Sie sind auf Seite 1von 100

Lecture 1: AMATH 231-F14-001

September 8, 2014

Preliminary Material

0.1

Review Vector Algebra and Introduce Tensor Notation

Motivation for vectors and tensors:


Previously learned vector notation and this makes things more compact
Easier than writing out each component
However, proving vector identities can still be very lengthy
Here we review some vector algebra and present tensor (indicial) notation
Basic Notation
Small letters with half vectors for vectors:
~a = (a1 , a2 ).
Capital letters denote matrices:

A=

a11 a12
a21 a22


.

One free index denotes a vector: ai where i = 1, 2.


Two free indices denotes a matrix: Aij where i = 1, 2 and j = 1, 2.
Einstein summation convention: A repeated index means we sum up over all the indices
2
X

ai b i

is reduced to

ai b i

i=1

Can work for vectors of any size 2, 3, 4, !


Review of Vector Algebra
We will assume that ~a, ~b, ~c are all vectors in three-space, s is a scalar and A is a 3 3 matrix.
(a) Addition and Subtraction of Vectors: These operations are commutative and associative.
~a + ~b =

3
X

(ai + bi )

in tensor notation becomes ai + bi

i=1

(b) Multiplication of Vectors by Numbers:


s~a =

3
X

sai

in tensor notation becomes sai

i=1

(c) Scalar Products: Dot


~a ~b =

3
X

ai b i

in tensor notation becomes ai bi

i=1

(d) Matrix Multiplication:


A~b =

3
X

Aij bj

in tensor notation becomes Aij bi

j=1

(e) Vector Products: Curl


~a ~b = (a2 b3 a3 b2 , a3 b1 a1 b3 , a1 b2 a2 b1 )
(f) Triple Scalar Products:


~a ~b ~c = a1 (b2 c3 b3 c2 ) + a2 (b3 c1 b1 c3 ) + a3 (b1 c2 b2 c1 )


~a ~b ~c = ???
At the moment we dont see any clever way to write some of these vectors in tensorial notation
but we will figure out a way of doing that when needed. Now, we begin with the course notes
and start our discussion of Vector Calculus.
Before we get to talking about curves lets discuss some motivation to know why we might
want to learn vector calculus. To do this we visit the course notes for MATH 212.

0.2

Motivation to study Vector Calculus

Often we can translate fundamental laws of nature into the language of mathematics and Vector
Calculus is a useful way to accomplish this.
example 0.1 For example, consider water flowing into a cylindrical pipe. How can we determine how
the mass of water in a particular section is changing in time? (Units of kilograms/second)
1) Measure the concentration of water in that section at every instant and find the rate of change of
mass with time.
2) Find out how much water is coming in and going out through the boundaries per second.
By the Conservation of Mass both must give us the same answer. But in the first we are considering
the whole volume whereas in the second we are only looking at the boundaries. This simple idea is used a
great deal to determine the Differential Equations (DE) that describe a given system.
In physics we have four laws of electromagnetism (EM) that can be stated in words as follows:
1) Gauss Law


Flux of electric field


through any closed surface
2


=

Net change
inside the volume

2) Faraday


circulation of an electric field


around the perimeter of a surface


=

flux of the magnetic


field through the surface

3) No magnetic dipoles


flux of a magnetic field


through any closed surface


=0

4) Amp`ere

flux of the
circulation of a magnetic
flux of the

d
field around the perimeter =

+ electric current
electric field

density through
dt
through the surface
of a surface
the surface

Please dont be scared off by these equations. In this course you will not be tested in your
knowledge of physics. However, by the end of the course if I give you these principles you
should be able to deriving the governing system of Partial Differential Equations (PDEs) that
describes the motion.

Lecture 2: AMATH 231-F14-001

September 10, 2014

Curves and Vector Fields

This is mostly taken from sections 1.2 and 1.3 from the MATH 212 notes or section 1.1.1 of the
AMATH 231 course notes.
Definition 1.1 A scalar-valued function is of the form,
f : [a, b] R

or

x = f (t).

Definition 1.2 A vector-valued function is of the form,


~g : [a, b] R3

1.1
1.1.1

or

tensor

~g (t) = (g1 (t), g2 (t), g3 (t)) = gi (t).

Curves in Rn
Curves as vector-valued functions

Both functions produce curves in 1D space.


The first maps a 1D domain ([a, b]) to a 1D domain from f (a) to f (b).
The second maps a 1D domain ([a, b]) to a 3D domain from ~g (a) to ~g (b).
Next we need to define what we mean by a field.
Definition 1.3 A scalar field is a function of 2, 3, , n variables that yields a scalar. For example,
z = f (x, y),
u = g(x, y, z),
w = h(x1 , , xn ).
The input can be a vector but the output is always a scalar. Examples of scalar fields include charge
density, temperature, height of a mountain and speed of the winds.
Definition 1.4 A vector field is a function of 2, 3, , n variables that yields a vector. For example,
f~(x, y) = (f1 (x, y), f2 (x, y)),
~g (x, y, z) = (g1 (x, y, z), g2 (x, y, z), g2 (x, y, z)).
The input can be a vector but the output is always a vector. Examples of vector fields include velocity.
example 1.5 Recall that we can parameterize a curve as,
~x(t) = (r cos(t), r sin(t)),

if

0t

2
.

To make things concrete we can think of t as time, is the frequency (units of radians/second) and the ~x(t)
is the position of a particle (object) in the plane. How does the path change if we change the frequency? The
vector function, ~x(t), (position), is determined by the parameter, t (time). We say that the vector-valued
function is parameterized with respect to t.
4

example 1.6 We can parameterize a line using a linear relationship between the parameter and the
vector-valued function
~x(t) = ~r0 + t~u, if 0 t a,
where ~r0 and ~u are constant vectors and a is a scalar. This is the equation of a line that passes through
position ~r0 at t = 0 and has a slope of ~u.
Definition 1.7 Let ~g : [a, b] R2 be a vector function given by
~g (t) = (x(t), y(t)).
We call ~g a path and its image C, the curve of the path. The input t is a parameter and the above is the
parametric representation of the path.
example 1.8 Consider the path ~g : [0, 2] R2 given by
~g (t) = (t, t2 ) = (x(t), y(t)).
The input is t [0, 2] and the output is x(t) = t and y(t) = t2 . We can graph this for different values of t
or find out what the relation is between x and y:
y = t2 = (t)2 = x2 .
Therefore, this curve is nothing more that a parabolic centred at the origin opening upwards.
example 1.9 Consider the path ~g : [0, a] R2 given by
~g (t) = (a cos t, b sin t)

with

0 t 2.

We can find the following relations,


x
= cos t,
a

and

y
= sin t.
b

From our knowledge of trigonometry we know that cos2 t+sin2 t = 1 and therefore we deduce the following
algebraic relation between x and y:
x2 y 2
+ 2 = 1,
a2
b
which is of course the equation of an ellipse.
See section 1.1.1 of the AMATH 231 course notes for more details on the examples and also
one involving a helix.
1.1.2

Representation of a curve

Not only is there not a unique representation of a curve but there is always an infinite number of
parameterizations. For example in the previous example with the ellipse, we can replace t with
2s and as long as we change the bounds appropriately we get the same curve. We say that the
two paths are different parameterizations of the same curve C.
5

To make this concrete let


h : [, ] [a, b]
be a continuous and increasing and therefore one-to-one function between these two intervals.
Then, given the path ~g : [a, b] Rn we define, ~g : [, ] Rn by
~g( ) = ~g (h( )),

on

The two paths ~g (t) and ~g( ) follow the same points and thus yield the same curve.
How do we interpret these two paths? If we assume that the paths denote the position of
a particle and we want the independent parameters t and to both be time in seconds, then
that means the two particles follow the same paths but at different rate. That is to say different
speeds.
example 1.10 Define the following path,
~g (t) = (cos t, sin t).
This is clearly a circular motion of radius one and frequency equal to one.
If we pick t = h( ) = 3 then on possible path is
~g( ) = ~g (h( )) = (cos 3 , sin 3 ).
This does not have a constant frequency and indeed, the frequency increases quadratically in time.
If instead we pick t = h( ) = a then another path is
~g( ) = ~g (h( )) = (cos a, sin a ).
This rotates with a constant frequency of a.

Lecture 3: AMATH 231-F14-001


1.1.3

September 12, 2014

Limits

Next, we begin to discuss the derivative of ~g (t). This is build up by looking at the derivative of
each component. First we discuss the definition of a limit of ~g (t).
Definition 1.11 Given a vector-valued function (path)
~g (t) = (g1 (t), , gn (t)),
defined in a neighbourhood of t0 and a constant vector
~ = (L1 , , Ln ),
L
then
~
lim ~g (t) = L

tt0

lim gi (t) = Li ,

tt0

for

i = 1, , n.

Definition 1.12 The vector-field ~g is continuous at t0 means that


i) ~g (t0 ) is defined
ii) limtt0 ~g (t) exists
iii) limtt0 ~g (t) = ~g (t0 ).
Note the following properties:
i) ~g (t) is continuous at t0 gi (t) is continuous at t0 for i = 1, 2, , n.
ii) Physical curves are necessarily continuous functions.
iii) In general paths need not be continuous.
1.1.4

Derivatives

Definition 1.13 Given ~g : [a, b] Rn , the derivative of ~g at t, denoted by,

d~g
(t)
dt

or ~g 0 (t), is defined by

1
[~g (t + t) ~g (t)] ,
t0 t

~g 0 (t) = lim
provided the limit exists.
Note:

i) If ~g 0 (t) exists then ~g (t) is differentiable at t.


ii) ~g 0 (t) exists ~gi0 (t) exists for i = 1, 2, n.
iii) The derivative is computed component-wise, ~g 0 (t) = (g10 (t), , gn0 (t).

Physical Interpretation of the derivative:


Consider a curve C in R3 and the path (position) of the particle described by ~x = ~g (t). From
the definition of the derivative we see that it is a change in distance divided by the change in
time. That is to say it is the displacement vector over time t divided by the time interval. That
is why if ~g (t) is the position then the velocity is
~v (t) = ~g 0 (t) = (x0 (t), y 0 (t), z 0 (t)) ,
and the speed is simply,
k~v (t)k =

p
x0 (t)2 + y 0 (t)2 + z 0 (t)2 .

Following the same rationale, the second derivative yields the acceleration vector of the particle,
~a(t) = ~g 00 (t) = (x00 (t), y 00 (t), z 00 (t)) .
Geometrical Interpretation of the derivative:
Suppose a curve is defined by ~x = ~g (t) with ~g 0 (t0 ) 6= 0, then ~g 0 (t0 ) is a vector in the direction of
the tangent line at the point ~x = ~g (t0 ). Draw a picture like that of Figure 1.10 showing the second
and tangent lines.
Question: Why do we need ~g 0 (t0 ) 6= 0?
Equation of the tangent line:
Consider a curve ~x = ~g (t) with ~g 0 (t0 ) 6= 0. Since we know that ~g 0 (t0 ) is the direction of the
tangent line and it must go through the point ~g (t0 ) at t = t0 we can write down the equation of
the tangent line as,
~ t0 = ~g (t0 ) + (t t0 )~g 0 (t0 ).
L
~ t0 to denote the tangent line at the point t0 since clearly it will
Note that we used the variable L
change in general.
Question: For what curve are all the tangent lines the same?
Definition 1.14 We say that a vector-value function ~g : [a, b] Rn is of of class C 1 or just C 1 if ~g (t)
is differentiable.
Definition 1.15 We say that a vector-value function ~g : [a, b] Rn is piecewise C 1 if there is a partition
of [a, b],
a = t0 < t1 < < tN = b,
such that ~g (t), when restricted to each open interval (ti , ti+1 ) for i = 0, 1, , N1 , coincides with the
function that is C 1 on the closed interval [ti , ti+1 ].
example 1.16 Find the speed of ~x(t) = (r cos t, r sin t).
The velocity is
~x0 (t) = r( sin t, cos t).
The speed is therefore,
k~x0 (t)k = |r|.
This means that the linear speed is linear proportional to the radius and the frequency. The units are of
course length/time.
8

example 1.17 Show the following



d
k~x(t)k2 = 2~x ~x0 .
dt
Expand using the definition and the product rule:

d
d
k~x(t)k2 =
(~x ~x)
dt
dt
= ~x0 (t) ~x + ~x ~x0
= 2~x ~x0 .
example 1.18 Compute the following
d
k~x(t)k.
dt
Expand using the power rule:

d
k~x(t)k2 = 2k~xkk~xk0
dt
2~x ~x0 = 2k~xkk~xk0 .
From this we deduce the result,
k~xk0 =

~x ~x0
.
k~xk

example 1.19 Example 1.5 from AMATH 231 Course Notes.


Consider the curves defined by
i) ~x = ~g1 (t) = (t|t|, t2 ),
ii) ~x = ~g2 (t) = (t, |t|),

1 t 1.
1 t 1.

Observe that we can get non-parameteric representation of these curves:


i) y = t2 = |t||t| = |t|t|| = |x|.
ii) y = |t| = |x|.
Therefore, both curves are the same.
Question: Are the parameterizations differentiable at t = 0?
Since we cannot compute the derivative of the absolute value at t = 0 the only hope we have is to use
first principles. First for ~g1 (t),
~x(h) ~x(0)
(h|h|, h2 ) (0, 0)
= lim
= lim (|h|, h) = (0, 0),
h0
h0
h0
h
h
lim

and second for ~g2 (t),


~x(h) ~x(0)
(h, |h|) (0, 0)
|h|
= lim
= lim (1, ),
h0
h0
h0
h
h
h
lim

which does not exist.


Question: What does this mean?
9

Lecture 4: AMATH 231-F14-001

September 15, 2014

To figure this out we can compute the derivatives for t < 0 and t > 0, respectively. For the first
function we get,
~g10 (t) = (2t, 2t),

for

t<0

~g10 (t) = (2t, 2t), for t > 0,

The speed is the same on either side and is k~g k = 2 2|t|. From this we see that the particle slows down
as it approaches t = 0 and stops. For the second function we obtain a different result,
~g20 (t) = (1, 1),
~g20 (t) = (1, 1),

for
for

t<0
t > 0.

The speed on either side is k~g k = 2. So the speed is continuous but the velocity is discontinuous. We see
that the particle always moves in the positive x-direction at the same speed. But for t < 0 it moves down
at a constant speed and then, when it gets at t = 0 it suddenly starts to move upwards at the same speed.
Clearly we see that some paths can be rather confusing and non-physical.
See example 1.6 in the AMATH 231 course notes for an example of a cycloid.
1.1.5

Arclength

It is easy to determine the length of a straight lines is easy but how do we find the length of any
curve in Rn ? To answer this let us consider a path given by
~x = ~g (t),

with

a t b.

The mapping is from an interval of the real line to Rn .


insert picture
As in calculus, we use what we know. At present all we know is how to find length of straight
lines. That is why we approximate the curve with straight lines to get an estimate. Then we use
more and more straight lines. This is the same idea used in Riemann sums!
The process boils down to the following steps:
1) To begin, we partition our interval [a, b] into n subintervals,
a = t0 < t1 < t2 < < tn1 < tn = b.
2) We linearly approximate the curve along each subinterval using the tangent line,
~xi+1 ~xi ti~g 0 (ti ),

with ti = ti+1 ti .

3) Find the length of each segment (si denotes the length of arc i),
si k~xi+1 ~xi k = k~g 0 (ti )kti ,

10

for all i

4) We sum over all the segments to get an an exact value of the arc length, for this partition,
sN =

N
1
X

k~g 0 (ti )kti .

i=1

We used sN to denote the arc length of the N -th piecewise constant approximation.
5) Take the limit as N or ti 0 to get the exact value for the arc length of the curve,
Z

s=

k~g 0 (t)k dt.

If the image C is the trajectory traversed by a particle then s is the distance travelled.
Note: At each point t [a, b] we have that k~g 0 (t)k dt computes the infinitesimal distance
travelled in a time dt. Integration then sums over all values of t and therefore yields the arc
length (or distance) exactly.
example 1.20 Compute the length of the path ~g1 (t) = (cos t, sin t) on t [0, 2] and ~g2 (t) = (cos 3t, sin 3t)
on t [0, 2].
Z

k~g10 (t)k dt

s(~g1 (t)) =

s(~g2 (t)) =
0

0
2

k~g20 (t)k dt

k( sin t, cos t)k dt =

sin t +

Z
cos2

0
2

t dt =

1 dt = 2
0

k3( sin 3t, cos 3t)k dt = 3

p
sin2 3t + cos2 3t dt = 6.

The second is three times bigger because it wraps around the circle three times.
example 1.21 What is the length of the curve y = f (x) on x [a, b]?
We parameterize the curve as ~g (t) = (x, f (x)) and x [a, b], which implies ~g 0 (x) = (1, f 0 (x)). The
formula yields,
Z bp
s(~g (x)) =
1 + (f 0 (x))2 dx.
a

example 1.22 Prove that the shortest distance between two points is a straight line. Suppose that the
points are A = (a, 0) and B = (b, 0).
We parameterize any smooth curve connecting A to B with the path ~g (t) = (t, f (t)). Recall that the
formula for arc length is
Z bp
1 + (f 0 (t))2 dt.
s(~g (t)) =
a

Clearly the distance between the points is |b a|. We must show that any other path has a distance that
is larger than that, i.e. s(~g (t)) > b a, where we assume b > a.
If the path is not exactly a straight line then there must be a point where f 0 (t) 6= 0. Since f (t) is
assumed to be C 1 , then f 0 (t) is continuous and there must be an interval (a0 , b0 ) around t where f 0 (t) 6= 0.

11

Thus,

p
1 + (f 0 (t))2 > 1 holds on (a0 , b0 ). But,
Z bp
s(~g (t)) =
1 + (f 0 (t))2 dt,
a
Z a0 p
Z b0 p
Z bp
0
2
0
2
=
1 + (f (t)) dt +
1 + (f (t)) dt +
1 + (f 0 (t))2 dt,
0
0
a
a
b
Z b
Z b0
Z a0
1 dt +
1 dt,
1 dt +
>
a0
b0
a
Z b
=
1 dt = b a.
a

Therefore, if the line is ever not straight, we necessarily get a length that is larger than that of a line and
we are done.
Lecture 5: AMATH 231-F14-001

September 17, 2014

example 1.23 Let ~g : [a, b] R2 by a C 1 path. The arc length function of ~g (t) is given by
Z t
s(t) =
k~g 0 ( )k d.
a

Observe that,
a) s(a) = 0.
b) s(b) = s(~g ).
c) s(t) is a monotonically decreasing function.
d)

ds
dt

= k~g 0 ( )k.

This turns out to be a very powerful tool. Based on our previous work we can parameterize any
path with its arc length. Unlike our previous parameterizations, we have that there is as unique
path that does this. But to do this we need to find t(s).
example 1.24 Compute the arc length function for the path ~g (t) = (3 cos t, 3 sin t) with t [0, 2] and
parameterize it by its arc length.
Solution:
Compute the derivative of the path,
~g 0 (t) = (3 sin t, 3 cos t)
Therefor the norm of this is k~g 0 (t)k = 3. Therefore, the arc length function is easily computed,
Z t
Z t
0
s(t) =
k~g (t)k dt =
3 dt = 3t.
0

In this case the inversion is easy, t = s/3.


