CLP-4 V ECTOR C ALCULUS

Joel F ELDMAN Andrew R ECHNITZER Elyse Y EAGER

T HIS DOCUMENT WAS TYPESET ON F RIDAY 28 TH A UGUST, 2020.

§§ Legal stuff
• Copyright c 2017–20 Joel Feldman, Andrew Rechnitzer and Elyse Yeager.

• This work is licensed under the Creative Commons Attribution-NonCommercial-

ShareAlike 4.0 International License. You can view a copy of the license at
http://creativecommons.org/licenses/by-nc-sa/4.0/.

• Links to the source files can be found at the text webpage

2
 C ONTENTS

1 Curves 1
1.1 Derivatives, Velocity, Etc. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Reparametrization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.4 Curves in Three Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.5 A Compendium of Curve Formula . . . . . . . . . . . . . . . . . . . . . . . . 29
1.6 Integrating Along a Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.7 Sliding on a Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
1.8 Optional — Polar Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
1.9 Optional — Central Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
1.10 Optional — Planetary Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
1.11 Optional — The Astroid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
1.12 Optional — Parametrizing Circles . . . . . . . . . . . . . . . . . . . . . . . . 48

2 Vector Fields 51
2.1 Definitions and First Examples . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.2 Optional — Field Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.2.1 More about r1 (t) ˆ v(r(t)) = 0 . . . . . . . . . . . . . . . . . . . . . . 65
2.3 Conservative Vector Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
2.4 Line Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
2.4.1 Path Independence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
2.5 Optional — The Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

3 Surface Integrals 91
3.1 Parametrized Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.2 Tangent Planes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
3.3 Surface Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.3.1 Parametrized Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.3.2 Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.3.3 Surfaces Givenť by Implicit Equations . . . . . . . . . . . . . . . . . . 111
3.3.4 Examples of S ρ dS . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

i
CONTENTS CONTENTS

3.3.5 Optional — Dropping Higher Order Terms in du, dv . . . . . . . . . 124

3.4 Interpretation of Flux Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . 126
3.4.1 Examples of Flux Integrals . . . . . . . . . . . . . . . . . . . . . . . . 127
3.5 Orientation of Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

4 Integral Theorems 140

4.1 Gradient, Divergence and Curl . . . . . . . . . . . . . . . . . . . . . . . . . . 140
4.1.1 Vector Identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.1.2 Vector Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
4.1.3 Interpretation of the Gradient . . . . . . . . . . . . . . . . . . . . . . . 155
4.1.4 Interpretation of the Divergence . . . . . . . . . . . . . . . . . . . . . 156
4.1.5 Interpretation of the Curl . . . . . . . . . . . . . . . . . . . . . . . . . 160
4.2 The Divergence Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
4.2.1 Optional — An Application of the Divergence Theorem — the Heat
Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
4.2.2 Variations of the Divergence Theorem . . . . . . . . . . . . . . . . . . 179
4.2.3 An Application of the Divergence Theorem — Buoyancy . . . . . . . 181
4.2.4 Optional — Torque . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
4.2.5 Optional — Solving Poisson’s Equation . . . . . . . . . . . . . . . . . 189
4.3 Green’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
4.4 Stokes’ Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
4.4.1 The Interpretation of Div and Curl Revisited . . . . . . . . . . . . . . 221
4.4.2 Optional — An Application of Stokes’ Theorem — Faraday’s Law . 223
4.5 Optional — Which Vector Fields Obey ∇ ˆ F = 0 . . . . . . . . . . . . . . . . 224
4.6 Optional — More Interpretation of Div and Curl . . . . . . . . . . . . . . . . 232
4.7 Optional — A Generalized Stokes’ Theorem . . . . . . . . . . . . . . . . . . . 241

A Trigonometry 254
A.1 Trigonometry — Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
A.2 Trigonometry — Special Triangles . . . . . . . . . . . . . . . . . . . . . . . . 254
A.3 Trigonometry — Simple Identities . . . . . . . . . . . . . . . . . . . . . . . . 255
A.4 Trigonometry — Add and Subtract Angles . . . . . . . . . . . . . . . . . . . 256
A.5 Inverse Trigonometric Functions . . . . . . . . . . . . . . . . . . . . . . . . . 257

B Powers and Logarithms 259

B.1 Powers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
B.2 Logarithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260

C Table of Derivatives 262

D Table of Integrals 264

E Table of Taylor Expansions 266

F 3d Coordinate Systems 268

F.1 Cartesian Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
F.2 Cylindrical Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
F.3 Spherical Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270

ii
CONTENTS CONTENTS

G Conic Sections and Quadric Surfaces 273

H Review of Linear Ordinary Differential Equations 276

iii
CONTENTS CONTENTS

iv
 Chapter 1

C URVES

We are now going to study vector-valued functions of one real variable. That is, we are
going to study functions that assign to each real number t (typically in some interval) a
vector1 r(t). For example


r ( t ) = x ( t ), y ( t ), z ( t )

might be the position of a particle at time t. As t varies, r(t) sweeps out a curve.

r(0)
r(1)
r(2)

While in some applications t will indeed be “time”, it does not have to be. It can be simply
a parameter that is used to label the different points on the curve that r(t) sweeps out. We
then say that r(t) provides a parameterization of the curve.

Example 1.0.1

While we will often use t as the parameter in a parametrized curve r(t), there is no need
to call it t. Sometimes it is natural to use a different name for the parameter. For example,
consider the circle x2 + y
2 = a2 . It is natural to use the angle θ in the sketch below to label
the point a cos θ , a sin θ on the circle.

1 We are going to use boldface letters, like r, to designate vectors. When writing by hand, it is clearer to
use arrows, like ~r, instead.

1
C URVES

y
` ˘
a cos θ , a sin θ
x2 ` y 2 “ a2

θ
x

That is,
0 ď θ ď 2π


r(θ ) = a cos θ , a sin θ
is a parametrization of the circle x2 + y2 = a2 . Just looking at the figure above, it is clear
that, as θ runs from 0 to 2π, r(θ ) traces out the full circle.
However beware that just knowing that r(t) lies on a specified curve does not guaran-
tee that, as t varies, r(t) covers the entire curve. For example, as t runs over the whole real
line, π2 arctan(t) runs over the interval (´1, 1). For all t,
c !
2 4
1 ´ 2 arctan2 (t)


r ( t ) = x ( t ), y ( t ) = a arctan(t) ,
π π

is well-defined and obeys x (t)2 + y(t)2 = a2 . But this r(t) does not cover the entire circle
because y(t) is always positive.
Example 1.0.1

Curves often arise as the intersection of two surfaces. For example, the intersection
y2 2
of the ellipsoid x2 + 2 + z3 = 1 with the paraboloid z = x2 + 2y2 is the blue curve in
the figure below. One way to parametrize such curves is to choose one of the three

z “ x2 ` 2y 2

y2 z2
x2 ` 2
` 3
“1

2
C URVES

coordinates x, y, z as the parameter, and solve the two given equations for the remaining
two coordinates, as functions of the parameter. Here are two examples.
Example 1.0.2

The set of all ( x, y, z) obeying

x3 ´ e3y =0
x2 ´ ey + z = 0

is a curve. We can choose to use y as the parameter and think of

x3 = e3y
x2 + z = ey

as a system of two equations for the two unknowns x and z, with y being treated as a
given constant, rather than as an unknown. We can now solve the first equation for x,
substitute the result into the second equation, and finally solve for z.

x3 = e3y ùñ x = ey
x2 + z = ey ùñ e2y + z = ey ùñ z = ey ´ e2y

So
r(y) = ey , y , ey ´ e2y

is a parametrization for the given curve.

Example 1.0.2

Example 1.0.3

The previous example was rigged so that it was easy to solve for x and z as functions of y.
In practice it is not always easy, or even possible, to do so. A more realistic example is the
set of all ( x, y, z) obeying

y2 z2
x2 + + =1
2 3
x2 + 2y2 = z

which is the blue curve in the figure above. Substituting x2 = z ´ 2y2 (from the second
equation) into the first equation gives

3 z2
´ y2 + z + =1
2 3
or, completing the square,
3 2 1 3 2 7
´ y + z+ =
2 3 2 4

3
 C URVES 1.1 D ERIVATIVES , V ELOCITY, E TC .

If, for example, we are interested in points ( x, y, z) on the curve with y ě 0, this can be
solved to give y as a function of z.
c
2 3 2 14
y= z+ ´
9 2 12

Then x2 = z ´ 2y2 also gives x as a function of z. If x ě 0,

c
4 3 2 14
x= z´
z+ +
c 9 2 6
4 4 2 1
= ´ z ´ z
3 9 3

The other signs of x and y can be gotten by using the appropriate square roots. In this
example, ( x, y, z) is on the curve, i.e. satisfies the two original equations, if and only if all
of (˘x, ˘y, z) are also on the curve.
Example 1.0.3

1.1IJ Derivatives, Velocity, Etc.

This being a Calculus text, one of our main operations is differentiation. We are now
interested in parametrizations r(t). It is very easy and natural to extend our definition of
derivative to r(t) as follows.

Definition 1.1.1.

The derivative of the vector valued function r(t) is defined to be

rptq
dr r(t + h) ´ r(t) rpt ` hq ´ rptq
r1 ( t ) = (t) = lim
dt hÑ0 h
rpt ` hq


when the limit exists. In particular, if r(t) = x (t) , y(t) , z(t) , then

r1 ( t ) = x 1 ( t ) , y1 ( t ) , z1 ( t )

That is, to differentiate a vector valued function of t, just differentiate each of its
components.

And of course differentiation interacts with arithmetic operations, like addition, in the
obvious way. Only a little more thought is required to see that differentiation interacts
quite nicely with dot and cross products too.

4
C URVES 1.1 D ERIVATIVES , V ELOCITY, E TC .

Theorem 1.1.2 (Arithmetic of differentiation).

Let

˝ a(t), b(t) be vector valued differentiable functions of t P R that take values

in Rn and
˝ α, β P R be constants and
˝ γ(t) and s(t) be real valued differentiable functions of t P R

Then
d
α a ( t ) + β b ( t ) = α a1 ( t ) + β b1 ( t )

(a) (linear combination)
dt
d
γ ( t ) b ( t ) = γ1 ( t ) b ( t ) + γ ( t ) b1 ( t )

(b) (multiplication by scalar function)
dt
d
a ( t ) ¨ b ( t ) = a1 ( t ) ¨ b ( t ) + a ( t ) ¨ b1 ( t )

(c) (dot product)
dt
d
a ( t ) ˆ b ( t ) = a1 ( t ) ˆ b ( t ) + a ( t ) ˆ b1 ( t )

(d) (cross product)
dt
d
a s ( t ) = a1 s ( t ) s1 ( t )



(e) (composition)
dt

Let’s think about the geometric significance of r1 (t). In particular, let’s think about the
relationship between r1 (t) and distances along the curve. The derivative r1 (t) is the limit
r(t+h)´r(t)
of h as h Ñ 0. The numerator, r(t + h) ´ r(t), is the vector with head at r(t + h)
and tail at r(t).

r(t)
r(t + h) − r(t) ≈ r′ (t) h

r(t + h)

When h is very small this vector

˝ has the essentially the same direction as the tangent vector to the curve at r(t) and
˝ has length being essentially the length of the part of the curve between r(t) and
r ( t + h ).

Taking the limit as h Ñ 0 yields that

˝ r1 (t) is a tangent vector to the curve at r(t) that points in the direction of increasing
t and ˇ dr ˇ
˝ if s(t) is the length of the part of the curve between r(0) and r(t), then ds ˇ ˇ
dt ( t ) = dt ( t ) .

This is worth stating formally.

5
C URVES 1.1 D ERIVATIVES , V ELOCITY, E TC .

Lemma 1.1.3.

Let r(t) be a parametrized curve.

(a) Denote by T̂(t) the unit tangent vector to the curve at r(t) pointing in the
direction of increasing t. If r1 (t) ‰ 0 then

r1 ( t )
T̂(t) =
|r1 (t)|

(b) Denote by s(t) the length of the part of the curve between r(0) and r(t). Then

ds ˇ dr ˇ
ˇ ˇ sptq
( t ) = ˇ ( t )ˇ rp0q
dt dt ˇ ˇ
żT
ˇ dr ˇ
s( T ) ´ s( T0 ) = ˇ ˇ
ˇ dt (t)ˇ dt rptq
T0
T̂ptq

(c) In particular, if the parameter happens to be arc length, i.e. if t = s, so that

ds
ds = 1, then ˇ ˇ
ˇ dr ˇ
ˇ ( s )ˇ = 1 T̂(s) = r1 (s)
ˇ ds ˇ

As an application, we have the

Lemma 1.1.4.


If r(t) = x (t) , y(t) , z(t) is the position of a particle at time t, then


position at time t = r(t) = x (t) , y(t) , z(t)
ds
velocity at time t = v(t) = r1 (t) = x1 (t) , y1 (t) , z1 (t) =


(t) T̂(t)
dt
b
ds
speed at time t = (t) = |v(t)| = |r (t)| = ( x1 (t)2 + y1 (t)2 + z1 (t)2
1
dt
acceleration at time t = a(t) = r2 (t) = v1 (t) = x2 (t) , y2 (t) , z2 (t)

and the distance travelled between times T0 and T is

żTˇ ˇ żTb
ˇ dr ˇ
s( T ) ´ s( T0 ) = ˇ (t)ˇ dt = ( x1 (t)2 + y1 (t)2 + z1 (t)2 dt
T0 dt T0

ds
Note that the velocity v(t) = r1 (t) is a vector quantity while the speed dt ( t ) = |r1 (t)| is
a scalar quantity.
Example 1.1.5

6
C URVES 1.1 D ERIVATIVES , V ELOCITY, E TC .

In general it can be quite difficult to compute arc lengths. So, as an easy warmup example,
we will compute the circumference of the circle x2 + y2 = a2 . We’ll also find a unit tangent
to the circle at any point on the circle. We’ll use the parametrization

0 ď θ ď 2π


r(θ ) = a cos θ , a sin θ

of Example 1.0.1. Using Lemma 1.1.3, but with the parameter t renamed to θ

r1 (θ ) = a ´ sin θ , cos θ

r1 ( θ )
= ´ sin θ , cos θ


T̂(θ ) = 1
|r (θ )|
ds ˇ ˇ
( θ ) = ˇr1 ( θ )ˇ = a
dθ
żΘ
ˇ 1 ˇ
s ( Θ ) ´ s (0) = ˇr (θ )ˇ dθ = aΘ
0

As2 s(Θ) is the arc length of the part of the circle with 0 ď θ ď Θ, the circumference of the
whole circle is
s(2π ) = 2πa
which is reassuring, since this formula has been known3 for thousands of years. The

y T̂pθq
` ˘
a cos θ , a sin θ
x2 ` y 2 “ a2

θ
x

formula s(Θ) ´ s(0) = aΘ also makes sense — the part of the circle with 0 ď θ ď Θ is the
Θ Θ
fraction 2π of the whole circle, and so should have length 2π ˆ 2πa. Also note that

r(θ ) ¨ T̂(θ ) = a cos θ , a sin θ ¨ ´ sin θ , cos θ = 0

so that the tangent to the circle at any point is perpendicular to the radius vector of the
circle at that point. This is another geometric fact that has been known4 for thousands of
years.

2 You might guess that Θ is a capital Greek theta. You’d be right.

3 The earliest known written approximations of π, in Egypt and Babylon, date from 1900–1600BC. The
first recorded algorithm for rigorously evaluating π was developed by Archimedes around 250 BC. The
first use of the symbol π, for the ratio between the circumference of a circle and its diameter, in print
was in 1706 by William Jones.
4 It is Proposition 18 in Book 3 of Euclid’s Elements. It was published around 300BC.

7
C URVES 1.1 D ERIVATIVES , V ELOCITY, E TC .

Example 1.1.5

Example 1.1.6

Consider the curve

r(t) = 6 sin(2t)ı̂ı + 6 cos(2t)̂ + 5tk̂
where the standard basis vectors ı̂ı = (1, 0, 0), ̂ = (0, 1, 0) and k̂ = (0, 0, 1). We’ll first
sketch it, by observing that
˝ x (t) = 6 sin(2t) and y(t) = 6 cos(2t) obey x (t)2 + y(t)2 = 36 sin2 (2t) + 36 cos2 (2t) =
2 2

lie on the cylinder x + y = 36 and
36. So all points of the curve
˝ as t increases, x (t), y(t) runs clockwise around the circle x2 + y2 = 36 and at the
same time z(t) = 5t just increases linearly.
Our curve is the helix

t“π
π
t“ 2

y
t“0

We have marked three points of the curve on the above sketch. The first has t = 0 and is
0ı̂ı + 6̂ + 0k̂. The second has t = π2 and is 0ı̂ı ´ 6̂ + 5π
2 k̂, and the third has t = π and is
0ı̂ı + 6̂ + 5π k̂. We’ll now use Lemma 1.1.3 to find a unit tangent T̂(t) to the curve at r(t)
and also the arclength of the part of curve between t = 0 and t = π.

r(t) = 6 sin(2t)ı̂ı + 6 cos(2t)̂ + 5tk̂

r1 (t) = 12 cos(2t)ı̂ı ´ 12 sin(2t)̂ + 5k̂
ds ˇ 1 ˇ b a
(t) = ˇr (t)ˇ = 122 cos2 (2t) + 122 sin2 (2t) + 52 = 122 + 52
dt
= 13
r1 ( t ) 12 12 5
T̂(t) = 1 = cos(2t)ı̂ı ´ sin(2t)̂ + k̂
|r (t))| 13 13 13
żπ
ˇ 1 ˇ
s ( π ) ´ s (0) = ˇr (t)ˇ dt = 13π
0

8
C URVES 1.1 D ERIVATIVES , V ELOCITY, E TC .

Example 1.1.6

Example 1.1.7

In this example, we find the perimeter of the astroid5

x2/3 + y2/3 = a2/3
A geometric construction of this curve, as well as a derivation of its equation is given in
the optional section 1.11 later. We’ll start by finding a convenient parametrization.
˝ To do so, notice that x2/3 + y2/3 = a2/3 looks somewhat like the equation of the circle
x 2 + y2 = a2 .
˝ The standard parametrization of the circle, namely x = a cos t, y = a sin t works
because of the elementary trig identity cos2 t + sin2 t = 1.
˝ If we can arrange that x (t)2/3 = a2/3 cos2 t and y(t)2/3 = a2/3 sin2 t, then the same
elementary trig identity will give x (t)2/3 + y(t)2/3 = a2/3 , as desired.
˝ But of course its easy to arrange that: just solve x (t)2/3 = a2/3 cos2 t for x (t), namely
x (t) = a cos3 t, and solve y(t)2/3 = a2/3 sin2 t for y(t), namely y(t) = a sin3 t.
Our parametrization is
r(t) = a cos3 tı̂ı + a sin3 t̂
By Lemma 1.1.3
r(t) = a cos3 tı̂ı + a sin3 t̂
r1 (t) = ´3a sin t cos2 tı̂ı + 3a sin2 t cos t̂
ds ˇ ˇ a
(t) = ˇr1 (t)ˇ = 9a2 sin2 t cos4 t + 9a2 sin4 t cos2 t
dt b
= 3a sin2 t cos2 t(cos2 t + sin2 t)
ˇ ˇ
= 3aˇ sin t cos tˇ
r1 ( t ) sin t cos t
´ cos tı̂ı + sin t̂


T̂(t) = =
|r (t))|
1 | sin t cos t|
= sgn sin t cos t ´ cos tı̂ı + sin t̂

Here sgn sin t cos t means “the sign of sin t cos t”, i.e +1 when sin t cos t ą 0 and ´1

when sin t cos t ă 0. So

# +
1 if sin t ą 0, cos t ą 0 or sin t ă 0, cos t ă 0
´ cos tı̂ı + sin t̂


T̂(t) =
´1 if sin t ą 0, cos t ă 0 or sin t ă 0, cos t ą 0
# +
1 if 0 ă t ă π2 or π ă t ă 3π 2 ´


= cos tı
ı̂ + sin t 
̂
´1 if π2 ă t ă π or 3π 2 ă t ă 2π

Before we go on to sketch the astroid and compute its perimeter, we can make a few
observations that will simplify our lives.

5 Astroid should not be confused with asteroid, though both words derive from the Greek word for star.

9
C URVES 1.1 D ERIVATIVES , V ELOCITY, E TC .

˝ The signs of both components of r(t) are the same as the signs of the components of
cos t ı̂ı + sin t ̂ ; and the signs of both components of r1 (t) are the same as the signs of
the components of ´ sin t ı̂ı + cos t ̂ . Consequently the astroid looks somewhat like a
circle in that
– when 0 ď t ď π2 , r(t) lies in the first quadrant and moves upward and to the
left as t increases and
– when π2 ď t ď π, r(t) lies in the second quadrant and moves downward and to
the left as t increases and
– when π ď t ď 3π 2 , r ( t ) lies in the third quadrant and moves downward and to
the right as t increases and
– when 3π 2 ď t ď 2π, r ( t ) lies in the fourth quadrant and moves upward and to
the right as t increases and
– r(2π ) = r(0) so that the astroid is a closed curve that circumnavigates the origin
exactly once as t runs from 0 to 2π.
˝ Something weird happens at those values of t where sin t cos t changes sign6 , i.e. at
t = 0, π2 , π, 3π
2 , etc. Namely T̂ ( t ) flips. To be precise

lim T̂ (t) = ı̂ı lim T̂ (t) = ´ı̂ı

tÑ0´ tÑ0+
lim T̂ (t) = ̂ lim T̂ (t) = ´̂
tÑπ/2´ tÑπ/2+

and so on. This signals cusps in the curve at t = 0, i.e. at r(0) = aı̂ı , and at t = π2 ,
i.e. at r( π2 ) = a̂ , and so on. So while the astroid looks somewhat like a circle, it has
cusps at ˘aı̂ı and ˘a̂ . Here is the sketch.

y
t = π/2
x2/3 + y 2/3 = a2/3

t=0
t=π x

t = 3π/2

˝ The astroid is invariant under reflections in the x-axis and in the y-axis. That is,
x2/3 + y2/3 = a2/3 is invariant under x Ñ ´x and also under y Ñ ´y. So to find the
whole perimeter, it suffices to find the arc length of the part of the astroid in the first
quadrant, and then multiply by 4.
ż π/2 ż π/2 ż π/2
ds
perimeter = 4 dt = 4 3a sin t cos t dt = 6a sin(2t) dt
0 dt 0 0
h cos(2t) iπ/2
= 6a ´ = 6a
2 0

6 Like a cross-walk sign.

10
C URVES 1.1 D ERIVATIVES , V ELOCITY, E TC .

Example 1.1.7

Example 1.1.8

In the last example, we found that the astroid had cusps at those points r(t) where the ve-
locity r1 (t) vanished. In this example, we will explore a little further what can happen
when r1 (t) = 0.
Suppose that you are out for a walk and that your position at time t is r(t). If at some
time you have nonzero velocity, it is very hard for you to change your direction of motion
discontinuously7 . On the other hand, when r1 (t) = 0, you are not moving at all and it is
easy for you to turn and leave in any direction you choose. You could reverse direction
completely, or make a sharp left turn, or not change direction at all. Here are examples of
all of these. They all have r1 (t) = 0. They are sketched below.

r1 ( t ) = ( t5 , t2 ) r11 (t) = (5t4 , 2t)

# + # +
( t2 , 0) if t ď 0 1 (2t, 0) if t ď 0
r2 ( t ) = r2 ( t ) =
(0, t2 ) if t ě 0 (0, 2t) if t ě 0

r3 ( t ) = ( t3 , 0) r13 (t) = (3t2 , 0)

r1 ptq
r2 ptq
r3 ptq
t“0 t“0 t“0

Example 1.1.8

Example 1.1.9

We’ll find the arc length of

?
r(t) = t cos t ı̂ı + t sin t ̂ + t k̂ 0ďtď 2

We’ll first sketch it, by observing that

˝ x (t) = t cos t, y(t) = t sin t and z(t) = t obey x (t)2 + y(t)2 = t2 = z(t)2 . So all points
2 2 2
of the curve lie on the cone
x + y = z and
˝ as t increases, x (t), y(t) runs counterclockwise around a “circle whose radius in-
creases linearly with t and at the same time z(t) also increases linearly.

Our curve is the “corkscrew”

7 For your velocity to jump discontinuously, your acceleration has to be infinite, which requires an infinite
force. You might not look so healthy afterwards

11
C URVES 1.1 D ERIVATIVES , V ELOCITY, E TC .

z
y “ z, x “ 0

By Lemma 1.1.3

r(t) = t cos t ı̂ı + t sin t ̂ + t k̂

r1 (t) = [cos t ´ t sin t]ı̂ı + [sin t + t cos t]̂ + k̂
ds ˇ ˇ
( t ) = ˇr1 ( t )ˇ
dt b
cos2 t ´ 2t sin t cos t + t2 sin2 t + sin2 t + 2t sin t cos t + t2 cos2 t + 1



=
a
= 2 + t2

Our goal, stated at the beginning of this example, was to compute

ż ? ż ?2 a
? 2ˇ ˇ
s ( 2) ´ s (0) = ˇr1 (t)ˇ dt = 2 + t2 dt
0 0

To evaluate the integral, we’ll use three ?

techniques that you learned in your first integral
calculus course. First, motivated by the 2 + t2 , we’ll use the trigonometric substitution
? ?
dt = 2 sec2 u du 2 + t2 = 2 1 + tan2 u = 2 sec2 u
 
t = 2 tan u
?
When t = 0, u = 0 and when t = 2, tan u = 1 so that u = π4 and

? ż π/4 a ? ż π/4
2
s ( 2) ´ s (0) = 2
2 sec u 2 sec u du = 2 sec3 u du
0 0

You may have evaluated this integral in first year. There are several ways of doing so.
Perhaps the most straight forward, but also most tedious, method is to rewrite the integral
as ż π/4
? cos u
s ( 2) ´ s (0) = 2 du
0 cos4 u
We recognize that this is a trigonometric integral that contains an odd power of cos u, so
we substitute w = sin u, dw = cos u du, cos2 u = 1 ´ w2 . When u = 0, w = 0 and when
u = π4 , w = ?12 so that
? ż 1/?2
dw
s ( 2) ´ s (0) = 2
0 (1 ´ w2 )2

12
 C URVES 1.2 R EPARAMETRIZATION

The integrand is now a rational function, i.e. a ratio of polynomials. So we apply partial
fractions.
? ż 1/?2
dw
s ( 2) ´ s (0) = 2
0 [(1 ´ w)(1 + w)]2
ż ?
1 1/ 2 h 1 1 i2
= + dw
2 0 1´w 1+w
ż ?
1 1/ 2 h 1 2 1 i
= + + dw
2 0 (1 ´ w)2 (1 ´ w)(1 + w) (1 + w)2
ż ?
1 1/ 2 h 1 1 1 1 i
= + + + dw
2 0 (1 ´ w )2 1 ´ w 1 + w (1 + w )2
?
1h 1 1 i1/ 2
= ´ ln |1 ´ w| + ln |1 + w| ´
2 1´w ?
1+w ?0
1 h 2w 1 + w i1/ 2 1h ? 2 + 1i
= + ln = 2 2 + ln ? « 2.2956
2 1 ´ w2 1´w 0 2 2´1
Ooof!
Example 1.1.9

1.2IJ Reparametrization
There are invariably many ways to parametrize a given curve. Kind of trivially, one
can always replace t by, for example, 3u. But there are also more substantial ways to
reparametrize curves. It often pays to tailor the parametrization used to the application of
interest. For example, we shall see in the next couple of sections that many curve formulae
simplify a lot when arc length is used as the parameter.
Example 1.2.1

Here are three different parametrizations of the semi-circle x2 + y2 = r2 , y ě 0.

˝ The first uses the polar angle θ as the parameter. We have already seen, in Example
1.0.1, the parametrization
y
` ˘
r cos θ , r sin θ
x2 ` y 2 “ r 2 0ďθďπ


r1 (θ ) = r cos θ , r sin θ
θ
x
˝ The second uses x as the parameter.
? Just solving x2 + y2 = r2 , y ě 0 for y as a
function of x, gives y( x ) = r2 ´ x2 and so gives the parametrization
y
` ? ˘
x , r 2 ´ x2 a
x2 ` y 2 “ r 2 x r2 ( x ) = x , r 2 ´ x 2 ´r ď x ď r

x
13
C URVES 1.2 R EPARAMETRIZATION

˝ The third uses arc length from (r, 0) as the parameter. We have seen, in Example
1.1.5, that the arc length from (r, 0) to r2 (θ ) is just s = rθ. So the point on the semi-
circle that is arc length s away from (r, 0) is
y
` ˘
r cos sr , r sin sr s  s s
x2 ` y 2 “ r 2 r3 ( s ) = r2 = r cos , r sin
r r r
s
x
with 0 ď s ď πr.
Example 1.2.1

We shall see that, for some purposes, it is convenient to use parametrization by arc
length. Here is a messier example in which we reparametrize a curve so as to use the arc
length as the parameter.
Example 1.2.2

We saw in Example 1.1.7, that, as t runs from 0 to π2 , r(t) = a cos3 t ı̂ı + a sin3 t ̂ runs
from ( a, 0) to (0, a) along the astroid x2/3 + y2/3 = a2/3 . Suppose that we want a new
parametrization R(s) chosen so that, as s runs from 0 to some appropriate value, R(s)
runs from ( a, 0) to (0, a) along x2/3 + y2/3 = a2/3 , with s being the arc length from ( a, 0)
to R(s) along x2/3 + y2/3 = a2/3 .

Rpsq
rptq
s

We saw, in Example 1.1.7, that, for 0 ď t ď π2 , ds 3a

dt = 2 sin(2t ) so that the arclength from
( a, 0) = r(0) to r(t) is
żt
3a 3a 
sin(2t1 ) dt1 = 1 ´ cos(2t)

s(t) =
0 2 4
which runs from 0, at t = 0, to 3a π
2 , at t = 2 . This is relatively clean and we can invert s ( t )
to find t as a function of s. The value, T (s), of t that corresponds to any given 0 ď s ď 3a 3
is determined by
3a  1  4s 
1 ´ cos 2T (s) ðñ T (s) = arccos 1 ´


s=
4 2 3a

14
 C URVES 1.3 C URVATURE

and
R(s) = r T (s) = a cos3 T (s) ı̂ı + a sin3 T (s) ̂

We can simplify cos3 T (s) and sin3 T (s) by just using trig identities to convert the

cos 2T (s) in s = 3a
4 1 ´ cos 2T ( s )





into cos T (s) ’s and sin T (s) ’s.

3a  3a  2s
1 ´ cos 2T (s) = 1 ´ 2 cos2 T (s) ´ 1 ðñ cos2 T (s) = 1 ´




s=
4 4 3a
3a  3a  2s
1 ´ cos 2T (s) = 1 ´ 1 ´ 2 sin2 T (s) ðñ sin2 T (s) =




s=
4 4 3a
Consequently the desired parametrization is
h 2s i3/2 h 2s i3/2 3a
R(s) = a 1 ´ ı̂ı + a ̂ 0ďsď
3a 3a 2
which is remarkably simple.
Example 1.2.2

1.3IJ Curvature
So far, when we have wanted to approximate a complicated curve by a simple curve near
some point, we drew the tangent line to the curve at the point. That’s pretty crude. In
particular tangent lines are straight — they don’t curve. We will get a much better idea
of what the complicated curve looks like if we approximate it, locally, by a very simple
“curvy curve” rather than by a straight line. Probably the simplest “curvy curve” is a
circle8 and that’s what we’ll use.
Definition 1.3.1.

(a) The circle which best approximates a given curve near a given point is called
the circle of curvature or the osculating circle9 at the point.

(b) The radius of the circle of curvature is called the radius of curvature at the
point and is normally denoted ρ.

(c) The curvature at the point is κ = 1/ρ.

(d) The centre of the circle of curvature is called centre of curvature at the point.

These definitions are illustrated in the figure below. It shows (part of) the osculating circle
at the point P. The point C is the centre of curvature.

8 Circles are good for studying “curvature”, because, unlike parabolas for example, the rate at which a
circle curves is uniform over the entire circle.
9 “Osculare” is the Latin verb “to kiss”. The German mathematician Gottfried Wilhelm (von) Leibniz
(1646–1716) named the circle the “circulus osculans”.

15
C URVES 1.3 C URVATURE

Note that when the curvature κ is large, the radius of curvature ρ is small and we
have a very curvy curve. On the other hand when the curvature κ is small, the radius
of curvature ρ is large and our curve is almost straight. In particular, straight lines have
curvature exactly zero.
We are now going to determine how to find the circle of curvature, starting by figuring
out what its radius should be. We’ll first look at curves10 that lie in the xy-plane and then
move on to curves in 3d. Consider the black curve in the figure below.

ds

ρ dθ

That figure also contains a (portion of a) red circle that fits the curve really well between
the two radial lines that are (a very small) angle dθ apart. So the arclength ds of the part
of the black curve between the two radial lines, should be (essentially) the same as the
arc length of the circle between the two radial lines, which is ρ |dθ|, where ρ is the radius
of the circle. (We put in absolute values to take into account the possibility that dθ could
be negative.) Thus ds = ρ |dθ|. When dθ is a macroscopic angle, this is of course an
approximation. But in the limit as dθ Ñ 0, we should end up with
ˇ ds ˇ
ˇ ˇ
ρ=ˇ ˇ
dθ
We now have a formula for the radius of curvature, but not in a very convenient form,
because to evaluate it we would need to know the arc length along the curve as a function
of the angle θ in the rightmost figure below. We’ll now spend some time developing more
convenient formulae for ρ. First consider the three figures below. They all show the same
curve as in the last figure. The leftmost figure just shows
˝ the curve of interest, which is the black curve, and
˝ the (blue) point of interest on the black curve. We want to find the curvature at that
point.

10 We’ll also assume that the curves of interest are smooth, with no cusps for example, and not straight,
so that the radius of curvature 0 ă ρ ă 8.

16
C URVES 1.3 C URVATURE

The middle figure shows the same curve and point of interest and also shows
˝ the red circle of curvature (i.e. best fitting circle) for the black curve at the blue dot.
˝ The red dot is the centre of curvature.
The rightmost figure shows the same black curve, blue point of interest and red circle of
curvature (at least part of it) somewhat enlarged.
˝ The angle θ is the angle between ı̂ı and the radius vector from the red dot (the centre
of curvature) to the blue dot (the point of interest).
˝ T̂ is the tangent vector to the black curve at the blue dot.
˝ The angle φ is the angle between ı̂ı and T̂. The vector T̂ is also tangent to the red circle.
As the tangent and radius vectors for ˇ circles
ˇ are perpendicular to each other11 , we
ˇ ds ˇ
have that φ = θ + π2 and hence ρ = ˇ dφ ˇ too.

θ
φ ı̂ı
T̂

ı̂ı

We are now in a position to develop a bunch of formulae for the radius of curvature ρ and
ˇ ˇ´1
1 ˇ ds ˇ
the curvature κ = ρ , that are more convenient than κ = ˇ dφ ˇ . These formulae will use
the
Notation 1.3.2.

If r(t) is a parametrized curve, then

dr
• v(t) = dt ( t ) is the velocity vector at r(t)
d2 r
• a(t) = dt2
(t) is the acceleration vector at r(t)

• T̂(t) is the unit tangent vector to the curve at r(t) that points in the direction
of increasing t.

• N̂(t) is the unit normal vector to the curve at r(t) that points toward the
centre of curvature.

• κ (t) is the curvature at r(t)

• ρ(t) is the radius of curvature at r(t)

11 We saw that in Example 1.1.5.

17
C URVES 1.3 C URVATURE

Theorem 1.3.3.

(a) Given12 s(φ), i.e. if we know the arc length along the curve as a function of
the angle13 φ = >(ı̂ı , T̂), then
ˇ ds ˇ ˇ ds ˇ´1 ˇ dφ ˇ
ˇ ˇ ˇ ˇ ˇ ˇ
ρ=ˇ ˇ κ=ˇ ˇ κ=ˇ ˇ
dφ dφ ds

(b) Given r(s), i.e. if we have a parametrization of the curve in terms of arc
length, then
dT̂
(s) = κ (s) N̂(s)
ds
where N̂(s) is the unit normal vector to the curve at r(s) that points toward
the centre of curvature.

(c) Given r(t), i.e. if we have a general parametrized curve, then

dT̂ ds ds d2 s  ds 2
= κ N̂ v(t) = (t) T̂(t) a(t) = T̂ + κ N̂
dt dt dt dt2 dt


(d) Given x (t) , y(t) , (for curves in the xy-plane)
ˇ ˇ ˇ dx d2 y dy d2 x ˇ
ˇ v(t) ˆ a(t) ˇ ˇ ˇ
ˇ ˇ dt dt2 ´ dt dt2
κ=ˇ ˇ =
ˇ ds 3 ˇ  dx
2
2 3/2


dt dt + dy
dt

(e) Given y( x ), (again for curves in the xy-plane)

ˇ d2 y ˇ
ˇ 2ˇ
dx
κ= 
dy
2 3/2
1+ dx

Proof. (a) Given s(φ), then

ˇ ds ˇ ˇ ds ˇ´1
ˇ ˇ ˇ ˇ
ρ=ˇ ˇ κ=ˇ ˇ
dφ dφ
ˇ ˇ
ˇ ds ˇ
As we are assuming that 0 ă ρ = ˇ dφ ˇ ă 8, the inverse function theorem says that we can
invert the function s(φ) (at least locally) to get φ as a function of s, and that
ˇ dφ ˇ
ˇ ˇ
κ=ˇ ˇ
ds

12 The equation s = s(φ) is called the “intrinsic equation of the curve”.

13 The notation >(ı̂ı , T̂) means “the angle between ı̂ı and T̂”.

18
C URVES 1.3 C URVATURE

(b) Given r(s), then, by Lemma 1.1.3.c, T̂(s) = r1 (s) is a unit tangent to the curve at r(s)
and
dT̂ dT̂ dφ
= (˚)
ds dφ ds
dφ
Now up to a sign ds is κ, and just because φ = >(ı̂ı , T̂), with T̂ a unit vector,

T̂ = cos φ ı̂ı + sin φ ̂

dT̂ (˚˚)
ùñ = ´ sin φ ı̂ı + cos φ ̂
dφ
dT̂
So dφ is a unit vector that is perpendicular14 to T̂, and hence to the curve at r(s), and

dT̂
(s) = κ (s) N̂(s) (:)
ds
with N̂(s) a unit normal vector to the curve at r(s). In fact, N̂(s) is the unit normal vector
to the curve at r(s) that points toward the centre of curvature.
To see that, look at the figures below15 , and note that substituting the sign information
from each figure into (˚) gives (:). For example, consider the figure on the lower left. In

dT̂
dφ T̂
dT̂
dφ T̂ φ

dφ
cos φ ą 0, ds
ą0 cos φ ą 0, dφ
ds
ă0

φ
φ
dT̂
T̂ dφ
dT̂
T̂ dφ
dφ dφ
cos φ ă 0, ds
ă0 cos φ ă 0, ds
ą0

that figure,
˝ the x component of T̂ is negative (T̂ is leftward pointing in the figure),

14 Think about why this should be the case. In particular, sketch T̂ and φ and think about what the sketch
says about ddφT̂ .
15 In each of the four figures, the arrow on the curve specifies the direction of increasing arc length s and
the red dot is the centre of curvature for the curve at the blue dot.

19
C URVES 1.3 C URVATURE

˝ which makes cos φ negative (see (˚˚)),

dT̂
˝ which makes the y component of dφ negative (see (˚˚) again),
dT̂
˝ so dφ is downward pointing,

so ddφT̂ = ´N̂ (the centre of curvature is the red dot above the curve) and
˝ as s increases (i.e. as you move in the direction of the arrow on the curve), φ de-
creases (on the far right hand part of the curve φ « π2 , while on the far left hand part
dφ ˇ dφ ˇ dφ
of the curve φ « 0), so ds ă 0 and κ = ˇ ds ˇ = ´ ds .
dT̂ dT̂ dφ
˝ So by (˚), ds = dφ ds = ´ N̂)(´κ ) = κ N̂.

In each of the three other figures we also end up with ddsT̂ = κ (s)N̂(s).
Note that if κ (s) = 0, then N̂(s) is not defined. This makes sense: if the curve is (locally) a
straight line, there is no “best fitting circle”.

(c) Given r(t), i.e. if we have a general parametrized curve, we can determine a unit
tangent vector by using Lemma 1.1.3:
dr ds r1 ( t )
v(t) = (t) = (t) T̂(t) ùñ T̂(t) = 1
dt dt |r (t)|

Then we can determine κ and N̂ by differentiating T̂(t) and using the chain rule:

dT̂ dT̂ ds ds |T̂1 (t)|

= = κ N̂ ùñ κ (t) =
dt ds dt dt |r1 (t)|
ds
Also, if we differentiate v(t) = dt T̂ ( t ), we get that the acceleration

d2 r d2 s ds dT̂ d2 s  ds 2
a(t) = 2 = 2 T̂ + = 2 T̂ + κ N̂
dt dt dt dt dt dt


(d) Given x (t) , y(t) , (for curves in the xy-plane), we can read off the curvature from
 ds   d2 s  ds 2 
v(t) ˆ a(t) = (t) T̂(t) ˆ T̂ + κ N̂
dt dt2 dt
 ds 3
=κ T̂ ˆ N̂ (since T̂ ˆ T̂ = 0)
dt
Think of T̂ and N̂ as 3d vectors that whose z-components happen to be zero. As T̂ and N̂
are mutually perpendicular unit vectors in the xy-plane, the cross-product T̂ ˆ N̂ will be
ˇ ˇ ˇ3
either +k̂ or ´k̂. In both cases, |v(t) ˆ a(t)ˇ = κ ˇ ds ˇ
dt . So
ˇ ˇ ˇ ˇ ˇ ˇ
ˇ v(t) ˆ a(t) ˇ ˇˇ  dx ı̂ı + dy ̂  ˆ  d2 x ı̂ı + d2 y ̂  ˇˇ ˇˇ  dx d2 y ´ dy d2 x k̂ ˇˇ
ˇ ˇ ˇ dt dt2 dt2 ˇ 2 dt dt2
κ=ˇ 3 ˇ=ˇ
dt
3 ˇ = ˇˇ dt dt 3
ˇ
ˇ
ˇ ds

ˇ ˇ ds


ˇ ˇ ds


ˇ
dt dt dt
ˇ dx d2 y dy d2 x ˇ
ˇ 2 ´ dt dt2
ˇ
=  dt
dt
2 3/2
dx 2
dt + dydt

20
C URVES 1.3 C URVATURE

(e) Given y( x ), again for curves in the xy-plane, we can parametrize the curve using x as
the parameter:


r(t) = X (t) , Y (t) with X (t) = t and Y (t) = y(t)
Then
dX d2 X dY dy d2 Y d2 y
=1 =0 = =
dt dt2 dt dx dt2 dx2
and ˇ dX d2 Y dY d2 X ˇ ˇ d2 y ˇ
ˇ 2 ´ 2
ˇ ˇ 2ˇ
κ =  dt
dt dt dt
3/2
= dx
dX 2 2 dy
2 3/2
+ dY

 
dt dt 1 + dx

Take another look at Theorem 1.3.3.c and note that

2
˝ the tangential component of acceleration, i.e. ddt2s , arises purely from change in speed
while
2
˝ the normal component of acceleration, i.e. κ ds dt , arises from curvature and is pro-
ds
portional to the square of the speed dt . Think about what you feel when you are
driving. That’s why velodromes and (car) race tracks often have banked corners.

Example 1.3.4

As a warm up example, and also a check that our formulae make sense, we’ll find the cur-
vature κ, radius of curvature, ρ, unit tangent vector, T̂, unit normal vector, N̂, and centre
of curvature of the parametrized curve
r(t) = a cos t ı̂ı + a sin t ̂
with the constant a ą 0. This is, of course, the circle of radius a centred on the origin. As
dr ds
v(t) = (t) = ´a sin t ı̂ı + a cos t ̂ ùñ (t) = |v(t)| = a
dt dt
we have that the unit tangent vector
v(t)
T(t) = = ´ sin t ı̂ı + cos t ̂
|v(t)|
Note, as a check, that this is indeed a vector of length one and is perpendicular to the
radius vector (as expected — the curve is a circle). As

dT̂
(t) = ´ cos t ı̂ı ´ sin t ̂
dt
we have that
ˇ dT̂ ˇ
dT̂ ˇ ( t )ˇ
dt ( t ) dt 1 1
N̂(t) = ˇ dT̂ ˇ = ´ cos t ı̂ı ´ sin t ̂ κ (t) = ds
= ρ(t) = =a
ˇ ( t )ˇ dt ( t )
a κ (t)
dt

Now look at the figure.

21
C URVES 1.3 C URVATURE

T̂
rptq

N̂

To get to the centre of curvature we should start from r(t) and walk a distance ρ(t), which
after all is the radius of curvature, in the direction N̂( T ), which is pointing towards the
centre of curvature. So the centre of curvature is
r(t) + ρ(t)N̂(t) = a cos t ı̂ı + a sin t ̂ + a ´ cos t ı̂ı ´ sin t ̂ = 0
   

This makes perfectly good sense — the radius of curvature is the radius of the original
circle and the centre of curvature is the centre of the original circle.
One alternative calculation of the curvature, using x (t) = a cos t, y(t) = a sin t, is
ˇ dx d2 y ˇ
ˇ (t) 2 (t) ´ dy (t) d2 2x (t)ˇ
κ (t) = hdt dt dt dt
2 3/2
i
dx

2 dy

dt ( t ) + dt ( t )
ˇ
ˇ ´ a sin t ´ a sin t ´ a cos t ´ a cos t ˇ


ˇ
=
2
2 3/2
´ a sin t + a cos t


1
=
a
?
Another alternative calculation of the curvature, using y( x ) = a2 ´ x2 (for the part of
the circle with y ą 0),
x x y( x ) ´ xy1 ( x ) y ( x )2 + x 2 a2
y1 ( x ) = ´ ? =´ y2 ( x ) = ´ = ´ = ´
a ´x
2 2 y( x ) y ( x )2 y ( x )3 y ( x )3
is
ˇ d2 y ˇ a2
ˇ 2 ( x )ˇ y ( x )3 a2 1
dx
κ (x) = h =h = 3/2 = a
dy
2 i3/2 x2
i3/2
y ( x )2 + x 2
1+ dx ( x ) 1+ y ( x )2

Example 1.3.4

Example 1.3.5

As a more computationally involved example, we’ll analyze

r(t) = cos t + t sin t ı̂ı + sin t ´ t cos t ̂ tą0

v(t) = t cos t ı̂ı + t sin t ̂

a(t) = cos t ´ t sin t ı̂ı + sin t + t cos t ̂

22
 C URVES 1.4 C URVES IN T HREE D IMENSIONS

We can read off from v(t), that

ds
(t) = |v(t)| = t
dt
d2 s
(t) = 1
dt2
v(t)
T(t) = = cos t ı̂ı + sin t ̂
|v(t)|

Next, from a(t), we read off that

a(t) = cos t ´ t sin t ı̂ı + sin t + t cos t ̂

and
d2 s ds
2
a(t) = 2 (t) T̂(t) + κ (t) (t) N̂(t) (by Theorem 1.3.3.c)
dt dt
= cos t ı̂ı + sin t ̂ + t2 κ (t)N̂(t)
ùñ t2 κ (t)N̂(t) = ´t sin t ı̂ı + t cos t ̂

so that
1
κ (t) = and N̂(t) = ´ sin t ı̂ı + cos t ̂
t
As an alternative calculation of the curvature, we have

|v(t) ˆ a(t)|
κ (t) =
( ds (t))3
ˇ dt
ˇ t cos t ı̂ı + t sin t ̂ ˆ cos t ´ t sin t ı̂ı + sin t + t cos t ̂ ˇ
 

ˇ
=
( ds
dt ( t ))
3
ˇ
 ˇ
ˇ t cos t sin t + t cos t ´ t sin t cos t ´ t sin t k̂ˇ


=
( ds
dt ( t ))
3

|t2 k̂| 1
= 3
=
t t
It pays to think before you calculate!
Example 1.3.5

1.4IJ Curves in Three Dimensions

So far, we have developed formulae for the curvature, unit tangent vector, etc., at a point
r(t) on a curve that lies in the xy-plane. We now extend our discussion to curves in R3 .
Fix any t. For t1 very close to t, r(t1 ), will, by the Taylor expansion to second order, be very
close to r(t) + r1 (t) (t1 ´ t) + 12 r2 (t) (t1 ´ t)2 , so that r(t1 ) almost lies in the plane through
r(t) that is determined by the two vectors r1 (t) and r2 (t). Thus, if we restrict our attention
to a very small part of the curve near the point of interest r(t), the curve will, to a very

23
C URVES 1.4 C URVES IN T HREE D IMENSIONS

good approximation lie in some plane. So we can still define, for example, the osculating
circle to the curve at r(t) to be the circle in that plane that fits the curve best near r(t). And
we still have the formulae

dr ds
v= = T̂
dt dt
dT̂
= κ N̂
ds
dT̂ ds
= κ N̂
dt dt
d2 r d2 s  ds 2
a = 2 = 2 T̂ + κ N̂
dt dt dt
 ds 3
vˆa = κ T̂ ˆ N̂
dt

The only16 difference is that v, a, T̂ and N̂ are now three component vectors rather than
two component vectors.
If we are lucky and our curve happens to lie completely in a single plane, the vectors
T̂(s) and N̂(s) are mutually perpendicular unit vectors that lie in the same plane, so that
their cross product B̂(s) = T̂(s) ˆ N̂(s) is a unit vector that is perpendicular to the plane.
By continuity, B̂(s) has to be a constant vector, i.e. be independent of s.
If, on the other hand, B̂(s) is not constant, then our curve doesn’t lie in a single plane,
and we can use the derivative

dB̂ d dT̂ dN̂

T̂ ˆ N̂ = ˆ N̂ + T̂ ˆ


=
ds ds ds ds
dN̂  dT̂ 
= T̂ ˆ since is parallel to N̂
ds ds

as a measure

˝ of how badly the curve fails to lie in a plane,

˝ i.e. how much the plane that fits the curve best near r(s) twists as s increases,

The cross product in ddsB̂ = T̂ ˆ ddsN̂ implies that dB̂

ds is perpendicular to T̂. In addition, dB̂
ds
must be perpendicular to B̂ because

d  dB̂
|B̂| = 1 ùñ 1 = B̂ ¨ B̂ ùñ 0 = B̂ ¨ B̂ = 2B̂ ¨

ds ds

dB̂
So ds ( s ) must be parallel to N̂(s).

16 However this can be a significant difference.

24
C URVES 1.4 C URVES IN T HREE D IMENSIONS

Definition 1.4.1.

(a) The binormal vector at r(s) is B̂(s) = T̂(s) ˆ N̂(s). The normal vector N̂(s)
is sometimes called the unit principal normal vector to distinguish it from the
binormal vector.

(b) We define the torsion τ (s) by

dB̂
(s) = ´τ (s)N̂(s)
ds
The negative sign is included so that τ (s) ą 0 indicates “right handed twist-
ing”. There will be an explanation of what this means in Example 1.4.4 below.

(c) The osculating plane at r(s) (the plane that fits the curve best at r(s)) is the
plane through r(s) with normal vector B̂(s). The equation of the plane is
(
B̂(s) ¨ ( x, y, z) ´ r(s) = 0

For each s, T̂(s), N̂(s) and B̂(s) are mutually perpendicular unit vectors. They form an
orthonormal basis for R3 , just as ı̂ı , ̂ and k̂ form an orthonormal basis for R3 . Furthermore
both (T̂(s) , N̂(s) , B̂(s)) and (ı̂ı , ̂ , k̂) are “right handed triples17 ”, meaning that B̂(s) =
T̂(s) ˆ N̂(s) and k̂ = ı̂ı ˆ ̂ .

k̂ B̂

̂
N̂
ı̂ı T̂

dT̂ dB̂
We have already computed ds and ds . It is now an easy matter to compute
dN̂ d
B̂(s) ˆ T̂(s)


=
ds ds
= ´τ (s)N̂(s) ˆ T̂(s) + B̂(s) ˆ κ (s)N̂(s)

= τ (s)B̂(s) ´ κ (s)T̂(s)
Now suppose that we have a curve that is parametrized by t rather than s. How do we
find the torsion τ? The most obvious method is to

3 ds 3
• recall that v ˆ a = κ ds
dt T̂ ˆ N̂ = κ dt B̂ and that B ( t ) is a unit vector. So

v(t) ˆ a(t)
B̂(t) =
|v(t) ˆ a(t)|

17 We shall stick to “right handed triples” to make it easier to get various signs right.

25
C URVES 1.4 C URVES IN T HREE D IMENSIONS

dB̂
• Having found B(t) we can differentiate it and use ds ( s ) = ´τ (s)N̂(s) and the chain
rule to give
dB ds
= ´τ N̂
dt dt
ds
from which we can read off τ, provided we know dt and N̂.
There is another, often more efficient, method to find the torsion τ that uses
da d  d2 s  ds 2 
= T̂ + κ N̂
dt dt dt2 dt
d3 s d2 s ds d   ds 2   ds 3
τ B̂ ´ κ T̂


= 3 T̂ + 2 κ N̂ + κ N̂ + κ
dt dt dt dt dt dt

3
While this looks a little complicated, notice that, with just one exception, namely κ ds
dt τ ( s ) B̂ ( s ),
every term on the right hand side is either in the direction T̂ or in the direction N̂ and so

3
is perpendicular to B̂. So, dotting with v ˆ a = κ ds dt B̂ gives

da  6
2 ds
vˆa ¨ =κ τ = |v ˆ a|2 τ
dt dt
and hence
v ˆ a ¨ da


dt
τ=
|v ˆ a|2
If the curvature18 κ (s) ą 0 and the torsion τ (s) are known, then the system of equa-
tions19
Equation 1.4.2 (Frenet–Serret Formulae).

dT̂
(s) = κ (s) N̂(s)
ds
dN̂
(s) = τ (s) B̂(s) ´ κ (s) T̂(s)
ds
dB̂
(s) = ´τ (s) N̂(s)
ds

is a first order linear system of ordinary differential equations

    
T̂(s) 0 κ (s) 0 T̂(s)
d 
N̂(s) = ´κ (s) 0 τ (s) N̂(s)
ds
B̂(s) 0 ´τ (s) 0 B̂(s)

18 As in two dimensions, if κ (s) = 0, then N̂(s) is not defined. This makes even more sense in three
dimensions than in two dimensions: if the curve is a straight line, there are infinitely many unit vectors
perpendicular to it and there is no way to distinguish between them.
19 The equations are named after the two French mathematicians who independently discovered them:
Jean Frédéric Frenet (1816–1900, the son of a wig maker), in his thesis of 1847 (actually he only gave
two of the three equations), and Joseph Alfred Serret (1819–1885) in 1851.

26
C URVES 1.4 C URVES IN T HREE D IMENSIONS

for the 9 component vector valued function (T̂(s) , N̂(s) , B̂(s)).

Any first order linear initial value problem

d
x(s) = M(s)x(s) x (0) = x0
ds
where x is an n-component vector and M(s) is an n ˆ n matrix with continuous entries,
has exactly one solution. If n = 1, so that x(s) and M(s) are just functions, this is easy to
see. Just let M(s) be the antiderivative of M(s) that obeys M(0) = 0. Then

d d
x(s) = M (s)x(s) ðñ e´M(s) x(s) ´ M(s)e´M(s) x(s) = 0
ds ds
d ´M(s)
 
ðñ e x(s) = 0
ds

by the product rule. So e´M(s) x(s) is a constant, which to satisfy the initial condition, must
be x0 . That is x(s) = x0 eM(s) . This argument can be generalized to any natural number n.
But that is beyond the scope of this book.
Since the Frenet-Serret formulae constitute a first order system of ordinary differential
equations for the vector (T̂(s) , N̂(s) , B̂(s)) and since any first order linear initial value
problem has a exactly one solution,

˝ the vector valued function (T̂(s) , N̂(s) , B̂(s)) is determined by the functions κ (s)
and τ (s) (assuming that they are continuous) together with the initial condition
(T̂(0) , N̂(0) , B̂(0)).
dr
˝ Furthermore, once you know T̂(s), then r(s) is determined by r(0) and ds (s) = T̂(s).
˝ So any smooth curve r(s) is completely determined by r(0), (T̂(0) , N̂(0) , B̂(0)),
κ (s) and τ (s).
˝ That is, up to translations (you can move between any two possible choices of r(0)
by a translation) and rotations (you can move between any two possible choices of
(T̂(0) , N̂(0) , B̂(0)) by a rotation) a curve is completely determined by the curvature
κ (s) ą 0 and the torsion τ (s). This result is called “The fundamental theorem of
space curves”.

Theorem 1.4.3 (The Fundamental Theorem of Space Curves).

Let κ (s) ą 0 and τ (s) be continuous. Then up to translations and rotations, there
is a unique curve with curvature κ (s) and torsion τ (s).

Example 1.4.4 (Right circular helix)

The right circular helix is the curve

r(t) = a cos t ı̂ı + a sin t ̂ + bt k̂

with a, b ą 0 as in the figure on the left below.

27
C URVES 1.4 C URVES IN T HREE D IMENSIONS

t“ 5π
2

t“ 3π
2
π
t“ 2
y

t“0
x

Here is why it is called a right helix rather than a left helix. If the helix is the thread of a
bolt that you are screwing into a nut, and you turn the bolt in the direction of the (curled)
fingers of your right hand (as in the figure20 on the right above), then it moves in the
direction of your thumb (as in the black arrow of the figure on the right above).
To determine the curvature and torsion of this curve we compute

v(t) = ´a sin t ı̂ı + a cos t ̂ + b k̂

a(t) = ´a cos t ı̂ı ´ a sin t ̂
da
(t) = a sin t ı̂ı ´ a cos t ̂
dt
From v(t) we read off

ds a a a b
= a2 + b2 T̂(t) = ´ ? sin t ı̂ı + ? cos t ̂ + ? k̂
dt a2 +b 2 2
a +b 2 a + b2
2

d2 s
From a(t), and the fact that dt2
= 0, we read off that
a
κ (t) = N̂(t) = ´ cos t ı̂ı ´ sin t ̂
a2 + b2
From
 
ı̂ı ̂ k̂
v(t) ˆ a(t) = det  ´a sin t a cos t b  = ab sin t ı̂ı ´ ab cos t ̂ + a2 k̂
´a cos t ´a sin t 0
|v(t) ˆ a(t)|2 = a2 b2 + a4 = a2 ( a2 + b2 )

we read off
b b a
B̂ = ? sin t ı̂ı ´ ? cos t ̂ + ? k̂
a2 +b 2 2
a +b 2 a + b2
2

20 This figure is a variant of https://commons.wikimedia.org/wiki/File:Right hand rule simple.png

28
 C URVES 1.5 A C OMPENDIUM OF C URVE F ORMULA

and
v ˆ a ¨ da


dt a2 b b
τ (t) = = = 2
|v ˆ a| 2 2 2 2
a (a + b ) a + b2
Note that, for the right handed helix, τ ą 0. Finally the centre of curvature is

1  a2 + b2   a2 + b2 
r(t) + N̂(t) = a ´ cos t ı̂ı + a ´ sin t ı̂ı + bt k̂
κ (t) a a
b2 b2
= ´ cos t ı̂ı ´ sin t ı̂ı + bt k̂
a a
which is another helix. In the figure below, the red curve is the original helix and the blue
curve is the helix traced by the centre of curvature.

Example 1.4.4

1.5IJ A Compendium of Curve Formula

In the following r(t) = x (t) , y(t) , z(t) is a parametrization of some curve. The vectors
T̂(t), N̂(t), and B̂(t) are the unit tangent, normal and binormal vectors, respectively, at
r(t). The tangent vector points in the direction of travel (i.e. direction of increasing t) and
the normal vector points toward the centre of curvature. The arc length from time 0 to
time t is denoted s(t). The binormal B̂(t) = T̂(t) ˆ N̂(t) is perpendicular to the plane
that fits the curve best at r(t). Some formulae use an arc length parametrization, which is
denoted r(s).
dr ds
˝ the velocity v(t) = dt ( t ) = dt ( t ) T̂ ( t )
v(t)
˝ the unit tangent vector T̂(t) = |v(t)|
(general parametrization)
dr
T̂(s) = ds ( s ) (arc length parametrization)

29
C URVES 1.5 A C OMPENDIUM OF C URVE F ORMULA

d2 r d2 s

2
˝ the acceleration a(t) = dt2
(t) = dt2
(t) T̂(t) + κ (t) ds
dt ( t ) N̂ ( t )
ˇ dr ˇ
˝ the speed ds
( t ) = |v ( t ) | = ˇ ( t )ˇ
dt dt
ş T ds şT a
˝ the arc length s( T ) = 0 dt (t) dt = 0 x1 (t)2 + y1 (t)2 + z1 (t)2 dt
ˇ ˇ |v(t) ˆ a(t)|
˝ the curvature κ (t) = ˇ ddtT̂ (t)ˇ/ ds
dt ( t ) =
( ds
dt ( t ))
3
ˇ dφ ˇ ˇ ˇ
κ (s) = ˇ ds (s)ˇ = ˇ ddsT̂ (s)ˇ
ˇ ˇ
˝ the unit normal vector N̂(t) = ddtT̂ (t)/ˇ ddtT̂ (t)ˇ N̂(s) = ddsT̂ (s)/κ (s)
1
˝ the radius of curvature ρ(t) = κ (t)

˝ the centre of curvature r(t) + ρ(t)N̂(t)

v(t) ˆ a(t) ¨ da


dt ( t )
˝ the torsion τ (t) =
|v(t) ˆ a(t)|2
v(t) ˆ a(t)
˝ the binormal B̂(t) = T̂(t) ˆ N̂(t) =
|v(t) ˆ a(t)|
dr
Under arclength parametrization (i.e. if t = s) we have T̂(s) = ds ( s ) and the Frenet-Serret
formulae
dT̂
(s) = κ (s) N̂(s)
ds
dN̂
(s) = τ (s) B̂(s) ´ κ (s) T̂(s)
ds
dB̂
(s) = ´τ (s) N̂(s)
ds

which in matrix form is

    
T̂(s) 0 κ (s) 0 T̂(s)
d 
N̂(s) = ´κ (s) 0 τ (s) N̂(s)
ds
B̂(s) 0 ´τ (s) 0 B̂(s)
When the curve lies entirely in the xy-plane the curvature is given by
ˇ dx 2 ˇ
ˇ (t) d 2y (t) ´ dy (t) d2 2x (t)ˇ
dt dt dt dt
κ (t) = h i3/2
dx

2 dy
2
dt ( t ) + dt ( t )

When the curve lies entirely in the xy-plane and the y-coordinate is given as a function,
y( x ), of the x-coordinate, the curvature is
ˇ d2 y ˇ
ˇ 2 ( x )ˇ
dx
κ (x) = h
dy
2 i3/2
1 + dx ( x )

dx d2 x
Notice that this follows from the previous formula since dx = 1 and dx2
= 0.

30
 C URVES 1.6 I NTEGRATING A LONG A C URVE

1.6IJ Integrating Along a Curve

Suppose that we have a curve C that is parametrized as r(t) with a ď t ď b. Suppose

further that C is actually a piece of wire and that the density (i.e. mass per unit length) of
the wire at the point r is ρ(r). How do we figure out the mass of C ? Of course we use the
standard Calculus divide and conquer strategy. We select a natural number n and

˝ divide the interval a ď t ď b into n equal subintervals, each of length ∆t = b´a

n . We
denote by t` = a + `∆t the right hand end of interval number `.
˝ Then



ˇ we approximate the length of the part of the curve between r t`´1 and r t`
by ˇr t`´1 ´ r t` ˇ and the mass of the part of the curve between r t`´1 and r t`


ˇ

by ρ r(t` ) ˇr t`´1 ´ r t` ˇ.

ˇ

ˇ

rptℓ´1 q rptℓ q

˝ This gives us, as an approximate mass for C of

n n ˇ
ˇ
ÿ ÿ
ˇ r t`´1 ´ r t` ˇ


ρ r(t` ) ˇr t`´1 ´ r t` ˇ = ρ r(t` ) ˇˇ ˇ∆t

ˇ

ˇ
t` ´ t`´1 ˇ
`=1 `=1

Then we take the limit as n Ñ 8. Assuming21 that r(t) is continuously differentiable and
that ρ(r) is continuous we get

żb

ˇˇ dr ˇˇ
Mass of C = ρ r(t) ˇ (t)ˇ dt
a dt

which we take to be a definition.

21 We could relax these conditions somewhat by instead assuming that r1 (t) and ρ(t) are bounded and are
continuous except at a finite number of points. (r1 (t) need not exist at all at those points.)

31
C URVES 1.6 I NTEGRATING A LONG A C URVE

Definition 1.6.1.

(a) For a parametrized curve x (t), y(t), z(t) , a ď t ď b, in R3 that we call C ,

and for a function f ( x, y, z), we define

ż żb

b
f ( x, y, z) ds = f x ( t ), y ( t ), z ( t ) x1 (t)2 + y1 (t)2 + z1 (t)2 dt
C a

In this notation the subscript C specifies the curve, and ds signifies arc length.

(b) For a curve y = f ( x ), a ď x ď b, in R2 that we call C, and for a function

g( x, y), we define
ż żb

b
g( x, y) ds = g x, f ( x ) 1 + f 1 ( x )2 dx
C a

Example 1.6.2

Suppose that we have a helical wire22

0 ď t ď 2π



r(t) = x (t) , y(t) , z(t) = a cos t , a sin t , bt

and that this wire has constant mass density ρ. Let’s find the centre of mass of the wire.
Recall that the centre of mass is x̄, ȳ, z̄) with, for example, x̄ being the weighted average
ş ş
xρds xds
x̄ = ş = ş (since ρ is constant)
ρds ds
ş ş
of x over the wire. Similarly ȳ = şyds and z̄ = şzds . For the given curve
ds ds



x (t) , y(t) , z(t) = a cos t , a sin t , bt
x1 (t) , y1 (t) , z1 (t) = ´ a sin t , a cos t , b



b a a
ds
(t) = x (t) + y (t) + z (t) = a2 sin2 t + a2 cos2 t + b2 = a2 + b2
1 2 1 2 1 2
dt
so that
ş ş2π ? ş2π
xds x ( t ) a 2 + b2 dt a cos(t) dt
x̄ = ş = 0 ş2π ? = 0 =0
ds a 2 + b2 dt 2π
0
ş ş2π ? ş2π
yds y ( t ) a 2 + b2 dt a sin(t) dt
ȳ = ş = 0 ş2π ? = 0 =0
ds a 2 + b2 dt 2π
0
ş ş2π ? ş2π
zds z ( t ) a 2 + b2 dt bt dt b h t2 i2π
z̄ = ş = 0 ş2π ? = 0 = = bπ
ds a 2 + b2 dt 2π 2π 2 0
0

22 For example, your favourite solenoid or spring or slinky.

32
 C URVES 1.7 S LIDING ON A C URVE

So the centre of mass is right on the axis of the helix, half way up, which makes perfect
sense.
Example 1.6.2

1.7IJ Sliding on a Curve

We are going to investigate the motion of a particle of mass m sliding on a frictionless23 ,
smooth curve that lies in a vertical plane. We will consider three scenarios:
˝ First, to set things up we’ll look at a bead sliding on a stiff wire.
˝ Then, we will imagine that we are skiing straight downhill and ask “Where on the
hill can we become airborne?”.
˝ Then we will imagine that we are skateboarding in a culvert (a large pipe) and ask
“When is it safe?”.

§§ The Sliding Bead

First, consider a bead of mass m that is sliding, without friction, on a stiff wire. According
to Newton’s law of motion
ma = F
where F is the net force being applied to the bead. The bead is subject to two forces.
The gravitational force is ´mg̂ . By definition, absence of friction means that the wire is
does not apply any force that is in the direction tangential to the wire. But, because it is
stiff, the wire never changes shape and instead applies just the right amount of force, in
the direction normal to the wire, that is needed to keep the bead on the wire24 without
bending the wire. Call this normal force W N̂.

