Calculus IIIvectorcalculus 2011

Chapter 1
vector calculus
Vector calculus is the study of vector fields and related scalar functions. For the most part we
focus our attention on two or three dimensions in this study. However, certain theorems are easily
extended to Rn . We explore these concepts in both Cartesian and the standard curvelinear coor-
diante systems. I also discuss the dictionary between the notations popular in math and physics 1
The importance of vector calculus is nicely exhibited by the concept of a force field in mechanics.
For example, the gravitational field is a field of vectors which fills space and somehow communi-
cates the force of gravity between massive bodies. Or, in electrostatics, the electric field fills space
and somehow communicates the influence between charges; like charges repel and unlike charges
attract all through the mechanism propagated by the electric field. In magnetostatics constant
magnetic fields fill space and communicate the magnetic force between various steady currents.
All the examples above are in an important sense static. The source charges2 are fixed in space
and they cause a motion for some test particle immersed in the field. This is an idealization. In
truth, the influence of the test particle on the source particle cannot be neglected, but those sort
of interactions are far too complicated to discuss in elementary courses.
Often in applications the vector fields also have some time-dependence. The differential and inte-
gral calculus of time-dependent vector fields is not much different than that of static fields. The
main difference is that what was once a constant is now some function of time. A time-dependent
vector field is an assignment of a vectors at each point at each time. For example, the electric and
magnetic fields that together make light. Or, the velocity field of a moving liquid or gas. Many
other examples abound.
The calculus for vector fields involves new concepts of differentiation and new concepts of integra-
tion.
1
this version of my notes uses the inferior math conventions as to be consitent with earlier math courses etc...
2
the concept of a charge really allows for electric, magnetic or gravitational although isolated charges exist only
for two of the aformentioned
1
2 CHAPTER 1. VECTOR CALCULUS
For differentiation, we study gradients, curls and divergence. The gradient takes a scalar field and
generates a vector field (actually, this is not news for us). The curl takes a vector field and generates
a new vector field which says how the given vector field curls about a point. The divergence takes a
given vector field and creates a scalar function which quantifies how the given vector field diverges
from a point. Many novel product rules exist for these operations and the algebra which links these
operations is rich and interesting as we already saw at the beginning of this course as we studied
vectors at a point.
On the topic of integration there are two types of integration we naturally consider for a vector
field. We can integrate along an oriented curve, this type of integration is called a line integral
even when the curve is not a line. Also, for a given surface, we can calculate a surface integral
of a vector field. The line integral measures how the vector field lines up along the curve of inte-
gration, the measures something called the circulation. The surface integrals value depends on
how the vector field pokes through the surface of integration, it measures something called flux.
Critical to both linea and surface integrals are parametric equations for curves and surfaces. We’ll
see that we need parametrics to calculate anything yet the answers are completely indpendent of
the parameters utilized. In other words, the surface and line integrals are coordinate free objects.
This is in considerable contrast to the types of integrals we studied in the previous part of this
course.
Between the study of differentiation and integration of vector fields we find the unifying theorems of
Greene, Gauss and Stokes. We study Green’s theorem to begin since it is simple and merely a two-
dimensional result. However, it is evidence of something deeper. Both Gauss and Stokes reduce to
Greene’s in certain context. There is also a fundmental theorem of line integrals which helps validate
my claim that intuitively the ∇f is basically the derivative for a function of several variables. In
addition, we will learn about how these integral theorems relate to the path-independence of vector
fields. We saw earlier that certain vector fields could not be gradients because they violate Clairaut’s
Theorem on their components. On the other hand, just because a vector field has components
consistent with Clairaut’s theorem we also saw that it was not necessarily the case they were the
gradient of some scalar function on their whole domain. We’ll find sufficient conditions to make
the Clairaut test work. We will find how to say with certainty a given vector field is the gradient
of some scalar function. More than just that, we’ll find a method to calculate the scalar function.
1.1. VECTOR FIELDS 3
1.1 vector fields

Definition 1.1.1.
A vector field on S ⊆ RnPis an assignment of a n-dimensional vector to each point in S. If
F~ = hF1 , F2 , . . . , Fn i = nj=1 Fj x
bj then the multivariate functions Fj : S → R are called
~
the component functions of F . In the cases of n = 2 and n = 3 we sometimes use the
popular notations
F~ = hP, Qi & F~ = hP, Q, Ri.
We have already encountered examples of vector fields earlier in this course.
Example 1.1.2. Let F~ = x b or G ~ = yb. These are constant vector fields on R 2 . At each point we
attach the same fixed vector; F~ (x, y) = h1, 0i or G(x,
~ y) = h0, 1i. In contrast, H~ = rb and I~ = θb are
2
non-constant and technically are only defined on the punctured plane R − {(0, 0)}. In particular,

