 sin t dt (now use y = a2 (t) = sin t on ∂R ∩ {y > 0})
0
Z π Z π  2 π 8
= 3
−2 sin t dt = (2 sin t cos t − 2 sin t) dt = − cos t + 2 cos t  = − ,
2 3
0 0 3 0 3

and we have proved that the right- and the left-hand sides of formula (96) are equal to each other, for this
~
choice of domain R and field F.
(4)
Solution of Exercise 3.12. k̂ × n̂ = k̂ × (τ̂ × k̂) = τ̂ (k̂ · k̂) − k̂(τ̂ · k̂) = τ̂ 1 − k̂0 = τ̂ .
Solution of Exercise 3.16. We compute the divergence ∇ ~ = ex + 1 and the iterated triple integral:
~ ·F
ZZ ZZZ Z 1 Z 1 Z 10
~ · d~S (101)
F = ~ ·F
∇ ~ dV = (ex + 1) dz dy dx = 20 e.
∂D D 0 −1 0

Solution of Exercise 3.17. We want to find a vector field F ~ defined in the ball B = {|~r| ≤ R} such
~ · n̂ = x + y on S = ∂B. The outward-pointing unit vector field n̂ on ∂B is n̂ = r̂ = 1 ~r, so a
that F 2 2
R
suitable vector field is F ~ = Rxı̂ + Ry̂. Thus we conclude:
ZZ ZZ ZZZ ZZZ
~ · d~S (101) ~ ·F
~ dV = 8
(x2 + y 2 ) dS = F = ∇ 2R dV = 2RVol(B) = πR4 .
S S B B 3

Solution of Exercise 3.18. We first decompose the boundary ∂C in the disc B = {x2 + y 2 < 1, z = 1}
and in the lateral part L = {x2 + y 2 = z 2 , 0 < z < 1}. We need to find a vector field F ~ such that
~ ~ 2 ~
∇ · F = |~r| and F · n̂ is “easy to integrate” on ∂D. Since the divergence must be a polynomial of degree
two, the easiest choice is to look for a field whose components are polynomials of degree three. Moreover,
the position field ~r has zero normal component on the lateral boundary L of C (draw a sketch to figure out
why), so a good F ~ may be a multiple of ~r. It is easy to figure out that a possible candidate is F
~ = 1 |~r|2~r,
5
2
whose divergence is |~r| , as desired. From the divergence theorem we have:
ZZZ ZZZ ZZ
(101)
2
|~r| dV = ~ ~
∇ · F dV = ~ · d~S
F
C C ∂C
ZZ ZZ
= ~
F · |{z}
n̂ dS + ~ · n̂ dS
B
|F{z } L
=k̂ =0
ZZ
1 2
= (x + y 2 + z 2 )z dS
B 5
ZZ
1 2
= (r + 1) dS using z = 1 on B and cylindrical coordinates
B 5
Z π Z 1   
1 2 1 1 1 3π
= (r + 1)r dr dθ = 2π + = ,
5 −π 0 5 4 2 10

which agrees with the result found in Exercise 2.68.

Solution of Exercise 3.19. (i) We compute the divergence and integrate in spherical coordinates:

∇~ ·F
~ = x2 + z 2 + y 2 = |~r|2 = ρ2 ,
ZZ ZZZ ZZZ Z π Z π Z 1
~ ~ ~ ~ 2 2 1 4
F · dS = ∇ · F dV = ρ ρ sin φ dρ dφ dθ = dθ sin φ dφ ρ4 dρ = 2π 2 = π.
∂B B B −π 0 0 5 5

(ii) Similarly, we find the flux through the boundary of CL = (0, L)3 integrating in Cartesian coordinates:
ZZ ZZZ Z L Z L Z L
~ d~S =
F· ~ ·F
∇ ~ dV = (x2 + y 2 + z 2 ) dz dy dx
∂CL CL 0 0 0
Z L Z L Z L
1 1 1 1
= (Lx2 + Ly 2 + L3 ) dy dx = (L2 x2 + L4 + L4 ) dx = (L5 + L5 + L5 ) = L5 .
0 0 3 0 3 3 3
1
Thus, the two fluxes are equal if L5 = 45 π, i.e. if L = ( 45 π) 5 ≈ 1.202.
A. Moiola, University of Reading 104 Vector calculus lecture notes, 2016–17