̂ wire

W N̂

px, yq ı̂ı
bead

T̂ ´mg̂ (gravity)

So, by Newton’s law,

m a = ´mg ̂ + W N̂

23 We are mathematicians — we like idealized situations.

24 This force is required to keep the bead from either passing through the wire or flying off the wire.

33
C URVES 1.7 S LIDING ON A C URVE

We’ll analyse this equation by splitting it into its tangential and normal components.
To extract the tangential component of Newton’s law, we dot it with v = |v|T̂. Since
T̂ ¨ N̂ = 0 this kills all normal components.
dv
mv ¨ = ´mg̂ ¨ v + W N̂ ¨ v
dt
1 d dy
m (v ¨ v) = ´mg
2 dt dt
Here we have used
˝ Theorem 1.1.2.c on the left hand side and
˝ that ̂ ¨ v is just the y component of v and
˝ that N̂ and v = |v|T̂ are perpendicular.
Moving everything to the left hand side of the equation gives
 
d 1 2
m|v| + mgy = 0
dt 2
and we conclude that the quantity

Equation 1.7.1 (Conservation of Energy).

1
E= m|v|2 + mgy
2

is a constant, independent of time. This

b is, of course, the principle of conservation of
energy. It determines the speed |v| = 2Em ´ 2gy of the bead as a function of the height y
(and of the energy E, which is determined by the initial conditions).
To extract the normal component of Newton’s law, we dot it with N̂:

ma ¨ N̂ = ´mg̂ ¨ N̂ + W

Since
d2 s  ds 2 d2 s
a= T̂ + κ N̂ = T̂ + κ|v|2 N̂
dt2 dt dt2
and T̂ and N̂ are perpendicular, this gives, after a little rearrangement,

Equation 1.7.2 (Normal Force).

W = mκ|v|2 + mg̂ ¨ N̂ = 2κ ( E ´ mgy) + mg̂ ¨ N̂

§§ The Skier
The difference between the bead on the wire and the skier on the hill is that while the hill
is capable of applying an upward normal force (i.e. it can push you upward to keep you
from falling to the centre of the Earth), it is not capable of applying a downward normal
force. That is the hill cannot pull down on you to keep you on the hill. Only gravity can
keep you grounded. There are two main possibilities.

34
C URVES 1.7 S LIDING ON A C URVE

N̂

T̂
N̂ T̂

˝ If the hill is concave downward as in the figure on the left above, then N̂ points
downward and the hill is allowed to have W ď 0 (which corresponds to the normal
force W N̂ pushing upward). If ever W ą 0, the hill would have to pull on you to
keep you on hill. It can’t, so you become airborne. Since ̂ ¨ N̂ ă 0, this happens
whenever
c
g
W ą 0 ðñ mκ|v|2 + mg̂ ¨ N̂ ą 0 ðñ |v| ą |̂ ¨ N̂|
κ

˝ If the hill is concave upward as in the figure on the right above, then N̂ points up-
ward and the hill is allowed to have W ě 0 (which corresponds to the normal force
W N̂ pushing upward). Since ̂ ¨ N̂ ą 0 we always have W = mκ|v|2 + mg̂ ¨ N̂ ą 0.
You never become airborne. On the other hand your knees may complain.

§§ The Skate Boarder

So far, Equations (1.7.1) and (1.7.2) apply to any stiff frictionless “wire”. We now specialize
to the special case of a skateboarder inside a circular culvert of radius a. Let’s put the
bottom of the circle at the origin (0, 0), so that the centre of the circle is at (0, a).

a ´ y φ aN̂
px, yq

a´y
In this case the curvature is κ = 1a and ̂ ¨ N̂ = cos φ = a so (1.7.1) and (1.7.2) simplify
to
c d
2  E 
|v| = ( E ´ mgy) = 2g ´y
m mg
2 mg 3mg  2 a 
W = ( E ´ mgy) + ( a ´ y) = E+ ´y
a a a 3mg 3

35
C URVES 1.7 S LIDING ON A C URVE

Imagine now that you start at the bottom of the culvert, that is at y = 0, with energy E ą 0.
As time progresses, y increases and consequently |v| and W both decrease, as, of course,
they should. This continues until one of the following three things happen.

(i) |v| hits 0, in which case you stop rising and start descending. The speed |v| is zero
E
when y = yS = mg . (The subscript “S” stands for “stop”.) Physicists say that when
you reach yS all of your kinetic energy ( 12 m|v|2 ) has been converted into potential
energy (mgy).

(ii) W hits zero. When you get higher than this, W becomes negative and the culvert
would have to pull on you to keep your feet on the culvert. As the culvert can only
push on you, you become airborne. The normal force W is zero when y = y A =
2 E a
3 mg + 3 . (The subscript “A” stands for “airborne”.)

(iii) y hits 2a. This is the summit of the culvert. You descend on the other side.

Which case actually happens is determined by the relative sizes of yS , y A and 2a.

2 E 1 E 2 E a E
˝ Comparing yS = 3 mg + 3 mg and y A = 3 mg + 3 , we see that yS ď y A ðñ mg ď a.
2 E a 2 a E
˝ Comparing y A = 3 mg + 3 and a = 3 a + 3 , we see that y A ď a ðñ mg ď a.
2 E a 5 a E 5
˝ Comparing y A = 3 mg + 3 and 2a = 3 a + 3 , we see that y A ď 2a ðñ mg ď 2 a.

So the conclusions are:

E
˝ If 0 ď mg ď a then 0 ď yS ď y A ď a . In this case you just oscillate between heights
0 and yS ď a in the bottom half of the culvert, as in the figure on the left below.
E
˝ If a ď mg ď 52 a then a ď y A ď yS , 2a . In this case you make it more than half way
to the top. But you become airborne at y = y A which is somewhere between the half
way mark y = a and the top y = 2a. At this point our model breaks down because
you are no longer in contact with the culvert. You just freely follow a parabolic arc
until you crash back into the culvert, as in the figure in the centre below.
˝ If 52 a ă mgE
then 2a ă y A ă yS . In this case you successfully go all the way
around the culvert, looping the loop, as in the figure on the right below. Note that,
E
as mg ą 52 a ą 2a, this requires significantly more energy than that required to reach
the top, i.e. to reach height 2a.

happy unhappy thrilled!

36
 C URVES 1.8 O PTIONAL — P OLAR C OORDINATES

1.8IJ Optional — Polar Coordinates

So far we have always vectors in two dimensions using the basis vectors ı̂ı and ̂ . This is not
always convenient. For example, when working in polar coordinates it is often convenient
to use basis vectors r̂(θ ), θ̂θ (θ ) which depend on the value of the current polar coordinate θ
— though one usually just writes r̂, θ̂θ , suppressing the dependence on θ from the notation.
When one is at the point with polar coordinates (r, θ ), these basis vectors are defined by

Equation 1.8.1.
θ̂θ r̂
r̂(θ ) = cos θ ı̂ı + sin θ ̂ θ
̂
θ̂θ (θ ) = ´ sin θ ı̂ı + cos θ ̂ r
θ
ı̂ı
p0, 0q

Note that this basis has two very nice properties.

1. |r̂(θ )| = |θ̂θ (θ )| = 1, r̂(θ ) K θ̂θ (θ ) (orthonormality)

dr̂ dθ̂θ
2. dθ ( θ ) = θ̂θ (θ ), dθ ( θ ) = ´r̂(θ )
dr̂
That dθ ( θ ) is some scalar multiple of θ̂θ (θ ) follows just from the fact that |r̂(θ )| = 1.

|r̂(θ )| = 1 ùñ r̂(θ ) ¨ r̂(θ ) = 1

dr̂ 1d
ùñ r̂(θ ) ¨ r̂(θ ) ¨ r̂(θ ) = 0


(θ ) =
dθ 2 dt
dr̂ dr̂
ùñ (θ ) K r̂(θ ) ùñ (θ ) k θ̂θ (θ )
dθ dθ

Similarly that ddθθ̂θ (θ ) is some scalar multiple of r̂(θ ) follows just from the fact that |θ̂θ (θ )| = 1.

Lemma 1.8.2.

If we parametrize a curve by giving its polar coordinates25 r (t) , θ (t) , then

(a) r(t) = r (t) r̂ θ (t)

(b) v(t) = dr dθ



dt ( t ) r̂ θ ( t ) + r ( t ) dt ( t ) θ̂ θ ( t )
θ
 2
2 
 
d2 θ
(c) a(t) = ddt2r (t) ´ r (t) dθ dr dθ


dt ( t ) r̂ θ ( t ) + r ( t ) dt 2 ( t ) + 2 dt ( t ) dt ( t ) θ
θ̂ θ ( t )

25 As usual r is the distance from the origin to the point and θ is angle between the x-axis and the vector
from the origin to the point.

37
C URVES 1.8 O PTIONAL — P OLAR C OORDINATES

It is standard to suppress the arguments t and θ (t) and write, for example,
dr dθ
v= r̂ + r θ̂θ
dt dt
But it is important to remember that the arguments really are there.
Proof. The vector from the origin to the point whose polar coordinates are (r, θ ) is r =
r r̂(θ ). So if we parametrize a curve by giving the polar coordinates at time t,


r(t) = r (t) r̂ θ (t)
dr
dr̂
dθ
v(t) = (t) r̂ θ (t) + r (t) θ (t) (t)
dt dθ dt
dr
dθ

= (t) r̂ θ (t) + r (t) (t) θ̂θ θ (t)
dt dt
d2 r dr dr̂ dθ dr dθ d2 θ  dθ 2 dθ̂θ
a(t) = 2 r̂ + + θ̂ + r 2 θ̂ + r
θ θ
dt dt dθ dt dt dt dt dt dθ
 d2 r  dθ 2   d2 θ dr dθ 
= ´r r̂ + r 2 + 2 θ̂θ
dt2 dt dt dt dt

Example 1.8.3

As an example, consider a bead that is sliding on a frictionless rod that has one end fixed
at the origin and that is rotating about the origin at a constant Ω rad/sec.

Ωt

Because the rod is frictionless, it is incapable of applying to the bead any force parallel to
the rod. So under Newton’s law, ma = F, the radial26 component of the acceleration of

the particle is exactly zero. So, if the polar coordinates of the bead at time t are r (t), θ (t) ,
then, by Lemma 1.8.2.c,
d2 r  dθ 2
´ r =0
dt2 dt
As the rod is rotating at Ω rad/sec, dθ
dt = Ω and

d2 r
2
´ Ω2 r = 0
dt

26 The θ̂θ component of the acceleration just tells us how much normal force the rod is applying to the bead
to keep it on the rod.

38
C URVES 1.8 O PTIONAL — P OLAR C OORDINATES

The general solution to this constant coefficient second order ordinary differential equa-
tion is
r (t) = AeΩ t + Be´Ω t
where A and B are arbitrary constants that are determined by initial conditions. Just as an
example, if r (0) = 1 and r1 (0) = 0, then A + B = 0 and A ´ B = 0, so that A = B = 12 and
1 Ωt
e + e´Ω t


r (t) =
2
If, again for example, θ (0) = 0, then θ (t) = Ωt and the bead follows the polar coordinate
curve
1
r (θ ) = eθ + e´θ


2
which is a spiral.

Example 1.8.3

Example 1.8.4 (Conic sections in polar coordinates)

In this example, we derive the equation of a general conic section in polar coordinates. A
conic section is the intersection of a plane with a cone. This is illustrated in the figures
below. For our current purposes, it is convenient to use the equivalent27 (and often used)

circle ellipse parabola hyperbola

definition that a conic section is the set of points P in the xy-plane

27 It is outside our scope to prove this equivalence.

39
C URVES 1.8 O PTIONAL — P OLAR C OORDINATES

• whose distance from a fixed point F (called the focus of the conic)
• is a constant multiple ε ě 0 (called the eccentricity of the conic)
• of the distance from P to a fixed line L (called the directrix of the conic).

Choose a coordinate system with the focus F of the conic being the origin and with the
directrix L being x = p for some p ą 0.

y L

P Q
r
θ
F x

If P has polar coordinates (r, θ ), then P has x-coordinate r cos θ. The point Q on the line
L in the figure above has x-coordinate p. So the distance from P to L, which is also the
distance from P to Q, is p ´ r cos θ. The distance from P to F is r. We require that the
distance from P to F is ε times the distance form P to L. So
εp
r = ε p ´ r cos θ ðñ r =


1 + ε cos θ

The numerator εp is usually renamed to ` giving the equation

`
r=
1 + ε cos θ

Example 1.8.4

Example 1.8.5 (Conic sections in polar coordinates, again)

We’ll now take the equation r = 1+ε `cos θ for a conic section in polar coordinates, from the
last example, and convert it to the more familiar Cartesian coordinates. Just by the defini-
tion of polar coordinates

r 1 + ε cos θ = ` ðñ r = ` ´ εx

ðñ x2 + y2 = `2 ´ 2ε` x + ε2 x2
ðñ (1 ´ ε2 ) x2 + 2ε` x + y2 = `2 (C)

Now consider separately four different cases, depending on the value of ε ě 0.

40
 C URVES 1.9 O PTIONAL — C ENTRAL F ORCES
y
• If ε = 0, (C) reduces to
ℓ
x 2 + y2 = `2 x

which is of course a circle of radius `.

• If 0 ă ε ă 1, completing the square in (C) gives
 ε ` 2 ε2 `2 `2
(1 ´ ε2 ) x + + y 2
= ` 2
+ =
1 ´ ε2 1 ´ ε2 1 ´ ε2
which is equivalent to y
ε` 2


x + 1´ε 2 y2
+ =1 x
`2
2
`2
2
p´rM , 0q
(1´ε2 ) 1´ε
p0, ´rm q
`
and is of course an ellipse with semi-major axis r M = 1´ε2
and semi-minor axis
rm = ? ` 2 .
1´ε y
• If ε = 1, (C) reduces to
pℓ{2, 0q
y2 = `2 ´ 2` x x
p0, ´ℓq
which is of course a parabola.
• If ε ą 1, the same computation as in the 0 ă ε ă 1 case gives
y
` 2
x ´ ε2ε´1


y2
2 ´ 2 =1
` `
x
(ε2 ´1)2 ε2 ´1

and is of course a hyperbola.

Example 1.8.5

1.9IJ Optional — Central Forces

One of the great triumphs of Newtonian mechanics was the explanation of Kepler’s laws28 ,
which said

28 The German astronomer Johannes Kepler (1571–1630) developed these laws during the course of an
attempt to relate the five extraterrestrial planets then known to the five Platonic solids. He based the
laws on a great number of careful measurements made by the Danish Astronomer Tycho Brahe (1546–
1601). Then Isaac Newton (English, 1642–1727) provided the explanation in 1687. Kepler also wrote a
paper entitled “On the Six-Cornered Snowflake”. Tycho Brahe lost his nose in a sword duel and wore
a prosthetic nose from then on. The story is that Brahe died from a burst bladder that resulted from his
refusing to leave the dinner table before his host.

41
C URVES 1.9 O PTIONAL — C ENTRAL F ORCES

1. The planets trace out ellipses about the sun as focus.

2. The radius vector r sweeps out equal areas in equal times.
3. The square of the period of each planet is proportional to the cube of the major axis
of the planet’s orbit.

Newton showed that all of these behaviours follow from the assumption that the accel-
eration a(t) of each planet obeys the law of motion ma = F where m is the mass of the
planet and
GMm
F=´ 3 r
r
is the “gravitational force” applied on the planet by the sun. Here G is a constant29 , called
the “gravitational constant” or the “universal gravitational constant”, M is the mass of the
sun, r is the vector from the sun to the planet and r = |r|.
In this section, we’ll show that some of these properties follow from the weaker as-
sumption that the acceleration a(t) of each planet obeys the law of motion ma = F with F
being a central force. That is, the assumption that F is parallel to r. The verification that
the other properties follow from the specific form of the gravitational force, proportional
to r´2 , will be delayed until the optional §1.10.
So, in this section, we assume that we have a parametrized curve r(t) and that this
curve obeys
d2 r

m 2 (t) = F r(t)
dt
where, for all r P R3 , F(r) is parallel to r. We shall show that

1. The path r(t) lies in a plane through the origin and that
2. the radius vector r sweeps out equal areas in equal times.

We’ll start by trying to guess what the plane is. Pretend that we know that r(t) lies in a
fixed plane through the origin. Then v(t) = dr
dt ( t ) lies in the same plane and r ( t ) ˆ v ( t ) is
perpendicular to the plane. If our path really does lie in a fixed plane, r(t) ˆ v(t) cannot
change direction — it must always be parallel to the normal vector to the plane. So let’s
define
Ω (t) = r(t) ˆ v(t)
and check how it depends on time. By the product rule,

Ω
dΩ d 1
r(t) ˆ v(t) = v(t) ˆ v(t) + r(t) ˆ a(t) = r(t) ˆ F r(t)) = 0



(t) = (A)
dt dt m
because r(t) and F r(t)) are parallel. So Ω (t) is30 in fact independent of t. It is a constant

vector that we’ll just denote Ω .

As r(t) ˆ v(t) = Ω , we have that r(t) is always perpendicular to Ω and

r(t) ¨ Ω = 0

29 Its value is about 6.67408 ˆ 10´11 m3 kg´1 sec´2 .

30 Physicists call m Ω (t) the angular momentum at time t and refer to (A) as (an example of) conservation
of angular momentum. Conservation of angular momentum is exploited in gyro-compasses and by ice
skaters (to spin faster/slower).

42
 C URVES 1.10 O PTIONAL — P LANETARY M OTION

˝ If Ω ‰ 0, this is exactly the statement that r(t) always lies in the plane through the
origin with normal vector Ω .
˝ If Ω = 0, then r(t) is always parallel to v(t) and there is some function α(t) such that
dr
(t) = v(t) = α(t) r(t)
dt
This is a first order, linear, ordinary differential equation that we can solve by using
an integrating factor. Set
żt
β(t) = α(t) dt
0
Then
dr dr
(t) = α(t) r(t) ðñ e´β(t) (t) ´ α(t)e´β(t) r(t) = 0
dt dt
d  ´β(t)
ðñ

e r(t) = 0
dt
ðñ e´β(t) r(t) = r(0)
ðñ r(t) = e β(t) r(0)
so that r(t) lies on a line through the origin. This makes sense — the particle is
always moving parallel to its radius vector.
This completes the verification that r(t) lies in a plane through the origin.
Now we show that the radius vector r(t) sweeps out equal areas in equal times. In
other words, we now verify that the rate at which r(t) sweeps
ˇ out area is independent of
time. To do so we rewrite the statement that |r(t) ˆ v(t)ˇ is constant in polar coordinates.


Writing r(t) = r (t)r̂ θ (t) and then applying Lemma 1.8.2.b gives that
ˇ ˇ ˇˇ  dr dθ ˇˇ dθ
constant = r ˆ vˇ = ˇrr̂ ˆ
ˇ r̂ + r θ̂θ ˇ = r2 since |r̂ ˆ r̂| = 0, |r̂ ˆ θ̂θ | = 1
dt dt dt
is constant. It now suffices to observe that r (t)2 dθ
dt ( t ) is exactly twice the rate at which r ( t )
sweeps out area. To see this, just look at the figure below. The shaded area is essentially a
wedge of a circular disk of radius r. (If r (t) were independent of t, it would be exactly a
dθ
wedge of a circular disk.) Its area is the fraction 2π of the area of the full disk, which is

dθ 1 rpt ` dtq
πr2 = r2 dθ
2π 2 dθ
rptq

1.10IJ Optional — Planetary Motion

We now return to the claim, made in §1.9 on central forces, that if r(t) obeys Newton’s
inverse square law
d2 r GM GM
2
= ´ 3 r = ´ 2 r̂
dt r r
then the curve obeys Kepler’s laws

43
C URVES 1.10 O PTIONAL — P LANETARY M OTION

1. r(t) runs over an ellipse having one focus at the origin and
2. r(t) sweeps out equal areas in equal times and
3. the square of the period is proportional to the cube of the major axis of the ellipse.

We just showed, in §1.9, that the fact that ´ GM r3

r is parallel to r implies that r(t) lies in a
plane through the origin and sweeps out equal area in equal times. We now verify the
remaining Kepler laws.
We start by just rewriting Newton’s laws above in polar coordinates. We saw in
Lemma 1.8.2.c, that if we write r(t) = r (t) r̂(t), then

d2 r  d2 r  dθ 2   d2 θ dr dθ  GM GM
2
= 2
´ r r̂ + r 2
+ 2 θ̂θ = ´ 3 r = ´ 2 r̂
dt dt dt dt dt dt r r

The r̂ and θ̂θ components of this equation are

d2 r  dθ 2 GM
2
´ r =´ 2
dt dt r
d2 θ dr dθ
r 2 +2 =0
dt dt dt
The second of these two equations only tells us that

d ! 2 dθ ) ! d2 θ dr dθ ) dθ
r = r r 2 +2 = 0 ùñ r2 = h, a constant
dt dt dt dt dt dt
dθ h
which we already knew. Substituting dt = r2
into the first equation gives

d2 r h 2 GM
´ = ´ (1.10.1)
dt2 r3 r2

This equations contains a lot of 1r ’s. So let’s set u = 1r . Furthermore, for the first of Kepler’s
laws, we really want r as a function of θ rather than t. So let’s make u a function of θ and
write
1
r (t) =
u(θ (t))
Then
dr 1 du
dθ du dθ h
(t) = ´ 2 (t) = ´h = 2 = hu2


θ (t) θ (t) since
dt u dθ dt dθ dt r
2
d r 2
d u 2
dθ 2d u
(t) = ´h 2 θ (t) (t) = ´h2 u θ (t)




2
θ (t)
dt dθ dt dθ 2
and our equation becomes

d2 u d2 u GM
´ h2 u2 ´ h2 u3 = ´GMu2 or +u = 2 (1.10.2)
dθ 2 dθ 2 h
This is a second order, linear, ordinary differential equation with constant coefficients.
Recall that the general solution of such an equation is the sum of a “particular solution”

44
C URVES 1.10 O PTIONAL — P LANETARY M OTION

(i.e. any one solution, which in this case we can take to be the constant function GM
h2
) plus
2
the general solution of the homogeneous equation u + u = 0, which one often writes as
A cos θ + B sin θ
with A and B arbitrary constants. In this particular application it is more convenient to
write the solution in a different, standard but less commonly used, form. Namely, we can
use the triangle

C B
α
A

to write A = C cos α and B = C sin α so that the general solution of the homogeneous
equation u2 + u = 0 becomes
C cos α cos θ + C sin α sin θ = C cos(θ ´ α)
with C and α being arbitrary constants. So the general solution to (1.10.2) is
GM
u(θ ) = + C cos(θ ´ α)
h2
and the general solution to (1.10.1) is
1
r (t) = GM
h2
+ C cos(θ (t) ´ α)
The angle α just shifts the zero point of our coordinate θ. By rotating our coordinate system
by α, we can arrange that α = 0 and then
1 ` h2 Ch2
r (t) = GM
= with `= , ε=
+ C cos(θ (t)) 1 + ε cos θ GM GM
h2
As we saw in Example 1.8.4, this is exactly the equation of a conic section with eccentricity
ε.
That leaves only the last of Kepler’s laws, relating the period to the semi-major axis.
As we are talking about planets, whose orbits remain bounded, our conic section must be
a circle or ellipse, rather than a parabola or hyperbola. Looking back at Example 1.8.5, we
see that the semi-major and semi-minor axes of our ellipse are
` `
a= b=?
1 ´ ε2 1 ´ ε2
The period T of our orbit is just the length of time it takes the radius vector r(t) to sweep
out the area of the ellipse31 , which is πab. As the rate at which the radius vector is sweep-
h
ing out area is 12 r2 dθ
dt = 2 , we have
 πab 2 4π 2 a2 b2 4π 2 a2 b2 4π 2 3 b2
T2 = = = = a since ` =
h/2 h2 GM` GM a
31 You probably computed the area of an ellipse in first year calculus. If not, you should be able to do it
now in a few lines.

45
 C URVES 1.11 O PTIONAL — T HE A STROID

1.11IJ Optional — The Astroid

Imagine a ball of radius a/4 rolling around the inside of a circle of radius a. The curve
traced by a point P painted on the inner circle (that’s the blue curve in the figures below)
is called an astroid32 . We shall find its equation.

P
P P

Define the angles θ and φ as in the figure in the left below.

̂ Q ̂
φ θ
θ ı P
ı̂ ı̂ı
O O

32 The name “astroid” comes from the Greek word “aster”, meaning star, with the suffix “oid” meaning
“having the shape of”. The curve was first discussed by Johann Bernoulli in 1691–92.

46
C URVES 1.11 O PTIONAL — T HE A STROID

That is
˝ the vector from the centre, O,
of the circle of radius a to the centre, Q, of the ball of
3
radius a/4 is 4 a cos θ, sin θ and
˝ the vector from the centre, Q, of the ball of radius a/4 to the point P is 14 a cos φ, ´ sin φ

As θ runs from 0 to π2 , the point of contact between the two circles travels through one
quarter of the circumference of the circle of radius a, which is a distance 14 (2πa), which,
in turn, is exactly the circumference of the inner circle. Hence if φ = 0 for θ = 0 (i.e. if P
starts on the x-axis), then for θ = π2 , P is back in contact with the big circle at the north
pole of both the inner and outer circles. That is, φ = 3π π
2 when θ = 2 . (See the figure on
the right above.) So φ = 3θ and P has coordinates
3
1 a
a cos θ, sin θ + a cos φ, sin φ = 3 cos θ + cos 3θ, 3 sin θ ´ 3 sin θ



4 4 4
As, recalling your double angle, or even better your triple angle, trig identities,
cos 3θ = cos θ cos 2θ ´ sin θ sin 2θ
= cos θ [cos2 θ ´ sin2 θ ] ´ 2 sin2 θ cos θ
= cos θ [cos2 θ ´ 3 sin2 θ ]
sin 3θ = sin θ cos 2θ + cos θ sin 2θ
= sin θ [cos2 θ ´ sin2 θ ] + 2 sin θ cos2 θ
= sin θ [3 cos2 θ ´ sin2 θ ]
we have
3 cos θ + cos 3θ = cos θ [3 + cos2 θ ´ 3 sin2 θ ] = cos θ [3 + cos2 θ ´ 3(1 ´ cos2 θ )] = 4 cos3 θ
3 sin θ ´ sin 3θ = sin θ [3 ´ 3 cos2 θ + sin2 θ ] = sin θ [3 ´ 3(1 ´ sin2 θ ) + sin2 θ ] = 4 sin3 θ
and the coordinates of P simplify to
x (θ ) = a cos3 θ y(θ ) = a sin3 θ
Oof! As x2/3 + y2/3 = a2/3 cos2 θ + a2/3 sin2 θ , the path traced by P obeys the equation

x2/3 + y2/3 = a2/3

which is surprisingly simple, considering what we went through to get here.
There remains the danger that there could exist points ( x, y) obeying the equation
x + y2/3 = a2/3 that are not of the form x = a cos3 θ, y = a sin3 θ for any θ. That is, there
2/3

is a danger that the parametrized curve x = a cos3 θ, y = a sin3 θ covers only a portion of
x2/3 + y2/3 = a2/3 . We now show that the parametrized curve x = a cos3 θ, y = a sin3 θ in
fact covers all of x2/3 + y2/3 = a2/3 as θ runs from 0 to 2π.
?
2 ?
2
First, observe that x2/3 = 3 x ě 0 and y2/3 = 3 y ě 0. Hence, if ( x, y) obeys
x2/3 + y2/3 = a2/3 , then necessarily 0 ď x2/3 ď a2/3 and so ´a ď x ď a. As θ runs from
0 to 2π, a cos3 θ takes all values between ´a and a and hence takes all possible values of
3/2
x. For each x P [´a, a], y takes two values, namely ˘[ a2/3 ´ x2/3 ] . If x = a cos3 θ0 =
a cos3 (2π ´ θ0 ), the two corresponding values of y are precisely a sin3 θ0 and ´a sin3 θ0 =
a sin3 (2π ´ θ0 ).

47
 C URVES 1.12 O PTIONAL — PARAMETRIZING C IRCLES

1.12IJ Optional — Parametrizing Circles

We now discuss a simple strategy for parametrizing circles in three dimensions, starting
with the circle in the xy-plane that has radius ρ and is centred on the origin. This is easy
to parametrize:
z
k̂

y r(t) = ρ cos t ı̂ı + ρ sin t ̂ 0 ď t ď 2π

ρ̂
ρı̂ı
x

Now let’s move the circle so that its centre is at some general point c. To parametrize
this new circle, which still has radius ρ and which is still parallel to the xy-plane, we just
translate by c:
k̂
z

c
ρ̂ r(t) = c + ρ cos t ı̂ı + ρ sin t ̂ 0 ď t ď 2π
ρı̂ı
y

x
Finally, let’s consider a circle in general position. The secret to parametrizing a general
circle is to replace ı̂ı and ̂ by two new vectors ı̂ı 1 and ̂ 1 which

(a) are unit vectors,

(b) are parallel to the plane of the desired circle and
(c) are mutually perpendicular.

z k̂1
ρ̂1
c r(t) = c + ρ cos t ı̂ı 1 + ρ sin t ̂ 1 0 ď t ď 2π
1
ρı̂ı
y

x
To check that this is correct, observe that

˝ r(t) ´ c is parallel to the plane of the desired circle because both ı̂ı 1 and ̂ 1 are parallel
to the plane of the desired circle and r(t) ´ c = ρ cos t ı̂ı 1 + ρ sin t ̂ 1

48
C URVES 1.12 O PTIONAL — PARAMETRIZING C IRCLES

˝ r(t) ´ c is of length ρ for all t because

|r(t) ´ c |2 = (r(t) ´ c ) ¨ (r(t) ´ c )
= (ρ cos t ı̂ı 1 + ρ sin t ̂ 1 ) ¨ (ρ cos t ı̂ı 1 + ρ sin t ̂ 1 )
= ρ2 cos2 t ı̂ı 1 ¨ ı̂ı 1 + ρ2 sin2 t ̂ 1 ¨ ̂ 1 + 2ρ cos t sin t ı̂ı 1 ¨ ̂ 1
= ρ2 (cos2 t + sin2 t) = ρ2
since ı̂ı 1 ¨ ı̂ı 1 = ̂ 1 ¨ ̂ 1 = 1 (ı̂ı 1 and ̂ 1 are both unit vectors) and ı̂ı 1 ¨ ̂ 1 = 0 (ı̂ı 1 and ̂ 1 are
perpendicular).
To find such a parametrization in practice, we need to find the centre c of the circle, the
radius ρ of the circle and two mutually perpendicular unit vectors, ı̂ı 1 and ̂ 1 , in the plane
of the circle. It is often easy to find at least one point p on the circle. Then we can take
p´c
ı̂ı 1 = |p´c| . It is also often easy to find a unit vector, k̂1 , that is normal to the plane of the
circle. Then we can choose ̂ 1 = k̂1 ˆ ı̂ı 1 . We’ll illustrate this now.
Example 1.12.1

Let C be the intersection of the sphere x2 + y2 + z2 = 4 and the plane z = y.

˝ The intersection of any plane with any sphere is a circle. The plane in question passes
through the centre of the sphere, so C has the same centre and same radius as the
sphere. So C has radius 2 and centre (0, 0, 0).
˝ Notice that the point (2, 0, 0) satisfies both x2 + y2 + z2 = 4 and z = y and so is on
C. We may choose ı̂ı 1 to be the unit vector in the direction from the centre (0, 0, 0) of
the circle towards (2, 0, 0). Namely ı̂ı 1 = (1, 0, 0).
˝ Since the plane of the circle is z ´ y = 0, the vector ∇ (z ´ y) = (0, ´1, 1) is perpen-
dicular to the plane of C. So we may take k̂1 = ?12 (0, ´1, 1).
˝ Then ̂ 1 = k̂1 ˆ ı̂ı 1 = ?1 (0, ´1, 1) ˆ (1, 0, 0) = ?1 (0, 1, 1).
2 2

Substituting in c = (0, 0, 0), ρ = 2, ı̂ı 1 = (1, 0, 0) and ̂ 1 = ?1 (0, 1, 1) gives

2
z

1
1 r(t) = 2 cos t (1, 0, 0) + 2 sin t ? (0, 1, 1)
k̂
̂1 2
y sin t sin t
= 2 cos t, ? , ? 0 ď t ď 2π


1
ı̂ı 2 2

x ? ?
To check this, note that x = 2 cos t, y = 2 sin t, z = 2 sin t satisfies both x2 + y2 + z2 = 4
and z = y.
Example 1.12.1

Example 1.12.2

Let C be the circle that passes through the three points (3, 0, 0), (0, 3, 0) and (0, 0, 3).

49
C URVES 1.12 O PTIONAL — PARAMETRIZING C IRCLES

˝ All three points obey x + y + z = 3. So the circle lies in the plane x + y + z = 3. We

guess, by symmetry, or by looking at the figure below, that the centre of the circle is
at the centre of mass of the three points, which is 31 [(3, 0, 0) + (0, 3, 0) + (0, 0, 3)] =
(1, 1, 1). We must check this and can do so by checking that (1, 1, 1) is equidistant
from the three points:
ˇ ˇ ˇ ˇ ?
ˇ(3, 0, 0) ´ (1, 1, 1)ˇ = ˇ(2, ´1, ´1)ˇ = 6
ˇ ˇ ˇ ˇ ?
ˇ(0, 3, 0) ´ (1, 1, 1)ˇ = ˇ(´1, 2, ´1)ˇ = 6
ˇ ˇ ˇ ˇ ?
ˇ(0, 0, 3) ´ (1, 1, 1)ˇ = ˇ(´1, ´1, 2)ˇ = 6

This tells us both that (1, 1, 1) is indeed the centre (as only the centre
?is equidistant
from any three distinct points on a circle) and that the radius of C is 6.
˝ We may choose ı̂ı 1 to be the unit vector in the direction from the centre (1, 1, 1) of the
circle towards (3, 0, 0). Namely ı̂ı 1 = ?16 (2, ´1, ´1).
˝ Since the plane of the circle is x + y + z = 3, the vector ∇ ( x + y + z) = (1, 1, 1) is
perpendicular to the plane of C. So we may take k̂1 = ?13 (1, 1, 1).
˝ Then ̂ 1 = k̂1 ˆ ı̂ı 1 = ?1 (1, 1, 1) ˆ (2, ´1, ´1) = ?1 (0, 3, ´3) = ?1 (0, 1, ´1).
18 18 2
? 1
Substituting in c = (1, 1, 1), ρ = 6, ı̂ı = ?1 (2, ´1, ´1) and ̂ 1 = ?1 (0, 1, ´1) gives
6 2
z

? 1 ? 1
r(t) = (1, 1, 1) + 6 cos t ? (2, ´1, ´1) + 6 sin t ? (0, 1, ´1)
6 2
? ?
= 1 + 2 cos t, 1 ´ cos t + 3 sin t, 1 ´ cos t ´ 3 sin t

x y
2π 4π



To check this, note that r(0) = (3, 0, 0), r 3 =? (0, 3, 0) and r 3 = (0, 0, 3) since
?
3
cos 2π
3 = cos 4π
3 = ´ 12 , sin 2π
3 = 2 and sin 4π
3 = ´ 23 .
Example 1.12.2

50
 Chapter 2

V ECTOR F IELDS

2.1IJ Definitions and First Examples

In the last chapter, we studied vector valued functions of a single variable, like, for exam-
ple, the velocity v(t) of a particle at time t. Suppose however that we are interested in a
fluid. There is a, possibly different, velocity at each point in the fluid. So the velocity of
a fluid is really a vector valued function of several variables. Such a function is called a
vector field.

Definition 2.1.1.

(a) A vector field in the plane is a rule which assigns to each point ( x, y) in a
subset, D, of the xy-plane, a two component vector v( x, y).

(b) A vector field in space is a rule which assigns to each point ( x, y, z) in a subset
of R3 , a three component vector v( x, y, z).

Here are two typical applications that naturally involve vector fields.

• If v( x, y, z) is the velocity of a moving fluid at position ( x, y, z), then v is called a

velocity field.

• If F( x, y, z) is the force at position ( x, y, z), then F is called1 a force field.

Example 2.1.2 (The Point Source)

Imagine

˝ The whole world is filled with an incompressible fluid. Call it water.

1 No, force fields are not only a sci-fi trope. Gravity is an example of a force field.

51
V ECTOR F IELDS 2.1 D EFINITIONS AND F IRST E XAMPLES

˝ Somehow you find a way to produce still more water at the origin. Say you create
4πm litres per second.
˝ This forces the water to flow outward. Let’s suppose that it flows symmetrically
outward from the origin.
Let’s find the resulting vector field v( x, y, z). As the flow is to be symmetric, the velocity
of the water at the point ( x, y, z)
˝ has to be pointing radially outward from the origin. That is, the direction of the
velocity vector v( x, y, z) has to be the unit radial vector

xı̂ı + y̂ + zk̂

r̂( x, y, z) = a
x 2 + y2 + z2

˝ The magnitude of the velocity, i.e. the speed |v( x, y, z)| of the water, has to depend
only on the distance from the origin. That is, the speed can only be some function of
b
r ( x, y, z) = x2 + y2 + z2

Thus the velocity field is of the form

v( x, y, z) = v r ( x, y, z) r̂( x, y, z)

We just have to determine the function v(r ). Fix any r ą 0 and concentrate on the sphere
x2 + y2 + z2 = r2 . It is sketched in red in the figure below. During a very short time

vprq dt thick

interval dt seconds, 4πm dt litres of water is created at the origin (which is the red dot).
As the water is incompressible, 4πm dt litres of water must exit through the sphere during
the same time interval to make room for the newly created water.
But, at the surface of the sphere the water is flowing radially outward with speed v(r ).
So during the time interval in question the water near the surface of the sphere moves
outward a distance
a v(r ) dt, and in particular the water that was in the thin spherical shell
r ´ v(r )a
dt ď x + y2 + z2 ď r at the beginning of the time interval exits through the
2

sphere x2 + y2 + z2 = r during the time interval. The shell is sketched in gray in the
figure above. The volume of water in the gray shell is essentially the surface area of the
shell, which is 4πr2 , times the thickness of the shell, which is v(r ) dt. So, equating the
volume of water created inside the sphere with the volume of water that exited the sphere,
4πm m
4πm dt = (4πr2 ) v(r ) dt ùñ v(r ) =


2
= 2
4πr r

52
V ECTOR F IELDS 2.1 D EFINITIONS AND F IRST E XAMPLES

Thus our vector field is

m
v( x, y, z) = r̂( x, y, z)
r ( x, y, z)2
If the world were two, rather than three dimensional2 , and the source created 2πm litres
per second, the same argument leads to

2πm m
ùñ v(r ) =


2πm dt = (2πr ) v(r ) dt =
2πr r
and to the vector field
b
m xı̂ı + y̂
v( x, y) = r̂( x, y) r ( x, y) = x 2 + y2 r̂( x, y) = a
r ( x, y) x 2 + y2

To get a mental image of what this field looks like, imagine sketching, for each point ( x, y),
m m
the vector r( x,y )
r̂( x, y) with its tail at ( x, y). Note that the vector r( x,y )
r̂( x, y)

˝ points radially outward and

m
˝ has length r( x,y )
which
˝ depends only on r = |( x, y)| and
˝ is very long when ( x, y) is near the origin and
˝ decreases in length like 1r as r = |( x, y)| increases.

Here is a sketch of a bunch of such vectors.

Figure 2.1.1.

vector field ~v “ m rr̂

2 You might want to think about what happens in d dimensions for general d.

53
V ECTOR F IELDS 2.1 D EFINITIONS AND F IRST E XAMPLES

Note that as |( x, y)| Ñ 0, the magnitude of the velocity |v( x, y)| Ñ 8. This is a conse-
quence of our idealized assumption that we are producing water at a single point (the
origin).
Example 2.1.2

Example 2.1.3 (The Vortex)

In this example, we sketch the vector field

v( x, y) = Ω ´ yı̂ı + x̂

where Ω is just a strictly positive constant. We give an efficient procedure for getting a
rough sketch, which still provides a pretty realistic picture of the vector field, and which
also generalises to other vector fields. First concentrate on the horizontal component ı̂ı ¨
v( x, y) of the vector field and determine in which part of the xy-plane it is zero, in which
part it is positive and in which part it is negative.
$
’
&= 0 if y = 0
ı̂ı ¨ v( x, y) = ´Ωy ă 0 if y ą 0
’
%
ą 0 if y ă 0

Next repeat with the vertical component.

$
’
&= 0 if x = 0
̂ ¨ v( x, y) = Ωx ă0 if x ă 0
’
%
ą0 if x ą 0

This naturally divides the xy-plane into nine parts according to whether each of the com-
ponents is positive, 0 or negative —
ˇ (
˝ ı̂ı ¨ v ą 0 and ̂ ¨ v ą 0 in ( x, y) P R2 ˇ y ă 0, x ą 0
ˇ (
˝ ı̂ı ¨ v ą 0 and ̂ ¨ v = 0 in ( x, y) P R2 ˇ y ă 0, x = 0
ˇ (
˝ ı̂ı ¨ v ą 0 and ̂ ¨ v ă 0 in ( x, y) P R2 ˇ y ă 0, x ă 0
ˇ (
˝ ı̂ı ¨ v = 0 and ̂ ¨ v ą 0 in ( x, y) P R2 ˇ y = 0, x ą 0
˝ and so on

Now think of v( x, y) as being the velocity at ( x, y) of a flowing fluid.

˝ Look at the firstˇ bullet point above.

( It says that in the first of the nine parts, namely
ˇ
( x, y) P R y ă 0, x ą 0 , which is the fourth quadrant, the horizontal com-
2

ponent ı̂ı ¨ v ą 0 signifying that the fluid is flowing rightwards. Indicate this in the
sketch by drawing a rightward pointing horizontal arrow at some generic point in
the middle of the fourth quadrant. (It’s the blue arrow in the figure below.) The ver-
tical component ̂ ¨ v ą 0 signifying that the fluid is also moving upwards. Indicate
this in the sketch by drawing an upward pointing vertical arrow at the same generic
point in the fourth quadrant. (It’s the red arrow in the figure below.)

54
V ECTOR F IELDS 2.1 D EFINITIONS AND F IRST E XAMPLES

˝ Next, look at the second bullet

ˇ point above.( It says that on the second of the nine
parts, namely ( x, y) P R2 ˇ y ă 0, x = 0 , which is the bottom half of the y-axis,
the horizontal component ı̂ı ¨ v ą 0, signifying that the fluid is moving rightwards.
Indicate this in the sketch by drawing a rightward pointing horizontal arrow at some
generic point in the middle of the bottom half of the y-axis. (It’s the second blue
arrow in the figure below.) The vertical component ̂ ¨ v = 0 signifying that the fluid
has no vertical motion at all. Indicate this in the sketch by not drawing any vertical
arrow on the bottom half of the y-axis.

˝ and so on

By the time we have looked at all nine regions we will have built up the following sketch.

55
V ECTOR F IELDS 2.1 D EFINITIONS AND F IRST E XAMPLES

Figure 2.1.2.
y

From this sketch we see that, for example, in the first quadrant, the fluid is moving up-
wards and to the left. We also see that there is one point, namely (0, 0), where the vector
field is exactly zero. It’s the black dot in the centre of the figure above. Putting all of
this accumulated wisdom together, and also noticing that v( x, y) is smaller when ( x, y) is
closer to the origin and v( x, y) is larger when ( x, y) is farther from the origin, we come up
with this better sketch of the vector field.

Figure 2.1.3.
y

vector field v “ Ωp´yı̂ı ` x̂q

This shows the field swirling around the origin in a counterclockwise direction. Hence
the name “vortex”.
Example 2.1.3

56
V ECTOR F IELDS 2.1 D EFINITIONS AND F IRST E XAMPLES

Example 2.1.4 (The Undamped Nonlinear Pendulum)

In this example, we illustrate another way in which vector fields arise. Model a pendulum
by a mass m that is connected to a hinge by an idealized rod that is massless3 and of fixed
length `. Denote by θ the angle between the rod and vertical. The forces acting on the

θ
ℓ
τ

mg

mass are

˝ gravity and
˝ the tension in the rod, whose magnitude, τ, automatically adjusts itself so that the
distance between the mass and the hinge is fixed at `.

In the optional4 Section 2.5, we show that the angle θ (t) obeys the second order nonlinear5
differential equation
d2 θ g
+ sin θ = 0
dt2 `
It is often much more convenient to deal with first order, rather than second order, differ-
ential equations. The second order pendulum equation above may be reformulated6 as a
system of first order ordinary differential equations, by the simple expedient of defining

x (t) = θ (t) y(t) = θ 1 (t)

So x (t) is the angle at time t and y(t) is the angular velocity at time t. Then,

x1 (t) = θ 1 (t) = y(t)

g
y1 (t) = θ 2 (t) = ´ sin x (t)
`

3 While we are idealizing, let’s put everything in a vacuum.

4 In the optional Section 2.5 we also include frictional forces. In this example, we do not, so set the β of
Section 2.5 to zero here.
5 It is common, when considering only small amplitude oscillations, to approximate sin θ by θ. This
converts our nonlinear differential equation into a linear differential equation.
6 This “hack” generalizes easily and is commonly used when generating, by computer, approximate
solutions to higher order ordinary differential equations.

57
V ECTOR F IELDS 2.1 D EFINITIONS AND F IRST E XAMPLES

Usually, one does not write in the (t) dependence explicitly.

x1 = y
g
y1 = ´ sin x
`
The right hand sides form the vector field
g 
v ( x, y) = y , ´ sin x


`
We can sketch this vector field, just as we sketched the vector field of Example 2.1.3. Not-
ing that the horizontal component
$
’
&= 0 if y = 0
ı̂ı ¨ v( x, y) = y ą 0 if y ą 0
’
%
ă 0 if y ă 0

and the vertical component.

$
’
&= 0 if x = 0, ˘π, ˘2π, ¨ ¨ ¨
g
̂ ¨ v( x, y) = ´ sin x ą0 if ´π ă x ă 0, π ă x ă 2π, etc.
` ’
%
ă0 if 0 ă x ă π, 2π ă x ă 3π, etc.

we have
˝ rightward motion7 when y ą 0
˝ leftward motion when y ă 0
˝ downward motion when 0 ă x ă π, 2π ă x ă 3π, ¨ ¨ ¨ and
˝ upward motion when ´π ă x ă 0, π ă x ă 2π, ¨ ¨ ¨ .
This gives us the collection of arrows in the figure

y“0

x “ ´π x“0 x“π x “ 2π

Our full sketch will be less cluttered if we make all arrows the same length. This gives

7 Note that this is rightward motion of the point ( x, y), not of the pendulum itself.

58
 V ECTOR F IELDS 2.2 O PTIONAL — F IELD L INES

direction field for x1 “ y, y 1 “ ´2 sin x

which is a sketch of what is called the direction field of our vector field (see below).
In the next section, we’ll learn how to use vector field sketches to sketch solution tra-
jectories.
Example 2.1.4

Definition 2.1.5.

The direction field of a vector field v( x, y, z) is the vector field

$
& v( x,y,z) if v( x, y, z) ‰ 0
|v( x,y,z)|
V( x, y, z) =
%0 if v( x, y, z) = 0

2.2IJ Optional — Field Lines

Suppose that we drop a tiny stick into a river8 with the velocity field of the flowing water
being v( x, y). We are assuming, for simplicity, that the velocity field does not depend9 on
time t. The stick will move along with the water10 . When the stick is at r, its velocity will
be the same as the velocity of the water at r, which is v(r). Thus if the stick is at r(t) at
time t, we will have
dr

= v r(t)
dt
The stick will trace out a path, parametrized by r(t).

8 Think Poohsticks.
9 This is not such an unreasonable assumption. The flow often changes on a larger time scale.
10 This is also not an unreasonable approximation.

59
V ECTOR F IELDS 2.2 O PTIONAL — F IELD L INES

Definition 2.2.1.

A path that is parametrized by a function r(t) that obeys

dr

= v r(t)
dt
is called a

˝ field line or integral curve (for general vector fields) or a

˝ stream line or flow line (when the vector field v is being thought of as a
velocity field) or a
˝ line of force (when the vector field v is being thought of as a force field)

of the vector field v.

Example 2.2.2 (Flow Line Sketch for the Vortex of Example 2.1.3)

Consider the vortex vector field, v( x, y) = Ω ´ yı̂ı + x̂ of Example 2.1.3. Once we

sketched the vector field, as in Figure 2.1.3, or even made the “skeleton sketch” of Fig-
ure 2.1.2, we can get rough idea of what the stream lines look like just by following the
arrows. For example, suppose that we start a stream line (i.e. drop the stick into the
stream) on the positive x-axis. Looking at Figure 2.1.2, which is repeated here,

the stick

˝ starts by moving in the +y direction, i.e. straight upward.

˝ As it moves farther into the first quadrant it develops a larger and larger negative
x-component of velocity. So it also moves leftwards toward the y-axis.
˝ Eventually it crosses the positive y-axis moving in the ´x direction, i.e. to the left.
˝ As it moves farther into the second quadrant it develops a larger and larger negative
y-component of velocity. So it also moves downwards toward the x-axis.
˝ Eventually it crosses the negative x-axis moving in the ´y direction, i.e. straight
downward.

60
V ECTOR F IELDS 2.2 O PTIONAL — F IELD L INES

˝ As it moves farther into the third quadrant it develops a larger and larger positive
x-component of velocity. So it also moves rightward towards the y-axis.
˝ Eventually it crosses the negative y-axis moving in the + x direction, i.e. to the right.
˝ As it moves farther into the fourth quadrant it develops a larger and larger positive
y-component of velocity. So it also moves upwards toward the x-axis.
With this type of analysis we cannot tell if the streamline, which is the red line in the figure
above, will return to the x-axis
˝ exactly at its starting point, forming a closed curve, or
˝ inside its starting point, spiralling inwards, or
˝ outside its starting point, spiralling outwards.
Example 2.2.2

While the above procedure is a good way to get a qualitative feel for trajectories, we
can develop more precise, detailed descriptions of field lines by working analytically. As
we saw above, thinking of r(t) as the position at time t of a stick dropped into water
whose velocity at ( x, y) is v( x, y), the velocity

of the stick at time t will be the same as the
velocity of the water at r(t), which is v r(t) . Thus r(t) will obey the system of first order
differential equations
Equation 2.2.3.
dr

(t) = v r(t)
dt



Notice that if we reparametrize r(t), say to R(u) = r t(u) , then R1 (u) = r1 t(u) t1 (u)





is parallel to (though not necessarily equal to) r1 t(u) = v r (t(u) = v R(u) . So if we
only care about the curve traced out by the stick, and not about when the stick is at each
point of the path, then it suffices to impose the weaker condition11 that, when
the stick
is at r(t), its velocity r1 (t) is parallel to (though not necessarily equal to) v r(t) . In three


dimensions, r1 (t) is parallel to v r(t) when the cross product is zero:

Equation 2.2.4.
r1 ( t ) ˆ v r ( t ) = 0

In two dimensions
we can still use the cross product by the simple expedient of thinking
of r1 (t) and v r(t) as three component vectors whose third components are zero.
A more convenient way to implement the weaker “just parallel” condition, involves
reparametrizing our streamline. Suppose that we are in two dimensions with r1 (t) =
dx dy
and fix some t0 . If dx 12



dt ( t ) , dt ( t ) and v ( r ) = v 1 ( r ) , v 2 ( r ) dt ( t0 ) is nonzero , we can
reparametrize the curve (at least near r(t0 )) so as to use x, rather than t as the parameter.
To do so, we

11 We’ll have a more careful discussion of this in the optional §2.2.1.

dy dy
12 If dx dx
dt ( t0 ) = 0, but dt ( t0 ) ‰ 0, we should use y rather than x as the parameter. If dt ( t0 ) = dt ( t0 ) = 0,
then r(t) = r(t0 ) for all t and the streamline doesn’t move. It is just a single point.

61
V ECTOR F IELDS 2.2 O PTIONAL — F IELD L INES

˝ solve x = x (t) for t as a function of x. Call the solution T ( x ). Then

˝ the point on the curve which has x-coordinate x is R( x ) = X ( x ) , Y ( x ) with




X ( x ) = x and Y ( x ) = y T ( x ) .

Then the condition that R1 ( x ) = 1, Y 1 ( x ) is parallel to v R( x ) says that R1 ( x ) is a scalar



multiple of v R( x ) so that there is a nonzero number c( x ) so that R1 ( x ) = c( x )v R( x ) .
That is

1, Y 1 ( x ) = c( x )v1 x, Y ( x ) , c( x )v2 x, Y ( x )

or equivalently

1 Y1 ( x) c( x ) v2 x, Y ( x ) v2 x, Y ( x )
Y (x) = =

=

1 c( x ) v1 x, Y ( x ) v1 x, Y ( x )

This is exactly the statement that y = Y ( x ) is a solution of the differential equation

dy v2 x, y
(x) =

dx v1 x, y

dy
It is conventional to pretend13 that dx is the ratio of dy and dx and rewrite the differential
equation14 as

dx dy
=
v1 ( x, y) v2 ( x, y)

Here is a summary of the discussion we have just completed. It extends to three di-
mensions in an obvious way.

dy
13 Of course dx is not the ratio of dy and dx. However pretending that it is provides a simple way to
remember the technique that is used to solve the equation. You have used this mnemonic device before
when you learned how to solve separable differential equations.
14 Here is another nonrigorous, but intuitive way to come up with this equation. Suppose that our stick is
dy
at ( x, y) and has velocity dx time interval dt the stick moves by dx (t) , dy



dt ( t ) , dt ( t ) . In a tiny dt dt (t) dt =
(dx, dy), which is parallel to v1 ( x, y) , v2 ( x, y) if v (dx = v (dy


x,y) x,y)
.
1 2

62
V ECTOR F IELDS 2.2 O PTIONAL — F IELD L INES

Equation 2.2.5.
Use the symbol k to stand for “is parallel to”.
In two dimensions
 dx dy 

(t) , (t) k v1 (r(t)) , v2 (r(t)
dt dt
 dx dy 
ðñ (t) , 0 ˆ v1 (r(t)) , v2 (r(t)) , 0 = 0


(t) ,
dt dt
dx dy
ðñ =
v1 ( x, y) v2 ( x, y)

and in three dimensions

 dx dy dz 

(t) , (t) , (t) k v1 (r(t)) , v2 (r(t)) , v3 (r(t))
dt dt dt
 dx dy dz 
ðñ (t) ˆ v1 (r(t)) , v2 (r(t)) , v3 (r(t)) = 0


(t) , (t) ,
dt dt dt
dx dy dz
ðñ = =
v1 ( x, y, z) v2 ( x, y, z) v3 ( x, y, z)

Let us apply this to two examples, in which the stream lines of the vortex field of
Example 2.1.3 are found by two different methods.

Example 2.2.6 (Stream lines for the vortex field using r1 (t) k v(r(t)))

In this example we will find the stream lines for the vortex field, v( x, y) = Ω ´ yı̂ı + x̂


1
of Example 2.1.3, by using
the requirement that, on a stream line, the velocity vector r (t)
must be parallel to v r(t) . By (2.2.5) one way to express this requirement mathematically
is

dx dy
=
´Ωy Ωx

This is a simple separable differential equation. We can solve it by cross multiplying and
integrating both sides. (Recall that Ω is a constant.)

ż ż
Ωx dx = ´Ωy dy ðñ Ω x dx = ´Ω y dy

2 Ωx =
2
ðñ 1
´ 12 Ωy2 + C1
2 2
ðñ x + y = C

where C1 and C = Ω2 C1 are just arbitrary constants. So the stream lines of the vortex field
are exactly circles centred on the origin.

63
V ECTOR F IELDS 2.2 O PTIONAL — F IELD L INES

We can come to exactly the same conclusion by using the cross product formulation of
(2.2.4).
 dx dy 
(t) , 0 ˆ v1 (r(t)) , v2 (r(t)) , 0 = 0


(t) ,
dt dt
 dx dy 
ðñ (t) ı̂ı + (t) ̂ ˆ ´ Ωy(t) ı̂ı + Ωx (t) ̂ = 0


dt dt
 dx dy 
ðñ Ωx (t) (t) + Ωy(t) (t) k̂ = 0
dt dt
dx dy
ðñ Ωx (t) (t) + Ωy(t) (t) = 0
dt dt
d 1 
ðñ Ωx (t) + 2 Ωy(t) = 0
2 1 2
(Go ahead and evaluate the derivative.)
dt 2
ðñ 12 Ω x (t)2 + y(t)2 = C1

ðñ x (t)2 + y(t)2 = C

Example 2.2.6

Example 2.2.7 (Stream lines for the vortex field using r1 (t) = v(r(t)))

This time we will find the stream lines for the vortex field, v( x, y) = Ω ´ yı̂ı + x̂ of Ex-

ample 2.1.3, by using (2.2.3), which is

dx
= ´Ωy
dt
dy
= Ωx
dt
We can convert this system of first order linear ordinary differential equations into a single
second order linear constant coefficient differential equation15 , by differentiating the first

15 In Example 2.1.4 we converted a second order ordinary differential equation into a system of first order
ordinary differential equations. We are now just reversing the procedure we used there.

64
 V ECTOR F IELDS 2.2 O PTIONAL — F IELD L INES

d2 x
equation, to get dt2
= ´Ω dy
dt , and then substituting in the second equation to get

d2 x
2
+ Ω2 x = 0
dt
This equation is a special case of the ordinary differential equation treated in Example H.3
of the Appendix H, entitled “Review of Linear Ordinary Differential Equations”. In fact it
is exactly (H.5h ) with R = 0, L = C = Ω1 . So the general solution is (H.7) with ρ = 0 and
ν = Ω, which is
x (t) = A cos(Ωt ´ θ )
with A and θ being arbitrary constants16 . Then

1 dx
y(t) = ´ = A sin(Ωt ´ θ )
Ω dt
giving us the familiar circular stream lines.
Example 2.2.7

2.2.1 §§ More about r1 (t) ˆ v r(t) = 0

Here is a lemma that gives a more precise version of “if we only care about the curve
traced out by the stick, and not about when the stick
is at each point of the path, then it
suffices to impose the weaker condition r1 (t) ˆ v r(t) = 0”.

Lemma 2.2.8.

Lat a ă b and let v(r) be a vector

field. Assume that, for all a ă u ă b, R(u) is
1
defined, both R (u) and v R(u) are continuous and nonzero and

R1 ( u ) ˆ v R ( u ) = 0

ˇ (
Then R(u) ˇ a ă u ă b is contained in a field line.

Proof. As R1 (u) ˆ v R(u) = 0 and both R1 (u) and v R(u) are nonzero, there is an a(u)

such that
R1 ( u ) = a ( u ) v R ( u )

R1 (u)¨v(R(u))
This a(u) = v(R(u))¨v(R(u)) is necessarily nonzero and continuous. Since a(u) is nonzero
and continuous, it never changes sign. That is, either a(u) ą 0 for all u, or a(u) ă 0 for all
u. Let T (u) be an antiderivative of a(u). Then T (u) is strictly monotone (and continuous)

16 Even if you don’t know how x (t) = A cos(Ωt ´ θ ) was arrived at, you should be able to easily verify
that it really does obey x2 + Ω2 x = 0.

65
V ECTOR F IELDS 2.2 O PTIONAL — F IELD L INES


and hence invertible. That is, there is a continuous function U (t) that obeys U T (u) = u
for all a ă u ă b and T U (t) = t for all t in the range of U. Set r(t) = R U (t) . Then

1 1
r1 ( t ) = R1 U ( t ) U 1 ( t ) = a U ( t ) v R U ( t )








= a U (t) v r(t)

1
T U (t) a U (t)


= v r(t)

So r(t) is a field line and R(u) = r( T (u)) is a reparametrization of r(t).

Here are a couple of examples that show that bad things can happen if we drop the
requirement that v(R(u)) is nonzero.
Example 2.2.9


Let the vector field v( x, y) be identically zero. Then any field line x (t) , y(t) must obey

x1 (t) = 0 y1 ( t ) = 0

which forces both x (t) and y(t) to be constants. So each field line

is just a single point. On
the other hand every nonconstant R(u) obeys R (u) ˆ v R(u) = 0 but is not contained
1

in a field line.
Example 2.2.9

Now here is a more interesting example.

Example 2.2.10

Consider the vector field v( x, y) = x ı̂ı . This vector field takes the value 0 at each point on
the y-axis, is a positive multiple of ı̂ı at every point of the right half-plane and is a negative
multiple of ı̂ı at every point of the left half-plane. Let’s find the field lines. Any field line
must obey
dx dy
(t) = x (t) (t) = 0
dt dt
So y(t) must be a constant. We can solve the linear ordinary differential equation dx
dt ( t ) =
x (t) by moving the x (t) to the left hand side, and multiplying by the (integrating factor)
e´t . This gives
dx
e´t (t) ´ e´t x (t) = 0
dt
By the product rule, this is the same as
d ´t

e x (t) = 0
dt
which forces e´t x (t) to be a constant. So our field lines are Cet , D , with C and D being

arbitrary constants. Note that

• if C = 0, the field line is just the single point (0, D ) on the y-axis. It is illustrated by
the black dot in the figure below.

66
 V ECTOR F IELDS 2.3 C ONSERVATIVE V ECTOR F IELDS

• If C ą 0, then as t runs from ´8 to +8, the field line covers the horizontal half-line
ˇ (
( x, D ) ˇ x ą 0

in the right half-plane. It is illustrated by the red line in the figure below.

• If C ă 0, then as t runs from ´8 to +8, the field line covers the horizontal half-line
ˇ (
( x, D ) ˇ x ă 0

in the left half-plane. It is illustrated by the blue line in the figure below.

vector field v “ xı̂ı

On the other hand, fix any constant D and set R(u) = uı̂ı + D̂ . Then

R1 (u) ˆ v R(u) = ı̂ı ˆ uı̂ı ) = 0

ˇ (
But as u runs from ´8 to +8, R(u) runs over the full line ( x, D ) ˇ ´ 8 ă x ă 8 . It
is not contained in any single field line and, in fact, completely covers three different field
lines.
Example 2.2.10

2.3IJ Conservative Vector Fields

Not all vector fields are created equal. In particular, some vector fields are easier to work
with than others. One important class of vector fields that are relatively easy to work
with, at least sometimes, but that still arise in many applications are “conservative vector
fields”.

67
V ECTOR F IELDS 2.3 C ONSERVATIVE V ECTOR F IELDS

Definition 2.3.1.

(a) The vector field F is said to be conservative if there exists a function ϕ such
that F = ∇ ϕ. Then ϕ is called a potential for F. Note that if ϕ is a potential for
F and if C is a constant, then ϕ + C is also a potential for F.

(b) If F = ∇ ϕ is a conservative field with potential ϕ and if C is a constant, then

the set of points that obey ϕ( x, y, z) = C is called an equipotential surface.
Similarly, in two dimensions, the set of points that obey ϕ( x, y) = C is called
an equipotential curve.

Warning 2.3.2.

In physics, when a vector field is of the form F = ´∇ ϕ, then ϕ is called a potential

for F. Note the minus17 sign in F = ´∇ ϕ.
Ò

Example 2.3.3 (Potential energy)

The “conservative” in “conservative vector field” has nothing to do with politics. It comes
from “conservation of energy”. Here is how. Suppose that you have a particle of mass m
moving in a force field F that happens to be of the form
F = ∇ ϕ for some function ϕ. If
the position of the particle a time t is x (t), y(t), z(t) , then, by Newton’s law of motion,
dv
ùñ


ma = F ( t ) = F x ( t ), y ( t ), z ( t )
m
dt
dv
ùñ


m ( t ) = ∇ ϕ x ( t ), y ( t ), z ( t )
dt
Now dot both sides with v(t).
dv
ùñ m v(t) ¨ ( t ) = v ( t ) ¨ ∇ ϕ x ( t ), y ( t ), z ( t )


dt
Bϕ Bϕ
= x1 (t) x ( t ), y ( t ), z ( t ) + y1 ( t )



x ( t ), y ( t ), z ( t )
Bx By
Bϕ
+ z1 ( t )


x ( t ), y ( t ), z ( t )
Bz
d
Next use = 2v ¨ dv
dt v ¨ v dt on the left hand side and the chain rule on the right hand side.
d 1
  d 

ùñ mv(t) ¨ v(t) = ϕ x ( t ), y ( t ), z ( t )
dt 2 dt
d 1

ùñ mv(t) ¨ v(t) ´ ϕ x (t), y(t), z(t) = 0
dt 2
1
ùñ m|v(t)|2 ´ ϕ x (t), y(t), z(t) = const


2

17 Physicists introduce this minus sign in order to eliminate the minus sign in the next footnote.

68
V ECTOR F IELDS 2.3 C ONSERVATIVE V ECTOR F IELDS

So 12 m|v(t)|2 ´ ϕ x (t), y(t), z(t) , which is called the energy18 of the particle at time t,

does not actually depend on time —
it is conserved. Let’s call the initial
energy E. That is,
1 2 1 2
E = 2 m|v(0)| ´ ϕ x (0), y(0), z(0) . Then 2 m|v(t)| ´ ϕ x (t), y(t), z(t) = E for all t and,
in particular
1
ϕ x (t), y(t), z(t) = m|v(t)|2 ´ E ě ´E


2


So even without having to ˇfind x (t) , y(t) , z(t) , we know that our particle can never
(
escape the region ( x, y, z) ˇ ϕ( x, y, z) ě ´E .
Example 2.3.3

Example 2.3.4 (Gravity)

The gravitational force that a body of mass M at the origin exerts on a body of mass m at
r = ( x, y, z) is
GMm
F(r) = ´ 3 r
r
a
where r = |r| = x + y + z and G is the gravitational constant. This force is conserva-
2 2 2

tive, with potential ϕ(r) = GMm

r . To verify that this is correct, observe that
B B GMm 1 GMm(2x ) GMm
ϕ(r) = a =´ =´ 3 x
Bx Bx x2 + y2 + z2 2 2 2
2 [x + y + z ] 3/2 r
B B GMm 1 GMm(2y) GMm
ϕ(r) = a =´ =´ 3 y
By By x + y + z
2 2 2 2 2 2
2 [x + y + z ] 3/2 r
B B GMm 1 GMm(2z) GMm
ϕ(r) = a =´ = ´ z
Bz Bz x2 + y2 + z2 2 [ x2 + y2 + z2 ]3/2 r3

Example 2.3.4

We have already found conservation of energy very helpful a couple of times in Section
1.7 (Sliding on a Curve). So, at this point, there are probably several questions gnawing
away at you.
• Is every vector field conservative?
• If not, is there an easy way to tell whether or not a vector field is conservative?
• If we know that a given vector field is conservative, how do find a potential for it?
Have no fear. We will consider those questions in some detail shortly. But first, a couple
of more examples.
Example 2.3.5

In this example we will show that the vector field F( x, y) = x ı̂ı ´ y ̂ is conservative and
find both its potential and its field lines.

1 2
18 2 m|v ( t )| is the kinetic energy and ´ϕ is the potential energy. See Warning 2.3.2.

69
V ECTOR F IELDS 2.3 C ONSERVATIVE V ECTOR F IELDS

(a) The potential: Our vector field F( x, y) = x ı̂ı ´ y ̂ is conservative if we can find a ϕ
obeying

Bϕ
( x, y) = x
Bx
Bϕ
( x, y) = ´y
By

B
Recall that, when taking the partial derivative Bx the coordinate y is viewed as a con-
stant. So the first of these equations is satisfied if and only if there is a ψ(y), which
does not depend on x, so that

x2
ϕ( x, y) = + ψ(y)
2

For this to also satisfy the second equation, we need

Bϕ B  x2 
´y = ( x, y) = + ψ ( y ) = ψ1 ( y )
By By 2

which is the case if and only if there is a constant C with

y2
ψ(y) = ´ +C
2

So, for any choice of the constant C,

x 2 y2
´ +C
2 2

is a potential. In particular, taking C = 0, one possible potential is

x 2 y2
ϕ( x, y) = ´
2 2

Some equipotential curves for this potential are sketched in (c) below. They are the
blue curves.