~ x y ~ −y x
H(x, y) = p , p & I(x, y) = p , p .
x2 + y 2 x2 + y 2 x2 + y 2 x2 + y 2
Of course, given any differentiable f : S ⊆ R 2 → R we can create the vector field ∇f = h∂x f, ∂y f i
which is normal to the level curves of f .
Example 1.1.3. Let F~ = x ~ = yb or H

b or G ~ = zb. These are constant vector fields on R 3 .
At each point we attach the same fixed vector; F~ (x, y, z) = h1, 0, 0i or G(x,
~ y, z) = h0, 1, 0i and
~ ~ ~
H(x, y, z) = h0, 1, 0i. In contrast, I = rb and J = θ are non-constant and technically are only
b
defined on R 3 − {(0, 0, z) | z ∈ R}. In particular,

~ y, z) = p x y ~ y, z) = p −y x
I(x, , p & J(x, , p .
x2 + y 2 x2 + y 2 x2 + y 2 x2 + y 2
Similarly, the spherical coodinate frame ρb, φ,

b θb are vector fields on the domain of their definition. Of
3
course, given any differentiable f : S ⊆ R → R we can create the vector field ∇f = h∂x f, ∂y f, ∂z f i
which is normal to the level surfaces of f .
The flow-lines or streamlines are paths for which the velocity field matches a given vector field.
Sometimes these paths which line up with the vector field are also called integral curves.
Definition 1.1.4.
In particular, given a vector field F~ = hP, Q, Ri we say ~r(t) = hx(t), y(t), z(t)i is an integral
curve of F~ iff

~ dx dy dz dx dx dz
F (~r(t)) = , , a.k.a = P, = Q, = R.
dt dt dt dt dt dt
We need to integrate each component of the vector field to find this curve. Of course, given that
P, Q, R are typically functions of x, y, z the ”integration” requires thought. Even in differential
equations(334) the general problem of finding integral curves for vector fields is beyond our standard
techniques for all but a handful of well-behaved vector fields. That said, the streamlines for the
examples below are geometrically obvious so we can reasonably omit the integration.
Example 1.1.5. Suppose F~ = x b then obviously x(t) = xo + t, y = yo , z = zo is the streamline

of xb through (xo , yo , zo ). Likewise, I think you can calculate the steamlines for yb and zb without
much trouble. In fact, any constant vector field F~ = ~vo simply has streamlines which are lines with
direction vector ~vo .
Example 1.1.6. The magnetic field around a long steady current in the postive z-direction is
~ θ, z) = µo I θ.
conveniently written as B(r, b The streamlines are circles which are centered on the
2πr
axis and point in the θb direction.
Example 1.1.7. If a charge Q is distributed uniformly through a sphere of radius R then the
electric field can be show to me a function of the distance from the center of the sphere alone.
Placing that center at the origin gives
(
kρ
~ 3 0≤ρ≤R
E(ρ) = ρb Rk
ρ2
ρ≥R
The streamlines are simply lines which flow radially out from the origin in all directions.
Challenge: in electrostatics the density of streamlines (often called fieldlines in physics ) is used to
measure the magnitude of the electric field. Why is that reasonable?
Example 1.1.8. The other side of the thinking here is that given a differential equation we could
use the plot of the vector field to indicate the flow of solutions. We can solve numerically by playing
a game of directed connect the dots which is the multivariate analog of Euler’s method for solving
dy/dx = f (x, y).
dx dy dz
= 2x(y 2 + z 2 ), = 2x2 y, = 2x2 z
dt dt dt
We’d look to match the curve up with the vector-field plot of F~ = h2x(y 2 + z 2 ), 2x2 y, 2x2 zi. This
particular field is a gradient field with F~ = ∇f for f (x, y, z) = x2 (y 2 + z 2 ). Solutions to the
differential equations describe paths which are orthogonal to the level surfaces of f since the paths
are parallel to ∇f .
Perhaps you can see how this way of thinking might be productive towards analyzing otherwise
intractable problems in differential equations. I merely illustrate here to give a bit more breadth to
the concept of a vector field. Of course Stewart3 has pretty pictures with real world jutsu so you
should read that if this is not real to you without those comments.
3
we are covering chapter 17 from here on out
1.2. GRAD, CURL AND DIV 5
1.2 grad, curl and div

In this section we investigate a few natural derivatives we can construct with the operator ∇. Later
we will explain what these derivatives mean. First, the computation:
Definition 1.2.1.
Suppose f is a scalar function on R 3 then we defined the gradient vector field
grad(f ) = ∇f = h∂x f, ∂y f, ∂z f i
We studied this before, recall that we can compactly express this by