Therefore, the parameterization by arc length is,
s
s
~g 0 (s) = (3 cos , 3 sin )
3
3
and note that k~g 0 (s)k = 1 by design, as we will prove shortly.
12

example 1.25 Compute the arc length function for the path ~g (t) = (a cos t, a sin t, bt) with t [0, 2]
with a, b > 0 and parameterize it by its arc length.
Solution:
ELFS (Exercise Left for the Student)
1.1.6

Acceleration and Curvature

Now we have a method to parameterize any path, say ~g (t), by its arc length, ~g (s) (at least in
principle). Observe that if we differentiate with respect to arc length we get, after using the
chain rule,
d
d
dt
~g (s) = ~g (t) =
ds
dt
ds

d
~g (t)
dt
ds
dt

~g 0 (t)
k~g 0 (t)k

Therefore, the length of this is unity,


0


d


~g (s) = ~g (t) = 1.
ds
k~g 0 (t)k
Definition 1.26 For any path ~g (t) we can define the unit tangent vector to ~g (t) as
~g 0 (t)
T~ (t) = 0
.
k~g (t)k
This is the unique choice of tangent vectors that has a length of one for any value of t.
Note that T~ (t) = ~g 0 (s).
example 1.27 Let ~g (t) = (2t3/2 , 2t) from (0, 0) to (2, 2)
a) Find the arc length function s and then ~g (s).
b) Check

ds
dt

= k~g 0 (t)k

c) Find T~ (t) = ~g 0 (s).


Solution:
a) The derivative of the path is ~g 0 (t) = (3t1/2 , 2) and thus k~g 0 (t)k =

9t + 4.

With this we can calculate the arc length function,


Z t
t


2
2 
s=
9t + 4 dt =
(9t + 4)3/2 0 =
(9t + 4)3/2 8 .
27
27
0
Next, we invert to solve for t in terms of s,
27
s + 8,
2

2/3
1 27
4
t=
s+8
.
9 2
9

(9t + 4)3/2 =

We can substitute this into the path to obtain our unique parameterization in terms of the arc length,



2/3
 3/2 2 27
2 
8
3/2
~g (s) = (2
(9t + 4) 8
,
s+8
)
27
9 2
9
13

b) We differentiate the arc length with respect to t to verify that it is equal to k~g 0 (t)k =

9t + 4,

ds
2 27
=
(9t + 4)1/2 = 9t + 4.
dt
27 2
c) We compute the unit tangent vector,
~g 0 (t)
(3t1/2 , 2)
T~ (t) = 0
=
.
k~g (t)k
9t + 4
Physical Quantities
Let ~g (t) be a C 2 path describing the trajectory of the motion.
Speed is

ds
dt

= k~g 0 (t)k is the rate of change of arc length with time.

Velocity is the speed times the unit tangent vector.


~v (t) = ~g 0 (t) = k~g 0 (t)k

~g 0 (t)
ds
= k~g 0 (t)kT~ (t) = T~ (t).
0
k~g (t)k
dt

Note that ds/dt is the magnitude and T~ is the unit tangent vector.
Acceleration: sum of two components and directions.


d ds ~
d
T (t) ,
~a(t) = ~v (t) =
dt
dt dt
 
d ds ~
ds d ~
~a(t) =
T (t) +
T (t),
dt dt
dt dt
d2 s
ds d ~
~a(t) = 2 T~ (t) +
T (t).
dt
dt dt
Orthogonality property: Since we know that kT~ (t)k2 = T~ T~ = 1 we can differentiate this
to show that the unit tangent vector is orthogonal to its derivative,
T~ T~ 0 = 0.
The acceleration has a part parallel to T~ and a second parallel to T~ 0
Tangential acceleration (speeding up in the same direction)
d2 s ~
T (t) .
dt2

~aT =
Normal acceleration (tendency to turn)
~aN =

ds d ~
T (t).
dt dt

14

Can simplify the derivative of the unit tangent vector,


d~
d
d
ds
T (t) = T~ (t) = k~g 0 (t)k T~ (t).
dt
ds
dt
ds
Normal component of acceleration can be rewritten as,
ds
dt

2

d2 s
~a = 2 T~ (t) +
dt


~aN =

d ~
T (t) .
ds

Acceleration can then be written as,


ds
dt

2

d ~
T (t).
ds

Lecture 6: AMATH 231-F14-001

September 19, 2014

Definition 1.28 ~g (t) is a C 2 path on R2 (or R3 ) parameterized by arc length. The curvature (s) of ~g (s)
is defined as,





0
~
T
(t)


d
~ =
(s)
T
,
ds
k~g 0 (t)k
using the above identity. Geometrically, that means that the curvature is the magnitude of the rate of
change of the unit tangent vector with respect to arc length.
example 1.29 Find the curvature of the line ~g (t) = (at, bt).
Solution:
(a, b)
~g 0 (t)
=
T~ (t) = 0
.
k~g (t)k
a2 + b 2
Since this is a constant we then deduce that the curve is zero, as it should be for a straight line,


~0
T (t)
(s) =
= 0.
k~g 0 (t)k
example 1.30 Find the curvature of the circle ~g (t) = (a cos t, a sin t).
Solution:
(a sin t, a cos t)
~g 0 (t)
T~ (t) = 0
=
= ( sin t, cos t).
k~g (t)k
a
Next, we can compute the curvature from the formula


~0
T (t)
k( cos t, sin t)k
1
(s) =
=
= .
0
k~g (t)k
a
a
The curvature of a circle of radius a is 1/a. In the limit as a we have a straight line and the
curvature is zero. This confirms what we verified in the previous example.
15

example 1.31 Find the curvature of the helix ~g (t) = (a cos t, a sin t, bt) with a, b > 0 and t 0.
Solution:
~g 0 (t)
(a sin t, a cos t, b)
1

T~ (t) = 0
=
=
(a sin t, a cos t, b).
2
2
2
k~g (t)k
a +b
a + b2
Next, we can compute the curvature from the formula


~0
T (t)
k(a cos t, a sin t, 0)k
a
(s) =
=
= 2
.
0
2
2
k~g (t)k
a +b
a + b2
In the limit where b = 0 we recover the result for a circle. But in general we find that if b 6= 0 then the
curvature of the helix tends to be smaller than that of a circle.
exercise 1.32 Find the curvature of the parabola y = x2 . [Solution: (t) =

1.2
1.2.1

2
]
(1+4t2 )3/2

Vector Fields
Examples from physics

example 1.33 Gravitational field due to a spherical body of mass M is a vector field
~r
F~ (~x) = GM 3 ,
r

with

r = k~rk ,

where G is the gravitational constant and F~ is the force exerted on a test body of unit mass.
example 1.34 Velocity of a fluid is a vector field, such as the atmosphere or the oceans.
example 1.35 Density of a fluid is a scalar field.
1.2.2

Field lines of a vector field

Assume that F~ (~x) determines the velocity of a fluid at position ~x.


Definition 1.36 Let F~ (~x) be a continuous vector field defined on a subset R2 or R3 . A flow (field) line
of F is a path ~g (t) such that
d
~g (s) = F~ (~g (t)).
dt
Note: Flow lines are called integral curves if a path. That is to say they are the curves that satisfy this
above Differential Equation (DE).
Calculating the flow lines is equivalent to answering the following question: given a velocity
field F~ , what trajectory do particles follow? For example, consider a river with a winding coastline. If we drop a light nerf ball in the river, what path does it traverse?

16

example 1.37 The field lines ~x = ~g (t) satisfies ~g 0 (t) = F~ (~g (t)). Then we deduce,
d
~x = (x0 (t), y 0 (t)) = (y(t), x(t)),
dt
that goes through the point (, ). The flow lines yield two DE,
dx
dy
= y, and
= x.
dt
dt
If we differentiate the first we can then substitute in the second to get,
d2 x
d2 x
dy
= x, or
=
+ x = 0.
dt2
dt
dt2
This is the equation for the simple harmonic oscillator that we can solve exactly.
We look for a solution of the form et and find that the characteristic equation is,
2 + 1 = 0,
and so the roots are = 1. Therefore, the solution can be written as,
x = aeit + beit .
But using Eulers formula and redefining our constants we can write this in terms of real functions,
x = a cos t + b sin t.
To satisfy the initial conditions we pick a = and b = and therefore our solution is,
x = cos t + sin t.
Crash Course in ODEs
For second order DE with constant coefficients of the form,
ax00 + bx0 + c = 0,
we look for solutions of the form x = et to obtain our characteristic equation (after dividing
through by the exponential)
a2 + b + c = 0.
The roots are,
1,2 =

b2 4ac
.
2a

example 1.38 The field lines ~x = ~g (t) satisfies ~g 0 (t) = F~ (~g (t)). Then we deduce,
d
~x = (x0 (t), y 0 (t)) = (x(t), y(t)),
dt
that goes through the point (, ). The flow lines yield two DE,
dx
dy
= x, and
= y.
dt
dt
These equations separate and we can solve each independently. The exact solution is,
x = et ,

and
17

y = et .

Lecture 7: AMATH 231-F14-001

September 22, 2014

Line Integrals and Greens Theorem

2.1
2.1.1

Line Integral of a Scalar-Field


Motivation and Definitions

So far we are able to compute the arclength of a curve given its path ~g (t). What if we want to do
more? Suppose that f (x, y) is the density of a hanging wire and t parameterizes the wire. What
is the mass of the wire?
Definition 2.1 The curve C in Rn is defined by ~x = ~g (t) on a t b where ~g is C 1 and f (~x) is C 0 . The
integral of f along path C is,
Z

f ds =

f ds =
C

~g

f (~g (t)) k~g 0 (t)k dt.

In worlds we say, f is evaluated on the curve C giving f (~g (t)) and the symbol ds is the arc length and
therefore yields k~g 0 (t)k dt. The left-hand side is new but the right-hand side is a standard Riemann
integral. Note that in the special case where f = 1 we recover the formula for arc length,
ds =

ds =

s=

~g

k~g 0 (t)k dt.

Consider a curve like in Figure 2.2 of the course notes.


1) We divide our interval [a, b] into N subintervals.
2) Linearly approximate the curve on each segment.
3) For the i-th segment we can write the approximate arc length as,
s(~g (ti )) k~g 0 (ti )k ti .
4) The mass of a small segment (that is nearly linear) is equal to the density times the arc
length. If ti is sufficiently small then the density is nearly constant and therefore the
mass is
mi f (~g (ti )) k~g 0 (ti )k ti .
This says that locally mass = density length.
5) To find the mass of the entire curve we sum up over all the segments and then take the
limit as N . This results in the above equation.
example 2.2 Compute the line integral

R
~g

f ds of f (x, y, z) = xyz along the curve

~g (t) = ( sin t, 2 cos t, sin t),


18

on

t [0, /2].

Solution:
Must compute the derivative,

~g 0 (t) = ( cos t, 2 sin t, cos t),

then the magnitude of the derivative,


p

cos2 t + 2 sin2 t + cos2 t = 2.

k~g 0 (t)k =
Now we can unravel the line integral,

Z
f ds =
~g

f (~g (t)) k~g 0 (t)k dt,

/2

( 2 sin2 t cos t) 2 dt,

= 2 sin t cos t|/2


0 ,
/2
2
= sin3 t 0 ,
3
2
= .
3
Is it bad to have a negative answer? It depends on what were computing. Here, no.
example 2.3 Compute the line integral

R
~g

f ds of f (x, y) = 2x + y along the paths:

a) Clockwise quarter circle: ~g1 (t) = (cos t, sin t) on t [0, /2].


b) Along a line segment: ~g2 (t) = (1, 0) + t(1, 1) = (1 t, t) on t [0, 1].
c) Along two line segments: ~g3 (t) = (1 t, 0) on t [0, 1] and ~g4 (t) = (0, t) on t [0, 1].
Solution:
a) Compute the derivative and its magnitude,
~g10 (t) = ( sin t, cos t),

k~g1 k = 1.

Therefore, the line integral becomes,


Z

Z
f ds =

~g1

f (~g (t)) k~g 0 (t)k dt,

/2

(2 cos t + sin t)1 dt,


0

= 2 sin t cos t|/2


0 ,
= (2 0) (0 1) = 3.
b) Compute the derivative and its magnitude
~g20 (t) = (1, 1),

=
19

k~g20 (t)k =

2.

Now we compute the line integral


Z 1
(2 t) dt,
f ds = 2

Z
~g2

=
=

1
2 (2t t2 /2) 0 ,

2(2 1/2) = 3/ 2.

c) The same thing but now for two line segments:


~g30 (t) = (1, 0),

k~g30 (t)k = 1.

~g40 (t) = (0, 1),

k~g40 (t)k = 1.

Sum the two line integrals,


Z

Z
f ds +

~g3

2(1 t) dt +

f ds =
~g4

t dt,
0


1
= 2t t2 + t2 /2 0 dt,
= 2 1 + 1/2 = 3/2.
Even though the starting and ending points are the same the line integral depends on the path, and
so the answers are different.
Definition 2.4 The average value of a scalar field is,
1
hf i =
s(~g )

1
f ds =
s(~g )
~g

f (~g (t)) k~g 0 (t)k dt.

Consistency Property:
If ~g (t) : a t b is C 1 and ~g( ) : is C 1 and they parameterize the same curve C then
the line integral of any function f over the two paths must be equal:
Z

Z
f ds =

f (~g1 (t)) k~g10 (t)k

dt =

f (~g2 ( )) k~g20 ( )k d.

example 2.5 Compute the average temperature of a wire in the shape


~g (t) = (cos t, t/10, sin t),
if the temperature is T (~x) = x2 + y + z 2 .
Solution:
1+

.
2

20

on

t [0, 10],

Lecture 8: AMATH 231-F14-001

2.2
2.2.1

September 24, 2014

Line Integral of Vector Field


Motivation and Definition

Suppose a force field F~ (~x) moves a particle along a curve C given by ~g : [a, b] R2 from ~g (a) to
~g (b). What is the work done by F~ ? (Recall that work done is equal to the force dotted with the
displacement.)
Definition 2.6 The curve C in Rn is given by ~x = ~g (t), a t b, ~g is C 1 and F~ is continuous. The line
integral of F~ along C is,
Z
Z b
Z
~
~
F~ (~g (t)) ~g 0 (t) dt.
F d~x =
F d~x =
~g

Note: F~ (~x)d~x = F~ (~g (t))~g 0 (t) dt is the work (force times distance) done by F~ along the segment denoted
by dx.
Method proceeds as before but with one major difference:
1) We divide our interval [a, b] into N subintervals.
2) Linearly approximate the curve on each segment and compute ~xi = ~xi+1 ~xi .
3) The line integral (work done) on the i-th segment is
F~ (~xi ) ~xi .
This says that locally work = force displacement.
4) Rewrite the displacement in terms of the path ~xi ~g 0 (ti )ti and substitute into the above
F~ (~xi ) ~xi = F~ (~xi ) ~g 0 (ti )ti .
5) Sum over all the segments and then take the limit as N .
example 2.7 Compute the work done by F~ acting upon a particle that moves along the trajectory ~g (t) :
[a, b] R3 is given by the path integral,
Z b
W =
F~ (~g (t)) ~g 0 (t) dt.
a

where F~ = (y, x, 1) if we consider the following curves:


a) ~g (t) = (cos t, sin t, 0), t [0, ].
b) ~g (t) = (t, t, t), t [0, 1].
c) ~g (t) = (cos t, sin t, t), t [0, 2].
Solution:
21

a) We must evaluate the vector field along the path and compute the derivative of the path,
F~ (~g (t)) = (y(t), x(t), 1) = ( sin t, cos t, 1),
~g 0 (t) = ( sin t, cos t, 0).
Now we can evaluate the integral,
Z
Z
0
W =
F~ (~g (t)) ~g (t) dt =
0

sin2 t + cos2 t dt = .

b) Similarly,
F~ (~g (t)) = (y(t), x(t), 1) = (t, t, 1),
~g 0 (t) = (1, 1, 1),
and the work becomes
Z
W =

F~ (~g (t)) ~g 0 (t) dt =

t + t + 1 dt = 1.

~g (t) = (t, t, t), t [0, 1].


c) Finally,
F~ (~g (t)) = (y(t), x(t), 1) = ( sin t, cos t, 1),
~g 0 (t) = ( sin t, cos t, 1),
and the work becomes
Z
W =

F~ (~g (t)) ~g 0 (t) dt =

sin2 t + cos2 t + 1 dt = 4.

Geometric Interpretation:
The path integral can be rewritten using the dot product
Z b
Z b
Z


~
0
~
~
F (~g (t)) ~g (t) dt. =
F d~x =
F (~g (t)) k~g 0 (t)k cos (t) dt.
a

~g

The integrand, and therefore the integral, is largest for paths such that is closest to zero. Physically, that means the most work is done by a vector field that along a path that is parallel to the
vector field. These optimal paths are the flow lines of F~ .
Consistency requirement:
If ~g and ~g are two parameterizations of the same curve then
Z b
Z
0
F~ (~g (t)) ~g (t) dt. =
F~ (~g(t)) ~g0 (t) dt,
a

where ~g : [a, b] Rn and ~g : [, ] Rn .


Alternative Notation:
If F~ = (F1 , F2 ) and ~g 0 (t) dt = (x0 (t), y 0 (t)) dt = (dx, dy), then we could write,
Z
Z
Z
~
F d~x = (F1 , F2 ) (dx, dy) =
F1 dx + F2 dy.
C

22

example 2.8 Evaluate the line integral,


Z
I = (3x + 4y) dx + (2x + 3y 2 ) dy,
C

around the circle x2 + y 2 = 4.


Solution:
We can parameterize the circle as ~g (t) = (2 cos t, 2 sin t) and t [0, 2]. We compute the derivative to be,
~g 0 (t) = (2 sin t, 2 cos t),
then substitute the resultant into the integral,
Z
I = (3x + 4y) dx + (2x + 3y 2 ) dy,
ZC 2


=
2(3 cos t + 4 sin t)(2 sin t) + 2(2 cos t + 6 sin2 t)(2 cos t) dt,
Z0 2


=
12 cos t sin t 16 sin2 t + 8 cos2 t + 24 sin2 t cos t dt,
Z0 2


=
12 cos t sin t 8 (1 cos 2t) + 4 (1 + cos 2t) + 24 sin2 t cos t dt,
0




2

1
1
2
3
= 6 sin t 8 t sin 2t + 4 t + sin 2t + 8 sin t
,
2
2
0
= 8.
Properties of Line Integrals:
(i) Linearity:
Z

~ d~x =
(F~ + G)

F~ d~x +

~ d~x.
G

(ii) Additivity: C is a C 1 curve and C = C1 C2 and C1 C2 = then


Z
Z
Z
~
~
F d~x =
F d~x +
F~ d~x.
C

C1

C2

(iii) Reversal of Orientation: If you reverse the orientation of a path C we denote that with C
and we have that
Z
Z
~
F d~x =
F~ d~x.
C

Note that each curve has two possible orientations.


Proof of Reversal of Orientation
See proposition 2.2 of the course notes. Not too difficult but technical.