(b) The field lines (Optional): Recalling (2.2.5), the field lines of the vector field F( x, y) =
x ı̂ı ´ y ̂ are determined by

dx dy
= ðñ ´ydx = xdy
x ´y
ðñ xdy + ydx = 0
ðñ d( xy) = 0 by the product rule
ðñ xy = C

70
V ECTOR F IELDS 2.3 C ONSERVATIVE V ECTOR F IELDS

for some constant C. If you are not comfortable with the use of the product rule above,
here is another way to write the same computation.
dy y dy
= ´ ðñ x = ´y
dx x dx
dy
ðñ x +y = 0
dx
d
ðñ ( xy) = 0 by the product rule
dx
ðñ xy = C

Some field lines are sketched in (c) below. They are the red curves. Note that they
appear to cross the equipotential curves, the blue curves, at right angles. We shall see
in Lemma 2.3.6, below, that this is not a coincidence. Also note that, while the above
computation tells what the field lines are, it does not give us the direction of motion
along the field lines. We determine the direction of motion next.
(c) Direction of motion (Optional): The sign data
$ , $ ,
&ą 0 if x ą 0 /
’ . &ą 0
’ if y ă 0 /
.
ı̂ı ¨ F( x, y) = x = 0 if x = 0 ̂ ¨ F( x, y) = ´y = 0 if y = 0
’
% /
- ’
% /
-
ă 0 if x ă 0 ă0 if y ą 0

is visually displayed in the figure on the left below. The arrows in the figure on the
left gives us the direction of motion along the field lines (in red) in the figure on the
right below. Some equipotential curves are also sketched (in blue) in the figure on the
right below.

Example 2.3.5

We have just seen one example of a conservative vector field for which the field lines
appear to cross the equipotential curves at right angles. Here is a result which says that
that was no accident. The field lines of conservative fields always cross the equipotential
surfaces at right angles.

71
V ECTOR F IELDS 2.3 C ONSERVATIVE V ECTOR F IELDS

Lemma 2.3.6 (Optional).

If F is a conservative vector field, then the field lines of F are perpendicular to the
equipotential surfaces of F.

Proof. Let F = ∇ ϕ. Pick any point r0 and any nonzero vector
T that is tangent to the
equipotential surface at r0 . That equipotential surface is ϕ x, y, z = ϕ(r0 ). Consider any


curve r(t) = x (t), y(t), z(t) that

˝ lies in the equipotential surface of F through r0 , so that ϕ r(t) = ϕ(r0 ) for all t, and

also obeys
˝ r(0) = r0 and
˝ dr
dt (0) = T.


Differentiating ϕ r(t) = ϕ(r0 ) with respect to t and applying the chain rule gives

d 

ϕ x ( t ), y ( t ), z ( t ) = 0
dt
Bϕ
dx Bϕ
dy Bϕ
dz
x ( t ), y ( t ), z ( t ) (t) + x ( t ), y ( t ), z ( t ) (t) + x ( t ), y ( t ), z ( t ) (t) = 0
Bx dt By dt Bz dt
Bϕ Bϕ Bϕ

Notice that the left hand side is exactly the dot product of Bx , By , Bz = ∇ ϕ with
dx dy dz dr


dt dt dt = dt . So
, ,

dr
∇ ϕ r(t) ¨ (t) = 0
dt

dr
F r(t) ¨ (t) = 0
dt
Then set t = 0 to get
F r0 ¨ T = 0

This says that the vector T (which is tangent to the equipotential surface at r0 ) is perpen-
dicular to the vector field at r0 (which is a tangent vector to the field line of F through
r0 ).

Here is another example in which we try to find a potential for a vector field.
Example 2.3.7

Let’s try to find a potential for the vortex vector field v( x, y) = Ω ´ yı̂ı + x̂ of Example

2.1.3. The potential would have to obey

Bϕ
( x, y) = ´Ωy
Bx
Bϕ
( x, y) = Ωx
By

72
V ECTOR F IELDS 2.3 C ONSERVATIVE V ECTOR F IELDS

We proceed just as we did in Example 2.3.5. The first of these equations is satisfied if and
only if there is a ψ(y), which does not depend on x, so that

ϕ( x, y) = ´Ωxy + ψ(y)

For this to also satisfy the second equation, we need

Bϕ B  
Ωx = ( x, y) = ´ Ωxy + ψ(y) = ´Ωx + ψ1 (y) ðñ ψ1 (y) = 2Ωx
By By

If Ω ‰ 0, the right hand side of this equation depends on x while the left hand side
in
independent of x, no matter what ψ is. So no ψ can work, and v( x, y) = Ω ´ yı̂ı + x̂ is
not conservative.
Example 2.3.7

The previous example shows that not all vector fields are conservative. That answers
the first of the questions that we posed just before Example 2.3.5. The next theorem pro-
vides a simple screening test for conservativeness, which partially answers the second
question. The easy way to remember the screening test uses the curl, which we now de-
fine.

Definition 2.3.8.

The curl of a vector field F( x, y, z) is denoted by ∇ ˆ F( x, y, z) and is defined by

 BF BF2   BF BF3   BF BF 
3 1 2
∇ˆF = ´ ı̂ı + ´ ̂ + ´ 1 k̂
By Bz Bz Bx Bx By
ı̂ı
 
̂ k̂
B B B
= det  Bx By Bz

F1 ( x, y, z) F2 ( x, y, z) F3 ( x, y, z)

The determinant in the second row is really just a mnemonic device used to make
it easy to remember the expression after the equals sign in the first row. One must
be careful about the signs in this definition — the determinant helps with that.

73
V ECTOR F IELDS 2.3 C ONSERVATIVE V ECTOR F IELDS

Theorem 2.3.9 (Screening test for conservative vector fields.).

(a) Assume that F1 ( x, y) and F2 ( x, y) are continuously differentiable. If the vector

field F1 ( x, y)ı̂ı + F2 ( x, y)̂ is conservative, then we must have

BF1 BF
= 2
By Bx

(b) Assume that F1 ( x, y, z), F2 ( x, y, z) and F3 ( x, y, z) are continuously differen-

tiable. If the vector field F1 ( x, y, z)ı̂ı + F2 ( x, y, z)̂ + F3 ( x, y, z)k̂ is conservative,
then
BF1 BF BF1 BF BF2 BF
= 2 = 3 = 3
By Bx Bz Bx Bz By
Equivalently,
 BF BF2   BF BF3   BF BF 
3 1 2
∇ˆF = ´ ı̂ı + ´ ̂ + ´ 1 k̂ = 0
By Bz Bz Bx Bx By

That is, F is curl free.

Proof. (a) If the vector field F1 ( x, y)ı̂ı + F2 ( x, y)̂ is conservative, then there is a potential
ϕ( x, y) such that
Bϕ
( x, y) = F1 ( x, y)
Bx
Bϕ
( x, y) = F2 ( x, y)
By
B B
Applying By to the first equation and Bx to the second equation gives

B2 ϕ BF
= 1
ByBx By
B2 ϕ BF
= 2
BxBy Bx
B2 ϕ B2 ϕ
Recall that, for any twice continuously differentiable function, ByBx = BxBy . So the two left
hand sides are equal, and the two right hand sides must also be equal.
(b) If the vector field F1 ( x, y, z)ı̂ı + F2 ( x, y, z)̂ + F3 ( x, y, z)k̂ is conservative, then there is a
potential ϕ( x, y, z) such that
Bϕ
( x, y, z) = F1 ( x, y, z)
Bx
Bϕ
( x, y, z) = F2 ( x, y, z)
By
Bϕ
( x, y, z) = F3 ( x, y, z)
Bz
We proceed just as in (a).

74
V ECTOR F IELDS 2.3 C ONSERVATIVE V ECTOR F IELDS

B B
• Applying By to the first equation and Bx to the second equation gives
$ 2 ,
& B ϕ = BF1 . BF1 BF2
ByBx By
ùñ =
2
% B ϕ = BF2 - By Bx
BxBy Bx

B B
• Applying Bz to the first equation and Bx to the third equation gives
$ 2 ,
& B ϕ = BF1 . BF1 BF
= 3
BzBx Bz
ùñ
% B2 ϕ = BF3 - Bz Bx
BxBz Bx

B B
• Applying Bz to the second equation and By to the third equation gives
$ 2 ,
& B ϕ = BF2 . BF2 BF
BzBy Bz
ùñ = 3
% B 2ϕ BF3 - Bz By
ByBz = By

Combining the three bullet points gives ∇ ˆ F = 0.

Warning 2.3.10.

As always, we have to be careful with the flow of logic19 . The screening test
in Theorem 2.3.9 is a one-way test. If, for example, BF
By ‰ Bx then the vector
1 BF2

field F cannot be conservative. But if BF1 BF2

By = Bx Theorem 2.3.9 does not guarantee
that F is conservative. In fact there are fields that are not conservative but do
obey BF 1 BF2
By = Bx . We’ll see one in Example 2.3.14, below. We shall later find some
BF1 BF2
additional regularity conditions which, when combined with By = Bx , do imply
conservativeness. See Theorem 2.4.8, below.

Example 2.3.11 (Example 2.3.7 revisited)

In Example 2.3.7, we attempted to find a potential for the vector field

v( x, y) = Ω ´ yı̂ı + x̂

In the end we showed that, if Ω ‰ 0, no potential could exist, i.e. v( x, y) is not conserva-
tive. Had we known the screening test of Theorem 2.3.9.a, we could have concluded that
v( x, y) is not conservative by simply observing that
Bv1 B
= ´ Ωy)= ´Ω
By By
Bv2 B
= Ωx ) = +Ω
Bx Bx

19 Use your favourite search engine to look up a list of common logical errors. One is “affirming the
consequent”. An example would be concluding that because Shakespeare is dead, Elvis, who is also
dead, must also be Shakespeare.

75
V ECTOR F IELDS 2.3 C ONSERVATIVE V ECTOR F IELDS

are not equal, unless Ω = 0. But Ω = 0 makes a rather boring vector field.
Example 2.3.11

Example 2.3.12

Determine whether or not the vector field

F( x, y, z) = yı̂ı ´ z̂ + x k̂

is conservative. If it is conservative, find a potential.

Solution. Let’s start by applying the screening test Theorem 2.3.9.b. Since

ı̂ı
 
̂ k̂
∇ ˆ F = det  Bx
B B
By
B 
Bz = ı̂ı ´ ̂ ´ k̂
y ´z x

is not 0, the vector field F cannot be conservative.

Example 2.3.12

Example 2.3.13

Determine whether or not the vector field

F( x, y, z) = (y2 + 2xz2 ´ 1)ı̂ı + (2x + 1)y ̂ + (2x2 z + z3 )k̂

is conservative. If it is conservative, find a potential.

Solution. Again start by applying the screening test Theorem 2.3.9.b. This time

ı̂ı
 
̂ k̂
∇ ˆ F = det  B
Bx
B
By
B
Bz
 = 0ı̂ı ´ (4xz ´ 4xz)̂ + (2y ´ 2y)k̂
y2 + 2xz2 ´ 1 (2x + 1)y 2x2 z + z3
=0

So F passes the screening test. Let’s look for a function ϕ( x, y, z) obeying

Bϕ
( x, y, z) = y2 + 2xz2 ´ 1
Bx
Bϕ
( x, y, z) = (2x + 1)y (˚)
By
Bϕ
( x, y, z) = 2x2 z + z3
Bz
B
The partial derivative Bx treats y and z as constants. So ϕ( x, y, z) obeys the first equation
if and only if there is a function ψ(y, z) with

ϕ( x, y, z) = xy2 + x2 z2 ´ x + ψ(y, z)

76
V ECTOR F IELDS 2.3 C ONSERVATIVE V ECTOR F IELDS

This ϕ( x, y, z) will also obey the second equation if and only if

B Bψ
xy2 + x2 z2 ´ x + ψ(y, z) = (2x + 1)y ðñ 2xy +


(y, z) = (2x + 1)y
By By
Bψ
ðñ (y, z) = y
By
y2
ðñ ψ(y, z) = + ζ (z)
2
for some function ζ (z) which depends only on z. At this stage we know that
y2
ϕ( x, y, z) = xy2 + x2 z2 ´ x + ψ(y, z) = xy2 + x2 z2 ´ x + + ζ (z)
2
obeys the first two equations of (˚), for any function ζ (z). Finally to have the third equation
of (˚) also satisfied, we also need to chose ζ (z) to obey
B y2
xy2 + x2 z2 ´ x + + ζ (z) = 2x2 z + z3 ðñ 2x2 z + ζ 1 (z) = 2x2 z + z3


Bz 2
ðñ ζ 1 (z) = z3
z4
ðñ ζ (z) =
+C
4
for any constant C. So one possible potential, namely that with C = 0 is
y2 z4
ϕ( x, y, z) = xy2 + x2 z2 ´ x + +
2 4
Note, as a check20 , that
∇ ϕ( x, y, z) = y2 + 2xz2 ´ 1 ı̂ı + 2xy + y)̂ + 2x2 z + z3 k̂

as desired.
Example 2.3.13

 
y x 
Example 2.3.14 Optional: First look at ´ x2 +y2 ı̂ı + x 2 + y2
̂

Now is a good time to reread the Warning 2.3.10. In this example we will show that the
vector field
y x
F( x, y) = ´ 2 2
ı̂ı + 2 ̂ defined for all ( x, y) in R2 except ( x, y) = (0, 0)
x +y x + y2
passes the screening test of Theorem 2.3.9.a. We will also begin to see why it is not conser-
vative on the domain R2 zt(0, 0)u. To verify the screening test, we compute
B  y  ( x2 + y2 ) ´ y(2y) y2 ´ x 2
´ 2 = ´ =
By x + y2 ( x 2 + y2 )2 ( x 2 + y2 )2
B  x  ( x2 + y2 ) ´ x (2x ) y2 ´ x 2
= =
Bx x2 + y2 ( x 2 + y2 )2 ( x 2 + y2 )2

20 It is always worth doing this check.

77
V ECTOR F IELDS 2.3 C ONSERVATIVE V ECTOR F IELDS

and observe that the two right hand sides are identical. So the screening test is passed.
In order for F to be conservative on the domain R2 zt(0, 0u, there must exist a function
Bϕ Bϕ
ϕ( x, y), that, together with both partial derivatives Bx ( x, y) and By ( x, y), is defined for all
( x, y) in R2 except ( x, y) = (0, 0), and obeys
y
Bϕ y ´ x2 B  y
( x, y) = ´ 2 = = arctan
Bx x + y2 1+ x
y
2 Bx x
1
Bϕ x x B  y
( x, y) = = = arctan
By x 2 + y2 1+ x
y
2 By x

by the chain rule, because

B y y B y 1
=´ 2 =
Bx x x By x x
y
It looks like we have found a potential, namely arctan x . But there is a problem. Recall
y y
that, by definition, arctan x is an angle θ ( x, y) that obeys tan θ ( x, y) = arctan x ; but for
y
any ( x, y) P R2 zt(0, 0u there are infinitely many angles having the tangent x . To define
ϕ( x, y) we have to select exactly one such angle. It is impossible to do so in such a way
that ϕ( x, y) is continuous on all of R2 zt(0, 0u.
To see why, fix any r ą 0, and imagine that you are walking on the circle x2 + y2 = r2
y
in the xy-plane. At time θ, you are at x = r cos θ, y = r sin θ and then x = tan θ and you
are allowed to define ϕ( x, y) = θ + kπ, for any integer k.
Suppose that at time θ = 0 you choose k = 0. That is, you choose ϕ(r, 0) = 0. Now
start walking, choosing an allowed ϕ( x, y), i.e. choosing a k, for each point ( x, y) that
you cross. Because ϕ( x, y) has to vary continuously21 with ( x, y), you have to continue
choosing k = 0. But you run off a cliff as θ approaches 2π, because then
˝ you are approaching (r, 0) from below, as in the figure below, and
˝ because you are choosing k = 0, ϕ( x, y) is just a little less than 2π, but
˝ you have already chosen ϕ(r, 0) = 0, not 2π. So ϕ( x, y) has a jump discontinuity22
along the positive x-axis.

y
ϕ “ 5π{8 ϕ “ 3π{8
ϕ “ 6π{8 ϕ “ 2π{8
ϕ “ 7π{8 ϕ “ π{8
ϕ“π ϕ“0
x
ϕ “ 9π{8 ϕ “ 15π{8
ϕ “ 10π{8 ϕ “ 14π{8
ϕ “ 11π{8 ϕ “ 13π{8

21 If ϕ( x, y) is not continuous, its gradient does not exist, and ϕ cannot be a potential.
22 Those who have taken some complex analysis may recognize this as the branch cut in ln z.

78
 V ECTOR F IELDS 2.4 L INE I NTEGRALS

If you are having trouble following this argument, don’t worry about it. We will return
with a less hand-wavy argument later.
Example 2.3.14

2.4IJ Line Integrals

We have already seen, in §1.6, one type of integral along curves. We are now going to see
a second, that turns out to have significant connections to conservative vector fields. It
arose from the concept of “work” in classical mechanics.
Suppose that we wish to find the work done by a force F(r) moving a particle along
a path r(t). During the “infinitesimal time interval23 ” from t to t + dt the particle moves
from r(t) to r(t) + dr with dr = drdt ( t ) dt. By definition, the work done during that in-
finitesimal time interval is

dr
F r(t) ¨ dr = F r(t) ¨ (t) dt


dt

The total work done during the time interval from t0 to t1 is then
ż t1

dr
Work = F r(t) ¨ (t) dt
t0 dt

There are some useful shorthand notations for this work.

Notation 2.4.1.

Denote by C the parametrized path r(t) with t0 ď t ď t1 . Then

ż ż ż t1

dr
F ¨ dr = F r(t) ¨ (t) dt


F1 dx + F2 dy + F3 dz =
C C t0 dt
ű
If C is a closed path, the notation C F ¨ dr is also used.

In the event that F is conservative, and we know the potential ϕ, the following theorem
provides a really easy way to compute “work integrals”. The theorem is a generalization
of the fundamental theorem of calculus, and indeed some people call it the fundamental
theorem of line integrals.

23 Yes, yes. We should first consider short time intervals ∆t ą 0 and then take the limit ∆t Ñ 0 at the
end. But you have undoubtedly used this type of argument so many times before that you would be
thoroughly bored by it.

79
V ECTOR F IELDS 2.4 L INE I NTEGRALS

Theorem 2.4.2.

Let F = ∇ ϕ be a conservative vector field. Then if C is any curve that starts at P0

and ends at P1 , we have24
ż
F ¨ dr = ϕ( P1 ) ´ ϕ( P0 )
C

Proof. Let r(t) = x (t), y(t), z(t) , t0 ď t ď t1 , be any parametrization of C with r(t0 ) = P0

and r(t1 ) = P1 . Then, by definition,

ż ż t1 ż t1

dr
dr
F ¨ dr = F r(t) ¨ (t) dt = ∇ ϕ r(t) ¨ (t) dt
C t0 dt t0 dt
ż t1 h
Bϕ
dx Bϕ
dy
= x ( t ), y ( t ), z ( t ) (t) + x ( t ), y ( t ), z ( t ) (t)
t0 Bx dt By dt
Bϕ
dz i
+ x ( t ), y ( t ), z ( t ) (t) dt
Bz dt
ż t1 h
d
i
= ϕ x (t), y(t), z(t) dt by the chain rule in reverse
t0 dt
= ϕ r(t1 ) ´ ϕ r(t0 ) = ϕ( P1 ) ´ ϕ( P0 )

by the fundamental theorem of calculus.

ş
Observe that, in Theorem 2.4.2, the value, ϕ( P1 ) ´ ϕ( P0 ), of the integral C F ¨ dr de-
pended only on the endpoints P0 and P1 of the curve, not on the path that the curve
followed to get to P0 from P1 . We shall see, in Theorem 2.4.7, below, that this happens
only for conservative vector fields. Here are several examples of line integrals of vector
fields that are not conservative.
Example 2.4.3

Set P0 = (0, 0), P1 = (1, 1) and25

F( x, y) = xy ı̂ı + (y2 + 1) ̂

We shall consider three curves, all starting at P0 and ending at P1 .

(a) Let C1 be the straight line from P0 to P1 .

(b) Let C2 be the path, made from two straight lines, which follows the x-axis from P0 to
(1, 0) and then follows the line x = 1 from (1, 0) to P1 .

(c) Let C3 be the part of the parabola x = y2 from P0 to P1 .

24 So ϕ acts a bit like the antiderivative of first year calculus.

25 The reader should check that this vector field is not conservative.

80
V ECTOR F IELDS 2.4 L INE I NTEGRALS

y
P1
C3

C2
C1

P0 C2 x

ş
We shall calculate the work Ci F ¨ dr for each of the curves.
(a) We parametrize C1 by r(t) = t ı̂ı + t ̂ with t running from 0 to 1. Then x (t) = t and
y(t) = t so that
dr
F r(t) = t2 ı̂ı + (t2 + 1) ̂


and (t) = ı̂ı + ̂
dt
so that
ż ż1 ż1

dr
F ¨ dr = F r(t) ¨ (t) dt = t2 ı̂ı + (t2 + 1) ̂ ¨ [ı̂ı + ̂ ] dt
 
C1 0 dt 0
ż1
 2 
= 2t + 1 dt
0
5
=
3
(b) We split C2 into two parts, C2,x running from P0 to (1, 0) along the x-axis and then C2,y
running from (1, 0) to P1 along the line x = 1. We parametrize C2,x by r( x ) = x ı̂ı with
x running from 0 to 1 and C2,y by r(y) = ı̂ı + y ̂ with y running from 0 to 1. Then26
ż ż ż
F ¨ dr = F ¨ dr + F ¨ dr
C2 C2,x C2,y
ı̂ı
hkkkikkkj ̂
hkkkkkikkkkkj
ż1 ż1
 d  d
( x )(0) ı̂ı + (02 + 1) ̂ ¨ (1)(y) ı̂ı + (y2 + 1) ̂ ¨



= x ı̂ı dx + ı̂ı + y ̂ dy
0 dx 0 dy
ż1 ż1
y2 + 1 dy


= 0 dx +
0 0
4
=
3

(c) We parametrize C3 by r(t) = t2 ı̂ı + t ̂ with t running from 0 to 1. Then x (t) = t2 and
y(t) = t so that
dr
F r(t) = t3 ı̂ı + (t2 + 1) ̂


and (t) = 2t ı̂ı + ̂
dt

26 You might like to think about why we can split up the integral like this.

81
V ECTOR F IELDS 2.4 L INE I NTEGRALS

so that
ż ż1 ż1
3 2
F ¨ dr = t ı̂ı + (t + 1) ̂ ¨ [2t ı̂ı + ̂ ] dt = 2t4 + t2 + 1 dt
   
C3 0 0
2 1 26
= + +1 =
5 3 15

Note that,ş despite şthe fact that şC1 , C2 and C3 all start at P0 and all end at P1 , the three
integrals C1 F ¨ dr, C2 F ¨ dr and C3 F ¨ dr all have different values.
Example 2.4.3

Example 2.4.4

Set27
F( x, y) = 2y ı̂ı + 3x ̂
This time we consider two curves.

(a) Let C1 be circle x2 + y2 = 1, traversed once counterclockwise, starting at (1, 0).

(b) Let C2 be (trivial) curve which just consists of the single point (1, 0).
ş
We shall calculate the work C F ¨ dr for each curve.
i

(a) We parametrize C1 by r(t) = cos t ı̂ı + sin t ̂ with t running from 0 to 2π, just as we did
in Example 1.0.1. Then
¿ ż 2π ż 2π
F ¨ dr = 2 sin t ı̂ı + 3 cos t ̂ ¨ [´ sin t ı̂ı + cos t ̂ ] dt = ´ 2 sin2 t + 3 cos2 t dt
   
0 0
C1

You could evaluate these integrals using double angle trig identities like you did
in first year calculus. But there is a sneaky, much easier, way. Because sin2 t and
cos2 t are translates of each other, and both are periodic of period π, the two integrals
ş2π 2
ş2π 2
0 sin t dt and 0 cos t dt represent the same area and so are equal. See the figure
below.

y y
y “ sin x 2
1
y “ cos2 x
1

x x
π{2 π 3π{2 2π π{2 π 3π{2 2π

27 Again, the reader should verify that this vector field is not conservative.

82
V ECTOR F IELDS 2.4 L INE I NTEGRALS

Thus
ż 2π ż 2π ż 2π ż 2π
2 2 1 2 1
sin t + cos2 t dt =

sin t dt = cos t dt = dt = π
0 0 0 2 2 0

and
¿ ż 2π ż 2π
2
F ¨ dr = ´2 sin t dt + 3 cos2 t dt = π
0 0
C1

dr
ş
(b) We parametrize C2 by r(t) = ı̂ı for all t. Then dt ( t ) = 0 and C2 F ¨ dr = 0.

Again,
ş ş the fact that C1 and C2 both start at (1, 0) and end at (1, 0), the two integrals
despite
C1 F ¨ dr and C2 F ¨ dr are different.
Example 2.4.4

Example 2.4.5 (Example 2.3.14, again.)

In Example 2.3.14, we saw that the vector field

y x
F( x, y) = ´ ı
ı̂ + ̂ defined for all ( x, y) in R2 except ( x, y) = (0, 0)
x 2 + y2 x 2 + y2

passed the screening test of Theorem 2.3.9.a, and yet was not conservative. In this exam-
ple, we will see that this F violates the conclusion of Theorem 2.4.2, thereby providing a
second proof that F( x, y) is not conservative on R2 with (0, 0) removed. For the curve C ,
of Theorem 2.4.2, we use the circle parametrized by x = a cos θ, y = a sin θ, 0 ď θ ď 2π.
Then dx = ´a sin θ dθ and dy = a cos θ dθ so that
ż ż 2π ż 2π
1 x dy ´ y dx 1 a2 cos2 θ dθ + a2 sin2 θ dθ 1
2 2
= 2
= dθ = 1
2π C x +y 2π 0
2 2 2
a cos θ + a sin θ 2π 0

The curve C has initial point

ˇ
P0 = ( a cos θ, a sin θ )ˇθ =0 = ( a, 0)

and final point

ˇ
P1 = ( a cos θ, a sin θ )ˇθ =2π = ( a, 0) = P0

So, if F were conservative with potential ϕ, Theorem 2.4.2 would give that
ż
1 x dy ´ y dx
= ϕ( P1 ) ´ ϕ( P0 ) = 0
2π C x2 + y2

Consequently, F can’t be conservative.

Example 2.4.5

83
 V ECTOR F IELDS 2.4 L INE I NTEGRALS

2.4.1 §§ Path Independence

This brings us to the following question. Let F be any fixed vector field. When is it true
that, given any two fixed points P0 and P1 , the integrals
ż ż
F ¨ dr = F ¨ dr
C C1

for all curves C , C 1 that startşat P0 and end at P1 ? When can we ignore the path taken? If
this is the case we say that “ C F ¨ dr is independent of the path chosen” and we write
ż P1 ż
F ¨ dr = F ¨ dr
P0 C

for any path C from P0 to P1 . The point of this section is that there is an intimate relation
between path independence and conservativeness of vector fields, that we will get to in
Theorem 2.4.7.
For simplicity, we will consider only vector fields that are defined and continuous on
all of R2 (i.e. the xy-plane) or R3 (i.e. the usual three dimensional world). Some discussion
about what happens for vector fields that are defined only on part of R2 or R3 is given in
the optional §4.5.
First we show that if there is one pair of (not necessarily distinct) points P0 , P1 such
that ż ż
F ¨ dr = F ¨ dr
C1 C2
for all curves C1 , C2 that start at P0 and end at P1 , then it is also true that, for any other pair
of points P01 , P11 ż ż
F ¨ dr = F ¨ dr
C11 C21

for all curves C11 , C21 that start at P01 and end at P11 . This might seem unlikely at first, but the
idea of the proof is really intuitive.

Theorem 2.4.6.

Let F be a vector field that is defined and continuous on all of R2 (or R3 ). Let P0 ,
P1 , P01 , P11 be any four points in R2 (or R3 ). Assume that
ż ż
F ¨ dr = F ¨ dr
C1 C2

for all curves C1 , C2 that start at P0 and end at P1 . Then

ż ż
F ¨ dr = F ¨ dr
C11 C21

for all curves C11 , C21 that start at P01 and end at P11 .

84
V ECTOR F IELDS 2.4 L INE I NTEGRALS

Proof. Let C11 and C21 be any two curves that start at P01 and end at P11 . We start by choosing

P1
C11
P01 Cr
Cℓ P11
P0 C21

any two (auxiliary) curves

˝ C` that starts at P0 and ends at P01 and

˝ Cr that starts at P11 and ends at P1 .

and then we define the curves

˝ C1 to be C` , followed by C11 , followed by Cr and

˝ C2 to be C` , followed by C21 , followed by Cr .

Then both C1 and C2 start at P0 and end at P1 , so that, by hypothesis,

ż ż
F ¨ dr = F ¨ dr
C1 C2

and, from the construction of C1 and C2 ,

ż ż ż ż ż ż
F ¨ dr + F ¨ dr + F ¨ dr = F ¨ dr + F ¨ dr + F ¨ dr
C` C1 Cr C` C21 Cr
ż ż1
ùñ F ¨ dr = F ¨ dr
C11 C21

as desired.

We are now ready for our main theorem on conservative fields.

Theorem 2.4.7.

Let F be a vector field that is defined and continuous on all of R2 (or R3 ). Then
the following three statements are equivalent.

(a) F is conservative. That is, there exists a function ϕ such that F = ∇ ϕ.

ű
(b) The integral C F ¨ dr = 0 for any closed curve C .
ş
(c) The integral
ş F ¨şdr is path independent. That is, for any points P0 , P1 we
have C1 F ¨ dr = C2 F ¨ dr for all curves C1 , C2 that start at P0 and end at P1 .

That is, if any one of the three statements are true, then all three are true.

85
V ECTOR F IELDS 2.4 L INE I NTEGRALS

Proof. It suffices for us to prove28 that

˝ the truth of (a) implies the truth of (b) and
˝ the truth of (b) implies the truth of (c) and
˝ the truth of (c) implies the truth of (a).
That’s exactly what we will do.
(a) ùñ (b): Let C be a closed curve that starts at P0 and then ends back at P0 . Then, by
Theorem 2.4.2 with P1 = P0 ,
¿
F ¨ dr = ϕ( P0 ) ´ ϕ( P0 ) = 0
C
ű
(b) ùñ (c): Pick any point P0 and set P1 = P0 . Then we ş are assuming that C F ¨ dr = 0 for
all curves that start at P0 and end at P1 . In particular C F ¨ dr takes the same value for all
curves that start at P0 and end at P1 . So Theorem 2.4.6 immediately yields property (c).
(c) ùñ (a): We are to show that F is conservative. We’ll start by guessing ϕ and then we’ll
verify that, for our chosen ϕ, we really do have F = ∇ ϕ. Our guess for ϕ is motivated
by
ş Theorem 2.4.2. If our F really is conservative, its potential is going to have to obey
C F ¨ dr = ϕ ( P1 ) ´ ϕ ( P0 ) for any curve C that starts at P0 and ends at P1 . Let’s choose
P0 = 0. Remembering, from Definition 2.3.1.a, that adding a constant to a potential
ş always
yields another potential, we can always choose ϕ(0) = 0. Then ϕ( P1 ) = şC F ¨ dr for any
curve C that starts at 0 and ends at P1 . So define, for each point x, ϕ(x) = C F ¨ dr for any
curve C that starts at 0 and ends at x.
We now verify that, for this chosen ϕ, we really do have F = ∇ ϕ. Fix any point x and
any curve Cx that starts at the origin and ends at x. For any vector u, let Du be the curve
with parametrization
ru (t) = x + tu 0ďtď1
This curve is a line segment that starts at x at t = 0 andş ends at x + u at t = 1. Observe
that r 1u (t) = u. Recall that, by assumption, ϕ(x + su) = C F ¨ dr for any curve C that starts
at 0 and ends at x + su. So ż
ϕ(x + su) = F ¨ dr
Cx +Dsu
where Cx + Dsu is the curve which first follows Cx from the origin to x and then follows
Dsu from x to x + su. We have
ż ż ż
F ¨ dr = F ¨ dr + F ¨ dr
Cx + Dsu Cx Dsu
ż ż1
= F ¨ dr + F(x + tsu) ¨ (su) dt
Cx 0

In the second integral, make the change of variables τ = ts, dτ = sdt. This gives
ż żs
ϕ(x + su) = F ¨ dr + F(x + τu) ¨ u dτ
Cx 0

28 This is a pretty efficient, and standard, way to structure the proof of the equivalence of three statements.

86
V ECTOR F IELDS 2.4 L INE I NTEGRALS

By the fundamental theorem of calculus, applied to the second integral,

d ˇ ˇ
ˇ ˇ
ϕ(x + su)ˇ = F(x + su) ¨ uˇ = F(x) ¨ u
ds s =0 s =0

Applying this with u = ı̂ı , ̂ , k̂ gives us

 Bϕ Bϕ Bϕ 
(x) = F(x) ¨ ı̂ı , F(x) ¨ ̂ , F(x) ¨ k̂


(x) , (x) ,
Bx By Bz
which is
∇ ϕ(x) = F(x)
as desired.
Using this result, we can completely characterize conservative fields on R2 and R3 .

Theorem 2.4.8.

Let F be a vector field that is defined and has continuous first order partial deriva-
tives on all of R2 (or R3 ). Then F is conservative if and only if it passes the
screening test ∇ ˆ F = 0, i.e. is curl free.

Warning 2.4.9.

Note that in Theorem 2.4.8 we are assuming that F passes the screening test on
all of R2 or R3 . We have already seen, in Example 2.3.14, that if the screening test
fails at even a single point, for example because the vector field is not defined at
that point, then F need not be conservative. We’ll explore what happens in such
cases in the (optional) §4.5. We’ll see that something can be salvaged.

Proof of Theorem 2.4.8. We’ll give the proof for the R2 case. The proof for the R3 case is
very similar. We have already seen, in Theorem 2.3.9, that if F is conservative, then it
passes the screening test and there is nothing more to do.
So we now have to assume that F obeys BF By ( x, y ) = Bx ( x, y ) on all of R and prove
1 BF2 2

that it is conservative. We’ll do so using the strategy of Example 2.3.13 to find a function
ϕ( x, y), that obeys
Bϕ
( x, y) = F1 ( x, y)
Bx
Bϕ
( x, y) = F2 ( x, y)
By
B
The partial derivative Bx treats y as a constant. So ϕ( x, y) obeys the first equation if and
only if there is a function ψ(y) with
żx
ϕ( x, y) = F1 ( X, y) dX + ψ(y)
0

87
 V ECTOR F IELDS 2.5 O PTIONAL — T HE P ENDULUM

This ϕ( x, y) will also obey the second equation if and only if

Bϕ
F2 ( x, y) = ( x, y)
By
ż
B  x 
= F1 ( X, y) dX + ψ(y)
By 0
żx
BF1
= ( X, y) dX + ψ1 (y)
0 By

So we have to find a ψ(y) that obeys

żx
BF1
ψ (y) = F2 ( x, y) ´
1
( X, y) dX
0 By

This looks bad — no matter what ψ(y) is, the left hand side is independent of x, while
it looks like the right hand side depends on x. Fortunately our screening test hypothesis
now rides in to the rescue29 . (We haven’t used it yet, and it has to come in somewhere.)
żx żx
BF1 BF2
F2 ( x, y) ´ ( X, y) dX = F2 ( x, y) ´ ( X, y) dX
0 By 0 Bx
ˇX = x
ˇ
= F2 ( x, y) ´ F2 ( X, y)ˇ
X =0
= F2 (0, y)

In going from the first line to the second line we used the fundamental theorem of calculus.
So choosing ż y
ψ(y) = F2 (0, Y ) dY + C
0
for any constant C, does the trick.

2.5IJ Optional — The Pendulum

Model a pendulum by a mass m that is connected to a hinge by an idealized rod that is
massless and of fixed length `. Denote by θ the angle between the rod and vertical. The

̂

θ ı̂ı
ℓ
τ βℓ dθ
dt θ
ℓ cos θ ℓ

mg ℓ sin θ

29 or bails us out, or saves our bacon, or . . .

88
V ECTOR F IELDS 2.5 O PTIONAL — T HE P ENDULUM

forces acting on the mass are

˝ gravity, which has magnitude mg and direction (0, ´1),

˝ tension in the rod, whose magnitude, τ, automatically adjusts itself so that the dis-
tance between the mass and the hinge is fixed at ` and whose direction, (´ sin θ, cos θ ),
is always parallel to the rod and
˝ possibly some frictional forces, like friction in the hinge and air resistance. We shall
assume that the total frictional force has magnitude proportional to the speed30 of
the mass and has direction opposite to the direction of motion of the mass.

If we use a coordinate system
centered on the hinge, the ( x, y) coordinates of the mass at
time t are ` sin θ (t), ´ cos θ (t) . Hence its velocity vector is

d 
dθ
` sin θ (t), ´ cos θ (t) = ` cos θ (t), sin θ (t)


v(t) = (t)
dt dt

and the total frictional force is ´β`(cos θ, sin θ ) dθ

dt , for some constant β. The acceleration
vector of the mass is

d d2 θ  dθ 2
a(t) = v(t) = `(cos θ, sin θ ) 2 + `(´ sin θ, cos θ )
dt dt dt
so that Newton’s law of motion, F = ma, now tells us

d2 θ  dθ 2
ma(t) = m`(cos θ, sin θ ) + m `( ´ sin θ, cos θ )
dt2 dt
dθ
= F = mg(0, ´1) + τ (´ sin θ, cos θ ) ´ β`(cos θ, sin θ )
dt
To eliminate the (unknown) coefficient τ we dot this equation with (cos θ, sin θ ), which
extracts the component parallel to the direction of motion of the mass. Dotting with
2
(cos θ, sin θ ) gives m` ddt2θ = ´mg sin θ ´ β` dθ
dt or

d2 θ β dθ g
2
+ + sin θ = 0
dt m dt `
which is the equation of motion of the (nonlinear) pendulum. In general, it can be hard
to analyse nonlinear differential equations. But if the amplitude of oscillation is small
enough that we may approximate sin θ by θ we get the equation of motion of the linear
pendulum31 which is

d2 θ β dθ g
2
+ + θ=0
dt m dt `

30 The dependence of air resistance (drag) on the speed v is relatively complex. At low speed drag tends
to be approximately proportional to v, while at high speed it tends to be approximately proportional to
v2 .
2 g
31 When β = 0, this equation reduces to the equation ddt2θ + ` θ = 0, which occurs in many different
applications, and whose solutions exhibit simple harmonic motion.

89
V ECTOR F IELDS 2.5 O PTIONAL — T HE P ENDULUM

These equations may be reformulated as systems of first order ordinary differential equa-
tion, that is as equations for the flow lines of a vector field, by the simple expedient of
defining (as we did in Example 2.1.4)

x (t) = θ (t) y(t) = θ 1 (t)

d2 θ
Then, for the full, nonlinear, equation dt2
+ β dθ
m dt + g` sin θ = 0

x1 (t) = θ 1 (t) = y(t)

g β
y1 (t) = θ 2 (t) = ´ sin x (t) ´ y(t)
` m
The solutions of this first order system of ordinary differential equations are flow lines for
the vector field

 g β 
V ( x, y) = y , ´ sin x ´ y
` m
When β = 0, this is exactly the vector field of Example 2.1.4.

90
 Chapter 3

S URFACE I NTEGRALS

3.1IJ Parametrized Surfaces

For many applications we will need to use integrals over surfaces. One obvious one is just
computing surface areas. Another is computing the rate at which fluid traverses a surface.
The first step is to simply specify surfaces carefully.
There are three common ways to specify a surface in three dimensions.
(a) Graph of a function: Probably the most common way to specify a surface is to give its
equation in the form
z = f ( x, y) ( x, y) P D Ă R2
Here “( x, y) P D Ă R2 ” just means that ( x, y) runs over the subset D of R2 . For
example, if the surface is the top half of the sphere of radius one centred on the origin
b
z = 1 ´ x 2 ´ y2 with x2 + y2 ď 1

(b) Implicitly: We can also specify that the surface is the set of points ( x, y, z) that satisfy
the equation G ( x, y, z) = 0, or, more generally1 , satisfy the equation G ( x, y, z) = K,
with K a constant. For example, the sphere of radius one centred on the origin is the
set of points that obey
x 2 + y2 + z2 = 1
We shall explore this surface a little more in Example 3.1.2 below.

(c) Range of a function: Probably the most useful way to specify a surface, when one needs
to integrate over the surface, is as the range of a function

r : D Ă R2 Ñ R3
(u, v) P D ÞÑ r(u, v) = x (u, v) , y(u, v) , z(u, v)

1 Of courses we can always convert the equation G ( x, y, z) = K into H ( x, y, z) = 0 with H ( x, y, z) =

G ( x, y, z) ´ K. But it often more convenient to use G ( x, y, z) = K.

91
S URFACE I NTEGRALS 3.1 PARAMETRIZED S URFACES

The upper line means that r is a function which is defined on the subset D of R2 and
which assigns to each point on D a point in R3 . The second line means that the func-

tion r assigns to the element (u, v) of D the element r(u, v) = x (u, v) , y(u, v) , z(u, v)
in R3 . Such a surface is called a parametrized surface — each point of the surface is
labelled by the values of the two parameters u and v. Parametrized surfaces are of
course the two parameter analog of parametrized curves. Examples of parametrized
surfaces come next.

Example 3.1.1

One simple, even trivial, way to parametrize the surface which is the graph

z = f ( x, y) ( x, y) P D Ă R2

is to choose x and y as the parameters. That is, to choose

r(u, v) = u, v, f (u, v) , (u, v) P D ( x, y) P D

or r( x, y) = x, y, f ( x, y) ,

Example 3.1.1
Let’s do something a bit more substantial.
Example 3.1.2 (Sphere)

The sphere of radius 1 centred on the origin is the set of points ( x, y, z) that obey

G ( x, y, z) = x2 + y2 + z2 = 1

We cannot express this surface as the graph of a function because, for each ( x, y) with
x2 + y2 ă 1, there are two z’s that obey x2 + y2 + z2 = 1, namely
b
z = ˘ 1 ´ x 2 ´ y2

On the other hand, locally, this surface is the graph of a function. This means that, for any
point ( x0 , y0 , z0 ) on the sphere, all points of the surface that are sufficiently near ( x0 , y0 , z0 )
can be expressed in one of the forms z = f ( x, y)?or x = g(y, z), or y = h( x, z). For example,
the part of the sphere that is within a distance 2 of the point (0, 0, 1) is
ˇ ? (
( x, y, z) ˇ x2 + y2 + z2 = 1, |( x, y, z) ´ (0, 0, 1)| ă 2
ˇ (
= ( x, y, z) ˇ x2 + y2 + z2 = 1, x2 + y2 + (z ´ 1)2 ă 2
ˇ (
= ( x, y, z) ˇ x2 + y2 + z2 = 1, x2 + y2 + z2 ´ 2z + 1 ă 2
ˇ (
= ( x, y, z) ˇ x2 + y2 + z2 = 1, z ą 0
ˇ b (
ˇ
= ( x, y, z) z = 1 ´ x2 ´ y2 , x2 + y2 ă 1

This is illustrated in the figure below which shows the y = 0 section of the sphere?x2 +
y2 + z2 = 1 and also the y = 0 section of the set of points that are within a distance 2 of
(0, 0, 1). (They are the points inside the dashed circle.)

92
S URFACE I NTEGRALS 3.1 PARAMETRIZED S URFACES

x2 ` y 2 ` pz ´ 1q2 “ 2
p0, 0, 1q

p1, 0, 0q x
x2 ` y 2 ` z 2 “ 1

Similarly, as illustrated
? schematically in the figure below, the part of the sphere that is
within a distance 2 of the point (1, 0, 0) is
ˇ ? (
( x, y, z) ˇ x2 + y2 + z2 = 1, |( x, y, z) ´ (1, 0, 0)| ă 2
ˇ (
= ( x, y, z) ˇ x2 + y2 + z2 = 1, ( x ´ 1)2 + y2 + z2 ă 2
ˇ (
= ( x, y, z) ˇ x2 + y2 + z2 = 1, x2 ´ 2x + 1 + y2 + z2 ă 2
ˇ (
= ( x, y, z) ˇ x2 + y2 + z2 = 1, x ą 0
ˇ b (
= ( x, y, z) ˇ x = 1 ´ y2 ´ z2 , y2 + z2 ă 1

The figure below shows the y = 0 section of the sphere?x2 + y2 + z2 = 1 and also the y = 0
section of the set of points that are within a distance 2 of (1, 0, 0). (Again, they are the
points inside the dashed circle.)

z
p0, 0, 1q

p1, 0, 0q x
x2 ` y 2 ` z 2 “ 1 px ´ 1q2 ` y 2 ` z 2 “ 2

We can parametrize the unit sphere by using spherical coordinates, which you should
have seen before. As a reminder, here is a figure showing the definitions of the three
spherical coordinates

ρ = distance from ( x, y, z) to (0, 0, 0)

ϕ = angle between the line (0, 0, 0) ( x, y, z) and the z axis
θ = angle between the line (0, 0, 0) ( x, y, 0) and the x axis

93
S URFACE I NTEGRALS 3.1 PARAMETRIZED S URFACES

z
p0, 0, zq ρ sin ϕ
px, y, zq
ρ cos ϕ
ϕ ρ
y
px, 0, 0q θ
px, y, 0q
x ρ sin ϕ sin θ

and here are two more figures giving the side and top views of the previous figure.

z
ρ sin ϕ px, y, zq
p0, 0, zq y
θ ρ sin ϕ
ρ cos ϕ
ρ ρ sin ϕ cos θ
ϕ
px, y, 0q
px, 0, 0q
px, y, 0q ρ sin ϕ sin θ
side view x top view

From the figure, we see that Cartesian and spherical coordinates are related by

x = ρ sin ϕ cos θ
y = ρ sin ϕ sin θ
z = ρ cos ϕ

The points on the sphere x2 + y2 + z2 = 1 are precisely the set of points with ρ = 1. So we
can use the parametrization


r(θ, ϕ) = sin ϕ cos θ , sin ϕ sin θ , cos ϕ

Here is how to see that as ϕ runs over (0, π ) and θ runs over [0, 2π ), r(θ, ϕ) covers the
whole sphere x2 + y2 + z2 = 1 except for the north pole (ϕ = 0 gives the north pole for all
values of θ) and the south pole (ϕ = π gives the south pole for all values of θ).

˝ Fix θ and have ϕ run over the interval 0 ă ϕ ď π/2. Then r(θ, ϕ) traces out one
quarter of a circle starting at the north pole r(θ, 0) = (0, 0, 1) (but excluding the
north pole itself) and ending at the point r(θ, π/2) = (cos θ, sin θ, 0) in the xy-plane.

94
S URFACE I NTEGRALS 3.1 PARAMETRIZED S URFACES

ϕ
y
θ

• Keep θ fixed at the same value and extend the interval over which ϕ runs to 0 ă ϕ ă
π. Now r(θ, ϕ) traces out a semi-circle starting at the north pole r(θ, 0) = (0, 0, 1),
ending at the south pole r(θ, π ) = (0, 0, ´1) (but excluding both the north and south
poles themselves) and passing through the point r(θ, π/2) = (cos θ, sin θ, 0) in the
xy-plane.

y
θ

• Finally have θ run over 0 ď θ ă 2π. Then the semicircle rotates about the z-axis,
sweeping out the full sphere, except for the north and south poles.

Recall that ϕ is the angle between the radius vector and the z-axis. If you hold ϕ fixed
and increase θ by a small amount dθ, r(θ, ϕ) sweeps out the red circular arc in the figure
on the left below. If you hold ϕ fixed and vary θ from 0 to 2π, r(θ, ϕ) sweeps out a line of
latitude. The figure on the right below gives the lines of latitude (or at least the parts of
π 2π 3π 4π
those lines in the first octant) for ϕ = 10 , 10 , 10 , 10 and 5π π
10 = 2 .

95
S URFACE I NTEGRALS 3.1 PARAMETRIZED S URFACES

z
z

dθ

ϕ
y y
θ

x x

On the other hand, if you hold θ fixed and increase ϕ by a small amount dϕ, r(θ, ϕ) sweeps
out the red circular arc in the figure on the left below. If you hold θ fixed and vary ϕ from
0 to π, r(θ, ϕ) sweeps out a line of longitude. The figure on the right below gives the lines
π 2π 3π 4π
of longitude (or at least the parts of those lines in the first octant) for θ = 0, 10 , 10 , 10 , 10
5π π
and 10 = 2 .

z
z

ϕ dϕ
y y
θ

x x

Example 3.1.2

Example 3.1.3 (Cylinder)

The surface x2 + z2 = 1 is an infinite cylinder. Part of this cylinder in the first octant is
sketched below.

96
S URFACE I NTEGRALS 3.1 PARAMETRIZED S URFACES

z y“c

x2 ` z 2 “ 1 y

Note that the section of this cylinder that lies in the xz-plane, and in fact in any plane
y = c, is the circle x2 + z2 = 1. We can of course parametrize this circle by x = cos θ,
z = sin θ. So we can parametrize the whole cylinder by using θ and y as parameters.

0 ď θ ă 2π, ´8 ă y ă 8


r(θ, y) = cos θ , y , sin θ

Example 3.1.3

Example 3.1.4 (Surface of Revolution)

In this example, we are going to parametrize a surface of revolution. In your first integral
calculus course, you undoubtedly encountered many surfaces created by rotating a curve
y = f ( x ) about the x-axis or the y-axis. In this course, we are used to having the z-axis,
rather than the y-axis, run vertically. So in this example, we’ll parametrize the surface
constructed by rotating the curve

z = g(y) = ey 0ďyď1

about the z-axis. Exactly the same procedure can be used to parametrize surfaces created
by rotating about the x-axis or the y-axis too.
We start by just sketching the curve, considering the yz-plane as the plane x = 0 in R3 .
The specified curve is the red curve in the figure below. Concentrate on any one point on
that curve. It is the blue dot at (0, Y, eY ) in the figure. When our curve is rotated about

p0, Y, eY q

97
S URFACE I NTEGRALS 3.1 PARAMETRIZED S URFACES

the z-axis, the blue dot sweeps out a circle. The circle that the blue dot sweeps out

˝ lies in the horizontal plane z = eY and

˝ is centred on the z-axis and
˝ has radius Y.

We can parametrize the circle swept out in the usual way. Here is a top view of the circle,
with the parameter, named θ, indicated.

p0, Y, eY q
θ y
Y
pY sin θ, Y cos θ, eY q

x top view

The coordinates of the red dot are Y sin θ , Y cos θ , eY . This also gives a parametrization

of the surface of revolution

x (Y, θ ) = Y sin θ
y(Y, θ ) = Y cos θ
z(Y, θ ) = eY
0 ď Y ď 1, 0 ď θ ă 2π

Notice, by way of checks, that

˝ when θ = 0,
x (Y, 0) , y(Y, 0) , z(Y, 0) = (0, Y, eY )

runs over the entire desired curve (namely z = g(y), 0 ď y ď 1), when Y runs over
0 ď Y ď 1 and
˝ for any fixed 0 ď Y ď 1, x (Y, θ ) , y(Y, θ ) , z(Y, θ ) runs over the circle x2 + y2 = Y 2 ,

in the plane z = eY , when θ runs over 0 ď θ ă 2π.

Also notice that

x (Y, θ )2 + y(Y, θ )2 = Y 2
so that b
Y= x (Y, θ )2 + y(Y, θ )2
and
?
x (Y,θ )2 +y(Y,θ )2
z(Y, θ ) = eY = e

98
S URFACE I NTEGRALS 3.1 PARAMETRIZED S URFACES

That is, the surface of revolution is contained in the (infinite) surface

?
x 2 + y2
z=e

Remembering that 0 ď Y ď 1, we have that 1 ď z = eY ď e. Thus the surface of revolution

is
? 2 2
z = e x +y 1ďzďe

Finally here is a sketch of the part of the surface in the first octant, x, y, z ě 0.

?
z “ e x `y
2 2

Example 3.1.4

Example 3.1.5 (Torus)

In this example, we are going to parametrize a donut (well, its surface), or an inner tube.

The formal mathematical name for the surface of a donut is a torus. Our strategy will be
to first parametrize the section of the torus in the right half of the yz-plane, and then built
up the full torus by rotating the circle about the z-axis. The section is a circle, sketched
below. We’ll assume that the centre of the circle is a distance R from the z-axis, and that

99
S URFACE I NTEGRALS 3.1 PARAMETRIZED S URFACES

r
θ
p0,R,0q
y

the circle has radius r. Then the red dot on the circle is at
x=0
y = R + r cos θ
z = r sin θ
In particular the red dot is a distance r sin θ above the xy-plane and is a distance R + r cos θ
from the z-axis. So when we rotate the section about the z-axis, the red dot sweeps out a
circle which is sketched below.

y
x

The circle that the red dot sweeps out

˝ lies in the plane z = r sin θ and
˝ is centred on the z-axis and
˝ has radius ρ = R + r cos θ.
We can parametrize the circle swept out in the usual way. Here is a top view of the circle,
with the parameter, named ψ, indicated.

p0, ρ, r sin θq
y
ψ
pρ cos ψ, ρ sin ψ, r sin θq

x top view

100
 S URFACE I NTEGRALS 3.2 TANGENT P LANES

So the parametrization of the circle swept out by the red dot, and also the parametrization
of the torus, is

x = ρ cos ψ = ( R + r cos θ ) cos ψ

y = ρ sin ψ = ( R + r cos θ ) sin ψ
z = r sin θ

or

r(θ, ψ) = ( R + r cos θ ) cos ψ ı̂ı + ( R + r cos θ ) sin ψ ̂ + r sin θ k̂ 0 ď θ, ψ ă 2π

Example 3.1.5

3.2IJ Tangent Planes

If you are confronted with a complicated surface and want to get some idea of what it
looks like near a specific point, probably the first thing that you will do is find the plane
that best approximates the surface near the point. That is, find the tangent plane to the
surface at the point. In general, a good way to specify a plane is to supply

˝ a nonzero vector n (called a normal vector) perpendicular to the plane2 (to determine
the orientation of the plane) and
˝ one point ( x0 , y0 , z0 ) on the plane.

If ( x, y, z) is any other point on the plane, then the vector

( x, y, z) ´ ( x0 , y0 , z0 ) = ( x ´ x0 , y ´ y0 , z ´ z0 )

lies entirely in the plane and so is perpendicular to n. This gives the following very neat
the equation for the plane.
n
px, y, zq
n ¨ ( x ´ x0 , y ´ y0 , z ´ z0 ) = 0
px0 , y0 , z0 q

The following theorem provides formulae for normal vectors n to general surfaces, assum-
ing first that the surface is parametrized, second that the surface is a graph and finally the
surface is given by an implicit equation. The formulae are developed in the proof of the
theorem.

2 Alternatively, you could find two vectors that are in the plane (and not parallel to each other), and then
construct a normal vector by taking their cross product.

101
S URFACE I NTEGRALS 3.2 TANGENT P LANES

Theorem 3.2.1 (Normal vectors to surfaces).

(a) Let

r : D Ă R2 Ñ R3
(u, v) P D ÞÑ r(u, v) = x (u, v) , y(u, v) , z(u, v)

be a parametrized surface and let ( x0 , y0 , z0 ) = r(u0 , v0 ) be a point on the

surface. Set
B ˇ  Bx By Bz
ˇ

Tu = r(u, v0 )ˇ = ( u0 , v0 ) , ( u0 , v0 ) , ( u0 , v0 )
Bu u = u0 Bu Bu Bu
B ˇ  Bx By Bz
ˇ

Tv = r ( u0 , v )ˇ = ( u0 , v0 ) , ( u0 , v0 ) , ( u0 , v0 )
Bv v = v0 Bv Bv Bv

Then
ˇ ˇ
ˇ ı̂ı ̂ k̂ ˇ
ˇ ˇ
ˇ Bx By ˇ
n = Tu ˆ Tv = det ˇˇ Bu (u0 , v0 ) ˇ
Bz
Bu ( u0 , v0 ) Bu ( u0 , v0 )ˇ
ˇ ˇ
ˇ Bx (u , v ) By Bz ˇ
Bv 0 0 Bv ( u0 , v0 ) Bv ( u0 , v0 )

is normal (i.e. perpendicular) to the surface at ( x0 , y0 , z0 ).

(b) Let ( x0 , y0 , z0 ) = f ( x0 , y0 ) be a point on the the surface z = f ( x, y). Then,

n = ´ f x ( x0 , y0 ) ı̂ı ´ f y ( x0 , y0 ) ̂ + k̂

is normal to the surface at ( x0 , y0 , z0 ).

(c) Consider the surface given implicitly by the equation G ( x, y, z) = K, where

K is a constant. Let ( x0
, y0 , z0 ) be a point on the surface and assume that the
gradient ∇ G x0 , y0 , z0 ‰ 0. Then


n = ∇ G x0 , y0 , z0

is normal to the surface at ( x0 , y0 , z0 ).

Note that none of the normal vectors n above need be of unit length.

Note that if we
apply part (c) to G ( x, y, z) = z ´ f ( x, y) we get the normal vector
n = ∇ G x0 , y0 , z0 = ´ f x ( x0 , y0 ) ı̂ı ´ f y ( x0 , y0 ) ̂ + k̂, which is the same as the normal
vector provided by part (b). Of course they had to be at least parallel.
Proof. (a) First fix v = v0 and let u vary. Then
u ÞÑ r(u, v0 ) = x (u, v0 ) , y(u, v0 ) , z(u, v0 )

is a curve on the surface (the red curve in the figure on the right below) that passes through
( x0 , y0 , z0 ) (the black dot in the figure) when u = u0 . The tangent vector to this curve at

102
S URFACE I NTEGRALS 3.2 TANGENT P LANES

Tv

Tu

( x0 , y0 , z0 ), which is also a tangent vector to the surface at ( x0 , y0 , z0 ), is

B ˇ  Bx By Bz
ˇ

Tu = r(u, v0 )ˇ = ( u0 , v0 ) , ( u0 , v0 ) , ( u0 , v0 )
Bu u = u0 Bu Bu Bu
It is the red arrow in the figure on the right above.
Next fix u = u0 and let v vary. Then
v ÞÑ r(u0 , v) = x (u0 , v) , y(u0 , v) , z(u0 , v)

is a curve on the surface (the blue curve in the figure on the right above) that passes
through ( x0 , y0 , z0 ) when v = v0 . The tangent vector to this curve at ( x0 , y0 , z0 ), which is
also a tangent vector to the surface at ( x0 , y0 , z0 ), is
B ˇ  Bx By Bz
ˇ

Tv = r ( u0 , v )ˇ = ( u0 , v0 ) , ( u0 , v0 ) , ( u0 , v0 )
Bv v = v0 Bv Bv Bv
It is the blue arrow in the figure on the right above.
We now have two vectors, namely Tu and Tv , that are tangent to the surface at ( x0 , y0 , z0 ).
So their cross product
ˇ ˇ
ˇ ı
ı̂ 
̂ k̂ ˇ
ˇ ˇ
ˇ Bx By ˇ
n = Tu ˆ Tv = det ˇˇ Bu (u0 , v0 ) Bu (u0 , v0 ) Bu (u0 , v0 )ˇˇ
Bz
ˇ ˇ
ˇ Bx (u , v ) By (u , v ) Bz (u , v ) ˇ
Bv 0 0 Bv 0 0 Bv 0 0

is normal (i.e. perpendicular) to the surface at ( x0 , y0 , z0 ). Note however that this vector
need not be normalized. That is, it need not be of unit length.
(b) Next assume that the surface is given by the equation z = f ( x, y). Then, renaming u
to x and v to y, we may reuse part (a):


r( x, y) = x, y, f ( x, y)


parametrizes the surface and, at x0 , y0 , z0 = f ( x0 , y0 ) ,
Br

Tx = ( x0 , y0 ) = 1 , 0 , f x ( x0 , y0 )
Bx
Br

Ty = ( x0 , y0 ) = 0 , 1 , f y ( x0 , y0 )
By

103
S URFACE I NTEGRALS 3.2 TANGENT P LANES

and
ˇ ˇ
ˇ ı̂ı ̂ k̂ ˇ
ˇ ˇ
ˇ ˇ
n = T x ˆ Ty = det ˇ1 0 f x ( x0 , y0 )ˇ = ´ f x ( x0 , y0 ) ı̂ı ´ f y ( x0 , y0 ) ̂ + k̂
ˇ ˇ
ˇ0 1 f y ( x 0 , y 0 ) ˇ

(c) Finally assume that the surface is given implicitly by the equation G ( x, y, z) = 0
or, more generally
by the equation, G ( x, y, z) = K, where K is a constant. If r(t) =
x (t) y(t) , z(t) is any curve with r(0) = ( x0 , y0 , z0 ) that lies on the surface, then


G r(t) = K for all t
d
ùñ G x (t), y(t), z(t) = 0


for all t
dt
Applying the chain rule gives
BG
dx BG
dy BG
dz
x ( t ), y ( t ), z ( t ) (t) + x ( t ), y ( t ), z ( t ) (t) + x ( t ), y ( t ), z ( t ) (t) = 0
Bx dt By dt Bz dt
dx dy dz



The left hand side is exactly the dot product of BG Bx , BG BG
By , Bz = ∇ G with dt dt dt =
, ,
dr
dt , so that
∇ G r ( t ) ¨ r1 ( t ) = 0


for all t
ùñ ∇ G x0 , y0 , z0 ¨ r (0) = 0

1


This tell us that ∇ G x0 , y0 , z0 is perpendicular to r1 (0), which is a tangent vector to G = K
at ( x0 , y0 , z0 ). This is true for all curves r
(t) on G = K and so is true for all tangent vectors
to G = K at ( x0 , y0 , z0 ). So ∇ G x0 , y0 , z0 is a normal vector to G ( x, y, z) = K at ( x0 , y0 , z0 ).

Example 3.2.2

Consider the surface

x = x (u, v) = u cos v
y = y(u, v) = u sin v
z = z(u, v) = u
Observe that
x (u, v)2 + y(u, v)2 = u2 = z(u, v)2
So our surface is also
G ( x, y, z) = x2 + y2 ´ z2 = 0
We shall sketch it shortly. But first, let’s find it’s tangent plane at ( x0 , y0 , z0 ) = r(u0 , v0 ). In
fact, let’s do it twice. Once using the parametrization and once using its implicit equation.
First, using the parametrization r(u, v) = u cos v ı̂ı + u sin v ̂ + u k̂, we have
Br
Tu = (u0 , v0 ) = cos v0 ı̂ı + sin v0 ̂ + k̂
Bu
Br
Tv = (u0 , v0 ) = ´u0 sin v0 ı̂ı + u0 cos v0 ̂
Bv

104
S URFACE I NTEGRALS 3.2 TANGENT P LANES

so that
n = cos v0 ı̂ı + sin v0 ̂ + k̂ ˆ ´ u0 sin v0 ı̂ı + u0 cos v0 ̂

= ´ u0 cos v0 , ´u0 sin v0 , u0 ) = (´x0 , ´y0 , z0 )

Next using the implicit equation G ( x, y, z) = x2 + y2 ´ z2 = 0, we have the normal vector
∇ G x0 , y0 , z0 = (2x0 , 2y0 , ´2z0 ) = ´2(´x0 , ´y0 , z0 )

Of course the two vectors (´x0 , ´y0 , z0 ) and ´2(´x0 , ´y0 , z0 ) are parallel to each other. Ei-
ther can be used as a normal vector and the tangent plane to x2 + y2 ´ z2 = 0 at ( x0 , y0 , z0 )
is
0 = n ¨ ( x ´ x0 , y ´ y0 , z ´ z0 ) = ´x0 ( x ´ x0 ) ´ y0 (y ´ y0 ) + z0 (z ´ z0 )
provided ( x0 , y0 , z0 ) ‰ 0. In the event that ( x0 , y0 , z0 ) = 0 the “tangent plane equation”
reduces to 0 = 0 and there is clearly a problem.
More generally, if Tu ˆ Tv = 0 (or ∇ G ( x0 , y0 , z0 ) = 0), then either3
˝ the surface fails to have a tangent plane at ( x0 , y0 , z0 ), or
˝ our parametrization is screwy4 there. For example, we can parametrize the xy-plane,
z = 0, by r(u, v) = u cos v ı̂ı + u sin v ̂ . (This is just polar coordinates.) Then Tu =
cos v0 ı̂ı + sin v0 ̂ and Tv = ´u0 sin v0 ı̂ı + u0 cos v0 ̂ , so that Tu ˆ Tv = u0 k̂ is 0 when
u0 = 0. But the plane z = 0 is its own tangent plane everywhere.
The surface of current interest is x2 + y2 = z2 . The intersection of this surface with the
horizontal plane z = z0 is x2 + y2 = z20 , which is the circle of radius |z0 | centred on x =
y = 0. So our surface is a stack of circles. The radius of the circle in the xy-plane is zero.
The radius increases linearly as we move away from the xy-plane. Our surface is a cone.
It does not have a tangent plane at (0, 0, 0).

Example 3.2.2

Example 3.2.3

This time we shall find the tangent planes to the surface

x 2 + y2 ´ z2 = 1

3 We saw the same dichotomy when considering what happened for a curve when r1 (t) = 0. See Example
1.1.8.
4 Of course “screwy” is not a mathematically precise word. One way a parametrization r(u, v) could be
“screwy” is if it failed to give a one-to-one correspondence between parameter values (u, v) and points
on (part of) the surface. For example, polar coordinates r(u, v) = u cos v ı̂ı + u sin v ̂ give r(0, v) = (0, 0)
for all values of v.

105
S URFACE I NTEGRALS 3.2 TANGENT P LANES

As for the cone of the last example, the intersection

b of this surface with the horizontal
2
plane z = z0 is a circle — the circle of radius 1 + z0 centred on x = y = 0. Our surface is
again a stack of circles. The radius of the circle in the xy-plane is 1. The radius increases
as we move away from the xy-plane. Here is a sketch of the surface.

It is called a hyperboloid5 of one sheet.