3
X
∇f = (∂i f )b
xi
i=1
where ∂i = ∂/∂xi and x1 = x, x2 = y and x3 = z. Moreover, we have also shown previously in

notes or homework that the gradient has the following important properties:
∇(f + g) = ∇f + ∇g, & ∇(cf ) = c∇f, & ∇(f g) = (∇f )g + f (∇g)
Together these say that ∇ is a derivation of differentiable functions on Rn .
Definition 1.2.2.
Suppose F~ = hF1 , F2 , F3 i is a vector field. We define:
∂F1 ∂F2 ∂F3

Div(F~ ) = ∇ • F~ = + + .
∂x ∂y ∂z
More compactly, we can express the divergence by

3
X
∇ • F~ = ∂i Fi .
i=1
You can prove that the divergence satisfies the following important properties:
∇ • (F~ + G)
~ = ∇ • F~ + ∇ • G
~ & ∇ • (cF~ ) = c∇ • F~ , & ~ = ∇f • G
∇ • (f G) ~ + f∇ • G
~
For example,
3
X 3
X 3
X
∇ • (F~ + cG)
~ = ∂i (Fi + cGi ) = ∂i Fi + c ∂i Gi = ∇ • F~ + c∇ • G.
~
i=1 i=1 i=1
Linearity of the divergence follows naturally from linearity of the partial derivatives.
Definition 1.2.3.
Suppose F~ = hF1 , F2 , F3 i is a vector field. We define:

~ ~ ∂F3 ∂F2 ∂F1 ∂F3 ∂F2 ∂F1
Curl(F ) = ∇ × F = − , − , −
∂y ∂z ∂z ∂x ∂x ∂y
More compactly, using the antisymmetric symbol ijk 4 ,

3
X
∇ × F~ = ijk (∂i Fj )b
xk .
i,j,k=1
You can prove that the curl satisfies the following important properties:
∇ × (F~ + G)
~ = ∇ × F~ + ∇ × G ~ & ∇ × (cF~ ) = c∇ × F~ , & ∇ × (f G) ~ = ∇f × G
~ + f ∇ × G.
~
For example,
3
X
∇ × (F~ + cG)
~ = ijk ∂i (Fj + cGj )b
xk
i,j,k=1
3
X 3
X
= ijk (∂i Fj )b
xk + c ijk (∂i Gj )b
xk
i,j,k=1 i,j,k=1
= ∇ × F~ + c∇ × G.
~
Linearity of the curl follows naturally from linearity of the partial derivatives.
It is fascinating how many of the properties of ordinary differentiation generalize to the case of
vector calculus. The main difference is that we now must take more care to not commute things
that don’t commute or confuse functions with vector fields. For example, while it is certainly true
that A~ ·B~ =B ~ ·A~ it is not even sensible to ask the question does ∇ · A
~=A
~ · ∇ ? Notice ∇ · A
~ is a
~
function while A · ∇ is an operator, apples and oranges.
The proposition below lists a few less basic identities which are at times useful for differential vector
calculus.
Proposition 1.2.4.
Let f, g, h be real valued functions on R and F~ , G,
~ H~ be vector fields on R then (assuming
all the partials are well defined )
(i.) ∇ · (F~ × G)
~ =G ~ · (∇ × F~ ) − F~ · (∇ × G)
~
(ii.) ∇(F · G) = F × (∇ × G) + G × (∇ × F~ ) + (F~ · ∇)G
~ ~ ~ ~ ~ ~ + (G ~ · ∇)F~
(iii.) ∇(F~ × G)
~ = (G~ · ∇)F~ − (F~ · ∇)G ~ + F~ (∇ · G)
~ − G(∇
~ · F~ )
4
recall we used this before to better denote harder calculations involving the cross-product
1.2. GRAD, CURL AND DIV 7
Proof: Consider (i.), let F~ = Fi ei and G

~ = P Gi ei as usual,
P
∇ · (F~ × G)
~ = P ∂k [(F~ × G)
P
~ k]
= P ∂k [ijk Fi Gj ]
= P ijk [(∂k Fi )Gj + Fi (∂k Gj )]
(1.1)
= ijk (∂i Fj )Gk + −Fj ikj (∂i Gk )
= Gk (∇ × F~ ) − Fj (∇ × G) ~
P
~ ~ ~
= G · (∇ × F ) − F · (∇ × G).~
where the sums