~ = yzı̂ + xz̂ + xy k̂, from which we compute

Solution of Exercise 3.23. The gradient of the field is ∇f
the triple integral
ZZZ Z 1 Z 1 Z 1
~ dV =
∇f (yzı̂ + xz̂ + xy k̂) dx dy dz
D 0 0 0
Z 1 Z 1 Z 1 Z 1 Z 1 Z 1 Z 1 Z 1 Z 1
1
= ı̂ 1 dx y dy z dz + ̂ x dx 1 dy z dz + k̂ x dx y dy 1 dz = (ı̂ + ̂ + k̂).
0 0 0 0 0 0 0 0 0 4

The field f vanishes on the three faces of D incident to the origin (for example the lower face belongs to
the plane {z = 0}, so f (~r) = xy0 = 0). In the upper face z = 1 and n̂ = k̂, thus the contribution of this
face to the surface integral at the right-hand side of (102) is
Z 1 Z 1
1
xy1k̂ dy dx = ı̂.
0 0 4

Similarly, from the faces lying in the planes {x = 1} and {y = 1} we have the contributions 41 ı̂ and 14 ̂,
respectively. The sum of the three contributions give the value of the triple integral computed above.
Solution of Exercise 3.24. We recall that harmonic means that ∆f = 0 (Section 1.3.4), and the
∂f
normal derivative is ∂n ~ . Taking g = f in Green’s first identity, we see that
= n̂ · ∇f
ZZZ ZZZ ZZZ ZZ
~ |2 dV = ~ · ∇f
~ dV = − ∂f
|∇f ∇f f ∆f dV + f dS = 0.
D D D |{z} ∂n
∂D |{z}
=0
=0

~ |2 is non-negative (|∇f
Since the scalar field |∇f ~ (~r)|2 ≥ 0 for all ~r ∈ D), this equation implies that
~ ~
∇f = 0 everywhere in D. (More in detail, if ∇f 6= ~0 was true in a portion of D, this would imply that
~
RRR
~ |2 > 0 gave a positive contribution to the integral
|∇f ~ |2 dV that can not be cancelled by negative
|∇f
D
contributions from other parts of the domain as |∇f ~ |2 is never negative, so the integral in the formula
above could not vanish.) Since the gradient of f vanishes, all the partial derivatives of f are zero, so f
can not depend on any of the variables x, y and z, which means it is constant.
Solution of Exercise 3.20. We first note that the vector field F ~ := ~r/|~r|3 is well-defined in all of R3
except at the origin. We also know from Exercise 1.62 that ∇ ~ ·F~ = ~0 in all points ~r 6= ~0. Thus, from
RR RRR
divergence theorem we immediately conclude for the case ~0 ∈ / D: ∂D |~r|~r 3 · d~S = D
~ · ~r 3 dV =
∇ |~r|
RRR
D
0 dV = 0.
On the other hand, if ~0 ∈ D we cannot apply directly the divergence theorem to F ~ because this field
~
is not defined at the origin. We denote by B a ball centred at 0 and contained in D (it exists because D
is open) and we call R its radius. The outward-pointing unit normal vector on ∂B is n̂B (~r) = ~r/|~r|, so
ZZ ZZ ZZ ZZ
~ ~ ~ ~r ~r 1 Area(∂B) 4πR2
F · dS = F · n̂B dS = · dS = dS = = = 4π.
∂B ∂B r|3 |~r|
∂B |~ ∂B R
2 R2 R2

We now apply the divergence theorem in D \ B, where F ~ is well-defined and ∇

~ ·F
~ = 0. So we have that the
~ through the boundary of D \ B vanishes. The boundary of this set is made of two components:
flux of F
∂D, which gives the integral we want to compute, and ∂B, for which we have just computed the flux of F.~
We need to be careful with the signs: on ∂B, the vector field n̂B we have used above is outward-pointing
for B, so inward-pointing for D \ B, so the outward-pointing vector for D \ B satisfies n̂ = −n̂B . In
formulas:
ZZZ ZZ ZZ ZZ ZZ
0= ~ ~
∇ · F dV = ~ ~
F · dS = ~ ~
F · dS = ~ ~
F · dS + ~ · d~S
F
D\B ∂(D\B) ∂D∪∂B ∂D ∂B
ZZ ZZ ZZ
= ~ · d~S +
F ~ · (−n̂B ) dS =
F ~ · d~S − 4π,
F
∂D ∂B ∂D
RR
which gives ~ · dS
F ~ = 4π. (This is sometimes called Gauss’ theorem.)
∂D
A. Moiola, University of Reading 105 Vector calculus lecture notes, 2016–17