Using the implicit equation G ( x, y, z) = x2 + y2 ´ z2 = 1, we have
∇ G x0 , y0 , z0 = (2x0 , 2y0 , ´2z0 ) = 2( x0 , y0 , ´z0 )

and we may take ( x0 , y0 , ´z0 ) as a normal vector at ( x0 , y0 , z0 ). So the tangent plane to

x2 + y2 ´ z2 = 1 at ( x0 , y0 , z0 ) is
0 = n ¨ ( x ´ x0 , y ´ y0 , z ´ z0 ) = x0 ( x ´ x0 ) + y0 ( y ´ y0 ) ´ z0 ( z ´ z0 )
This time n = ( x0 , y0 , ´z0 ) ‰ 0, so that we have a tangent plane, at every point of the
surface. In particular, the vanishing of n = ( x0 , y0 , ´z0 ) at ( x0 , y0 , z0 ) = (0, 0, 0) is not a
problem because (0, 0, 0) is not on the surface.
Example 3.2.3

Example 3.2.4 (Optional — Parametrizing the Hyperboloid of One Sheet)

The hyperboloid of one sheet, x2 + y2 ´ z2 = 1, has a symmetry. It is invariant un-

der rotation about the z-axis. So it is natural to parametrize the surface using cylindrical
coordinates. z
px, y, zq
x = r cos θ
y = r sin θ z
y
r
z=z θ
px, y, 0q
x
2 2 2 2 2
? surface x + y?´ z = 1 is r ´ z = 1, and we could
In cylindrical coordinates the
2 2
parametrize it by r(θ, z) = 1 + z cos θ ı̂ı + 1 + z sin θ ̂ + z k̂. Alternatively, we can
eliminate the square roots in the parametrization by exploiting the hyperbolic trig func-
tions
1 1
sinh u = eu ´ e´u cosh u = eu + e´u



2 2

5 There are also hyperboloids of two sheets. See Appendix G.

106
 S URFACE I NTEGRALS 3.3 S URFACE I NTEGRALS

The functions have properties6 that are very similar to those of sin θ and cos θ.

d d
cosh u = sinh u sinh u = cosh u cosh2 u ´ sinh2 u = 1
du du
We can set r = cosh u, z = sinh u to yield the parametrization

r(θ, u) = cosh u cos θ ı̂ı + cosh u sin θ ̂ + sinh u k̂

As an exercise in working with hyperbolic trig functions, we’ll use this parametrization to
find n̂.

x = cosh u cos θ xu = sinh u cos θ xθ = ´ cosh u sin θ

y = cosh u sin θ yu = sinh u sin θ yθ = cosh u cos θ
z = sinh u zu = cosh u zθ = 0

So
ˇ ˇ
ˇ ı̂ı ̂ k̂ ˇˇ
ˇ
ˇ ˇ
n = Tu ˆ Tθ = det ˇ sinh u cos θ sinh u sin θ cosh uˇ
ˇ ˇ
ˇ´ cosh u sin θ cosh u cos θ 0 ˇ
= ´ cosh2 u cos θ , ´ cosh2 u sin θ , sinh u cosh u

Example 3.2.4

3.3IJ Surface Integrals

We are now going to define two types of integrals over surfaces.
ť
˝ Integrals that look like S ρ dS are used to compute the area and, when ρ is, for
example, a mass density, the mass of the surface S.
ť
˝ Integrals that look like S F ¨ n̂ dS, with n̂( x, y, z) being a unit vector that is perpen-
dicular to S at ( x, y, z), are called flux integrals. We shall see in §3.4, thatť when v is
the velocity field of a moving fluid and ρ is the density of the fluid, then S ρv ¨ n̂ dS
is the rate at which fluid is crossing the surface S.

3.3.1 §§ Parametrized Surfaces

Suppose that we wish to integrate over part, S, of a surface that is parametrized by r(u, v).
We start by cutting S up into small pieces by drawing a bunch of curves of constant u (the
blue curves in the figure below) and a bunch of curves of constant v (the red curves in the
figure below).

6 This is no accident: cosh u = cos(iu) and sinh u = ´i sin(iu), where i is the usual complex number that
obeys i2 = ´1.

107
S URFACE I NTEGRALS 3.3 S URFACE I NTEGRALS

Concentrate on any one the small pieces. Here is a greatly magnified sketch.

u“u0 u“u0 `du

v varying v varying

P2 P3 u varying
v“v0 `dv

P1 u varying
P0 v“v0

For example, the lower red curve was constructed by holding v fixed at the value v0 ,
varying u and sketching r(u, v0 ), and the upper red curve was constructed by holding v
fixed at the slightly larger value v0 + dv, varying u and sketching r(u, v0 + dv). So the
four intersection points in the figure are

P2 = r(u0 , v0 + dv) P3 = r(u0 + du, v0 + dv)

P0 = r(u0 , v0 ) P1 = r(u0 + du, v0 )

Now if
R(t) = r(u0 + tdU , v0 + tdV )
(where dU and dV are any small constants) then, by Taylor expansion,


r u0 + dU , v0 + dV = R(1)
« R (0 ) + R 1 (0 ) t ´ 0 t =1



Br Br
= r(u0 , v0 ) + (u0 , v0 ) dU + (u0 , v0 ) dV
Bu Bv
Applying this three times, once with dU = du, dV = 0, once with dU = 0 dV = dv, and
once with dU = du, dV = dv,

P0 = r(u0 , v0 )
Br
P1 = r(u0 + du, v0 ) « r ( u0 , v0 ) +
(u0 , v0 ) du
Bu
Br
P2 = r(u0 , v0 + dv) « r(u0 , v0 ) + (u0 , v0 ) dv
Bv
Br Br
P3 = r(u0 + du, v0 + dv) « r(u0 , v0 ) + (u0 , v0 ) du + (u0 , v0 ) dv
Bu Bv

108
 S URFACE I NTEGRALS 3.3 S URFACE I NTEGRALS

We have dropped all Taylor expansion terms that are of degree two or higher in du, dv.
The reason is that, in defining the integral, we take the limit du, dv Ñ 0. Because of that
limit, all of the dropped terms contribute exactly 0 to the integral. We shall not prove this.
But we shall show, in the optional §3.3.5, why this is the case.
The small piece of our surface with corners P0 , P1 , P2 , P3 is approximately a parallelo-
gram with sides
P2 P3
ÝÝÑ ÝÝÑ Br
P0 P1 « P2 P3 « (u0 , v0 ) du
Bu ÝÝÝÑ
P0 P2
ÝÝÑ ÝÝÑ Br P1
P0 P2 « P1 P3 « (u0 , v0 ) dv P0 θÝÝÝÑ
Bv P0 P1
ÝÝÑ ÝÝÑ
Denote by θ the angle
ˇ ˇ between the vectors P0 P1 and P0 P2 . The ˇbase ˇof the parallelogram,
ÝÝÑ ÝÝÑ ÝÝÑ
P0 P1 , has length ˇ P0 P1 ˇ, and the height of the parallelogram is ˇ P0 P2 ˇ sin θ. So the area of
the parallelogram is7
ÝÝÑ ÝÝÑ ˇÝÝÑ ÝÝÑˇ
| P0 P1 | | P0 P2 | sin θ = ˇ P0 P1 ˆ P0 P2 ˇ
ˇ ˇ
ˇ Br Br ˇ
ˇ
« ˇ (u0 , v0 ) ˆ (u0 , v0 )ˇˇdudv
Bu Bv
Br Br
Furthermore, Bu (u0 , v0 ) and Bv (u0 , v0 ) are tangent vectors to the curves r(t , v0 ) and
Br Br
r(u0 , t) respectively. Both of these curves lie in S. So Bu (u0 , v0 ) and Bv (u0 , v0 ) are tan-
gent vectors to S at (u0 , v0 ) and the cross product Bu (u0 , v0 ) ˆ Bv (u0 , v0 ) is perpendicular
Br Br

to S at (u0 , v0 ). We have found both dS and n̂ dS, where n̂ is a unit normal vector to the
surface.
Equation 3.3.1.
For the parametrized surface r(u, v),

Br Br
n̂ dS = ˘ (u , v) ˆ (u , v) dudv
Bu Bv
ˇ ˇ
ˇ Br Br ˇ
dS = ˇˇ (u , v) ˆ (u , v)ˇˇ dudv
Bu Bv

The ˘ sign in (3.3.1) is there because there are two unit normal vectors at each point of a
surface, one on each side of the surface. Typically, the application itself tells you which of
the two normal vectors should be used. We shall see many examples shortly.

3.3.2 §§ Graphs
The surface which is the graph z = f ( x, y) can be parametrized by

r( x, y) = x ı̂ı + y ̂ + f ( x, y) k̂

7 As we mentioned above, the approximation below becomes exact when the limit du, dv Ñ 0 is taken in
the definition of the integral. See the optional §3.3.5.

109
S URFACE I NTEGRALS 3.3 S URFACE I NTEGRALS

As
Br Bf Br Bf
= ı̂ı + k̂ and = ̂ + k̂
Bx Bx By By
we have  
ı̂ı ̂ k̂
Br Br 1 0
Bf 
ˆ = ´ f x ( x, y) ı̂ı ´ f y ( x, y) ̂ + k̂

= det  Bx 

Bx By 
Bf
0 1 By

So, (3.3.1) gives the following.

Equation 3.3.2.
For the surface z = f ( x, y),

n̂ dS = ˘ ´ f x ( x, y) ı̂ı ´ f y ( x, y) ̂ + k̂ dxdy

 

b
dS = 1 + f x ( x, y)2 + f y ( x, y)2 dxdy

Similarly, for the surface x = g(y, z),

n̂ dS = ˘ ı̂ı ´ gy (y, z) ̂ ´ gz (y, z) k̂ dydz

 

b
dS = 1 + gy (y, z)2 + gz (y, z)2 dydz

and for the surface y = h( x, z),

n̂ dS = ˘ ´ h x ( x, z) ı̂ı + ̂ ´ hz ( x, z) k̂ dxdz

 

b
dS = 1 + h x ( x, z)2 + hz ( x, z)2 dxdz

Again, in any given application, some care must be taken in choosing the sign in (3.3.2),
so as to get the appropriate normal
b vector.
The formulae like dS = 1 + f x ( x, y)2 + f y ( x, y)2 dxdy in (3.3.2) have geometric in-
terpretations. The red parallelogram in the sketch

z
n̂ k̂
θ

dx
x dy

110
 S URFACE I NTEGRALS 3.3 S URFACE I NTEGRALS
b
represents a little piece of our surface. It has area dS = 1 + f x ( x, y)2 + f y ( x, y)2 dxdy.
The blue parallelogram in the same sketch represents the projection of the red parallelo-
gram onto the xy-plane. It has area dxdy. The vector n̂ in the sketch is a unit normal for
the red parallelogram. We have seen that it is parallel to
Br Br
ˆ = ´ f x ( x, y) ı̂ı ´ f y ( x, y) ̂ + k̂
Bx By
so that the angle θ between n̂ and k̂ obeys
(´ f x ( x, y) ı̂ı ´ f y ( x, y) ̂ + k̂) ¨ k̂
cos θ = ˇ ˇ
ˇ ´ f x ( x, y) ı̂ı ´ f y ( x, y) ̂ + k̂ˇ |k̂|
1
=b
1 + f x ( x, y)2 + f y ( x, y)2
b
The geometric interpretation of dS = 1 + f x ( x, y)2 + f y ( x, y)2 dxdy is that the area dS
of a little piece of surface is the area of its projection on the xy-plane times the factor cos1 θ
where θ is the angle between n̂ (which is perpendicular to the surface) and k̂ (which is
perpendicular to the xy-plane). Notice that
˝ when θ is close to zero, which corresponds the f being almost constant and our
surface being almost parallel to the xy-plane, dS reduces to almost dxdy.
˝ On the other hand, in the limit θ Ñ π2 , which corresponds to f x and/or f y becoming
infinite and our surface becoming perpendicular to the xy-plane, dS becomes “in-
finity times” dxdy. In this case, we should represent our surface either in the form
x = g(y, z) or in the form y = h( x, z), rather than in the form z = f ( x, y).

3.3.3 §§ Surfaces Given by Implicit Equations

Finally suppose that the surface is given by the equation G ( x, y, z) = K, with K a constant.
Suppose further that at some point on the surface BG Bz ‰ 0. Then near that point we may
8
solve the equation G ( x, y, z) = K for z as a function of x and y. That is, the surface also
obeys z = f ( x, z) for a function f ( x, y) that satisfies


G x, y, f ( x, y) = K
near the point. Differentiating this with respect to x and y gives, by the chain rule,
B h
i


0= G x, y, f ( x, y) = Gx x, y, f ( x, y) + Gz x, y, f ( x, y) f x ( x, y)
Bx
B h
i


0= G x, y, f ( x, y) = Gy x, y, f ( x, y) + Gz x, y, f ( x, y) f y ( x, y)
By

8 This is called the implicit function theorem. We will not prove it. But it is not so hard to understand
why it is true, if one thinks in terms of the Taylor expansion of G about the point. For simplicity, let’s
suppose that the point is (0, 0, 0) and G happens to be exactly equal to its first order Taylor expansion
about (0, 0, 0). That is, G ( x, y, z) = A + Bx + Cy + Dz, for some constants A, B, C, D. Since (0, 0, 0) is
on the surface, A = K. As BBGz = D ‰ 0 we can easily solve G ( x, y, z) = K for z as a function of x and y.
Namely z = D1 (´Bx ´ Cy). The general proof is based on the fact that, under reasonable hypotheses,
the first order Taylor expansion is a good approximation to G near (0, 0, 0).

111
 S URFACE I NTEGRALS 3.3 S URFACE I NTEGRALS

which implies



Gx x, y, f ( x, y) Gy x, y, f ( x, y)
f x ( x, y) = ´
f y ( x, y) = ´

Gz x, y, f ( x, y) Gz x, y, f ( x, y)
and



Gx x, y, f ( x, y) Gy x, y, f ( x, y)
´ f x ( x, y) ı̂ı ´ f y ( x, y) ̂ + k̂ =
ı̂ı +
̂ + k̂
Gz x, y, f ( x, y) Gz x, y, f ( x, y)


∇ G x, y, f ( x, y)
=

Gz x, y, f ( x, y)
So, by (3.3.1),
Equation 3.3.3.
For the surface G ( x, y, z) = K, when Gz ( x, y, z) ‰ 0,


∇ G x, y, z
n̂ dS = ˘ dxdy
∇ G x, y, z ¨ k̂

ˇ
ˇ
ˇ ∇ G x, y, z ˇ
dS = ˇˇ
ˇ dxdy
∇ G x, y, z ¨ k̂ ˇ

Similarly, for the surface G ( x, y, z) = K, when Gx ( x, y, z) ‰ 0,

∇ G x, y, z
n̂ dS = ˘
dydz
∇ G x, y, z ¨ ı̂ı
ˇ
ˇ
ˇ ∇ G x, y, z ˇ
dS = ˇˇ
ˇ dydz
∇ G x, y, z ¨ ı̂ı ˇ

and for the surface G ( x, y, z) = K, when Gy ( x, y, z) ‰ 0,

∇ G x, y, z
n̂ dS = ˘
dxdz
∇ G x, y, z ¨ ̂
ˇ
ˇ
ˇ ∇ G x, y, z ˇ
dS = ˇˇ
ˇ dxdz
∇ G x, y, z ¨ ̂ ˇ

If, for some point ( x0 , y0 , z0 ) we have Gx ( x0 , y0 , z0 ) = Gy ( x0 , y0 , z0 ) = Gz ( x0 , y0 , z0 ) = 0,

we also have problem! Often this is a sign that our surface is not smooth at ( x0 , y0 , z0 ) and
in fact does not have a normal vector there. For an example of this, see Example 3.2.2.
ť
3.3.4 §§ Examples of S ρ dS
We’ll start by computing, in several different ways, the surface area of the hemisphere
x 2 + y2 + z2 = a2 zě0

112
S URFACE I NTEGRALS 3.3 S URFACE I NTEGRALS

(with a ą 0). You probably know, from high school, that the answer is 12 ˆ 4πa2 = 2πa2 .
But you have probably not seen a derivation of this answer. Note that, since x2 + y2 =
a2 ´ z2 on the hemisphere, the set ˇ of ( x, y)’s for(which there is a z with ( x, y, z) on the
hemisphere is exactly ( x, y) P R2 ˇ x2 + y2 ď a2 .

Example 3.3.4 (Area of a hemisphere — using cylindrical coordinates)

Let’s parametrize the hemisphere x2 + y2 + z2 = a2 , z ě 0, using as parameters the polar

coordinates r, θ of the cylindrical coordinates

z
px, y, zq
x = r cos θ
y = r sin θ z
y
r
z=z θ
px, y, 0q
x

and then apply (3.3.1). In cylindrical coordinates the equation x2 + y2 + z2 = a2 becomes

r2 + z2 = a2 , and the condition x2 + y2 ď a2 is 0 ď r ď a, 0 ď θ ă 2π.
So the hemisphere can be parametrized by

a
x (r, θ ) , y(r, θ ) , z(r, θ ) = r cos θ , r sin θ , a2 ´ r2

with 0 ď r ď a, 0 ď θ ă 2π

?
Note that we selected the positive solution z = a2 ´ r2 of r2 + z2 = a2 in order to satisfy
the condition that z ě 0. Since

 Bx By Bz   r 
, , = cos θ , sin θ , ´ ?
Br Br Br a2 ´ r 2
 Bx By Bz 
, , = (´r sin θ , r cos θ , 0)
Bθ Bθ Bθ

(3.3.1) yields

 Bx By Bz   Bx By Bz 
n̂ dS = ˘ , , ˆ , , drdθ
Br  Br Br Bθ Bθ Bθ 
ı̂ı ̂ k̂
 cos θ
= ˘ det  ?
sin θ ´ 2r 2   drdθ a ´r
´r sin θ r cos θ 0
 r2 cos θ r2 sin θ 
=˘ ? , ? , r drdθ
d a ´r a2 ´ r 2 d
2 2

r4 a2 r 2 ar
dS = + r 2 drdθ = drdθ = ? drdθ
a ´r
2 2 a ´r
2 2
a2 ´ r 2

113
S URFACE I NTEGRALS 3.3 S URFACE I NTEGRALS

So the area of the hemisphere is

ż a ż 2π ża
ar r
dr dθ ? = 2πa dr ?
0 0 a2 ´ r 2 0 a2 ´ r 2
ż0
´du/2
= 2πa ? with u = a2 ´ r2 , du = ´2r dr
a 2 u
h ? i0
= 2πa ´ u
a2
= 2πa2
as it should be.
Example 3.3.4

Example 3.3.5 (Area of a hemisphere — using an implicit equation)

This time we’ll compute the area of the hemisphere by using that, if ( x, y, z) is on the hemi-
sphere, then G ( x, y, z) = a2 with G ( x, y, z) = x2 + y2 + z2 . Since


∇ G ( x, y, z) = 2x , 2y , 2z
(3.3.3) yields
ˇ
ˇ
ˇ ∇ G x, y, z ˇ
dS = ˇˇ
ˇ dxdy
∇ G x, y, z ¨ k̂ ˇ
ˇ
ˇ
ˇ 2x , 2y , 2z ˇ
= ˇˇ ˇ dxdy
ˇ
2z
a
x 2 + y2 + z2
= dxdy
|z|
a
=a dxdy on x2 + y2 + z2 = a2
a2 ´ x 2 ´ y2
ť
So the area is x2 +y2 ďa2
? a
dxdy. To evaluate this integral, we switch to polar
a2 ´x2 ´y2
coordinates, substituting x = r cos θ, y = r sin θ. This gives
ij ża ż 2π
a a
area = a dxdy = dr r dθ ?
a ´x ´y
2 2 2 0 0 a ´ r2
2
x2 +y2 ďa2
ża
r
= 2πa dr ?
0 a2 ´ r2
We already showed, in Example 3.3.4, that the value of this integral is 2πa2 .
Example 3.3.5

Example 3.3.6 (Area of a hemisphere — using spherical coordinates)

Of course “integrating over a sphere” cries out for spherical coordinates. So this time we

114
S URFACE I NTEGRALS 3.3 S URFACE I NTEGRALS

parametrize the hemisphere x2 + y2 + z2 = a2 , z ě 0, using as parameters the angular

coordinates θ, ϕ of the spherical coordinates z
p0, 0, zq ρ sin ϕ
px, y, zq
ρ cos ϕ
x = ρ sin ϕ cos θ ϕ ρ
y = ρ sin ϕ sin θ y
z = ρ cos ϕ θ
px, 0, 0q
px, y, 0q
x ρ sin ϕ sin θ

and then apply (3.3.1). In spherical coordinates the equation x2 + y2 + z2 = a2 becomes

just ρ2 = a2 , and the condition z ě 0 is 0 ď ϕ ď π2 , 0 ď θ ă 2π. So the hemisphere can be
parametrized9 by
π
0ďϕď , 0 ď θ ă 2π



x (θ, ϕ) , y(θ, ϕ) , z(θ, ϕ) = a sin ϕ cos θ , a sin ϕ sin θ , a cos ϕ
2
Since
 Bx By Bz 
= ´ a sin ϕ sin θ , a sin ϕ cos θ , 0


, ,
Bθ Bθ Bθ
 Bx By Bz 
, , = ( a cos ϕ cos θ , a cos ϕ sin θ , ´a sin ϕ)
Bϕ Bϕ Bϕ
(3.3.1) yields
 Bx By Bz   Bx By Bz 
n̂ dS = ˘ , , ˆ , , dθdϕ
Bθ Bθ Bθ Bϕ Bϕ Bϕ
= ˘ ´ a sin ϕ sin θ , a sin ϕ cos θ , 0 ˆ ( a cos ϕ cos θ , a cos ϕ sin θ , ´a sin ϕ) dθdϕ



= ˘ ´ a2 sin2 ϕ cos θ , ´a2 sin2 ϕ sin θ , ´a2 sin ϕ cos ϕ dθdϕ

2
= ¯a sin ϕ sin ϕ cos θ , sin ϕ sin θ , cos ϕ dθdϕ
b
2
dS = a sin ϕ sin2 ϕ cos2 θ + sin2 ϕ sin2 θ + cos2 ϕ dθdϕ
= a2 sin ϕ dθdϕ
So the area of the hemisphere is
żπ ż 2π ż π iπ/2
2 2
h
2 2
a dϕ dθ sin ϕ = 2πa dϕ sin ϕ = 2πa2 ´ cos ϕ
0 0 0 0
2
= 2πa

9 As we have noted before, the spherical coordinate system really breaks down at ϕ = 0, because ρ = 1,
ϕ = 0 gives the same point, namely the north pole (0, 0, 1), for all values of θ. We should really treat our
integral like an improper integral, first integrating over ε ă ϕ ď π2 and then taking the limit ε Ñ 0+ .
However the breakdown of the spherical coordinate system at ϕ = 0, just like the breakdown of polar
coordinates at r = 0, rarely causes problem and it is routine to skip the “improper integral” step.

115
S URFACE I NTEGRALS 3.3 S URFACE I NTEGRALS

Example 3.3.6

There is an easier way to do this, using a little geometry.

Example 3.3.7 (Area of a hemisphere — using spherical coordinates again)

We are now going to again compute the surface area of the hemisphere using spherical co-
ordinates. But this time instead of determining dS using the canned formula (3.3.1), we
are going to read it off of a sketch.
Sketch the part of the hemisphere that is in the first octant, x ě 0, y ě 0, z ě 0. Slice it
up into small pieces by drawing in curves of constant θ (the blue lines in the figure below)
and curves of constant ϕ (the red lines in the figure below). Each piece is approximately

z
a sin ϕ dθ

a dϕ

a little rectangle. Concentrate on one of them, like the piece with the thick sides in the
figure above. The area, dS, of that piece is (essentially) the product of its height and its
width. Each of the two sides of the piece is
˝ a segment of a circle of radius a (a fat blue line in both the figure above and in the
figure on the left below)
˝ that subtends an angle dϕ
dϕ dϕ
˝ and hence is the fraction 2π of a full circle of radius a and hence is of length 2π 2πa =
adϕ.
The top of the piece is
˝ a segment of a circle of radius a sin ϕ (a fat red line in both the figure above and in
the figure on the right below)
˝ that subtends an angle dθ
dθ
˝ and hence is the fraction 2π of a full circle of radius a sin ϕ and hence is of length
dθ
2π 2πa sin ϕ = a sin ϕdθ.

These are drawn in the figure below.

116
S URFACE I NTEGRALS 3.3 S URFACE I NTEGRALS

z z
a sin ϕ dθ

dθ
a a dϕ a sin ϕ

ϕ dϕ ϕ a
y y

x x

So the area of our piece is

a sin ϕdθ = a2 sin ϕ dθdϕ

dS = adϕ

This is exactly the same formula that we found for dS in Example 3.3.6 so that we will, yet
again, get that the area of a hemisphere of radius a is 2πa2 . (Phew!)
Example 3.3.7

But wait! We can do it again, by yet another method!

Example 3.3.8 (Area of a hemisphere — using z = f ( x, y))

We’ll compute the area of the hemisphere one last time10 . This time we’ll use that the
equation of the hemisphere is
b
z = f ( x, y) = a2 ´ x2 ´ y2 with ( x, y) running over x2 + y2 ď a2

So (3.3.2) yields
b
dS = 1 + f x ( x, y)2 + f y ( x, y)2 dxdy
d
 ´x 2  ´y 2
= 1+ a + a dxdy
a2 ´ x 2 ´ y2 a2 ´ x 2 ´ y2
d
x 2 + y2
= 1+ 2 dxdy
a ´ x 2 ´ y2
d
a2
= dxdy
a2 ´ x 2 ´ y2
ť
So the area is x2 +y2 ďa2
? a
dxdy. We already found, in Example 3.3.5, that the
a2 ´x2 ´y2
value of this integral in 2πa2 .

10 We promise!

117
S URFACE I NTEGRALS 3.3 S URFACE I NTEGRALS

Example 3.3.8

Let’s do some more substantial examples, where the integrand is not 1.

Example 3.3.9
ť
Problem: Evaluate S x2 y2 z2 dS where S is the part of the cone x2 + y2 = z2 with 0 ď z ď 1.
Solution 1. We can express S as

b
z = f ( x, y) = x 2 + y2 x 2 + y2 ď 1 x2 ` y 2 “ z 2

Now since
x y
f x ( x, y) = a f y ( x, y) = a
x 2 + y2 x 2 + y2
(3.3.2) gives11
h x2 y2 i1/2 ?
dS = 1 + + dxdy = 2 dxdy
x 2 + y2 x 2 + y2
Our integral is then
ij ? ij
2 2 2
x y z dS = 2 x2 y2 ( x2 + y2 ) dxdy
S x2 +y2 ď1

Since we are integrating over a circular domain, let’s convert to polar coordinates.
ij ? ż 2π ż1
2 2 2
x y z dS = 2 dθ dr r (r cos θ )2 (r sin θ )2 r2
0 0
S
"ż # "ż
?
#
2π 1
= 2 dθ cos2 θ sin2 θ dr r7
0 0
? ż 2π ? ż 2π
2 2 2 2
= dθ cos θ sin θ = dθ sin2 (2θ )
8 32 0
? ż02π
2
dθ 1 ´ cos(4θ )
 
=
64 0

Remembering12 that the integral of cos(θ ), or cos(4θ ), over a full period is 0, we end up
with ij ? ?
2 π 2
x2 y2 z2 dS = (2π ) =
64 32
S

11 This answer for dS is a very clean. Think about why. Hint: review the discussion following (3.3.2).
12 If you have forgotten why, sketch the graph.

118
S URFACE I NTEGRALS 3.3 S URFACE I NTEGRALS

Solution 2. We may parametrize13 the cone by

r(z, θ ) = z cos θ ı̂ı + z sin θ ̂ + z k̂ 0 ď z ď 1, 0 ď θ ď 2π

The because
Br Br
= cos θ ı̂ı + sin θ ̂ + k̂ and = ´z sin θ ı̂ı + z0̧sθ ̂
Bz Bz
(3.3.1) yields14
 
ı̂ı ̂ k̂
n̂ dS = ˘ det  cos θ sin θ 1  dzdθ
´z sin θ z cos θ 0
= ˘ ´ z cos θ ı̂ı ´ z sin θ ̂ + z k̂ dzdθ
 
?
dS = 2z dzdθ

So our integral becomes

ij ? ż 2π ż1
2 2 2
x y z dS = 2 dθ dz z(z cos θ )2 (z sin θ )2 z2
0 0
S
? ż 2π ż1
= 2 dθ dz z7 cos2 θ sin2 θ
? ż0 2π 0
2
= dθ cos2 θ sin2 θ
8 0
We evaluated this integral in Solution 1. So again
ij ?
2 2 2 π 2
x y z dS =
32
S

Example 3.3.9

Let’s do something more celestial.

Example 3.3.10

Consider a spherical shell of radius a with mass density µ per unit area. Think of it as a
hollow planet15 . We are going to determine the gravitational force that it exerts on a parti-
cle of mass m a distance b away from its centre. This particle can be either outside the shell
(b ą a) or inside the shell (b ă a). We can choose the coordinate system so that the centre
of the shell is at the origin and the particle is at (0, 0, b). By Newton’s law of gravitation,

13 We did so previously, with different variable names, in Example 3.2.2.

14 Again the formula for dS is very neat. Think about why.
15 A favourite of science fiction and fantasy writers. Plug “subterranean fiction” into your favourite search
engine. While you’re at it, also try “gravity train”. We’ll look at it in the optional Example 3.3.11.

119
S URFACE I NTEGRALS 3.3 S URFACE I NTEGRALS

the force exerted on the particle by a tiny piece of the shell of surface area dS located at r
is
G (µdS) m
(r ´ (0, 0, b))
|r ´ (0, 0, b)|3
Here G is the gravitational constant, µdS is the mass of the tiny piece of shell, m is the
mass of the particle and r ´ (0, 0, b) is the vector from the particle to the piece of shell. If

p0, 0, bq dS
r ´ p0, 0, bq
r ´ p0, 0, bq
dS p0, 0, bq r

we work in spherical coordinates, as we did in Example 3.3.6,

dS = a2 sin ϕ dϕdθ

and

r = a sin ϕ cos θ ı̂ı + a sin ϕ sin θ ̂ + a cos ϕ k̂

r ´ (0, 0, b) = a sin ϕ cos θ ı̂ı + a sin ϕ sin θ ̂ + ( a cos ϕ ´ b) k̂
|r ´ (0, 0, b)|2 = a2 + b2 ´ 2ab cos ϕ

The total force is then

żπ ż 2π
2 a sin ϕ cos θ ı̂ı + a sin ϕ sin θ ̂ + ( a cos ϕ ´ b) k̂
F = Gµma dϕ dθ sin ϕ 3/2
a2 + b2 ´ 2ab cos ϕ

0 0

Note for future reference that the square root in [ a2 + b2 ´ 2ab cos ϕ]3/2 is the positive square
root because [b2 + a2 ´ 2ab cos ϕ]1/2 is the length of r ´ (0, 0, b), which is positive.
This integral is a little different than other integrals that we have encountered so far in
that the integrand is a vector. By definition16 ,
ij ij ij ij
 
G1 ı̂ı + G2 ̂ + G3 k̂ dS = ı̂ı G1 dS + ̂ G2 dS + k̂ G3 dS
S S S S

so we just have to compute the three components separately.

ť ť ť
16 Under this definition we still have (A + B) dS = A dS + B dS.

120
S URFACE I NTEGRALS 3.3 S URFACE I NTEGRALS

In our case, the ı̂ı and ̂ components

ż π
" ż 2π #
a sin ϕ
F ¨ ı̂ı = Gµma2 dϕ sin ϕ  3/2 dθ cos θ
0 a2 + b2 ´ 2ab cos ϕ 0
żπ " ż 2π #
a sin ϕ
F ¨ ̂ = Gµma2 dϕ sin ϕ  3/2 dθ sin θ
0 a2 + b2 ´ 2ab cos ϕ 0

ş2π ş2π
are both zero17 because 0 cos θ dθ = 0 sin θ dθ = 0 so that
żπ ż 2π
2 a cos ϕ ´ b
F = Gµma k̂ dϕ dθ sin ϕ  3/2
0 a2 + b2 ´ 2ab cos ϕ
0
żπ
a cos ϕ ´ b
= 2πGµma2 k̂ dϕ sin ϕ  3/2
0 a2 + b2 ´ 2ab cos ϕ

To evaluate this integral we substitute

a2 + b2 ´ u
u = a2 + b2 ´ 2ab cos ϕ du = 2ab sin ϕ dϕ cos ϕ =
2ab
When ϕ = 0, u = ( a ´ b)2 and when ϕ = π, u = ( a + b)2 , so
ż ( a + b )2 a2 +b2 ´u
πGµma 2b ´b
F= k̂ du 3/2
b ( a´b)2 u
πGµma h a2 ´ b2  u´1/2 u1/2 i(a+b)2
= k̂ ´
b 2b ´1/2 1/2(2b) (a´b)2

Recalling that u1/2 is the positive square root,

πGµma h b2 ´ a2  1 a + b  b2 ´ a2  1 |a ´ b| i
F= k̂ ´ ´ +
b b a+b b b |a ´ b| b

If b ą a, so that |a ´ b| = b ´ a

πGµma h b ´ a a + b a + b b ´ a i G (4πa2 µ)m

F= k̂ ´ ´ + =´ k̂
b b b b b b2
If b ă a, so that |a ´ b| = a ´ b

πGµma h b ´ a a + b a + b a ´ b i
F= k̂ ´ + + =0
b b b b b
The moral18 is

17 Think about why the ı̂ı and ̂ components should both be zero. Think symmetry.
18 These two results appeared in Isaac Newton’s Principia Mathematica (1687). They are known as New-
ton’s “superb theorems”.

121
S URFACE I NTEGRALS 3.3 S URFACE I NTEGRALS

˝ if the particle is inside the shell, it feels no gravitational force at all, and

˝ if the particle is outside the shell, it feels the same gravitational force as it would if
the entire mass of the shell (4πa2 µ) were concentrated at the centre of the shell.

Example 3.3.10

Example 3.3.11 (Optional — Gravity Train)

The “Gravity Train19 ” refers to the following curious, though admittedly not very practi-
cal, thought experiment.

• Pretend that the Earth is a perfect sphere of radius R and that it has a constant mass
density ρ.

• Pick any two distinct points on the surface of the Earth. Call them V and M.

• Bore a tunnel straight through the Earth from V to M.

• Place a train in the tunnel at V. Assume that the only forces acting on the train are
gravity, G, and a normal force, N, that the tunnel imposes on the train to keep it
in the tunnel. In particular, there are no frictional forces, like air resistance, and the
train does not have an engine. Release the train and assume that it does not melt as
it passes through the centre of the Earth.

N
M V
G
O

What happens?
We’ll simplify our analysis of the motion of the train by picking a convenient coordi-
nate system.
˝ First translate our coordinate system so that the centre of the Earth, call it O, is at the
origin, (0, 0, 0).
˝ Then rotate our coordinate system about the origin so that the origin, V and M all
lie in the xz-plane.

19 The British physicist and architect (he was Surveyor to the City of London and chief assistant to Christo-
pher Wren) Robert Hooke (1635–1703) wrote about the gravity train idea in a letter to Isaac Newton. A
gravity train was used in the 2012 movie Total Recall.

122
S URFACE I NTEGRALS 3.3 S URFACE I NTEGRALS

˝ Then rotate our coordinate system about the y-axis so that

?V and M have
the same z-
coordinate Z ě 0. So the coordinates of V and M are ˘ R2 ´ Z2 , 0 , Z . Let’s sup-
? ?
R2 ´ Z2 , 0 , Z and M is at ´ R2 ´ Z2 , 0 , Z . It really doesn’t



pose that V is at
?
matter which is which, but we can always arrange that it is V at + R2 ´ Z2 , 0 , Z

by rotating around the z-axis by 180˝ if necessary.

? N ?
p´ R2 ´ Z 2 , 0 , Zq p R2 ´ Z 2 , 0 , Zq
G
x
p0, 0, 0q

The y- and z-coordinates of the train are always fixed at 0 and Z, respectively. So let’s call
the x-coordinate at time t x (t), and look at the x-component of Newton’s law of motion.

ma = G + N

It is
mx2 (t) = G ¨ ı̂ı
because the normal force N has no ı̂ı component. Recall that Newton’s law of gravity says
that
GMm
G=´ r
|r|3
where G is the gravitational constant, r is the vector from O to the train, and m is the mass
of the train. In this case, because of our computation in Example 3.3.10, the train only feels
gravity from shells of the Earth that are inside the train, so that M is the mass of the part

|r| G
x

123
 S URFACE I NTEGRALS 3.3 S URFACE I NTEGRALS

of the Earth whose distance to the centre of the Earth is no more than |r|. So

4
M= π|r|3 ρ
3
and
Gm 4
mx2 (t) = ´ π|r|3 ρ r ¨ ı̂ı
|r|3 3
so that
4πGρ
x2 (t) + x (t) = 0
3
This is exactly the differential equation of simple harmonic motion. We have seen it be-
4πGρ
fore in Example 2.2.7. Except for the constant 3 , it is identical to the equation solved in
Example H.4 of the Appendix H, entitled “Review of Linear Ordinary Differential Equa-
tions”. The general solution is
c ! c !
4πGρ 4πGρ
x (t) = C1 cos t + C2 sin t
3 3

? release the train, from rest, at t = 0, then

with C1?and C2 being arbitrary constants. If we
x (0) = R ´ Z and x (0) = 0 so that C1 = R2 ´ Z2 , C2 = 0 and
2 2 1

c
a
!
4πGρ
x (t) = R2 ´ Z2 cos t
3
b
?

4πGρ
The train reaches M when x (t) = ´ R2 ´ Z2 . That is, when cos 3 t = ´1. So the
transit time from V to M is d d
3 3π
T=π =
4πGρ 4Gρ

Notice that this transit time depends only on the gravitational constant G and the density
of the Earth ρ. In particular it is completely independent of

˝ where V and M are and, in particular,

˝ how close together V and M are, and also of
˝ the radius of the Earth.

In the case of the Earth, the transit time is about 42 minutes.

Example 3.3.11

3.3.5 §§ Optional — Dropping Higher Order Terms in du, dv

In the course of deriving (3.3.1), that is, n̂dS and dS formulae for

124
S URFACE I NTEGRALS 3.3 S URFACE I NTEGRALS

u“u0 u“u0 `du

v varying v varying

P2 P3 u varying
v“v0 `dv

P1 u varying
P0 v“v0

we approximated, for example, the vectors

ÝÝÑ Br Br
P0 P1 = r(u0 + du, v0 ) ´ r(u0 , v0 )= (u0 , v0 ) du + E1 « (u0 , v0 ) du
Bu Bu
ÝÝÑ Br Br
P0 P2 = r(u0 , v0 + dv) ´ r(u0 , v0 ) = (u0 , v0 ) dv + E2 « (u0 , v0 ) dv
Bv Bv
where E1 is bounded20 by a constant times du2 and E2 is bounded by a constant times dv2 .
That is, we assumed that we could just drop E1 and E2 .
So we approximated
ˇÝÝÑ ÝÝÑˇ ˇˇh Br i h Br
ˇ
iˇ
ˇ P0 P1 ˆ P0 P2 ˇ = ˇ (u0 , v0 ) du + E1 ˆ (u0 , v0 ) dv + E2 ˇ
Bu Bv
ˇ Br Br ˇ
ˇ ˇ
= ˇ (u0 , v0 ) du ˆ (u0 , v0 ) dv + E3 ˇ
Bu Bv
ˇ Br Br ˇ
ˇ ˇ
« ˇ (u0 , v0 ) du ˆ (u0 , v0 ) dvˇ
Bu Bv
where the length of the vector E3 is bounded by a constant times du2 dv + du dv2 . We’ll
now see why dropping terms like E3 does not change the value of the integral at all21 .
Suppose that our domain of integration consists of all (u, v)’s in a rectangle of width
A and height B, as in the figure below. Subdivide the rectangle into a grid of n ˆ n

v
du

dv
B

small subrectangles by drawing lines of constant v (the red lines in the figure) and lines of

20 Remember the error in the Taylor polynomial approximations.

21 See the optional §1.1.6 of the CLP-2 text for an analogous argument concerning Riemann sums.

125
 S URFACE I NTEGRALS 3.4 I NTERPRETATION OF F LUX I NTEGRALS

constant v (the blue lines in the figure). Each subrectangle has width du = An and height
dv = Bn . Now suppose that in setting up the integral we make, for each subrectangle, an
error that is bounded by some constant times
 A 2 B A  B 2 AB( A + B)
2 2
du dv + du dv = + =
n n n n n3
Because there are a total of n2 subrectangles, the total error that we have introduced, for
all of these subrectangles, is no larger than a constant times

AB( A + B) AB( A + B)
n2 ˆ 3
=
n n
When we define our integral by taking the limit n Ñ 0 of the Riemann sums, this error
converges to exactly 0.

3.4IJ Interpretation of Flux Integrals

We defined, in §3.3, two types of integrals overť surfaces. We have seen, in §3.3.4, some
applications that lead to integrals
ť of the type S ρ dS. We now look at one application that
leads to integrals of the type S F ¨ n̂ dS. Recall that integrals of this type are called flux
integrals. Imagine a fluid with
• the density of the fluid (say in kilograms per cubic meter) at position ( x, y, z) and
time t being ρ( x, y, z, t) and with

• the velocity of the fluid (say in meters per second) at position ( x, y, z) and time t
being v( x, y, z, t).
We are going to determine the rate (say in kilograms per second) at which the fluid is
flowing through a tiny piece dS of surface at ( x, y, z). During a tiny time interval of length
dt about time t, fluid near dS moves v( x, y, z, t)dt. The green line in the figure below is a
side view of dS and n̂ = n̂( x, y, z) is a unit normal vector to dS. So during that tiny time

n̂ vdt
vdt
θ

vdt vdt

interval
˝ the red line moves to the green line and
˝ the green line moves to the blue line so that
˝ the fluid filling the dark grey region below the green line crosses through dS and
moves to light grey region above the green line.

126
 S URFACE I NTEGRALS 3.4 I NTERPRETATION OF F LUX I NTEGRALS

If we denote by θ the angle between n̂ and vdt,

˝ the volume of fluid that crosses through dS during the time interval dt is the volume
whose side view is the dark grey region below the green line. This region has base
dS and height |vdt| cos θ and so has volume
|v( x, y, z, t)dt| cos θ dS = v( x, y, z, t) ¨ n̂( x, y, z) dt dS
because n̂( x, y, z) has length one.
˝ The mass of fluid that crosses dS during the time interval dt is then
ρ( x, y, z, t)v( x, y, z, t) ¨ n̂( x, y, z) dt dS
˝ and the rate at which fluid is crossing through dS is
ρ( x, y, z, t)v( x, y, z, t) ¨ n̂( x, y, z) dS
Integrating dS over a surface S, we conclude that

Lemma 3.4.1.

The rate at which fluid is crossing through a surface S is the flux integral
ij
ρ( x, y, z, t)v( x, y, z, t) ¨ n̂( x, y, z) dS
S

Here ρ is the density of the fluid, v is the velocity field of the fluid, and n̂( x, y, z)
is a unit normal to S at ( x, y, z). If the flux integral is positive the fluid is crossing
in the direction n̂. If it is negative the fluid is crossing opposite to the direction
of n̂.

3.4.1 §§ Examples of Flux Integrals

Example 3.4.2 (Point Source)

In Example 2.1.2, we found that the vector field of a point source22 (in three dimensions)
that creates 4πm liters per second is
m
v( x, y, z) = r̂( x, y, z)
r ( x, y, z)2
where b
xı̂ı + y̂ + zk̂
r ( x, y, z) = x 2 + y2 + z2 r̂( x, y, z) = a
x 2 + y2 + z2
We sketched it in Figure 2.1.1. We’ll now compute the flux of this vector field across a
sphere centred on the origin. Suppose that the sphere has radius R. Then the outward23

22 You can imagine that a very small pipe pumps water to the origin.

127
S URFACE I NTEGRALS 3.4 I NTERPRETATION OF F LUX I NTEGRALS

n̂

pointing Lemma 3.4.1 normal at a point ( x, y, z) on the sphere is

xı̂ı + y̂ + zk̂ xı̂ı + y̂ + zk̂

n̂( x, y, z) = r̂( x, y, z) = a =
x 2 + y2 + z2 R

Note that r̂( x, y, z) ¨ r̂( x, y, z) = 1 and that, on the sphere, r ( x, y, z) = R. So the flux of v
outward through the sphere is
ij ij
m
v ¨ n̂ dS = r̂( x, y, z) ¨ r̂( x, y, z) dS
r ( x, y, z)2
S S
ij
m m
= 2
dS = 2 4πR2
R R
S
= 4πm

This is the rate at which fluid is exiting the sphere. In our derivation of the vector field we
assumed that the fluid is incompressible, so it is also the rate at which the point source is
creating fluid.
Example 3.4.2

Example 3.4.3 (Vortex)

In Figure 2.1.3, we sketched the vector field (in two dimensions)

v( x, y) = Ω ´ yı̂ı + x̂

We’ll now compute the flux of this vector field across a circle C centred on the origin.
Suppose that the circle has radius R. By definition, in two dimensions, the flux of a

23 It doesn’t really matter which unit normal we pick here. We just have to be clear which one we’re
using. With the outward normal, the flux gives the rate at which fluid crosses the sphere in the outward
direction. If we were to use the inward pointing normal, the flux would give the rate at which fluid
crosses the sphere in the inward direction.

128
S URFACE I NTEGRALS 3.4 I NTERPRETATION OF F LUX I NTEGRALS

n̂

ş
vector field across a curve C is C v ¨ n̂ ds.
This is the natural analog of the flux in three dimensions — the surface S has been
replaced by the curve C, and the surface area dS of a tiny piece of S has been replaced by
the arc length ds of a tiny piece of C.
The outward pointing unit normal at a point ( x, y) on our circle C is

xı̂ı + y̂ xı̂ı + y̂

n̂( x, y) = a =
x +y2 2 R

So
Ω
v( x, y) ¨ n̂( x, y) = ´ yı̂ı + x̂ ¨ xı̂ı + y̂ = 0



R
and the flux across C is
ż
v ¨ n̂ ds = 0
C

This should not be a surprise — no fluid is crossing C at all. This is exactly what we would
expect from looking at the arrows in Figure 2.1.3 or at the stream lines in Example 2.2.6.
Example 3.4.3

Example 3.4.4
ť
Problem: Evaluate S F ¨ n̂ dS where

F( x, y, z) = ( x + y) ı̂ı + (y + z) ̂ + ( x + z) k̂
ˇ (
and S is the boundary of V = ( x, y, z) ˇ 0 ď x2 + y2 ď 9, 0 ď z ď 5 , and n̂ is the
outward normal24 to S.

Solution. The volume V looks like a tin can of radius 3 and height 5. It is natural to

24 It is necessary that the problem specify, one way or another, whether

ť n̂ is the inward pointing normal
or the outward pointing normal. Without this, the meaning of S F ¨ n̂ dS is ambiguous. Think about
where the orientation of the normal vector gets used in your solution.

129
S URFACE I NTEGRALS 3.4 I NTERPRETATION OF F LUX I NTEGRALS

n̂ “ k̂
St

n̂
Ss

Sb
n̂ “ ´k̂

decompose its surface S into three parts

ˇ (
St = ( x, y, z) ˇ 0 ď x2 + y2 ď 9, z = 5 = the top
ˇ (
Sb = ( x, y, z) ˇ 0 ď x2 + y2 ď 9, z = 0 = the bottom
ˇ (
Ss = ( x, y, z) ˇ x2 + y2 = 9, 0 ď z ď 5 = the side

We’ll compute the flux through each of the three parts separately and then add them
together.
The Top: On the top, the outward pointing normal to S is n̂ = k̂ and dS = dxdy. This is
probably intuitively obvious. But if it isn’t, you can always derive it by parametrizing the
top by r( x, y) = x ı̂ı + y ̂ + 5 k̂ with x2 + y2 ď 9. So the flux through the top is
ij ij ij
F ¨ n̂ dS = ( x + z) dxdy = ( x + 5) dxdy
St x2 +y2 ď9 x2 +y2 ď9
z =5

ť
The integral x2 +y2 ď9 x dxdy = 0 since x is odd and the domain of integration is symmet-
ric about x = 0. So
ij ij
F ¨ n̂ dS = 5 dxdy = 5π (3)2 = 45π
St x2 +y2 ď9

The Bottom: On the bottom, the outward pointing normal to S is n̂ = ´k̂ and dS = dxdy.
So the flux through the bottom is
ij ij ij
F ¨ n̂ dS = ´ ( x + z) dxdy = ´ x dxdy = 0
Sb x2 +y2 ď9 x2 +y2 ď9
z =0

again since x is odd and the domain of integration is symmetric about x = 0.

The Side: We can parametrize the side by using cylindrical coordinates.

0 ď θ ă 2π, 0 ď z ď 5


r(θ, z) = 3 cos θ , 3 sin θ , z

130
S URFACE I NTEGRALS 3.4 I NTERPRETATION OF F LUX I NTEGRALS

Then, using (3.3.1),

Br n̂
= (´3 sin θ , 3 cos θ , 0)
Bθ
Br r
= (0 , 0 , 1) θ
Bz
Br Br
n̂ dS = ˆ dθ dz
Bθ Bz
top view
= (3 cos θ , 3 sin θ , 0) dθ dz

Note that n̂ = (cos θ , sin θ , 0) is outward pointing25 , as desired. Continuing,

F x (θ, z) , y(θ, z) , z(θ, z) = 3(cos θ + sin θ ) ı̂ı + (3 sin θ + z) ̂ + (3 cos θ + z) k̂
(
F ¨ n̂ dS = 9 cos2 θ + 3 sin θ cos θ + 9 sin2 θ + 3z sin θ dθ dz
(
= 9 + 3/2 sin(2θ ) + 3z sin θ dθ dz

So the flux through the side is

ij ż 2π ż5
(
F ¨ n̂ dS = dθ dz 9 + 3/2 sin(2θ ) + 3z sin θ
0 0
Ss
ż 2π ż5 ż 2π ż 2π
=9 dθ dz since sin θ dθ = sin(2θ ) dθ = 0
0 0 0 0
= 9 ˆ 2π ˆ 5 = 90π

and the total flux is

ij ij ij ij
F ¨ n̂ dS = F ¨ n̂ dS + F ¨ n̂ dS + F ¨ n̂ dS = 45π + 0 + 90π = 135π
S St Sb Ss

Example 3.4.4

Example 3.4.5
ť
Problem: Evaluate S F ¨ n̂ dS where F( x, y, z) = x4ı̂ı + 2y2 ̂ + zk̂, S is the half of the sur-
face 14 x2 + 91 y2 + z2 = 1 with z ě 0, and n̂ is the upward pointing unit normal.

25 To check, draw, in your head, a sketch of the top view of the can. “Top view” just means “ignore the
z-coordinate”. The top view of the can is a circle of radius 3. Then, at a generic point, r = (cos θ, sin θ ),
on the can, draw the unit normal n̂ = (cos θ , sin θ ) with its tail at r. It is pointing away from the origin,
just like r is. That is, n̂ is pointing outward.

131
S URFACE I NTEGRALS 3.4 I NTERPRETATION OF F LUX I NTEGRALS

z
1 2
4
x ` 19 y 2 ` z 2 “ 1

Solution 1. We start by parametrizing the surface, which is half of an ellipsoid. By way of

motivation for the parametrization, recall that spherical coordinates, with ρ = 1, provide
a natural way to parametrize the sphere x2 + y2 + z2 = 1. Namely x = cos θ sin ϕ, y =
sin θ sin ϕ, z = cos ϕ. The reason that these spherical coordinates work is that the trig
identity cos2 α + sin2 α = 1 implies

x2 + y2 = cos2 θ sin2 ϕ + sin2 θ sin2 ϕ = sin2 ϕ

and then
x2 + y2 + z2 = sin2 ϕ + cos2 ϕ = 1

The equation of our ellipsoid is

 x 2  y 2
+ + z2 = 1
2 3
y
so we can parametrize the ellipsoid by replacing x with 2x and y with 3 in our parametriza-
tion of the sphere. That is, we choose the parametrization

x (θ, ϕ) = 2 cos θ sin ϕ

y(θ, ϕ) = 3 sin θ sin ϕ
z(θ, ϕ) = cos ϕ

with (θ, ϕ) running over 0 ď θ ď 2π, 0 ď ϕ ď π/2. Note that

1 1
x (θ, ϕ)2 + y(θ, ϕ)2 + z(θ, ϕ)2 = 1
4 9
as desired.
Then, using (3.3.1),
 Bx By Bz 
, , = (´2 sin θ sin ϕ , 3 cos θ sin ϕ , 0)
Bθ Bθ Bθ
 Bx By Bz 
, , = (2 cos θ cos ϕ , 3 sin θ cos ϕ , ´ sin ϕ)
Bϕ Bϕ Bϕ
 Bx By Bz   Bx By Bz 
n̂ dS = ´ , , ˆ , , dθdϕ
Bθ Bθ Bθ Bϕ Bϕ Bϕ
= ´(´3 cos θ sin2 ϕ, ´2 sin θ sin2 ϕ, ´6 sin ϕ cos ϕ)dθdϕ

132
S URFACE I NTEGRALS 3.4 I NTERPRETATION OF F LUX I NTEGRALS

The extra minus sign in n̂ dS was put there to make the z component of n̂ positive. (The
problem specified that n̂ is to be upward unit normal.) As
F x (θ, ϕ) , y(θ, ϕ) , z(θ, ϕ) = 24 cos4 θ sin4 ϕ ı̂ı + 2 ˆ 32 sin2 θ sin2 ϕ ̂ + cos ϕ k̂

we have
h i
4 5 6 2 3 4 2
F ¨ n̂ dS = 3 ˆ 2 cos θ sin ϕ + 2 ˆ 2 ˆ 3 sin θ sin ϕ + 6 sin ϕ cos ϕ dθ dϕ

and the desired integral

ij żπ ż 2π
2
h i
4 5 6 2 3 4 2
F ¨ n̂ dS = dϕ dθ 3 ˆ 2 cos θ sin ϕ + 2 ˆ 2 ˆ 3 sin θ sin ϕ + 6 sin ϕ cos ϕ
0 0
S
ş2π ş2π
Since 0 cosm θ dθ = 0 sinm θ dθ = 0 for all odd26 natural numbers m,
ij ż π/2 ż 2π ż π/2 h 1 iπ/2
2 2 3
F ¨ n̂ dS = dϕ dθ 6 sin ϕ cos ϕ = 12π dϕ sin ϕ cos ϕ = 12π ´ cos ϕ
0 0 0 3 0
S
= 4π
The integral was evaluated by guessing (and checking) that ´ 31 cos3 ϕ is an antiderivative
of sin ϕ cos2 ϕ. It can also be done by substituting u = cos ϕ, du = ´ sin ϕ dϕ.
Solution 2. This time we’ll parametrize the half-ellipsoid using a variant of cylindrical
coordinates.
x (r, θ ) = 2r cos θ
y(r, θ ) = 3r sin θ
a
z(r, θ ) = 1 ´ r2
with (r, θ ) running over 0 ď θ ď 2π, 0 ď r ď 1. Because we built the factors of 2 and 3
into x (r, θ ) and y(r, θ ), we have
x (r, θ )2 y(r, θ )2
+ = r2 cos2 θ + r2 sin2 θ = r2
4 9
x (r, θ )2 y(r, θ )2
ùñ + + z(r, θ )2 = 1
4 9
as desired. Further z(r, θ ) ě 0 by our choice of square root in the definition of z(r, θ ).
So, using (3.3.1),
 Bx By Bz 
, , = (´2r sin θ, 3r cos θ, 0)
Bθ Bθ Bθ
 Bx By Bz   r 
, , = 2 cos θ, 3 sin θ, ´ ?
Br Br Br 1 ´ r2
 Bx By Bz   Bx By Bz 
n̂dS = ´ , , ˆ , , dr dθ
Bθ Bθ Bθ Br Br Br
 3r2 cos θ 2r2 sin θ 
=´ ´ ? ,´? , ´6r dr dθ
1 ´ r2 1 ´ r2

26 Look at the graphs of cosm ϕ and sinm ϕ.

133
S URFACE I NTEGRALS 3.4 I NTERPRETATION OF F LUX I NTEGRALS

Once again, the extra minus sign in n̂dS was put there to make the z component of n̂
positive. Continuing,
a
F x (r, θ ) , y(r, θ ) , z(r, θ ) = 24 r4 cos4 θ ı̂ı + 2 ˆ 32 r2 sin2 θ ̂ + 1 ´ r2 k̂

6 4 a
4? r 2 2? r
h i
5 3
F ¨ n̂ dS = 3 ˆ 2 cos θ + 2 3 sin θ + 6r 1 ´ r dr dθ
2
1 ´ r2 1 ´ r2
ş2π ş2π
Again using that 0 cosm θ dθ = 0 sinm θ dθ = 0 for all odd natural numbers m,
ż ż 1 ż 2π a
F ¨ n̂ dS = dr dθ 6r 1 ´ r2
S 0 0
ż1 a h 1 i1
2 3/2
= 12π dr r 1 ´ r = 12π ´ (1 ´ r )
2
0 3 0
= 4π
1 2 3/2
? was evaluated by guessing (and checking) that ´ 3 (1 ´ r ) is an antideriva-
The integral
tive of r 1 ´ r2 . It can also be done by substituting u = 1 ´ r2 , du = ´2r dr.
Solution 3. The surface is of the form G ( x, y, z) = 0 with G ( x, y, z) = 41 x2 + 19 y2 + z2 ´ 1.
Hence, using (3.3.3),
x 2y
∇G 2 ı̂ı + 9 ̂ + 2z k̂
x y 
n̂dS = dx dy = dx dy = ı̂ı + ̂ + k̂ dx dy
∇ G ¨ k̂ 2z 4z 9z
 x5 2y3 
ùñ F ¨ n̂ dS = + + z dx dy
4z 9z
It is true that n̂dS, and consequently F ¨ n̂ dS become infinite27 as z Ñ 0. So we should
really treat the integral as an improper integral, first integrating over z ě ε and then taking
the limit ε Ñ 0+ . But, as we shall see,bthe singularity is harmless. So it is standard to gloss
2 y2 2 y2
over this point. On S, z = z( x, y) = 1 ´ x4 ´ 9 and x4 + 9 ď 1, so
ż ij 
x5 2y3 
F ¨ n̂ dS = + + z( x, y) dx dy
S 4z( x, y) 9z( x, y)
x2 y2
4 + 9 ď1
5 2y3
Both 4z(xx,y) and 9z( x,y) are odd under x Ñ ´x, y Ñ ´y and the domain of integration is
even under x Ñ ´x, y Ñ ´y, so their integrals are zero and
ż ij
F ¨ n̂ dS = z( x, y) dx dy
S
x2 y2
4 + 9 ď1
ij c
x 2 y2
= 1´ ´ dx dy
4 9
x2 y2
4 + 9 ď1

To evaluate this integral, first make the change of variables28 x = 2X, dx = 2dX, y = 3Y,

27 That’s because the ellipsoid is becoming vertical as z Ñ 0, so that x and y are not really good parameters
there.
28 The reader interested in general changes of variables in multidimensional integrals should look up
“Jacobian determinant”.

134
 S URFACE I NTEGRALS 3.5 O RIENTATION OF S URFACES

dy = 3dY to give ż ij a
F ¨ n̂ dS = 1 ´ X 2 ´ Y 2 6 dX dY
S
X 2 +Y 2 ď1
Then switch to polar coordinates, X = r cos θ, Y = r sin θ, dXdY = r drdθ to give
ż ż 1 ż 2π a ż1 a h 1 i1
2 3/2
F ¨ n̂ dS = dr dθ 6r 1 ´ r = 12π
2 dr r 1 ´ r = 12π ´ (1 ´ r )
2
S 0 0 0 3 0
= 4π
b
2 y2
Solution 4. The surface is of the form z = f ( x, y) with f ( x, y) = 1 ´ x4 ´ 9 . Hence,
using (3.3.2),
 
h Bf xı y
Bf i ı̂ + 
̂
n̂dS = ´ ı̂ı ´ ̂ + k̂ dx dy =  b 4 9
+ k̂ dx dy
Bx By x2 y2
1´ 4 ´ 9

2y3
c 
x5
+ x 2 y2 
ùñ F ¨ n̂ dS =  b 4 9
+ 1´ ´ dx dy
x2 y2 4 9
1´ 4 ´ 9

Note that our unit normal is upward pointing, as required. As in Solution 3, by the odd-
ness of the x5 and y3 terms in the integrand,
ż ij

2y3
c 
x5 2 2
+
F ¨ n̂ dS =  4? 9 + 1 ´ x ´ y  dx dy
S ¨¨¨ 4 9
x2 y2
4 + 9 ď1
ij c
x 2 y2
= 1´ ´ dx dy
4 9
x2 y2
4 + 9 ď1

Now continue as in Solution 3.

Example 3.4.5

3.5IJ Orientation of Surfaces

One thing that made the flux integrals of the last section possible is that we could choose
sensible unit normal vectors n̂. In this section, we explain this more carefully.
Consider the sphere x2 + y2 + z2 = 1. We can think of this surface as having two sides
— an inside (the side you see when you are living inside the sphere) and an outside (the
side you see when you are living outside the sphere). Concentrate on one point ( x0 , y0 , z0 )
on the sphere. The surface x2 + y2 + z2 = 1 has precisely two unit normal vectors at
( x0 , y0 , z0 ), namely
n̂+ = +( x0 , y0 , z0 ) and n̂´ = ´( x0 , y0 , z0 )

135
S URFACE I NTEGRALS 3.5 O RIENTATION OF S URFACES

We can view n̂+ as being associated to (or attached to) the outside of the sphere and n̂´ as
being associated to (or attached to) the inside of the sphere. Note that, as we move over
the sphere, both n̂+ and n̂´ change continuously.
Definition 3.5.1.

An oriented surface is a surface together with a continuous function

N̂ : S Ñ R3

such that, for each point p of S, N̂( p) is a unit normal to S at p.

Example 3.5.2 (Sphere)

ˇ (
One orientation of the sphere S = ( x, y, z) ˇ x2 + y2 + z2 = 1 is
N̂( x, y, z) = ( x, y, z)
It associates to each point p of S the outward pointing unit normal to S at p. We can think
of S with this orientation as being the outer side of S.ˇ (
The other orientation of the sphere S = ( x, y, z) ˇ x2 + y2 + z2 = 1 is
N̂( x, y, z) = ´( x, y, z)
It associates to each point p of S the inward pointing unit normal to S at p. We can think
of S with this orientation as being the inner side of S.
While this discussion might seem inordinately picky, it turns out that not all surfaces
can be oriented. Our next example exhibits one.
Example 3.5.2

Example 3.5.3 (Optional — The Möbius Strip)

There are some surfaces S for which it is not possible to choose a continuous orientation
map N̂ : S Ñ R3 . Such surfaces are said to be non-orientable. The most famous non-
orientable surface is the Möbius29 strip30 , which you can construct as follows. Take a
rectangular strip of paper. Lay it flat and then introduce a half twist so that the arrow on

29 August Ferdinand Möbius (1790–1868) was a German mathematician and astronomer. He was a de-
scendant of Martin Luther and a student of Gauss.
30 Another famous non-orientable surface is the Klein bottle. You can easily find discussions of it using
your favourite search engine.

136
S URFACE I NTEGRALS 3.5 O RIENTATION OF S URFACES

the right hand end points upwards, rather than downwards. Then glue the two ends of
the strip together, with the two arrows coinciding. That’s the Möbius strip.

Let’s parametrize it. Think of the strip of paper that we used to construct it as consist-
ing of a backbone (the horizontal black line in the figure below) with a bunch of ribs (like
the thick blue line in the figure) emanating from it. When we glue the two ends of the

w
θ

strip together, the black line forms a circle. If the strip has length `, the circle will have
`
circumference ` and hence radius 2π . We’ll parametrize it as the circle

`
r̂(θ ) where r̂(θ ) = cos(θ ) ı̂ı + sin(θ ) ̂
2π
This circle is in the xy-plane. It is the black circle in the figure below. (The figure only
shows the part of the circle in the first octant, i.e. with x, y, z ě 0.) Now we’ll add in the

k̂
y
θℓ
2π
r̂pθq

r̂pθq
x

`
blue ribs. We’ll put the blue rib, that is attached to the backbone at 2π r̂(θ ), in the plane
that contains the vectors r̂(θ ) and k̂. A side view of the plane that contains the vectors
r̂(θ ) and k̂ is sketched in the figure below. To put the half twist into the strip of paper, we

137
S URFACE I NTEGRALS 3.5 O RIENTATION OF S URFACES

k̂ upv, θ, ϕq
ϕ
ℓ r̂pθq
2π
r̂pθq

want the blue rib to rotate about the backbone by 180˝ , i.e. π radians, as θ runs from 0 to
2π. That will be the case if we pick the angle ϕ in the figure to be θ/2. The vector that is
running along the blue rib in the figure is
u(v, θ, ϕ) = v cos( ϕ) r̂(θ ) + v sin( ϕ) k̂
where the length, v, of the vector is a parameter. If the width of our original strip of paper
is w, then as the parameter v runs from ´w/2 to +w/2, the tip of the vector u(v, θ, ϕ) runs
over the entire blue rib. So, choosing ϕ = θ/2, our parametrization of the Möbius strip is
`
r(θ, v) = r̂(θ ) + u(v, θ, θ/2)
2π
` w w
0 ď θ ă 2π, ´ ďvď



= r̂(θ ) + v cos θ/2 r̂(θ ) + v sin θ/2 k̂
2π 2 2
where r̂(θ ) = cos(θ ) ı̂ı + sin(θ ) ̂ .
Now that we have parametrized the Möbius strip, let’s return to the question of ori-
entability. Recall, from Definition 3.5.1, that, if the Möbius strip were orientable, there
would exist a continuous function N̂ which assigns to each point r of the strip a unit nor-
mal vector N̂(r) at r. First, we’ll find the normal vectors to the surface using (3.3.1). The
partial derivatives
Br ` 1 v v
r̂ (θ ) + v cos θ/2 r̂1 (θ ) ´ sin θ/2 r̂(θ ) + cos θ/2 k̂




(θ, v) =
Bθ 2π 2 2
Br


(θ, v) = cos θ/2 r̂(θ ) + sin θ/2 k̂
Bv
are relatively messy, so let’s just consider the case v = 0 (i.e. find the normal vectors a the
backbone). Then
Br ` 1
(θ, 0) = r̂ (θ )
Bθ 2π
Br


(θ, 0) = cos θ/2 r̂(θ ) + sin θ/2 k̂
Bv
Since
r̂1 (θ ) ˆ r̂(θ ) = ´ sin(θ ) ı̂ı + cos(θ ) ̂ ˆ cos(θ ) ı̂ı + sin(θ ) ̂ = ´k̂

r̂1 (θ ) ˆ k̂ = ´ sin(θ ) ı̂ı + cos(θ ) ̂ ˆ k̂ = r̂(θ )

we have
Br Br `  
(θ, 0) ˆ (θ, 0) = ´ cos θ/2 k̂ ´ sin θ/2 r̂(θ )



Bθ Bv 2π

138
S URFACE I NTEGRALS 3.5 O RIENTATION OF S URFACES

As k̂ and r̂(θ ) are mutually perpendicular unit vectors, cos θ/2 k̂ ´ sin θ/2 r̂(θ ) has

length one, and the two unit normal vectors to the Möbius strip at r(θ, 0) are
 
˘ cos θ/2 k̂ ´ sin θ/2 r̂(θ )

So, for each θ, N̂ r(θ, 0) must be either

cos θ/2 k̂ ´ sin θ/2 r̂(θ ) ´ cos θ/2 k̂ ´ sin θ/2 r̂(θ )






or


Imagine walking along the Möbius strip. The normal vector N̂ r(θ, v) is our body when


we are at r(θ, v) — our feet are at the tail of the vector N̂ r(θ, v) and our head is at the ar-
` ı ` ı




row of N̂ r(θ, v) . We start walking at r(0, 0) = 2π ı̂ . Our body, N̂ 2π ı̂ = N̂ r(0, 0) has to
be one of ˘ cos(0) k̂ ´ sin(0) r̂(0) = ˘k̂. Let’s suppose that N̂ r(0, 0) = +k̂. (We start

upright.) Now we
start walking along the backbone of the Möbius strip,

increasing
θ. Be-
cause N̂ r(θ, 0) has to be continuous, N̂ r(θ, 0) has to be + cos /2 k̂ ´ sin θ/2 r̂(θ ) .