23

Lecture 9: AMATH 231-F14-001

September 26, 2014

example 2.9 Work equals gain in kinetic energy


Suppose that F~ (~x) is the force acting on a particle m that moves along the path ~g (t) on the interval
t [a, b]. Newtons second law implies that,
F~ (~g (t)) = m~a(t) = m~v 0 (t) = m~g 00 (t).
The work done by the force of gravity is,
Z
W =

F~ d~x =

F~ (~g (t)) ~g (t) dt =


0

m~v 0 (t) ~v (t) dt.

~g

But the product rule yields,



d
d
k~v k2 =
(~v ~v ) = 2~v ~v 0 .
dt
dt
When we substitute this into the line integral we obtain,
Z
W =
a

m
m~v (t) ~v (t) dt =
2
0

Z
a


d
k~v k2 dt.
dt

By a fundamental theorem of calculus then we can evaluate this easily,


W =

m
m
k~v (b)k2 k~v (a)k2 .
2
2

That is to say the work done on an object is equal to its final (kinetic) energy minus the initial (kinetic)
energy. This is completely independent on what path it takes. This leads to our next topic.

2.3

Path Independent Line Integrals (Conservative Fields)

Definition 2.10 The image of a one to one, piecewise C 1 mapping (path) ~g : [a, b] R2 is said to be a
simple curve. That is to say there are no intersections.
Definition 2.11 The image of a piecewise C 1 mapping (path) ~g : [a, b] R2 that is one to one on [a, b]
and ~g (a) = ~g (b) is called a simple closed curve. These curves can be clockwise or counterclockwise.
Definition 2.12 A domain D is connected if every pair of points in D can be joined by a continuous
curve lying strictly in D.
Definition 2.13 A simply connected domain D is a connected domain in which every simple closed curve
can be continuously shrunk to a point in D without ever leaving D.
Definition 2.14 A positive orientation is one that is counter-clockwise.
We have see many examples of line integrals that depend on path and that is very typical.
However, there is a special family of line integrals that are path independent. These occur when
the vector field is said to be conservative. Gravitational fields are conservative whereas air resistance is non-conservative (path dependent).
24

~ x) in a domain D, then F~ is a conservative vector field in D and is a


Definition 2.15 If F~ (~x) = (~
potential for F~ for D.
example 2.16 Say,
F~ (~x) = km

p
,p
x2 + y 2
x2 + y 2

= km

~r
,
k~rk

is a conservative (gradient) vector field on R2 {(0, 0)} since if we define


p
~
= km x2 + y 2 , =
F~ = .
Check:
Suppose the definition from above and evaluate
~ = (km
~
~ k~rk .

k~rk) = km
How do we evaluate the above? We can use the definition,



~
~r ~r,
~r ~r ,
k~rk =
x
y
!
~
r
~
r
~r y
~r x
,
=
.
~r ~r ~r ~r
But since ~r = (x, y) we have that the partial derivatives are the unit vectors in each direction,
~ k~rk = 1 (~r ex , ~r ey ) = ~r .

k~rk
k~rk
Therefore, returning to our above calculations,
~ = km
~ k~rk = km ~r .

k~rk
~
Thus, we have demonstrated that F~ (~x) = .
example 2.17 Show that the velocity field ~u = (y, x) of rigid body motion about the z-axis is nonconservative if 6= 0.
Solution:
If the field is conservative then ~u = (y, x) = and therefore,

= y,
x

and

= x.
y

Method 1:
Integrate the first equation with respect to x to get,
= xy + f (y).
Then, differentiate with respect to y and match with the second equation,

= x + f 0 (y) = x.
y
25

But there is no function f 0 (y) that can satisfy this equation. Therefore it is impossible and so our field
must be non-conservative.
Method 2:
If ~u is C 1 and exists then must be C 2 and therefore we require that,
2
2
=
.
xy
yx
But, from the above equations we have,
 
2

=
=
xy
y x
 
2

=
=
yx
x y

(y) = ,
y

(x) = .
x

Since these are not equal a function cannot exist and therefore a conservation function does not exist.
Theorem 2.18 Necessary condition for a conservative field in R2 to exist.
If F~ = (F1 (x, y), F2 (x, y)) is a conservative vector field in the domain D then,

F1 (x, y) =
F2 (x, y).
y
x
Proof:
If F~ = then, by definition,
F~ = (F1 , F2 ) =


,
x y

But if F~ is C 1 then must be C 2 and thus,


F1

F2
=
=
=
.
y
yx
xy
x

26

Lecture 10: AMATH 231-F14-001

September 29, 2014

Theorem 2.19 Necessary condition for a conservative vector field in R3 .


If F~ = (F1 , F2 , F3 )(x, y, z) is a conservative vector field in D in R3 then we must have everywhere in D
that,

F1 =
F2 ,
F2 =
F3 ,
F3 =
F1 .
y
x
z
y
x
z
Or, this can be more easily stated as saying that the curl of the vector field is zero,
~ F~ = ~0.

The proof is not very difficult and is analogous to the 2D result from before.
example 2.20 Show that F~ (x, y) = (x, y) is a conservative vector field in R2 and find the potential .
Solution:
From a previous theorem,

= x, and
= y.
x
y
Integrate the first equation,
1
= x2 + f (y).
2
Differentiate with respect to y and then use the second equation,

= f 0 (y) = y,
y
and therefore f (y) = 12 y 2 + c. Therefore, the vector field is conservative and the potential function is,
=

x2 y 2

+ c.
2
2

2
example 2.21 Find the potential function of F~ (x, y, z) = (xy sin z, x2
Solution:
From a previous theorem we surmise that

= xy sin z,
x

x2 e y
=
,
y
2
z
y

e
= 2 x cos z.
z
z
Integrate the first equation,
x2 y
x sin z + f (y, z).
2
Differentiate with respect to y and then use the second equation,
=

x2 f
x2 e y
=
+
=
.
y
2
y
2
z
27

ey ey
,
2 z2

x cos z)

Therefore, f = ez + g(z). We plug this int our expression for , differentiate with respect to z and then
use our final equation to obtain that
x cos z

ey
ey
0
+
g
(z)
=
x cos z.
z2
z2

Therefore, we can pick g 0 (z) = 0 and so the potential function is


=

ey
x2 y
x sin z +
+ constant.
2
z

Theorem 2.22 Necessary and sufficient conditions for a vector field to be a gradient vector.
~ for
Let F~ be a C 1 vector field defined on an open simply connected set U , a subset of R2 . Then F~ = f
some function f if and only if
F2
F1
=
.
x
y
~ F~ = ~0.
In R3 we require that
Theorem 2.23 Generalization of the Fundamental Theorem of Calculus.
Let f : Rn R be a C 1 function and let ~g : [a, b] Rn be a piecewise C 1 path. Then
Z
~ d~x = f (~g (b)) f (~g (a)).
f
~g

Proof:
By the definition of a line integral we have
Z
Z b
~
~ (~g (t)) ~g 0 (t) dt.
f d~x =
f
~g

But observe that the chain rule yields,


d
~ (~g (t)) ~g 0 (t).
(f (~g (t))) = f
dt
Therefore, we obtain our desired result,
Z
Z b
~ d~x =
~ (~g (t)) ~g 0 (t) dt.
f
f
~g

d
(f (~g (t))) dt.
a dt
= f (~g (b)) f (~g (a)).
=

Observe that to prove the above we needed to use the Fundamental Theorem of Calculus for one variable
and the chain rule for paths.
Corollary 2.24 As above but if ~g is a closed path then
I
~ d~x = 0.
f
~g

Note: we typically denote the line integral over a closed path with a circle on the integral, as used above.
28

example 2.25 Compute the work of the electrostatic force,


Qq ~r
F~ =
,
40 krk3
acting on a change q that moves from point A to point B along the path ~g .
Solution:
Easy to show if we realize that
~ (~r),
F~ = V

with

V (~r) =

Qq 1
.
40 k~rk

Then the work is computed as follows:


Z

F~ d~x,
Z
~ (~r) d~x,
= V

W =

~g

~g
~
r(b)

= V (~r)|~r(a) ,
= V (~r(a)) V (~r(b)),


1
1
Qq

.
=
40 k~r(a)k k~r(b)k
Lecture 11: AMATH 231-F14-001

October 1, 2014

Theorem 2.26 Properties of a gradient vector field


Let F~ be a C 1 vector field defined on an open, connected set U , a subset of Rn . Then the following are
equivalent:
~
a) F~ is a gradient field (F~ = )
b) For any oriented simple closed path ~g ,
I

F~ d~x = 0.

~g

c) For any two orientated simple curves ~g1 , ~g2 , having the same initial and final positions,
Z
Z
~
F d~x =
F~ d~x
~g1

~g2

If U is simply connected then the above are equivalent to


d)

F2
F1
=
x
y
and the analogue for R3 .
29

The equivalence of a) and c) is the First Fundamental Theorem of Line Integrals.


Proof:
a) = b)
~ in D and say C is parameterized by ~g (t) for
This is equivalent to a previous Corollary. Assume F~ =
a t b and then evaluate the line integral,
Z b
I
~
~ ~g 0 (t) dt,
F d~x =

~g

d
((~g (t))) dt,
a dt
= (~g (b)) (~g (a)),
=

= 0,
since the curve is assumed to be closed.
b) = c)
Suppose we have any orientated simple closed curve C. We can split it up into C = C1 (C2 ). Then
I
0 = F~ d~x,
Z
Z~g
~
F~ d~x,
F d~x
0=
C2
Z
ZC1
F~ d~x =
F~ d~x.
C1

C2

Therefore, the line integral is path independent.


c) = a)
Suppose that the line integral is path independent. Call the initial and final points P0 = (x0 , y0 ) and
P = (x, y), respectively. Then decompose the curve C into C1 C2 where C1 is the line from (x0 , y0 ) to
(x0 , y) and C2 is from (x0 , y) to (x, y). The two paths can be written as,
y0 t y,

~g1 (t) = (x0 , t),

x0 t x.

~g2 (t) = (t, y),

We can then define the following potential function and also decompose the line integral into the line
integral over each subpath
Z
Z
Z
~
~
(x, y) =
F d~x =
F d~x +
F~ d~x.
C

C1

C2

We can evaluate the integrals to obtain,


Z y
Z x
(x, y) =
(F1 , F2 ) (0, 1)dt +
(F1 , F2 ) (1, 0)dt,
y0
x0
Z y
Z x
(x, y) =
F2 (x0 , t) dt +
F1 (t, y) dt.
y0

x0

If we then compute the x partial derivative we get,

= F1 (x, y).
x
30

To show this for the y partial derivative we need to construct a path that first goes along x and then along
y. Please try it and convince yourself that you understand how this works.
Therefore, we can show that F~ is conservative since we can write it as the gradient of a function.
Part d) follows from the definition of a conservative function if you make the additional assumption.

2.4

Greens Theorem

Theorem 2.27 Let D be a bounded subset of R2 whose boundary D is a piecewise C 1 simple closed
curve oriented counter clockwise (positive). If F~ = (F1 , F2 ) is C 1 on D D then

ZZ 
I
F
F
2
1
F~ d~x =

dxdy
x
y
D
D
Note: Greens Theorem relates a line integral around the perimeter of the domain to an area
integral within the domain. What are these quantities that we are summing up?
Suppose we can describe D in the following form,
f (x) y g(x),
a x b,
and H(x, y) is continuous function on D D then the double integral over the domain can be
written as,
#
Z b "Z y=g(x)
ZZ
H(x, y) dy dx.
H(x, y) dxdy =
D

y=f (x)

x=a

This will be useful in the proof.


Proof of Greens Theorem:
First approach that Greens Theorem has special cases for (F1 , 0) and (0, F2 ). If we can prove
each of them independently, then we can add them to get the total result.
To do this we need to evaluate the line integral. Let us suppose that it can be bounded in
two different ways. The first is as mentioned before. That is to say we assume that C1 and C2
parameterize the bottom and top of the region, respectively, then we get,
~g1 = (x, f (x)),
~g2 = (x, g(x)),

= ~g10 = (1, f 0 (x)),


=

~g20

= (1, g (x)),

on

axb

on

axb

where on both curves we have a x b. This parameterization implies that D = C1 (C2 ).


Alternatively, we could assume that we have left and right bounds on the functions
~g3 = (h(y), y),
~g4 = (k(y), y),

= ~g30 = (h0 (y), 1),


=

~g20

= (k (y), 1),

31

on

c y d,

on

c y d.

For the first part, consider the special case with F~ = (F1 , 0):
I
Z
Z
~
~
F d~x =
F d~x
F~ d~x,
C1
b

C2

(F1 (x, g(x)), 0) (1, g 0 (x))dx,


(F1 (x, f (x)), 0) (1, f (x))dx
a
a
Z b
Z b
=
F1 (x, f (x))dx
F1 (x, g(x))dx,
a
a
Z b
[F1 (x, f (x)) F1 (x, g(x))] dx.
=
0

But observe that the following double integral takes on the same form, after some manipulation,
#
ZZ
Z b "Z g(x)
F1
F1

dxdy =
dy dx,
y
D
a
y=f (x) y
Z b
[F1 (x, g(x)) F1 (x, f (x))] dx,
=
a
Z b
[F1 (x, f (x)) F1 (x, g(x))] dx,
=
a
I
=
F~ d~x.
D

In the above equation we used the previous result just before to obtain the desired result.
The other case we consider is F~ = (0, F2 ):
Z
Z
I
~
~
F~ d~x,
F d~x
F d~x =
C1
d

C2

(0, F2 (k(y), y)) (k (y), 1)dy

=
c

(0, F2 (h(y), y)) (h0 (y), 1)dy,

c
d

[F2 (k(y), y)) F2 (h(y), y)] dy.

=
c

But we can rewrite the other double integral as


#
ZZ
Z d "Z k(y)
F2
F2
dydx =
dx dy,
D x
c
x=h(y) x
Z d
=
[F2 (k(y), y) F2 (h(y), y)] dy,
c
I
=
F~ d~x.
D

In the above equation we used the previous result just before to obtain the desired result.
By linearity we can sum these two results and get the theorem as stated above.

32

Lecture 12: AMATH 231-F14-001

October 3, 2014

example 2.28 Verify Greens Theorem if F~ = (x2 xy, xy y 2 ) if D is the triangular region with
vertices (0, 0), (1, 1), (2, 0).
Solution:
Label the curves on the bottom, right and left as C1 , C2 , C3 , respectively. Note that the boundary
is the union of the three, D = C1 C2 C3 .
LHS:
I
Z
Z
Z
F~ d~x =
F~ d~x +
F~ d~x +
F~ d~x
D

C1

C2

C3

We must parameterize each path,


C1 : ~g1 (t) = (t, 0)

0 t 2,

C2 : ~g2 (t) = (2 t, t)

0 t 1,

C3 : ~g3 (t) = (1 t, 1 t)

0 t 1.

We evaluate each integral,


Z

F~ d~x =

F~ d~x =

F~ d~x =

~g1

~g3

2
8
1 3
(t , 0) (1, 0)dt = t = ,
3 0 3
2

~g2

4
((2 t)2 (2 t)t, (2 t)t t2 ) (1, 1)dt = ,
3

(0, 0) (1, 1)dt = 0.


0

Therefore, the final answer is the sum of the three integrals,


I
4
F~ d~x = .
3
D
RHS:
Since the boundary is piecewise C 1 and a simple closed curve and F~ is C 1 on D D we can
use Greens theorem,

I
ZZ 
ZZ
F2 F1
~
F d~x =

dxdy
(y + x) dxdy.
x
y
D
D
D
We can parameterize the area with,
y x 2 y,

on

33

0 y 1.

This allows us to set up the double integer,


ZZ
Z 1 Z
(y + x) dxdy =
D

2y


(y + x) dx dy,

2y

1 2
dy,
=
xy + x
2
0
y

Z 1
1
1 2
2
2
(2 y)y + (2 y) (y + y ) dy,
=
2
2
0
Z 1


1 y 2 dy,
=2
0

1
1 3
4
=2 y y
= .
3
3
0
Z

Lecture 13: AMATH 231-F14-001

October 6, 2014

Corollary 2.29 If we make all the assumptions of Greens theorem and further assume that,
F2 F1

= 0,
x
y
or F~ is a conservative field for all points in D then,
I
F~ d~x = 0.
D

As we have already shown but now we verify with Greens Theorem.


example 2.30 Define U = R2 {(0, 0)} and

F~ (~x) =
Show that

y
x
,
x2 + y 2 x2 + y 2


.

F2 F1

= 0,
x
y

but that

F~ d~x 6= 0,

if we define D to be the perimeter of the unit circle.


Solution:
First we verify that the difference of partial derivatives is zero,
F2 F1
(x2 + y 2 ) 2x2 (x2 + y 2 ) 2y 2
0

=
+
= 2
= 0.
2
2
2
2
2
2
x
y
(x + y )
(x + y )
(x + y 2 )2
To compute the line integral we define the path of the unit circle ~g (t) = (cos t, sin t) on 0 t
2 and evaluate the line integral,
I
Z 2
Z 2
F~ d~x =
( sin t, cos t) ( sin t, cos t) dt =
dt = 2 6= 0.
0

34

Question: Is Greens Theorem wrong?


No! The conditions require that the boundary is a piecewise C 1 simple closed curve and that
F~ is C 1 on the boundary and interior. The problem here is that the origin is not included and
therefore the domain is not simply connected.
If C is any simple closed curve that has (0, 0) in its interior than one can show that
I
F~ d~x = 2.
D

If C is any simple closed curve that does not contain (0, 0) then,
I
F~ d~x = 0.
D

~ for the potential


At all points except (0, 0) we have that F~ is conservative and therefore F~ =
function
y
= arctan
.
x
We can only apply Greens Theorem to curves that do not contain the origin.
example 2.31 Define F~ (~x) = (2xy, 1 + x2 ) and
a) Test that F~ is conservative
b) If yes, find the potential function.
Solution:
a) F~ is C 1 everywhere because it consists of polynomials.
For any open simply connected set U a subset of R2 ,
F2 F1

= 2x 2x = 0.
x
y
Thus, F~ is a conservative field.
~ and therefore,
b) Define the potential function F~ =

= 2xy,
x

and

= 1 + x2 .
y

We integrate the first equation and get = x2 y + f (y). When we compute the partial
derivative with respect to y we then deduce that x2 + f 0 (y) = 1 + x2 . Therefore we require
f (y) = y and a potential function is,
= x2 y + y.
The level curves of (curves where is constant) are called equipotential lines.

35

2.5

Vorticity and Circulation

Definition 2.32 Given a C 1 field F~ : R2 R2 and a simple closed path C defined by ~g : [a, b] R2 , the
line integral,
I
= F~ d~x,
~g

is called the circulation of F~ around ~g .


Physical Interpretation of Circulation:
Suppose that we pick F~ = ~u to be the velocity of an object. The circulation is the line integral of
the velocity dotted with the tangent vector. The quantity that is integrated is


~
F d~x = F~ (~g (t)) ~g 0 (t)dt.
~g

If the velocity is always orthogonal to the path then the circulation is zero. If the velocity is
tangent to the path then the circulation is largest. Therefore, the circulation is the tendency for the
velocity field to flow (circulate) counter clockwise around the path.
The three possible cases are:
a) If > 0

fluid moves counter clockwise.

b) If < 0

the fluid moves clockwise.

c) If = 0

there is no net circulation.