Solution of Exercise 3.30. As usual, we parametrise the unit circumference ∂S with the curve ~a(t) =
cos tı̂ + sin t̂, 0 < t < 2π. The circulation in (106) reads
I Z 2π

~ · d~r =
F (2x − y)ı̂ − yz 2 ̂ − y 2 z k̂ · (− sin tı̂ + cos t̂) dt
∂S 0
Z 2π
= (−2 cos t sin t + sin2 t) dt since on ∂S x = cos t, y = sin t, z = 0,
0
Z 2π  
1
= − sin 2t + (1 − cos 2t) dt = π.
0 2
p
~ is ∇
The curl of F ~ ×F ~ = k̂ and the surface S is the graph of the field g(x, y) = 1 − x2 − y 2 over the
disc R = {xı̂ + y̂, x2 + y 2 < 1}. Thus, using formula (74) for the flux through a graph surface, we have
ZZ ZZ  
~ ~ ~
(∇ × F) · dS =
(74)
~ 1 ∂g − (∇
~ × F)
− (∇ ~ 2 ∂g + (∇
~ × F) ~ × F)
~ 3 dA
S ∂x ∂y
ZZR
= (0 + 0 + 1) dA = Area(R) = π,
R

and both sides of (106) give the same value.

Solution of Exercise 3.31. The desired integral is the circulation of the field |~r|2 ı̂ along the boundary of
the paraboloid S, so we apply Stokes’ theorem and exploit the computation already done in Exercise 2.55:
I I ZZ ZZ
(46) (106)
4
|~r|2 dx = |~r|2 ı̂ · d~r = ~ × (|~r|2 ı̂) · d~S (31)
∇ =
Ex. 2.55
2(~r × ı̂) · d~S = − .
∂S ∂S S S 3

Solution of Exercise 3.32. We can prove the assertion in two different ways. The first option is to
use Stokes’ theorem to transform the line integral to a surface integral, then to use the product rule for
the curl, and to verify that the terms obtained are zero:
I ZZ ZZ


~ · d~r (106)
f ∇f = ∇ ~ ) · d~S (31)
~ × (f ∇f = ~ × ∇f
∇f ~ +f ∇ ~ × ∇f
~ · d~S = 0.
∂S S S | {z } | {z }
=~
0, (~ u=~
u×~ 0) =~
0, (26)

Alternatively, we can use the chain rule to reduce the desired integral to the line integral of a gradient,
which vanishes by the fundamental theorem of vector calculus since the path of integration is a loop:
I I
~ · d~r (14) 1 ~ 2 ) · d~r (49)
f ∇f = ∇(f = 0.
∂S 2 ∂S
A. Moiola, University of Reading 106 Vector calculus lecture notes, 2016–17

C Further exercises for tutorials and personal revision

Some of these exercises will be solved in the tutorials. Which exercises you should try to solve before each
tutorial will be announced in the lectures.
Exercises for Section 1.
Exercise C.1. (Exercise 4 in assignment 1 of MA2VC 2013–14.) Show that, for any unit vector n̂, and any
~ perpendicular to n̂, the identity
vector w

n̂ × (n̂ × w
~ ) = −w
~

holds true. You can make use of the identities in Section 1.1. Demonstrate this identity for the vectors
3 4
n̂ = ı̂ + k̂, ~ = 3̂.
w
5 5
Exercise C.2. Compute and draw the level sets of the scalar field f (~r) = xey .
Exercise C.3. Draw some paths of the curves

~a(t) = (λ + t2 )ı̂ + t̂, ~b(t) = tı̂ + (λt − λ2 )̂, ~c(t) = t cos(t + λ)ı̂ + t sin(t + λ)̂

for different values of λ ∈ R. Are they level sets of any scalar field?
Exercise C.4. Let û and ŵ be two unit vectors orthogonal to each other. Show that the curve ~a(t) =
û cos t + ŵ sin t lies on the unit sphere for all t ∈ R. Which curve is this?
Exercise C.5. Draw the path of the curve
(cos t − sin t)ı̂ + (cos t + sin t)̂
~a(t) = , 0 ≤ t ≤ 2π.
| cos t| + | sin t|