P above are taken over the indices which are repeated in the given expressions. In
physics the is often removed and the einstein index convention or implicit summation convention
is used to free the calculation of cumbersome summation symbols. The proof of the other parts of
this proposition can be handled similarly, although parts (viii) and (ix) require some thought so I
may let you do those for homework5 .
Proposition 1.2.5.
If f is a differentiable R-valued function and F~ is a differentiable vector field then
(i.) ∇ · (∇ × F~ ) = 0
(ii.) ∇ × ∇f = 0
(iii.) ∇ × (∇ × F~ ) = ∇(∇ · F~ ) − ∇2 F~
Before the proof, let me briefly indicate the importance of (iii.) to physics. We learn that in the
absence of charge and current the electric and magnetic fields are solutions of
~ = 0,
∇•E ~ = −∂t B,
∇×E ~ ~ = 0,
∇•B ~ = µo o ∂t E
∇×B ~
If we consider the curl of the curl equations we derive,

~ = ∇ × (−∂t B)
∇ × (∇ × E) ~ ⇒ ∇(∇ · E)
~ − ∇2 E
~ = −∂t (∇ × B)
~ ⇒ ∇2 E
~ = µo o ∂ 2 E.
~
t
~ = ∇ × (µo o ∂t E)
∇ × (∇ × B) ~ ⇒ ∇(∇ · B)
~ − ∇2 B
~ = µo o ∂t (∇ × E)
~ ⇒ ∇2 B
~ = µo o ∂t2 B.
~
These are wave equations. If you study the physics of waves you might recognize that the speed
√
of the waves above is v = 1/ µo o . This is the speed of light. We have shown that the speed
of light apparently depends only on the basic properties of space itself. It is indpendent of the
x, y, z coordinates so far as we can see in the usual formalism of electromagnetism. This math was
only possible because Maxwell added a term called the displacement current in about 1860. Not
many years later radio and TV was invented and all because we knew to look for the possiblility
thanks to this mathematics. That said, the notation used above was not common in Maxwell’s
time. His original presentation of what we now call Maxwell’s Equations was given in terms of 20
scalar partial differential equations. Now we enjoy the clarity and precision of the vector formalism.
5
relax fall 2011 students this did not happen to you
You might be interested to know that Maxwell (like many of the greatest 19-th century physicists)
was a Christian. Like Newton, they viewed their enterprise as revealing God’s general revelation.
Certainly their goal was not to remove God from the picture. They understood that the existence
of physical law does not relegate God to non-existence. Rather, we just get a clearer picture on
how He created the world in which we live. Just a thought. I have a friend who used to wear a
shirt with Maxwell’s equations and a taunt ”let there be light”, when he first wore it he thought
he was cleverly debunking God by showing these equations removed the need for God. Now, after
accepting Christ, he still wore the shirt but the equations don’t mean the same to him any longer.
The equations are evidence of God rather than his god.
Proof: I like to use parts (i.) and (ii.) for test questions at times. They’re pretty easy, I
leave
P3 them to the reader. The6 proof of (iii.) is a bit deeper. We need the well-known identity
j=1 ikj lmj = δil δkm − δkl δim
3
X
∇ × (∇ × F~ ) = ijk ∂i (∇ × F~ )j x
bk
i,j,k=1
3
X 3
X
= ijk ∂i ( lmj ∂l Fm )b
xk
i,j,k=1 l,m=1
3
X
= ijk lmj (∂i ∂l Fm )b
xk
i,j,k,l,m=1
3
X
= −ikj lmj (∂i ∂l Fm )b
xk
i,j,k,l,m=1
3
X
= (−δil δkm + δkl δim )(∂i ∂l Fm )b
xk
i,k,l,m=1
3
X 3
X
= (−δil δkm ∂i ∂l Fm )b
xk + (δkl δim ∂i ∂l Fm )b
xk
i,k,l,m=1 i,k,l,m=1
3
X 3
X
= −∂i ∂i (Fk x
bk ) + (∂i ∂k Fi )b
xk
i,k=1 i,k=1
3
X 3
X 3
X X3
=− ∂i ∂i ( Fk x
bk ) + ∂k ( ∂i Fi )b
xk
i=1 k=1 k=1 i=1
= −∇ F + ∇(∇ • F~ ).
2~

6
this is actually just the first in a whole sequence of such identities linking the antisymmetric symbol and the
kronecker deltas... ask me in advanced calculus, I’ll show you the secret formulas
1.3. LINE INTEGRALS 9
1.3 line integrals