θ

We keep increasing θ. By continuity, N̂ r(θ, 0) has to be + cos θ/2 k̂ ´ sin θ/2 r̂(θ )

for bigger and bigger θ. Eventually we get to θ = 2π, i.e. to

` ` `
r(2π, 0) = r̂(2π ) = ı̂ı = r̂(0) = r(0, 0)
2π 2π 2π
We are back to our starting point. Continuity has forced

ˇˇ
ˇˇ
= + cos /2 k̂ ´ sin /2 r̂(θ ) ˇ = ´k̂




N̂ r(2π, 0) = N̂ r(θ, 0) ˇ θ θ
θ =2π θ =2π

` ı



So we have arrived back upside down. That’s a problem — N̂ r(2π, 0) = N̂ 2π ı̂ and
` ı
ı̂ = +k̂, not ´k̂. So the Möbius strip is not orientable. The


we have already defined N̂ 2π
interested reader should look up M. C. Escher’s Möbius Strip II (Red Ants).
Example 3.5.3

139
 Chapter 4

I NTEGRAL T HEOREMS

4.1IJ Gradient, Divergence and Curl

“Gradient, divergence and curl”, commonly called “grad, div and curl”, refer to a very
widely used family of differential operators and related notations that we’ll get to shortly.
We will later see that each has a “physical” significance. But even if they were only short-
hand1 , they would be worth using.
For example, one of Maxwell’s equations (relating the electric field E and the magnetic
field B) written without the use this notation is

 BE BE2   BE3 BE1   BE BE  1  BB1 BB2 BB3 

3 2
´ ı̂ı ´ ´ ̂ + ´ 1 k̂ = ´ ı̂ı + ̂ + k̂
By Bz Bx Bz Bx By c Bt Bt Bt

The same equation written using this notation is

1 BB
∇ˆE = ´
c Bt

The shortest way to write (and easiest way to remember) gradient, divergence and curl
B
uses the symbol “∇ ” which is a differential operator like Bx . It is defined by

B B B
∇ = ı̂ı + ̂ + k̂
Bx By Bz

and is called “del” or “nabla”. Here are the definitions.

1 Good shorthand is not only more brief, but also aids understanding “of the forest by hiding the trees”.

140
I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

Definition 4.1.1.

(a) The gradient of a scalar-valued function f ( x, y, z) is the vector field

Bf Bf Bf
grad f = ∇ f = ı̂ı + ̂ + k̂
Bx By Bz

Note that the input, f , for the gradient is a scalar-valued function, while the
output,∇ f , is a vector-valued function.

(b) The divergence of a vector field F( x, y, z) is the scalar-valued function

BF1 BF2 BF3

div F = ∇ ¨ F = + +
Bx By Bz

Note that the input, F, for the divergence is a vector-valued function, while
the output, ∇ ¨ F, is a scalar-valued function.

(c) The curl of a vector field F( x, y, z) is the vector field

 BF BF2   BF3 BF1   BF BF 
3 2
curl F = ∇ ˆ F = ´ ı̂ı ´ ´ ̂ + ´ 1 k̂
By Bz Bx Bz Bx By

Note that the input, F, for the curl is a vector-valued function, and the output,
∇ ˆ F, is a again a vector-valued function.
(d) The Laplacian2 of a scalar-valued function f ( x, y, z) is the scalar-valued func-
tion
2 B2 f B2 f B2 f
∆f = ∇ f = ∇ ¨∇f = 2 + 2 + 2
Bx By Bz
The Laplacian of a vector field F( x, y, z) is the vector field

B2 F B2 F B2 F
∆F = ∇ 2 F = ∇ ¨ ∇ F = + 2+ 2
Bx2 By Bz

Note that the Laplacian maps either a scalar-valued function to a scalar-

valued function, or a vector-valued function to a vector-valued function.

The gradient, divergence and Laplacian all have obvious generalizations to dimensions
other than three. That is not the case for the curl. It does have a, far from obvious, gener-
alization, which uses differential forms. Differential forms are well beyond our scope, but
are introduced in the optional §4.7.
Example 4.1.2

2 Pierre-Simon Laplace (1749–1827) was a French mathematician and astronomer. He is also the Laplace
of Laplace’s equation, the Laplace transform, and the Laplace-Bayes estimator. He was Napoleon’s
examiner when Napoleon attended the Ecole Militaire in Paris.

141
 I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

As an example of an application in which both the divergence and curl appear, we have
Maxwell’s equations3 4 5 , which form the foundation of classical electromagnetism.
∇ ¨ E = 4πρ
∇¨B = 0
1 BB
∇ˆE+ =0
c Bt
1 BB 4π
∇ˆB´ = J
c Bt c
Here E is the electric field, B is the magnetic field, ρ is the charge density, J is the current
density and c is the speed of light.
Example 4.1.2

4.1.1 §§ Vector Identities

d
Two computationally extremely important properties of the derivative dx are linearity and
the product rule.
d
df dg
a f ( x ) + bg( x ) = a ( x ) + b ( x )
dx dx dx
d
df dg
f ( x ) g( x ) = ( x ) g( x ) + f ( x ) ( x )
dx dx dx
Gradient, divergence and curl also have properties like these, which indeed stem (often
easily) from them. First, here are the statements of a bunch of them. (A memory aid and
proofs will come later.) In fact, here are a very large number of them. Many are included
just for completeness. Only a relatively small number are used a lot. They are in red.
Theorem 4.1.3 (Gradient Identities).

(a) ∇ ( f + g) = ∇ f + ∇ g

(b) ∇ (c f ) = c ∇ f , for any constant c

(c) ∇ ( f g) = (∇
∇ f ) g + f (∇
∇ g)
(d) ∇ ( f /g) = g ∇ f ´ f ∇ g /g2 at points x where g(x) ‰ 0.

(e) ∇ (F ¨ G) = F ˆ (∇
∇ ˆ G) ´ (∇
∇ ˆ F) ˆ G + (G ¨ ∇ )F + (F ¨ ∇ )G
Here6
BF BF BF
( G ¨ ∇ ) F = G1 + G2 + G3
Bx By Bz

3 To be picky, these are Maxwell’s equations in the absence of a material medium and in Gaussian units.
4 One important consequence of Maxwell’s equations is that electromagnetic radiation, like light, propa-
gate at the speed of light.
5 James Clerk Maxwell (1831–1879) was a Scottish mathematical physicist. In a poll of prominent physi-
cists, Maxwell was voted the third greatest physicist of all time. Only Newton and Einstein beat him.

142
I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

Theorem 4.1.4 (Divergence Identities).

(a) ∇ ¨ (F + G) = ∇ ¨ F + ∇ ¨ G

(b) ∇ ¨ (cF) = c ∇ ¨ F, for any constant c

(c) ∇ ¨ ( f F) = (∇
∇f)¨F + f ∇ ¨F
(d) ∇ ¨ (F ˆ G) = (∇
∇ ˆ F) ¨ G ´ F ¨ (∇
∇ ˆ G)

Theorem 4.1.5 (Curl Identities).

(a) ∇ ˆ (F + G) = ∇ ˆ F + ∇ ˆ G

(b) ∇ ˆ (cF) = c ∇ ˆ F, for any constant c

(c) ∇ ˆ ( f F) = (∇
∇f)ˆF + f ∇ ˆF
(d) ∇ ˆ (F ˆ G) = F(∇
∇ ¨ G) ´ (∇
∇ ¨ F)G + (G ¨ ∇ )F ´ (F ¨ ∇ )G
Here
BF BF BF
( G ¨ ∇ ) F = G1 + G2 + G3
Bx By Bz

Theorem 4.1.6 (Laplacian Identities).

(a) ∇ 2 ( f + g) = ∇ 2 f + ∇ 2 g

(b) ∇ 2 (c f ) = c ∇ 2 f , for any constant c

(c) ∇ 2 ( f g) = f ∇ 2 g + 2∇ f ¨ ∇ g + g ∇ 2 f

Theorem 4.1.7 (Degree Two Identities).

(a) ∇ ¨ (∇
∇ ˆ F) = 0 (divergence of curl)

(b) ∇ ˆ (∇
∇f) = 0 (curl of gradient)

(c) ∇ ¨ f t∇ g ˆ ∇ hu = ∇ f ¨ (∇∇ g ˆ ∇ h)

(d) ∇ ¨ ( f ∇ g ´ g∇ f ) = f ∇ 2 g ´ g ∇ 2 f

(e) ∇ ˆ (∇ ∇ ¨ F) ´ ∇ 2 F
∇ ˆ F) = ∇ (∇ (curl of curl)

6 ∇ F) does not make sense because

This is really the only definition that makes sense. For example G ¨ (∇
you can’t take the gradient of a vector-valued function.

143
I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

Memory Aid. Most of the vector identities (in fact all of them except Theorem 4.1.3.e, The-
orem 4.1.5.d and Theorem 4.1.7) are really easy to guess. Just combine the conventional
linearity and product rules with the facts that

˝ if the left hand side is a vector (scalar), then the right hand side must also be a vector
(scalar) and
˝ the only valid products of two vectors are the dot and cross products and
˝ the product of a scalar with either a scalar or a vector cannot be either a dot or cross
product and
˝ A ˆ B = ´B ˆ A. (The cross product is antisymmetric.)

For example, consider Theorem 4.1.4.c, which says ∇ ¨ ( f F) = (∇

∇ f ) ¨ F + f ∇ ¨ F.

˝ The left hand side, ∇ ¨ ( f F), is a scalar, so the right hand side must also be a scalar.
˝ The left hand side, ∇ ¨ ( f F), is a derivative of the product of f and F, so, mimick-
ing the product rule, the right hand side will be a sum of two terms, one with F
multiplying a derivative of f , and one with f multiplying a derivative of F.
˝ The derivative acting on f must be ∇ f , because ∇ ¨ f and ∇ ˆ f are not well-defined.
To end up with a scalar, rather than a vector, we must take the dot product of ∇ f
and F. So that term is (∇∇ f ) ¨ F.
˝ The derivative acting on F must be either ∇ ¨ F or ∇ ˆ F. We also need to multiply
by the scalar f and end up with a scalar. So the derivative must be a scalar, i.e. ∇ ¨ F
and that term is f t∇ ¨ Fu.
˝ Our final guess is ∇ ¨ ( f F) = (∇∇ f ) ¨ F + f ∇ ¨ F, which, thankfully, is correct.

Proof of Theorems 4.1.3, 4.1.4, 4.1.5, 4.1.6 and 4.1.7. All of the proofs (except for those of The-
orem 4.1.7.c,d, which we will return to later) consist of

˝ writing out the definition of the left hand side and

˝ writing out the definition of the right hand side and
˝ observing (possibly after a little manipulation) that they are the same.

For Theorem 4.1.3.a,b, Theorem 4.1.4.a,b, Theorem 4.1.5.a,b and Theorem 4.1.6.a,b, the
computation is trivial — one line per identity, if one uses some efficient notation. Rename
the coordinates x, y, z to x1 , x2 , x3 and the standard unit basis vectors ı̂ı , ̂ , k̂ to ı̂ı 1 , ı̂ı 2 , ı̂ı 3 . Then
ř
∇ = 3n=1 ı̂ı n Bx
B
n
and the proof of, for example, Theorem 4.1.4.a is

3
ÿ B
∇ ¨ (F + G) = ı̂ı n ¨ (F + G)
Bxn
n =1
ÿ 3 3
ÿ
B B
= ı̂ı n ¨ F + ı̂ı n ¨ G = ∇ ¨ F + ∇ ¨ G
Bxn Bxn
n =1 n =1

For Theorem 4.1.3.c,d, Theorem 4.1.4.c, Theorem 4.1.5.c and Theorem 4.1.6.c, the compu-

144
I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

tation is easy — a few lines per identity. For example, the proof of Theorem 4.1.5.c is
3
ÿ 3
ÿ
B B
∇ ˆ ( f F) = ı̂ı n ˆ ( f F) = f tı̂ı n ˆ Fu


Bxn Bxn
n =1 n =1
ÿ 3 3
ÿ
Bf B
= ı̂ı n ˆ F + f ı̂ı n ˆ F (by Theorem 1.1.2.b)
Bxn Bxn
n =1 n =1
∇f)ˆF + f ∇ ˆF
= (∇

For Theorem 4.1.4.d, the computation is also easy if one uses the fact that

a ¨ (b ˆ c) = (a ˆ b) ¨ c

which is Lemma 4.1.8.a below.

That leaves the proofs of Theorem 4.1.3.e, Theorem 4.1.5.d, Theorem 4.1.7.a,b,c,d,e, which
we write out explicitly.

Theorem 4.1.3.e:
First write out the left hand side as
3
ÿ 3
ÿ  ÿ3
B  BF  BG 
∇ (F ¨ G) = ı̂ı n (F ¨ G) = ı̂ı n ¨G + ı̂ı n F ¨
Bxn Bxn Bxn
n =1 n =1 n =1

Then rewrite a ˆ (b ˆ c) = (c ¨ a)b ´ (b ¨ a)c, which is Lemma 4.1.8.b below, as

(c ¨ a)b = a ˆ (b ˆ c) + (b ¨ a)c
BF BG
Applying it once with b = ı̂ı n , c = Bxn , a = G and once with b = ı̂ı n , c = Bxn , a = F gives

3 
ÿ  ÿ3 
BF  BF BG  BG
  
∇ (F ¨ G) = G ˆ ı̂ı n ˆ + (G ¨ ı̂ı n ) + F ˆ ı̂ı n ˆ + (F ¨ ı̂ı n )
Bxn Bxn Bxn Bxn
n =1 n =1
= G ˆ (∇
∇ ˆ F) + (G ¨ ∇ )F + F ˆ (∇
∇ ˆ G) + (F ¨ ∇ )G

Theorem 4.1.5.d:
We use the same trick. Write out the left hand side as
3
ÿ 3
ÿ  ÿ3
B  BF  BG 
∇ ˆ (F ˆ G) = ı̂ı n ˆ (F ˆ G) = ı̂ı n ˆ ˆG + ı̂ı n ˆ F ˆ
Bxn Bxn Bxn
n =1 n =1 n =1

Applying a ˆ (b ˆ c) = (c ¨ a)b ´ (b ¨ a)c, which is Lemma 4.1.8.b below,

3 h
ÿ 3
BF BFn i ÿ h BGn BG i
∇ ˆ (F ˆ G) = Gn ´ G + F ´ Fn
Bxn Bxn Bxn Bxn
n =1 n =1
= (G ¨ ∇ )F ´ (∇
∇ ¨ F)G + (∇
∇ ¨ G)F ´ (F ¨ ∇ )G

145
I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

Theorem 4.1.7.a:
Substituting in
 BF BF2   BF3 BF1   BF BF 
3 2
∇ˆF = ´ ı̂ı ´ ´ ̂ + ´ 1 k̂
By Bz Bx Bz Bx By
gives
B  BF3 BF2  B  BF3 BF1  B  BF2 BF1 
∇¨ ∇ˆF = ´ ´ ´ ´


+
Bx By Bz By Bx Bz Bz Bx By
B 2 F3 B 2 F2 B 2 F3 B 2 F1 B 2 F2 B 2 F1
= ´ ´ + + ´
BxBy BxBz ByBx ByBz BzBx BzBy
=0
because the two red terms have cancelled, the two blue terms have cancelled and the two
black terms have cancelled.

Theorem 4.1.7.b:
Substituting in
Bf Bf Bf
∇f = ı̂ı + ̂ + k̂
Bx By Bz
gives

B B f B B f  B B f B B f  B B f B B f 
∇ˆ ∇f = ´ ı̂ı ´ ´ ̂ + ´ k̂ = 0
By Bz Bz By Bx Bz Bz Bx Bx By By Bx

Theorem 4.1.7.c:
By Theorem 4.1.4.c, followed by Theorem 4.1.4.d,

∇ ¨ f (∇
∇ g ˆ ∇ h) = ∇ f ¨ (∇ ∇ g ˆ ∇ h) + f ∇ ¨ (∇
∇ g ˆ ∇ h)
 

= ∇ f ¨ (∇
∇ g ˆ ∇ h) + f (∇∇ ˆ ∇ g) ¨ ∇ h ´ ∇ g ¨ (∇
∇ ˆ ∇ h)
 

By Theorem 4.1.7.b, ∇ ˆ ∇ g = ∇ ˆ ∇ h = 0, so

∇ ¨ f (∇
∇ g ˆ ∇ f ) = ∇ f ¨ (∇
∇ g ˆ ∇ h)
 

Theorem 4.1.7.d:
By Theorem 4.1.4.c,

∇ ¨ ( f ∇ g ´ g∇ f ) = (∇
∇ f ) ¨ (∇
∇ g) + f ∇ ¨ (∇
∇ g) ´ (∇
∇ g) ¨ (∇
∇ f ) + g ∇ ¨ (∇
∇f)
= f ∇2 g ´ g ∇2 f

Theorem 4.1.7.e:
3
ÿ 3
 ÿ 3
ÿ 3
ÿ
B B
B 2 Fn

∇ ˆ (∇
∇ ˆ F) = ı̂ı ` ˆ ı̂ı m ˆ ı̂ı n Fn = ı̂ı ` ˆ ı̂ı m ˆ ı̂ı n
Bx` Bxm Bx` Bxm
`=1 m =1 n =1 `,m,n=1

146
I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

Using a ˆ (b ˆ c) = (c ¨ a)b ´ (b ¨ a)c, we have

ı̂ı ` ˆ ı̂ı m ˆ ı̂ı n = (ı̂ı ` ¨ ı̂ı n )ı̂ı m ´ (ı̂ı ` ¨ ı̂ı m )ı̂ı n = δ`,nı̂ı m ´ δ`,mı̂ı n

where7
#
1 if m = n
δm,n =
0 if m ‰ n

Hence

3
ÿ 3
ÿ
B 2 Fn B 2 Fn
∇ ˆ (∇
∇ ˆ F) = δ`,nı̂ı m ´ δ`,mı̂ı n
Bx` Bxm Bx` Bxm
`,m,n=1 `,m,n=1
ÿ3 3
ÿ
B BFn B 2 Fn
= ı̂ı m ´ ı̂ı n
Bxm Bxn Bxm 2
m,n=1 m,n=1
2
∇ ¨ F) ´ ∇ F
= ∇ (∇

Lemma 4.1.8.

(a) a ¨ (b ˆ c) = (a ˆ b) ¨ c

(b) a ˆ (b ˆ c) = (c ¨ a)b ´ (b ¨ a)c

Proof. (a) Here are two proofs. For the first, just write out both sides

a ¨ (b ˆ c) = ( a1 , a2 , a3 ) ¨ (b2 c3 ´ b3 c2 , b3 c1 ´ b1 c3 , b1 c2 ´ b2 c1 )
= a1 b2 c3 ´ a1 b3 c2 + a2 b3 c1 ´ a2 b1 c3 + a3 b1 c2 ´ a3 b2 c1
(a ˆ b) ¨ c = ( a2 b3 ´ a3 b2 , a3 b1 ´ a1 b3 , a1 b2 ´ a2 b1 ) ¨ (c1 , c2 , c3 )
= a2 b3 c1 ´ a3 b2 c1 + a3 b1 c2 ´ a1 b3 c2 + a1 b2 c3 ´ a2 b1 c3

and observe that they are the same.

For the second proof, we again write out both sides, but this time we express them in

7 δm,n is called the Kronecker delta function. It is named after the German number theorist and logician
Leopold Kronecker (1823–1891). He is reputed to have said “God made the integers. All else is the
work of man.”

147
I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

terms of determinants.
 
ı̂ı ̂ k̂
a ¨ b ˆ c = ( a1 , a2 , a3 ) ¨ det b1 b2 b3 
c1 c2 c3
     
b2 b3 b1 b3 b b2
= a1 det ´ a2 det + a3 det 1
c2 c3 c1 c3 c1 c2
 
a1 a2 a3
= det b1 b2 b3 

c1 c2 c3
 
ı̂ı ̂ k̂
a ˆ b ¨ c = det  a1 a2 a3  ¨ (c1 , c2 , c3 )
b1 b2 b3
     
a2 a3 a1 a3 a a2
= c1 det ´ c2 det + c3 det 1
b2 b3 b1 b3 b1 b2
 
c1 c2 c3
= det a1 a2 a3 

b1 b2 b3

Exchanging two rows in a determinant changes the sign of the determinant. Moving the
top row of a 3 ˆ 3 determinant to the bottom row requires two exchanges of rows. So the
two 3 ˆ 3 determinants are equal.
(b) The proof is not exceptionally difficult — just write out both sides and grind. Substi-
tuting in
b ˆ c = (b2 c3 ´ b3 c2 )ı̂ı ´ (b1 c3 ´ b3 c1 )̂ + (b1 c2 ´ b2 c1 )k̂
gives, for the left hand side,
 
ı̂ı ̂ k̂
a ˆ (b ˆ c) = det  a1 a2 a3 
b2 c3 ´ b3 c2 ´b1 c3 + b3 c1 b1 c2 ´ b2 c1

= ı̂ı a2 (b1 c2 ´ b2 c1 ) ´ a3 (´b1 c3 + b3 c1 )

 

´̂ a1 (b1 c2 ´ b2 c1 ) ´ a3 (b2 c3 ´ b3 c2 )

 

+k̂ a1 (´b1 c3 + b3 c1 ) ´ a2 (b2 c3 ´ b3 c2 )

 

On the other hand, the right hand side

(a ¨ c)b ´ (a ¨ b)c = ( a1 c1 + a2 c2 + a3 c3 )(b1ı̂ı + b2 ̂ + b3 k̂) ´ ( a1 b1 + a2 b2 + a3 b3 )(c1ı̂ı + c2 ̂ + c3 k̂)

= ı̂ı a1 b1 c1 + a2 b1 c2 + a3 b1 c3 ´ a1 b1 c1 ´ a2 b2 c1 ´ a3 b3 c1
 

+̂ a1 b2 c1 + a2 b2 c2 + a3 b2 c3 ´ a1 b1 c2 ´ a2 b2 c2 ´ a3 b3 c2
 

+k̂ a1 b3 c1 + a2 b3 c2 + a3 b3 c3 ´ a1 b1 c3 ´ a2 b2 c3 ´ a3 b3 c3
 

= ı̂ı [ a2 b1 c2 + a3 b1 c3 ´ a2 b2 c1 ´ a3 b3 c1 ]
+̂ [ a1 b2 c1 + a3 b2 c3 ´ a1 b1 c2 ´ a3 b3 c2 ]
+k̂ [ a1 b3 c1 + a2 b3 c2 ´ a1 b1 c3 ´ a2 b2 c3 ]

148
I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

The last formula that we had for the left hand side is the same as the last formula we had
for the right hand side.

Example 4.1.9 (Screening tests)

We have seen the vector identity Theorem 4.1.7.b before. It says that if a vector field F is
of the form F = ∇ ϕ for some some function ϕ (that is, if F is conservative), then
∇ ˆ F = ∇ ˆ (∇
∇ ϕ) = 0
Conversely, we have also seen, in Theorem 2.4.8, that, if F is defined and has continuous
first order partial derivatives on all of R3 , and if ∇ ˆ F = 0, then F is conservative. The
vector identity Theorem 4.1.7.b is our screening test for conservativeness.
Because its right hand side is zero, the vector identity Theorem 4.1.7.a is suggestive. It
says that if a vector field F is of the form F = ∇ ˆ A for some some vector field A, then
∇ ¨ F = ∇ ¨ (∇
∇ ˆ A) = 0
When F = ∇ ˆ A, A is called a vector potential for F. We shall see in Theorem 4.1.12, below,
that, conversely, if F(x) is defined and has continuous first order partial derivatives on all
of R3 , and if ∇ ¨ F = 0, then F has a vector potential8 . The vector identity Theorem 4.1.7.a
is indeed another screening test.
As an example, consider the Maxwell’s equations
∇¨B = 0
1 BB
∇ˆE+ =0
c Bt
that we saw in Example 4.1.2. The first equation implies that (assuming B is sufficiently
smooth and decaying) there is a vector field A, called the magnetic potential, with B =
∇ ˆ A. Substituting this into the second equation gives
1B  1 BA 
0 =∇ˆE+ ∇ˆA =∇ˆ E+
c Bt c Bt
So E + 1c BA
Bt passes the screening test of Theorem 4.1.7.b and there is a function ϕ (called
the electric potential) with
1 BA
E+ = ´∇ ϕ
c Bt
We have put in the minus sign just to provide compatibility with the usual physics termi-
nology.
Example 4.1.9

Example 4.1.10

Problem: Let r( x, y, z) = x ı̂ı + y ̂ + z k̂ and let ψ( x, y, z) be an arbitrary function. Verify that
∇ ¨ r ˆ ∇ψ = 0

8 Does this remind you of Theorem 2.4.8? It should.

149
 I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

Solution. By the vector identity Theorem 4.1.4.d,

∇ ¨ r ˆ ∇ ψ = (∇ ∇ ˆ r) ¨ ∇ ψ ´ r ¨ ∇ ˆ (∇



∇ ψ)
By the vector identity Theorem 4.1.7.b, the second term is zero. Now since
 Bz By   Bz Bx   By Bx 
∇ˆr = ´ ı̂ı ´ ´ ̂ + ´ k̂ = 0
By Bz Bx Bz Bx By
the first term is also zero. Indeed ∇ ¨ r ˆ ∇ ψ = 0 holds for any curl free r( x, y, z).

Example 4.1.10

4.1.2 §§ Vector Potentials

We’ll now further explore the vector potentials that were introduced in Example 4.1.9.
First, here is the formal definition.

Definition 4.1.11.

The vector field A is said to be a vector potential for the vector field B if

B = ∇ˆA

As we saw in Example 4.1.9, if a vector field B has a vector potential, then the vector
identity Theorem 4.1.7.a implies that ∇ ¨ B = 0. This fact deserves to be called a theorem.

Theorem 4.1.12 (Screening test for vector potentials).

If there exists a vector potential for the vector field B, then

∇¨B = 0

Of course, we’ll consider the converse soon. Also note that the vector potential, when
it exists, is far from unique. Two vector fields A and à are both vector potentials for the
same vector field if and only if
∇ ˆ A = ∇ ˆ Ã ðñ ∇ ˆ (A ´ Ã) = 0
That is, if and only if the difference A ´ Ã passes the conservative field screening test of
Theorems 2.3.9 and 2.4.8. In particular, if A is one vector potential for a vector field B (i.e.
if B = ∇ ˆ A), and if ψ is any function, then
∇ ˆ (A + ∇ ψ) = ∇ ˆ A + ∇ ˆ ∇ ψ = B
by the vector identity Theorem 4.1.7.b. That is, A + ∇ ψ is another vector potential for B.
To simplify computations, we can always choose ψ so that, for example, the third com-
Bψ ş
ponent of A + ∇ ψ, namely A + ∇ ψ ¨ k̂ = A3 + Bz , is zero — just choose ψ = ´ A3 dz.

We have just proven

150
I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

Lemma 4.1.13.

If the vector field B has a vector potential, then, in particular, there is a vector
potential A for B with9 A3 = 0.

Here is an example which exploits this choice to simplify the computations used to find a
vector potential.
Example 4.1.14

Let
B = yz ı̂ı + zx ̂ + xy k̂
This vector field has been set up carefully to obey
B B B
∇¨B = (yz) + (zx ) + ( xy) = 0
Bx By Bz
and so passes the screening test of Theorem 4.1.12.
Let’s try and find a vector potential for B. That is, let’s try and find a vector field
A = A1 ı̂ı + A2 ̂ + A3 k̂ that obeys ∇ ˆ A = B, or equivalently,
BA3 BA2
´ = B1 = yz
By Bz
BA3 BA1
´ + = B2 = zx
Bx Bz
BA2 BA1
´ = B3 = xy
Bx By
This system is nasty to solve because every equation contains more than one of the three
unknowns, A1 , A2 , A3 . Let us take advantage of our observation above that, if any vector
potential exists, then, in particular, a vector potential A exists that also obeys A3 = 0. So
let’s also require that A3 = 0. Then the equations above simplify to
BA2
´ = yz
Bz
BA1
= zx
Bz
BA2 BA1
´ = xy
Bx By
This system is much easier because, now that we have chosen A3 = 0, the first equation
contains only a single unknown, namely A2 and we can find all A2 ’s that obey the first
equation simply by integrating with respect to z:
yz2
A2 = ´ + N ( x, y)
2

9 There is nothing special about the subscript 3 here. By precisely the same argument, we could come up
with another vector potential whose second component is zero, and with a third vector potential whose
first component is zero.

151
I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

Note that, because BBz treats x and y as constants, the constant of integration N is allowed
to depend on x and y.
Similarly, the second equation contains only a single unknown, A1 , and is easily solved
by integrating with respect to z. The second equation is satisfied if and only if

xz2
A1 = + M( x, y)
2
for some function M.
Finally, the third equation is also satisfied if and only if M( x, y) and N ( x, y) obey

B  yz2  B  xz2 
´ + N ( x, y) ´ + M( x, y) = xy
Bx 2 By 2

which simplifies to
BN BM
( x, y) ´ ( x, y) = xy
Bx By
This is one linear equation in two unknowns, M and N. Typically, we can easily solve
one linear equation in one unknown. So we are free to eliminate one of the unknowns by
setting, for example, M = 0, and then choose any N that obeys
BN
( x, y) = xy
Bx
x2 y
Integrating with respect to x gives, as one possible choice, N ( x, y) = 2 . So we have
found a vector potential. Namely

xz2  yz2 x2 y 
A= ı̂ı + ´ + ̂
2 2 2
One can, and indeed should, quickly check that ∇ ˆ A = B.
Example 4.1.14

Let’s do another.
Example 4.1.15

Let
B = (2x ) ı̂ı + (2z ´ 2x ) ̂ + (2x ´ 2z) k̂
This vector field obeys
B B B
∇¨B = (2x ) + (2z ´ 2x ) + (2x ´ 2z) = 0
Bx By Bz

and so passes the screening test of Theorem 4.1.12. We’ll now find a vector potential
A = A1 ı̂ı + A2 ̂ + A3 k̂ for B. As in the last example, we’ll simplify the computations by
further requiring10 that A3 = 0.

10 Of course, we could equally well pick A1 = 0 or A2 = 0.

152
I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

The requirements that ∇ ˆ A = B and A3 = 0 come down to

BA2
´ = 2x
Bz
BA1
= 2z ´ 2x
Bz
BA2 BA1
´ = 2x ´ 2z
Bx By

Because BBz treats x and y as constants, the first equation is satisfied if and only if there is a
function N ( x, y)
A2 = ´2xz + N ( x, y)
and second equation is satisfied if and only if there is a function M( x, y)

A1 = z2 ´ 2xz + M( x, y)

Finally, the third equation is also satisfied if and only if M ( x, y) and N ( x, y) obey

B   B  
´ 2xz + N ( x, y) ´ z2 ´ 2xz + M( x, y)) = 2x ´ 2z
Bx By
BN BM
ðñ ´2z + ( x, y) ´ ( x, y) = 2x ´ 2z
Bx By
BN BM
ðñ ( x, y) ´ ( x, y) = 2x
Bx By

All of the z’s in this equation have cancelled out11 , and we can choose, for example,
M( x, y) = 0 and N ( x, y) = x2 . So we have found a vector potential. Namely

A = (z2 ´ 2xz)ı̂ı + ( x2 ´ 2xz)̂

Again it is a good idea to check that ∇ ˆ A = B.

Example 4.1.15

We can use exactly the strategy of the last examples to prove

Theorem 4.1.16.

Let B be a vector field that is defined and has all of its first order partial deriva-
tives continuous on all of R3 . Then there exists a vector potential for B if and
only if it passes the screening test ∇ ¨ B = 0.

Proof. We already know that the existence of a vector potential implies that ∇ ¨ B = 0. So
we just have to assume that ∇ ¨ B = 0 and prove that this implies the existence of a vector

11 If the z’s had not cancelled out, no N ( x, y) and M ( x, y), which after all are independent of z, could
satisfy the equation. That would have been a sure sign of a user error.

153
I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

field A that obeys ∇ ˆ A = B. Hence we need to solve

BA3 BA2
´ = B1 ( x, y, z)
By Bz
BA3 BA1
´ + = B2 ( x, y, z)
Bx Bz
BA2 BA1
´ = B3 ( x, y, z)
Bx By

We’ll explicitly find such an A using exactly the strategy of Example 4.1.14. In particular,
we’ll look for an A that also has A3 = 0. Then the equations simplify to

BA2
´ = B1 ( x, y, z)
Bz
BA1
= B2 ( x, y, z)
Bz
BA2 BA1
´ = B3 ( x, y, z)
Bx By

The first equation is satisfied if and only if

żz
A2 ( x, y, z) = ´ B1 ( x, y, z̃) dz̃ + N ( x, y)
0

for some function N ( x, y). And the second equation is satisfied if and only if
żz
A1 ( x, y, z) = B2 ( x, y, z̃) dz̃ + M( x, y)
0

So all three equations are satisfied if and only only if we can find M( x, y) and N ( x, y) that
obey

A2 ( x,y,z) A1 ( x,y,z)
hkkkkkkkkkkkkkkkkkkikkkkkkkkkkkkkkkkkkj hkkkkkkkkkkkkkkkkikkkkkkkkkkkkkkkkj
ż z ż
 B  z
B  
´ B1 ( x, y, z̃) dz̃ + N ( x, y) ´ B2 ( x, y, z̃) dz̃ + M ( x, y) = B3 ( x, y, z)
Bx 0 By 0

which is the case if and only if

żz
BN BM BB1 BB 
( x, y) ´ ( x, y) = B3 ( x, y, z) + ( x, y, z̃) + 2 ( x, y, z̃) dz̃
Bx By 0 Bx By

Oof! At first sight, it looks like we have a very big problem here. No matter what N and
M we pick the left hand side will depend on x and y only — not on z. But it appears like
the right hand side depends on z too. Fortunately the screening test (which we have not
used to this point in the proof) rides to the rescue and ensures that the right hand actually
does does not depend on z. By the screening test,

BB1 BB2 BB3

∇¨B = + + =0
Bx By Bz

154
 I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

and we have
BB1 BB2 BB
+ =´ 3
Bx By Bz
so that the right hand side is
żz
BB3  h iz̃=z
B3 ( x, y, z) + ´ ( x, y, z̃) dz̃ = B3 ( x, y, z) + ´ B3 ( x, y, z̃ = B3 ( x, y, 0)
0 Bz z̃=0

by the fundamental theorem of calculus. So we just have to choose M and N to obey

BN BM
( x, y) ´ ( x, y) = B3 ( x, y, 0)
Bx By
şx
For example, M = 0, N ( x, y) = 0 B3 ( x̃, y, 0) dx̃ work. So not only have we proven that a
vector potential exists, but we have found a formula for it.

Warning 4.1.17.

Note that in Theorem 4.1.12 we are assuming that B passes the screening test
on all of R3 . If that is not the case, for example because the vector field is not
defined on all of R3 , then B can fail to have a vector potential. An example (the
point source) is provided in Example 4.4.8.

4.1.3 §§ Interpretation of the Gradient

In this section we’ll develop an interpretation of the gradient ∇ f (r0 ). This should just be
a review of material that you have seen before.
Suppose that
you are moving through space and that your position at time t is r(t) =
x (t), y(t), z(t) . As you move along, you measure, for example, the temperature. If the
temperature at position
( x, y, z) is f ( x, y, z), then the temperature that you measure at time
t is f x (t), y(t), z(t) . So the rate of change of temperature that you feel is

d

f x ( t ), y ( t ), z ( t )
dt
Bf
dx Bf
dy Bf
dz
= x ( t ), y ( t ), z ( t ) (t) + x ( t ), y ( t ), z ( t ) (t) + x ( t ), y ( t ), z ( t ) (t)
Bx dt By dt Bz dt
(by the chain rule)
= ∇ f r(t) ¨ r (t)

1
ˇ
ˇ ˇ ˇ
= ˇ∇ f r(t) ˇ ˇr1 (t)ˇ cos θ


where θ is the angle between the gradient vector ∇ f r(t) and the velocity vector r1 (t).
This is the rate of change per unit time. We ˇcan get
ˇ the rate of change per unit distance
travelled by moving with speed one, so that r (t)ˇ = 1 and then
ˇ 1

d
ˇ
f r(t) = ˇ∇ f r(t) ˇ cos θ

ˇ
dt

155
 I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

If, at a given moment t = t0 , you are at r(t0 ) = r0 , then

d
ˇˇ ˇ ˇ
f r(t) ˇ = ˇ∇ f (r0 )ˇ cos θ
dt t = t0

Recall that θ is the angle between our direction of motion and the gradient vector ∇ f (r0 ).
So to maximize the rate of change of temperature that we feel, as we pass through r0 ,
we should choose our direction of motion to be the direction of the the gradient vector
∇ f (r0 ). In conclusion

Equation 4.1.18.
"
direction of maximum rate of
∇ f (r0 ) has direction =
change of f at r0
"
magnitude of maximum rate of
has magnitude =
change (per unit distance) of f at r0

4.1.4 §§ Interpretation of the Divergence

In this section we’ll develop an interpretation of the divergence ∇ ¨ v(r0 ) of the vector field
v(r) at the point r0 . We shall do so in two steps.

˝ First we’ll express ∇ ¨ v(r0 ) in terms of flux integrals.

˝ Then we’ll use the interpretation of flux integrals given in Lemma 3.4.1 to get an
interpretation of ∇ ¨ v(r0 ).

Think of v( x, y, z) as the velocity of a fluid at ( x, y, z) and fix any point r0 = ( x0 , y0 , z0 ).

Let, for any ε ą 0, Sε be the sphere

˝ centered at r0
˝ of radius ε.
˝ Denote by n̂( x, y, z) the outward normal to Sε at ( x, y, z).

We shall prove, in Lemma 4.1.20, below, that we can write ∇ ¨ v(r0 ) as the limit
ij
1
∇ ¨ v( x0 , y0 , z0 ) = lim 4 3 v( x, y, z) ¨ n̂( x, y, z) dS
εÑ0 πε
3
Sε

Once we have that lemma we can use that

˝ 34 πε3 is the volume of the interior of the sphere Sε and

ť
˝ by Lemma 3.4.1, Sε v( x, y, z) ¨ n̂( x, y, z) dS is the rate12 at which fluid is exiting Sε

to conclude that

12 Lemma 3.4.1 is being applied with the density ρ set equal to one, so, more precisely, the rate is the
number of units of volume of fluid exiting Sε per unit time

156
I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

Equation 4.1.19.
"
rate at which fluid is exiting an infinitesimal sphere
∇ ¨ v ( r0 ) =
centred at r0 , per unit time, per unit volume
= strength of the source at r0

Here is the critical computation.

Lemma 4.1.20.

ij
1
∇ ¨ v( x0 , y0 , z0 ) = lim 4 3 v( x, y, z) ¨ n̂( x, y, z) dS
εÑ0 πε
3
Sε

Proof. (Optional). 13
By translating our coordinate system, it suffices to consider r0 = ( x0 , y0 , z0 ) = (0, 0, 0).
Then
ˇ ( 1
Sε = ( x, y, z) ˇ |( x, y, z)| = ε n̂( x, y, z) = ( x, y, z)
ε
We expand v( x, y, z) in a Taylor expansion in powers of x, y, and z, to first order, with
second order error term.

v( x, y, z) = A + B x + C y + D z + R( x, y, z)

where
Bv Bv Bv
A = v(0, 0, 0) B= (0, 0, 0) C= (0, 0, 0) D= (0, 0, 0)
Bx By Bz

and the error term R( x, y, z) is bounded by a constant times14 x2 + y2 + z2 . In particular

there is a constant K so that, on Sε ,

|R( x, y, z)| ď Kε2

So
ij ij
1
v( x, y, z) ¨ n̂( x, y, z) dS = A + B x + C y + D z + R( x, y, z) ¨ ( x, y, z) dS


ε
Sε Sε

13 There is another, easier to understand, proof of this result given in §4.4.1. We cannot give that proof
here because it uses the divergence theorem, which we will get to later in the chapter.
14 Terms like xy, xz and yz are not needed because, for example, |xy| ď 12 ( x2 + y2 ). This inequality is

2
equivalent to |x| ´ |y| ě 0.

157
I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

Multiply out the dot product so that the integrand becomes

A ¨ ı̂ı x + A ¨ ̂ y + A ¨ k̂ z
+ B ¨ ı̂ı x2 + B ¨ ̂ xy + B ¨ k̂ xz
+ C ¨ ı̂ı xy + C ¨ ̂ y2 + C ¨ k̂ yz
+ D ¨ ı̂ı xz + D ¨ ̂ yz + D ¨ k̂ z2
+ R( x, y, z) ¨ ( x, y, z)
That’s a lot of terms. But most of them integrate to zero, simply because the integral of
an odd function over an even domain is zero. Because Sε is invariant under x Ñ ´x and
under y Ñ ´y and under z Ñ ´z we have
ij ij ij ij ij ij
x dS = y dS = z dS = xy dS = xz dS = yz dS = 0
Sε Sε Sε Sε Sε Sε

which is a relief. We are now left with

ij ij
1
v( x, y, z) ¨ n̂( x, y, z) dS = B ¨ ı̂ı x2 + C ¨ ̂ y2 + D ¨ k̂ z2 dS


ε
Sε Sε
ij
1
+ R( x, y, z) ¨ ( x, y, z) dS
ε
Sε

As well Sε is invariant15 under the interchange of x and y and also under the interchange
of x and z. Consequently
ij ij ij ij
2 2 2 1  2
x + y2 + z2 dS

x dS = y dS = z dS =
3
Sε Sε Sε S
ijε
1
= ε2 dS since x2 + y2 + z2 = ε2 on Sε
3
Sε
4
= πε4
3
since the surface area of the sphere Sε is 4πε2 . So far, we have
ij ij
4 3
1
v( x, y, z) ¨ n̂( x, y, z) dS = πε B ¨ ı̂ı + C ¨ ̂ + D ¨ k̂ + R( x, y, z) ¨ ( x, y, z) dS
3 ε
Sε Sε
ij
4 1
= πε3∇ ¨ v(0) + R( x, y, z) ¨ ( x, y, z) dS (review the definitions of B, C, D)
3 ε
Sε

which implies
ij ij
1 3
lim 4 v( x, y, z) ¨ n̂( x, y, z) dS = ∇ ¨ v(0) + lim R( x, y, z) ¨ ( x, y, z) dS
εÑ0 πε3 εÑ0 4πε4
3
Sε Sε

15 Spheres have lots of symmetry!

158
I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

Finally, it suffices to recall that |R( x, y, z)| ď Kε2 and, on Sε , |( x, y, z)| = ε, so that
ˇij ˇ ij
3 ˇˇ ˇ 3
R( x, y, z) ¨ ( x, y, z) dSˇˇ ď |R( x, y, z)| |( x, y, z)| dS
4πε4 ˇ 4πε4
Sε S
ijε
3 3
ď 4
Kε3 dS = 4
Kε3 4πε2 )
4πε 4πε
Sε
= 3Kε

converges to zero as ε Ñ 0. So we are left with the desired result.

Example 4.1.21

Here is a sketch of the vector field v( x, y, z) = x ı̂ı + y ̂ + z k̂ and a sphere centered on the
origin, like Sε .

This velocity field has fluid being created and pushed out through the sphere. We have

∇ ¨ v(0) = 3

consistent with our interpretation (4.1.19).

Example 4.1.21

Example 4.1.22

Here is a sketch of the vector field v( x, y, z) = ´y ı̂ı + x ̂ and a sphere centered on the ori-
gin, like Sε .

159
 I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

This velocity field just has fluid going around in circles. No fluid actually crosses the
sphere. The divergence

∇ ¨ v(0) = 0

consistent with our interpretation (4.1.19).

Example 4.1.22

Example 4.1.23

Here is a sketch of the vector field v( x, y, z) = ı̂ı and a sphere centered on the origin, like
Sε .

This velocity field just has fluid moving uniformly to the right. Fluid enters the sphere
from the left and leaves through the right at precisely the same rate, so that the net rate at
fluid crosses the sphere is zero. The divergence

∇ ¨ v(0) = 0

again consistent with our interpretation (4.1.19).

Example 4.1.23

4.1.5 §§ Interpretation of the Curl

We’ll now develop the interpretation of the curl, or more precisely, of ∇ ˆ v(r0 ) ¨ n̂ for any
unit vector n̂. As we did in developing the interpretation of divergence, we’ll

˝ first express ∇ ˆ v(r0 ) ¨ n̂ as a limit of integrals, and

˝ then we’ll interpret the integrals.

To specify the integrals involved, let Cε be the circle which

˝ is centered at r0
˝ has radius ε
˝ lies in the plane through r0 perpendicular to n̂

160
I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

˝ is oriented in the standard way with respect to n̂. Imagine standing on the circle
with your feet on the plane through r0 perpendicular to n̂, with the vector from your
feet to your head in the same direction as n̂ and with your left arm point towards r0 .
Then your are facing in the positive direction for Cε .

n̂

ε
Cε

We shall show in Lemma 4.1.25, below, that

¿
1
∇ ˆ v(r0 ) ¨ n̂ = lim 2 v(r) ¨ dr
εÑ0 πε
Cε
Now let’s
ű work on interpreting the right hand side, and in particular on interpreting the
integral Cε v(r) ¨ dr, which is called the circulation of v around Cε . Place a tiny paddle-
wheel in the fluid with its axle running along n̂ and its paddles along Cε , as in the figure
below, except that the paddlewheel is really expensive and has a lot more than just four

n̂

t̂
ε
Cε

paddles. Pretend16 that you are one of the paddles. If the paddlewheel is rotating at Ω
radians per unit time, then you are moving at speed Ωε. If you are at r, the component of
the fluid velocity in your direction of motion, i.e. tangential to Cε , is v(r) ¨ dr dr
ds , because ds ,
with s denoting arc length along the circle, is a unit vector tangential to Cε . All paddles
have to move at the same speed. So the speed of the paddles, Ωε, should be the average
ds around the circle. Thus the rate of rotation, Ω, of the paddlewheel should
value of v(r) ¨ dr
be determined by
ű dr
ű
C v ( r ) ¨ ds v(r) ¨ dr
Ωε = ε ű ds
= Cε
Cε ds 2πε

Consequently, ∇ ˆ v(r0 ) ¨ n̂ is the limit as ε (the radius of the paddlewheel) tends to zero
of
¿
1
v(r) ¨ dr = 2Ω
πε2
Cε
That’s our interpretation.

16 Method acting might help you here.

161
I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

Equation 4.1.24.
If a fluid has velocity field v and you place an infinitesimal paddlewheel at r0
with its axle in direction n, then it rotates at 12 ∇ ˆ v(r0 ) ¨ n̂ radians per unit time.
In particular, to maximize the rate of rotation, orient the paddlewheel so that
n̂ k ∇ ˆ v(r0 ).

There will be some examples at the end of this section. First, we show

Lemma 4.1.25.
¿
1
∇ ˆ v(r0 ) ¨ n̂ = lim 2 v(r) ¨ dr
εÑ0 πε
Cε

Proof. (Optional). 17
Just as we did in the proof of Lemma 4.1.20, we can always translate our coordinate
system so that r0 = ( x0 , y0 , z0 ) = (0, 0, 0). We can also rotate our coordinate system so that
n̂ = k̂. Because r0 = (0, 0, 0) and n̂ = k̂, so that Cε lies in the xy-plane, we can parametrize
Cε by
r(t) = ε cos t ı̂ı + ε sin t ̂
Again as we did in the proof of Lemma 4.1.20, expand v( x, y, z) in a Taylor expansion in
powers of x, y, and z, to first order, with second order error term.
v( x, y, z) = A + B x + C y + D z + R( x, y, z)
where
Bv Bv Bv
A = v(0, 0, 0) B= (0, 0, 0) C= (0, 0, 0) D= (0, 0, 0)
Bx By Bz
and the error term R( x, y, z) is bounded by a constant times x2 + y2 + z2 . In particular
there is a constant K so that, on Cε ,
|R( x, y, z)| ď Kε2
So
¿ ż 2π
v(r) ¨ dr = A + B ε cos t + C ε sin t + R(r(t)) ¨ ´ ε sin t ı̂ı + ε cos t ̂ dt



0
Cε

Again, multiply out the dot product so that the integrand becomes
´ εA ¨ ı̂ı sin t + εA ¨ ̂ cos t
´ ε2 B ¨ ı̂ı sin t cos t + ε2 B ¨ ̂ cos2 t
´ ε2 C ¨ ı̂ı sin2 t
+ ε2 C ¨ ̂ sin t cos t
+ R(r(t)) ¨ ´ ε sin t ı̂ı + ε cos t ̂

17 There is another, easier to understand, proof of this result given in §4.4.1. We cannot give that proof
here because it uses Stokes’ theorem, which we will get to later in the chapter.

162
I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

Again most of these terms integrate to zero, because

ż 2π ż 2π
sin t dt = cos t dt = 0
0 0
ż 2π ż
1 2π
sin t cos t dt = sin(2t) dt = 0
0 2 0

and the sin2 t and cos2 t terms are easily integrated using (see Example 2.4.4)
ż 2π ż 2π ż
2 2 1 2π  2
sin t + cos2 t dt = π

sin t dt = cos t dt =
0 0 2 0
So we are left with
¿ ż 2π
2 2
v(r) ¨ dr = πε B ¨ ̂ ´ πε C ¨ ı̂ı + R(r(t)) ¨ ´ ε sin t ı̂ı + ε cos t ̂ dt


0
Cε

which implies that

¿ ż 2π
1 Bv2 Bv1 1
lim 2 v(r) ¨ dr = (0, 0, 0) ´ (0, 0, 0) + lim 2 R(r(t)) ¨ ´ ε sin t ı̂ı + ε cos t ̂ dt


εÑ0 πε Bx By εÑ0 πε 0
Cε
ż 2π
1
= ∇ ˆ v(0, 0, 0) ¨ k̂ + lim 2 R(r(t)) ¨ ´ ε sin t ı̂ı + ε cos t ̂ dt



εÑ0 πε 0

Finally, it suffices to recall that |R( x, y, z)| ď Kε2 , so that

ˇż ˇ ż 2π
1 ˇˇ 2π
ˇ 1
R(r(t)) ¨ ´ ε sin t ı̂ı + ε cos t ̂ dtˇˇ ď |R(r(t))| dt
πε2 ˇ 0 πε 0
ż
1 2π 2 1
ď Kε dt = Kε2 (2π )
πε 0 πε
= 2Kε

converges to zero as ε Ñ 0.

Here are some examples. We will use the same vector fields as in Examples 4.1.21,
4.1.22 and 4.1.23. In all examples, we shall orient the paddlewheel so that n̂ = k̂ and
sketch the top view, so that the paddlewheel looks like

163
I NTEGRAL T HEOREMS 4.1 G RADIENT, D IVERGENCE AND C URL

Example 4.1.26

Here is a sketch of the vector field v( x, y, z) = x ı̂ı + y ̂ + z k̂ and a circle centered on the
origin, like Cε .

This velocity field has fluid moving parallel to the paddles, so the paddlewheel should
not rotate at all. The computation
 
ı̂ı ̂ k̂
∇ ˆ v(0) = det  BBx ByB B  = 0 ùñ ∇ ˆ v(0) ¨ k̂ = 0

Bz
x y z

is consistent with our interpretation (4.1.24).

Example 4.1.26

Example 4.1.27

Here is a sketch of the vector field v( x, y, z) = ´y ı̂ı + x ̂ and a circle centered on the origin,
like Cε .

This velocity field has fluid going around in circles, counterclockwise. So the paddlewheel
should rotate counterclockwise too. That is, it should have positive angular velocity. Our
interpretation (4.1.24) predicts an angular velocity of half
 
ı̂ı ̂ k̂
∇ ˆ v(0) ¨ k̂ = det  Bx  ¨ k̂ = 2k̂ ¨ k̂ = 2
B B B 
By Bz
´y x 0

164
 I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

which is indeed positive18 .

Example 4.1.27

Example 4.1.28

Here is a sketch of the vector field v( x, y, z) = ı̂ı and a circle centered on the origin, like Cε .

The fluid pushing on the top paddle tries to make the paddlewheel rotate clockwise. The
fluid pushing on the bottom paddle tries to make the paddlewheel rotate counterclock-
wise, at the same rate. So the paddlewheel should not rotate at all. Our interpretation
(4.1.24) predicts an angular velocity of
 
ı̂ı ̂ k̂
1 1
¨ ∇ ˆ v(0) ¨ k̂ = det  BBx BBy BBz  ¨ k̂ = 0 ¨ k̂ = 0
 
2 2
1 0 0
as expected.
Example 4.1.28

4.2IJ The Divergence Theorem

The rest of this chapter concerns three theorems: the divergence theorem, Green’s theo-
rem and Stokes’ theorem. Superficially, they look quite different from each other. But, in
fact, they are all very closely related and all three are generalizations of the fundamental
theorem of calculus żb
df
(t) dt = f (b) ´ f ( a)
a dt
The left hand side of the fundamental theorem of calculus is the integral of the derivative
of a function. The right hand side involves only values of the function on the boundary of
the domain of integration. The divergence theorem, Green’s theorem and Stokes’ theorem
also have this form, but the integrals are in more than one dimension. So the derivatives
are multidimensional, like the curl and divergence, and the integrands can involve vector
fields.

18 Even for small values of 2.

165
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

• For the divergence theorem, the integral on the left hand side is over a (three di-
mensional) volume and the right hand side is an integral over the boundary of the
volume, which is a surface.

• For Green’s and Stokes’ theorems, the integral on the left hand side is over a (two
dimensional) surface and the right hand side is an integral over the boundary of the
surface, which is a curve.

The divergence theorem is going to relate a volume integral over a solid V to a flux
integral over the surface of V. First we need a couple of definitions concerning the allowed
surfaces. In many applications solids, for example cubes, have corners and edges where
the normal vector is not defined. On the other hand, to be able to compute a flux integral
over a surface, we certainly need that the set of points where the normal vector is not
well-defined is small enough that the existence of the flux integral is not jeopardized. This
is the case for “piecewise smooth” surfaces, which we now define.

Definition 4.2.1.

(a) A surface is smooth if it has a parametrization r(u, v) with continuous partial

Br
derivatives Bu Br
and Bv Br
and with Bu ˆ Bv
Br
nonzero.

(b) A surface is piecewise smooth if it consists of a finite number of smooth pieces

that meet along sharp curves and at sharp corners.

Here are sketches of a smooth surface (a sausage) and a piecewise smooth surface (an
ice-cream cone), followed by the divergence theorem19 .

19 It is also known as Gauss’s theorem. Johann Carl Friedrich Gauss (1777–1855) was a German math-
ematician. Throughout the 1990’s Gauss’s portrait appeared on the German ten-mark banknote. In
addition to Gauss’s theorem, the Gaussian distribution (the bell curve), degaussing and the CGS unit
for the magnetic field, and the crater Gauss on the Moon are named in his honour.

166
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

Theorem 4.2.2 (Divergence Theorem).

Let

• V be a bounded solid with a piecewise smooth surface20 BV

• F be a vector field that has continuous first partial derivatives at every point
of V.

Then
ij ¡
F ¨ n̂ dS = ∇ ¨ F dV
BV V

where n̂ is the outward unit normal of BV.

Like the fundamental theorem of calculus, the divergence theorem expresses the integral
of a derivative of a function (in this case a vector-valued function) over a region in terms
of the values of the function on the boundary of the region.

Warning 4.2.3.

Note that in Theorem 4.2.2 we are assuming that the vector field F has continu-
ous first partial derivatives at every point of V. If that is not the case, for example
because F is not defined on all of V, then the conclusion ˇ of the divergence ( the-
orem can fail. An example is F = |r|r 3 , V = ( x, y, z) ˇ x2 + y2 + z2 ď 1 . See
Example 4.2.7.

Proof. We have to show that

ij  ¡ 
 BF1 BF2 BF3 
F1 ı̂ı + F2 ̂ + F3 k̂ ¨ n̂ dS = + + dV
Bx By Bz
BV V

Note that the left hand side is a sum of three terms — one involving F1 , one involving F2
and one involving F3 — and the right hand side is a sum of three terms — one involving
F1 , one involving F2 and one involving F3 . We’ll just show that the F3 terms on the left
hand side and right hand side are equal, i.e. that
ij ¡
BF3
F3 k̂ ¨ n̂ dS = dV
Bz
BV V

Showing that the F1 terms match and the F2 terms match is done in the same way21 .
Special Geometry

20 We are going to consistently use the notation B(thing) to denote the boundary of (thing).
21 Mutatis mutandis.

167
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

We’ll first assume that the solid has the special form
ˇ (
V = ( x, y, z) ˇ B( x, y) ď z ď T ( x, y), ( x, y) P R xy

where R x,y is some subset of the xy-plane. We can further assume that, for each ( x, y) P R xy ,
we have B( x, y) ď T ( x, y). After we’re finished with this special case, we’ll handle the
general case. ť
Let’s work on BV F3 k̂ ¨ n̂ dS first. As in the figure below, the surface BV consists of

z
z “ T px, yq

S
B

z “ Bpx, yq

y
Rxy
x BRxy

three pieces — the top, the bottom and the side. We’ll consider each in turn.
ˇ (
˝ The top is T = ( x, y, z) ˇ z = T ( x, y), ( x, y) P R xy . By (3.3.2), on T

n̂ dS = + ´ Tx ( x, y) ı̂ı ´ Ty ( x, y) ̂ + k̂ dxdy

 

As n̂ is to be the outward normal, it must point upwards on T . That’s why we have

chosen, and emphasised, the “+” sign. So k̂ ¨ n̂ dS = dxdy and
ij ij
F3 k̂ ¨ n̂ dS = F3 ( x, y, T ( x, y)) dxdy
T R xy

ˇ (
˝ The bottom is B = ( x, y, z) ˇ z = B( x, y), ( x, y) P R xy . By (3.3.2), on B

n̂ dS = ´ ´ Bx ( x, y) ı̂ı ´ By ( x, y) ̂ + k̂ dxdy

 

As n̂ is to be the outward normal, it must point downwards on B . That’s why we

have chosen the “´” sign. So k̂ ¨ n̂ dS = ´dxdy and
ij ij
F3 k̂ ¨ n̂ dS = ´ F3 ( x, y, B( x, y)) dxdy
B R xy

168
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM
ˇ (
˝ The side is S = ( x, y, z) ˇ ( x, y) P BR xy , B( x, y) ď z ď T ( x, y) . It runs vertically.
Hence on S the normal vector to BV is parallel to the xy-plane so that k̂ ¨ n̂ = 0 and
ij
F3 k̂ ¨ n̂ dS = 0
S

So all together
ij ij ij ij
F3 k̂ ¨ n̂ dS = F3 k̂ ¨ n̂ dS + F3 k̂ ¨ n̂ dS + F3 k̂ ¨ n̂ dS
BV T B S
ij
F3 ( x, y, T ( x, y)) ´ F3 ( x, y, B( x, y)) dxdy + 0
 
= (BV)
R xy

Now let us examine

¡ ij ż T ( x,y)
BF3 BF3
dV = dxdy dz ( x, y, z)
Bz B( x,y) Bz
V R xy
ij
F3 ( x, y, T ( x, y)) ´ F3 ( x, y, B( x, y)) dxdy
 
= (V)
R xy

by the fundamental theorem of calculus. That’s exactly what we had to show. The inte-
grals (BV) and (V) are equal.
General Geometry
Now we’ll drop the assumption on V that we imposed in the “Special Geometry”
section above. The key idea that makes the proof work is that we can cut up any22 V into
pieces, each of which does obey the special assumption that we just considered. Consider,
for example, the sausage shaped solid in the figure on the left below.

V1
S1
Sc
S2
V2

Call the sausage V. Cut it into two halves by running a cleaver horizontally through its
centre. This splits the solid V into two halves, V1 and V2 as in the figure on the right above.
It also splits the boundary BV of V into two halves S1 and S2 , also as in the figure on the
right above. Note that
˝ the boundary, BV1 , of V1 is the union of S1 and the shaded disk Sc (the cut introduced
by the cleaver). On the cut Sc , the outward pointing normal to V1 is ´k̂.

22 We are assuming that V is “reasonable”.

169
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

˝ The boundary, BV2 , of V2 is the union of S2 and the shaded disk Sc . On the cut Sc ,
the outward pointing normal to V2 is +k̂.

Now both V1 and V2 do satisfy the assumption of the “Special Geometry” section above.
So
¡ ¡ ¡
BF3 BF3 BF3
dV = dV + dV
Bz Bz Bz
V V V2
ij1 ij
= F3 k̂ ¨ n̂ dS + F3 k̂ ¨ n̂ dS
BV1 BV2
ij ij ij ij
= F3 k̂ ¨ n̂ dS + F3 k̂ ¨ n̂ dS + F3 k̂ ¨ n̂ dS + F3 k̂ ¨ n̂ dS
S1 S2 n̂
Sc Sc
n̂

The second and fourth integrals are identical except that n̂ = ´k̂ in the second integral
and n̂ = +k̂ in the fourth integral. So they cancel exactly and
¡ ij ij ij
BF3
dV = F3 k̂ ¨ n̂ dS + F3 k̂ ¨ n̂ dS = F3 k̂ ¨ n̂ dS
Bz
V S1 S2 BV

as desired.

Example 4.2.4
ť
Problem: Evaluate the flux integral S F ¨ n̂ dS where n̂ is the outward normal to S, which
is the surface of the hemispherical region n̂

ˇ ( V
V= ( x, y, z ˇ x2 + y2 + z2 ď a2 , z ě 0 S

n̂
and
F = xz2 ı̂ı + ( x2 y ´ z3 ) ̂ + 2xy + y2 z + ecos y k̂

Solution. The ecos y in F suggests that a direct evaluation of the integral is difficult. So
we’ll use a little trickery to to evaluate it. Not surprisingly, considering that we have just
proven the divergence theorem, the trick is to apply the divergence theorem23 . Since

BF1 BF2 BF3 B
B
B

∇¨F = xz2 + x 2 y ´ z3 + 2xy + y2 z + ecos y


+ + =
Bx By Bz Bx By Bz
= z2 + x 2 + y2

23 It’s almost as though someone rigged the example with this in mind.

170
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

The divergence theorem tell us that

ij ¡
F ¨ n̂ dS = x2 + y2 + z2 dV

S V

Spherical coordinates are perfect for this integral. (See Appendix F.3, if you need to refresh
your memory.)
¡ ż 2π ż π/2 ża
2 2 2
dρ ρ2 sin ϕ ρ2


x +y +z dV = dθ dϕ
0 0 0
V
 ż 2π  ż π/2  ż a 
4
= dθ sin ϕ dϕ ρ dρ
0 0 0
 h iπ/2 h ρ5 i a
= 2π ´ cos ϕ
0 5 0
2πa 5
=
5

Example 4.2.4

Example 4.2.5
ť
Problem: Evaluate the flux integral S F ¨ n̂ dS where n̂ is the outward normal to S, which
is the part of the surface z2 = x2 + y2 with 1 ď z ď 2, and where

F = 3x ı̂ı + (5y + ecos x ) ̂ + z k̂

Solution. Again the ecos x in F suggests that a direct evaluation is difficult24 and again we’ll
apply the divergence theorem. But this time S is not the boundary of a solid V. It is the
portion of the cone outlined in red in the figure on the left below and does not have a top
or bottom “cap”. Fortunately, there is a solid V whose boundary, while not being equal

D2 n̂ “ k̂

S S V
n̂ n̂
D1
n̂ “ ´k̂

24 In fact, it is possible to evaluate this integral directly, if one recognizes that the ugly part of the integrand
is odd under y Ñ ´y and integrates to exactly zero.

171
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

to S, at least contains S. It is (unsurprisingly)

ˇ (
V = ( x, y, z) ˇ x2 + y2 ď z2 , 1 ď z ď 2
and is sketched in the figure on the right above. The boundary, BV, is the union of S and
the two disks
ˇ (
D1 = ( x, y, z) ˇ x2 + y2 ď z2 , z = 1
ˇ (
D2 = ( x, y, z) ˇ x2 + y2 ď z2 , z = 2
So the divergence theorem gives
¡ ij ij ij ij
∇ ¨ F dV = F ¨ n̂ dS = F ¨ n̂ dS + F ¨ n̂ dS + F ¨ n̂ dS
V BV S D1 D2

which implies
ij ¡ ij ij
F ¨ n̂ dS = ∇ ¨ F dV ´ F ¨ n̂ dS ´ F ¨ n̂ dS
S V D1 D2

The point of this exercise is that the left hand side, which is not easy to evaluate directly,
is the integral we want, while the three integrals on the right hand side are all easy to
evaluate. We do so now. The outward normal to (the horizontal disk) D2 is +k̂. So
ij ij ij
F ¨ n̂ dS = F ¨ k̂ dS = z dS
D2 D2 D2

As z = 2 on D2 , and D2 is a disk of radius 2,

ij
F ¨ n̂ dS = 2Area( D2 ) = 2π22 = 8π
D1

Similarly, the outward normal to (the horizontal disk) D1 is ´k̂. So

ij ij ij
F ¨ n̂ dS = ´ F ¨ k̂ dS = ´ z dS
D1 D1 D1

As z = 1 on D1 , and D1 is a disk of radius 1,

ij
F ¨ n̂ dS = Area( D1 ) = ´π12 = ´π
D1

Finally, as ∇ ¨ F = 3 + 5 + 1 = 9
¡
∇ ¨ F dV = 9 Vol(V )
V

The volume of V can be easily computed using the first year technique25 of slicing V into
thin horizontal pancakes like that sketched in the figure below.

25 You can review in §1.6 of the CLP-2 text.

172
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

z“2

z“1

The pancake at height z has

˝ thickness dz,
˝ a circular cross-section of radius z (remember that the outer boundary of V has equa-
tion x2 + y2 = z2 ), and hence has
˝ cross-sectional area πz2 and
˝ volume πz2 dz.
So
¡ ż2 h πz3 i2
2 7
∇ ¨ F dV = 9 Vol(V ) = 9 πz dz = 9 = 9 ˆ π = 21π
1 3 1 3
V

and, all together

ij ¡ ij ij
F ¨ n̂ dS = ∇ ¨ F dV ´ F ¨ n̂ dS ´ F ¨ n̂ dS = 21π ´ (´π ) ´ 8π = 14π
S V D1 D2

Example 4.2.5

Example 4.2.6
ť
Problem: Evaluate the flux integral S F ¨ n̂ dS where n̂ is the upward normal to S, which

2
is the part of z = x2 + y2 with 0 ď z ď 1, and
2

F = x + ey ı̂ı + (y + cos z) ̂ + k̂

Solution. This integral can be evaluated in much the same way as we evaluated the integral
of Example 4.2.6. We first define a solid V whose boundary BV contains S. A good, and
hopefully obvious, choice is
ˇ
2 (
V = ( x, y, z) ˇ x2 + y2 ď z, 0 ď z ď 1
The boundary of V is the union of S, with outward pointing normal ´n (recall that the
problem specifies that the symbol n̂ refers to the upward point normal) and the disk
ˇ
2 (
D = ( x, y, z) ˇ z = 1, x2 + y2 ď 1

with outward pointing normal k̂.