~ /g) where f and g 6= 0 are

Exercise C.6. Find a quotient rule for the gradient, i.e. give a formula for ∇(f
scalar fields. Hint: use parts 2 and 3 of Proposition 1.33.
Exercise C.7. • Compute the vector field n̂ of unit length defined on the sphere of radius R > 0, which is
orthogonal to the sphere itself and points outwards. (This is called “outward-pointing unit normal vector
field”.) Hint: use part 4 of Proposition 1.33.
• Compute the outward-pointing unit normal vector field on the boundary of the parabolic “cup”
{x2 + y 2 < z < 1}.
~ = exyz ı̂. Show that F
Exercise C.8. • Fix F ~ and ∇
~ ×F
~ are orthogonal to each other at each point of R3 .

• Prove that any vector field with only one non-zero component is orthogonal to its own curl.
• Find a vector field that is not orthogonal to its own curl.
Exercise C.9. Find a scalar field f defined on the complement of the x axis, whose direction of maximal
increase is (~r − xı̂)/(y 2 + z 2 )1/2 .
Exercise C.10. (Exercises 1–2 in assignment 1 of MA2VC 2013–14.) Prove the following identity:


∇~ · ~rf g = 3f g + (~r · ∇f
~ )g + (~r · ∇g)f,
~

where ~r = xı̂ + y̂ + z k̂ is the position vector, and f (~r) and g(~r) are two scalar fields.
Hint: you can either use the vector differential identities in the boxes of Propositions 1.52 and 1.55, or the
definitions of gradient and divergence.
Demonstrate the above identity for the scalar fields f = exy and g = y 4 z.
Exercise C.11. Prove identity (34) in the notes (curl of a vector product).

Exercise C.12. • Find a vector field F ~ such that: F ~ is irrotational, F ~ is not solenoidal, all its streamlines
(recall caption to Figure 7) are straight lines passing through the origin.
~ r) is parallel to ~r in all points ~r 6= ~0.
This last condition means that F(~

• [Harder!] Find a vector field G ~ (defined on a suitable domain) such that: G ~ is solenoidal, G
~ is not
irrotational, all its streamlines are straight lines passing through the origin.
 A. Moiola, University of Reading 107 Vector calculus lecture notes, 2016–17

Exercise C.13. Compute the total derivative of the field f (~r) = (x2 + y 2 )β , for β ∈ R, evaluated along the
2
curve ~a(t) = cos 2πtı̂ + sin 2πt̂ + e−t k̂.
Interpret the result geometrically.
Exercise C.14. Let F : (0, ∞) → R be a smooth function. Let f : R3 \ {~0} → R be the scalar field defined
by f (~r) = F (|~r|).
~ .
• Compute ∇f
2F ′ (|~r|)
• Prove that ∆f = F ′′ (|~r|) + |~r| . Hint: use one of the identities of Section 1.4.

Exercise C.15. Let F ~ : R3 → R3 be a conservative force field, i.e. a conservative field with F
~ = −∇ψ,~ where
the scalar field ψ is the potential energy. Let a particle of mass m move with trajectory ~a(t) according to
Newton’s law F(~~ a(t)) = m d2~2a (t) (force equals mass times acceleration). Define the kinetic energy of the
 a 2 dt
particle T (t) = 21 m d~ 
dt (t) . Prove that the total energy E(t) = ψ(~ a(t)) + T (t) of the particle is constant in
time.
Hint: you only need to use chain and product rule for the total derivative.
Exercises for Section 2.
Exercise
p C.16. Consider the curve ~a(t) = t cos t ı̂ + t sin t ̂ for t ∈ R[0, 4π] and the scalar field f (~r) =
1/ 1 + |~r|2 . Draw the path Γ of the curve, compute the line integral Γ f ds and the total derivative of f
evaluated along ~a.
~ r) = y 2 ı̂ + 2xy̂ along the five curves of
Exercise C.17. Compute the line integral of the vector field F(~
Example 2.10.
Exercise C.18. Compute the integral of the scalar field f = x + y on the square Q with vertices ı̂, ̂, −ı̂, −̂.
Exercise C.19. Compute the area of the “eye” domain {xı̂ + y̂, 1 − y 2 < |x| < 2(1 − y 2 )}.
Hint: draw a sketch of the domain. This domain is neither y-simple nor x-simple, to compute the double
integral you need to use two of the fundamental properties of integrals in list (40) in the notes.
~
Exercise C.20. Compute the area of T(R), ~
where R = {xı̂ + y̂, 0 < x < 1, 0 < y < 1} and T(x, y) =
3
(x + y)ξ̂ + y η̂.
Exercise C.21. Find a change of variables from the trapezoid R with vertices −2ı̂, 2ı̂, ı̂ + ̂, −ı̂ + ̂ to the
square Q with vertices ~0, ξ̂, ξ̂ + η̂, η̂. RR 2
Use the change of variables you have found to compute R e(2−y) dx dy.