If the circulation is zero that does not imply there is no movement along the boundary. It simply
means that there is no net movement. There could be regions of clockwise movement but they
are essentially balanced out by regions of counter clockwise movement.
Definition 2.33 The vorticity of a vector field F~ = (u(x, y), v(x, y)) is


v u
=

,
x y
which, if F~ is the velocity, is the tendency for objects in the flow to rotation counter clockwise. We will
show that the vorticity is interpreted to be the circulation per unit area.
Physical Interpretation of Vorticity:
The vorticity (or rotation rate) of a fluid is defined as the sum of the rotation rate of two orthogonal lines. By convention we choose the vorticity to be positive if the rotation counter (anti)
clockwise. Figure 1 illustrates two initially orthogonal lines each of which is deformed due to
the orthogonal component of velocity. If we consider the vertical line for instance, and if u1 is
uniform on this line there will be a translation but no rotation. Its only due to the vertical shear
of u1 do you get a component of rotation from this axis. A similar argument applies to the horizontal line which implies that it is the horizontal gradient in the vertical velocity, that induces
rotation. Each of these act in opposite directions however and a typical outcome after a short
time is illustrated in Figure 2.
36

Figure 1: Two orthogonal lines are shown in the xy-plane and how gradients in the velocities
can give rise to local rotation, and thus vorticity.

Figure 2: This is what could transpire after some short time if the gradients in the previous plots
are both positive.

37

Following this argument, we observe that the vorticity about the z-axis, which we denote
with 3 is
d d
+
dt
dt





u1
1
1
1
u2

x2 dt +
x1 dt
=
dt x2
x2
x1 x1
u2 u1

.
=
x1 x2

3 =

Analogously, we can define a vorticity component about the other two orthogonal axes. This
motives the following definition.
Vorticity exists if the vector field changes in space Suppose that F~ = (u, v) is a velocity and
consider cases with constants , :
(u, v) = (0, x)

= ,

(u, v) = (y, 0)

= ,

(u, v) = (y, x)

= + .

The vorticity is a sum of the two and is the sum of the tendency for the fluid to rotate counter
clockwise due to both components of velocity.
Suppose the velocity field is V~ (x, y) = (u(x, y), v(x, y)). If V~ is C 1 in a small disk of radius
  1 denoted by D with boundary D , then Greens theorem implies

ZZ 
I
v
u

dxdy,
V~ d~x =
x y
D
D


v u

=
(P )area(D ).
x y
In the above we used the Mean Value Theorem to approximate the double integral with the
value of the integrand at point P , somewhere in the disk. We can divide by the area and invert
this to get,
H


V~ d~x

v u
D

= 2.
(P ) =
x y
area(D )

Therefore, we have that the circulation is equal to the vorticity times the area contained within.
Physical Interpretation of Greens Theorem:
Suppose F~ is the velocity field of a fluid, say the velocity in your sink, then Greens theorem
states,

I
ZZ 
F
F
2
1
F~ d~x =

dxdy
x
y
D
D
The LHS is the line integral of the velocity field along a boundary D. It computes the
tangental component of velocity and sums it up over all parts of the boundary. This yields
the circulation of F~ along the boundary D.
The RHS is the double integral of the vorticity over the region D, which is the area contained within the boundary D. This computes the vorticity at every point in the domain
and then integrates (sums) this up over the entire domain.
38

Combining these two ideas yields the following interpretation for Greens theorem: The
total vorticity (like rotation) in D is equal to the circulation along the boundary D.

Figure 3: A cartoon to try and illustrate what Greens Theorem states.


example 2.34 Pick ~v = (y, x) with > 0 with units of 1/time. Previously, the field (flow) lines have
been shown to be x2 + y 2 = c2 , for some constant Lc.
At (1, 0) the velocity is upwards and therefore the rotation is counter clockwise.
The vorticity is easily computed,
=

v2 v1

= + = 2 > 0.
x
y

By Greens theorem the circulation is,



I
ZZ 
F2 F1
~
=
F d~x =

dxdy = 2c2 > 0.


x
y
D
D
y
x
example 2.35 Pick ~v = ( x2 +y
2 , x2 +y 2 ) on (x, y) 6= (0, 0) with > 0 with units of 1/time. The flow
lines are the same as before, x2 + y 2 = c2 , for some constant c.

39

At (1, 0) the velocity is upwards and therefore the rotation is counter clockwise.
The vorticity is computed to be zero,
=

v2 v1

= 0.
x
y

We can compute the circulation around a circle of radius c directly from the definition,
I
=
~v d~x,
D
Z 2
1
( sin t, cos t) (c sin t, c cos t) dt,
=
c
0
Z 2
1 dt,
=
0

= 2.
It can be shown that if we compute the circulation around any closed counter that contains
the origin, then we will get a non-zero value. This is in contrast to the circulation around any
closed contour that does not contain the origin where the result is zero,

I
ZZ 
F
F
2
1
=
F~ d~x =

dxdy = 0.
x
y
D
D
The distinction in the two cases comes form the fact that the velocity is singular at the origin.

40

Lecture 14: AMATH 231-F14-001

October 8, 2014

Surfaces and Surface Integrals

3.1

Parameterized Surfaces

Some discussion and Figure 3.1 form the course notes.


3.1.1

Surfaces as Vector-Valued Functions

Definition 3.1 A parameterized surface is a map ~g : D R2 R3 given by,


~g (u, v) = (x(u, v), y(u, v), z(u, v))
Surface is denoted by and is the image of ~g . If ~g is C 1 we call a C 1 -surface. Each component is C 1 .
example 3.2 Given f : D R where D R2 such that z = f (x, y), each (x, y) in R2 has a height
f (x, y). We can parameterize this surface as,
~g (u, v) = (u, v, f (u, v)).
Draw a picture like that of Figure 3.1 that shows the xyz space with D in the xy plane with (x, y)
and the surface above with f (x, y) and some curvature to the surface.
example 3.3 Surfaces can also be defined implicitly by h(x, y, z) = 0, such as x2 + y 2 + z 2 = a2 .
3.1.2

The Tangent Plane

We can fix one of the two parameters. For example if we set u = u0 , then we get a curve that
passes through (u0 , v0 ),
~
(v)
= ~g (u0 , v) = (x(u0 , v), y(u0 , v), z(u0 , v)) .
Or, we can set v = v0 and get a different curve that also passes through (u0 , v0 ),
~ (u) = ~g (u, v0 ) = (x(u, v0 ), y(u, v0 ), z(u, v0 )) .
Given what we already know we can find tangent vectors to the surface:
~g
T~u : ~ 0 (u) =
(u, v0 ),
u
~g
T~v : ~ 0 (v) =
(u0 , v).
v
These two (usually) nonparallel vectors determine the tangent plane to the surface.
A normal vector to the surface at (u0 , v0 ) is
~ (u0 , v0 ) = T~u (u0 , v0 ) T~v (u0 , v0 ),
N
~ (u0 , v0 ) = ~ 0 (u0 ) ~ 0 (v0 ).
N
41

The unit normal vector can be written as,


T~u T~v
~n

=
~

~
T

T
u
v
Draw a picture of a surface with tangent vectors Tu , Tv .
We require that the two tangent vectors are linearly independent otherwise this breaks down.
For a counter example consider the parameterization of a cone ~x = ~g (u, v) = (u cos v, u sin v, u).
Compute tangent vectors and consider what happens at u = 0.
example 3.4 Given ~g (u, v) = (u, v, f (u, v)) with (u, v) D find two tangent vectors to the surface and
a normal vector.
Solution:
We use the formulas above,




~g
f
~g
f
= 1, 0,
, and
= 0, 1,
.
u
u
v
v
Finally, the normal vector is the cross product of these two,
~ = ~g ~g ,
N
u v

i j k
,
= 1 0 f
u
f
0 1 v


f
f
= , ,1
u v
We can check this with what we learned in MATH 237. If we define h(x, y, z) = z f (x, y) = 0 then a
normal vector is found by the 3D gradient operation,


f
f
~
h = , , 1
u v
To find the equation of the tangent plane we use the fact that any line in the tangent play is necessarily
orthogonal to the normal. Therefore, if ~g (u0 , v0 ) = (u0 , v0 , f (u0 , v0 )) then
~ (~x ~g (u0 , v0 ))
0=N


f
f
0 = , , 1 (u u0 , v v0 , z f (u0 , v0 )),
u v
f
f
z = f (u0 , v0 ) + (u u0 ) (u0 , v0 ) + (v v0 ) (u0 , v0 ).
u
v
The equation of the tangent plane can be written as,
z = f (u0 , v0 ) + (u u0 )

f
f
(u0 , v0 ) + (v v0 ) (u0 , v0 ).
u
v

42

Lecture 15: AMATH 231-F14-001


3.1.3

October 10, 2014

Surface Area

Consider a surface parameterized by ~x = ~g (u, v) which is C 1 . The surface area can be built up
using the following steps:
Partition our domain into small rectangles Duv using the standard Cartesian grid. This then
partitions the surface into small surface elements (Draw a Figure like 3.6).
If u and v are sufficiently small then the surface element is very similar to a plane and
the edges are parallel vectors. Therefore, we can approximate the surface element with a
small plane parallelogram.


~ ~
~
~
Recall: the area of a parallelogram defined by the vectors A and B is A B , that is to
say the magnitude of the cross product,


~ ~ ~
A B = A kBk sin ,
where is the angle between the two vectors.
Using our linear approximation we can obtain,
~ (u) ~g (u0 , v0 ),
A
u

and

~ (v) ~g (u0 , v0 ),
B
v

Therefore, the surface area, S, is approximately,




~g

~
g
uv
S
(u
,
v
)

(u
,
v
)
0
0
0
0
u

v
Next we sum over all the surface elements determined by the partition we have chosen
and finally take the limit as N and u, v 0.
This motivates the following definition for the surface area.
Definition 3.5 The surface area of the surface ~x = ~g (u, v), (u, v) Duv , where ~g is C 1 , is defined by

ZZ
~g ~g


S=
u v dudv.
Duv
Locally, this computes the tangent vectors at a point, the surface area of the parallelogram and then
integrates that over all the points.
example 3.6 Calculate the surface area of the cone of radius b and height h. See example 3.4 of the course
notes. The surface is given by
~g (r, ) = (br cos , br sin , hr)
with (r, ) Duv = {(r, ) : 0 r 1,

0 2}.
43

Solution:
We compute the tangent vectors and their cross product:
~g
= (b cos , b sin , h),
r
~g
= (br sin , br cos , 0),

~ = r(bh cos , bh sin , b2 ).


N
The norm the unit vector is

b2 + h2 br. Therefore, the surface area is,


ZZ
S=
1d,

Z 1 Z 2
=
b2 + h2 br ddr,
0 0

= b2 + h2 b[r2 ]10 = b b2 + h2 ,

which is the surface area of a cone. Note that at r = 0, the point we have that one of the tangent
vectors is the zero vector. That does not have a direction, and therefore where we cannot find a
normal vector.

44

Lecture 16: AMATH 231-F14-001

October 15, 2014

example 3.7 Suppose ~g : Duv R3 is defined by


~g (u, v) = a(sin u cos v, sin u sin v, cos u),
with Duv = {(u, v)|0 u , 0 v 2}.
Verify that x2 + y 2 + z 2 = a2 . [Parameterization of the sphere of radius a centred at the origin.]
Then compute the surface area of the sphere.
Solution:
We check the algebraic identity:
x2 + y 2 + z 2 = a2 sin2 u cos2 v + a2 sin2 u sin2 v + a2 cos2 u,
= a2 sin2 u + a2 cos2 u,
= a2 .
Before we can use the formula for surface area we need to compute the tangent vectors and the
normal vector,
~g
= a(cos u cos v, cos u sin v, sin u),
u
~g
= a( sin u sin v, sin u cos v, 0),
v
when then yields,
~g ~g

= a2 (sin2 u cos v, sin2 u sin v, sin u cos u) = (a sin u)~x.


u v
Therefore, the surface element becomes,


~g ~g
2


u v = (a sin u) k~xk = a sin u.
Finally, we substitute into the formula for the surface are,
ZZ
Z 2 Z
d =
a2 sin u dudv,

= 2a [ cos u]0 ,
2

= 4a2 .
Of course this confirms with what we already know.
3.1.4

Orientation of Surface

Given a surface described by ~x = ~g (u, v) on (u, v) Duv where ~g is one to one and C 1 and the
normal vector is,
~ = ~g ~g ,
N
u v
45

is non-zero for all (u, v) Duv . Then the normal vector varies continuously and if it starts
pointing on one side and move it around it cannot ever point in the other direction at the same
point. Surfaces where you can have this peculiar behaviour are non-orientable. An example is a
Moebius band.
We will assume that our surfaces are orientable.
When dealing with a surface with boundary is a piecewise smooth closed curve, we
relate the orientation of the two. In particular, when viewed form the side of on which the
normal vector points, the boundary is orientated counter-clockwise. [Kind of like the right
hand rule for vectors.]

3.2
3.2.1

Surface Integrals
Scalar Fields

Suppose we have that the density of a thin fluid on a surface is with units of mass/unit area,
then can we figure out what is the mass of the fluid?
We can revisit our derivation of surface area and make one minor modification. Instead of
simply computing the surface area on each surface element, we would like to compute the mass,
which is the product of the local density times the surface area. Following the same rational as
before motivates the following definition.
Definition 3.8 If a surface R3 is parameterized by
~g (u, v) = (x(u, v), y(u, v), z(u, v)),

(u, v) Duv R2 ,

and f : R is a continuous function. Then, the surface integral of f over is




ZZ
ZZ
~g ~g

dudv.
f d =
f (~g (u, v))


u
v

Duv
If f has units of stuff/unit area then the surface integral of f is going to have units of stuff.
Compare this to the line integral of a scalar function from before,
Z

Z
f ds =

~g

f (~g ) k~g 0 (t)k dt.

On the LHS, we change the line for a surface and a ds for a d. On the RHS we change the norm
of the derivative of the path to the norm of the cross product of tangent vectors. Yes, this is more
work but it is doable!
example 3.9 Example 3.5 of the course notes considers a sphere and asks to compute the surface integral
of z 2 . There calculations are slightly longer but its a good exercise that I encourage you to go through to
get more practice.
RR p
example 3.10 Find x2 + y 2 + 1 d where is the helicoidal surface,
~g (u, v) = (u cos v, u sin v, v),

46

0 u 1,

0 v 2.

Solution:
We find the tangent vectors and normal vector:
~g
= (cos v, sin v, 0),
u
~g
= (u sin v, u cos v, 1),
v
~ = (sin v, cos v, u).
N
Then we compute the surface integral,
ZZ p
Z 1 Z 2

x2 + y 2 + 1 d =
u2 + 1 1 + u2 dvdu,

Z 2
Z0 1 0
2
dv,
(u + 1) du
=
0

1
= [ u3 + u]10 2,
3
8
= .
3
Theorem 3.11 Surface Integrals are independent of Parameterization
Let
RR be a smooth surface and let g : Duv R be a real-value continuous function. Then surface integral
f d does not depend on the parameterization of the surface, proved the parameterization is smooth.

Lecture 17: AMATH 231-F14-001

October 17, 2014


RR

example 3.12 Find the formula for the integral h d where is determined by the smooth function
z = f (x, y) defined on Duv R2 and h(x, y, z) : R3 R is continuous.
Solution:
The surface is parameterized by ~g (x, y) = (x, y, f (x, y)). Previously, we found the surface element for this type of function. Then, all we have to do is evaluate the integrand on the surface
and multiply by the surface element, to obtain
s
 2  2
ZZ
ZZ
f
f
h d =
h(u, v, f (u, v)) 1 +
+
dA.
u
v
Duv

example 3.13 Find the surface area of a torus denoted by


~g (u, v) = ((R + cos v) cos u, (R + cos v) sin u, sin v),

0 u, v 2.

where R, are constants. R denotes the radius of the torus from the centre and is the radius along each
segment.
Solution: We compute the tangent vectors and the normal vector,
~g
= ((R + cos v) sin u, (R + cos v) cos u, 0),
u
~g
= ( sin v cos u, sin v sin u, cos v),
v
~ = (R + cos v) (cos u cos v, sin u cos v, sin v) .
N
47

The norm of the unit vector is



p
~
N = (R + cos v) cos2 v + sin2 v = (R + cos v).
Therefore, we can now evaluate the surface area,
Z 2 Z 2
(R + cos v) dvdu,
S=
0
0

Z 2 Z 2
(R + cos v) dv du,
=
0
0
Z 2
[(Rv + sin v)]2
=
0 du,
0
Z 2
2Rdu,
=
0

= 4 2 R.
3.2.2

Vector Fields

Suppose that F~ is a velocity field defined in an around a surface that has unit outward normal
~n
. The dot product of these two vectors
F~ (~g (u, v)) ~n
(~g (u, v)),
projects the velocity field normal to the surface and yield the component of the velocity moving
through the surface. We call this the flux of F~ through the surface at ~g (u, v). If we sum up this
quantity over an entire surface we get the total (net) outward flux of the velocity field through
the surface,
ZZ
F~ ~n
d.

When would we need this concept? Suppose we have water flowing and we want to find out
how much is flowing through a given region. That would be exactly what we described above.
We will considering some applications but first lets figure out how to compute this formally.
Definition 3.14 Consider a C 1 orientated surface given by ~g (u, v) with (u, v) D and a vector field
F~ continuous on the surface. The surface integral of F~ over is


ZZ
ZZ
~g ~g
~
~
~
F n
d =
F (~g (u, v))

dudv
u v

Duv
Note:


ZZ

F~ ~n
d =

ZZ

~g
u

F~ (~g (u, v))


~g
Duv
u

~g
v



~g ~g
dudv


~g u
v
v

and so the norms cancel and therefore why it doesnt appear in the surface integral of a vector field.
example 3.15 Compute the outward flux of F~ through the surface if F~ = (y, x, z 2 ) and the surface
is the helicoid ~g (u, v) = (u cos v, u sin v, v) with Duv = {(u, v) : 0 u 1, 0 v /2}
48

Solution:
From a previous example we have that
~g ~g

= (sin v, cos v, u).


u v
Now, we can compute the flux,
ZZ
ZZ
F~ ~n
d =
(u sin v, u cos v, v 2 ) (sin v, cos v, u) dvdu,

D
Z 1 Z /2
=
u + uv 2 dudv,
0


/2
1 3
uv + uv
=
du
3
0
0

Z 1
1  3
u + u
=
du
2 3
3
0

1
1 2 1 2  3
= u + u
2 2 6
3
0
3

= + .
4 48
Z

Lecture 18: AMATH 231-F14-001

October 20, 2014

example 3.16 Compute the outward flux of F~ through the surface if F~ = (xyz, 0, 0) and the surface
is the portion of the denoted by x2 + y 2 + z 2 = 4 in the first quadrant.
Solution:
The surface mapping for the sphere is
~g (u, v) = (2 cos v cos u, 2 cos v sin u, 2 sin v)
on
Duv = {(u, v) : 0 u /2, 0 v /2}
We compute the normal vector,
~g
u
~g
v
~
N
~
N

= (2 cos v sin u, 2 cos v cos u, 0),


= (2 sin v cos u, 2 sin v sin u, 2 cos v),

= 4 cos2 v cos u, cos2 v sin u, cos v sin v ,
= 4 cos v (cos v cos u, cos v sin u, sin v) .