173
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

k̂
D
z“1

n̂ S

So the divergence theorem gives

¡ ij ij
∇ ¨ F dV = ´ F ¨ n̂ dS + F ¨ k̂ dS
V S D
which implies
ij ¡ ij
F ¨ n̂ dS = ´ ∇ ¨ F dV + F ¨ k̂ dS
S V D
¡ ij
=´ 2 dV + dS
V D
D is a circular disk of radius 1, and so has area π. To evaluate the volume integral we slice
V into horizontal pancakes with the pancake at height z having a circular cross-section of

2
radius z1/4 . (Recall that the boundary of V has x2 + y2 = z.) So
ij ż1
? 2 π
F ¨ n̂ dS = ´2 π z dz + π = ´2π ˆ + π = ´
0 3 3
S
Again, you can see that the actual integration is quite easy. All of the work (or at least all
of the thinking) happens in the setup.
Example 4.2.6

Example 4.2.7

In Warning 4.2.3 we emphasised that the conclusion of the divergence Theorem 4.2.2 can
fail if the vector field F is not defined at even a single point of V. Here is an example. Set
r
F= 3 where r = x ı̂ı + y ̂ + z k̂
|r|
ˇ (
and V = ( x, y, z) ˇ x2 + y2 + z2 ď 1 . Then, if ( x, y, z) ‰ 0,
B x B y B z
∇ ¨ F( x, y, z) = + +
Bx x2 + y2 + z2 3/2 By x2 + y2 + z2 3/2 Bz x2 + y2 + z2 3/2
    

x + y2 + z2 ´ x 23 (2x ) x + y2 + z2 ´ y 32 (2y) x + y2 + z2 ´ z 32 (2z)

 2   2   2 
=  5/2 +  5/2 +  5/2
x 2 + y2 + z2 x 2 + y2 + z2 x 2 + y2 + z2
=0

174
 I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM
ˇ (
On the other hand, the boundary of V is the unit sphere BV = ( x, y, z) ˇ x2 + y2 + z2 = 1 .
r
The outward unit normal to BV is n̂ = |r| so that
ż ż ż ż
r r 1
F ¨ n̂ dS = ¨ dS = dS = dS
BV |r|=1 |r|3 |r| |r|=1 |r|2 |r|=1
= 4π ‰ 0

Example 4.2.7

4.2.1 §§ Optional — An Application of the Divergence Theorem — the Heat

Equation
§§§ Derivation of the Heat Equation
Let T ( x, y, z, t) be the temperature at time t at the point ( x, y, z) in some object B . The heat
equation26 is the partial differential equation that describes the flow of heat energy and
consequently the behaviour of T. We now use the divergence theorem to derive the heat
equation from two physical “laws”, that we assume are valid:

˝ The amount of heat energy required to raise the temperature of an object by ∆T

degrees is CM ∆T where, M is the mass of the object and C is a positive physical
constant determined by the material contained in the object. It is called the specific
heat, or specific heat capacity27 , of the object.
˝ Think of heat energy as a moving fluid. We will rig its velocity field so that heat
flows in the direction opposite to the temperature gradient. Precisely, we choose
its velocity field to be ´κ∇ T ( x, y, z, t). Here κ is another positive physical constant
called the thermal conductivity of the object. So the rate at which heat is conducted
across an element of surface area dS at ( x, y, z) in the direction of its unit normal n̂
is given by ´κ n̂ ¨ ∇ T ( x, y, z, t) dS at time t. For example, in the figure

n̂ cold

dS

∇T hot

26 The heat equation was formulated by the French mathematician and physicist Jean-Baptiste Joseph
Fourier in 1807. He lived from 1768 to 1830, a period which included both the French revolution and
the reign of Napoleon. Indeed Fourier served on his local Revolutionary Committee, was imprisoned
briefly during the Terror, and was Napoleon Bonaparte’s scientific advisor on his Egyptian expedition
of 1798. Fourier series and the Fourier transform are named after him. Fourier is also credited with
discovering the greenhouse effect.
27 Heat is now understood to arise from the internal energy of the object. In an earlier theory, heat was
viewed as measuring an invisible fluid, called the caloric. The amount of caloric that an object could
hold was called its “heat capacity” by the Scottish physician and chemist Joseph Black (1728–1799).

175
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

the temperature gradient, which points in the direction of increasing temperature, is

opposite n̂. Consequently the flow rate ´κ n̂ ¨ ∇ T ( x, y, z, t) dS is positive, indicating
flow in the direction of n̂. This is just what you would expect — heat flows from
hot regions to cold regions. Also the rate of flow increases as the magnitude of
the temperature gradient increases. This also makes sense (and is reminiscent of
Newton’s law of cooling).
Let V Ă B be any three dimensional region in the object and denote by BV the surface of V
and by n̂ the outward normal to BV. The amount of heat that enters V across an infinites-
imal piece dS of BV in an infinitesimal time interval dt is ´ ´ κ n̂ ¨ ∇ T ( x, y, z, t) dS dt.

The amount of heat that enters V across all of BV in the time interval dt is given by the
integral
n̂
dS
BV
ij
κ n̂ ¨ ∇ T ( x, y, z, t) dS dt V
BV

In this same time interval the temperature at a point ( x, y, z) in V changes by BT Bt ( x, y, z, t ) dt.

If the density of the object at ( x, y, z) is ρ( x, y, z), the amount of heat energy required to
increase the temperature of an infinitesimal volume dV of the object centred at ( x, y, z) by
BT BT
Bt ( x, y, z, t ) dt is C ( ρdV ) Bt ( x, y, z, t ) dt. The amount of heat energy required to increase
the temperature by BT Bt ( x, y, z, t ) dt at all points ( x, y, z ) in V is then
¡
BT
Cρ ( x, y, z, t) dV dt
Bt
V

Assuming that the object is not generating or destroying28 heat itself, this must be same
as the amount of heat that entered V in the time interval dt. That is
ij ¡
BT
κ n̂ ¨ ∇ T dS dt = Cρ dV dt
Bt
BV V

Now we cancel the common factor of dt. We can then rewrite the left hand side as an
integral over V by applying the divergence theorem giving
¡ ¡
BT
κ∇ ¨ ∇ T dV = Cρ dV
Bt
V V

As both integrals are over the same volume V, we have

¡ ¡ ¡ h
BT BT i
κ∇ ¨ ∇ T dV ´ Cρ dV = 0 ùñ κ∇ 2 T ´ Cρ dV = 0 (H)
Bt Bt
V V V

28 The caloric theory of heat was itself destroyed by the cannon boring experiment of 1798. In this experi-
ment the American/British physicist Benjamin Thompson (1753–1814) boiled water just using the heat
generated by friction during the boring of a cannon.

176
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

2 2 2
where ∇ 2 = ∇ ¨ ∇ = BBx2 + BBy2 + BBz2 is the Laplacian. This must be true for all volumes V
in the object and for all times t. We claim that this forces

BT
κ∇ 2 T ( x, y, z, t) ´ Cρ ( x, y, z, t) = 0
Bt
for all ( x, y, z) in the object and all t.
Suppose that to the contrary there was a point ( x0 , y0 , z0 ) in the object and a time t0
with, for example, κ∇ 2 T ( x0 , y0 , z0 , t0 ) ´ Cρ BT Bt ( x0 , y0 , z0 , t0 ) ą 0. By continuity, which we
are assuming, κ∇ T ( x, y, z, t0 ) ´ Cρ Bt ( x, y, z, t0 ) must remain close to κ∇ 2 T ( x0 , y0 , z0 , t0 ) ´
2 BT

Cρ BT
Bt ( x0 , y0 , z0 , t0 ) when ( x, y, z ) is close to ( x0 , y0 , z0 ). So we would have

BT
κ∇ 2 T ( x, y, z, t0 ) ´ Cρ ( x, y, z, t0 ) ą 0
Bt
for all ( x, y, z) in some small ball B centered on ( x0 , y0 , z0 ). Then, necessarily,
¡ h
BT i
κ∇ ¨ ∇ T ( x, y, z, t0 ) ´ Cρ ( x, y, z, t0 ) dV ą 0
Bt
B

which violates (H) for V = B. This completes our derivation of the heat equation, which
is

Equation 4.2.8.
BT
( x, y, z, t) = α∇ 2 T ( x, y, z, t)
Bt
κ
where α = Cρ is called the thermal diffusivity.

§§§ An Application of the Heat Equation

As an application, we look at the temperature a short distance below the surface of the
Earth. For simplicity, we make the Earth flat29 and we assume that the temperature, T,
depends only on time, t, and the vertical coordinate, z. Then the heat equation simplifies
to
BT B2 T
(z, t) = α 2 (z, t) (HE)
Bt Bz
We choose a coordinate system having the surface of the Earth at z = 0 and having z
increase downward. We also assume that the temperature T (0, t) at the surface of the
Earth is primarily determined by solar heating and is given by

T (0, t) = T0 + TA cos(σt) + TD cos(δt) (IC)

Here T0 is the long term average of the temperature at the surface of the Earth, TA cos(σt)
gives seasonal temperature variations and TD cos(δt) gives daily temperature variations.

29 Insert sarcastic footnote here.

177
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

air
z “ 0, T p0, tq “ T0 ` TA cospσtq ` TD cospδtq
earth
Tt “ αTzz
z

We measure time in days so that δ = 2π and σ = 1 2π 2π

year = 365days . Then TA cos( σt )
has period one year and TD cos(δt) has period one day. The solution to the initial value
problem (HE)+(IC) can be found by separation of variables, a standard topic in courses on
partial differential equations. The solution is
c b c
?σ  σ  δ  δ 
´ z
T (z, t) = T0 + TA e´ 2α z cos σt ´ z + TD e 2α cos δt ´ z (SLN)
2α 2α
Whether or not you can find this solution, you can, and should, check that (SLN) satisfies
both (HE) and (IC).
Now let’s see what we can learn from the solution (SLN). For any fixed z, the time
average of T (z, t) is T0 (just because the average value if cosine is zero), the same as the
average temperature at the surface z = 0. That is, under the hypotheses that we have
made, the long term average temperature at any depth z is is the same as the long term
average temperature ? at the surface.
b 
σ
z
The term TA e´ 2α cos σt ´ σ
2α z

˝ oscillates in time with a period of one year, just like TA cos(σt)

?σ
˝ has an amplitude TA e´ 2α z which is TA at the surface ? and decreases exponentially
as z increases. Increasing the depth z by a distance 2α/σ causes the amplitude of
the oscillation to decrease by a factor of 1/e. Both of these first two bullet points are
probably very consistent with your intuition. But this term also has a third property
that you may find less obvious. It has
z
˝ has a time lag of ?2ασ with respect to TA cos(σt). The surface term TA cos(σt) takes
? σ valuewhen?t = 0, /σ, /σ, ¨ ¨ ¨ . At depth z, the corresponding
its maximum 2π 4π
?
term TA e´ 2α z cos σt ´ σ/2α z takes its maximum value when σt ´ σ/2α z =
? ? ?
0, 2π, 4π, ¨ ¨ ¨ so that t = z/ 2ασ, 2π/σ + z/ 2ασ, 4π/σ + z/ 2ασ, ¨¨¨.
b  b 
δ
´ z
Similarly, the term TD e 2α cos δt ´ δ
2α z

˝ oscillates in time with a period of one day, just like TD cos(δt)

˝ has an amplitude
? which is TD at the surface and decreases by a factor of 1/e for each
increase of 2α/δ in depth.
˝ has a time lag of ?z with respect to TD cos(δt).
2αδ
For water α is approximately 0.012 m2 /day. This α gives
c c
2α 2α z z
« 1.2 m « 0.062 m ? « 49 z days ? « 2.6 z days
σ δ 2ασ 2αδ

178
 I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

for z measured in centimeters. So at a depth of a couple of meters, the temperature is

pretty constant in time. What variation there is lags the surface variations by several
months.

4.2.2 §§ Variations of the Divergence Theorem

Here are a couple useful variations of the divergence theorem.

Theorem 4.2.9 (Variations on the divergence theorem).

If V is a solid with surface BV, then

ij ¡
F ¨ n̂ dS = ∇ ¨ F dV
BV V
ij ¡
f n̂ dS = ∇ f dV
V
ij BV ¡
n̂ ˆ F dS = ∇ ˆ F dV
BV V

where n̂ is the outward unit normal of BV.

Memory Aid. All three formulae can be combined into

ij ¡
n̂ ˚ F̃ dS = ∇ ˚ F̃ dV
BV V

where ˚ can be either ¨, ˆ or nothing. When ˚ = ¨ or ˚ = ˆ, then F̃ = F. When ˚ is nothing,

F̃ = f .
Proof. The first formula is exactly the divergence theorem and was proven in Theorem 4.2.2.
To prove the second formula, set F = f a, where a is any constant vector, and apply the
divergence theorem.
ij ¡
f a ¨ n̂ dS = ∇ ¨ ( f a) dV
BV V
¡
∇ f ) ¨ a + f lo ¨ aon dV
 
= (∇ ∇
omo (by the vector identity Theorem 4.1.4.c)
V =0
¡
= ∇ f ) ¨ a dV
(∇
V

To get the third line, we just used that a is a constant, so that all of its derivatives are zero.
Rewrite ¡ ¡
∇ f ) ¨ a dV =
(∇ a ¨ (∇
∇ f ) dV
V V

179
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

Since a is a constant, we can factor it out of both integrals, so

ij ¡
a¨ f n̂ dS = a ¨ ∇ f dV
BV V
"ij ¡ *
ùñ a ¨ f n̂ dS ´ ∇ f dV = 0
BV V

In
ť particular,ţ choosing a = ı̂ı , ̂ and k̂, we see that all three components of the vector
BV f n̂ dS ´ V ∇ f dV are zero. So
ij ¡
f n̂ dS ´ ∇ f dV = 0
BV V

which is what we wanted show.

To prove the third formula, apply the divergence theorem, but with F replaced by a ˆ F,
where a is any constant vector.
ij ¡
(a ˆ F) ¨ n̂ dS = ∇ ¨ (a ˆ F) dV
BV V
¡
F ¨ looomooon
∇ ˆ a) ´a ¨ (∇
∇ ˆ F) dV
 
= (∇ (by the vector identity Theorem 4.1.4.d)
V =0
¡ ¡
=´ a ¨ (∇
∇ ˆ F) dV = ´a ¨ ∇ ˆ F dV
V V

To get the third line, we again used that a is a constant, so that all of its derivatives are
zero. For all vectors (a ˆ b) ¨ c = a ¨ (b ˆ c) (in case you don’t remember this, it was
Lemma 4.1.8.a) so that
(a ˆ F) ¨ n̂ = a ¨ (F ˆ n̂)
and
ij ¡
a¨ F ˆ n dS = ´a ¨ ∇ ˆ F dV
BV V
"ij ¡ *
ùñ a ¨ F ˆ n dS + ∇ ˆ F dV = 0
BV V

In
ť particular, choosing
ţ a = ı̂ı , ̂ and k̂, we see that all three components of the vector
BV F ˆ n dS + V ∇ ˆ F dV are zero. So
¡ ij ij
∇ ˆ F dV = ´ F ˆ n dS = n̂ ˆ F dS
V BV BV

which is what we wanted show.

180
 I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

4.2.3 §§ An Application of the Divergence Theorem — Buoyancy

In this section, we use the divergence theorem to show that when you immerse an object
in a fluid the net effect of fluid pressure acting on the surface of the object is a vertical
force (called the buoyant force) whose magnitude equals the weight of fluid displaced by
the object. This is known as Archimedes’ principle30 .
We shall also show that the buoyant force acts through the “centre of buoyancy” which
is the centre of mass of the fluid displaced by the object. The design of self-righting31 boats
exploits the fact that the centre of buoyancy and the centre of gravity, where gravity acts,
need not be the same.
We start by computing the total force due to the pressure of the fluid pushing on the
object. Recall that pressure
˝ is the force per unit surface area that the fluid exerts on the object
˝ acts perpendicularly to the surface
˝ pushes on the object
Thus the force due to pressure that acts on an infinitesimal piece of the object’s surface at
r = ( x, y, z) with surface area dS and outward normal n̂ is ´p(r) n̂dS. The minus sign
is there because pressure is directed into the object. If the object fills the volume V and
has surface BV, then the total force on the object due to fluid pressure, called the buoyant
force, is ij
B=´ p(r) n̂ dS
BV
ţ
We now wish to apply a variant of the divergence theorem to rewrite B = ´ V ∇ p dV.
But there is a problem with this: p(r) is the fluid pressure at r and is only defined where
there is fluid. In particular, there is no fluid32 inside the object, so p(r) is not defined for
any r in the interior of V.
So we pretend that we remove the object from the fluid and we call P(r) the fluid
pressure at r when there is no object in the fluid. We also make the assumption that at any
point r outside of the object, the pressure at r does not depend on whether the object is in
the fluid or not. In other words, we assume that
#
P(r) if r is not in V
p(r) =
not defined if r is in the V

This assumption is only an approximation to reality, but, in practice, it is a very good

approximation. So, by Theorem 4.2.9,
ij ij ¡
B=´ p(r) n̂ dS = ´ P(r) n̂ dS = ´ ∇ P(r) dV (4.2.1)
BV BV V

Our next job is to compute ∇ P. Concentrate on an infinitesimal cube of fluid whose

edges are parallel to the coordinate axes. Call the lengths of the edges dx, dy and dz and

30 The interested reader should do a net search for the story of Archimedes and the golden crown.
31 The first design of a self-righting boat was entered by William Wouldhave in a lifeboat design compe-
tition organised by South Shield’s Law House committee in 1789.
32 A cup of tea in the galley doesn’t count.

181
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

the position of the centre of the cube ( x, y, z). The forces applied to the various faces of the
cube by the pressure of fluid outside the cube are illustrated in the figure

´P px, y, z ` dz
2
q dxdy k̂

dy dy
P px, y ´ 2
, zq dxdz ̂ ´P px, y ` 2
, zq dxdz ̂

P px, y, z ´ dz
2
q dxdy k̂

The total force due to the pressure acting on the cube is the sum

dx dx
´P x+ , y, z dydz ı̂ı + P x ´



, y, z dydz ı̂ı
2 2
dy
dy

´ P x, y + , z dxdz ̂ + P x, y ´ , z dxdz ̂
2 2
dz dz

´ P x, y, z + dxdy k̂ + P x, y, z ´


dxdy k̂
2 2
of the forces acting on the six faces. Consider the ı̂ı component and rewrite it as

dx dx
´P x+ , y, z dydz ı̂ı + P x ´



, y, z dydz ı̂ı
2 2
P( x + 2 , y, z) ´ P( x ´ dx
dx
2 , y, z ) ı
=´ ı̂ dxdydz
dx
BP
= ´ ( x, y, z) ı̂ı dxdydz
Bx
Doing this for the other components as well, we see that the total force due to the pressure
acting on the cube is
! BP BP BP )
´ ( x, y, z) ı̂ı + ( x, y, z) ̂ + ( x, y, z) k̂ dxdydz = ´∇ P( x, y, z) dxdydz
Bx By Bz

We shall assume that the only other force acting on the cube is gravity and that the fluid
is stationary (or at least not accelerating). Hence the total force acting on the cube is zero.
If the fluid has density ρf , then the cube has mass ρf dxdydz so that the force of gravity is
´gρf dxdydz k̂. The vanishing of the total force now tells us that

´∇ P(r) dxdydz ´ gρf dxdydz k̂ = 0 ùñ ∇ P(r) = ´gρf k̂

Subbing this into (4.2.1) gives

¡
B = g k̂ ρf dV = gM f k̂
V

182
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

ţ
where M f = V ρf dV is the mass of the fluid displaced by the object — not the mass of
the object itself. Thus the buoyant force acts straight up and has magnitude equal to gM f ,
which is also the magnitude of the force of gravity acting on the fluid displaced by the
object. In other words, it is the weight of the displaced fluid. This is exactly Archimedes’
principle.
We next consider the rotational motion of our submerged object. The physical law
that determines the rotational motion of a rigid body about a point r0 is analogous to the
familiar Newton’s law, m dv dt = F, that determines the translational motion of the object.
For the rotational law of motion,

˝ the mass m is replaced by a physical quantity, characteristic of the object, called the
moment of inertia, and
˝ the ordinary velocity v is replaced by the angular velocity, which is a vector whose
length is the rate of rotation (i.e. angle rotated per unit time) and whose direction is
parallel to the axis of rotation (with the sign determined by a right hand rule), and
˝ the force F is replaced by a vector called the torque about r0 . A force F applied at
r = ( x, y, z) produces the torque33 (r ´ r0 ) ˆ F about r0 .

This is derived in the optional §4.2.4, entitled “Torque”, and is all that we need to know
about rotational motion of rigid bodies in this discussion.
Fix any point r0 . The total torque about r0 produced by force of pressure acting on the
surface of the submerged object is
ij ij
(r ´ r0 ) ˆ ´ p(r)n̂ dS = n̂ ˆ P(r) (r ´ r0 ) dS



T=
BV BV

Recall that in these integrals r = ( x, y, z) is the position of the infinitesimal piece dS of the
surface S. Applying the cross product variant of the divergence theorem in Theorem 4.2.9,
followed by the vector identity Theorem 4.1.5.c, gives
¡ ¡
(
∇ ˆ P(r) (r ´ r0 ) dV = ∇ P ( r ) ˆ ( r ´ r0 ) + P ( r ) ∇ ˆ (r ´ r0 ) dV


T= loooooomoooooon
V V =0
¡
= ∇ P(r) ˆ (r ´ r0 ) dV
V

since ∇ ˆ r0 = 0, because r0 is a constant, and

ı̂ı
 
̂ k̂
∇ ˆ r = det  Bx
B B
By
B 
Bz =0
x y z

33 This is what Archimedes was referring to when he said “Give me a lever and a place to stand and I will
move the earth.”

183
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

We have already found that ∇ P(r) = ´gρf k̂. Substituting it in gives

¡ ¡
T=´ gρf k̂ ˆ (r ´ r0 ) dV = ´gk̂ ˆ ρf (r ´ r0 ) dV
V V
"¡ ¡ * "¡ *"ţ *
V rρf dV
= ´gk̂ ˆ rρf dV ´ r0 ρf dV = ´g ρf dV k̂ ˆ ţ ´ r0
V ρf dV
V V V
"ţ * "ţ *
V rρf dV V rρf dV
= ´B ˆ ţ ´ r0 = ţ ´ r0 ˆ B
V ρf dV V ρf dV

So the torque generated at r0 by pressure over the entire surface is the same as a force B
all applied at the single point ţ
rρf dV
CB = ţV
V ρf dV
This point is called the centre of buoyancy. It is the centre of mass of the displaced fluid.
The moral of the above discussion is that the buoyant force, B, on a rigid body

˝ acts straight upward,

˝ has magnitude equal to the weight of the displaced fluid and
˝ acts at the centre of buoyancy, which is the centre of mass of the displaced fluid.

As above, denoting by ρb the density of the object, the torque about r0 due to gravity acting
on the object is

¡ "ţ * "¡ *
 
rρb dV
(r ´ r0 ) ˆ (´gρb k̂) dV = ţV ´ r0 ˆ ´g ρb dV k̂
V bρ dV
V V

So the gravitational force, G,

˝ acts straight down, ţ

˝ has magnitude equal to the weight gMb = g V rρb dV (where ρb is the density of
the object) of the object and ţ
rρb dV
˝ acts at the centre of mass, CG = ţV , of the object.
V ρb dV

Because the mass distribution of the object need not be the same as the mass distribution of
the displaced fluid, buoyancy and gravity may act at two different points. This is exploited
in the design of self-righting boats.
These boats are constructed with a heavy, often lead (which is cheap and dense), keel.
As a result, the centre of gravity is lower in the boat than the center of buoyancy, which,
because the displaced fluid has constant density, is at the geometric centre of the boat. As
the figure below illustrates, a right side up configuration of such a boat is stable, while an
upside down configuration is unstable. The boat rotates so as to keep the centre of gravity
straight below the centre of buoyancy. To see this pretend that you are holding on to the
boat with one hand holding the centre of buoyancy and the other hand holding the centre
of gravity. Use your hands to apply forces in the directions of the arrows and think about
how the boat will respond.

184
 I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

G G

4.2.4 §§ Optional — Torque

In this section, we derive the properties of torque that we used in the last example. New-
ton’s law of motion says that the position r(t) of a single particle moving under the in-
fluence of a force F obeys mr2 (t) = F. Similarly, the positions ri (t), 1 ď i ď n, of a set of
particles moving under the influence of forces Fi obey mr2i (t) = Fi , 1 ď i ď n. Very often
systems of interest consist of some small number of rigid bodies. Suppose that we are
interested in the motion of a single rigid body, say a piece of wood. The piece of wood is
made up of a huge number34 of atoms. So the system of equations determining the motion
of all of the individual atoms in the piece of wood is huge. On the other hand, we shall
see that because the piece of wood is rigid, its configuration is completely determined by
the position of, for example, its centre of mass and its orientation (we won’t get into what
precisely is meant by “orientation”, but it is certainly determined by, for example, the po-
sitions of a few of the corners of the piece of wood). To be precise, we shall extract from the
huge system of equations that determine the motion of all of the individual atoms, a small
system of equations that determine the motion of the centre of mass and the orientation.
We’ll do so now.
Imagine a piece of wood moving in R3 . Furthermore, imagine that the piece of wood

consists of a huge number of particles joined by a huge number of weightless but very
strong35 steel rods. The steel rod joining particle number one to particle number two just
represents a force acting between particles number one and two. Suppose that

˝ there are n particles, with particle number i having mass mi ,

˝ at time t, particle number i has position ri (t),

34 Just 12 grams of carbon contains about 6 ˆ 1023 atoms.

35 Mathematicians and their idealizations! Really the rods just represent the atomic/chemical forces that
hold the wood together.

185
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

˝ at time t, the external force (gravity and the like) acting on particle number i is Fi (t),
and
˝ at time t, the force acting on particle number i, due to the steel rod joining parti-
cle number i to particle number j is Fi,j (t). If there is no steel rod joining particles
number i and j, just set Fi,j (t) = 0. In particular, Fi,i (t) = 0.

The only assumptions that we shall make about the steel rod forces are

(A1) for each i ‰ j, Fi,j (t) = ´F j,i (t). In words, the steel rod joining particles i and j
applies equal and opposite forces to particles i and j.
(A2) for each i ‰ j, there is a function Mi,j (t) such that Fi,j (t) = Mi,j (t) ri (t) ´ r j (t) .
 

In words, the force due to the rod joining particles i and j acts parallel to the line
joining particles i and j. For (A1) to be true, that is to have Mi,j (t) ri (t) ´ r j (t) =


´M j,i (t) r j (t) ´ ri (t) , we need Mi,j (t) = M j,i (t).

 

Newton’s law of motion, applied to particle number i, now tells us that

n
ÿ
mi r2i (t) = Fi (t) + Fi,j (t) (Ni )
j =1

Adding up all of the equations (Ni ), for i = 1, 2, 3, ¨ ¨ ¨ , n gives

n
ÿ n
ÿ ÿ
mi r2i (t) = Fi ( t ) + Fi,j (t) (ΣNi )
i =1 i =1 1ďi,jďn
ř
The sum Fi,j (t) contains F1,2 (t) exactly once and it also contains F2,1 (t) exactly
1ďi,jďn
once
ř and these two terms cancel exactly, by assumption (A1). In this way, all terms in
Fi,j (t) with i ‰ j exactly cancel. All terms with i = j are assumed to be zero. So
1ďi,jďn
ř
Fi,j (t) = 0 and the equation (ΣNi ) simplifies to
1ďi,jďn

n
ÿ n
ÿ
mi r2i (t) = Fi ( t ) (ΣNi )
i =1 i =1

ř
n
1 ř
n
Phew! Denote by M = mi the total mass of the body, by R(t) = M mi ri (t) the centre
i =1 i =1
ř
n
of mass36 of the body and by F(t) = Fi (t) the total external force acting on the system.
i =1
In this notation, equation (ΣNi ) can be written as

Equation 4.2.10.
MR2 (t) = F(t)

36 Note that this is just the weighted average (no pun intended) of the positions of the particles.

186
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

The upshot is that the centre of mass of the system moves just like a single particle of mass
M subject to the total external force. This is why we can often replace an extended object
by a point mass at its centre of mass.
Now take the cross product of ri (t) and equation (Ni ) and sum over i. This gives
n
ÿ n
ÿ ÿ
mi ri (t) ˆ r2i (t) = ri ( t ) ˆ Fi ( t ) + ri (t) ˆ Fi,j (t) (Σri ˆ Ni )
i =1 i =1 1ďi,jďn

By the assumption (A2)

r1 (t) ˆ F1,2 (t) = M1,2 (t) r1 (t) ˆ r1 (t) ´ r2 (t)

 

r2 (t) ˆ F2,1 (t) = M2,1 (t) r2 (t) ˆ r2 (t) ´ r1 (t)

 

= ´M1,2 (t) r2 (t) ˆ r1 (t) ´ r2 (t)

 

so that

r1 (t) ˆ F1,2 (t) + r2 (t) ˆ F2,1 (t) = M1,2 (t) r1 (t) ´ r2 (t) ˆ r1 (t) ´ r2 (t) = 0
   

because the cross product of any two parallel vectors is zero.

ř
The last equation says that the i = 1, j = 2 term in ri (t) ˆ Fi,j (t) exactly cancels
1ďi,jďn
ř
the i = 2, j = 1 term. In this way all of the terms in ri (t) ˆ Fi,j (t) with i ‰ j cancel.
1ďi,jďn
ř
Each term with i = j is exactly zero because Fii = 0. So ri (t) ˆ Fi,j (t) = 0 and
1ďi,jďn
(Σri ˆ Ni ) simplifies to
n
ÿ n
ÿ
mi ri (t) ˆ r2i (t) = ri ( t ) ˆ Fi ( t ) (Σri ˆ Ni )
i =1 i =1

At this point it makes sense to define vectors

n
ÿ
L(t) = mi ri (t) ˆ r1i (t)
i =1
ÿ n
T(t) = ri ( t ) ˆ Fi ( t )
i =1

because, in this notation, (Σri ˆ Ni ) becomes

Equation 4.2.11.
d
L(t) = T(t)
dt

Equation (4.2.11) plays the rôle of Newton’s law of motion for rotational motion. T(t)
is called the torque and plays the rôle of “rotational force”. L(t) is called the angular
momentum (about the origin) and is a measure of the rate at which the piece of wood

187
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

is rotating. For example, if a particle of mass m is traveling in a circle of radius ρ in the

xy-plane at ω radians per unit time, then r(t) = ρ cos(ωt)ı̂ı + ρ sin(ωt)̂ and
mr(t) ˆ r1 (t) = m ρ cos(ωt)ı̂ı + ρ sin(ωt)̂ ˆ ´ ωρ sin(ωt)ı̂ı + ωρ cos(ωt)̂
   

= mρ2 ω k̂
is proportional to ω, which is the rate of rotation about the origin and is in the direction
k̂, which is normal to the plane containing the circle.
In any event, in order for the piece of wood to remain stationary, equations (4.2.10) and
(4.2.11) force F(t) = T(t) = 0.
Now suppose that the piece of wood is a seesaw37 that is supported on a fulcrum at
p. The forces consist of gravity, ´mi gk̂, acting on particle number i, for each 1 ď i ď n,
and the force Φ imposed by the fulcrum that is pushing up on the particle at p. The total

´m1 g k̂ ´m2 g k̂ ´m3 g k̂ ´m4 g k̂

ř
n
external force is F = Φ ´ mi gk̂ = Φ ´ Mgk̂. If the seesaw is to remain stationary, this
i =1
must be zero so that Φ = Mgk̂.
The total torque (about the origin) is
n
ÿ  n
ÿ 
T = pˆΦ´ mi gri ˆ k̂ = g Mp ´ mi ri ˆ k̂
i =1 i =1
If the seesaw is to remain stationary, this must also be zero. This will be the case if the
fulcrum is placed at
n
1 ÿ
p= mi ri
M
i =1
which is just the centre of mass of the piece of wood.
More generally, suppose that the external forces acting on the piece of wood consist of
Fi , acting on particle number i, for each 1 ď i ď n, and a “fulcrum force” Φ acting on a
ř
n
particle at p. The total external force is F = Φ + Fi . If the piece of wood is to remain
i =1
ř
n
stationary, this must be zero so that Φ = ´ Fi . The total torque (about the origin) is
i =1
n
ÿ n
ÿ
T = pˆΦ+ ri ˆ Fi = ( ri ´ p ) ˆ Fi
i =1 i =1

37 Or teeter-totter for those who speak a different English dialect.

188
 I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

If the piece of wood is to remain stationary, this must also be zero. That is, the torque
about point p due to all of the forces Fi , 1 ď i ď n, must be zero.

4.2.5 §§ Optional — Solving Poisson’s Equation

In this section we shall use the divergence theorem find a formula for the solution of
Poisson’s equation
∇ 2 ϕ = 4πρ
Here ρ = ρ(r) is a given (continuous) function and ϕ is the unknown function that we
wish to find. This equation arises, for example, in electrostatics, where ρ is the charge
density and ϕ is the electric potential.
The main step in finding this solution formula will be to consider an

arbitrary (smooth) function ϕ and an

arbitrary (smooth) region V in R3 and an
arbitrary point r0 in the interior of V

and to find an auxiliary formula which expresses ϕ(r0 ) in terms of

∇ 2 ϕ(r), with r running over V and

∇ ϕ(r) and ϕ(r), with r running only over BV.

This auxiliary formula, which we shall derive below, is

"¡ ij ij *
1 ∇ 2 ϕ(r) 3 r ´ r0 ∇ ϕ(r)
ϕ ( r0 ) = ´ d r´ ϕ(r) ¨ n̂ dS ´ ¨ n̂ dS (V)
4π |r ´ r0 | |r ´ r0 |3 |r ´ r0 |
V BV BV

When we take the limit as V expands to fill all of R3 then, assuming that ϕ and ∇ ϕ go to
zero sufficiently quickly38 at 8, the two integrals over BV will converge to zero and we
will end up with the formula
¡
1 ∇ 2 ϕ(r) 3
ϕ ( r0 ) = ´ d r
4π |r ´ r0 |
R3

This expresses ϕ evaluated at an arbitrary point, r0 , of R3 in terms of ∇ 2 ϕ(r), with r

running over R3 , which is exactly what we want, since ∇ 2 ϕ = 4πρ for any solution of
Poisson’s equation. So once we have proven (V) we will have proven39

38 Suppose, for example, that, for large |r ´ r0 |, |ϕ(r)| is bounded by a constant times 1/|r´r0 | and |∇ ϕ(r)|
is bounded by a constant times 1/|r´r0 |2 . Then, if BV is the sphere of radius R centred on r0 , BV has
surface area 4πR2 and the two integrals over BV are bounded by a constant times 1/R.
39 Note that the theorem does not claim that the ϕ defined in the theorem obeys ∇ 2 ϕ = 4πρ. It does, but
the proof is beyond our scope.

189
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

Theorem 4.2.12.

Assume that ρ(r) is continuous and decays sufficiently quickly as r Ñ 8. If ϕ

obeys ∇ 2 ϕ = 4πρ on R3 , and ϕ and ∇ ϕ decay sufficiently quickly as r Ñ 8,
then ¡
ρ(r) 3
ϕ ( r0 ) = ´ d r
|r ´ r0 |
R3

for all r0 in R3 .

Let

r( x, y, z) = x ı̂ı + y ̂ + z k̂
r0 = x0 ı̂ı + y0 ̂ + z0 k̂
1
We shall exploit three properties of the function |r´r0 | . The first two properties are

1 r ´ r0
∇ =´ (P1)
|r ´ r0 | |r ´ r0 |3
1 r ´ r0
∇2 = ´∇ ¨ =0 (P2)
|r ´ r0 | |r ´ r0 |3
and are valid for all r ‰ r0 . Verification of the first property is a simple one line computa-
tion. Verification of the second property is a simple three line computation.
1
The other property of |r´r that we shall use is the following. Let Sε be the sphere of
0|
radius ε centered on r0 . Then, for any continuous function ψ(r),
ij ij ij
ψ(r) 1 ψ ( r0 ) ψ ( r0 )
lim dS = lim p ψ(r) dS = lim dS = lim 4πε2
εÑ0+ |r ´ r0 | p εÑ0+ ε εÑ0+ ε p εÑ0+ ε p
Sε Sε Sε
$
’
&4πψ(r0 ) if p = 2
= 0 if p ă 2 (P3)
’
%
undefined if p ą 2

Derivation of (V):
Here is the derivation of (V). Let Vε be the part of V outside of Sε . Note that the

Sε V

r0 Vε

190
I NTEGRAL T HEOREMS 4.2 T HE D IVERGENCE T HEOREM

boundary BVε of Vε consists of two parts — the boundary BV of V and the sphere Sε —
r´r0
and that the unit outward normal to BVε on Sε is ´ |r´r , because it points towards r0 and
0|
hence outside of Vε .
Recall the vector identity Theorem 4.1.7.d, which says

∇ ¨ ( f ∇ g ´ g∇ f ) = f ∇ 2 g ´ g ∇ 2 f
1
Applying this identity with f = |r´r0 | and g = ϕ gives

=0 by (P2)
hkkkkikkkkj
 1 1  ∇2 ϕ 1
∇¨ ∇ ϕ ´ ϕ∇ = ´ ϕ ∇2
|r ´ r0 | |r ´ r0 | |r ´ r0 | |r ´ r0 |
∇2 ϕ
=
|r ´ r0 |
which is the integrand of the first integral on the right hand side of (V). So, by the diver-
gence theorem
¡ ¡
∇2 ϕ  1 1 
dV = ∇¨ ∇ ϕ ´ ϕ∇ dV
|r ´ r0 | |r ´ r0 | |r ´ r0 |
Vε V
ijε 
1 1 
= ∇ ϕ ´ ϕ∇ ¨ n̂ dS
|r ´ r0 | |r ´ r0 |
BV
ij 
1 1   r ´ r0 
+ ∇ ϕ ´ ϕ∇ ¨ ´ dS (M)
|r ´ r0 | |r ´ r0 | |r ´ r0 |
Sε

To see the connection between (M) and the rest of (V), note that,

˝ by (P1), the first term on the right hand side of (M) is

ij  ij ij
1 1  ∇ ϕ(r) r ´ r0
∇ ϕ ´ ϕ∇ ¨ n̂ dS = ¨ n̂ dS + ϕ(r) ¨ n̂ dS
|r ´ r0 | |r ´ r0 | |r ´ r0 | |r ´ r0 |3
BV BV BV
(R1)
which is 4π times the second and third terms on the right hand side of (V),
1 r´r0
˝ and substituting in ∇ |r´r = ´ |r´r 3 , from (P1), and applying (P3) with p = 2, the
0| 0|
limit of the second term on the right hand side of (M) is
ij 
1 1   r ´ r0 
lim ∇ ϕ ´ ϕ∇ ¨ ´ dS
εÑ0+ |r ´ r0 | |r ´ r0 | |r ´ r0 |
Sε
ij
1
= ´ lim ∇ ϕ ¨ ( r ´ r0 ) + ϕ
 
dS
εÑ0+ |r ´ r0 |2
Bε
h i
= ´4π ∇ ϕ ¨ (r ´ r0 ) + ϕ
r = r0
= ´4πϕ(r0 ) (R2)

191
 I NTEGRAL T HEOREMS 4.3 G REEN ’ S T HEOREM

So applying40 limεÑ0+ to (M) and substituting in (R1) and (R2) gives

¡ ij ij
∇2 ϕ ∇ ϕ(r) r ´ r0
dV = ¨ n̂ dS + ϕ(r) ¨ n̂ dS ´ 4πϕ(r0 )
|r ´ r0 | |r ´ r0 | |r ´ r0 |3
V BV BV

which is exactly equation (V).

4.3IJ Green’s Theorem

Our next variant of the fundamental theorem of calculus is Green’s41 theorem, which
relates an integral, of a derivative of a (vector-valued) function, over a region in the xy-
plane, with an integral of the function over the curve bounding the region. First we need
to define some properties of curves.

Definition 4.3.1.

(a) A curve C with parametrization r(t), a ď t ď b, is said to be closed if r( a) =

r ( b ).

(b) A curve C is said to be simple if it does not cross itself. More precisely, if r(t),
a ď t ď b, is a parametrization of the curve and if a ď t1 , t2 ď b obey t1 ‰ t2
and tt1 , t2 u ‰ ta, bu, then r(t1 ) ‰ r(t2 ). That is, if r(t1 ) = r(t2 ), then either
t1 = t2 or t1 = a, t2 = b, or t1 = b, t2 = a.

(c) A curve C is piecewise smooth if it has a parametrization r(t) which

• is continuous and which

• is differentiable except possibly at finitely many points with
• the derivative being continuous and nonzero except possibly at finitely
many points.

Here are sketches of some examples.

∇2 ϕ ţ ∇2 ϕ
40 You might worry about the singularity in |r´r | when applying limεÑ0+ to Vε |r´r | dV. That this
0 0
singularity is harmless may be seen using spherical coordinates centred on r0 . Then dV contains a
factor of |r ´ r0 |2 (see §F.3), which completely eliminates the singularity.
41 George Green (1793–1841) was a British mathematical-physicist. He spent much of the early part of
his life working in his father’s bakery and grain mill. He was finally admitted as an undergraduate to
Cambridge in 1832, aged nearly forty.

192
I NTEGRAL T HEOREMS 4.3 G REEN ’ S T HEOREM

simple curve simple closed curve not a simple piecewise smooth

curve curve

And here is Green’s theorem.

Theorem 4.3.2 (Green’s Theorem).

Let

• R be a finite region in the xy-plane,

• the boundary, C, of R consist of a finite number of piecewise smooth, simple
closed curves
– that are oriented (i.e. arrows are put on C) consistently with R in the
sense that if you walk along C in the direction of the arrows, then R is
on your left

y y
C
R C
R
x
C
x

• F1 ( x, y) and F2 ( x, y) have continuous first partial derivatives at every point

of R.

Then ¿ ij 
BF2 BF1 
´
 
F1 ( x, y) dx + F2 ( x, y) dy = dxdy
Bx By
C R

Warning 4.3.3.

Note that in Theorem 4.3.2 we are assuming that F1 and F2 have continuous first
partial derivatives at every point of R. If that is not the case, for example because
F1 or F2 is not defined on all of R, then the conclusion of Green’s
ˇ 2 theorem
( can
y x ˇ 2
fail. An example is F1 = ´ x2 +y2 , F2 = x2 +y2 , R = ( x, y) x + y ď 1 . See
Examples 4.3.7 and 4.3.8.

193
I NTEGRAL T HEOREMS 4.3 G REEN ’ S T HEOREM

Here are three notational remarks before we start the proof.

˝ One way to remember the integrand on the right hand side is to write it as (∇
∇ ˆ F) ¨ k̂.
˝ Many people use M instead of F1 andN insteadof F2 . Then Green’s theorem becomes
ű   ť BN BM
C M ( x, y ) dx + N ( x, y ) dy = R Bx ´ By dxdy
ű ş
˝ The symbol C is just an alternate notation for C that is sometimes used when C is a
closed curve. See Notation 2.4.1.
Proof. We prove the result by reformulating it as a divergence theorem statement. To that
end, we define
ˇ (
V = ( x, y, z) ˇ ( x, y) P R, 0 ď z ď 1
G( x, y, z) = F2 ( x, y) ı̂ı ´ F1 ( x, y) ̂

Notice that V is exactly the volume obtained by expanding R vertically upward by one
unit.

z
z“1

V
y
R
x C

The definition of G does not contain a typo — the x-component of G really is F2 and the
y-component of G really is ´F1 . (More or less the reverse of what you would normally
write down.)
These definitions have been rigged so that the divergence theorem applied to G and
V, namely
ij ¡
G ¨ n̂ dS = ∇ ¨ G dV
BV V

gives us exactly Green’s theorem.

BF1
Since ∇ ¨ G = BF
Bx ´ By , the right hand side is just
2

¡ ij ż1
∇ ¨ G dV = dxdy dz ∇ ¨ G
0
V R
ij ż1  BF
2 BF1 
= dxdy dz ( x, y) ´ ( x, y)
0 Bx By
R
ij  BF
2 BF1 
= dxdy ( x, y) ´ ( x, y)
Bx By
R

194
I NTEGRAL T HEOREMS 4.3 G REEN ’ S T HEOREM

because the integrand is independent of z. This is exactly the right hand side of Green’s
theorem.
Now for the left hand side. The boundary, BV, of V is the union of the (flat) bottom,
the (flat) top and the (curved) side. The outward unit normal on the (horizontal, flat) top
is +k̂ and the outward unit normal on the (horizontal, flat) bottom is ´k̂ so that
ij ij ij ij
G ¨ n̂ dS = G ¨ k̂ dS + G ¨ (´k̂) dS + G ¨ n̂ dS
BV top bottom side
ij
= G ¨ n̂ dS
side

We have used the fact that the k̂ component of G is exactly zero to discard the integrals
over the top and bottom of BV. To evaluate the integral over the side, we’ll parametrize
the side. Suppose that r(t) = x (t)ı̂ı + y(t)̂ , a ď t ď b, is a parametrization of C, with the
arrow in the figure above giving the direction of increasing t. Then we can use

R(t, z) = r(t) + zk̂ = x (t)ı̂ı + y(t)̂ + zk̂ a ď t ď b, 0 ď z ď 1

as a parametrization of the side. We’ll use (3.3.1) to determine n̂ dS for the side. Since

BR
(t, z) = x1 (t)ı̂ı + y1 (t)̂
Bt
BR
(t, z) = k̂
Bz
gives

BR BR
n̂ dS = (t, z) ˆ (t, z) dtdz
Bt Bz

= x1 (t)ı̂ı + y1 (t)̂ ˆ k̂ dtdz
= ´ x1 (t)̂ + y1 (t)ı̂ı dtdz

Note that with this choice of ˘ sign (that is, BR

Bt ˆ Bz dtdz rather than ´ Bt ˆ Bz dtdz), the
BR BR BR

vector n̂ really is the outward pointing normal, as we see from the sketch

z
z“1

k̂ V
y
R
x C
BR
n̂
Bt
“ r1

195
I NTEGRAL T HEOREMS 4.3 G REEN ’ S T HEOREM

We can now compute the surface integral directly.

ij ij
G ¨ n̂ dS = G ¨ n̂ dS
BV side
żb ż1
dz G R(t, z) ¨ ´ x1 (t)̂ + y1 (t)ı̂ı



= dt
a 0
żb ż1
dz F2 ( x (t), y(t)) ı̂ı ´ F1 ( x (t), y(t)) ̂ ¨ ´ x1 (t)̂ + y1 (t)ı̂ı



= dt
a 0
żb
dt F2 ( x (t), y(t)) y1 (t) + F1 ( x (t), y(t)) x1 (t)
 
=
a
since the integrand is independent of z
¿
 
= F1 ( x, y) dx + F2 ( x, y) dy
C

This is exactly the left hand side of Green’s theorem.

Example 4.3.4

Problem: Evaluate ¿
( x ´ xy) dx + (y3 + 1) dy


C
where C is the curve given in the figure

y p1, 1q p2, 1q

R C

p1, 0q p2, 0q x

ˇ (
Solution. Let R = ( x, y) ˇ 1 ď x ď 2, 0 ď y ď 1 . By Green’s theorem
¿ ij h
3 B B i
( x ´ xy) dx + (y + 1) dy = (y3 + 1) ´ ( x ´ xy) dxdy

Bx By
C R
ż2 ż1 ˇ2
x2 ˇˇ 3
= dx dy x = ˇ =
1 0 2 1 2

Example 4.3.4

Here is a simple corollary of Green’s theorem that tells how to compute the area en-
closed by a curve in the xy-plane.

196
I NTEGRAL T HEOREMS 4.3 G REEN ’ S T HEOREM

Corollary 4.3.5.

Let

• R be a finite region in the xy-plane whose boundary

• C consists of a finite number of piecewise smooth, simple closed curves.
• Orient C (i.e. put arrows on C) so that if you walk along C in the direction
of the arrows, then R is on your left.

Then ¿ ¿ ¿
1
xdy = ´ xdy ´ ydx
 
Area( R) = ydx =
2
C C C

Proof. This is just Green’s theorem applied first with F = x ̂ , then with F = ´y ı̂ı and
finally with F = 12 ´ y ı̂ı + x ̂ . For all three of these F’s,


BF2 BF1
´ =1
Bx By

so that Green’s theorem gives

¿ ij  ij
BF2 BF1 
´
 
F1 ( x, y) dx + F2 ( x, y) dy = dxdy = dxdy = Area( R)
Bx By
C R R

Example 4.3.6

In this example we will use Green’s theorem to compute the area enclosed by the astroid
x2/3 + y2/3 = a2/3 .

In Example 1.1.7 we found the parametrization

r(t) = x (t) ı̂ı + y(t) ̂ = a cos3 tı̂ı + a sin3 t̂ 0 ď t ď 2π

197
I NTEGRAL T HEOREMS 4.3 G REEN ’ S T HEOREM

for the astroid. So, by Corollary 4.3.5,

¿ ż
1  1 2π 
xdy ´ ydx = x (t)y1 (t) ´ y(t) x1 (t) dt
 
Area =
2 2 0
C
ż 2π
3a2
cos3 t sin2 t cos t + sin3 t cos2 t sin t dt
 
=
2 0
ż
3a2 2π
cos2 t sin2 t cos2 t + sin2 t dt
 
=
2 0
ż
3a2 2π
= cos2 t sin2 t dt
2 0
ż ż
3a2 2π 2 3a2 2π
= sin (2t) dt = [1 ´ cos(4t)] dt
8 0 16 0
3 2
= a π
8

Example 4.3.6

Example 4.3.7 (Trick Question)

Problem: Evaluate ¿
B ¨ dr
C
where
´yı̂ı + x̂
B=
x 2 + y2
and C is the curve

x (t) = sin(cos t)
y(t) = sin(sin t)
z(t) = 0

with 0 ď t ď 2π.
Solution. First let’s think about the curve C. If the curve were just X (t) = cos t, Y (t) = sin t,
Z (t) = 0, it would be the unit circle centred on the origin in the xy-plane, traversed
counterclockwise. For ´ π2 ď u ď π2 , the function sin u increases monotonically with u and
is of the same sign as u so that, since | sin t| ď 1 ă π2 ,
˝ x (t) = sin cos t) has the same sign as X (t) = cos t and is increasing at precisely the
same t’s as is X (t)
˝ y(t) = sin sin t) has the same sign as Y (t) = sin t and is increasing at precisely the
same t’s as is Y (t)
So the extra sine in our parametrization of C just distorts the circle, straightening the sides
a little as depicted here.

198
I NTEGRAL T HEOREMS 4.3 G REEN ’ S T HEOREM

It looks like our problem is a straightforward Green’s theorem problem like Example 4.3.4.
Let’s just try using the strategy of Example 4.3.4. Because
BB2 BB1 B x B ´y
´ = ´
Bx By Bx x + y
2 2 By x + y2
2

1 2x2 1 2y2
= 2 ´ + ´
x + y2 ( x 2 + y2 )2 x 2 + y2 ( x 2 + y2 )2
( x2 + y2 ) ´ 2x2 + ( x2 + y2 ) ´ 2y2
=
( x 2 + y2 )2
=0
it looks like Green’s theorem gives us, trivially,
¿ ¿ ij 
BB2 BB1 
B ¨ dr = ´
 
B1 dx + B2 dy = dxdy = 0
Bx By
C C R
where R is the region inside our curve C.
That was easy — but it’s also very wrong! Our next steps are to
ű
• verify that C B ¨ dr ‰ 0, and
• explain why we got the wrong answer, and
• modify our computation so as to give the correct answer. We’ll do this in Example
4.3.8.
ű
Verification that C B ¨ dr ‰ 0:
Since
x1 (t) = ´ cos(cos t) sin t
y1 (t) = cos(sin t) cos t
z1 ( t ) = 0
our integral is
¿ ¿ ż 2π
B ¨ dr = B1 x (t), y(t) x1 (t) + B2 x (t), y(t) y1 (t) dt
  


B1 dx + B2 dy =
0
C C
ż 2π
sin(sin t) cos(cos t) sin t + sin(cos t) cos(sin t) cos t
= dt
0 sin2 (cos t) + sin2 (sin t)
This is a very ugly looking integral42 . But even if we can’t evaluate the integral, we can

42 Indeed!

199
I NTEGRAL T HEOREMS 4.3 G REEN ’ S T HEOREM

ű
see that the integrand is strictly positive, and that forces C B ¨ r ą 0. Because
π
0 ď | sin t|, | cos t| ď 1 ă
2
˝ cos(cos t) ą 0, and sin(sin t) has the same sign as sin t, and sin(sin t) is zero if and
only if sin t = 0. So the first term in the numerator,
cos(cos t) sin(sin t) sin t ě 0
and is zero if and only if sin t = 0
˝ cos(sin t) ą 0, and sin(cos t) has the same sign as cos t, and sin(cos t) is zero if and
only if cos t = 0. So the second term in the numerator,
cos(sin t) sin(cos t) cos t ě 0
and is zero if and only if cos t = 0.
˝ There is no t for which both sin t and cos t are simultaneously zero. So the whole
numerator
sin(sin t) cos(cos t) sin t + sin(cos t) cos(sin t) cos t ą 0
is strictly positive.
Since the integrand is strictly positive, the integral is strictly positive.
Why we got the wrong answer:
BB1
In our initial and wrong calculation above, we assumed that BB Bx ( x, y ) ´ By ( x, y ) = 0 at
2

all points ( x, y) of the region R inside C. That’s not true. While it is true for most points,
it is not true for all points. The vector field B( x, y) is not defined at ( x, y) = (0, 0). So
BB2 BB1
Bx ( x, y ) ´ By ( x, y ) is also not defined at ( x, y ) = (0, 0). That’s enough to invalidate
Green’s theorem. Read the statement of Theorem 4.3.2 again carefully.
Example 4.3.7

Example 4.3.8 (Example 4.3.7, again.)

Problem: Evaluate ¿
B ¨ dr
C
where
´yı̂ı + x̂
B=
x 2 + y2
and C is the curve
x (t) = sin(cos t)
y(t) = sin(sin t)
z(t) = 0
with 0 ď t ď 2π.
Solution. This is the same integral
ű that we computed incorrectly in Example 4.3.7. We’ll
use two ingredients to compute C B ¨ dr correctly.

200
I NTEGRAL T HEOREMS 4.3 G REEN ’ S T HEOREM

• Let a ą 0 and denote by Ca the clockwise oriented circle in the xy-plane

ű that is of
radius a and is centered on the origin. We can explicitly compute Ca B ¨ dr. To do so
just parametrize Ca by x (t) = a cos t, y(t) = a sin t, z(t) = 0. Then x1 (t) = ´a sin t,
y(t) = a cos t and
¿ ż 2π h ż 2π
´a sin t ı̂ı + a cos t ̂ i 
B ¨ dr = ¨ ´

a sin t ı
ı̂ + a cos t 
̂ dt = dt = 2π
0 a2 cos2 t + a2 sin2 t 0
Ca

• Pick an a that is small enough that Ca lies entirely inside C and apply Green’s theo-
rem with the region, R a , that is between C and Ca . The curve bounding R a has two

C
Ra
´Ca

components — C and Ca , but now Ca is oriented clockwise. (Recall that, in Green’s

theorem, when you walk along a boundary curve in the direction of the arrow, R a
BB1
has to be on your left.). Use ´Ca to denote Ca oriented clockwise. BB Bx ( x, y ) ´ By ( x, y )
2

really is zero at all points ( x, y) of the region R a . So Green’s theorem gives

ij  ¿ ¿
BB2 BB1 
0= ´ dxdy = B ¨ dr + B ¨ dr
Bx By
R C ´Ca
¿a ¿
= B ¨ dr ´ B ¨ dr
C Ca

and so
¿ ¿
B ¨ dr = B ¨ dr = 2π
C Ca

Example 4.3.8

201
 I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

4.4IJ Stokes’ Theorem

Our last variant of the fundamental theorem of calculus is Stokes’43 theorem, which is like
Green’s theorem, but in three dimensions. It relates an integral over a finite surface in R3
with an integral over the curve bounding the surface.

Theorem 4.4.1 (Stokes’ Theorem).

Let

• S be a piecewise smooth oriented surface (i.e. a unit normal n̂ has been cho-
sen at each point of S and this choice depends continuously on the point)
• the boundary, BS, of the surface S consist of a finite number of piecewise
smooth, simple curves that are oriented consistently with n̂ in the sense
that
– if you walk along BS in the direction of the arrow on BS,
– with the vector from your feet to your head having direction n̂
– then S is on your left hand side.

z
n̂

S
BS

• F be a vector field that has continuous first partial derivatives at every point
of S.

Then ¿ ij
F ¨ dr = ∇ ˆ F ¨ n̂ dS
BS S

Note that
• in Stokes’ theorem, S must be an oriented surface. In particular, S may not be a
Möbius strip. (See Example 3.5.3.)

43 Sir George Gabriel Stokes (1819–1903) was an Irish physicist and mathematician. In addition to Stokes’
theorem, he is known for the Navier-Stokes equations of fluid dynamics and for his work on the wave
theory of light. He gave evidence to the Royal Commission on the Use of Iron in Railway Structures
after the Dee bridge disaster of 1847.

202
I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

• If S is part of the xy-plane, then Stokes’ theorem reduces to Green’s theorem. Our
proof of Stokes’ theorem will consist of rewriting the integrals so as to allow an
application of Green’s theorem.

• If BS is simple closed curve and

– when you look at BS from high on the z-axis, it is oriented counterclockwise

(look at the figure in Theorem 4.4.1), then
– n̂ is upward pointing, i.e. has positive z-component, at least near BS.

Proof. Write F = F1 ı̂ı + F2 ̂ + F3 k̂. Both integrals involve F1 terms and F2 terms and F3
terms. We shall show that the F1 terms in the two integrals agree. In other words, we
shall assume that F = F1ı̂ı . The proofs that the F2 and F3 terms also agree are similar. For
simplicity, we’ll assume44 that the boundary of S consists of just a single curve, and that
we can
˝ pick a parametrization of S with

ˇ (
S= r(u, v) = x (u, v), y(u, v), z(u, v) ˇ (u, v) in R Ă R2
Br
and with r(u, v) orientation preserving in the sense that n̂ dS = + Bu ˆ Bv
Br
du dv. Also
˝ pick a parametrization of the curve, BR, bounding R as u(t), v(t) , a ď t ď b, in

such a way that when you walk along BR in the direction of increasing t, then R is
on your left.
Then the curve BS bounding S can be parametrized as R(t) = r u(t), v(t) , a ď t ď b.

z n̂
v

rpu, vq
S R
BS pu, vq

BR
y

u
x

The orientation of R(t):

We’ll now verify that the direction of increasing t for the parametrization R(t) of BS is
the direction of the arrow on BS in the figure on the left above. By continuity, it suffices to
check the orientation at a single point.
Find a point (u0 , v0 ) on BR where the forward pointing tangent vector is a positive
multiple of ı̂ı . The horizontal arrow on BR in the figure on the left below is at such a point.


Suppose that t = t0 at this point — in other words, suppose that (u0 , v0 ) = u(t0 ), v(t0 ) .

44 Otherwise, decompose S into simpler pieces, analogously to what we did in the proof of the divergence
theorem.

203
I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

Because the forward pointing tangent vector to BR at (u0 , v0 ), namely u1 (t0 ), v1 (t0 ) , is

of ı̂ı , we have u (t0 ) ą 0 and v (t0 ) = 0. The tangent vector to BS at

a positive multiple 1 1

R(t0 ) = r u0 , v0 , pointing in the direction of increasing t, is

d Br Br
R1 ( t0 ) = r u ( t ), v ( t ) ˇt = t = u 1 ( t 0 ) ( u 0 , v 0 ) + v 1 ( t 0 ) ( u 0 , v 0 )

ˇ
dt 0 Bu Bv
Br
= u1 ( t0 ) ( u0 , v0 )
Bu
Br
and so is a positive multiple of Bu (u0 , v0 ). See the figure on the right below.
If we now walk along a path in the uv-plane which starts at (u0 , v0 ), holds u fixed at u0
and increases v, we move into the interior of R starting at (u0 , v0 ). Correspondingly, if we
walk along the path, r(u0 , v), in R3 with v starting at v0 and increasing, we move into the
Br
interior of S. The forward tangent to this new path, Bv (u0 , v0 ), points from r(u0 , v0 ) into
the interior of S. It’s the blue arrow in the figure on the left below.

v z
Br Br
rpu, vq Bu
ˆ Bv
Br
Bv
R Br
Bu

BR pu0 , v0 q pu1 pt0 q, 0q Rpt0 q “ rpu0 , v0 q

BS
y
u
x

Now imagine that you are walking along BS in the direction of increasing t. At time
t0 you are at R(t0 ). You point your right arm straight ahead of you. So it is pointing
Br
in the direction Bu (u0 , v0 ). You point your left arm out sideways into the interior of S.
Br
It is pointing in the direction Bv (u0 , v0 ). If the direction of increasing t is the same as
the forward direction of the orientation of BS, then the vector from our feet to our head,
Br
which is Bu (u0 , v0 ) ˆ Bv
Br
(u0 , v0 ), should be pointing in the same direction as n̂. And since
n̂ dS = + Bu ˆ Bv du dv, it is.
Br Br

Now, with our parametrization and orientation sorted out, we can examine the inte-
grals.

The surface integral:

Since F = F1ı̂ı , so that
 
ı̂ı ̂ k̂
BF1 BF1

∇ ˆ F = det  BBx B B  = 0, ,´

By Bz  Bz By
F1 0 0

204
I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

and
 
ı̂ı ̂ k̂
Br Br  Bx By Bz 
n̂ dS = ˆ du dv = det  Bu Bu Bu 
Bu Bv By
Bx Bz
 Bv Bv Bv
By Bz Bz By Bz Bx Bx Bz Bx By By Bx
    
= ´ ı̂ı + ´ ̂ + ´ k̂
Bu Bv Bu Bv Bu Bv Bu Bv Bu Bv Bu Bv
and
ij ij 
BF1 BF1 Br Br

∇ ˆ F ¨ n̂ dS = 0, ,´ ¨ ˆ du dv
Bz By Bu Bv
S R
ij !
BF1 Bz Bx Bx Bz BF1 Bx By By Bx )
   
= ´ ´ ´ du dv
Bz Bu Bv Bu Bv By Bu Bv Bu Bv
R

Now we examine the line integral and show that it equals this one.
The line integral:
¿ żb 

 d
F ¨ dr = F r u(t), v(t) ¨ r u(t), v(t) dt


a dt
BS
żb 

 h Br
du Br
dv i
= F r u ( t ), v ( t ) ¨ u ( t ), v ( t ) (t) + u ( t ), v ( t ) (t) dt
a Bu dt Bv dt
We can write this as the line integral
¿ żbh

du dv i
M(u, v) du + N (u, v) dv = M u ( t ), v ( t ) (t) + N u(t), v(t)) (t) dt
a dt dt
BR

around BR, if we choose

Br
Bx
M(u, v) = F r(u, v) ¨ (u, v) = F1 x (u, v), y(u, v), z(u, v) (u, v)
Bu Bu

Br
Bx
N (u, v) = F r(u, v) ¨ (u, v) = F1 x (u, v), y(u, v), z(u, v) (u, v)
Bv Bv

Finally, we show that the surface integral equals the line integral:
By Green’s Theorem, we have
¿ ¿
F ¨ dr = M(u, v) du + N (u, v) dv
BS
ij ! BR
BN BM )
= ´ dudv
Bu Bv
R
ij !
B 
 Bx B2 x
= F1 x (u, v), y(u, v), z(u, v) + F1
Bu Bv BuBv
R
B 
 Bx B2 x )
´ F1 x (u, v), y(u, v), z(u, v) ´ F1 dudv
Bv Bu BvBu

205
I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

ij !
BF1 Bx BF1 By BF1 Bz  Bx B2 x
= + + + F1
Bx Bu By Bu Bz Bu Bv BuBv
R
 BF Bx BF By BF Bz  Bx
1 B2 x )
´ + 1 + 1 ´F1 dudv
Bx Bv By Bv Bz Bv Bu BvBu
ij !
BF1 By BF1 Bz  Bx  BF1 By BF1 Bz  Bx )
= + ´ + du dv
By Bu Bz Bu Bv By Bv Bz Bv Bu
R
ij
= ∇ ˆ F ¨ n̂ dS
S

which is the conclusion that we wanted.

Before we move on to some examples, here are a couple of remarks.
ű ť
• Stokes’ theorem says that C F ¨ dr = S ∇ ˆ F ¨ n̂ dS for any (suitably oriented)
surface whose boundary is C. So if S1 and S2 are two different (suitably oriented)
surfaces having the same boundary curve C, then
ij ij
∇ ˆ F ¨ n̂ dS = ∇ ˆ F ¨ n̂ dS
S1 S2

For example, if C is the unit circle

ˇ (
C= ( x, y, z) ˇ x2 + y2 = 1, z = 0
oriented counterclockwise when viewed from above, then both
ˇ (
S1 = ( x, y, z) ˇ x2 + y2 ď 1, z = 0
ˇ (
S2 = ( x, y, z) ˇ z ě 0, x2 + y2 + z2 = 1
with upward pointing
ť ť unit normal vectors, have boundary C. So Stokes’ tells us that
S1 ∇ ˆ F ¨ n̂ dS = S2 ∇ ˆ F ¨ n̂ dS.

n̂

S2
n̂

S1
C

ť ť
It should not be a surprise that S1 ∇ ˆ F ¨ n̂ dS = S2 ∇ ˆ F ¨ n̂ dS, for the following
reason. Let ˇ (
V = ( x, y, z) ˇ x2 + y2 + z2 ď 1, z ě 0
be the solid between S1 and S2 . The boundary BV of V is the union of S1 and S2 .

206
I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

N̂

S2
V

S1

N̂

But beware that the outward pointing normal to BV (call it N̂) is +n̂ on S2 and ´n̂
on S1 . So the divergence theorem gives

ij ij ij ij
∇ ˆ F ¨ n̂ dS ´ ∇ ˆ F ¨ n̂ dS = ∇ ˆ F ¨ N̂ dS + ∇ ˆ F ¨ N̂ dS
S2 S1 S S1
ij2
= ∇ ˆ F ¨ N̂ dS
BV
¡
∇ ¨ ∇ ˆ F dV


=
V by the divergence theorem
=0

by the vector identity Theorem 4.1.7.a.

• As a second remark, supposeű that the vector field F obeys ∇ ˆ F = 0 everywhere.

Then Stokes’ theorem forces C F ¨ dr = 0 are around all closed curves C, which im-
plies that F is conservative, by Theorem 2.4.7. So Stokes’ theorem provides another
proof of Theorem 2.4.8.

Here is an easy example which shows that Stokes’ can be very useful when ∇ ˆ F
simplifies.

Example 4.4.2
ű
Problem: Evaluate C F ¨ dr where F = 2z + sin x146 ı̂ı ´ 5z ̂ ´ 5y k̂ and the curve C is the



circle x2 + y2 = 4, z = 1, oriented counterclockwise when viewed from above.

Solution. The x146 in F will probably make a direct evaluation of the integral difficult. So
we’ll use Stokes’ theorem. To do so we need a surface S with BS = C. The simplest is just
the flat disk
ˇ (
S = ( x, y, z) ˇ x2 + y2 ď 4, z = 1

207
I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

z
n̂

S
C
y

Since
 
ı̂ı ̂ k̂
∇ ˆ F = det  B B B
 
Bx By Bz 
2z + sin x146 ´5z ´5y


" # " #
B B  B B
 B B
= ı̂ı det By Bz ´ ̂ det Bx
146

Bz + k̂ det Bx
By
´5z ´5y 2z + sin x ´5y 2z + sin x 146 ´5z
= 2̂
and the normal to S is k̂, Stokes’ theorem gives
¿ ij ij
F ¨ dr = ∇ ˆ F ¨ n̂ dS = (2̂ ) ¨ k̂ dS = 0
C S S

Example 4.4.2

Now we’ll repeat the last example with a harder curve.