In this section I describe several integrals over curves. These integrals capture something intrinsic
between the given function or vector field and the given curve. For convenience we state the
definitions in terms of a particular type of parametrized path, but we prove that the choice of
the parameter is independent from the value of the integral. Placing a parameter on a curve is
a choice ofR coordinates and our integrals are coordinate free. For the integral with respect to
arclength C f ds, or scalar line integral, we are adding up some f scalar function along the curve.
On the other hand, the line-integral C F~ • d~r is taken along an oriented curve and it computes the
R
amount which the vector field F~ points tangentially to the curve. From a physical perspective,
the line-integral gives the work done by a force over a curve. This is much more general than the
simplisitic constant force or constant direction idea of work you have seen in previous portions of
the calculus sequence. 7 . The organization of this section is as follows: we begin with curves and
some terminology, the define the integral with respect to arclength, the line integral and finally we
investigate connections between the two concepts as well as the application of differential notation
as an organizing principle for quick assembly of a line-integral.
1.3.1 curves and paths

Definition 1.3.1.
A path in R3 is a continuous function ~γ with connected domain I such that ~γ : I ⊂ R → R3 .
If ∂I = {a, b} then we say that ~γ (a) and ~γ (b) are the endpoints of the path ~γ . When ~γ
has continuous derivatives of all orders we say it is a smooth path (of class C ∞ ), if it has
at least one continuous derivative we say it is a differentiable path( of class C 1 ). When
I = [a, b] then the path is said to go from ~γ (a) = P to ~γ (b) = Q and the image C = ~γ ([a, b])
is said to be an oriented curve C from P to Q. The same curve from Q to P is denoted
−C. We say C and −C have opposite orientations.
Hopefully most of this is already familar from our earlier work on parametrizations. I give another
example just in case.
Example 1.3.2. The line-segment L from (1, 2, 3) to (5, 5, 5) has parametric equations x = 1 +
4t, y = 2 + 3t, z = 3 + 2t for 0 ≤ t ≤ 1. In other words, the path ~γ (t) = h1 + 4t, 2 + 3t, 3 + 2ti covers
the line-segment L. In constrast −L goes from (5, 5, 5) to (1, 2, 3) and we can parametrize it by
x = 5−4u, y = 5−3u, z = 5−2u or in terms of a vector-formula ~γreverse (u) = h5−4u, 5−3u, 5−2ui.
How are these related? Observe:
~γreverse (0) = ~γ (1) & ~γreverse (1) = ~γ (0)
Generally, ~γreverse (t) = ~γ (1 − t).

We can generalize this construction to other curves. If we are given C from P to Q parametrized
by ~γ : [a, b] → R3 then we can parametrize −C by ~γreverse : [a, b] → R3 defined by ~γreverse (t) =
Rb
7 ~ • 4~
recall W = F x from the beginning of this course and W = F (x)dx from calculus I or II
a
~γ (a + b − t). Clearly we have ~γreverse (a) = ~γ (b) = Q whereas ~γreverse (b) = ~γ (a) = Q. Perhaps it is
interesting to compare these paths at a common point,
γ(t) = ~γreverse (a + b − t)
The velocity vectors naturally point in opposite directions, (by the chain-rule)
d~γ d~γreverse
(t) = − (a + b − t).
dt dt
Example 1.3.3. Suppose ~γ (t) = hcos(t), sin(t)i for π ≤ t ≤ 2π covers the oriented curve C. If we
wish to parametrize −C by β~ then we can use
~ = ~γ (3π − t) = hcos(3π − t), sin(3π − t)i

β(t)
~ = h− cos(t), − sin(t)i for π ≤ t ≤ 2π. You can easily verify

Simplifying via trigonometry yields β(t)
that β~ covers the lower half of the unit-circle in a CW-fashion, it goes from (1.0) to (−1, 0)
What I have just described is a general method to reverse a path whilst keeping the same domain
for the new path. Naturally, you might want to use a different domain after you change the
parametrization of a given curve. Let’s settle the general idea with a definition. This definition
describes what we allow as a reasonable reparametrization of a curve.
Definition 1.3.4.
Let ~γ1 : [a1 , b1 ] → R3 be a path. We say another path ~γ2 : [a2 , b2 ] → R3 is a

reparametrization of ~γ1 if there exists a bijective (one-one and onto), continuous func-
tion u : [a1 , b1 ] → [a2 , b2 ] with continuous inverse u−1 : [a2 , b2 ] → [a1 , b1 ] such that
~γ1 (t) = ~γ2 (u(t)) for all t ∈ [a1 , b1 ]. If the given curve is smooth or k-times differentiable then
we also insist that the transition function u and its inverse be likewise smooth or k-times
differentiable.
In short, we want the allowed reparametrizations to capture the same curve without adding any
artificial stops, starts or multiple coverings. If the original path wound around a circle 10 times
then we insist that the allowed reparametrizations also wind 10 times around the circle. Finally,
let’s compare the a path and its reparametrization’s velocity vectors, by the chain rule we find:
d~γ1 du d~γ2
~γ1 (t) = ~γ2 (u(t)) ⇒ (t) = (u(t)).
dt dt dt
This calculation is important in the section that follows. Observe that:
1. if du/dt > 0 then the paths progress in the same direction and are consitently oriented
2. if du/dt < 0 then the paths go in opposite directions and are oppositely oriented
Reparametrizations with du/dt > 0 are said to be orientation preserving.