49

We now can compute the flux,


ZZ
ZZ
~
~
F n
d =
(8 cos2 v sin v cos u sin u, 0, 0) 4 cos v (cos v cos u, cos v sin u, sin v) dvdu,

ZD Z
= 32
cos4 v sin v cos2 u sin u dudv,
D
Z /2
Z /2
2
cos u sin u du
cos4 v sin v dv,
= 32
0

/2 

/2
1
1
3
5
= 32 cos u
cos v
,
3
5
0
0
32
= .
15
RR
example 3.17 Compute F~ n
d with F~ = (x, 0, z) and is the surface of a cube bounded by
x = 0, y = 0, z = 0 and x = 1, y = 1, z = 1 and n
is the outward normal.


Solution:
If n
is parallel to x, y, z maybe it is simpler to evaluate the flux using F~ n
rather than using the
definition that we presented.
We parameterize each surface,
1 : n
1 = (0, 0, 1),
2 : n
2 = (0, 0, 1),

and
and

~g1 (u, v) = (u, v, 0)


~g2 (u, v) = (u, v, 1)

0 u, v 1
0 u, v 1

3 : n
3 = (0, 1, 0),

and

~g3 (u, v) = (u, 0, v)

0 u, v 1

4 : n
4 = (0, 1, 0),

and

~g4 (u, v) = (u, 1, v)

0 u, v 1

5 : n
5 = (1, 0, 0),

and

~g3 (u, v) = (0, u, v)

0 u, v 1

~g4 (u, v) = (1, u, v)

0 u, v 1.

6 : n
6 = (1, 0, 0),

and

With these it is easy enough to evaluate the line integrals over a vector field,
ZZ
ZZ
~
F n
1 d =
(u, 0, 0) (0, 0, 1) d = 0,
1
1
ZZ
ZZ
F~ n
2 d =
(u, 0, 1) (0, 0, 1) d = 1,
2
2
ZZ
ZZ
F~ n
3 d =
(u, 0, v) (0, 1, 0) d = 0,
3
3
ZZ
ZZ
F~ n
4 d =
(u, 0, v) (0, 1, 0) d = 0,
4
4
ZZ
ZZ
F~ n
5 d =
(0, 0, v) (1, 0, 0) d = 0,
5
5
ZZ
ZZ
F~ n
6 d =
(1, 0, v) (1, 0, 0) d = 1.
6

RR
Therefore the line integral over the whole cube is F~ n
d = 2.
From the above example we see that we can compute surface integrals of piecewise C 1 surfaces but it is a lot of work. Later, we will see that sometimes this can be rewritten in terms of a
volume integral. In that case we only have one integral that can be easier to compute.
50

example 3.18 Compute

RR

F~ n
d over given by,

~g (u, v) = (u, v, u2 + v 2 ),



D = (u, v)|u2 + v 2 4

and F~ = (y 3 , x3 , 3z 2 ).
Solution:
Compute the normal vector,
~g
= (1, 0, 2u),
u
~g
= (0, 1, 2v),
v
~g ~g

= (2u, 2v, 1).


u v
Now, we can compute the surface integral of the vector field,
ZZ
ZZ
~
(v 3 , u3 , 3(u2 + v 2 )2 ) (2u, 2v, 1) dudv,
F n
d =

Z ZD
=
(2uv(u2 + v 2 ) + 3(u2 + v 2 )2 ) dudv.
D

To evaluate this integral it us easier to transform to polar co-ordinates, Dr = {(r, )|0 r 2, 0 2}


Z 2 Z 2
ZZ
2r4 cos sin + 3r4 rdrd,
F~ n
d =
0
0

2
Z 2 
1 6
1 6
=
r cos sin + r
d,
3
2 0
0
Z
26 2
2 cos sin + 3 d,
=
6 0
2
26  2
=
cos + 3 0 ,
6
= 64.
3.2.3

Properties of surface integrals

Similar to line integrals, surface integrals posses the properties of linearity and additivity.
Also, if the surface is piecewise C 1 (the surface is a union of several surfaces who are each
C 1 ), the integral of can be decomposed into the integral over each component.
3.2.4

Applications

Definition 3.19 The centre of mass (or centre of gravity) of a surface is a point on which balances
when supported at that point.
The point need not be part of the surface. For example consider a ring of constant density. The
centre of mass is the centre, which is not part of the ring.
51

Mathematically, the three co-ordinates can be defined as


RR
x d
x = RR
,
d

RR
y d
y = RR
,
d

RR
z d
,
z = RR
d

where is the density of the surface. Note that in each case we are dividing by the total mass of
the surface.
These expressions are called moments and in particular they are the first moments since we
have linear functions of x, y, z in the numerator. In particular the numerators of the three expressions are the moments with respect to yz-, xz- and xy-planes, respectively. These are often
denoted with Myz , Mxz , Mxy .
Definition 3.20 The moment of inertial measures the rotational inertia of a body. That is the opposition
we feel when we try and change the rotation of the body about a particular axis. The moments of inertia
about the x, y, z axes are denoted by Ix , Iy , Iz and are the second moments, defined by,
ZZ
(y 2 + z 2 ) d,
Ix =
Z Z
Iy =
(x2 + z 2 ) d,
Z Z
(x2 + y 2 ) d.
Iz =

example 3.21 Find the moment of inertia of a torus about the z axis if the density is constant and equal
to one. Recall the torus can be defined as
~g (u, v) = ((R + cos v) cos u, (R + cos v) sin u, sin v),

0 u, v 2.

where R, are constants.


Solution:
Recall that we previously computed

p
~
N = (R + cos v) cos2 v + sin2 v = (R + cos v).
Therefore, we can compute the moment of inertia:
ZZ
Iz =
(x2 + y 2 ) d,

Z 2
Z 2
=
(R + cos v)3 dvdu,

Z0 2 0 Z 2
3
3
2
2
=
du
(R + 3R cos v + R(1 + cos 2v) + (1 sin v) cos v) dv
2
0
0

2
3
1
1 3
3
2
= 2 R v + 3R sin v + R(v + sin 2v) (sin v sin v)
2
2
3
0
 2

2
= 2 R 2R + 3 .
52

Lecture 19: AMATH 231-F14-001

October 22, 2014

Gauss and Stokes Theorems

In this chapter we present the basics of Vector Differential Calculus. But first we review indicial
notation and add one more idea.
Index Notation
Here we recall some indicial (tensoral) notation that we mentioned in the first week and define
a few more.
We write vectors ~a as ai , where i is a free index.
Dot product ~a ~b = ai bi .
Matrix multiplicaiton A~b = Aij bj .
The alternating tensor ink is defined as

1 if ijk = 123, 231, 312

ijk =
0 if 2 or more indices repeat

1 if ink = 321, 213, 132


Cross product: (sum over i and j but k is free)
(~a ~b)k = ijk ai bj

4.1
4.1.1

The vector differential operator


Divergence and curl of a vector field

Definition 4.1 Let f : U R3 R be a differential function. The gradient is defined as,


~ = f ex + f ey + f ez = f
f
x
y
z
xi
This is a vector where the coordinate in each direction corresponds to the partial derivative of the function.
Physical Interpretation of Gradient:
This we have seen before, in MATH 237, shows us the direction in which the field is increasing most rapidly. Think of it as the steepest part of a hill or valley.
Definition 4.2 Let F~ = (F1 , F2 , F3 ) : U R3 R3 be a differential function. The curl is,

ex

ey

ez

~ F~ =

F1 F2 F3

F3 F2 F1 F3 F2 F1

y
z z
x x
y

This is a vector quantity.


53


= ijk

Fj
xi

Physical Interpretation of Curl


If we consider the case where F~ = (u(x, y), v(x, y), 0) is a velocity field, then we see that there
is only one component of the curl that is nonzero, the vertical component is v/x u/y. This,
as we have seen before, is the vorticity of a two dimensional flow. What is different is that this is
a vector pointing in the z direction, which I point out is orthogonal to the xy-plane.
The other components are similarly defined but in different directions. For a velocity field
F~ = (u, v, w) then the x component of the curl is the vorticity about the x-axis due to motion
in the yz plane. Finally, the y-component of curl is the vorticity about the y axis due to the
~ F~ is the
motion in the xz-plane. That is why, if we assume F~ to be a velocity, then its curl,
three-dimensional vorticity, which shows us the local rotation rate about each axis.
Definition 4.3 Let F~ = (F1 , F2 , F3 ) : U R3 R be a differential function. The divergence is,



~
~
F =
, ,
(F1 , F2 , F3 ),
x y z


F1 F2 F3
+
+
,
=
x
y
z
Fi
=
.
xi
This is a scalar quantity, which is the sum of partial derivatives.
Physical Interpretation of Divergence:
This is something we have not seen before. To see what it does consider a couple of examples.
Suppose we have the two-dimensional case of F~ = (x, y). We saw that the field lines of this
consists of paths that are moving away from the original at an exponential rate. We can say the
paths are diverging. Note that in this case the divergence is 2. If we change the sign of the RHS
then we will have that the paths will converge to the origin, and the divergence is 2.
Instead consider, F~ = (y, x). The solutions to this vector field are circular paths. If we
compute the divergence we find it is zero.
These are just examples but they should hopefully give you some intuition about what the
divergence of a vector field means. It gives us the idea as to how the solutions are going to
spread (if divergence is positive), converge (if divergence is negative) or simply rotate around if
the divergence is zero.
Lecture 20: AMATH 231-F14-001
4.1.2

October 24, 2014

~
Identities involving

G1
~ + g) = f
~ + g
~
(f
G2
~ g) = f g
~ + g f
~
(f
G3
~ F~ G)
~ = (F~ )
~ G
~ + (G
~ )
~ F~ + F~ (
~ G)
~ +G
~ (
~ F~ ),
(
54

D1
~ (F~ + G)
~ =
~ F~ +
~ G
~

D2
~ (f F~ ) = f
~ F~ + f
~ F~

D3


~ F~ G
~ = ijk Fi Gj ,

xk
Fi
Gj
= Gj
ijk + Fi
ijk ,
xk
xk
Gj
Fi
kij + Fi
jki , permute indices
= Gj
xk
xk
~ (
~ F~ ) F~ (
~ G),
~
=G

C1
~ (F~ + G)
~ =
~ F~ +
~ G
~

C2
~ (f F~ ) = f
~ F~ + f
~ F~

C3









~
~
~
~
~
~
~
~
~
~
~
~
~
~
~
(F G) = G F F G + F G G F

Z1
~ (f
~ ) = ~0

Z2
~ (
~ F~ ) = 0

~ G
~ and (G
~ )
~ F~ . These can be rewritten in component
Note that above we saw terms like (F~ )
form as,



~ = (F1 , F2 , F3 )
F~
, ,
,
x y z

= F1
+ F2
+ F3 ,
x
y
z

= Fk
.
xi
~ operator does not commute in general. For example
It is very important to note that the
~ =
~ F~ .
F~
6
Definition 4.4 The Laplacian 2 f of f is defined as
2
~ (f
~ ) = f = f
2 f
xi xi
x2i

55

Another important identify is the curl of the curl:






~
~ F~ =
~
~ F~ 2 F~ .

~
The Laplacian commutes with :
L1
~ 2 f ) = 2 (f
~ )
(
L2
~ 2 F~ ) = 2 (
~ F~ )
(
L3
~ (2 F~ ) = 2 (
~ F~ )

~ and
~ F~ = 0 then 2 = 0, or
example 4.5 If F~ =
2 2 2
+
+ 2 = 0.
x2 y 2
z
That is to say that satisfies Laplaces equation.
Theorem 4.6 Helmholtz Decomposition
Let F~ be a C 2 vector field on a bounded volume V R3 . Then F~ can be decomposed into a curl-free and
a divergent-free component:
~ +
~ .
~
F~ =
~ as the potential function (scalar) and the stream vector.
We define and
Note that if compute the divergence and curl we get,
2 F = 2 ,
and
~ F~ =
~
~
~ = 2 ,
~

where we assume that the stream vector is divergence fee.


Physical Interpretation of Helmholtz decomposition:
This theorem, which is not in the AMATH 231 course notes, is a simple and deep physical
interpretation. It says that every vector field (e.g. velocity or force fields) can be thought of as a
divergent part and a rotational part. These two are distinct. A vector field could be only divergent, only rotational or a combination of the two. If turns out that if we impose the divergence
and curl of a vector field then we can determine the potential function and stream vector.

56

Lecture 21: AMATH 231-F14-001

October 27, 2014

Applications
~
In physics, electromagnetic fields are composed of electric fields E(x,
y, z, t) and a magnetic
~
field H(x, y, z, t). By choosing the units appropriately Maxwells equations read,
~
E
t
~
H
t
~
~
E
~ H
~

~ H
~ 4 J,
~
= c
~ E,
~
= c
= 4,
= 0.

Where  is the charge density, J~ is the current vector and c is the speed of light in a vacuum.
These are a famous example of a system of linear partial differential equations (PDEs). The
~ H,
~ J.
~ If we suppose that the current is set, then we have two
three variables that appear are E,
unknowns but four equations. Isnt that over prescribed? In this case no because we need to
determine the rotational and divergent part of each of the electric and magnetic fields.
4.1.3

~ in curvilinear coordinates
Expressing

We are very used to doing a lot of things in Cartesian coordinates. But, as we have previously
seen, there is more to life than Cartesian coordinates. For example, there are cylindrical coordinates and spherical coordinates. Those are just two. One can define whatever coordinate system
they want as long as the different variables are orthogonal in some sense.
In Cartesian coordinates our orthonormal basis is {
e1 , e2 , e3 } and the del operator takes the
form,
~ = e1 + e2 + e3

x1
x2
x3
A general curvilinear co-ordinate system (v1 , v2 , v3 ) in R3 is related to the Cartesian co-ordinate
system by,
x = f (v1 , v2 , v3 ),
y = g(v1 , v2 , v3 ),
z = h(v1 , v2 , v3 ).
Alternatively, this could be written in vector form as ~x = F~ (~v ).
If k, l, m are constants then the following are curves (dimension one),
x = f (v1 , l, m),
y = g(k, v2 , m),
z = h(k, l, v3 ).
But we know how to find unit tangent vectors of any curves, based on what we learned in
Chapter 1, and they are,
~
x
vi


ei =
~x
vi
57


~x
A convention is to define hi = v
for i = 1, 2, 3.
i
Note that above we have three unit vectors. If they are linearly independent, then they can
be used to span a basic for R3 . If they form a basis then any vector can be written as a sum of
vectors decomposed in terms of that basis.
For example, if we wanted to rewrite the gradient in terms of our new basis we can write,


~ = u
~ ei ei
u
Note that above we are using indical (tensoral) notation to sum up over i = 1, 2, 3.
We can also find the partial derivative of u with respect to each co-ordinate by using the
Chain Rule:
u xj
u
=
,
vi
xj vi
~ ~x ,
= u
vi
~
= hi ei u.
This can be rewritten as,

~ = 1 u .
ei u
hi vi

From this we can generalize to say that the gradient operator can be rewritten in terms of any
basis ei as,
~ = = 1 ei ,

xi
hi vi

~ = ex

+ ey
+ ez
= e1
+ e2
, + e3
,
x1
x2
x3
h1 v1 h2 v2 h3 v3
where we recall for completeness that,


~x

hi =
vi ,

and

ei =

Lecture 22: AMATH 231-F14-001

1 ~x
.
hi vi
October 29, 2014

Midterm.
Lecture 23: AMATH 231-F14-001

October 31, 2014

Cylindrical Coordinates
Draw a Picture:
x = cos ,
y = sin ,
z = z.

58

This can be written compactly as,


~x(~v ) = ( cos , sin , z),
where ~v = (, , z).
We derive the unit vectors and the gradient in cylindrical co-ordinates.
e =

1 ~x
1
= (cos , sin , 0) = (cos , sin , 0).
h
h

Note that in the above we used the fact that h = 1.


e =

1
1 ~x
= ( sin , cos , 0) = ( sin , cos , 0),
h
h

since = .
ez =

1 ~x
1
= (0, 0, 1) = (0, 0, 1),
hz z
hz

where hz = 1.
Therefore, putting this together we get that in cylindrical co-ordinates we have that the gradient operator can be written as,


1

~ = e
+ e
+ ez

.

z
Spherical Coordinates
Draw a Picture:
x = r sin cos ,
y = r sin sin ,
z = r cos .
This can be written as,
~x(~v ) = (r sin cos , r sin sin , r cos ),
where ~v = (r, , ).
We derive the unit vectors and the gradient in cylindrical co-ordinates.
er =

1 ~x
1
= (sin cos , sin sin , cos ) = (sin cos , sin sin , cos ).
hr r
hr

Note that in the above we used the fact that hr = 1.


e =

1 ~x
1
= (r sin sin , r sin cos , 0) = ( sin , cos , 0),
h
h

since h = r sin .
e =

1 ~x
1
=
(r cos cos , r cos sin , r sin ) = (cos cos , cos sin , sin ),
h
h
59

where h = r.
Therefore, putting this together we get that in cylindrical co-ordinates we have that the gradient operator can be written as,



1
1

~
= er
+ e
+
e
.
r r r sin
Divergence in Spherical Coordinates
Consider
F~ = Fr er + F e + F e .
We expand the divergence as we would any dot product, keeping in mind that the order matters,
and that the unit vectors can depend on the coordiantes


1
~
~
F =
ei
(Fj ej ) ,
hi vi
Fj
ej
1 Fj
ej + ei
.
= ei
hi vi
hi vi
Remember that we must sum over all the indices because each is repeated.
The first term is easy enough to evaluate because, by definition, our basis is orthonormal and
therefore ei ej = 0 if i 6= j and 1 if i = j,
ej
~ F~ = 1 Fi + 1 ei

Fj .
hi vi
hi vi
Together these can be written in vector form as,


1
1
1
F
F
F
1
2
3
~ F~ =
+
+

h1 v1
h2 v2
h3 v3



e1

e2

e3
1
F1 +
F2 +
+ e1
F3
h1
x1
x1
1



e1

e2

e3
1
F1 +
F2 +
F3
+ e2
h2
x2
x2
x2



e1

e2

e3
1
F1 +
F2 +
F3 .
+ e1
h3
x1
x3
x3
We return to the case of spherical coordinates. We use our definition for the unit vectors,
compute their derivatives, and then compute the dot products:

er

er

er
= e ,
= sin
e ,
= 0,

e
=
er ,
= cos e ,
= 0,

e
=0
= sin er cos e ,
= 0.

r
We can use these equations to obtain,


1 F
~ F~ = 1 r2 Fr + 1

(F sin ) +
.
2
r r
r sin
r sin
60

Lecture 24: AMATH 231-F14-001

4.2
4.2.1

November 3, 2014

Gauss Theorem
The Theorem

Theorem 4.7 Let be a bounded subset of R3 whose boundary, is a single piecewise smooth orientated closed surface. If F~ is C 1 on then
ZZ
ZZZ
~
~
F dV =
F~ n
d,

where n
is the unit outward normal to .
Note: The volume between 2 concentric spheres does not have a single piecewise smooth oriented closed surface.
Idea of Proof:
Assume a special volume of the form f` (x, y) z fu (x, y) with (x, y) Dxy .
If F~ = (F1 , F2 , F3 ) and n
= (
n1 , n
2, n
3 ) then the equation can be written as, (summing over
the is),
ZZZ X
ZZ X
3
3
Fi
Fi n
i d.
dV =
i=1 xi
i=1
This is equivalent to saying that for each component i we have,
ZZ
ZZZ
Fi
(Fi ei )
ni d.
dV =

xi
Note we are not using tensor notation in this equation.
We prove this for i = 3 (z direction), but the others can be shown analogously.
We begin with the LHS,
!
ZZZ
Z fu (x,y)
ZZ
F3
F3
dV =
dA,
z
z
f` (x,y)
Dxy
ZZ
=
[F3 (x, y, fu (x, y)) F3 (x, y, f` (x, y))] dA.
Dxy

Now we consider the RHS and divide the surface integral into an upper (1 ) and lower (2 )
surface. The surfaces can be parameterized as,
~x = ~g1 (x, y) = (x, y, fu (x, y)),
~x = ~g2 (x, y) = (x, y, f` (x, y)).
The normal vectors can be computed to yield,


fu fu
~g1 ~g1

=
,
,1 ,
x
y
x
y


~g2 ~g2
f` f`

=
,
, 1
x
y
x y
61

Note that we change the sign of the second normal vector to pick the outward normal, as stated
in the theorem.
The surface integral can be decomposed into two parts:
ZZ
ZZ
ZZ
(F3 e3 ) n
d =
(F3 e3 ) n
d +
(F3 e3 ) n
d.