Example 4.4.3
ű
Problem: Evaluate C F ¨ dr where F = 2z + sin x146 ı̂ı ´ 5z ̂ ´ 5y k̂ and the curve C is the



intersection of x2 + y2 + z2 = 4 and z = y, oriented counterclockwise when viewed from

above.
Solution. The surface x2 + y2 + z2 = 4 is the sphere of radius 2 centred on the origin and
z = y is a plane which contains the origin. So C, being the intersection of a sphere with
a plane through the centre of the sphere, is a circle, with centre (0, 0, 0) and radius 2. The
part of the circle in the first octant is sketched on the left below. The x146 in F will probably

z z

n̂
C y S C y

x x

208
I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

make a direct evaluation of the integral difficult. So we’ll use Stokes’ theorem. To do so
we need a surface S with BS = C. The simplest is the flat disk
ˇ (
S = ( x, y, z) ˇ x2 + y2 + z2 ď 4, z = y
The first octant of S is shown in the figure on the right above. We saw in the last Example
4.4.2 that
∇ ˆ F = 2̂
So Stokes’ theorem gives
¿ ij ij
F ¨ dr = ∇ ˆ F ¨ n̂ dS = 2 ̂ ¨ n̂ dS
C S S
ť
We’ll evaluate the integral 2 S ̂ ¨ n̂ dS in two ways. The first way is more efficient, but
also requires more insight. Since ∇ (z ´ y) = k̂ ´ ̂ , the upward unit normal to the plane
z ´ y = 0, and hence to S, is n̂ = ?12 (k̂ ´ ̂ ). Consequently the integrand
 ´̂ + k̂  1
̂ ¨ n̂ = ̂ ¨ ? = ´?
2 2
is a constant and we do not need a formula for n̂ dS:
¿ ij ? ij ? ?
F ¨ dr = 2 ̂ ¨ n̂ dS = ´ 2 dS = ´ 2Area(S) = ´ 2π 22
C S S
?
= ´4 2π
ť
Alternatively, we can evaluate the integral S ̂ ¨ n̂ dS using our normal protocol. As S
is part of the plane z = f ( x, y) = y,
n̂ dS = ˘ ´ f x ı̂ı ´ f y ̂ + k̂ dxdy = ˘(´̂ + k̂) dxdy

To get the upward pointing normal pointing normal, we take the + sign so that n̂ dS =
(´̂ + k̂) dxdy. As ( x, y, z) runs over
ˇ ( ˇ (
S = ( x, y, z) ˇ x2 + y2 + z2 ď 4, z = y = ( x, y, z) ˇ x2 + 2y2 ď 4, z = y
ˇ 2 2 (
= ( x, y, z) ˇ x4 + y2 ď 1, z = y
ˇ 2 2 (
( x, y) runs over the elliptical disk R = ( x, y) ˇ x4 + y2 ď 1 . The part of this ellipse in
the first octant is the shaded region in the figure below. This ellipse has semiaxes a = 2

n̂
y

209
I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM
? ?
and b = 2 and hence area πab = 2 2π. So
¿ ij ij ij
F ¨ dr = 2 ̂ ¨ n̂ dS = 2 ̂ ¨ (´̂ + k̂) dxdy = ´2 dxdy = ´2Area( R)
C S R R
?
= ´4 2π

Example 4.4.3

Example 4.4.4
ű
Problem: Evaluate C F ¨ dr where F = ( x + y) ı̂ı + 2( x ´ z) ̂ + (y2 + z) k̂ and C is the ori-
ented curve obtained by going from (2, 0, 0) to (0, 3, 0) to (0, 0, 6) and back to (2, 0, 0) along
straight line segments.

C2
C3
y
C1
x

Solution 1. In this first solution, we’ll evaluate the integral directly. The first line segment
(C1 in the figure above) may be parametrized as
(
r(t) = (2, 0, 0) + t (0, 3, 0) ´ (2, 0, 0) = 2 ´ 2t , 3t , 0 0ďtď1

So the integral along this segment is

ż1 ż1 ż1
dr 2
F(r(t)) ¨ dt = (2 + t , 2(2 ´ 2t) , (3t) ) ¨ (´2 , 3 , 0) dt = (8 ´ 14t) dt
0 dt 0 0
h i1
= 8t ´ 7t2 = 1
0

The second line segment (C2 in the figure above) may be parametrized as
(
r(t) = (0, 3, 0) + t (0, 0, 6) ´ (0, 3, 0) = 0 , 3 ´ 3t , 6t 0ďtď1

. So the integral along this segment is

ż1 ż1
dr
F(r(t)) ¨ 3(1 ´ t) , ´12t , 9(1 ´ t)2 + 6t ¨ (0, ´3, 6) dt


dt =
0 dt 0
ż1 h i1
= [36t + 54(1 ´ t)2 + 36t] dt = 18t2 ´ 18(1 ´ t)3 + 18t2
0 0
= 54

210
I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

The final line segment (C3 in the figure above) may be parametrized as
(
r(t) = (0, 0, 6) + t (2, 0, 0) ´ (0, 0, 6) = (2t , 0 , 6 ´ 6t) 0ďtď1

So the line integral along this segment is

ż1 ż1
dr
F(r(t)) ¨ 2t , 4t ´ 12(1 ´ t) , 6(1 ´ t) ¨ (2, 0, ´6) dt


dt =
0 dt 0
ż1 h i1
2 2
= [4t ´ 36(1 ´ t)] dt = 2t + 18(1 ´ t) = ´16
0 0

The full line integral is

¿
F ¨ dr = 1 + 54 ´ 16 = 39
C

Solution 2. This time we shall apply Stokes’ Theorem. The curl of F is

 
ı̂ı ̂ k̂
∇ ˆ F = det  B B B
 = (2y + 2)ı̂ı ´ (0 ´ 0)̂ + (2 ´ 1)k̂ = 2(y + 1)ı̂ı + k̂
 
Bx By Bz
x + y 2( x ´ z ) y2 + z

The curve C is a triangle and so is contained in a plane. Any plane has an equation of
the form Ax + By + Cz = D. Our plane does not pass through the origin (look at the
figure above) so the D must be nonzero. Consequently we may divide Ax + By + Cz = D
through by D giving an equation of the form ax + by + cz = 1.

˝ Because (2, 0, 0) lies on the plane, a = 21 .

˝ Because (0, 3, 0) lies on the plane, b = 13 .
˝ Because (0, 0, 6) lies on the plane, c = 16 .
y
So the triangle is contained in the plane 2x + 3 + 6z = 1. It is the boundary of the surface S
y
that consists of the portion of the plane 2x + 3 + 6z = 1 that obeys x ě 0, y ě 0 and z ě 0.
Rewrite the equation of the plane as z = 6 ´ 3x ´ 2y. For this surface

n̂ dS = (3ı̂ı + 2̂ + k̂) dx dy

by (3.3.2), and we can write

ˇ (
S= ( x, y, z) ˇ x ě 0, y ě 0, z ě 0, z = 6 ´ 3x ´ 2y
ˇ (
= ( x, y, z) ˇ x ě 0, y ě 0, 6 ´ 3x ´ 2y ě 0, z = 6 ´ 3x ´ 2y

As ( x, y, z) runs over S, ( x, y) runs over the triangle

ˇ (
R= ( x, y, z) ˇ x ě 0, y ě 0, 3x + 2y ď 6
ˇ (
= ( x, y, z) ˇ x ě 0, 0 ď y ď 23 (2 ´ x )

211
I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

z “ 6 ´ 3x ´ 2y
S
y
R
3x ` 2y “ 6
x z“0

Using horizontal strips as in the figure below,

¿ ij
F ¨ dr = ∇ ˆ F ¨ n̂ dS
y
C S
ij
3
= [2(y + 1)ı̂ı + k̂] ¨ [3ı̂ı + 2̂ + k̂] dx dy
R
ij ż3 ż 1 x “ 13 p6 ´ 2yq
3 (6´2y )
= [6y + 7] dx dy = dy dx [6y + 7]
0 0
R 2 x
ż3 ż
1 1 3
= dy [6y + 7][6 ´ 2y] = dy [´12y2 + 22y + 42]
0 3 3 0
1h i3 
3 2
´ 4y + 11y + 42y = ´ 4 ˆ 9 + 11 ˆ 3 + 42 = 39

=
3 0

Alternatively, using vertical strips as in the figure below,

¿ ij y
F ¨ dr = [6y + 7] dx dy 3
C R
ż2 ż 3
y “ 32 p2 ´ xq
2 (2´x )
= dx dy [6y + 7]
0 0
2 x
ż2 h 32 i2
3 i h 27 1 21 1
= dx 3 2 (2 ´ x )2 + 7 (2 ´ x ) = ´ (2 ´ x )3 ´ (2 ´ x )2
0 2 2 4 3 2 2 0

9 21
= 8+ 4 = 39
4 4

Example 4.4.4

Example 4.4.5
ű
Problem: Evaluate C F ¨ dr where F = (cos x + y + z) ı̂ı + ( x + z) ̂ + ( x + y) k̂ and C is the

212
I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

intersection of the surfaces

y2 z2
x2 +
+ =1 and z = x2 + 2y2
2 3
oriented counterclockwise when viewed from above.
y2 2
Solution. First, let’s sketch the curve. x2 + 2 + z3 = 1 is an ellipsoid centred on the origin
and z = x2 + 2y2 is an upward opening paraboloid that passes through the origin. They
are sketched in the figure below. The paraboloid is red. Their intersection, the curve C, is

z “ x2 ` 2y 2

y2 z2
x2 ` 2
` 3
“1

the blue curve in the figure. It looks like a deformed45 circle.

One could imagine parametrizing C. For example, substituting x2 = z ´ 2y2 into

2
the equation of the ellipsoid gives ´ 32 y2 + 13 (z + 32 = 74 . This can be solved to give y as
a function of z and then x2 = z ´ 2y2 also gives x as a function of z. However this would
clearly yield, at best, a really messy integral. So let’s try Stokes’ theorem.
In fact, since
 
ı̂ı ̂ k̂
∇ ˆ F = det  B B B
 
Bx By Bz

cos x + y + z x + z x + y
= ı̂ı 1 ´ 1) ´ ̂ 1 ´ 1 + k̂ 1 ´ 1

=0


This
ű F is conservative! (In fact F = ∇ sin x + xy + xz + yz .) As C is a closed curve,
C F ¨ dr = 0.
Example 4.4.5

Example 4.4.6
ť
Problem: Evaluate S G ¨ n̂ dS where G = (2x ) ı̂ı + (2z ´ 2x ) ̂ + (2x ´ 2z) k̂ and
ˇ (
( x, y, z) ˇ z = 1 ´ x2 ´ y2 (1 ´ y3 ) cos x ey , x2 + y2 ď 1


S=

45 By Salvador Dali?

213
I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

with upward pointing normal

Solution 1. The surface S is sketched below. It is a pretty weird surface. About the only

z “ p1 ´ x2 ´ y 2qp1 ´ y 3qey cos x

x2 ` y 2 “ 1, z “ 0

simple thing about it is that its boundary, BS, is the circle x2 + y2 = 1, z = 0. It is clear that
we should not try to evaluate the integral directly46 . In this solution we will combine the
divergence theorem with the observation that

B B B
∇¨G = (2x ) + (2z ´ 2x ) + (2x ´ 2z) = 0
Bx By Bz

to avoid ever having work with the surface S. Here is an outline of what we will do.

˝ We first select a simple surface S1 whose boundary BS1 is also the circle x2 + y2 = 1,
z = 0. A nice simple choice of S1 , and the surface that we will use, is the disk
ˇ (
S1 = ( x, y, z) ˇ x2 + y2 = 1, z = 0

˝ Then we define V to be the solid whose top surface is S and whose bottom surface
is S1 . So the boundary of V is the union of S and S1 .

n̂ V
n̂

S1

46 That way lies pain.

214
I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

˝ For S1 , we will use the upward pointing normal n̂ = k̂, which is minus the outward
pointing normal to BV on S1 . So the divergence theorem says that
¡ ij ij
∇ ¨ G dV = G ¨ n̂ dS ´ G ¨ n̂ dS
V S S1

The left hand side is zero because, as we have already seen, ∇ ¨ G = 0. So

ij ij
G ¨ n̂ dS = G ¨ n̂ dS
S S1
ť
˝ Finally, we compute S1 G ¨ n̂ dS.

We saw an argument like this (with G = ∇ ˆ F) in the first remark following the proof of
Theorem 4.4.1.
So all that we have to do now is compute
ij ij ij ij ij
G ¨ n̂ dS = G ¨ n̂ dS = G ¨ k̂ dS = (2x ´ 2z) dxdy = (2x ) dxdy
S S1 S1 x2 +y2 ď1 x2 +y2 ď1
z =0 z =0

=0

simply because the integrand is odd under x Ñ ´x.

Solution 2. In this second solution we’ll use Stokes’ theorem instead of the divergence
theorem. To do so, we have to express G in the form ∇ ˆ F. So the first thing to do is to
check if G passes the screening test, Theorem 4.1.12, for the existence of vector potentials.
That is, to check if ∇ ¨ G = 0. It is. We saw this in Solution 1 above.
Next, we have to find a vector potential. In fact, we have already found, in Example
4.1.15, that
F = (z2 ´ 2xz)ı̂ı + ( x2 ´ 2xz)̂

is a vector potential for G, which we can quickly check.

Parametrizing C by r(t) = cos t ı̂ı + sin t ̂ , 0 ď t ď 2π, Stokes’ theorem gives (recalling
that z = 0 on C)
ij ij ¿ ż 2π

dr
G ¨ n̂ dS = ∇ ˆ F ¨ n̂ dS = F ¨ dr = F r(t) ¨ dt
0 dt
S S C
ż 2π
cos2 t (cos t) dt


=
0

Of course this integral

can be evaluated by using that one antiderivative of the integrand
3 2 1 3
cos t = 1 ´ sin t cos t is sin t ´ 3 sin t and that this antiderivative is zero at t = 0 and
at t = 2π. But it is easier to observe that the integral of any odd power of sin t or cos t over
any full period is zero. Look, for example, at the graphs of sin3 x and cos3 x, below.

215
I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

y y
y “ sin3 x y “ cos3 x
1 1

x x
π{2 π 3π{2 2π π{2 π 3π{2 2π

´1 ´1

Either way
ij
G ¨ n̂ dS = 0
S

Example 4.4.6

Example 4.4.7

ű
In this example we compute, in three different ways, C F ¨ dr where
F = (z ´ y) ı̂ı ´ ( x + z) ̂ ´ ( x + y) k̂
and C is the curve x2 + y2 + z2 = 4, z = y oriented counterclockwise when viewed from
above.

Direct Computation: ű
In this first computation, we parametrize the curve C and compute C F ¨ dr directly. The
plane z = y passes through the origin, which is the centre of the sphere x2 + y2 + z2 = 4. So
C is a circle which, like the sphere, has radius 2 and centre (0, 0, 0). We use a parametriza-
tion of the form
r(t) = c + ρ cos t ı̂ı 1 + ρ sin t ̂ 1 0 ď t ď 2π
where

216
I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

˝ c = (0, 0, 0) is the centre of C,

˝ ρ = 2 is the radius of C and
˝ ı̂ı 1 and ̂ 1 are two vectors that
(a) are unit vectors,
(b) are parallel to the plane z = y and
(c) are mutually perpendicular.

k̂1
̂1
y
ı̂ı1
p2, 0, 0q
x

The trickiest part is finding suitable vectors ı̂ı 1 and ̂ 1 :

˝ The point (2, 0, 0) satisfies both x2 + y2 + z2 = 4 and z = y and so is on C. We may
choose ı̂ı 1 to be the unit vector in the direction from the centre (0, 0, 0) of the circle
towards (2, 0, 0). Namely ı̂ı 1 = (1, 0, 0).
˝ Since the plane of the circle is z ´ y = 0, the vector ∇ (z ´ y) = (0, ´1, 1) is perpen-
dicular to the plane of C. So k̂1 = ?12 (0, ´1, 1) is a unit vector normal to z = y. Then
̂ 1 = k̂1 ˆ ı̂ı 1 = ?1 (0, ´1, 1) ˆ (1, 0, 0)
?1 (0, 1, 1) is a unit vector that
= is perpendicu-
2 2
lar to ı̂ı 1 and k̂ . Since ̂ is
1 1
perpendicular to k̂1 , it is parallel to z = y.
Substituting in c = (0, 0, 0), ρ = 2, ı̂ı 1 = (1, 0, 0) and ̂ 1 = ?12 (0, 1, 1) gives
1  sin t sin t 
r(t) = 2 cos t (1, 0, 0) + 2 sin t ? (0, 1, 1) = 2 cos t, ? , ? 0 ď t ď 2π
2 2 2
? ?
To check that this parametrization is correct, note that x = 2 cos t, y = 2 sin t, z = 2 sin t
satisfies both x2 + y2 + z2 = 4 and z = y. ?
At t = 0,
r(0) = ? (2, 0,
? 0). As t increases, z(t) = 2 increases and r(t) moves upwards
towards r π2 = (0, 2, 2). This is the desired counterclockwise direction. Now that we
have a parametrization, we can set up the integral.
? ?

r(t) = 2 cos t, 2 sin t, 2 sin t
? ?
r 1 (t) = ´ 2 sin t, 2 cos t, 2 cos t

F r(t) = z(t) ´ y(t), ´x (t) ´ z(t), ´x (t) ´ y(t)

? ? ? ?
2 sin t ´ 2 sin t, ´2 cos t ´ 2 sin t, ´2 cos t ´ 2 sin t


=
? ?
= ´ 0, 2 cos t + 2 sin t, 2 cos t + 2 sin t


 ?  ? ?
F r(t) ¨ r 1 (t) = ´ 4 2 cos2 t + 4 cos t sin t = ´ 2 2 cos(2t) + 2 2 + 2 sin(2t)

 

217
I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

by the double angle formulae sin(2t) = 2 sin t cos t and cos(2t) = 2 cos2 t ´ 1. So
¿ ż 2π
F ¨ dr = F r(t) ¨ r 1 (t) dt


0
C
ż 2π
 ? ?
´ 2 2 cos(2t) + 2 2 + 2 sin(2t) dt

=
0
h? ? i2π
= ´ 2 sin(2t) + 2 2t ´ cos(2t)
0
?
= ´4 2π
Oof! Let’s do it an easier way.
Stokes’ Theorem
To apply Stokes’ theorem we need to express C as the boundary BS of a surface S. As
ˇ (
C = ( x, y, z) ˇ x2 + y2 + z2 = 4, z = y
is a closed curve, this is possible. In fact there are many possible choices of S with BS = C.
Three possible S’s (sketched below) are
ˇ (
S = ( x, y, z) ˇ x2 + y2 + z2 ď 4, z = y
ˇ (
S1 = ( x, y, z) ˇ x2 + y2 + z2 = 4, z ě y
ˇ (
S2 = ( x, y, z) ˇ x2 + y2 + z2 = 4, z ď y

S1
S

S2

The first of these, which is part of a plane, is likely to lead to simpler computations than
the last two, which are parts of a sphere. So we choose what looks like the simpler way.
In preparation for application of Stokes’ theorem, we compute ∇ ˆ F and n̂ dS. For the
latter, we apply the formula n̂ dS = ˘(´ f x , ´ f y , 1) dxdy (of Equation (3.3.2)) to the surface
z = f ( x, y) = y. We use the + sign to give the normal a positive k̂ component.
 
ı̂ı ̂ k̂
∇ ˆ F = det  BBx B B
 
By Bz

z ´ y ´x ´ z ´x ´ y
= ı̂ı ´ 1 ´ (´1) ´ ̂ ´ 1 ´ 1 + k̂ ´ 1 ´ (´1)

= 2 ̂
n̂ dS = (0, ´1, 1) dxdy
∇ ˆ F ¨ n̂ dS = (0, 2, 0) ¨ (0, ´1, 1) dxdy = ´2 dxdy

218
I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

The integration variables are x and y and, by definition, the domain of integration is
ˇ (
R = ( x, y) ˇ ( x, y, z) is in S for some z

To determine precisely what this domain of integration is, we observe that since z = y on
S ˇ ( ˇ (
S = ( x, y, z) ˇ x2 + 2y2 ď 4, z = y ùñ R = ( x, y) ˇ x2 + 2y2 ď 4
So the
? domain of integration
? is an ellipse with semimajor axis a = 2, semiminor axis
b = 2 and area πab = 2 2π. The integral is then
¿ ij ij ?
F ¨ dr = ∇ ˆ F ¨ n̂ dS = (´2) dxdy = ´2 Area ( R) = ´4 2π
C S R

Remark (Limits of integration):

If the integrand were more complicated, we would have to evaluate the integral over R by
expressing it as an iterated integrals with the correct limits of integration. First suppose
that we slice upaR using thin vertical
a slices. On each such slice, x is essentially constant and
y runs from ´ (4 ´ x )/2 to (4 ´ x2 )/2. The leftmost such slice would have x = ´2
2
and the rightmost such slice would have x = 2. So the correct limits with this slicing are
y
x2 ` 2y 2 “ 4 ij ż2 ż ?(4´x2 )/2
f ( x, y) dxdy = dx ? dy f ( x, y)
x ´2 ´ (4´x2 )/2
R

If, instead, we slice up R using thin

a horizontala slices, then, on each such slice, y is essen-
tially constant
? and x runs from ´ 4 ´ 2y to 4 ?
2 ´ 2y2 . The bottom such slice would have
y = ´ 2 and the top such slice would have y = 2. So the correct limits with this slicing
are
y
x2 ` 2y 2 “ 4 ij ż ?2 ż ?4´2y2
f ( x, y) dxdy = ? dy ? dx f ( x, y)
x ´ 2 ´ 4´2y2
R

Note that the integral with limits

y
ż ?2 ż2
? dy dx f ( x, y)
x ´ 2 ´2

corresponds to a slicing with x running from ´2 to 2 on every slice. This corresponds to a

rectangular domain of integration, not what we have here.
Stokes’ Theorem, Again:

219
I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

Since the integrand is just a constant (after Stoking — not the original ť integrand) and S is
so simple (because we chose it wisely), we can evaluate the integral S ∇ ˆ F ¨ n̂ dS with-
out ever determining dS explicitly and without ever setting up any limits of integration.
We already know that ∇ ˆ F = 2 ̂ . Since S is the level surface z ´ y = 0, the gradient
∇ (z ´ y) = ´̂ + k̂ is normal to S. So n̂ = ?12 (´̂ + k̂) and
¿ ij ij ij ? ?
1
F ¨ dr = ∇ ˆ F ¨ n̂ dS = (2̂ ) ¨ ? (´̂ + k̂) dS = ´ 2 dS = ´ 2 Area (S)
2
C S S S
ű ?
As S is a circle of radius 2, C F ¨ dr = ´4 2π, yet again.
Example 4.4.7

Example 4.4.8

In Warning 4.1.17, we stated that if a vector field fails to pass the screening test ∇ ¨ B = 0
at even a single point, for example because the vector field is not defined at that point,
then B can fail to have a vector potential. An example is the point source
r̂( x, y, z)
B( x, y, z) =
r ( x, y, z)2
of Example 3.4.2. Here, as usual,
b
xı̂ı + y̂ + zk̂
r ( x, y, z) = x2 + y2 + z2 r̂( x, y, z) = a
x 2 + y2 + z2
This vector field is defined on all of R3 , except for the origin, and its divergence
B B B
     
x y z
∇¨B = + +
Bx ( x2 + y2 + z2 )3/2 By ( x2 + y2 + z2 )3/2 Bz ( x2 + y2 + z2 )3/2
3x2 3y2
   
1 1
= ´ + ´
( x2 + y2 + z2 )3/2 ( x2 + y2 + z2 )5/2 ( x2 + y2 + z2 )3/2 ( x2 + y2 + z2 )5/2
3z2
 
1
+ ´
( x2 + y2 + z2 )3/2 ( x2 + y2 + z2 )5/2
3 3( x 2 + y2 + z2 )
= 2 ´
( x + y2 + z2 )3/2 ( x2 + y2 + z2 )5/2
is zero everywhere except at the origin, where it is not defined.
This vector field cannot
ˇ have a vector potential
( on its domain of definition, i.e. on
ˇ
R zt(0, 0, 0)u = ( x, y, z) ( x, y, z) ‰ (0, 0, 0) . To see this, suppose to the contrary that
3

it did have a vector potential A. Then its flux through any closed surface47 (i.e. surface
without a boundary) S would be
ij ij ¿
B ¨ n̂ dS = ∇ ˆ A ¨ n̂ dS = A ¨ dr = 0
S S BS

47 If you are uncomfortable with the surface not having a boundary, poke a very small hole in the surface,
giving it a very small boundary. Then take the limit as the hole tends to zero.

220
 I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

by Stokes’ theorem. But we found in Example 3.4.2, with m = 1, that the flux of B through
any sphere centred on the origin is 4π.
Example 4.4.8

4.4.1 §§ The Interpretation of Div and Curl Revisited

In sections 4.1.4 and 4.1.5 we derived interpretations of the divergence and of the curl.
Now that we have the divergence theorem and Stokes’ theorem, we can simplify those
derivations a lot.

§§§ Divergence
Let ε ą 0 be a tiny positive number, and then let
ˇ (
Bε ( x0 , y0 , z0 ) = ( x, y, z) ˇ ( x ´ x0 )2 + (y ´ y0 )2 + (z ´ z0 )2 ă ε2

be a tiny ball of radius ε centred on the point ( x0 , y0 , z0 ). Denote by

ˇ (
Sε ( x0 , y0 , z0 ) = ( x, y, z) ˇ ( x ´ x0 )2 + (y ´ y0 )2 + (z ´ z0 )2 = ε2

its surface. Because Bε ( x0 , y0 , z0 ) is really small, ∇ ¨ v is essentially constant in Bε ( x0 , y0 , z0 )

and we essentially have
¡
∇ ¨ v dV = ∇ ¨ v( x0 , y0 , z0 ) Vol Bε ( x0 , y0 , z0 )

Bε ( x0 ,y0 ,z0 )

Of course we are really making an approximation here, based on the assumption that
v( x, y, z) is continuous and so takes values very close to v( x0 , y0 , z0 ) everywhere on the
domain of integration. The approximation gets better and better as ε Ñ 0 and a more
precise statement is
ţ
Bε ( x0 ,y0 ,z0 ) ∇ ¨ v dV
∇ ¨ v( x0 , y0 , z0 ) = lim

εÑ0 Vol Bε ( x0 , y0 , z0 )

By the divergence theorem, we also have

¡ ij
∇ ¨ v dV = v ¨ n̂ dS
Bε ( x0 ,y0 ,z0 ) Sε ( x0 ,y0 ,z0 )

Think of the vector field v as the velocity of a moving fluid which has density one. We
have already seen, in §3.4, that the flux integral for a velocity field has the interpretation
ij "
the volume of fluid leaving Bε ( x0 , y0 , z0 ) through
v ¨ n̂ dS =
Sε ( x0 , y0 , z0 ) per unit time
Sε ( x0 ,y0 ,z0 )

221
I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

We conclude that, as we said in (4.1.19),

the rate at which fluid is exiting Bε ( x0 , y0 , z0 )
∇ ¨ v( x0 , y0 , z0 ) = lim

εÑ0 Vol Bε ( x0 , y0 , z0 )
"
rate at which fluid is exiting an infinitesimal sphere
=
centred at ( x0 , y0 , z0 ), per unit time, per unit volume
= strength of the source at ( x0 , y0 , z0 )
If our world is filled with an incompressible fluid, a fluid whose density is constant and
so never expands or compresses, we will have ∇ ¨ v = 0.

§§§ Curl
Again let ε ą 0 be a tiny positive number and let Dε ( x0 , y0 , z0 ) be a tiny flat circular disk
of radius ε centred on the point ( x0 , y0 , z0 ) and denote by Cε ( x0 , y0 , z0 ) its boundary circle.
Let n̂ be a unit normal vector to Dε . It tells us the orientation of Dε . Give the circle Cε the
corresponding orientation using the right hand rule. That is, if the fingers of your right
hand are pointing in the corresponding direction of motion along Cε and your palm is
facing Dε , then the thumb is pointing in the direction n̂.

n̂

ε
Cε

Because Dε ( x0 , y0 , z0 ) is really small, ∇ ˆ v is essentially constant on Dε ( x0 , y0 , z0 ) and

we essentially have
ij
∇ ˆ v ¨ n̂ dS = ∇ ˆ v( x0 , y0 , z0 ) ¨ n̂ Area Dε ( x0 , y0 , z0 )

Dε ( x0 ,y0 ,z0 )

= πε2 ∇ ˆv( x0 , y0 , z0 ) ¨ n̂
Again, this is really an approximate statement which gets better and better as ε Ñ 0. A
more precise statement is
ť
Dε ( x0 ,y0 ,z0 ) ∇ ˆ v ¨ n̂ dS
∇ ˆv( x0 , y0 , z0 ) ¨ n̂ = lim
εÑ0 πε2
By Stokes’ theorem, we also have
ij ¿
∇ ˆ v ¨ n̂ dS = v ¨ dr
Dε ( x0 ,y0 ,z0 ) Cε ( x0 ,y0 ,z0 )
ű
Again, think of the vector field v as the velocity of a moving fluid. Then Cε v ¨ dr is called
the circulation of v around Cε .
To measure the circulation experimentally, place a small paddle wheel in the fluid,
with the axle of the paddle wheel pointing along n̂ and each of the paddles perpendicular
to Cε and centred on Cε . Each paddle moves tangentially to Cε . It would like to move with

222
 I NTEGRAL T HEOREMS 4.4 S TOKES ’ T HEOREM

n̂

t̂
ε
Cε

the same speed as the tangential speed v ¨ t̂ (where t̂ is the forward pointing unit tangent
vector to Cε at the location of the paddle) of the fluid at its location. But all paddles
are rigidly fixed to the axle of the paddle wheel and so must all move with the same
speed. That common speed will be the average value of v ¨ t̂ around Cε . If ds represents
an element of arc length of Cε , the average value of v ¨ t̂ around Cε is
¿ ¿
1 1
vT = v ¨ t̂ ds = v ¨ dr
2πε 2πε
Cε Cε

since dr has direction t̂ and length ds so that dr = t̂ds, and since 2πε is the circumference
of Cε . If the paddle wheel rotates at Ω radians per unit time, each paddle travels a distance
Ωε per unit time (remember that ε is the radius of Cε ). That is, v T = Ωε. Combining all
this information,
ť
Dε ( x0 ,y0 ,z0 ) ∇ ˆ v ¨ n̂ dS
∇ ˆv( x0 , y0 , z0 ) ¨ n̂ = lim
εÑ0
ű πε2
v ¨ dr
= lim Cε 2
εÑ0 πε
2πε v T
= lim
εÑ0 πε2
2πε (Ωε)
= lim
εÑ0 πε2
= 2Ω
so that
Ω = 12 ∇ ˆv( x0 , y0 , z0 ) ¨ n̂
The component of ∇ ˆv( x0 , y0 , z0 ) in any direction n̂ is twice the rate at which the paddle
wheel turns when it is put into the fluid at ( x0 , y0 , z0 ) with its axle pointing in the direction
n̂. The direction of ∇ ˆv( x0 , y0 , z0 ) is the axle direction which gives maximum rate of
rotation and the magnitude of ∇ ˆv( x0 , y0 , z0 ) is twice that maximum rate of rotation. For
this reason, ∇ ˆ v is called the “vorticity”.

4.4.2 §§ Optional — An Application of Stokes’ Theorem — Faraday’s Law

Magnetic induction refers to a physical process whereby an electric voltage is created (“in-
duced”) by a time varying magnetic field. This process is exploited in many applications,
including electric generators, induction motors, induction cooking, induction welding
and inductive charging. Michael Faraday48 is generally credited with the discovery of

48 Michael Faraday (1791–1867) was an English physicist and chemist. He ended up being an extremely
influential scientist despite having only the most basic of formal educations.

223
 I NTEGRAL T HEOREMS 4.5 O PTIONAL — W HICH V ECTOR F IELDS O BEY ∇ ˆ F = 0

magnetic induction. Faraday’s law is the following. Let S be an oriented surface with
boundary C. Let E and B be the (time dependent) electric and magnetic fields and define
¿ B
E ¨ dr = voltage around C
C
ij
B ¨ n̂ dS = magnetic flux through S S
S C
Then the voltage around C is the negative of the rate of change of the magnetic flux
through S. As an equation, Faraday’s Law is
¿ ij
B
E ¨ dr = ´ B ¨ n̂ dS
Bt
C S

We can reformulate this as a partial differential equation. By Stokes’ Theorem,

¿ ij
E ¨ dr = ∇ ˆ E) ¨ n̂ dS
(∇
C S

so Faraday’s law becomes

ij 
BB 
∇ˆE+ ¨ n̂ dS = 0
Bt
S

This is true for all surfaces S. So the integrand, assuming that it is continuous, must be
zero.  
To see this, let G = ∇ ˆ E + Bt . Suppose that G(x0 ) ‰ 0. Pick a unit vector n̂ in
BB

the direction of G(x0 ). Let S be a very small flat disk centered on x0 with normal n̂ (the
vector we picked). Then G(x0 ) ¨ n̂ ą 0 and, by continuity, G(x) ¨ n̂ ą 0 for all x on S, if we
ť  
have picked S small enough. Then S ∇ ˆ E + BB Bt ¨ n̂ dS ą 0, which is a contradiction.
So G = 0 everywhere and we conclude that
BB
∇ˆE+ =0
Bt
This is one of Maxwell’s electromagnetic field equations49 .

4.5IJ Optional — Which Vector Fields Obey ∇ ˆ F = 0

We already know that if a vector field F passes the screening test ∇ ˆ F = 0 on all of R2 or
R3 , then there is a function ϕ with F = ∇ ϕ. That is, F is conservative. We are now going
to take a first look at what happens50 when F passes the screening test ∇ ˆ F = 0 only on

49 For the others, see Example 4.1.2

50 Russell Crowe posed a related question in the movie A Beautiful Mind. The movie is based on the life of
the American mathematician John Nash, who won a Nobel Prize in Economics.

224
I NTEGRAL T HEOREMS 4.5 O PTIONAL — W HICH V ECTOR F IELDS O BEY ∇ ˆ F = 0

some proper subset D of Rn , n = 2 or 3. We will just scratch the surface of this topic —
there is a whole subbranch of Mathematics (cohomology theory, which is part of algebraic
topology) concerned with a general form of this question. We shall imagine that we are
given a vector field F that is only defined on D and we shall assume
˝ that D is a connected, open subset of Rn with n = 2 or n = 3 (see Definition 4.5.1,
below)
˝ that all first order derivatives of all vector fields and functions that we consider are
continuous and
˝ that all curves we consider are piecewise smooth. A curve is piecewise smooth if it
is a union of a finite number of smooth curves C1 , C2 , ¨ ¨ ¨ , Cm with the end point of Ci
being the beginning point of Ci+1 for each 1 ď i ă m. A curve is smooth51 if it has a
parametrization r(t), a ď t ď b, whose first derivative r1 (t) exists, is continuous and
is nonzero everywhere.

C1 C2
C3

Definition 4.5.1.

Let n ě 1 be an integer.

(a) Let a P Rn and ε ą 0. The open ball of radius ε centred on a is

ˇ (
Bε (a) = x P Rn ˇ |x ´ a| ă ε

Note the strict inequality in |x ´ a| ă ε.

(b) A subset O Ă Rn is said to be an “open subset of Rn ” if, for each point a P O ,

there is an ε ą 0 such that Bε (a) Ă O . Equivalently, O is open if and only if it
is a union of open balls.

(c) A subset D Ă Rn is said to be (pathwise) connected if every pair of points in

D can be joined by a piecewise smooth curve in D .

Here are some examples to help explain this definition.

Example 4.5.2

ˇ (
(a) The “open rectangle” O = ( x, y) P R2 ˇ 0 ă x ă 1, 0 ă y ă 1 is an open
subset of R2 because any point a = ( x0 , y0 ) P O is a nonzero distance, namely d =

51 The word “smooth” does not have a universal meaning in mathematics. It is used with different mean-
ings in different contexts. We are here using one of the standard definitions. Another standard defini-
tion requires that all derivatives of all orders are continuous.

225
I NTEGRAL T HEOREMS 4.5 O PTIONAL — W HICH V ECTOR F IELDS O BEY ∇ ˆ F = 0
(
min x0 , 1 ´ x0 , y0 , 1 ´ y0 away from the boundary of O . So the open ball Bd/2 (a) is
contained in O .
ˇ (
(b) The “closed rectangle” C = ( x, y) P R2 ˇ 0 ď x ď 1, 0 ď y ď 1 is not an open subset
of R2 . For example, 0 = (0, 0) is a point in C . No matter what ε ą 0 we pick, the open
ball Bε (0) is not contained in C because Bε (0) contains the point (´ 2ε , 0), which is not
in C .

` ˘
C Bε px0 , 0q
O

ˇ (
(c) The x-axis, X = ( x, y) P R2 ˇ y = 0 , in R2 is not an open subset of R2 because for
any point ( x0 , 0) P X and any ε ą 0, the ball Bε ( x0 , 0) contains points with nonzero

y-coordinates and so is not contained in X .

(d) The union of open balls

ˇ (
B1 (0, 0) Y B1 (2, 0) = ( x, y) P R2 ˇ x2 + y2 ă 1 or ( x ´ 2)2 + y2 ă 1

is not connected, since any continuous path from, for example, (2, 0) to (0, 0) must
leave the union. In the figure on the left below, an “empty disk” has been sketched at
(1, 0) just to emphasise that the point (1, 0) is not in the union.

(e) On the other hand the union of “closed balls”

ˇ (
( x, y) P R2 ˇ x2 + y2 ď 1 or ( x ´ 2)2 + y2 ď 1

is connected. For example, the straight line segment from (2, 0) to (0, 0) remains in the
the union.

p0,0q p2,0q p0,0q p2,0q

Example 4.5.2

Many, but not all, of the basic facts that we developed, in §2.4.1, about conservative
fields in Rn also applies (with the same proofs) to fields on D .

226
I NTEGRAL T HEOREMS 4.5 O PTIONAL — W HICH V ECTOR F IELDS O BEY ∇ ˆ F = 0

Theorem 4.5.3.

For a vector field F on D Ă Rn ,

F is conservative on D ðñ F = ∇ ϕ on D , for some function ϕ

ż
ðñ for each P0 , P1 P D , the integral F ¨ dr takes
C
the same value for all curves C from P0 to P1
¿
ðñ F ¨ dr = 0 for all closed curves C in D
C
ùñ ∇ ˆ F = 0 on D

Note that the last line of this theorem contains only a one way implication.
Combining this with Stokes’ Theorem 4.4.1 (when n = 3, or Green’s Theorem 4.3.2
when n = 2) gives us the following two consequences.

Theorem 4.5.4.

(a) If D has the property that

every closed curve C in D is the boundary

(H)
of a bounded oriented surface, S , in D

then
F is conservative on D ðñ ∇ ˆ F = 0 on D

(b) For any D , if ∇ ˆ F = 0 on D , then F is locally conservative. This means that

for each point x0 P D , there is an ε ą 0 and a function ϕ such that F = ∇ ϕ on
Bε (x0 ).

Proof. (a) This is simply because if ∇ ˆ F = 0 on D and if the curve C = B S , with S an

oriented surface in D , then Stokes’ theorem gives
ż ż ij
F ¨ dr = F ¨ dr = ∇ ˆ F ¨ n̂ dS = 0
C BS
S

So F is conservative by Theorem 4.5.3.

(b) This is true simply because Bε (x0 ) satisfies property (H).

Example 4.5.5

Here are some examples of D ’s that violate (H).

227
I NTEGRAL T HEOREMS 4.5 O PTIONAL — W HICH V ECTOR F IELDS O BEY ∇ ˆ F = 0
ˇ (
˝ When D = D1 = ( x, y) P R2 ˇ 0 ă |( x, y)| ă 3 (an open ball with its centre
removed), then the circle x2 + y2 = 4 is a curve in D that is not the boundary of a
surface in D . The circle x2 + y2 = 4 is the boundary of the disk x2 + y2 ă 4, but the
disk x2 + y2 ă 4 is not contained in D because the point (0, 0) is in the disk and not
in D . See the figure on the left below.
ˇ (
˝ When D = D2 = ( x, y, z) P R3 ˇ |( x, y, z)| ă 2, |( x, y)| ą 0 (an open ball with the
z-axis removed), then the circle x2 + y2 = 1, z = 0 is a curve in D that is not the
boundary of a surface in D . The circle is the boundary of many different surfaces
in R3 , but each contains a point on the z-axis and so is not contained in D . See the
figure in the centre below.

D2 D3
D1

On the other hand, here is an example which does satisfy (H).

ˇ (
˝ Let D = D3 = ( x, y, z) P R3 ˇ 0 ă |( x, y, z)| ă 2 (an open ball with its centre
removed). For ˇexample the circle x2 +(y2 = 1, z = 0 is the boundary of the surface
( x, y, z) P R3 ˇ x2 + y2 + z2 = 1, z ą 0 Ă D . See the figure on the right above.

Example 4.5.5

This leaves the question of what happens when (H) is violated. We’ll just look at one
example, which however gives the flavour of the general theory.
The punctured disk is
ˇ (
D= ( x, y) P R2 ˇ 0 ă |( x, y)| ă 1
D
We’ll start by looking at one particular vector field, which passes the screening test, but
which cannot possibly be conservative. The field, which we saw in Example 2.3.14, is
y x
Θ=´ 2 ı
ı̂ + ̂
x + y2 x 2 + y2
with domain of definition D . We’ll first check that it passes the screening test:
!B  x  B  y )
∇ ˆΘ = ´ ´ 2 k̂
Bx x2 + y2 By x + y2
! 1 2x2   1 2y2 )
= ´ + ´ k̂
x 2 + y2 ( x 2 + y2 )2 x 2 + y2 ( x 2 + y2 )2
=0

228
I NTEGRAL T HEOREMS 4.5 O PTIONAL — W HICH V ECTOR F IELDS O BEY ∇ ˆ F = 0

Next we’ll check that it cannot be conservative. Denote by Cε the circle x2 + y2 = ε2 , with
counterclockwise orientation. Parametrize Cε by r(θ ) = ε cos θ ı̂ı + ε sin θ ̂ with 0 ď θ ď 2π.
Then

ż ż 2π

dr
Θ ¨ dr = Θ r(θ ) ¨ (θ ) dθ
Cε 0 dθ
ż 2π 
1 1 
´ sin θ ı̂ı + cos θ ̂ ¨ ´ ε sin θ ı̂ı + ε cos θ ̂ dθ


= (E1)
0 ε ε
ż 2π
= dθ
0
= 2π

ş By Theorem 4.5.3, Θ cannot be conservative on the punctured disk since the

is not zero.
integral Cε Θ ¨ dr around the closed curve Cε is nonzero.
Next we’ll check that it is locally conservative. That is, it can be written in the form
∇ θ ( x, y) near any point ( x0 , y0 ) in its domain. Define θ ( x, y) to be the polar angle of ( x, y)
with, for example, ´π ă θ ă π. This θ is defined on all of D , except for the negative real
axis. The domain of definition, Dπ , is sketched on the left below. If ( x0 , y0 ) happens to lie

Dπ D0

on the negative real axis, just replace ´π ă θ ă π by a different interval of length 2π, like
0 ă θ ă 2π. The domain of definition of θ would then change to the D0 , sketched on the
right above.
It’s now a simple matter to check that ∇ θ ( x, y) = Θ ( x, y) on the domain of definition
of θ. If x ‰ 0, then, from the figure below,

y px, yq
?
x2 `y 2
y
θpx, yq
x x

229
I NTEGRAL T HEOREMS 4.5 O PTIONAL — W HICH V ECTOR F IELDS O BEY ∇ ˆ F = 0

y
we have that tan θ ( x, y) = x , and cos θ ( x, y) = ? x
, so that
x 2 + y2

B y hB i y
tan θ ( x, y) = ´ 2 ùñ θ ( x, y) sec2 θ ( x, y) = ´
Bx x Bx x2
B y y x2 y
ùñ θ ( x, y) = ´ 2 cos2 θ ( x, y) = ´ 2 2 = ´
Bx x x x + y2 x 2 + y2
B 1 hB i 1
tan θ ( x, y) = ùñ θ ( x, y) sec2 θ ( x, y) =
By x By x
B 1 1 x2 x
ùñ θ ( x, y) = cos2 θ ( x, y) = = 2
By x 2
x x +y 2 x + y2
If x = 0, then we must have y ‰ 0 (since (0, 0) is not in the domain of definition to θ), and
we can use cot θ ( x, y) = yx instead and arrive at the same result.
So far we have just looked at one vector field on D . We are now ready to consider any
vector field F on D that passes the screening test ∇ ˆ F = 0 on D . We claim that there is a
function ϕ on D such that
¿
1
F = αF Θ + ∇ ϕ where αF = F ¨ dr (E2)
2π
Cε

The significance of this claim is that it says that if a vector field on D passes the screening
test on D , then, either it is conservative (that’s the case if and only if αF = 0) or, if it fails to
be conservative, then it differs from a conservative field (namely ∇ ϕ) only by a constant
(namely αF ) times the fixed vector field Θ . That is, there is only one nonconservative
vector field on D that passes the screening test, up to multiplication by constants and
addition of conservative fields. This is a nice simple surprise.
Observe that in the definition of αF , we did not specify the radius ε of the circle Cε
to be used for the integration curve. That’s because the answer to the integral does not
( take any 0 ă ε ă ε ă 1 and consider the surface
depend on the choice 1
ˇ 1 of ε. To see this,
S = ( x, y) P R ˇ ε ă |( x, y)| ă ε . It is completely contained in D . The boundary
2

´Cε1

Cε

of S consists of two parts. The outside part is Cε , oriented counterclockwise as usual.

The inside part is Cε1 , but oriented clockwise. It is usually denoted ´Cε1 . So, by Stokes’
theorem,
¿ ¿ ¿ ¿ ¿ ij
F ¨ dr ´ F ¨ dr = F ¨ dr + F ¨ dr = F ¨ dr = ∇ ˆ F ¨ n̂ dS = 0
Cε C ε1 Cε ´C ε1 BS S

230
I NTEGRAL T HEOREMS 4.5 O PTIONAL — W HICH V ECTOR F IELDS O BEY ∇ ˆ F = 0

Finally to verify the claim (E2), we check thatű the vector field G = F ´ αFΘ is conser-
vative on D . To do so, it suffices to check that C G ¨ dr = 0 for any closed curve C in D . In
fact we can restrict our attention to curves C that are simple, closed, counterclockwise ori-
ented curves on D . A curve is called simple if it does not cross itself. Closed curves which
are not simple can be split upű into simple closed subcurves. And changing the orientation
of C just changes the sign of C G ¨ dr = 0, which does not affect whether it is zero or not.
ű let C be a simple, closed, counterclockwise2 oriented curve
So in D . We need to verify
that C G ¨ dr = 0. Any simple closed curve in R divides R into three mutually disjoint
2

subsets52 — C itself, the set of points inside C and the set of points outside C . Since (0, 0)
is not on C , it must be either outside C , as in the figure of the left below, or inside C as in
the figure on the right below.

Cε

C C
D D

˝ Case 1: (0, 0) outside C . In this case C is the boundary of a set, S, which is completely
contained in D , namely all of the points inside C . So, by Stokes’ theorem,
¿ ¿ ij ij
G ¨ dr = F ´ αFΘ ¨ dr = ∇ ˆ F ¨ n̂ dS ´ αF ∇ ˆ Θ ¨ n̂ dS = 0 ´ αF 0

C BS S S
=0

˝ Case 2: (0, 0) inside C . Since (0, 0) is not on C , we can choose ε small enough that the
circle Cε lies completely inside C . Then the curve C ´ Cε is the boundary of a set, S,
which is completely contained in D , namely the part of D that is between Cε and C .
So, by Stokes’ theorem,
¿ ¿ ¿ ¿ ij
G ¨ dr ´ G ¨ dr = G ¨ dr = G ¨ dr = ∇ ˆ G ¨ n̂ dS = 0
C Cε C ´C ε BS S

since ∇ ˆ G = ∇ ˆ F ´ αF∇ ˆ Θ = 0 on D . Hence

¿ ¿ ¿ ¿
G ¨ dr = G ¨ dr = F ¨ dr ´ αF Θ ¨ dr = 2παF ´ αF (2π ) = 0
C Cε Cε Cε

by the definition, (E2), of αF and (E1).

52 This, intuitively obvious, but hard to prove, result is called the Jordan curve theorem. It is named after
the French mathematician Camille Jordan (1838–1922), who first proved it.

231
 I NTEGRAL T HEOREMS 4.6 O PTIONAL — M ORE I NTERPRETATION OF D IV AND C URL

So G is conservative on D and F is of the form (E2) on D .

The ideas that we have explored here can be generalised quite a bit. For example, if we
had a disk with n ą 1 punctures, we could use arguments like those above to show that
any vector field F that passes the screening test has to be of the form
n
ÿ
F = ∇ϕ + α` Θ `
`=1

with Θ ` simply being the above Θ translated so as to be centered on the `th puncture.

4.6IJ Optional — More Interpretation of Div and Curl

We are now going to determine, in much more detail than before53 , what the divergence
and curl of a vector field tells us about the flow of that vector field.
Consider a (possibly compressible) fluid with velocity field v(x, t). Pick any time t0
and a really tiny piece of the fluid; assume that, at time t0 , it is a cube with corners at

ˇ ( εêp1q
x0 + n1 εê(1) + n2 εê(2) + n3 εê(3) ˇ n1 , n2 , n3 P t0, 1u
εêp2q
εêp3q x0

Here ε ą 0 is the length of each edge of the cube and is assumed to be really small.
The vectors ê(1) , ê(2) and ê(3) are three mutually perpendicular unit vectors that give the
orientation of the edges of the cube. The vectors from the corner x0 to its three nearest
neighbour corners are εê(1) , εê(2) and εê(3) .
As time progresses, the chunk of fluid moves. In particular, the corners move. Let
us denote by εb1 (t) the vector, at time t, joining the n1 = n2 = n3 = 0 corner to the
n1 = 1, n2 = n3 = 0 corner. Define εb2 (t) and εb3 (t) similarly. For times very close to t0
we can think of our chunk of fluid as being essentially a parallelepiped with edges εbk (t).

εb3 ptq εb1 ptq

εb2 ptq

53 We’ll also use some more mathematics than before. In this section, we’ll use eigenvalues and eigenvec-
tors and solve some simple systems of ordinary differential equations. We’ll also need to use a lot of
subscripts and superscripts. It only looks intimidating.

232
I NTEGRAL T HEOREMS 4.6 O PTIONAL — M ORE I NTERPRETATION OF D IV AND C URL

By concentrating on the edges εbk (t) of the chunk of fluid, rather than the corners, we are
ignoring any translations that the chunk of fluid might have undergone. We want, instead,
to determine how the size and orientation of the parallelepiped changes as t increases.
At time t0 , bk = ê(k) . The velocities of the corners of the chunk of fluid at time t0 are

v x0 + n1 εê(1) + n2 εê(2) + n3 εê(3) , t0

In particular, at time t0 , the tail of εbk has velocity v(x0 , t0 ) and the head of εbk has velocity
v(x0 + εê(k) , t0 ). Consequently (using a Taylor approximation),

dbk ř
3 Bv
(k)
(t0 ) = v x0 + εê(k) , t0 ´ v x0 , t0 = x0 , t0 ê j + O(ε2 )



ε ε
dt j=1 Bx j

and so
dbk 3 Bv
ř
(k)
( t0 ) = x0 , t0 ê j + O(ε)
dt j=1 Bx j

The notation O(εn ) represents a function that is bounded by a constant times εn for all
ř3 Bv
(k)
sufficiently small ε. That is, we are saying that db
dt
k
( t 0 ) is j = 1 Bx x 0 , t 0 ê j plus a small
j
(k)
error that is bounded by a constant time ε. The notation ê j just refers to the jth component
of the vector ê(k) .
Define the 3 ˆ 3 matrix V by
Bvi

Vi,j = x0 , t0 (M)
Bx j
Then we can write the above more compactly:

dbk
( t0 ) = V b k ( t0 ) + O ( ε )
dt
We study the behaviour of bk (t) for small ε and t close to t0 , by studying the behaviour of
the solutions to the initial value problems

dbk
(t) = V bk (t) bk (t0 ) = ê(k) (IVP)
dt
To warm up, we first look at two two-dimensional examples. In both examples, the ve-
locity field v( x, y) is linear in ( x, y). Consequently, in these examples, v x0 + εê(k) , t0 ´


ř
(k)
v x0 , t0 is exactly 3j=1 ε Bx

Bv
x0 , t0 ê j and the solution to (IVP) coincides with the exact
j
bk (t). Following each example, we discuss a broad class of V ’s that generate behaviour
similar to that example.

Example 4.6.1 (v( x, y) = 2xı̂ı + 3y̂ )

In this example
 
2 0
V=
0 3

233
I NTEGRAL T HEOREMS 4.6 O PTIONAL — M ORE I NTERPRETATION OF D IV AND C URL

The solution to the initial value problem

b11 (t) = 2b1 (t) b1 (0) = β 1

 
β
b ( t ) = V b ( t ) b (0) = 1
1
or equivalently
β2 b12 (t) = 3b2 (t) b2 (0) = β 2

is
b1 (t) = e2t β 1 e2t 0
 
or equivalently b(t) = b (0)
b2 (t) = e3t β 2 0 e3t

If one chooses ê(1) = ı̂ı and ê(2) = ̂ , the edges, b1 (t) = e2t ê(1) and b2 (t) = e3t ê(2) , of the
chunk of fluid never change direction. But their lengths do change. The relative rate of
change of length per unit time, | db dt ( t )|/|bk ( t )|, is 2 for b1 and 3 for b3 . In the figure below,
k

the darker rectangle is the initial square. That is, the square with edges bk (t0 ) = ê(k) . The
lighter rectangle is that with edges bk (t) for some t a bit bigger than t0 .

Example 4.6.1
As time increases the initial cube becomes a larger and larger rectangle.
Example 4.6.2 (Example 4.6.1, generalized.)

The behaviour of Example 4.6.1 is typical of V ’s that are symmetric matrices, i.e. that
obey54 Vi,j = V j,i for all i, j. Any d ˆ d symmetric matrix55 (with real entries)
˝ has d real eigenvalues
˝ has d mutually orthogonal real unit eigenvectors.
Denote by λk , 1 ď k ď d, the eigenvalues of V and choose d mutually perpendicular real
unit vectors, ê(k) , that obey V ê(k) = λk ê(k) for all 1 ď k ď d. Then

bk (t) = eλk (t´t0 ) ê(k)

Bv
54 In terms of our original vector field, this condition is that BBvx i x0 , t0 = B x j x0 , t0 . So, in three dimen-



j i


sions, it comes down to the requirement that ∇ ˆ v be zero at the point x0 , t0 .
55 This was proven by the French mathematician and physicist Augustin-Louis Cauchy (1789–1857) in
1829.

234
I NTEGRAL T HEOREMS 4.6 O PTIONAL — M ORE I NTERPRETATION OF D IV AND C URL

obeys
dbk
(t) = λk eλk (t´t0 ) ê(k) = eλk (t´t0 ) V ê(k) = V bk (t) and bk (t0 ) = ê(k)
dt
So bk (t) = eλk (t´t0 ) ê(k) satisfies (IVP) for all t and 1 ď k ď d.
If we start, at time t0 , with a cube whose edges, ê(k) , are eigenvectors of V , then as time
progresses the edges, bk (t), of the chunk of fluid never change direction. But their lengths
change with the relative rate of change of length per unit time being λk for edge number
k. This rate of change may be positive (the edge grows with time) or negative (the edge
shrinks in time) depending on the sign of λk .
The volume of the chunk of fluid at time t is V (t) = eλ1 (t´t0 ) ¨ ¨ ¨ eλd (t´t0 ) . The relative
rate of change of volume per unit time is V 1 (t)/V (t) = λ1 ¨ ¨ ¨ + λd , the sum of the d eigen-
ř
values. The sum of the eigenvalues of any d ˆ d matrix V is given by its trace id=1 Vi,i .
For the matrix (M)
ÿ d
V 1 ( t0 ) Bvi
x0 , t0 = ∇ ¨ v x0 , t0



=
V ( t0 ) Bxi
i =1

So, at least when the matrix V defined in (M) is symmetric, the divergence ∇ ¨ v x0 , t0

gives the relative rate of change of volume per unit time for our tiny chunk of fluid at time
t0 and position x0 . Thus when ∇ ¨ v = 0 the volume is fixed. In particular, this is the case
when the fluid is incompressible.
Example 4.6.2
In fact we can relax the symmetry condition.
Example 4.6.3 (Example 4.6.1, generalized yet again.)

For any d ˆ d matrix V , the solution of

b1 ( t ) = V b ( t ) b ( t0 ) = e

is
b(t) = eV (t´t0 ) e
where the exponential of a d ˆ d matrix B is defined by the power series
1 3 ř
8 1
e B = 1 + B + 21 B2 + B +¨¨¨ = Bn
3! n =0 n!

with 1 denoting the d ˆ d identity matrix. This sum converges56 for all d ˆ d matrices
B. Furthermore it easy to check, using the power series, that eV (t´t0 ) obeys dt d V (t´t0 )
e =
Ve V ( t´t 0 ) and is the identity matrix when t = t0 . So b(t) = e V ( t´t 0 ) e really does obey
b (t) = V b(t) and b(t0 ) = e.
1

Pick any d vectors ek , 1 ď k ď d, and define bk (t) = eV (t´t0 ) ek . Also let E be the d ˆ d
matrix whose kth column is ek and E(t) be the d ˆ d matrix whose kth column is bk (t).

56 The proof is not so hard, though we’ll only outline it. Just denote by β the magnitude of the largest
matrix element of B. Then use the definition of the matrix product to prove that the largest matrix
element of Bn has magnitude at most (dβ)n .

235
I NTEGRAL T HEOREMS 4.6 O PTIONAL — M ORE I NTERPRETATION OF D IV AND C URL

Then the volume of the parallelepiped with edges ek , 1 ď k ď d, is V (t0 ) = det E and the
volume of the parallelepiped with edges bk (t), 1 ď k ď d, is

V (t) = det E(t) = det eV (t´t0 ) E = det eV (t´t0 ) det E = det eV (t´t0 ) V (t0 )

Of course now we have to compute the determinant of the exponential of a matrix. Luck-
ily, there is an easy way to do this. For any d ˆ d matrix B, det e B = etr B , so57
d
ÿ
(t´t0 ) tr V V 1 ( t0 )
V (t) = e V ( t0 ) ñ = tr V = Vi,i
V ( t0 )
i =1

So, for any
matrix V defined as in (M) and any choice of ê(k) , 1 ď k ď d, the divergence
∇ ¨ v x0 , t0 gives the relative rate of change of volume per unit time for our tiny chunk of
fluid at time t0 and position x0 .
Example 4.6.3

Example 4.6.4 (v( x, y) = ´yı̂ı + x̂ )

In this example
0 ´1
 
V=
1 0
The solution58 to
b11 (t) = ´b2 (t) b1 (0) = β 1
 
β
1
b (t) = V b(t) b (0) = 1 or equivalently
β2 b21 (t) = b1 (t) b2 (0) = β 2
is
b1 (t) = β 1 cos t ´ β 2 sin t cos t ´ sin t
 
or equivalently b(t) = b (0)
b2 (t) = β 1 sin t + β 2 cos t sin t cos t
Consequently the vector b(t) has the same length as b(0). The angle between b(t) and
b(0) is just t radians. So, in this example, no matter what direction vectors ê(k) we pick,
the chunk of fluid just rotates at one radian per unit time. In the figure below, the outlined
rectangle is the initial square. That is, the square with edges bk (t0 ) = ê(k) . The shaded
rectangle is that with edges bk (t) for some t a bit bigger than t0 .

57 Again, we won’t prove this. But for a diagonal matrix, it is easy — just compute both sides. So for a
diagonalizable matrix it is also easy — diagonalize.
58 You can find the solution either by guessing, or by using eigenvalues and eigenvectors.

236
I NTEGRAL T HEOREMS 4.6 O PTIONAL — M ORE I NTERPRETATION OF D IV AND C URL

Example 4.6.4

Example 4.6.5 (Example 4.6.4, generalized.)

The behaviour of Example 4.6.4 is typical of V ’s that are antisymmetric matrices, i.e. that
obey Vi,j = ´V j,i for all i, j. As we have already observed, for any d ˆ d matrix V , the
solution of b1 (t) = V b(t), b(0) = e is b(t) = eV t e. We now show that if V is a 3 ˆ 3
antisymmetric matrix, then eV t is a rotation.
Assuming that V is not the zero matrix (in which case eV t is the identity matrix for
all t), we can find a number Ω ą 0 and a unit vector k̂ = (k1 , k2 , k3 ) (not necessarily the
standard unit vector parallel to the z-axis) such that

´Ωk3 Ωk2
 
0
V =  Ωk3 0 ´Ωk1  (R)
´Ωk2 Ωk1 0

This is easy. Because

b V is antisymmetric, all of the entries on its diagonal must be zero.
Define Ω to be V1,2 + V1,3
2 2 + V 2 and k = ´V /Ω, k = V /Ω, k = ´V /Ω. Also,
2,3 1 2,3 2 1,3 3 1,2
let ı̂ı be any unit vector orthogonal to k̂ (again, not necessarily the standard one) and ̂ =
k̂ ˆ ı̂ı . So ı̂ı , ̂ , k̂ is a right-handed system of three mutually perpendicular unit vectors.
Observe that, for any vector e = (e1 , e2 , e3 )

´Ωk3 Ωk2 k 2 e3 ´ k 3 e2
    
0 e1
V e =  Ωk3 0 ´Ωk1  e2  = Ω k3 e1 ´ k1 e3  = Ωk̂ ˆ e
´Ωk2 Ωk1 0 e3 k 1 e2 ´ k 2 e1

In particular,

V ı̂ı = Ωk̂ ˆ ı̂ı = Ω̂ V ̂ = Ωk̂ ˆ ̂ = ´Ωı̂ı V k̂ = Ωk̂ ˆ k̂ = 0

V 2ı̂ı = ΩV ̂ = ´Ω2ı̂ı V 2 ̂ = ´ΩV ı̂ı = ´Ω2 ̂ V 2 k̂ = V 0 = 0
V 3ı̂ı = ΩV 2 ̂ = ´Ω3 ̂ V 3 ̂ = ´ΩV 2ı̂ı = Ω3ı̂ı V 3 k̂ = V 2 0 = 0
V 4ı̂ı = ΩV 3 ̂ = Ω4ı̂ı V 4 ̂ = ´ΩV 3ı̂ı = Ω4 ̂ V 4 k̂ = V 3 0 = 0

and so on. For all odd n ě 1,

V nı̂ı = (´1)(n´1)/2 Ωn ̂ V n ̂ = ´(´1)(n´1)/2 Ωnı̂ı V n k̂ = 0

and all even n ě 2,

V nı̂ı = (´1)n/2 Ωnı̂ı V n ̂ = (´1)n/2 Ωn ̂ V n k̂ = 0

237
I NTEGRAL T HEOREMS 4.6 O PTIONAL — M ORE I NTERPRETATION OF D IV AND C URL

Hence we can write

ř
8 1 ř (´1)n/2 ř (´1)(n´1)/2
eV tı̂ı = (V t)nı̂ı = (Ωt)nı̂ı + (Ωt)n ̂
n =0 n! n even n! n odd n!
= cos(Ωt) ı̂ı + sin(Ωt) ̂
ř
8 1 ř (´1)n/2 ř (´1)(n´1)/2
eV t ̂ = (V t)n ̂ = (Ωt)n ̂ ´ (Ωt)nı̂ı
n =0 n! n even n! n odd n!
= ´ sin(Ωt) ı̂ı + cos(Ωt) ̂
ř
8 1
eV t k̂ = (V t)n k̂ = k̂
n =0 n!

So eV t is rotation by an angle Ωt about the axis k̂.

Example 4.6.5

Example 4.6.6 (Example 4.6.5, continued.)

Whether or not the matrix V defined in (M) is antisymmetric, the related matrix with
entries
Ai,j = 12 Vi,j ´ V j,i

is. When V is antisymmetric, A and V coincide. The matrix A is (to write it out explicitly)
Bv1
Bv2
Bv1
Bv3
´ ´


0 x 0 , t 0 x 0 , t 0 x 0 , t 0 x 0 , t 0
1 Bv1
Bx2 Bx1 Bx3
Bx
Bv3
1
A = ´ Bx

Bv2 Bv2
´


x 0 , t 0 + Bx x 0 , t 0 0 Bx x 0 , t 0 Bx x 0 , t 0
2 Bv12 3 2


Bv31
Bv3
Bv2
´ Bx3 x0 , t0 + Bx1 x0 , t0 ´ Bx3 x0 , t0 + Bx2 x0 , t0



0

Comparing this with (R), we see that

Ωk̂ = 12 ∇ ˆ v x0 , t0

V defined
So, at least when the matrix ˇ in (M) is antisymmetric,
1ˇ
our tiny cube rotates about
ˇ
the axis with ∇ ˆ v x0 , t0 at rate 2 ∇ ˆ v x0 , t0 .

ˇ

Example 4.6.6

Remark:
In the generalization, Example 4.6.5, of Example 4.6.4, we only considered dimension 3. It
is a nice exercise in eigenvalues and eigenvectors to handle general dimension. Here are
the main facts about antisymmetric matrices with real entries that are used.

˝ All eigenvalues of antisymmetric matrices are either zero or pure imaginary.

˝ For antisymmetric matrices with real entries, the nonzero eigenvalues come in com-
plex conjugate pairs. The corresponding eigenvectors may also be chosen to be com-
plex conjugates.

Choose as basis vectors (like ı̂ı , ̂ , k̂ above)

238
I NTEGRAL T HEOREMS 4.6 O PTIONAL — M ORE I NTERPRETATION OF D IV AND C URL

˝ the eigenvectors of eigenvalue 0 (they act like k̂ above)

˝ the real and imaginary parts of each complex conjugate pair of eigenvectors (they
act like ı̂ı , ̂ above)
Resumé so far:
We have now seen that
• when the matrix V defined in (M) is symmetric and the direction vectors ê(k) of
the cube are eigenvectors of V , then, at time t0 , the chunk of fluid is not changing
orientation but is changing volume at instantaneous relative rate ∇ ¨ v x0 , t0 and

• when the matrix V defined in (M) is antisymmetric, then, at time t0 , the chunk of
fluid is not changing shape or size but is rotating about the axis ∇ ˆ v x0 , t0 at rate


1ˇ
ˇ
ˇ
2 ∇ ˆ v x0 , t0 . For this reason, ∇ ˆ v is often referred to as a “vorticity” meter.

ˇ

These agree with our earlier interpretations of divergence and curl.

The general case:

Now consider a general matrix V . It can always be written as the sum

V = S+A
of a symmetric matrix S and an antisymmetric matrix A. Just define

Si,j = 21 Vi,j + V j,i Ai,j = 12 Vi,j ´ V j,i

As we have already observed, the solution of

b1 ( t ) = V b ( t ) b (0) = e

is
b ( t ) = eV t e = e( A+S)t e
If S and A were ordinary numbers, we would have e( A+S)t = e At eSt . But for matrices this
need not be the case, unless S and A happen to commute59 . For arbitrary matrices, it is
still true that h in
e( A+S)t = lim e At/n eSt/n
nÑ8

This is called the Lie60 product formula. It shows that our tiny chunk of fluid mixes
together the behaviours of A and S, scaling a bit, then rotating a bit, then scaling a bit and
so on.
Example 4.6.7 ( v( x, y) = 2yı̂ı .)

In this example
     
0 2 0 1 0 1
V= = S+A with S= A=
0 0 1 0 ´1 0

59 By definition, the matrices S and A commute when AS = SA.

60 This formula is named after the Norwegian mathematician Marius Sophus Lie (1842–1899). In 1870, he
was arrested and held in prison in France for a month, because he was suspected of being a German
spy. His mathematics notes were thought to be top secret coded messages.