Exercise C.22. Let S be the surface parametrised by the chart X(u, ~ w) = uı̂ + w u ̂ + wk̂ defined over the
triangular region R = {0 < w < u < 1}. Compute the flux through S of the field F ~ = ̂ + 2xk̂.
Rewrite the surface as the graph of a scalar field g(x, y) over a suitable region Re and compute again the
same flux using formula (74).
2 2
Exercise C.23. Compute the triple integral of f = x
p + y over the domain D lying inside the ball of radius
a > 0 centred at the origin and above the cone z = x + y 2 .
2

Exercise C.24. Compute the area of the region bounded by ~a(t) = t(2π − t)(cos tı̂ + sin t̂), 0 ≤ t ≤ 2π.
Exercise C.25. London and Astana (Kazakhstan) approximately lie on the parallel 51◦ N , and have longitude
0◦ and 71◦ E, respectively. Describe the location of the two cities in a suitable special coordinates system.
Compute the distance between the two cities, if the distance is measured:
(i) along a straight (underground) segment,
(ii) on the Earth surface along the parallel 51◦ N ,
(iii) along the shortest surface path.
Assume that Earth is a sphere of radius 6371 km.
Exercises for Section 3.
Exercise C.26. Compute the area of the region bounded by the curve ~a(t) = sin tı̂ + sin 2t̂, 0 ≤ t ≤ 2π.
~ = xyı̂ + yz̂ + zxk̂ through the boundary of the cube C = (0, 2)3 =
Exercise C.27. Compute the flux of F
{0 < x, y, z < 2}.
A. Moiola, University of Reading 108 Vector calculus lecture notes, 2016–17

~ = 1 z 2 k̂ and the upper half unit ball B = {z >

Exercise C.28. Demonstrate the divergence theorem for F 2
0, |~r| < 1}.

Exercise C.29. Let F ~ be a vector field, and a and ǫ be two positive numbers such that F(~~ r) · ~r ≥ ǫ for all
~ is not solenoidal in the ball B of radius a centred at the origin.
~r ∈ R3 with |~r| = a. Prove that F

Exercise C.30. Prove the following “integration by parts formula” for the curl: given a domain D ⊂ R3 and
~ G,
two vector fields F, ~
ZZZ ZZZ ZZ
~ × F)
(∇ ~ ·G~ dV = ~ · (∇
F ~ × G)
~ dV + ~ × n̂) · F
(G ~ dS.
D D ∂D

~ = zyı̂ and the quadrilateral S with vertices

Exercise C.31. Demonstrate Stokes’ theorem for the field F
ı̂, ı̂ + 12 ̂ − k̂, ̂, 2k̂.

Exercise C.32. Consider the field G~ = 2x2 y 2 z̂ − 2x2 yz 2 k̂ and a triangle T with the three vertices on the x-,
y- and z-axis, respectively. Show that the flux of G~ through T is zero.
~ = xı̂ + y̂ − 2z k̂ through the cylindrical surface
Exercise C.33. Use Stokes’ theorem to compute the flux of G
2 2
S = {x + y = 1, 0 < z < H}, where H > 0.