The first integral can be computed as follows,




ZZ
ZZ
~g1 ~g1
(F3 e3 ) n
d =
(F3 (~g1 (x, y))
e3 )

dA,
x
y
1
Dxy
ZZ
F3 (x, y, fu (x, y)) dA.
=
Dxy

Similarly we find for the lower surface,




ZZ
ZZ
~g2 ~g2
(F3 (~g2 (x, y))
e3 )

dA,
(F3 e3 ) n
d =
x
y
2
Dxy
ZZ
=
F3 (x, y, f` (x, y)) dA.
Dxy

When we combine these we get,


ZZ
ZZ
(F3 e3 ) n
d =

(F3 (x, y, fu (x, y)) F3 (x, y, f` (x, y))) dA.


Dxy

This proves the desired result.


Physical Interpretation of Gauss Theorem
To make things concrete consider a sphere of radius  centred at a point ~x denoted by  with
boundary  . Gauss theorem states,
ZZZ
ZZ
~
~
F dV =
F~ n
d,


Draw a picture of a sphere that denotes the centre, radius, volume and boundary
By the mean value theorem there exists a ~c such that,
ZZZ
~ F~ dV =
~ F~ (~c)V ( ),

since F~ is a C 1 function. If we apply Gauss theorem and divide through by the volume and get
ZZ
1
~ F~ (~c) =

F~ n
dV.
V ( ) 
In the limit as  0 we must have by continuity that ~c ~x and therefore,
ZZ
1
~
~
F (~x) = lim+
F~ n
dV.
0 V ( )

62

This shows that the divergence at a point is equal to the flux per unit volume of F~ . That is to say
the rate at which stuff is moving away from the point.
~ F~ > 0(< 0) there is a net outward (inward) flux at the point.
If
~ F~ > 0(< 0) at a point means that the fluid is
If you consider F~ to be a velocity then
expanding (contracting).
~ F~ is the spreading (divergence) at a point per unit volume.
In summary,
~ ~
That means
RRR that F dV is the total spreading at a point.
~
~
Finally,
F dV is the total divergence over the volume .

Gauss theorem states that the total divergence in a volume is equal to the net flux through the
boundary of the volume. Again, this is a conservation principle as we saw in Greens theorem
but instead of conserving circulation we are conserving the flux of whatever F~ represents.
exampleRR4.8 If F~ = (2x, y 2 , z 2 ) and is the unit sphere centred at the origin defined by x2 + y 2 + z 2 = 1
evaluate F~ n
d.
Solution:
Since F~ is C 1 and is a single piecewise smooth orientated closed surface we can apply Gauss
theorem. Therefore, instead of computing the integral over the surface we can compute the
~ F~ =
volume integral of the divergence. First note that the divergence at any point in space is
2 + 2y + 2z. Then we can evaluate the integral,
ZZZ
ZZZ
ZZZ
ZZZ
~
~
2z dV,
2y dV +
2 dV +
F dV =

4
= 2( ) + 0 + 0,
3
= 0.
Thus, we conclude that

RR

F~ n
d = 83 .

example 4.9 Use Gauss theorem to evaluate

RR

(x2 + y + z) d if is the shape x2 + y 2 + z 2 = 1.

Solution:
To use Gauss theorem we must first find F~ such that F~ ~n = (x2 + y + z). Since we know that
the unit outward normal is (x, y, z) we must conclude that F~ = (x, 1, 1). It then follows that the
~ F~ = 1.
divergence is
The integral is straightforward to evaluate,
ZZ
ZZZ
4
2
(x + y + z) d =
1 dV = .
3

Lets revisit this example.


RR
example 4.10 Compute F~ n
d with F~ = (x, 0, z) and is the surface of a cube bounded by
x = 0, y = 0, z = 0 and x = 1, y = 1, z = 1 and n
is the outward normal.

63

Solution:
Using Gauss theorem this is a lot easier,
ZZZ
ZZ
~
~ F~ dV,
F n
d, =

ZZZ
=
2 dV,

= 2.

64

Lecture 27: AMATH 231-F14-001


4.2.2

November 10, 2014

Conservation Laws

Suppose (~x, t) is the density of a fluid [mass/volume] and is C 1 . The total mass of the volume
at time t is
ZZZ
(~x, t) dV.

Take ~j(~x, t) to be the flux of mass [rate of flow of mass/unit area]. By this definition, the flux
through the boundary is
ZZ
~j n
d,

if n
is the unit outwards normal. If it is positive (negative) the mass tends to leave (build) the
volume inclosed.
Conservation of Mass
In words, it can be stated as,

Rate at which
Rate at which
mass in = mass leaves .
increases
across
I realize that the negative on the RHS seems strange but it is a convention we use because of the
fact that we consider the outward normal vector.
What we wrote in words can now be translated in terms of mathematics,
ZZ
ZZZ
d
~j n
d.
dV =
dt

If we further assume that is fixed and is C 1 then the LHS can be rewritten as,
ZZZ
ZZZ
d

dV =
dV
dt
t

By Gauss theorem, we can rewrite the RHS of the conservation law in terms of a volume
integral,
ZZ
ZZZ
~ j dV.
~j n
d =

Recall that to apply Gauss theorem we need that ~j is C 1 and has no holes, that is to say the
boundary is a simple smooth closed surface.
It might seem that we have done a lot of work for nothing but thats not the case. Using the
two identities above we can combine these two terms into one since they both involve a triple
integral. We do exactly this below,
ZZZ
ZZ
d
~j n
dV =
d,
dt

ZZZ
ZZZ

~ j dV,
dV =


ZZZ 
~
+ j d = 0.
t
65

What we do next is a fundamental idea that appears again and again. If you get it once it will
make life much easier in serving equations of motion, as is done a lot in AMATH 353.
So far the only assumption that we made on is that it is constant. That means that this
equation applies to any volume. since and ~j are C 1 , the integrand is continuous. This then
implies that the integrand is zero everywhere,
~
+ j = 0.
t
You can argue this by contradiction. If this is non-zero at a point ~x, then by continuity, it must
be non-zero over an interval. If we pick to be a subset of that interval then we obtain that the
volume integral is non-zero and therefore the contradiction.
One simple and natural choice is to take the mass flux to be ~j = ~u, which is the product of
the density times the velocity, In this case the governing equation is
~
+ (~u) = 0.
t
This is called the continuity equation and is one of the equations that governs the motion of a
compressible fluid, such as air.
A second choice is that the flux is due to the diffusion of the motion as a result of molecular
~ then the resulting equation
bumping. If we assume Fouriers law, which states that ~j = k ,
can be written as,



~ k
~
= k
= k2 .
t
This is called the diffusion equation and describes how things diffuse due to the molecular motion
of molecules.
Lecture 28: AMATH 231-F14-001
4.2.3

November 12, 2014

The Generalized Divergence Theorem

Suppose that R3 with boundary , which is a simple piecewise smooth oriented closed
surface, and F~ is C 1 on except possibly at one point ~a and ~a
/ . If we surround ~a
by a surface H lying entirely in ~a then
ZZZ
ZZ
ZZ
~
~
~
F dV =
F n
d
F~ n
d.
H

Note: H = {~x|~x and ~x


/ H}.
Physically, this says that the divergence in a volume is equal to the amount going out the
inner and outer boundaries.
Theorem 4.11 Gauss Law
~
Suppose is as above and O
/ then,

ZZ
~r n

4
d =
3
0
r

if
if

~0
~0
/

where ~r is the position and ~n is the unit outward normal vector on . Observe that ~r/r3 is clearly C 1
everywhere except at ~0.
66

Proof:
We consider two cases:
(i) If ~0 then define a sphere of radius , call it H around ~0. By the generalized Gauss
theorem,
 
ZZZ
ZZ
ZZ
~r
~r n

~r n

dV =
d.
d.
3
3
3
r
H
r
H r
But note that the divergence becomes,
~

Moreover,

ZZ
H

~r
r3

~r n

d =
3
r

 
~ ~r

~ 1 ,
= 3 + ~r
r
r3
3
~r ~r
= 3 3 5 ,
r
r
= 0.
ZZ
H

~r ~r
d =
r3 r

ZZ
H

1
d
r2

On a sphere of radius r = , which is a constant, and so the upper integral is easily evaluated,
ZZ
~r n

1
d = 2 42 = 4.
3

H r
Therefore, we conclude the following,
ZZ

~r n

d = 4,
3
r

~ . That is to say the singularity is contained within the volume.


if O
(ii) This is like the first case except that we use the standard Gauss theorem and we get,
ZZ
~r n

d = 0.
3
r

4.3

Stokes Theorem

Recall that Greens Theorem states the following and only applies to functions in R2 .
Theorem 4.12 Let D be a bounded subset of R2 whose boundary D is a piecewise C 1 simple closed
curve oriented counter clockwise (positive). If F~ = (F1 , F2 ) is C 1 on D D then

I
ZZ 
F
F
2
1
F~ d~x =

dxdy
x
y
D
D
The three-dimensional analogue is called Stokes Theorem.

67

4.3.1

The Theorem

Theorem 4.13 An orientable surface R3 is enclosed by a piecewise C 1 simple closed curve and
F~ = (F1 , F2 , F3 ) is C 1 on then
I
ZZ 

~
~ F~ n
F d~x =

d.

Note that is positively oriented from the perspective of n


.
Special Case:
Consider the two-dimensional case with F~ = (F1 , F2 , 0) and ~g (u, v) = (x(u, v), y(u, v), 0) is a flat
surface in the xy-plane. Clearly a unit normal going through the surface is n
= (0, 0, 1) = k and
Stokes Theorem reduces to
I
ZZ 

~
~ F~ n
F d~x =

d,


ZZ 
F2 F1

dxdy.
=
x
y

But this is precisely Greens Theorem.


Therefore, we deduce that Stokes Theorem is a generalization of Greens Theorem to curved
surfaces. The physical interpretation extends as well. The circulation around the boundary is
equation to the sum of the vorticity in the surface contained within the boundary.
Proof of Stokes Theorem:(outline)
Rather than doing it in general we consider the special case where the surface is parameterized
~ y) = (x, y, f (x, y)) with (x, y) Dxy . The boundary projected onto the xy-plane takes the
by, S(x,
form ~h(t) = (x(t), y(t), 0) for t1 t t2 . That means that the actual parameterization of the
~ = (x(t), y(t), f (x(t), y(t))) for t1 t t2 .
boundary is S(t)
As we have seen many times, we normal vector to the surface is,

~
~  f
S
S
f

= , ,1 .
x
y
x y
We prove the theorem in three steps.
Step 1 Rewrite the RHS of the theorem
 

ZZ 
I 

F3 F2 F1 F3 F2 F1
f
f
~
~
F n
d =

, , 1 dxdy.
y
z z
x x
y
x y

Step 2 Given that we parameterized with ~g (t) = (x(t), y(t), f (x(t), y(t))) for t1 t t2 we can
get an expression for the LHS.


I
Z t2
dx dy f dx f dy
~
F d~x =
(F1 , F2 , F3 )(x(t), y(t), f (x(t), y(t)))
, ,
+
dt,
dt dt x dt
y dt

t1



Z t2 
f dx
f dy
=
F1 + F3
+ F2 + F3
dt,
x dt
y dt
t1
Z
~ d~x.
=
G
Dxy

68

where we have defined the vector valued function




f
f
~ = F1 + F3 , F2 + F3
G
.
x
y
Step 3 Apply Greens Theorem, which we have already proved, and simplify
 



Z
ZZ

f
~
G d~x =
F2 + F3

F1 + F3
dA,
y
y
x
Dxy
Dxy x
 


ZZ
F3 F3 f f
2f
F2 F2 f
+
+
+
+ F3
=
x
z x
x
z x y
xy
D
 xy
 


F1 F1 f
F3 F3 f f
f

+
F3
dA,
y
z y
y
z y x
xy


ZZ
F3 f
F1 F1 f
F3 f
F2 F2 f
+
+

dA,
=
x
z x
x y
y
z y
y x
Dxy

 

ZZ
F3 F2 F1 F3 F2 F3
f
f
=

, , 1 dA.
y
z z
x x
y
x y
Dxy
Note that this matches term by term with the expression in Step 1 and therefore we are
done.
Lecture 29: AMATH 231-F14-001

November 14, 2014

Corollary 4.14 If 1 and 2 are two oriented piecewise C 1 surfaces with 1 = 2 = C, where C is a
simple closed curve and F~ is a C 1 vector field then,
ZZ 
ZZ 


~ F~ n
~ F~ n

1 d =

2 d
1

Proof:
We can use Stokes theorem to show each side is equal to the circulation around the common
boundary,
I
F~ d~x.
C

Alternative Proof:
Consider the surface that is formed from the union of the two
= 1 (2 )
Then by the divergence theorem we have,
ZZ 
ZZZ



~ F~ n
~
~ F~ d = 0,

1 d =

since the divergence of the curl is always zero.


Therefore, we conclude
ZZ 
ZZ

~
~
F n
1 d =
1


2

69


~
~
F n
2 d

~ A
~ with
example 4.15 What is the flux of F~ =
~ = (2z y, x z, y x),
A
through the hemisphere 1 defined by x2 + y 2 + z 2 = a2 with z y 0.
Solution:
We define 2 to be the surface of the plane z = y that cuts through the sphere. A normal to the
plane is (0, 1, 1) and therefore a unit normal is 12 (0, 1, 1). Since it is easier to parameterize the
tiled plane we prefer to deal with that surface.
By the corollary we have,
ZZ 
ZZ 


~
~
~
~
F n
1 d =
F n
2 d
1

~ A
~ = (2, 3, 2) and the dot product of the curl with the unit norma is
The curl of the flux is



1
1
~
~
F n
2 = (0, 1, 1) (2, 3, 2) = .
2
2

Therefore, we can evaluate the flux using Stokes theorem rather easily,
ZZ 
ZZ

1
1
~
~
F n
1 d =
d = a2 .
2
2
1
2
~ and is a C 2 scalar
Definition 4.16 A vector field F~ can be written in terms of a gradient, F~ = ,
field that we call a scalar potential or a potential function.
Below is a proposition similar to what we saw before but now it is in three-dimensions.
~ and is C 2 then
Proposition 4.17 If F~ =
i)

F~ d~x = (~b) (~a),

for any curve C joining ~a and ~b.


ii)

F~ d~x = 0,

for a simple closed curve C.


iii)
~ F~ = ~0.

~ F~ = ~0 in U R3 and U is simply connected then there exists a C 2


Theorem 4.18 If F~ is C 1 and
~ in U .
scalar potential such that F~ =

70

Proof:
Given U and a surface contained within such that = C then by Stokes theorem,
I
ZZ 

~ F~ n
F~ d~x =

d = 0.

Since the circulation is zero for any curve C, by the proposition above, the line integral is inde~
pendent of the path and therefore we can construct a scalar potential such that F~ = .
~ A
~ and A
~ is C 2 in U R3 then
Proposition 4.19 If F~ =
i)

ZZ

F~ n
d =

~ d~x,
A

where is any surface such that = C, (i.e. the integral is surface independent).
ii)

ZZ

F~ n
d = 0

for any closed surface U .


iii)
~ F~ = ~0.

example 4.20 Calculate the work done along C by the force field,
F~ = (x + y 2 , y + z 2 , z + x2 ).
and C is the triangle with vertices (1, 0, 0), (0, 1, 0), (0, 0, 1).
Method 1:
Since C is piecewise C 1 , we can compute this directly be decomposing the curve into three
curves,
I
Z
Z
Z
~
~
~
F d~x =
F d~x +
F d~x +
F~ d~x.
C

C1

C2

C3

Next, we must parameterize the three lines,


~g1 (t) = (1, 0, 0) + (1, 1, 0)t = (1 t, t, 0),

0 t 1,

~g2 (t) = (0, 1, 0) + (0, 1, 1)t = (0, 1 t, t),

0 t 1,

~g3 (t) = (0, 0, 1) + (1, 0, 1)t = (t, 0, 1 t),

0 t 1.

Now we can evaluate each line integral,


Z
Z 1
~
F d~x =
(1 t + t2 , t, (1 t)2 ) (1, 1, 0)dt,
C1
0
Z 1
=
1 + t t2 + t dt,
0
Z 1
=
1 + 2t t2 dt,
0

1
= t + t2 t3 /3 0 ,
= 1/3.
71

F~ d~x =

C2

((1 t)2 , 1 t + t2 , t) (0, 1, 1)dt,

0
1

1 + t t2 + t dt,

=
Z0 1

1 + 2t t2 dt,
0

1
= t + t2 t3 /3 0 ,
=

= 1/3.
Z

F~ d~x =

C3

(t, (1 t)2 , (1 t) + t2 ) (1, 0, 1)dt,

t 1 + t t2 dt,

=
Z0 1

1 + 2t t2 dt,
0

1
= t + t2 t3 /3 0 = 1/3.
=

Therefore,

F~ d~x = 1.

Method 2:
We use Stokes theorem,
I
C

ZZ 


~ F~ n

d,

ZZ 
  ~g ~g 
~
~

=
F
d.
x y

F~ d~x =

The curl of this function is (2z, 2x, 2y).