1.3.2 line-integral of scalar function

These are also commonly called the integral with respect to arclength. In lecture we framed
the need for this definition by posing the question of finding the area of a curved fence with height
f (x, y). It stood to reason that the infinitesimal area dA of the curved fence over the arclength ds
would simply be dA = f (x, y)ds. Then integration is used to sum all the little areas up. Moreover,
the natural calculation to accomplish this is clearly as given below:
Definition 1.3.5.
Let ~γ : [a, b] → C ⊂ Rn be a differentiable path and suppose that C ⊂ dom(f ) for a
continuous function f : dom(f ) → R then the scalar line integral of f along C is
Z Z b
f ds ≡ f (~γ (t)) ||~γ 0 (t)|| dt.
C a
We should check to make sure there is no dependence on the choice of parametrization above. If
there was then this would not be a reasonable definition. Suppose ~γ1 (t) = ~γ2 (u(t)) for a1 ≤ t ≤ b1
where u : [a1 , b1 ] → [a2 , b2 ] is differentiable and strictly monotonic. Note
Z b1 Z b1
d~γ1 du d~γ2
f (~γ1 (t))
dt = f (~γ2 (u(t)))
(u(t)) dt
a1 dt
a1 dt dt
Z b1
d~γ2 du
= f (~γ2 (u(t)))
(u(t)) · dt
a1 dt dt
If u is orientation preserving then du/dt > 0 hence u(a1 ) = a2 and u(b1 ) = b2 and thus
Z b1 Z b1
d~γ1 d~γ2 du
f (~γ1 (t))
dt = f (~γ2 (u(t)))
(u(t)) dt
a1 dt
a1 dt dt
Z b2
d~γ2
= f (~γ2 (u)
du.
a2 du
On the other hand, if du/dt < 0 then |du/dt| = −du/dt and the bounds flip since u(a1 ) = b2 and
u(b1 ) = a2
Z b1 Z b1
d~γ1 d~γ2 du
f (~γ1 (t))
dt = − f (~γ 2 (u(t))) (u(t)) dt dt

a1 dt a1
dt
Z a2
d~γ2
=− f (~γ2 (u) du.
b2 du
Z b2
d~γ2
= f (~γ2 (u) du.
a2 du
Note, the definition requires me to flip the bounds before I judge if we have the same result. This
is implicit in the statement in the definition that dom(~γ ) = [a, b] this forces a < b and hence the
integral in turn. Technical details aside we have derived the following important fact:
Z Z
f ds = f ds
C −C
The scalar-line integral of function with no attachment to C is indpendent of the orientation of the
curve. Given our original motivation for calculating the area of a curved fence this is not surprising.
One convenient notation calculation of the scalar-line integral is given by the dot-notation of New-
ton. Recall that dx/dt = ẋ hence ~γ = hx, y, zi has ~γ 0 (t) = hẋ, ẏ, żi. Thus, for a space curve,
Z Z b p
f ds ≡ f (x, y, z) ẋ2 + ẏ 2 + ż 2 dt.
C a
We can also calculate the scalar line integral of f along some curve which is made of finitely many
differentiable segments, we simply calculate each segment’s contribution and sum them together.
Just like calculating the integral of a piecewise continuous function with a finite number of jump-
discontinuities, you break it into pieces.
Furthermore, notice that if we calculate the scalar line integral of the constant function f = 1 then
we will obtain the arclength of the curve. More generally the scalar line integral calculates the
weighted sum of the values that the function f takes over the curve C. If we divide the result by
the length of C then we would have the average of f over C.
Example 1.3.6. Suppose the linear mass density of a helix x = R cos(t), y = R sin(t), z = t is given
by dm/dz = z. Calculate the total mass around the two twists of the helix given by 0 ≤ t ≤ 4π.
Z Z 4π p
mtotal on C = zds = z ẋ2 + ẏ 2 + ż 2 dt (1.2)
C 0
Z p 4π
= t R2 + 1 dt
0
√
t2 R2 + 1 4π