D Alphabetical index
∆ (Laplacian), 17 Colatitude, 64
· (scalar product), 3 Connected domain, 69
∂(ξ,η) Conservative field, 25
∂(x,y) (Jacobian determinant), 46
∂(ξ,η,σ) Contour integral, 32
∂(x,y,z) (Jacobian determinant, 3D), 53
curl operator, 19
∂D (boundary of a domain), 57 Curve, 9
∂S (boundary of a surface), 59 Cylindrical coordinates, 62
× (vector product), 4
~ (vector Laplacian), 17 ∂
∆ ∂x (partial derivative), 12
~ ∂
F · ∇ (advection operator), 23 ∂ û (directional derivative), 13
~ (nabla symbol), 13 ∂ ∂
∇ ∂n , ∂ n̂ (normal derivative), 13
~ (divergence), 18
∇· Del, 13
~ (curl operator), 19
∇× Direction (of a vector), 2
~ · ∇ (advection operator), 19 Directional derivative, 13
F
RRR Divergence, 18
RR D f dV , 52 Domain, 6
RRR f dA, 42 Double integral, 42
f dS, 53
R S
~ Field line, 9
RΓ F · d~r, 35
RRΓ f ds, 31 Flux, 58
H ∂D (integral on closed surfaces), 58 Fundamental theorem of vector calculus, 39
(contour integral, circulation), 32, 35
Gradient, 13
Additivity of integrals, 31 Graph surface, 8, 55
Advection operator ~ (Hessian), 17
HF
for scalar fields, 19 Helmholtz decomposition, 26
for vector fields, 23 Hessian, 17
Arc length, 27
Azimuth, 64 ı̂ (first canonical unit vector), 2
Induced orientation, 59
Boundary, 57 Intrinsic parametrisation of a curve, 27
Bounded domain, 69 Irrotational field, 25
Iterated integral, 44
Canonical basis, 2
Chain rule JF ~ (Jacobian matrix), 16
for gradient, 14, 28 ̂ (second canonical unit vector), 2
for partial derivatives, 13, 28 Jacobian determinant, 46
for total derivative, 27 Jacobian matrix, 16
Chart, 54 k̂ (third canonical unit vector), 2
Circulation, 35
Closed set, 6 Laplacian, 17
A. Moiola, University of Reading 109 Vector calculus lecture notes, 2016–17

Length (of a vector), 2 for the total derivative, 27

Length of a path, 34 for vector differential operators, 22
Level line, level surface, 8
Line integral ~r (position vector), 3
of scalar fields, 31 Region, 7
of vector fields, 35 Riemann sum, 33
Linearity
of differential operators, 20 Scalar field, 7
of gradient, 14 Scalar potential, 25
of integrals, 31 Scalar product, 3
of partial derivatives, 13 Second order partial derivatives, 16
of the total derivative, 27 Simply connected domain, 26, 69
Longitude, 64 Solenoidal field, 25
Loop, 9 Solid of revolution, 63
Spherical coordinates, 64
Magnitude (of a vector), 2 Star-shaped domain, 39, 61, 66
Streamline, 9
Nabla, 13 Surface, 54
Normal derivative, 13, 77 Surface area, 56
Surface integral, 53, 55
Open set, 6
Oriented path, 36 τ̂ (unit tangent vector to a curve), 36
Oriented surface, 56 Total derivative, 26
Triple integral, 52
Parametric surface, 54
Triple product, 5
Parametrisation
of a curve, 9 Unit normal to a surface, 15, 56
of a surface, 54 Unit normal to a two-dimensional region, 73
Partial derivative, 12 Unit tangent vector to a curve, 36
Path, 9
Piecewise smooth domain, 68 Vector field, 9
Polar coordinates, 60 Vector Laplacian, 17
Position vector, 3 Vector potential, 25
Product rule Vector product, 4
for partial derivatives, 13
for the gradient, 14 y-simple domain, 42

E References
[1] R.A. Adams, C. Essex, Calculus, a complete course, Pearson Canada, Toronto, 7th ed., 2010.
UoR library call number: FOLIO–515-ADA.
[2] E. Carlen, Multivariable Calculus, Linear Algebra and Differential Equation, lecture notes, 2013,
available on: http://www.math.rutgers.edu/~carlen/291F14/index.html.
[3] M. Corral, Vector Calculus, 2013, available on: http://www.mecmath.net/.
[4] O. Knill, http://www.math.harvard.edu/~knill/teach/index.html. Here you can find plenty of
materials: lecture notes, exercises and instructive illustrations. Click on one of the Math(s)21a links.
[5] S. Lang, Calculus of Several Variables, Springer Undergraduate Texts in Mathematics, 3rd ed., 1996.
[6] C.J. Smith, Vector Calculus Primer, 2011, Reading, available on Blackboard and on course web page.
[7] M. Spiegel, S. Lipschutz, D. Spellman, Vector Analysis, McGraw-Hill, 2nd ed., 2009.
A. Moiola, University of Reading 110 Vector calculus lecture notes, 2016–17