The normal vector is (1, 1, 1) and therefore the plane must be,
x + y + z = 1.
Thus, the surface is ~g (x, y) = (x, y, 1 x y) on 0 x 1 and 0 y 1. This yields that the
normal vector for this surface is (1, 1, 1).
Stokes theorem then yields
I
ZZ 
  ~g ~g 
~ F~

d,
F~ d~x =

x y
C

ZZ
1
=
(2x, 2y, 2z) (1, 1, 1) d,
3 ZZ
2
=
(x + y + z) d,
3 Z Z
2
=
1 d = 1.
3
Note that above we used the equation of the plane to simplify the integrand and then computed
the area of a right triangle with the two sides adjacent to the right angle equal to one.
72

Lecture 30: AMATH 231-F14-001

4.4

November 17, 2014

Maxwells Equations

In this subsection we derive Maxwells equations using the theorems of vector calculus.
4.4.1

Gauss Law

The electrostatic force field at ~x0 due to a charge Q at position ~xQ is,
~ x0 ) = Q ~x0 ~xQ .
E(~
40 k~x0 ~xQ k2
Note that 0 is the permittivity of free space. Building on this formula we can determine the
electrostatic force field at ~x0 due to a cluster of point charges of strength Qi each at position ~xi is
simply a sum of the previous formula:
~ x0 ) =
E(~

n
1 X
~x0 ~xi
Qi
.
40 i=1 k~x0 ~xi k2

By the same token, the electrostatic force field at ~x0 due to a continuous charge (~x) in a volume
V is the following integral
ZZZ
1
~x0 ~xi
~ x0 ) =
E(~
dV.
(~x)
40
k~x0 ~xi k2
V
~ x), is parallel to the direction of the current and has a
The current density vector field, J(~
magnitude of current/unit area. Using the relation J~ = ~u the total current through the surface
is,
ZZ
ZZ
~u n
d.
J~ n
d =
I=
RRR

Note also that the total charge in volume V is Q =


dV .
V
With this established we can state Gauss Theorem.
Theorem 4.21 Gauss Theorem
Physically it states that the flux of the magnetic field through the surface is equal to the total change
contained within the surface. Mathematically, it takes the form,
ZZ
ZZZ
1
1
~
En
d = Q =
dV.
0
0

V
As true as this might be it is hard to determine the force given the charge because this is a global relation.
To get a local, or differential, relation, we rewrite the LHS using the divergence theorem, combine the two
integrals and use our lemma,
ZZZ
ZZZ
1
~ E
~ dV =

dV,
0
V
V

ZZZ 
1
~ E
~ dV = 0,

0
V
~ E
~ = .

0
This is what we call Maxwells First Equation.
73

4.4.2

Faradays Law

~ around a simple closed curve equals the negaPhysically, this states that the circulation of E
tive of the rate of change of the magnetic flux through the surface bounded by .
Mathematically, this can be written using a line integral and a surface integral and then simplified using Stokes theorem:
I
ZZ
d
~
~ n
E d~x =
B
d,
dt

ZZ
ZZ
~
B
~
~
n
d,
En
d =

t
#
ZZ "
~

B
~ E
~+
n
d = 0,

~
~ E
~ = B .

t
This is Maxwells second law.
4.4.3

Gauss Law for Magnetism

Physically, there are no magnetic monopoles because wherever you have a North there must be
a corresponding South. This can be demonstrated when you take a magnet and break it and
youll find that each piece has a corresponding North and South.
This can be restated to say that the net flux of a magnetic field through any surface is zero.
Mathematically, this is easily stated as,
ZZ
~ n
B
d = 0.
V

If we apply Gauss Divergence Theorem we obtain,


ZZZ
~ B
~ dV = 0.

The differential form states,


~ B
~ = 0.

This is Maxwells Third law.


4.4.4

Amp`eres Law

Physically, it states that the circulation of the magnetic field around a simple closed curve
equals the rate of change of the electric flux through a surface plus the total current through
said surface.
Mathematically, this can be written as,
I
ZZ
ZZ
d
~
~
B d~x = 0 0
En
d + 0
J~ n
d.
dt

74

Note that 0 is the permeability of free space.


We apply Stokes theorem and simply using our favourite lemma:
ZZ 

ZZ
~

E
~ B
~ n

d 0 0
n
d 0
J~ n
d = 0,
t

#
Z Z "

~

E
~ B
~ 0 0
0 J~ n
d = 0,



~
~
~ B
~ = 0 0 E + 0 J.

t


ZZ

This is of course Maxwells Fourth law.

75

Lecture 25: AMATH 231-F14-001

November 5, 2014

Fourier Series and Fourier Transforms

Review Linear Algebra


For a finite dimensional vector space, Rn , we can find a complete orthogonal (orthonormal) basis
ei such that ei ej 6= 0 if i 6= j and ei ei = 1. These vectors form a basis, but the basis is not
unique.
Any vector ~x in the vector space can be written as a superposition of the basis in terms of
some coefficients ai ,
n
X
~x =
ai ei .
i=1

We can determine the coefficients using the dot product to project onto each basis vector,
!
n
X
~x ej =
ai ei ej = aj ej ej .
i=1

Therefore, we conclude that

~x ej
.
ej ej

aj =

5.1

Fourier Series

The set of functions f (x) : R R form an infinite dimensional vector space. As with finite
dimensional vector spaces, we can find a complete orthogonal (orthonormal) basis for this space,
which is also not unique.
One set of basis functions are {xm } for m = 0, 1, 2, 3, . If we express f (x) as a superposition
we get a Taylor series,

X
am xm .
f (x) =
m=0

However, these basis functions are not orthogonal and that makes things a bit harder.
Instead, we will consider f (x) defined on a finite interval say [, ] then a set of basis functions can be formed using trigonometric functions,
{cos nx, sin nx} ,

n = 0, 1, 2, .

Note that sin 0 = 0 which does not contribute anything and therefore does not need to be included. Also, cos 0 = 1, is just the constant function.
We will see soon that it is possible to express many functions as,
f (x) =

(an cos nx + bn sin nx) .

n=0

Or we could pull out the constant term and neglect the zero term to get

X
1
f (x) = a0 +
(an cos nx + bn sin nx) .
2
n=1

76

Note that in these two equations a0 is defined differently. The course notes use the second equation and so will we.
Why do we have the factor of 12 in front of a0 ? Believe it or not but it is to make our lives
easier, eventually.
Definition 5.1 A Fourier Series of a function f (x) is a series of the form,

X
1
f (x) = a0 +
(an cos nx + bn sin nx) .
2
n=1

where a0 , an , bn are constants for n = 1, 2, 3, .


One example of Fourier Series is,
sin

7x
.
L

Note that all the coefficients are zero except b7 = 1.


Definition 5.2 The coefficients an and bn in the Fourier Series are called the Fourier coefficients.
In this example, and others like it, we didnt have to do any work to find the coefficients. As
you can imagine, that is not usually the case. Another example is,
cos2

x
1 1
2x
= + cos ,
L
2 2
L

where we rewrite it as a Fourier Series using a double angle formula. All the coefficients are zero
except a0 = 1 and a2 = 12 .
A third example is,
x
sin4 .
L
This is not written in our standard form but we can use a double angle formula twice if we
wanted to. This works for certain cases but we need to find an approach that works for all
functions. We do this by defining an inner product, the analogue of the dot product, but for
functions.
5.1.1

Calculating Fourier Series

Definition 5.3 The inner product of two functions f (x), g(x) on the interval [, ] is,
Z
hf, gi =
f (x)g(x) dx.

Definition 5.4 We say that two functions f (x), g(x) are orthogonal on the interval [, ] if,
Z
hf, gi =
f (x)g(x) dx = 0.

77

Definition 5.5 The norm of a function f (x) on the interval [, ] can be defined in terms of the inner
product,
Z

f 2 (x) dx.

kf k = hf, f i =

To find our coefficients we can use the inner product to project onto each function. This can
be done using the following identities that are easy enough to verify,

Z
0 m 6= 0
hcos mx, cos nxi =
cos mx cos nx dx =
L m=n=0

2L m = n = 0,
Z
hcos mx, sin nxi =
cos mx sin nx dx = 0,


Z
0 m 6= 0
sin mx sin nx dx =
hsin mx, sin nxi =
L m = n = 0.

If we say that the orthogonal basis is denoted with en (x) for m = 0, 1, 2, then our series
representation of f (x) is,

X
an en (x).
f (x) =
n=0

As in linear algebra, to find the coefficients an we must project on to each basis vector. This is
done using the inner product,

X
an en , em i,
hf (x), em i = h

n=0

an hen , em i,

n=0

= am hem , em i,
after using the orthogonality. Therefore, the m-th coefficient is,
am =

hf, em i
.
hem , em i

This is the general form but this must be done for each basis vector.
For the Fourier basis mentioned above, using the orthogonality we find,
hf, 21 i
a0 = 1 1 ,
h2, 2i
R
f (x) 21 dx

= R 1 1 ,
dx
2
Z 2
1
=
f (x) dx.

78

hf, cos(mx)i
,
hcos2 (mx)i
R
f (x) cos(mx) dx
R
=
,
cos2 (mx) dx

Z
1
=
f (x) cos(mx) dx.

am =

hf, sin(mx)i
,
hsin2 (mx)i
R
f (x) sin(mx) dx
R
=
,
sin2 (mx) dx

Z
1
f (x) sin(mx) dx.
=

bm =

Lecture 26: AMATH 231-F14-001

November 7, 2014

example 5.6 Find the Fourier coefficients of the function,



1
0 < x < ,
f (x) =
1 < x < 0.
Solution:
Since f (x) is odd we see that an = 0 for all n. Thats the easy party (or at least can be easy if we
make this observation.
The next set of coefficients must be computed directly, with n = 1, 2, ,
Z
1
bn =
f (x) sin nx dx,

Z
2
=
sin nx dx,
0
2
=
[cos nx]0 ,
n
2
=
[1 cos n] ,
n

0 n is even
=
4
n is odd
n
Therefore, our Fourier series for this step function is,
f (x) =

5.1.2

4 X sin(2n 1)x
.
n=1 (2n 1)

Pointwise Convergence of a Fourier Series

Here we consider how the series converges to the function in question. First we need some
definitions.
79

Definition 5.7 A function f : [a, b] R is piecewise continuous (or C 1 ) means that there is a partition,
a = x0 < x1 < x2 < < xn = b,
such that f is continuous (or C 1 ) on each closed interval xi1 x xi for all i.
Definition 5.8 Given a function f : [, ] R, a period 2 extension of f , denoted by fp , is a function
of period 2 such that fn (x) = f (x) on < x < .
Theorem 5.9 Pointwise Convergence
If fp has period 2 and is piecewise-C 1 the the Fourier Series of fp converges point wise (i.e. for each x)
and
i) If fp is continuous at x, then the sum is fp (x)
ii) If fp is not continuous at x, then the sum is 12 [fp (x+ ) + fp (x )]
Note that we use x+ to denote the limit from the right and x to be the limit from the left.
Lecture 31: AMATH 231-F14-001

November 19, 2014

example 5.10 A Fourier Sine series (odd extension) of f (x) = 21 ( x) on 0 < x < is

X
sin nx
n=1

Sketch the graph of the period 2 function to which the series converges.
example 5.11 In the previous example if we set x = /2 then we obtain

X sin n

1
2
( ) =
,
2
2
n
n=1

X (1)n1
1 1 1
=
= 1 + + .
4
2n 1
3 5 7
n=1
Sketch the graph of the period 2 function to which the series converges.
Gibbs Phenomenon:
Our convergence theorem tells us that when we add all of the terms in the Fourier basis we will
recover the function we expanded. Practical constraints dictate that we cannot find all of the
terms and sum them up. That is why we can only sum up a finite number of basis functions.
When summing up the Fourier series for a continuous function the series converges nicely.
However, when summing the series for a continuously function oscillations tend to arise near
the point of discontinuity. The amplitude and wavelength of these wiggles tend to get smaller
and smaller with the number of terms but they will always persist. This is Gibbs Phenomenon.

80

Lecture 32: AMATH 231-F14-001


5.1.3

November 21, 2014

Symmetry Properties

We have already noted that knowing that a function is odd or even can simplify our calculations
since we know that cos and sin are even and odd, respectively.
Properties:
i) f is even means that f (x) = f (x) for all x.
ii) f is odd means that f (x) = f (x) for all x.
iii) The product of two even or two odd functions is even.
iv) The product of an even and an odd function is odd.
v) If f is even then

Z
f (x) dx = 2

vi) If f is odd then

f (x) dx.
0

f (x) dx = 0.
a

Special cases of Fourier series:


i) Cosine series:
Suppose that f is an even function.
Z
Z
1
2
an =
f (x) cos(nx) dx, =
f (x) cos(nx) dx,

0
Z
1
bn =
f (x) sin(nx) dx = 0.

Therefore, our cosine series take the form,

X
1
f (x) = a0 +
an cos nx
2
n=1

with an given by the formula above.


This implies that any function f (x) that is define on 0 < x < has a Fourier cosine series, it
does not need to be defined on < x < 0 But, since the series is defined on that interval
it can be thought of as the Fourier series of the even extension of f (x), which we define as

f (x) 0 < x < ,
feven =
f (x) < x < 0.

81

ii) Sine series:


Suppose that f is an odd function.
Z
1
an =
f (x) cos(nx) dx = 0,

Z
Z
1
2
bn =
f (x) sin(nx) dx =
f (x) sin(nx) dx

0
Therefore, our sine series take the form,
f (x) =

bn sin nx

n=1

with an given by the formula above.


This implies that any function f (x) that is define on 0 < x < has a Fourier sine series, it
does not need to be defined on < x < 0 It is the Fourier series of the odd extension of
f (x), which we define as,

f (x)
0 < x < ,
fodd =
f (x) < x < 0.
Symmetry about x = 2 . There are other different kinds of symmetry. One is to consider
whether the function is even or odd about the line x = 2 . If we consider f (x) defined on the
interval 0 < x < .
i) f is even about x =

means that
f ( x) = f (x),

ii) f is odd about x =

for all

x.

means that
f ( x) = f (x),

for all

x.

exercise 5.12 Show the following:


i) cos 2nx and sin(2n 1)x are even with respect to x = 2 .
ii) cos(2n 1)x and sin 2nx are odd with respect to x = 2 .
We can then simplify the cosine and sine series that are symmetric about x = 2 .
Symmetry about x =

and the cosine series.

i) If f ( x) = f (x) on 0 < x < then our cosine series further simplifies


Z
Z
2
4 /2
a2n =
f (x) cos(2nx) dx =
f (x) cos(2nx) dx,
0
0
Z
2
a2n1 =
f (x) cos((2n 1)x) dx = 0.
0
The cosine series reduces to,

X
1
f (x) = a0 +
a2n cos 2nx.
2
n=1

82

ii) If f ( x) = f (x) on 0 < x < then our cosine series further simplifies
Z
2
a2n =
f (x) cos(2nx) dx = 0,
0
Z
Z
2
4 /2
a2n1 =
f (x) cos((2n 1)x) dx =
f (x) cos((2n 1)x) dx
0
0
The cosine series reduces to,
f (x) =

a2n1 cos(2n 1)x.

n=1

Symmetry about x =

and the sine series.

i) If f ( x) = f (x) on 0 < x < then our sine series further simplifies


Z
2
b2n =
f (x) sin(2nx) dx = 0,
0
Z
Z
2
4 /2
b2n1 =
f (x) sin((2n 1)x) dx =
f (x) sin((2n 1)x) dx = 0.
0
0
The sine series reduces to,
f (x) =

b2n1 sin(2n 1)x.

n=1

ii) If f ( x) = f (x) on 0 < x < then our sine series further simplifies
b2n
b2n1

Z
Z
2
4 /2
=
f (x) sin(2nx) dx =
f (x) sin(2nx) dx,
0
0
Z
2
=
f (x) sin((2n 1)x) dx = 0.
0

The sine series reduces to,


f (x) =

a2n1 sin 2nx.

n=1

example 5.13 (course notes example 5.5) Find the Fourier sine series of
f (x) =

,
4

0 < x < .

Solution: First we define the odd extension,



0 < x < ,
4
fodd =
4 < x < 0.

83

The Fourier sine coefficients are,


bn =
=
=
=

Z
1
fodd (x) sin(nx) dx,

Z
2
fodd (x) sin(nx) dx,
0
Z
2

sin(nx) dx,
0 4
Z
1
sin(nx) dx.
2 0

Now since the constant function is even about x = /2 we recognize that b2n = 0 since sin(2nd)
is antisymmetric. The other coefficients are easily determined,
Z
1
sin((2n 1)x) dx,
b2n1 =
2 0
Z /2
sin((2n 1)x) dx,
=
0

1
[cos((2n 1)x)]/2
0 ,
2n 1
h
1
i
=
1 cos((2n 1) ) ,
2n 1
2
1
=
.
2n 1
=

Therefore the Fourier sine series is

X
sin(2n 1)x
n=1

2n 1

example 5.14 Find the Fourier cosine series of


f (x) =
Solution:

5.1.4

,
4

0 < x < .

.
4

Complex form of the Fourier series

Eulers formula states that,


ei = cos + i sin .
This shows that cos and sin are really just parts of the complex exponential. Since our Fourier
series is a sum of cosines and sines then we should be able to rewrite the sum in terms of an
exponential. One advantage is that it makes things more compact. One disadvantage is that we
have to deal with complex variables. But its good for us and a skill well worth learning so we
proceed.

84

Definition 5.15 The complex Fourier series of a -periodic function f (x) can be written as,
f (t) =

cn ein0 t ,

n=

where at the fundamental frequency is 0 = 2


.

The complex Fourier coefficients cn for n = 0, 1, 2, are,


1
cn =

f (t)ein0 t dt,

and satisfy the relation cn = cn . Note that the overbear denotes complex conjugation This relation is
what is required for the result of the infinite sum to be real.

85

Lecture 33: AMATH 231-F14-001

November 24, 2014

The complex basis functions, which we can denote as en (t) = ein0 t satisfy an orthogonality
condition. To derive that lets evaluate the following integral,
Z

/2

/2

ein0 t eim0 t dt,

en (t)em (t) dt =
/2

/2
/2

ei(nm)0 t dt,

=
/2

 i(nm)0 t  /2
1
, we assumed that n 6= m
e
/2
i(n m)

 i(nm)0
1
i(nm)0 2
2 e
=
,
e
i(n m)

 i(nm)
1
=
e
ei(nm) ,
i(n m)
2
=
[sin((n m))] .
(n m)
=

Therefore, if m 6= n we get the sine of an integer multiple of , which is necessarily zero.


In the case where m = n the integral is much easier since the integrand is one.
Combining these two results then yields the following orthogonality relation,
Z

/2


en (t)em (t) dt =

/2

if
if

m 6= n,
m = n.