=
2
0
p
= 8π 2 R2 + 1.
In constrast to total mass we could find the arclength by simply adding up ds, the total length L of
C is given by
Z Z 4π p
L= ds = ẋ2 + ẏ 2 + ż 2 dt
C 0
Z 4π p
= R2 + 1 dt
0
p
= 4π R2 + 1.
Definition 1.3.7.
Let C be a curve with length L then the average of f over C is given by
Z
1
favg = f ds.
L C
Example 1.3.8. The average mass per unit length of the helix with dm/dz = z as studied above
is given by Z
1 1 p
mavg = f ds = √ 8π 2 R2 + 1 = 2π.
L C 4π R2 + 1
Since z = t and 0 ≤ t ≤ 4π over C this result is hardly surprising.
Another important application of the scalar line integral is to find the center of mass of a wire.
The idea here is nearly the same as we discussed for volumes, the difference is that the mass is
distributed over a one-dimensional space so the integration is one-dimensional as opposed to two-
dimensional to find the center of mass for a planar laminate or three-dimensional to find the center
of mass for a volume.
Definition 1.3.9.
Let C be a curve with length L and R suppose dM/ds = δ is the mass-density of C. The total
mass of the curve found by M = c δds. The centroid or center of mass for C is found
at (x̄, ȳ, z̄) where
Z Z Z
1 1 1
x̄ = xδ ds, ȳ = yδ ds, z̄ = zδ ds.
M C M C M C
Often the centroid is found off the curve.
Example 1.3.10. Suppose x = RR cos(t), yR = R sin(t), z = h for 0 ≤ t ≤ π for a curve with δ = 1.

π
Clearly ds = Rdt and thus M = C δds = 0 Rdt = πR. Consider,
Z Z π
1 1
x̄ = xds = R2 cos(t)dt = 0
πR C πR 0
whereas,
Z Z π
π
1 1 2 1 2
2R
ȳ = yds = R sin(t)dt = (−R cos(t) =
πR C πR 0 πR 0 π
The reader can easily verify that z̄ = h hence the centroid is at (0, 2R
π , h).
Of course there are many other applications, but I believe

R these should suffice for our current
purposes. We will eventually learn that C F~ • T~ ds and C F~ • N ~ ds are also of interest, but we
R
should cover other topics before returning to these. Incidentally, it is pretty obvious that we have
the following properties for the scalar-line integral:
Z Z Z Z Z Z
(f + cg)ds = f ds + c gds & f ds = f ds + f ds
C C C C∪C̄ C C̄
R R
in addition if f ≤ g on C then C f ds ≤ C gds. I leave the proof to the reader.
1.3.3 line-integral of vector field
For those of you who know a little physics, the motivation to define this integral follows from our
desire to calculate the work done by a variable force F~ on some particle as it traverses C. In
particular, we expect the little bit of work dW done by F~ as the particle goes from ~r to ~r + d~r is
given by F~ • d~r . Then, to find total work, we integrate:
Definition 1.3.11.
Let ~γ : [a, b] → C ⊂ R3 be a differentiable path which covers the oriented curve C and
suppose that C ⊂ dom(F~ ) for a continuous vector field F~ on R3 then the vector line
integral of F~ along C is denoted and defined as follows:
Z Z b
d~γ
F~ • d~r = F~ (~γ (t)) • dt.
C a dt
This integral measures the work done by F~ over C. Alternatively, this is also called the circulation
of F~ along C, however that usage tends to appear in the case that C is a loop. A closed curve is
defined to be a curve which has the same starting and ending points. We can indicate the line-
integral is taken over a loop by the notation C F~ • d~r. As with the case of the scalar line integral
H
we ought to examine the dependence of the definition on the choice of parametrization for C. If
we were to find a dependence then we would have to modify the definition to make it reasonable.
Once more consider the reparametrization γ~2 of γ~1 by a strictly monotonic differentiable function
u : [a1 , b1 ] → [a2 , b2 ] where we have γ~1 (t) = γ~2 (u(t)). Consider,