F Common errors in vector calculus

This is a list of errors I found dozens of times in previous vector calculus assignments and exams. Many
of them are trivial and can easily be avoided, nonetheless they appear too often, watch out!
• The most common source of mistakes is treating scalar quantities as vector ones and vice versa. E.g.
summing or equating scalars to vectors, writing the divergence of a scalar field, evaluating fields at a
number and curves at a vector. One of the main aims of this module is to learn how to manipulate
these objects, so take the greatest care to avoid these mistakes! Remember: never confuse scalars
with vectors.
• An unbelievably large number of errors happen because brackets are not used or are badly placed.
∂ ∂
Typical example: instead of ∂x (f g) you write ∂x f g, in the next line you have forgotten that also g
∂f
was inside the derivation and you go on with ∂x g, which is wrong. When in doubt, always write the
brackets.
• You should know very well that the signs “=”, “⇐⇒” and “⇒” cannot be used in place of each
other. The equal sign can only stay between two numbers, vectors, sets, matrices, or any two other
comparable mathematical objects. The implication ⇒ and the double implication ⇐⇒ (“if and only
if”) stay between two propositions, statements or equations (in which case they contain an equal sign).
• Whenever you use a mathematical object denoted by a certain letter for the first time in an exercise,
you must introduce it, unless it is known from a standard definition (or from the statement of the
exercise). E.g. the number π and the position vector ~r do not need to be defined every time and you
~ ~a, D . . . are not standard and you need to define
can use them directly; fields, curves, domains f, g, F,
them. Examples of typical correct sentences: “given a scalar field f . . . ”, “let D ⊂ R3 be a domain. . . ”,
“fix ~a(t) := t3 ı̂ for t ∈ (−1, 1). . . ”
• Every time you write an (either single or multiple) integral sign, you must also write (1 ) the correct
infinitesimal element (e.g. dt, dS, · d~r, dx dy dz, . . .) and (2) either a domain of integration (placed
RRR R 1−y
under it, e.g. D in D
· · · dV ) or the extremes of integration (e.g. 0 · · · dx, only for single integrals
or line integrals of conservative fields). In this module we never consider indefinite integrals.
• Read carefully the main verb of the exercise request: “prove” (or “show”), “compute” (or “find”,
“calculate”), “demonstrate”, “use”, “state”. . . are not all the same! E.g. if you are asked to demonstrate
a certain identity A = B, you have to compute both A and B and verify they are equal; if you are
asked to use the same identity to compute B, you only compute A and state the identity.
• Do not mix up the different theorems. You need to know and remember well the statements of all the
main ones, it is not so hard! (E.g. the divergence theorem is that one containing the divergence, not
surprisingly, and not the curl or the partial derivative in z.) You do not need to remember by hearth
the identities in Proposition 1.55.
• When you prove a vector identity (either differential as in Proposition 1.55 or integral like (103))
and you consider only the first component of the identity, you cannot ignore the second and third
components of the fields involved, nor the derivatives with respect to the second and third variables.
• Whenever you can use the differential identities of Proposition 1.52 and 1.55, use them! Once you
become comfortable with them, their use is much easier than the expansion in coordinates, see the
example in Remark 1.58.
~ ∇·,
• The symbols ∇, ~ ∇×,
~ ∆ are different objects; every time you use one of them in place of another
one you do a mistake.
• In scalar products and in the divergence, the dot symbol “ · ” is not optional, it must be written clearly.
• If at the beginning of an exercise you report the request, distinguish clearly which are the data you
have and which is the assertion you want to prove or demonstrate or the entity you want to compute.
If you simply list many formulas it is not clear what you are doing.
• Often common sense helps: a negative length, area or volume is surely wrong.
• You must always write what you are computing. E.g. if an exercise requires to compute the potential
ϕ, your solution should include either a sentence such as “the desired potential is equal to xyez ”, or a
formula like “ϕ = xyez ”, and not simply the result “xyez ”.
• In the notes there are several “warning” remarks describing frequent errors, read them carefully. See
1.4, 1.16, 1.27, 1.45, 1.49.
A. Moiola, University of Reading 111 Vector calculus lecture notes, 2016–17