Derivation of Formula for Fourier coefficients:


Armed with our orthogonality relation it is pretty straightforward to obtain the general formula
for the Fourier coefficients stated above. But before we do so it is useful to define the inner
product for complex functions as,
/2

Z
hf, gi =

f (t)g(t) dt.
/2

We begin with the Fourier series in complex form and project onto a particular basis function
(equivalent to multiplying by a basis function and integrating but more insightful to say project),
im0 t

hf (t), e

i=h
=

cn ein0 t , eim0 t i,

n=

cn hein0 t , eim0 t i,

n=

= cm .
The final equation above can be inverted to yield the desired result,
Z
1
1
im0 t
i=
f (t)eim0 t dt.
cm = hf (t), e


86

Relation between real and complex forms


To make a connection between our two formulations of Fourier series let us begin with the complex form and use Eulers formula to rearrange

f (t) =

cn ein0 t ,

n=
1
X

cn e

in0 t

+ c0 +

n=

cn ein0 t .

n=1

In the first series, if we substitute n = n and use the fact that cn = cn we get
f (t) =

cn e

in0 t

+ c0 +

n=1

= c0 +

cn ein0 t ,

n=1


 in0 t
cn e
+ cn ein0 t ,

n=1

= c0 +

X



cn ein0 t + cn ein0 t ,

n=1

= c0 +
= c0 +

X
n=1

[cn (cos(n0 t) + i sin(n0 t)) + cn (cos(n0 t) i sin(n0 t))] ,


[(cn + cn ) cos(n0 t) + i (cn cn ) sin(n0 t)] .

n=1

This is precisely the same form as our real Fourier series. If we match up the coefficients we get,
1
c 0 = a0 ,
2

an = cn + cn ,

bn = i(cn cn ).

From this you can invert to find that


1
cn = (an ibn ).
2
Lecture 34: AMATH 231-F14-001

November 26, 2014

The Fourier spectrum and Parsevals Theorem


A Fourier series decomposes a function into a sum of trigonometric functions (or complex
exponentials). Each term in the basis has a constant frequency. Therefore, a physical interpretation of a Fourier series is to decompose a function (or signal) f into its basic frequencies or
harmonics,
0 , 20 , 30 , .
We define 0 = 2/ as the fundamental (smallest) frequency. All other frequencies that are
permitted are an integer multiple of the fundamental frequency.
The amount of each frequency is contained in the magnitude of the coefficient to the complex
Fourier series cn . Since this is a complex number it has both a magnitude and a phase. The
amount of each harmonic that
87

Definition 5.16 The Fourier coefficients cn for a function f (x) are defined as the Fourier spectrum of f .
Definition 5.17 The magnitude of the Fourier coefficients squared, |cn |2 is defined as the Power spectrum
of f . It shows how much power we have in each frequency.
When considering a function f (t) we can think of it in the physical (time or space) domain or
we can think of it in the spectral (or frequency) domain, as is determined by cn .
One way to measure the strength of a signal in the physical domain is to compute the average
of the square over the interval in question,
Z

/2

f 2 (t)dt.

/2

In the Fourier domain we can consider a sum of the powers,

|cn |2 .

n=

The next theorem will show that these two are not independent and that they are in fact equal.
Theorem 5.18 (Parsevals Theorem for a -periodic function)
If a -period function f has a complex Fourier series,

f (t) =

cn ein0 t ,

n=

then
1

/2
2

f (t)dt =
/2

|cn |2 .

n=

Proof:
Begin with the Fourier series of f (t) and multiply by f and integrate over the domain
!
Z /2
Z /2

X
2
in0 t
cn f (t)e
dt,
f (t)dt =
/2

/2

n=

Z
cn

n=

=
=

/2

n=

!
f (t)ein0 t dt,

/2

cn cn
|cn |2 .

n=

This shows that the strength of the function in physical space is equal to its strength in Fourier
space.

88

example 5.19 Find the complex Fourier series and apply Parsevals theorem to the function,
 4x
0 < x < ,

f (x) =
4(x+)
< x < 0.

This curve consists of a line that is copied every and is piecewise C 1 .


The complex Fourier coefficients are,
Z
1
cn =
f (x)einx dx,
2
Z

Z 0
4x inx
4(x + ) inx
1
=
e
dx +
e
dx ,
2

Z

Z 0
4
inx
inx
= 2
xe
dx +
(x + )e
dx ,
2
0

Z

Z 0
2
inx
inx
= 2
xe
dx +
e
dx .

Each of these integrals can be evaluated. First the second,



0
Z 0

1 inx
 in

inx
e
[(1)n 1]

e
dx =
=
e 1 =
in
in
in

Then the second, which we integrate by parts,


Z
Z
Z
hx
i
i
inx
xe
dx = i
x sin nx dx = i
cos(nx)
+
cos(nx) dx,
n
n

2
i
= i cos(n) + 2 [sin(nx)] ,
n
n
2i
n
=
(1) .
n
Therefore, the complex Fourier coefficients are,

Z
Z 0
2
inx
inx
cn = 2
xe
dx +
e
dx ,



2 2i

n
n
= 2
(1) +
[(1) 1] ,

n
in
2
=
(2(1)n + [(1)n 1]) ,
in
2
=
((1)n 1) ,
in
4i
=
.
n
The term inside the brackets is equal to 0 for n odd and 2 for n even. Therefore, we deduce for
k = 1, 2, 3, ,
c2k =
89

2i
.
k

The constant case must be computed separately,


Z
1
c0 =
f (x) dx,
2
Z
4
= 2
x dx,
0


4 1 2
= 2.
= 2
x
2
0
Therefore, the complex Fourier series can be written as,
f (x) = 2 +

c2k ei2kx = 2 +

k=1

2 X i i2kx
e
k=1 k

Next we verify Parsevals theorem. The LHS is,


1
2

Z  2
1 4x
f (x) dx =
dx,
0

Z
16 2
x dx,
= 3
0
16
16
= 3 3 = .
3
3

The RHS is a series,

|cn |2 = c20 + 2

n=

|c2k |2 ,

k=1

2

X
2
=4+2
,
k
k=1

8 X 1
=4+ 2
,
k=1 k 2

8 2
,
2 6
4
16
=4+ = .
3
3
=4+

using the following identity,

X
1
2
= .
k2
6
k=1

Lecture 35: AMATH 231-F14-001

November 28, 2014

Applications:

90

1) Energetics Consider a two dimensional fluid that has a constant density of and velocity
(u(x, y) and v(x, y)) in the x and y directions, respectively. Given that the kinetic energy of
a particle is 12 m(u2 + v 2 ) you might not be too surprised to find that the kinetic energy of
the fluid over some horizontal doubly domain D is
ZZ
1
KE =
(u2 + v 2 ) dA.
2 D
Suppose that the velocities are x - and y -periodic in each direction and that we know the
Fourier coefficients for u and v are unm and vnm . That is to say that the Fourier decomposition of the two fields are
Z x Z x
1
unm ei(n1 x+m2 y) dxdy,
u(x, y) =
x y x x
Z x Z x
1
vnm ei(n1 x+m2 y) dxdy.
v(x, y) =
x y x x
Note that we have two indices because we can decomposition in both the x and y direction.
We have defined 1 and 2 to be the fundamental frequencies in the x and y directions,
respectively.
We can apply Parsevels theorem to u in the x-direction then we have for each component
of velocity
Z

1 2
x y X X
2
u dx =
|unm |2 ,
2 x /2
2 n= m=
Z

x y X X
1 2
2
v dx =
|vnm |2 .
2 x /2
2 n= m=
Therefore the kinetic energy can be written as a sum of the Fourier coefficients,


x y X X
KE =
|unm |2 + |vnm |2 .
2 n= m=

This tells us what the total energy is in terms of the Fourier coefficients. But if we consider
an individual term in the sum we have what the power is in a particular frequency, that is
to say a wave with a given wavelength in the x and y direction,

x y
|unm |2 + |vnm |2 .
KEnm =
2
This is referred to as the spectral energy density.
This is a very useful concept that allows us to understand what length scales are energetic. For example, the oceans are forced by large scale forcings such as the winds from
the atmosphere and the tides. This is essentially forcing the system at length scales that
are input on planetary scales, thousands of kilometres. But the dissipation happens due
to molecular velocity down at the micro scales, micrometres. Energy must transfer from
106 m to 106 m. There are a plethora of nonlinear processes that can do this and they are
far from completely understood. This is an integral part of understanding the dynamics of
the oceans, the atmosphere, their interactions and therefore our evolving climate system.
91

2) Biology In the oceans we have a vast range of species that can be as large as a whale or as
small as a phytoplankton. Each of these has a given size or length scale. Biological oceanographers are trying to understand how much mass of biology, what they call biomass, we
have in each size class. This could be understand by knowing what is the spectral density
of biomass in the ocean, say nm . This is something that must be measured but theoreticians build models that can try and predict how this is changing and what the real ocean
is tending to.
One hypothesis is that there is the same biomass in each size class. That is to say if we sum
up all the whales we get as many kilograms if we sum up all of the smallest phytoplankton.
This sounds remarkable and it is but you should not keep in mind that there are far fewer
whales and maybe this is true. This is an issue of ongoing research that we have yet to
resolve.

5.2

Convergence of series of functions

Whenever one sums an infinite number of functions one should be careful that the sum does
in fact converge. Otherwise there is no real point (as far as I can tell). As you may recall from
MATH 138, to determine the convergence of a series,

an (x),

n=1

then you need to consider the N -th partial sum,


SN (x) =

N
X

an (x).

n=1

To determine if the series converges is equivalent to asking if the limit of the sequence SN (x)
converges.
5.2.1

A deficiency of pointwise convergence

What is different about what we are doing here is that our series depends on x. Therefore, when
we consider convergence it will depend on the value of x. There are at least three different types
of convergence one can consider (see course notes for details):
Definition 5.20 A sequence of functions {fn } converges pointwise to f on [a, b] means that,
lim |fn (x) f (x)| = 0,

for all x in[a, b].


Any kind of convergence is certainly better than divergence but pointwise convergence is the
weakest kind of convergence one can have in our context.

92

example 5.21 Consider the sequence of functions {fn } defined by,


fn (x) =

2nx
,
1 + n 2 x2

0 x 1.

Shows that,
for al

lim fn (x) = 0,

x [0, 1],

but that
max |fn (x)| = 1,

for all

0x1

n N.

Solution:
If we consider x (0, 1] then we can verify the limit,


1
2x
2
lim fn (x) = lim
=
0

= 0.
1
n
n n
x
x2 + n2
Next, we consider x = 0 and find
lim fn (0) = lim 0 = 0.

By computing the derivative one can show that an extrema has to happen at x = n1 and that
maximum is 1. Therefore, the maximum of the curve for any finite n is one.
The resolution is that for a given point x the function will tend to zero but if x is very small
one might need a very large n for that to happen.
5.2.2

The maximum norm and mean square norm

Other definitions of convergence rely on using different measures, what we call norms and ensure convergence over an interval not just pointwise.
Definition 5.22 The maximum norm kf k on the space C[a, b] is defined by
kf k = max |f (x)|.
axb

for all x [a, b].


Definition 5.23 The mean square norm kf k2 on the space C[a, b] is defined by
b

Z

1/2
f (x) dx
2

kf k2 =
a

The two norms are related and the interested reader can see Proposition 5.1 of the course notes.

93

5.2.3

Uniform and Mean Square Convergence

Definition 5.24 A sequence {fn } in C[a, b] (the set of continuous functions on the interval [a, b]), converges uniformly to f C[a, b] means that,
lim kfn f k = 0,

or


lim


max |fn f | = 0,

axb

Definition 5.25 A sequence {fn } in C[a, b] converges in the mean to f C[a, b] means that,
lim kfn f k2 = 0,

or
Z

5.3
5.3.1

1/2

[fn (x) f (x)]

lim

= 0,

A Second Look at Fourier Series


Uniform and mean square convergence

Theorem 5.26 (Convergence of Fourier series)


i) If fp is piecewise continuous, then the Fourier series of f converges in the mean to f on any finite
interval.
ii) If fp is piecewise C 1 , then the Fourier series of f converges pointwise to fp (x) for all x R.
iii) If fp is piecewise C 1 and continuous, then the Fourier series of f converges uniformly to fp (x) for
any finite interval.
5.3.2

Integration of Fourier series

Theorem 5.27 The Fourier series of a 2 piecewise continuous function can be integrated term-by-term
over any finite interval. In particular if,

X
1
(an cos nt + bn sin nt),
fp (t) = a0 +
2
n=1

then
Z

Z
fp (t) dt =

in the mean,

Z x
X
1
a0 dt +
(an cos nt + bn sin nt) dt,
2
a
n=1

and the convergence is uniform.


This is the justification that we need to derive the formulas for the Fourier coefficients.

94

example 5.28 Given that

X sin nx
1
( x) =
,
2
n
n=1
show that,

0 < x < ,

1
2 X cos nx
( x)2 =
+
,
4
12 n=1 n2

0 < x < ,

uniformly on 0 < x < .

5.4

The Fourier transform and Fourier integral

Fourier series are very useful but are necessary restricted to functions f (t) that are periodic on
a closed interval. Next, we introduce a very similar idea but holds for functions defined on the
real line.
5.4.1

The definition

Consider a -periodic function f that are square-integrable on R


Z
f 2 (t) dt < ,

and satisfy,
lim f (t) = 0

For these functions we define the following.


Definition 5.29 The Fourier transform of f is defined by
Z
F(f ) = F () =
f (t)eit dt,

and the inverse transform allows us to obtain f (t) given F (),


Z
1
f (t) =
F ()eit dt.
2
Note that these look almost identical except for three differences: 1) the factor of 1/(2), 2) changing the
variable of integration from t to and 3) the sign in the exponent.
Derivation of Fourier Transforms from Fourier Series
Consider a periodic function f (t) that is say a bump localized at t = 0 and then tends to
zero at /2. In the limit as increases we have that the domain tends to the real line and the
function is a single bump at the origin,
lim f (t) = f (t).

The complex Fourier series of our periodic function is


f (t) =

X
n=

95

cn ein0 t ,

with the formula for the Fourier coefficient,


Z
1 /2
cn =
f (t)ein0 t dt,
/2
and the fundamental frequency is 0 = 2/tau.
Since f (t) agrees exactly with f (t) in the interval [ /2, /2] and is zero elsewhere, we could
rewrite the above formula as
Z
1
cn =
f (t)ein0 t dt.

Note that all we did is extend the domain of integration and replaced the function in the integrand. This formula is almost identical with our definition for the Fourier transform of f (t),
which we denote by F () if we relate,
1
F (n0 ).

What happens when ? In this case the fundamental frequency 0 tends to zero. Therefore we introduce a new variable,
2
=
.

The Fourier series of f (t) can be written as


cn =

f (t) =
=

cn ein0 t ,

n=

1
F (n0 )ein0 t ,

n=

1 X
=
F (n0 )ein0 t ,
2 n=

Finally, we take the limit as and hence 0 to obtain,

X
1
f (t) =
lim
F (n0 )ein0 t .
2 0 n=

This is a Riemann sum for the improper integral and we get


Z
1
f (t) =
F ()eit d.
2
It is very important that both the Fourier series and transforms come in pairs. The Fourier
series is
Z

X
1 /2
in0 t
cn =
f (t)e
dt,
f (t) =
cn ein0 t ,
/2
n=
and the Fourier transform is
Z

F () =

f (t)e

it

1
f (t) =
2

dt,

96

F ()eit ,

5.4.2

Calculating Fourier transforms using the definition

A couple of functions pop up a lot in the context of Fourier transforms and are defined below.
Definition 5.30 The rectangular window function is defined as,

1 if |t| < 12 ,
W (t) =
0 if |t| > 12 .
Definition 5.31 The sinc function is defined as

sinc(x) =

sin x
x

if
if

x 6= 0,
x = 0.

example 5.32 What is the Fourier transform of W (t)?


By definition we get,
Z

W (t)eit dt,

F () =

1
2

eit dt,

=
12

1
1 it 2
e
=
,
i
12

2  i/2
=
e
ei/2 ,
2i
2
= sin(/2),

= sinc(/2).


Therefore, F (W (t)) = sinc( 21 ).


Properties:
1) If f (t) is even then,
Z

f (t)eit dt,

F(f (t)) = F () =
Z

f (t) [cos(t) i sin(t)] dt,

=2

f (t) cos(t) dt.


0

2) If f (t) is odd then,


Z

f (t)eit dt,

F(f (t)) = F () =
Z

f (t) [cos(t) i sin(t)] dt,


Z
= 2i
f (t) sin(t) dt.
=

97

3) Linearity: (where c is a constant)


F(f + g) = F(f ) + F(g),
F(cf ) = cF(f ).
4) Scaling property: If F(f (t)) = F () then,
1
F(f (at)) = F ( ).
a a
Proof:
Z

F(f (at)) =

f (at)eit dt,

1
=
f ( )ei a d,
a
1  
= F
,
a
a
where we made the substitution = at and so d = adt.
5) Shift formula: If F(f (t)) = F () then
F(ei0 t f (t)) = F ( 0 )
Proof:
Suppose F(f (t)) = F () then use the definition,
Z
f (t)ei0 t eit dt,
F(f (t)) = F () =

Z
=
f (t)ei(0 )t dt,

= F( 0 ).
6) Derivative: The Fourier transform of the derivative is,
 
df
F
= iF ().
dt
Proof:
Suppose that F(f (t)) = F (), then consider the definition and use integration by parts,
  Z
df
df it
F
=
e
dt,
dt
dt
Z
 it 
= fe
+ i
f (t)eit dt,

= iF ().
98

This is very useful in solving differential equations.


example 5.33 Using the fact that F (W (t)) = sinc( 12 ) we can then use the scaling formula to deduce

1
F (W (at)) = sinc( ).
a
2a
example 5.34 Calculate F(f (t)) if f (t) = W (at) cos(0 t).
Use the definition,
F(f (t)) = F(W (at)) cos(0 t)),


1
= F W (at)) ei0 t + ei0 t ,
2
 1

1
= F W (at))ei0 t + F W (at))ei0 t ,
2 

2


1
0
+ 0
=
sinc
+ sinc
2a
2a
2a
example 5.35 When we studied Maxwells equations without a current and density charge we found
that each the magnetic and electric fields were governed by a particular PDE. That PDE is known as the
wave equation and can be written as,
2
2u
2 u
=c
.
t2
x2
In this question we assume that the x domain is infinite.
If we take the Fourier transform of this equation with respect to x (why not time?), and define U () =
F(u), we obtain,
 2 
 2 
u
u
F
= F c2 2 ,
2
t
x
2

F [u] = c2 (i)2 F [u] ,


t2
2U
+ c2 2 U = 0.
t2
Note that in the above we used the derivative property twice to get the above result.
It is very important that the variable U (, t) solves an ordinary differential equation, in contrast with
u(x, t) that solves a partial differential equation. The former is much easier to solve and we saw previously
that this is nothing more than the simple harmonic oscillator. The solution to this equation can be written
as,
U = A()eit + B()eiit .
If we have two initial conditions then we can determine A and B. This gives us an exact solution in
Fourier space. To obtain u(x, t) you then need to invert back to physical space. This is done with the help
of tables of transforms. The interested reader is direct to AMATH 353 for more details. (I have lecture
notes for the course if youre curious to see more details.)

99

5.4.3

Parsevals Theorem for a non-periodic function

Theorem 5.36 If F(f (t)) = F () then,


Z
Z
1
2
f (t) dt =
|F ()|2 d.
2

Derivation:
Begin with the definition of the Fourier transform
Z
F ()eit d,
f (t) =

multiply by f (t) and integrate over t,



Z
Z Z
1
2
it
f (t) dt =
f (t)F ()e d dt exchange order of integration,
2


Z
Z
1
it
f (t)e dt d,
=
F ()
2

Z
1
=
F ()F () d,
2
Z
1
=
|F ()|2 d.
2

100

Das könnte Ihnen auch gefallen