239
I NTEGRAL T HEOREMS 4.6 O PTIONAL — M ORE I NTERPRETATION OF D IV AND C URL

The solution to the full flow

 
β b11 (t) = 2b2 (t) b1 (0) = β 1
1
b (t) = V b(t) b (0) = 1 or equivalently
β2 b21 (t) = 0 b2 (0) = β 2

is  
b1 (t) = β 1 + 2β 2 t 1 2t
or equivalently b(t) = b (0)
b2 (t) = β 2 0 1
The solution to the S part of the flow
 
β b11 (t) = b2 (t) b1 (0) = β 1
1
b (t) = Sb(t) b (0) = 1 or equivalently
β2 b21 (t) = b1 (t) b2 (0) = β 2

is61
 
b1 (t) = β 1 cosh t + β 2 sinh t cosh t sinh t
or equivalently b(t) = b (0)
b2 (t) = β 1 sinh t + β 2 cosh t sinh t cosh t

The eigenvectors of S are

   
(1) 1 1 1 1
ê =? ê (2)
=?
2 1 2 ´1

The corresponding eigenvalues are +1 and ´1. The eigenvectors obey

   
St (1) cosh t sinh t (1) cosh t sinh t (2)
e ê = ê = et ê(1) St (2)
e ê = ê = e´t ê(2)
sinh t cosh t sinh t cosh t

Under the S part of the flow ê(1) scales by a factor of et , which is bigger than one for t ą 0
and ê(2) scales by a factor of e´t , which is smaller than one for t ą 0.
The solution to the A part of the flow
 
β b11 (t) = b2 (t) b1 (0) = β 1
1
b (t) = Ab(t) b (0) = 1 or equivalently
β2 b21 (t) = ´b1 (t) b2 (0) = β 2

is  
b1 (t) = β 1 cos t + β 2 sin t cos t sin t
or equivalently b(t) = b (0)
b2 (t) = ´β 1 sin t + β 2 cos t ´ sin t cos t
The A part of the flow rotates clockwise about the origin at one radian per unit time.
Here are some figures to help us visualize this. The first shows a square with edges
ê(1) , ê(2) and its image under the full flow t = 0.4 later. Under this full flow the vector
ê(k) Ñ e0.4V ê(k) . The second shows its image under 0.4 time units of the S-flow (that is,
ê(k) Ñ e0.4S ê(k) ). The third applies 0.4 time units of the A-flow to the shaded rectangle of
the middle figure. So the lightly shaded rectangle of the third figure has edges e0.4S ê(k)
and the darkly shaded rectangle has edges e0.4A e0.4S ê(k) .

1
et ´ e´t and cosh t = 1
e t + e ´t .



61 Recall that sinh t = 2 2

240
 I NTEGRAL T HEOREMS 4.7 O PTIONAL — A G ENERALIZED S TOKES ’ T HEOREM

êpkq and e0.4V êpkq êpkq and e0.4S êpkq e0.4S êpkq and e0.4A e0.4S êpkq

Of course e0.4A e0.4S ê(k) is not a very good approximation for e0.4( A+S) ê(k) . It is much better
n
to take e0.4A/n e0.4S/n ê(k) with n large. Each of the following figures shows two paral-


lelepipeds. In each, the shaded region has edges e0.4V ê(k) = e0.4( A+S) ê(k) and the outlined
n
region has edges e0.4A/n e0.4S/n ê(k) .


n“1 n“5 n “ 10

n
So we can see that, as n increases, e0.4A/n e0.4S/n ê(k) becomes a better and better approx-


imation to e0.4( A+S) ê(k) .

Example 4.6.7

4.7IJ Optional — A Generalized Stokes’ Theorem

As we have seen, the fundamental theorem of calculus, the divergence theorem, Greens’
theorem and Stokes’ theorem share a number of common features. There is in fact a single
framework which encompasses and generalizes all of them, and there is a single theorem
of which they are all special cases. We now give a bare bones introduction to this frame-
work and theorem. A proper treatment typically takes up a good part of a full course.
Here is an outline of what we shall do:
• First, we will define “differential forms”. To try and keep things as simple and con-
crete as possible, we’ll only define62 differential forms on R3 — all of our functions

62 In general, a differential form is defined on a “manifold”, which is an abstract generalization of a multi-

dimensional surface, like a sphere or a torus.

241
I NTEGRAL T HEOREMS 4.7 O PTIONAL — A G ENERALIZED S TOKES ’ T HEOREM

will be defined on R3 . Very roughly speaking, a k-form is what you write after the
integral sign of an integral over a k dimensional object. Here k is one of 0, 1, 2, 3.
As a example, a 1-form is an expression of the form F1 ( x, y, z) dx + F2 ( x, y, z) dy +
F3 ( x, y, z) dz. For k = 0, think of a point as a zero dimensional object and think of
evaluating a function at a point as “integrating the function over the point”.

• Then we will define some operations on differential forms, so that we can add them,
multiply them, differentiate them and, eventually, integrate them. The derivative of
a k-form ω is a (k + 1)-form that is denoted dω. It will turn out that

˝ differentiating a 0-form amounts to taking a gradient,

˝ differentiating a 1-form amounts to taking a curl, and
˝ differentiating a 2-form amounts to taking a divergence.

• Finally we will get to the generalized Stokes’ theorem which says that, if ω is a k-
form (with k = 0, 1, 2) and D is a (k + 1)-dimensional domain of integration, then
ż ż
dω = ω
D BD

It will turn out that

˝ when k = 0, this is just the fundamental theorem of calculus and

˝ when k = 1, this is both Green’s theorem and our Stokes’ theorem, and
˝ when k = 2, this is the divergence theorem.

Now let’s get to work. For simplicity, we will assume throughout this section that all
derivatives of all functions exist and are continuous. Our first task to define differential
forms.
As we said above we will define a 1-form as an expression of the form F1 ( x, y, z) dx +
F2 ( x, y, z) dy + F3 ( x, y, z) dz. When you learned the definition of the integral the symbol
“dx” was not given any mathematical meaning by itself. A meaning was given only to
ş şb
the collections of symbols “ f ( x ) dx” and “ a f ( x ) dx”. Later in this section, we will give
a meaning to dx. We will, in Definition 4.7.11, define a differentiation operator that we
will call d. Then dx will be that differentiation operator applied to the function f ( x ) = x.
However, until then we will have to treat dx and dy and dz just as symbols. Their sole
role in F1 ( x, y, z) dx + F2 ( x, y, z) dy + F3 ( x, y, z) dz is to allow us to distinguish63 F1 ( x, y, z),
F2 ( x, y, z) and F3 ( x, y, z).
Similarly, we will define a 2-form as an expression of the form F1 ( x, y, z) dy ^ dz +
F2 ( x, y, z) dz ^ dx + F3 ( x, y, z) dx ^ dy. Once again there is a symbol, namely “^”, that
we have not yet given a meaning to. We will, in Definition 4.7.3, define a product, called
the wedge product, with ^ as the multiplication symbol. Then dx ^ dy will be the wedge
product of dx and dy. Until then we will have to treat dy ^ dz, dz ^ dx and dx ^ dy just
as three more meaningless symbols.
Finally here is the definition.


63 We could also define, for example, a 1-form as an ordered list F1 ( x, y, z) , F2 ( x, y, z) , F3 ( x, y, z)
of three functions and just view
F1 ( x, y, z) dx + F2 ( x, y, z) dy + F3 ( x, y, z) dz as another notation for
F1 ( x, y, z) , F2 ( x, y, z) , F3 ( x, y, z) .

242
I NTEGRAL T HEOREMS 4.7 O PTIONAL — A G ENERALIZED S TOKES ’ T HEOREM

Definition 4.7.1.

(a) A 0-form is a function f ( x, y, z).

(b) A 1-form is an expression of the form

F1 ( x, y, z) dx + F2 ( x, y, z) dy + F3 ( x, y, z) dz

with F1 ( x, y, z), F2 ( x, y, z) and F3 ( x, y, z) being functions of three variables.

(c) A 2-form is an expression of the form

F1 ( x, y, z) dy ^ dz + F2 ( x, y, z) dz ^ dx + F3 ( x, y, z) dx ^ dy

with F1 ( x, y, z), F2 ( x, y, z) and F3 ( x, y, z) being functions of three variables.

(d) A 3-form is an expression of the form f ( x, y, z) dx ^ dy ^ dz, with f ( x, y, z)

being a function of three variables.

At this stage (there’ll more later), just think of “dx”, “dy”, “dz”, “dx ^ dy”, and
so on, as symbols. Do not yet attempt to attach any significance to them.

There are four operations involving differential forms — addition, multiplication (^),
differentiation (d) and integration. Here are their definitions. First, addition is defined,
and works, just the way that you would expect it to.

Definition 4.7.2 (Addition of differential forms).

(a) The sum of the 0-forms f and g is the 0-form f + g.

(b) The sum of two 1-forms is the 1-form

 
F1 dx + F2 dy + F3 dz
 
+ G1 dx + G2 dy + G3 dz
=( F1 + G1 ) dx + ( F2 + G2 ) dy + ( F3 + G3 ) dz

(c) The sum of two 2-forms is the 2-form

F1 dy ^ dz + F2 dz ^ dx + F3 dx ^ dy
 

+ G1 dy ^ dz + G2 dz ^ dx + G3 dx ^ dy
 

=( F1 + G1 ) dy ^ dz + ( F2 + G2 ) dz ^ dx + ( F3 + G3 ) dx ^ dy

(d) The sum of two 3-forms is the 3-form

f dx ^ dy ^ dz + g dx ^ dy ^ dz = f + g dx ^ dy ^ dz

243
I NTEGRAL T HEOREMS 4.7 O PTIONAL — A G ENERALIZED S TOKES ’ T HEOREM

There is one wrinkle in multiplication. It is not commutative, meaning that α ^ β need

not be the same as β ^ α. You have already seen some noncommutative products. If a and
b are two vectors in R3 , then a ˆ b = ´b ˆ a. Also, if A and B are two n ˆ n matrices, the
matrix product AB need not be the same as BA.

Definition 4.7.3 (Multiplication of differential forms).

We now define a multiplication rule for differential forms. If ω is a k-form and ω 1

is a k1 -form then the product will be a (k + k1 )-form and will be denoted ω ^ ω 1
(read “omega wedge omega prime”). It is determined by the following proper-
ties.

(a) If f is a function (i.e. a 0-form), then

 
f F1 dx + F2 dy + F3 dz = ( f F1 ) dx + ( f F2 ) dy + ( f F3 ) dz
f F1 dy ^ dz + F2 dz ^ dx + F3 dx ^ dy = ( f F1 ) dy ^ dz + ( f F2 ) dz ^ dx
 

+ ( f F3 ) dx ^ dy
f g dx ^ dy ^ dz = ( f g) dx ^ dy ^ dz
 

Traditionally, the ^ is not written when multiplying a differential form by a

function (i.e. a 0-form).

(b) ω ^ ω 1 is linear in ω and in ω 1 . This means that if ω = f 1 ω1 + f 2 ω2 , where

f 1 , f 2 are functions and ω1 , ω2 are forms, then

f 1 ω1 + f 2 ω2 ^ ω 1 = f 1 ( ω1 ^ ω 1 ) + f 2 ( ω2 ^ ω 1 )

Similarly,
ω ^ f 1 ω11 + f 2 ω21 = f 1 (ω ^ ω11 ) + f 2 (ω ^ ω21 )

(c) If ω is a k-form and ω 1 is a k1 -form then

1
ω ^ ω 1 = (´1)kk ω 1 ^ ω

That is, if at least one of k and k1 is even, then

ω ^ ω1 = ω1 ^ ω

(so that the wedge product is commutative) and if k and k1 are both odd then

ω ^ ω 1 = ´ω 1 ^ ω

(so that the wedge product is anticommutative). In particular, if ω is a d-form

with d odd
ω^ω = 0

244
I NTEGRAL T HEOREMS 4.7 O PTIONAL — A G ENERALIZED S TOKES ’ T HEOREM

Definition 4.7.3 (continued).

(d) The wedge product is associative. This means that

(ω ^ ω 1 ) ^ ω 2 = ω ^ ω 1 ^ ω 2

So the wedge product obeys most of the usual multiplication rules, with the one big ex-
ception that if ω is k-form and ω 1 is a k1 -form with k and k1 both odd then ω ^ ω 1 = ´ω 1 ^ ω.
The best way to get a handle on the wedge product is to work through some examples,
like these.
Example 4.7.4

Let ω = F1 dx + F2 dy + F3 dz and ω 1 = G1 dx + G2 dy + G3 dz be any two 1-forms. Their

product is

ω ^ ω 1 = F1 dx + F2 dy + F3 dz ^ G1 dx + G2 dy + G3 dz
   

= F1 dx ^ G1 dx + F1 dx ^ G2 dy + F1 dx ^ G3 dy

+ F2 dx ^ G1 dx + F2 dx ^ G2 dy + F2 dx ^ G3 dy

+ F3 dx ^ G1 dx + F3 dx ^ G2 dy + F3 dx ^ G3 dy

(by linearity, i.e. by part (b) of Definition 4.7.3)

= F1 G1 dx ^ dx + F1 G2 dx ^ dy + F1 G3 dx ^ dz
+ F2 G1 dy ^ dx + F2 G2 dy ^ dy + F2 G3 dy ^ dz
+ F3 G1 dz ^ dx + F3 G2 dz ^ dy + F3 G3 dz ^ dz
= F1 G2 ´ F2 G1 ) dx ^ dy + F3 G1 ´ F1 G3 ) dz ^ dx + F2 G3 ´ F3 G2 ) dy ^ dz

because
dx ^ dx = dy ^ dy = dz ^ dz = 0

and
dx ^ dy = ´dy ^ dx dx ^ dz = ´dz ^ dx dz ^ dy = ´dy ^ dz

Note that, if we view F = ( F1 , F2 , F3 ) and G = ( G1 , G2 , G3 ) as vectors, we can write the

product simply as

Equation 4.7.5.

F1 dx + F2 dy + F3 dz ^ G1 dx + G2 dy + G3 dz
   

= (F ˆ G)1 dy ^ dz + (F ˆ G)2 dz ^ dx + (F ˆ G)3 dx ^ dy

where we are using (F ˆ G)` to denote the `th component of the cross product F ˆ G. In
the special case that F3 = G3 = 0, we have

245
I NTEGRAL T HEOREMS 4.7 O PTIONAL — A G ENERALIZED S TOKES ’ T HEOREM

Equation 4.7.6.
 
F1 F2
F1 dx + F2 dy ^ G1 dx + G2 dy = F1 G2 ´ F2 G1 ) dx ^ dy = det dx ^ dy
   
G1 G2

We can now see why in the Definition 4.7.1.c of 2-forms

˝ there were no dx ^ dx or dy ^ dy or dz ^ dz terms — they are all zero and
˝ there were no dy ^ dx or dz ^ dy or dx ^ dz terms — they can all be rewritten using
dx ^ dy, dy ^ dz and dz ^ dx terms (or vice versa).
The reason that we chose to write the Definition 4.7.1.c as
F1 dy ^ dz + F2 dz ^ dx + F3 dx ^ dy
as opposed to in the form, for example,
f 1 dx ^ dy + f 2 dx ^ dz + f 3 dy ^ dz
was to make formulae like (4.7.5) work. The easy way to remember
F1 dy ^ dz + F2 dz ^ dx + F3 dx ^ dy
is to rename (in your head) x, y, z to x1 , x2 , x3 . Then the subscripts in the three terms of
F1 dx2 ^ dx3 + F2 dx3 ^ dx1 + F3 dx1 ^ dx2
are just 1, 2, 3 and 2, 3, 1 and 3, 1, 2 — the three cyclic permutations of 1, 2, 3.
Example 4.7.4

Example 4.7.7

The product of the (general) 1-form ω = F1 dx

 + F2 dy + F3 dz and the (general) 2-form
ω = G1 dy ^ dz + G2 dz ^ dx + G3 dx ^ dy (again note the numbering of the coeffi-
1


cients in the 2-form) is

ω ^ ω 1 = F1 dx + F2 dy + F3 dz ^ G1 dy ^ dz + G2 dz ^ dx + G3 dx ^ dy
   

= F1 G1 dx ^ dy ^ dz + F2 G2 dy ^ dz ^ dx + F3 G3 dz ^ dx ^ dy
= F1 G1 + F2 G2 + F3 G3 ) dx ^ dy ^ dz
Here we have used that, for 1-forms, α ^ β = ´β ^ α, so that
dy ^ dz ^ dx = ´dy ^ dx ^ dz = dx ^ dy ^ dz
dz ^ dx ^ dy = ´dx ^ dz ^ dy = dx ^ dy ^ dz
We have also used that any wedge product of three dtx or y or zu’s with at least two of the
coordinates being the same is zero. For example
dx ^ dz ^ dx = ´dx ^ dx ^ dz = 0
So

246
I NTEGRAL T HEOREMS 4.7 O PTIONAL — A G ENERALIZED S TOKES ’ T HEOREM

Equation 4.7.8.

F1 dx + F2 dy + F3 dz ^ G1 dy ^ dz + G2 dz ^ dx + G3 dx ^ dy
   

= F ¨ G dx ^ dy ^ dz

Example 4.7.7

Example 4.7.9

Combining Examples 4.7.4 and 4.7.7, we have the wedge product of any three (general)
1-forms F1 dx + F2 dy + F3 dz and G1 dx + G2 dy + G3 dz and H1 dx + H2 dy + H3 dz is

F1 dx + F2 dy + F3 dz ^ G1 dx + G2 dy + G3 dz ^ H1 dx + H2 dy + H3 dz
     

= F1 dx + F2 dy + F3 dz ^ (G ˆ H)1 dy ^ dz + (G ˆ H)2 dz ^ dx + (G ˆ H)3 dx ^ dy

   
(
= F1 (G ˆ H)1 + F2 (G ˆ H)2 + F3 (G ˆ H)3 dx ^ dy ^ dz
(
= F1 ( G2 H3 ´ G3 H2 ) + F2 ( G3 H1 ´ G1 H3 ) + F3 ( G1 H2 ´ G2 H1 dx ^ dy ^ dz

This can be expressed cleanly in terms of determinants. Recalling the rule for expanding
a determinant along its top row

Equation 4.7.10.

F1 dx + F2 dy + F3 dz ^ G1 dx + G2 dy + G3 dz ^ H1 dx + H2 dy + H3 dz
     
 
F1 F2 F3
= det  G1 G2 G3  dx ^ dy ^ dz
H1 H2 H3

Example 4.7.9

Our next operation is a differential operator which unifies and generalizes gradient,
curl and divergence.

247
I NTEGRAL T HEOREMS 4.7 O PTIONAL — A G ENERALIZED S TOKES ’ T HEOREM

Definition 4.7.11 (Differentiation of differential forms).

If ω is a k-form, then dω is a k + 1-form, with d being the unique64 such operator

that obeys

(a) d is linear. That is, if ω1 , ω2 are k-forms and a1 , a2 P R, then

d a1 ω1 + a2 ω2 = a1 dω1 + a2 dω2

(b) d obeys a “graded product rule”. Precisely, if ω (k) is a k-form and ω (`) is an
`-form, then

d ω (k) ^ ω (`) = dω (k) ^ ω (`) + (´1)k ω (k) ^ dω (`)

(c) If f ( x, y, z) is a 0-form, then

Bf Bf Bf
df = ( x, y, z) dx + ( x, y, z) dy + ( x, y, z) dz
Bx By Bz
= ∇ f ( x, y, z) ¨ dr where dr = dx ı̂ı + dy ̂ + dz k̂

(d) For any differential form ω,

d dω = 0

Example 4.7.12

(a) If f ( x, y, z) = x, then

Bx Bx Bx
df = ( x, y, z) dx + ( x, y, z) dy + ( x, y, z) dz = dx
Bx By Bz

That is, dx really is the operator d applied to the function x. Similarly, dy really is
the operator d applied to the function y and dz really is the operator d applied to the
function z.

(b) For any k-form ω

d ω ^ dx = dω ^ dx + (´1)k ω ^ d dx
 

= dω ^ dx

Similarly
d ω ^ dy = dω ^ dy d ω ^ dz = dω ^ dz
   

64 That d is unique just means that the action of d on any differential form is completely determined by
the four rules (a), (b), (c), (d). We will see in Example 4.7.12.b,c,d, that this is indeed the case.

248
I NTEGRAL T HEOREMS 4.7 O PTIONAL — A G ENERALIZED S TOKES ’ T HEOREM

(c) For any 1-form

d F1 dx + F2 dy + F3 dz = dF1 ^ dx + dF2 ^ dy + dF3 ^ dz

 
 BF BF BF   BF BF2 BF2 
1 2
= dx + 1 dy + 1 dz ^ dx + dx + dy + dz ^ dy
Bx By Bz Bx By Bz
 BF BF3 BF3 
3
+ dx + dy + dz ^ dz
Bx By Bz
 BF BF2   BF BF3   BF BF 
3 1 2
= ´ dy ^ dz + ´ dz ^ dx + ´ 1 dx ^ dy
By Bz Bz Bx Bx By
∇ ˆ F)1 dy ^ dz + (∇
= (∇ ∇ ˆ F)2 dz ^ dx + (∇ ∇ ˆ F)3 dx ^ dy

(d) For any 2-form

d F1 dy ^ dz + F2 dz ^ dx + F3 dx ^ dy
 

= dF1 ^ dy ^ dz + dF2 ^ dz ^ dx + dF3 ^ dx ^ dy


 BF BF BF 
1
= dx + 1 dy + 1 dz ^ dy ^ dz
Bx By Bz
 BF BF2 BF2 
2
+ dx + dy + dz ^ dz ^ dx
Bx By Bz
 BF BF3 BF3 
3
+ dx + dy + dz ^ dx ^ dy
Bx By Bz
 BF BF BF 
= 1
+ 2 + 3 dx ^ dy ^ dz
Bx By Bz
= ∇ ¨ F dx ^ dy ^ dz

(e) For any 3-form

 B f Bf Bf 
d f dx ^ dy ^ dz = dz ^ dx ^ dy ^ dz

dx + dy +
Bx By Bz
=0

Example 4.7.12

Example 4.7.13

In Definition 4.7.11.c, we defined, for any function f ( x, y, z) of three variables

Bf Bf Bf
df = ( x, y, z) dx + ( x, y, z) dy + ( x, y, z) dz
Bx By Bz

249
I NTEGRAL T HEOREMS 4.7 O PTIONAL — A G ENERALIZED S TOKES ’ T HEOREM

The analogous formulae65 for functions of one or two variables also apply.

df
d f (t) = (t) dt
dt
Bf Bf
d f (uv) = (u, v) du + (u, v) dv
Bu Bv

(a) Let F1 ( x, y, z) dx + F2 ( x, y, z) dy + F3 ( x, y, z) dz be a 1-form. Suppose that we substitute

x = x (t), y = y(t) and z = z(t), so that we are
restricting our 1-form to a parametrized
curve. Then, writing r(t) = x (t), y(t), z(t) ,




F1 x (t), y(t), z(t) dx (t) + F2 x (t), y(t), z(t) dy(t) + F3 x (t), y(t), z(t) dz(t)

dx
dy
dz
= F1 r(t) (t) dt + F2 r(t) (t) dt + F3 r(t) (t) dt
dt dt dt

dr
= F r(t) ¨ (t) dt
dt

(b) Let F1 ( x, y, z) dy ^ dz + F2 ( x, y, z) dz ^ dx + F3 ( x, y, z) dx ^ dy be a 2-form. Suppose

that we substitute x = x (u, v), y = y(u, v) and z = z(u, v), so that we are restricting

our 2-form to a parametrized surface. Then, writing r(u, v) = x (u, v), y(u, v), z(u, v) ,

F1 x (u, v), y(u, v), z(u, v) dy(u, v) ^ dz(u, v)

+ F2 x (u, v), y(u, v), z(u, v) dz(u, v) ^ dx (u, v)

+ F3 x (u, v), y(u, v), z(u, v) dx (u, v) ^ dy(u, v)

 By By   Bz Bz 
= F1 r(u, v) du + dv ^ du + dv
Bu Bv Bu Bv

 Bz Bz   Bx Bx 
+ F2 r(u, v) du + dv ^ du + dv
Bu Bv Bu Bv

 Bx Bx   By By 
+ F3 r(u, v) du + dv ^ du + dv
Bu Bv Bu Bv
h
By Bz By Bz
 
Bz Bx Bz Bx 

= F1 r(u, v) ´ + F2 r(u, v) ´
Bu Bv Bv Bu Bu Bv Bv Bu

 Bx By Bx By i
+ F3 r(u, v) ´ du ^ dv
Bu Bv Bv Bu
h
Br Br i
= F r(u, v) ¨ (u, v) ˆ (u, v) du ^ dv
Bu Bv

Example 4.7.13

Let us summarize what we have seen in the Example 4.7.12.

65 Indeed, you can view f (t) as a function of three variables that happens to be independent of two of
the three variables. Similarly you can view f (u, v) as a function of three variables that happens to be
independent of one of the three variables.

250
I NTEGRAL T HEOREMS 4.7 O PTIONAL — A G ENERALIZED S TOKES ’ T HEOREM

Lemma 4.7.14.

(a) For any 0-form

d f = ∇ f ( x, y, z) ¨ dr

(b) For any 1-form

 
d F1 dx + F2 dy + F3 dz
∇ ˆ F)1 dy ^ dz + (∇
= (∇ ∇ ˆ F)2 dz ^ dx + (∇
∇ ˆ F)3 dx ^ dy

(c) For any 2-form

d F1 dy ^ dz + F2 dz ^ dx + F3 dx ^ dy = ∇ ¨ F dx ^ dy ^ dz
 

(d) For any 3-form

d f dx ^ dy ^ dz = 0
 

Our final operation is integration of differential forms.

Definition 4.7.15 (Integration of differential forms).

(a) Let f ( x, y, z) be a 0-form and P = ( x0 , y0 , z0 ) P R3 be a point. Then

ż


f = f x0 , y0 , z0
P

More generally if, for each 1 ď i ď `, Pi = ( xi , yi , zi ) P R3 is a point and ni is

an integer, then
ż ÿ̀

f = ni f xi , yi , zi
Σi`=1 ni Pi i =1

(b) Let ω = F(r) ¨ dr = F1 ( x, y, z) dx + F2 ( x, y, z) dy + F3 ( x, y, z) dz
be a 1-form.

Let C be a curve that is parametrized by r(t) = x (t) , y(t) , z(t) , a ď t ď b.
Then, motivated by Example 4.7.13.a above,
ż żb ż

dr
ω= F r(t) ¨ (t) dt = F ¨ dr
C a dt C

251
I NTEGRAL T HEOREMS 4.7 O PTIONAL — A G ENERALIZED S TOKES ’ T HEOREM

Definition 4.7.15 (continued).

(c) Let ω = F1 ( x, y, z) dy ^ dz + F2 ( x, y, z) dz ^ dx + F3 ( x, y, z) dx ^ dy be a
2-form. Let S be an oriented
surface that is parametrized by r(u, v) =
x (u, v) , y(u, v) , z(u, v) , with (u, v) running over a region R in the uv-
plane. Assume that r(u, v) is orientation preserving in the sense that n̂ dS =
Br
+ Bu ˆ BvBr
du dv.Then, motivated by Example 4.7.13.b above,
ż ij h ij

Br Br i
ω= F r(u, v) ¨ (u, v) ˆ (u, v) du ^ dv = F ¨ n̂ dS
S Bu Bv
R S

(d) Let ω = f ( x, y, z) dx ^ dy ^ dz be a 3-form. Let V be a solid in R3 . Then

ż ¡
ω= f ( x, y, z) dxdydz
V
V

Finally, after all of these definitions, we have a very compact theorem that simulta-
neously covers the fundamental theorem of calculus, Green’s theorem. Stokes’ theorem
and the divergence theorem. Had we given all of our definitions in n dimensions, rather
than just three dimensions, it would cover a lot more. This general theorem is also called
Stokes’ theorem.
Theorem 4.7.16 (Stokes’ Theorem).

If ω is a k-form (with k = 0, 1, 2) and D is a (k + 1)-dimensional domain of inte-

gration, then ż ż
dω = ω
D BD
Here BD is the boundary of D (suitably oriented).

To see the connection between the general Stokes’ Theorem 4.7.16 and the Stokes’ and
divergence theorems of the earlier part of this chapter, here are the k = 1 and k = 2 cases
of Theorem 4.7.16 again.
• Let ω = F1 dx + F2 dy + F3 dz be a 1-form and let S be a piecewise smooth oriented
surface as in (our original) Stokes’ theorem 4.4.1. Then, by Lemma 4.7.14.b,
∇ ˆ F)1 dy ^ dz + (∇
dω = (∇ ∇ ˆ F)2 dz ^ dx + (∇
∇ ˆ F)3 dx ^ dy
So, by parts
ş ş with F replaced by ∇ ˆ F) and (b) of Definition 4.7.15, the con-
(c) (but
clusion D dω = BD ω of (the general) Stokes’ Theorem 4.7.16 is
ij ż ż ż
∇ ˆ F ¨ n̂ dS = dω = ω= F ¨ dr
S BS BS
S

which is the conclusion of (our original) Stokes’ theorem 4.4.1.

252
I NTEGRAL T HEOREMS 4.7 O PTIONAL — A G ENERALIZED S TOKES ’ T HEOREM

• ω = F1 ( x, y, z) dy ^ dz + F2 ( x, y, z) dz ^ dx + F3 ( x, y, z) dx ^ dy be a 2-form and let

V be a solid as in the divergence theorem 4.2.2. Then, by Lemma 4.7.14.c,

dω = ∇ ¨ F dx ^ dy ^ dz
ş
şSo, by parts (d) (with f = ∇ ¨ F) and (c) of Definition 4.7.15, the conclusion D dω =
BD ω of (the general) Stokes’ Theorem 4.7.16 is
¡ ż ż ij
∇ ¨ F dxdydz = dω = ω= F ¨ n̂ dS
V BV
V BV

which is the conclusion of the divergence theorem 4.2.2.

253
 Appendix A

T RIGONOMETRY

A.1IJ Trigonometry — Graphs

sin θ cos θ tan θ

1 1

´π ´ π2 π π 3π 2π ´π ´ π2 π π 3π 2π ´π ´ π2 π π 3π 2π
2 2 2 2 2 2

´1 ´1

A.2IJ Trigonometry — Special Triangles

254
 T RIGONOMETRY A.3 T RIGONOMETRY — S IMPLE I DENTITIES

From the above pair of special triangles we have

?
π 1 π 1 π 3
sin = ? sin = sin =
4 2 6 2 3 2
?
π 1 π 3 π 1
cos = ? cos = cos =
4 2 6 2 3 2
π π 1 π ?
tan = 1 tan = ? tan = 3
4 6 3 3

A.3IJ Trigonometry — Simple Identities

• Periodicity

sin(θ + 2π ) = sin(θ ) cos(θ + 2π ) = cos(θ )

• Reflection

sin(´θ ) = ´ sin(θ ) cos(´θ ) = cos(θ )

• Reflection around π/4

´ θ = cos θ ´ θ = sin θ
π

π


sin 2 cos 2

• Reflection around π/2

sin (π ´ θ ) = sin θ cos (π ´ θ ) = ´ cos θ

• Rotation by π

sin (θ + π ) = ´ sin θ cos (θ + π ) = ´ cos θ

• Pythagoras

sin2 θ + cos2 θ = 1
tan2 θ + 1 = sec2 θ
1 + cot2 θ = csc2 θ

• sin and cos building blocks

sin θ 1 1 cos θ 1
tan θ = csc θ = sec θ = cot θ = =
cos θ sin θ cos θ sin θ tan θ

255
 T RIGONOMETRY A.4 T RIGONOMETRY — A DD AND S UBTRACT A NGLES

A.4IJ Trigonometry — Add and Subtract Angles

• Sine
sin(α ˘ β) = sin(α) cos( β) ˘ cos(α) sin( β)

• Cosine
cos(α ˘ β) = cos(α) cos( β) ¯ sin(α) sin( β)

• Tangent
tan α + tan β
tan(α + β) =
1 ´ tan α tan β
tan α ´ tan β
tan(α ´ β) =
1 + tan α tan β
• Double angle
sin(2θ ) = 2 sin(θ ) cos(θ )
cos(2θ ) = cos2 (θ ) ´ sin2 (θ )
= 2 cos2 (θ ) ´ 1
= 1 ´ 2 sin2 (θ )
2 tan(θ )
tan(2θ ) =
1 ´ tan2 θ
1 + cos(2θ )
cos2 θ =
2
1 ´ cos (2θ )
sin2 θ =
2
1 ´ cos (2θ )
tan2 θ =
1 + cos(2θ )
• Products to sums
sin(α + β) + sin(α ´ β)
sin(α) cos( β) =
2
cos(α ´ β) ´ cos(α + β)
sin(α) sin( β) =
2
cos(α ´ β) + cos(α + β)
cos(α) cos( β) =
2
• Sums to products
α+β α´β
sin α + sin β = 2 sin cos
2 2
α+β α´β
sin α ´ sin β = 2 cos sin
2 2
α+β α´β
cos α + cos β = 2 cos cos
2 2
α+β α´β
cos α ´ cos β = ´2 sin sin
2 2

256
 T RIGONOMETRY A.5 I NVERSE T RIGONOMETRIC F UNCTIONS

A.5IJ Inverse Trigonometric Functions

arcsin x arccos x arctan x

Domain: ´1 ď x ď 1 Domain: ´1 ď x ď 1 Domain: all real numbers
Range: ´ π2 ď arcsin x ď π
2 Range: 0 ď arccos x ď π Range: ´ π2 ă arctan x ă π
2

π/2
π π
2

π/2
´1 1

´π/2 ´ π2
´1 1

Since these functions are inverses of each other we have

π π
arcsin(sin θ ) = θ ´ ďθď
2 2
arccos(cos θ ) = θ 0ďθďπ
π π
arctan(tan θ ) = θ ´ ďθď
2 2

and also

sin(arcsin x ) = x ´1 ď x ď 1
cos(arccos x ) = x ´1 ď x ď 1
tan(arctan x ) = x any real x

arccsc x arcsec x arccot x

Domain: |x| ě 1 Domain: |x| ě 1 Domain: all real numbers
Range: ´ π2 ď arccsc x ď π
2 Range: 0 ď arcsec x ď π Range: 0 ă arccot x ă π
arccsc x ‰ 0 arcsec x ‰ π2
π π π
2

π π
2 2
´1 1

´ π2
´1 1

257
T RIGONOMETRY A.5 I NVERSE T RIGONOMETRIC F UNCTIONS

Again
π π
arccsc(csc θ ) = θ ´ ďθď , θ‰0
2 2
π
arcsec(sec θ ) = θ 0 ď θ ď π, θ ‰
2
arccot(cot θ ) = θ 0ăθăπ

and

csc(arccsc x ) = x |x| ě 1
sec(arcsec x ) = x |x| ě 1
cot(arccot x ) = x any real x

258
 Appendix B

P OWERS AND L OGARITHMS

B.1IJ Powers
In the following, x and y are arbitrary real numbers, q is an arbitrary constant that is
strictly bigger than zero and e is 2.7182818284, to ten decimal places.

• e0 = 1, q0 = 1

ex qx
• e x +y = e x ey , e x´y = ey , q x +y = q x qy , q x´y = qy

1 1
• e´x = ex , q´x = qx

y
y
• ex = e xy , qx = q xy

d x d g( x ) d x
• dx e = ex , dx e = g1 ( x ) e g ( x ) , dx q = (ln q) q x
ş ş
• e x dx = e x + C, e ax dx = 1a e ax + C if a ‰ 0

ř
8
xn
• ex = n!
n =0

• lim e x = 8, lim e x = 0
xÑ8 xÑ´8

lim q x = 8, lim q x = 0 if q ą 1
xÑ8 xÑ´8

lim q x = 0, lim q x = 8 if 0 ă q ă 1
xÑ8 xÑ´8

• The graph of 2x is given below. The graph of q x , for any q ą 1, is similar.

259
 P OWERS AND L OGARITHMS B.2 L OGARITHMS

y y = 2x
6

2
1
x
−3 −2 −1 1 2 3

B.2IJ Logarithms
In the following, x and y are arbitrary real numbers that are strictly bigger than 0 (except
where otherwise specified), p and q are arbitrary constants that are strictly bigger than
one, and e is 2.7182818284, to ten decimal places. The notation ln x means loge x. Some
people use log x to mean log10 x, others use it to mean loge x and still others use it to mean
log2 x.

log x
• eln x = x, q q = x

• ln e x = x, logq q x = x for all ´8 ă x ă 8

ln x log p x log p x
• logq x = ln q , ln x = log p e , logq x = log p q

• ln 1 = 0, ln e = 1
logq 1 = 0, logq q = 1

• ln( xy) = ln x + ln y, logq ( xy) = logq x + logq y

x x
= ln x ´ ln y, logq = logq x ´ logq y



• ln y y

1 1
= ´ ln y, logq = ´ logq y



• ln y y

• ln( x y ) = y ln x, logq ( x y ) = y logq x

d
• dx ln x = 1x , d
dx logq x = 1
x ln q
ş ş
• ln x dx = x ln x ´ x + C, logq x dx = x logq x ´ lnxq + C,

• lim ln x = 8, lim ln x = ´8
xÑ8 xÑ0
lim logq x = 8, lim logq x = ´8
xÑ8 xÑ0

• The graph of log10 x is given below. The graph of logq x, for any q ą 1, is similar.

260
P OWERS AND L OGARITHMS B.2 L OGARITHMS

y
y = log10 x
1.0

0.5

x
1 5 10 15
−0.5

−1.0

261
 Appendix C

TABLE OF D ERIVATIVES

Throughout this table, a and b are constants, independent of x.

dF
F(x) F1 ( x ) = dx

a f ( x ) + bg( x ) a f 1 ( x ) + bg1 ( x )

f ( x ) + g( x ) f 1 ( x ) + g1 ( x )

f ( x ) ´ g( x ) f 1 ( x ) ´ g1 ( x )

a f (x) a f 1 (x)

f ( x ) g( x ) f 1 ( x ) g ( x ) + f ( x ) g1 ( x )

f ( x ) g( x )h( x ) f 1 ( x ) g ( x ) h ( x ) + f ( x ) g1 ( x ) h ( x ) + f ( x ) g ( x ) h1 ( x )

f (x) f 1 ( x ) g ( x )´ f ( x ) g1 ( x )
g( x ) g ( x )2

1 g1 ( x )
g( x )
´ g ( x )2

f g( x ) f 1 g ( x ) g1 ( x )

262
TABLE OF D ERIVATIVES

dF
F(x) F1 ( x ) = dx

a 0
dF
F(x) F1 ( x ) = dx
xa ax a´1
1
ln x x
g( x )a ag( x ) a´1 g1 ( x )
g1 ( x )
ln g( x ) g( x )
sin x cos x
1
loga x x ln a
sin g( x ) g1 ( x ) cos g( x )
arcsin x ?1
1´x2
cos x ´ sin x
1
arcsin g( x ) ? g (x)
1´g( x )2
cos g( x ) ´g1 ( x ) sin g( x )
arccos x ´? 1
1´x2
tan x sec2 x
1
arctan x 1+ x 2
csc x ´ csc x cot x
g1 ( x )
arctan g( x ) 1+ g ( x )2
sec x sec x tan x
arccsc x ´ ?1 2
|x| x ´1
cot x ´ csc2 x
arcsec x ?1
|x| x2 ´1
ex ex
arccot x ´ 1+1x2
e g( x ) g1 ( x ) e g ( x )

ax (ln a) a x

263
 Appendix D

TABLE OF I NTEGRALS

Throughout this table, a and b are given constants, independent of x and C is an arbitrary
constant.

ş
f (x) F(x) = f ( x ) dx

ş ş
a f ( x ) + bg( x ) a f ( x ) dx + b g( x ) dx + C

ş ş
f ( x ) + g( x ) f ( x ) dx + g( x ) dx + C

ş ş
f ( x ) ´ g( x ) f ( x ) dx ´ g( x ) dx + C

ş
a f (x) a f ( x ) dx + C

ş
u ( x ) v1 ( x ) u( x )v( x ) ´ u1 ( x )v( x ) dx + C

ş
f y ( x ) y1 ( x ) F y( x ) where F (y) = f (y) dy

a ax + C

x a +1
xa a +1 + C if a ‰ ´1

1
x ln |x| + C

g ( x ) a +1
g ( x ) a g1 ( x ) a +1 + C if a ‰ ´1

264
TABLE OF I NTEGRALS

ş
f (x) F(x) = f ( x ) dx
ş
f (x) F(x) = f ( x ) dx
ex ex + C
sin x ´ cos x + C
e g ( x ) g1 ( x ) e g( x ) + C
g1 ( x ) sin g( x ) ´ cos g( x ) + C
1
e ax a e ax + C
cos x sin x + C
1
ax ln a ax + C
tan x ln | sec x| + C
ln x x ln x ´ x + C
csc x ln | csc x ´ cot x| + C
?1 arcsin x + C
1´x2
sec x ln | sec x + tan x| + C
1
? g (x) arcsin g( x ) + C
1´g( x )2
cot x ln | sin x| + C
? 1
arcsin xa + C
a2 ´x2
sec2 x tan x + C
1
1+ x 2
arctan x + C
csc2 x ´ cot x + C
g1 ( x )
1+ g ( x )2
arctan g( x ) + C
sec x tan x sec x + C
1
a2 + x 2
1
a arctan xa + C
csc x cot x ´ csc x + C
?1 arcsec x + C (x ą 1)
x x2 ´1

265
 Appendix E

TABLE OF TAYLOR E XPANSIONS

Let n ě be an integer. Then if the function f has n + 1 derivatives on an interval that

contains both x0 and x, we have the Taylor expansion
1 2 1
f ( x ) = f ( x0 ) + f 1 ( x0 ) ( x ´ x0 ) + f ( x0 ) ( x ´ x0 )2 + ¨ ¨ ¨ + f ( n ) ( x0 ) ( x ´ x0 ) n
2! n!
1
+ f ( n +1) ( c ) ( x ´ x 0 ) n +1 for some c between x0 and x
( n + 1) !
The limit as n Ñ 8 gives the Taylor series
ÿ8
f ( n ) ( x0 )
f (x) = ( x ´ x0 ) n
n =0
n!
for f . When x0 = 0 this is also called the Maclaurin series for f . Here are Taylor series
expansions of some important functions.
ÿ8
1 n
x
e = x for ´8 ă x ă 8
n =0
n!
1 1 1
= 1 + x + x2 + x3 + ¨ ¨ ¨ + x n + ¨ ¨ ¨
2 3! n!
ÿ8
(´1) n
sin x = x2n+1 for ´8 ă x ă 8
n =0
(2n + 1)!
1 3 1 (´1)n 2n+1
= x´ x + x5 ´ ¨ ¨ ¨ + x +¨¨¨
3! 5! (2n + 1)!
ÿ8
(´1)n 2n
cos x = x for ´8 ă x ă 8
n =0
(2n)!
1 2 1 (´1)n 2n
= 1´ x + x4 ´ ¨ ¨ ¨ + x +¨¨¨
2! 4! (2n)!
1 ÿ8
= xn for ´1 ď x ă 1
1´x n =0
= 1 + x + x2 + x3 + ¨ ¨ ¨ + x n + ¨ ¨ ¨

266
TABLE OF TAYLOR E XPANSIONS

1 ÿ8
= (´1)n x n for ´1 ă x ď 1
1+x n =0
= 1 ´ x + x2 ´ x3 + ¨ ¨ ¨ + (´1)n x n + ¨ ¨ ¨
ÿ
8
1 n
ln(1 ´ x ) = ´ x for ´1 ď x ă 1
n
n =1
1 1
= ´x ´ 21 x2 ´ x3 ´ ¨ ¨ ¨ ´ x n ´ ¨ ¨ ¨
3 n
ÿ (´1)
8 n
ln(1 + x ) = ´ xn for ´1 ă x ď 1
n
n =1

1 2 1 3 (´1)n n
= x´ 2x + x ´¨¨¨´ x ´¨¨¨
3 n
p( p ´ 1) 2 p( p ´ 1)( p ´ 2) 3
(1 + x ) p = 1 + px + x + x +¨¨¨
2 3!
p( p ´ 1)( p ´ 2) ¨ ¨ ¨ ( p ´ n + 1) n
+ x +¨¨¨
n!

267
 Appendix F

3 D C OORDINATE S YSTEMS

F.1IJ Cartesian Coordinates

Here is a figure showing the definitions of the three Cartesian coordinates ( x, y, z)

px, y, zq

z
y
x
y px, y, 0q
x

and here are three figures showing a surface of constant x, a surface of constant x, and a
surface of constant z.

z z z
p0, 0, zq

y y y
px, 0, 0q p0, y, 0q
x x x
surface of constant x surface of constant y surface of constant z
(a plane) (a plane) (a plane)

Finally here is a figure showing the volume element dV in cartesian coordinates.

268
 3 D C OORDINATE S YSTEMS F.2 C YLINDRICAL C OORDINATES

dz

dx
dy
volume element dV “ dx dy dz

F.2IJ Cylindrical Coordinates

Here is a figure showing the definitions of the three cylindrical coordinates
r = distance from (0, 0, 0) to ( x, y, 0)
θ = angle between the the x axis and the line joining ( x, y, 0) to (0, 0, 0)
z = signed distance from ( x, y, z) to the xy-plane

px, y, zq

z
y
θ r
x px, y, 0q

The cartesian and cylindrical coordinates are related by

x = r cos θ y = r sin θ z=z
b
y
r = x 2 + y2 θ = arctan z=z
x
Here are three figures showing a surface of constant r, a surface of constant θ, and a surface
of constant z.

z z
z
p0, 0, zq

r y y y
θ
x x x
surface of constant r surface of constant θ surface of constant z
(a cylindrical shell) (a plane) (a plane)

269
 3 D C OORDINATE S YSTEMS F.3 S PHERICAL C OORDINATES

Finally here is a figure showing the volume element dV in cylindrical coordinates.

r dr
dz

r dθ
volume element dV “ r dr dθ dz

F.3IJ Spherical Coordinates

Here is a figure showing the definitions of the three spherical coordinates
ρ = distance from (0, 0, 0) to ( x, y, z)
ϕ = angle between the z axis and the line joining ( x, y, z) to (0, 0, 0)
θ = angle between the x axis and the line joining ( x, y, 0) to (0, 0, 0)

z
p0, 0, zq ρ sin ϕ
px, y, zq
ρ cos ϕ
ϕ ρ

y
px, 0, 0q θ
px, y, 0q
ρ sin ϕ sin θ
x

and here are two more figures giving the side and top views of the previous figure.

z
ρ sin ϕ px, y, zq y
p0, 0, zq
θ ρ sin ϕ
ρ cos ϕ
ρ ρ sin ϕ cos θ
ϕ px, y, 0q
px, 0, 0q
ρ sin ϕ sin θ
px, y, 0q
x
side view top view

The cartesian and spherical coordinates are related by

x = ρ sin ϕ cos θ y = ρ sin ϕ sin θ z = ρ cos ϕ
b a
y x 2 + y2
ρ = x 2 + y2 + z2 θ = arctan ϕ = arctan
x z

270
3 D C OORDINATE S YSTEMS F.3 S PHERICAL C OORDINATES

Here are three figures showing a surface of constant ρ, a surface of constant θ, and a
surface of constant ϕ.

z z z

ϕ
ρ
y y y
θ
x x x
surface of constant ρ surface of constant θ surface of constant ϕ
(a sphere) (a plane) (a cone)

Here is a figure showing the surface element dS in spherical coordinates

ρ sin ϕ dθ

ρ dϕ

x surface element dS “ ρ2 sin ϕ dθ dϕ

and two extracts of the above figure to make it easier to see how the factors ρ dϕ and
ρ sin ϕ dθ arise.

271
3 D C OORDINATE S YSTEMS F.3 S PHERICAL C OORDINATES

z z

ρ sin ϕ dθ

dθ
ρ sin ϕ
ρ dϕ

ϕ dϕ ϕ ρ
y y

x x

Finally, here is a figure showing the volume element dV in spherical coordinates

ρ sin ϕ dθ

ρ dϕ

dρ
x
volume element dV “ ρ2 sin ϕ dρ dθ dϕ

272
 Appendix G

C ONIC S ECTIONS AND

Q UADRIC S URFACES

A conic section is the curve of intersection of a cone and a plane that does not pass through
the vertex of the cone. This is illustrated in the figures below. An equivalent1 (and often

circle ellipse parabola hyperbola

used) definition is that a conic section is the set of all points in the xy-plane that obey
Q( x, y) = 0 with
Q( x, y) = Ax2 + By2 + Cxy + Dx + Ey + F = 0
being a polynomial of degree two2 . By rotating and translating our coordinate system the
equation of the conic section can be brought into one of the forms3
• αx2 + βy2 = γ with α, β, γ ą 0, which is an ellipse (or a circle),
• αx2 ´ βy2 = γ with α, β ą 0, γ ‰ 0, which is a hyperbola,
• x2 = δy, with δ ‰ 0 which is a parabola.

1 It is outside our scope to prove this equivalence.

2 Technically, we should also require that the constants A, B, C, D, E, F, are real numbers, that A, B, C are
not all zero, that Q( x, y) = 0 has more than one real solution, and that the polynomial can’t be factored
into the product of two polynomials of degree one.
3 This statement can be justified using a linear algebra eigenvalue/eigenvector analysis. It is beyond
what we can cover here, but is not too difficult for a standard linear algeba course.

273
C ONIC S ECTIONS AND Q UADRIC S URFACES

The three dimensional analogs of conic sections, surfaces in three dimensions given by
quadratic equations, are called quadrics. An example is the sphere x2 + y2 + z2 = 1. Here
are some tables giving all of the quadric surfaces.

elliptic parabolic hyperbolic

name sphere
cylinder cylinder cylinder
equation in x2 y2 x2 y2
a2
+ b2
=1 y = ax2 a2
´ b2
=1 x 2 + y2 + z2 = r 2
standard form
x = constant two lines one line two lines circle
cross-section
y = constant two lines two lines two lines circle
cross-section
z = constant ellipse parabola hyperbola circle
cross-section

sketch

elliptic elliptic
name ellipsoid
paraboloid cone
equation in x2
2 2 x2 y2 x2 y2 z2
a2
+ yb2 + zc2 = 1 a2
+ b2
= z
c a2
+ b2
= c2
standard form
x = constant two lines if x = 0
ellipse parabola
cross-section hyperbola if x ‰ 0
y = constant two lines if y = 0
ellipse parabola
cross-section hyperbola if y ‰ 0
z = constant ellipse ellipse ellipse
cross-section

sketch

274
C ONIC S ECTIONS AND Q UADRIC S URFACES

hyperboloid hyperboloid hyperbolic

name
of one sheet of two sheets paraboloid
equation in x2 y2 z2 x2
2 2 y2 x2
a2
+ b2
´ c2
=1 a2
+ yb2 ´ zc2 = ´1 b2
´ a2
= z
c
standard form
x = constant hyperbola hyperbola parabola
cross-section
y = constant hyperbola hyperbola ellipse
cross-section
z = constant two lines if z = 0
ellipse ellipse
cross-section hyperbola if z ‰ 0

sketch

275
Appendix H

R EVIEW OF L INEAR O RDINARY

D IFFERENTIAL E QUATIONS

Definition H.1.

(a) A differential equation is an equation for an unknown function that contains

the derivatives of that unknown function. For example y2 (t) + y(t) = 0 is a
differential equation for the unknown function y(t).

(b) A differential equation is called an ordinary differential equation (often short-

ened to “ODE”) if only ordinary derivatives appear. That is, if the un-
known function has only a single independent variable. A differential equa-
tion is called a partial differential equation (often shortened to “PDE”) if par-
tial derivatives appear. That is, if the unknown function has more than
one independent variable. For example y2 (t) + y(t) = 0 is an ODE while
B2 u 2 B 2 u ( x, t ) is a PDE.
Bt 2 ( x, t ) = c B x2

(c) The order of a differential equation is the order of the highest derivative that
appears. For example y2 (t) + y(t) = 0 is a second order ODE.

(d) An ordinary differential equation that is of the form

a0 (t)y(n) (t) + a1 (t)y(n´1) (t) + ¨ ¨ ¨ + an´1 (t)y1 (t) + an (t)y(t) = F (t) (H.1)

with given coefficient functions a0 (t), ¨ ¨ ¨ , an (t) and F (t) is said to be linear.
Otherwise, the ODE is said to be nonlinear. For example, y1 (t)2 + y(t) = 0,
y1 (t)y2 (t) + y(t) = 0 and y1 (t) = ey(t) are all nonlinear.

(e) The ODE (H.1) is said to have constant coefficients if the coefficients a0 (t), a1 (t),
¨ ¨ ¨ , an (t) are all constants. Otherwise, it is said to have variable coefficients. For
example, the ODE y2 (t) + 7y(t) = sin t is constant coefficient, while y2 (t) +
ty(t) = sin t is variable coefficient.

276
R EVIEW OF L INEAR O RDINARY D IFFERENTIAL E QUATIONS

Definition H.1 (continued).

(f) The ODE (H.1) is said to be homogeneous if F (t) is identically zero. Other-
wise, it is said to be inhomogeneous or nonhomogeneous. For example, the ODE
y2 (t) + 7y(t) = 0 is homogeneous, while y2 (t) + 7y(t) = sin t is inhomoge-
neous. A homogeneous ODE always has the trivial solution y(t) = 0.

(g) An initial value problem is a problem in which one is to find an unknown

function y(t) that satisfies both a given ODE and given initial conditions,
like y(0) = 1, y1 (0) = 0.

(h) A boundary value problem is a problem in which one is to find an unknown

function y(t) that satisfies both a given ODE and given boundary conditions,
like y(0) = 0, y(1) = 0.

The following theorem gives the form of solutions to the ODE (H.1).

Theorem H.2.

Assume that the coefficients a0 (t), a1 (t), ¨ ¨ ¨ , an´1 (t), an (t) and F (t) are continu-
ous functions and that a0 (t) is not zero.

(a) The general solution to the ODE (H.1) is of the form

y(t) = y p (t) + C1 y1 (t) + C2 y2 (t) + ¨ ¨ ¨ + Cn yn (t) (H.2)

where

˝ n is the order of (H.1)

˝ y p (t) is any solution to (H.1)
˝ C1 , C2 , ¨ ¨ ¨ , Cn are arbitrary constants
˝ y1 , y2 , ¨ ¨ ¨ , yn are n independent solutions to the homogenous equation

a0 (t)y(n) (t) + a1 (t)y(n´1) (t) + ¨ ¨ ¨ + an´1 (t)y1 (t) + an (t)y(t) = 0

associated to (H.1). “Independent” just means that no yi can be written

as a linear combination of the other y j ’s. For example, y1 (t) cannot be
expressed in the form b2 y2 (t) + ¨ ¨ ¨ + bn yn (t).

In (H.2), y p is called the “particular solution” and C1 y1 (t) + C2 y2 (t) + ¨ ¨ ¨ +

Cn yn (t) is called the “complementary solution”.

(b) Given any constants b0 , ¨ ¨ ¨ , bn´1 there is exactly one function y(t) that obeys
the ODE (H.1) and the initial conditions

y(0) = b0 y1 (0) = b1 ¨¨¨ y(n´1) (0) = bn´1

277
R EVIEW OF L INEAR O RDINARY D IFFERENTIAL E QUATIONS

Example H.3 (RLC circuit)

As an example of the most commonly used techniques for solving linear, constant coef-
ficient ODE’s, we consider the RLC circuit, which is the electrical circuit consisting of a
resistor of resistance R, a coil (or solenoid) of inductance L, a capacitor of capacitance C
and a voltage source arranged in series, as shown below. Here R, L and C are all nonneg-
ative constants.

R L
`
`
xptq iptq yptq
´ C
´

We’re going to think of the voltage x (t) as an input signal, and the voltage y(t) as an
output signal. The goal is to determine the output signal produced by a given input signal.
If i (t) is the current flowing at time t in the loop as shown and q(t) is the charge on the
di
capacitor, then the voltages across R, L and C, respectively, at time t are Ri (t), L dt (t) and
q(t)
y(t) = C . By the Kirchhoff’s law1 that says that the voltage between any two points has
to be independent of the path used to travel between the two points, these three voltages
must add up to x (t) so that

di q(t)
Ri (t) + L (t) + = x (t) (H.3)
dt C
Assuming that R, L, C and x (t) are known, this is still one differential equation in two
unknowns, i (t) and q(t). Fortunately, there is a relationship between the two. Namely
dq
i (t) = (t) = Cy1 (t) (H.4)
dt
This just says that the capacitor cannot create or destroy charge on its own; all charging of
the capacitor must come from the current. Substituting (H.4) into (H.3) gives

LCy2 (t) + RCy1 (t) + y(t) = x (t)

As a concrete example, we’ll take an ac voltage source and choose the origin of time so
that x (0) = 0, x (t) = E0 sin(ωt). Then the differential equation becomes

LCy2 (t) + RCy1 (t) + y(t) = E0 sin(ωt) (H.5)

This is a second order, linear, constant coefficient ODE. So we know, from Theorem H.2,
that the general solution is of the form y p (t) + C1 y1 (t) + C2 y2 (t), where
˝ y p (t), the particular solution, is any one solution to (H.5),

1 Gustav Robert Kirchhoff (1824–1887) was a German physicist.

278
R EVIEW OF L INEAR O RDINARY D IFFERENTIAL E QUATIONS

˝ C1 , C2 are arbitrary constants and

˝ y1 (t), y2 (t) are any two independent solutions of the corresponding homogeneous
equation
LCy2 (t) + RCy1 (t) + y(t) = 0 (H.5h )
So to find the general solution to (H.5), we need to find three functions: y1 (t), y2 (t) and
y p ( t ).

• Finding y1 (t) and y2 (t): The best way to find y1 and y2 is to guess them. Any
solution, yh (t), of (H.5h ) has to have the property that yh (t), RCy1h (t) and LCy2h (t)
cancel each other out for all t. We choose our guess so that yh (t), y1h (t) and y2h (t) are
all proportional to a single function of t. Then it will be easy to see if yh (t), RCy1h (t)
and LCy2h (t) all cancel. All derivatives of the function ert are again proportional to
ert . Hence we try yh (t) = ert , with the constant r to the determined. This guess is a
solution of (H.5h ) if and only if

LCr2 ert + RCrert + ert = 0 ðñ LCr2 + RCr + 1 = 0

?
´RC ˘ R2 C2 ´ 4LC (H.6)
ðñ r = ” r1,2
2LC
b
How we proceed depends on the sign of R2 C2 ´ 4LC. That is, whether R ą 2 CL or
b b
R ă 2 CL or R = 2 CL .
b
L
– Finding y1 (t) and y2 (t), when R ą 2 C: Then R2 C2 ´ 4LC ą 0, and r1 and r2
are two different real numbers. We may take y1 (t) = er1 t and y2 (t) = er2 t so
that the complimentary solution is C1 y1 (t) + C2 y2 (t) = C1 er1 t + C2 er2 t .
b
– Finding y1 (t) and y2 (t), when R ă 2 CL : Then R2 C2 ´ 4LC ă 0 and r1 and r2
are the two different complex numbers ´ρ ˘ iν, where
?
R 4LC ´ R2 C2
ρ= and ν=
2L 2LC
We may again take C1 er1 t + C2 er2 t as the complimentray solution. However we
can also rewrite C1 er1 t + C2 er2 t in terms of real valued functions by using that
e˘iθ = cos θ ˘ i sin θ:

C1 er1 t + C2 er2 t = e´ρt C1 eiνt + C2 e´iνt

 
( (
= e´ρt C1 cos(νt) + i sin(νt) + C2 cos(νt) ´ i sin(νt)


= e´ρt D1 cos(νt) + D2 sin(νt)

 

where2 D1 = C1 + C2 , D2 = i (C1 ´ C2 ). So we may also take y1 (t) = e´ρt cos(νt),

y2 (t) = e´ρt sin(νt) in the complementary solution.

2 Don’t make the mistake of thinking that C1 and C2 have to be real numbers, forcing D2 to be pure
imaginary. In most applications, D1 and D2 will be pure real and C1 and C2 will be complex.

279
R EVIEW OF L INEAR O RDINARY D IFFERENTIAL E QUATIONS

There is yet a third useful way to write the complementary solution. Think of
( D1 , D2 ) as a point in the xy-plane. Call the polar coordinates of that point A
and θ so that D1 = A cos θ and D2 = A sin θ. Then, using the trig identity
cos(α + β) = cos α cos β ´ sin α sin β, with α = νt and β = ´θ,

e´ρt D1 cos(νt) + D2 sin(νt) = e´ρt A cos(νt) cos θ + A sin(νt) sin θ

   
(H.7)
= Ae´ρt cos(νt ´ θ )
We have, in effect, replaced the two arbitrary constants D1 and D2 , whose val-
ues would normally be determined by initial conditions, by two other arbitrary
constants, R and θ, whose values would also normally be determined by initial
conditions.
b
– Finding y1 (t) and y2 (t), when R = 2 CL : Then R2 C2 ´ 4LC = 0 so that r1 = r2 .
We may take y1 = er1 t , but er2 t = er1 t is certainly not a second independent
solution. So we still need to find y2 . Here is a trick (called reduction of order3 )
for finding the other solutions: look for solutions of the form v(t)e´r1 t . Here
e´r1 t is the solution we have already found and v(t) is to be determined. To
R
save writing, set ρ = 2L so that r1 = r2 = ρ. To save writing also divide (H.5h )
R2
by LC and substitute that RL = 2ρ and LC 1
= 4L 2
2 = ρ . (Recall that we are

assuming that R2 = 4L C .) So (H.5h ) is equivalent to

y2h (t) + 2ρ y1h (t) + ρ2 yh (t) = 0

Substitute in
yh (t) = v(t)e´ρt
y1h (t) = ´ρv(t)e´ρt + v1 (t)e´ρt
y2h (t) = ρ2 v(t)e´ρt ´ 2ρv1 (t)e´ρt + v2 (t)e´ρt

So when yh (t) = v(t)e´ρt ,

y2h (t) + 2ρ y1h (t) + ρ2 yh (t)
= ρ2 ´2ρ2 + ρ2 v(t)e´ρt + ´ 2ρ + 2ρ v1 (t)e´ρt + v2 (t)e´ρt
   

= v2 (t)e´ρt
Thus v(t)e´ρt is a solution of (H.5h ) whenever the function v2 (t) = 0 for all t.

´ρt C1 and C2 , v
(t)´r=t C1 + C2 t has vanishing
But, for any values of the constants
second derivative so C1 + C2 t e = C1 + C2 t e 1 solves (H.5h ). This is of
the form C1 y1 (t) + C2 y2 (t) with y1 (t) = e´r1 t , the solution that we found first,
and y2 (t) = te´r1 t , a second independent solution. So we may take y2 (t) = ter1 t .
• Finding y p (t): The best way to find y p is to guess it. We guess that the circuit
responds to an oscillating input voltage with an output voltage that oscillates at the

3 The modern method of reduction of order was created by the French mathematician, physicist and
music theorist Jean le Rond d’Alembert (1717-1783). The interested reader can easily search out more
about his life.

280
R EVIEW OF L INEAR O RDINARY D IFFERENTIAL E QUATIONS

same frequency. So we try y p (t) = A sin(ωt ´ ϕ) with the amplitude A and phase ϕ
to be determined.
For y p (t) to be a solution, we need

LCy2p (t) + RCy1p (t) + y p (t) = E0 sin(ωt)

´LCω 2 A sin(ωt ´ ϕ) + RCω A cos(ωt ´ ϕ) + A sin(ωt ´ ϕ) = E0 sin(ωt)
= E0 sin(ωt ´ ϕ + ϕ)

and hence, applying sin( A + B) = sin A cos B + cos A sin B with A = ωt ´ ϕ and
B = ϕ,

1 ´ LCω 2 A sin(ωt ´ ϕ) + RCω A cos(ωt ´ ϕ)

= E0 cos( ϕ) sin(ωt ´ ϕ) + E0 sin( ϕ) cos(ωt ´ ϕ)

Matching coefficients of sin(ωt ´ ϕ) and cos(ωt ´ ϕ) on the left and right hand sides
gives

1 ´ LCω 2 A = E0 cos( ϕ)


(H.8)
RCω A = E0 sin( ϕ) (H.9)

It is now easy to solve for A and ϕ

(H.9) RCω
ùñ tan( ϕ) =
(H.8) 1 ´ LCω 2
RCω
ùñ ϕ = arctan
1 ´ LCω 2
b b
2 2
2
(H.8) + (H.9) ùñ 1´LCω 2 + R2 C2 ω 2 A = E0
E0
ùñ A = a
(1´LCω 2 )2 + R2 C2 ω 2

Naturally, different input frequencies ω give different output amplitudes A. Here is a

graph of A against ω, with all other parameters held fixed.

Note that there is a small range of frequencies that give a large amplitude response. This is
the phenomenon of resonance. It is exploited in the design of radio and television tuning

281
R EVIEW OF L INEAR O RDINARY D IFFERENTIAL E QUATIONS

circuitry. It has also been dramatically illustrated in, for example, the collapse4 of the
Tacoma narrows bridge.
Example H.3

Example H.4 (Boundary Value Problems)

By part (b) of Theorem H.2, an initial value problem consisting of an nth order linear ODE
with reasonable5 coefficients and n initial conditions always has exactly one solution. We
shall now see that a boundary value problem may have no solutions at all. Or it may have
exactly one solution. Or it may have infinitely many solutions. We shall start by finding
all solutions to the ODE
y2 + y = 0 (H.10)
We shall then impose various boundary conditions and see what happens.
The function y(t) = ert is a solution to (H.10) if and only if
r2 ert + ert = 0 ðñ r2 + 1 = 0 ðñ r = ˘i
where i (which electrical engineers often denote6 j) is a square root of ´1. Thus the general
solution to the second order linear ODE (H.10) is y(t) = C11 eit + C21 e´it , with C11 and C21
arbitrary constants. We may rewrite this general solution in terms of sin t and cos t by
substituting in
eit = cos t + i sin t e´it = cos t ´ i sin t
This gives
y(t) = C11 cos t + i sin t) + C21 (cos t ´ i sin t) = C1 cos t + C2 sin t
where C1 = C11 + C21 , and C2 = i (C11 ´ C21 ). Note that there is nothing stopping C11 and
C21 from being complex numbers. So there is nothing stopping C1 = C11 + C21 , and C2 =
i (C11 ´ C21 ) from being real numbers.
(a) Now consider the boundary value problem
y2 + y = 0 y (0) = 0 y(2π ) = 1 (H.11)
The function y(t) satisfies the ODE if and only if it is of the form
y(t) = C1 cos t + C2 sin t
for some constants C1 and C2 . A function of this form satisfies the boundary condition
y(0) = 0 if and only if
0 = y(0) = C1 cos 0 + C2 sin 0 = C1
A function of this form satisfies the boundary condition y(2π ) = 1 if and only if
1 = y(2π ) = C1 cos 2π + C2 sin 2π = C1
The two requirements C1 = 0 and C1 = 1 are incompatible. So the boundary value
problem (H.11) has no solution at all.

4 There are videos of the collapse on the web.

5 For example, continuous.
6 This is to avoid confusion with the current, which is typically called i.

282
R EVIEW OF L INEAR O RDINARY D IFFERENTIAL E QUATIONS

(b) Next consider the boundary value problem

π
y2 + y = 0 y (0) = 0 y =0 (H.12)
2
The function y(t) satisfies the ODE if and only if it is of the form

y(t) = C1 cos t + C2 sin t

for some constants C1 and C2 . A function of this form satisfies the boundary condition
y(0) = 0 if and only if

0 = y(0) = C1 cos 0 + C2 sin 0 = C1

π


A function of this form satisfies the boundary condition y 2 = 0 if and only if
π π π
0=y = C1 cos + C2 sin = C2
2 2 2
So we have a solution if and only if C1 = C2 = 0 and the boundary value problem
(H.12) has exactly one solution, namely y(t) = 0, which is a bit dull.

(c) Finally consider the boundary value problem

y2 + y = 0 y (0) = 0 y(2π ) = 0 (H.13)

The function y(t) satisfies the ODE if and only if it is of the form

y(t) = C1 cos t + C2 sin t

for some constants C1 and C2 . A function of this form satisfies the boundary condition
y(0) = 0 if and only if

0 = y(0) = C1 cos 0 + C2 sin 0 = C1

A function of this form satisfies the boundary condition y(2π ) = 0 if and only if

0 = y(2π ) = C1 cos(2π ) + C2 sin(2π ) = C1

So we have a solution if and only if C1 = 0 and the boundary value problem (H.13) has
infinitely many solutions, namely y(t) = C2 sin t with C2 being an arbitrary constant.

Example H.4

283