G References to the relevant textbook sections and exercises

Note sections Book sections Book exercises

Vectors, scalar product 1.1, 1.1.1 10.2 2–3, 16–31
Vector product 1.1.2 10.3 1–28
Scalar fields 1.2.1 12.1 11–42
Vector fields 1.2.2 15.1 1-8
Curves 1.2.3 11.1
11.3 1–12
Partial derivatives 1.3.1 12.3 1–12, 25–31
12.5
Gradient 1.3.2 12.7 1–6 (part a), 10–15
16.1
Jacobian matrix 1.3.3 12.6
Second-order partial derivatives, Laplacian 1.3.4 12.4 1–14
Divergence and curl operator 1.3.5,1.3.6 16.1 1–8
Vector differential identities 1.4.1, 1.4.2 16.2 8, 11, 13, 14
Conservative fields, scalar potentials 1.5 15.2 1–10
Irrotational and solenoidal fields, potentials 1.5 16.2 9, 10, 12, 15, 16
Total derivative of curves 1.6 11.1 1–14, 27–32
Chain rule 1.7 12.5 1–12, 15–22
Line integral of scalar fields 2.1.1 8.4 1–9
15.3 1–9
11.3 13–21
Line integral of vector fields 2.1.2, 2.1.3 15.4 1–12, 15–22
Double integrals 2.2.1 14.1 13–22
14.2 1–28
Change of variables 2.2.2 14.4 32–34
14.6
Triple integrals 2.2.3 14.5 1–20
Surface integrals 2.2.4 14.7 1–10
15.5 3, 4, 7–10, 13–16
Flux integrals 2.2.5 15.6 1–13
Polar coordinates 2.3.1 8.5 1–28
8.6 1-11
14.4 1–30
Cylindrical and spherical coordinates 2.3.2, 2.3.3 10.6 1–14
14.6 1–16
15.2 19–22
16.1 9–11
16.7 1–4, 14
Green’s theorem 3.1 16.3 1–7
Divergence theorem 3.2 16.4 1–14
Stokes’ theorem 3.3 16.5 1–12

Table 4: References to the relevant sections of the textbook [1] by Adams and Essex. The last column lists
the exercises you should be able to attempt. Some of them are straightforward, while others are quite hard,
or require some of the remarks marked with ⋆, or use a slightly different notation, or are presented in a
different way from what you are used to: solving them is even more instructive. In many cases, drawing
a sketch of the domain is crucial to solve the exercises.
A. Moiola, University of Reading 112 Vector calculus lecture notes, 2016–17

Contents
1 Fields and vector differential operators 2
1.1 Review of vectors in 3-dimensional Euclidean space . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 Scalar product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.2 Vector product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.3 ⋆ Open and closed subsets of R3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 Scalar fields, vector fields and curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.1 Scalar fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.2 Vector fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2.3 Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Vector differential operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3.1 Partial derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3.2 The gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3.3 The Jacobian matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.4 Second-order partial derivatives, the Laplacian and the Hessian . . . . . . . . . . . 16
1.3.5 The divergence operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.3.6 The curl operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.4 Vector differential identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.4.1 Second-order vector differential identities . . . . . . . . . . . . . . . . . . . . . . . 20
1.4.2 Vector product rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.5 Special vector fields and potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.6 Total derivatives of a curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.7 Chain rule for fields and curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.8 Review exercises for Section 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

2 Vector integration 31
2.1 Line integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.1.1 Line integrals of scalar fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.1.2 Line integrals of vector fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.1.3 Independence of path and line integrals for conservative fields . . . . . . . . . . . . 38
2.2 Multiple integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.2.1 Double integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.2.2 Change of variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.2.3 Triple integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.2.4 Surface integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.2.5 Unit normal fields, orientations and fluxes (surface integrals of vector fields) . . . . 56
2.3 Special coordinate systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.3.1 Polar coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.3.2 Cylindrical coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.3.3 Spherical coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

3 Green’s, divergence and Stokes’ theorems 68

3.1 Green’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.2 The divergence theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.3 Stokes’ theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

A General overview of the notes 82

B Solutions of the exercises 83

B.1 Exercises of Section 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
B.2 Exercises of Section 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
B.3 Exercises of Section 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

C Further exercises for tutorials and personal revision 106

D Alphabetical index 108

E References 109

F Common errors in vector calculus 110

G References to the relevant textbook sections and exercises 111