Z b1 Z b1
d~γ1 d
F~ (~γ1 (t)) • dt = F~ (~γ2 (u(t))) ~γ2 (u(t)) dt
a1 dt a1 dt
Z b1
d~γ2 du
= F~ (~γ2 (u(t))) (u(t)) · dt
a1 dt dt
Z u(b1 )
d~γ2
= F~ (~γ2 (u) du
u(a1 ) du
Rb
If u is orientation preserving then u(a1 ) = a2 and u(b1 ) = b2 and we find the integral a11 F~ (~γ1 (t)) • d~
γ1
dt dt =
R b2
~ γ2 (u) d~γ2 du. However, if u is orientation reversing then we find u(a1 ) = b2 and u(b1 ) = a2
a2 F (~
R b du Ra
hence 1 F~ (~γ1 (t)) • d~γ1 dt = − 2 F~ (~γ2 (u) d~γ2 du. Therefore, we find that
a1 dt b2 du
Z Z
F~ • d~r = − F~ • d~r
−C C
This is actually not surprising if you think about the motivation for the integral. The integral
measures how the vector field F~ points in the same direction as C. The curve −C goes in the
opposite direction thus it follows the sign should differ for the line-integral. Long story short, we
must take line-integrals with respect to oriented curves.
Is instructive to relate the line-integral and the integral with respect to arclength8 ,
b b
~γ 0 (t)
Z Z Z Z
d~γ 0
F~ • d~r = F~ (~γ (t)) • dt = F~ (~γ (t)) • ||~γ (t)|| dt = (F~ • T~ )ds.
C a dt a ||~γ 0 (t)|| C
As the last equality indicates, the vector line integral of F~ is given by the scalar line integral of the
tangential component F~ · T~ . Thus the vector line integral of F~ along C gives us a measure of how
much F~ points in the same direction as the oriented curve C. If the vector field always cuts the
path perpendicularly ( if it was normal to the curve ) then the vector line integral would be zero.
Example 1.3.12. Suppose F~ (x, y, z) = hy, z − 1 + x, 2 − xi and suppose C is an ellipse on the plane
z = 1 − x − y where x2 + y 2 = 4 and we orient C in the CCW direction relative to the xy-plane
with positive normal (imagine pointing your right hand above the xy-plane and your fingers curl
~
R
around the ellipse in the CCW direction). Our goal is to calculate C F d~r. To do this we must
•
first understand how to parametrize the ellipse:
x = 2 cos(t), y = 2 sin(t)
8
think about this equality with −C in place of C, why is this not a contradiction? On first glance you might think
only the lhs is orientation dependent.
gives x2 + y 2 = 4 and the CCW direction. To find z we use the plane equation,
z = 1 − x − y = 1 − 2 cos(t) − 2 sin(t)
Therefore,
~r(t) = h2 cos(t), 2 sin(t), 1 − 2 cos(t) − 2 sin(t)i
thus
d~r
= − 2 sin(t), 2 cos(t), 2[sin(t) − cos(t)]
dt
Evaluate F~ (x, y, z) = hy, z − 1 + x, 2 − xi at x = 2 cos(t), y = 2 sin(t) and z = 1 − 2 cos(t) − 2 sin(t)
to find
F~ (~r(t)) = 2 sin(t), −2 sin(t), 2[1 − cos(t)]
Now put it together,

I Z 2π
F~ • d~r = 2 sin(t), −2 sin(t)), 2[1 − cos(t)] • − 2 sin(t), 2 cos(t), 2[sin(t) − cos(t)] dt
C 0
Z 2π
2
= −4 sin (t) − 4 sin(t) cos(t) + 4[1 − cos(t)][sin(t) − cos(t)] dt
0
Z 2π
2 2
= −4 sin (t) − 4 sin(t) cos(t) + 4 sin(t) − 4 cos(t) − 4 cos(t) sin(t) + 4 cos (t) dt
0
Z 2π
2
cos (t) − sin2 (t) dt

=4
0
Z 2π
1 1
=4 (1 + cos(2t) − (1 − cos(2t)) dt
0 2 2
Z 2π
1 1
=4 (1 + cos(2t) − (1 − cos(2t)) dt
0 2 2
Z 2π

=4 cos(2t) dt
0
= 0.
The example above indicates how we apply the definition of the line-integral directly. Sometimes
it is convenient to use differential notation. If C is parametrized by ~r = hx, y, zi for a ≤ t ≤ b we
define the integrals of the differential forms P dx, Qdy and Rdz in the following way:
Definition 1.3.13.
Let ~r : [a, b] → C ⊂ R3 be a differentiable path which covers the oriented curve C and
suppose that C ⊂ dom(hP, Q, Ri) for a continuous vector field hP, Q, Ri on R3 then we
define
Z Z b Z Z b Z Z b
dx dy dz
P dx = P (~r(t)) dt, Qdx = Q(~r(t)) dt, Rdx = R(~r(t)) dt,
C a dt C a dt C a dt
1.4. CONSERVATIVE VECTOR FIELDS 17
These areR not basic calculations and in and of themselves they are not terribly interesting. I suppose
that the C P dx measures the work done by the x-vector-component of F~ = hP, Q, Ri whereas the
~
R R
C Qdyl and the C Rdz measure the work done by the y and z vector components of F = hP, Q, Ri.
Primarily, these are interesting since when we add them we obtain the full line-integral:
Z Z

hP, Q, Ri d~r =
• P dx + Qdy + Rdz
C C
I invite the reader to verify the formula above. I will illustrate its use in many examples to follow.
It should be emphasized that these are just notation to organize the line integral.
Example 1.3.14. Calculate C F~ • d~r for F~ (x, y, z) = hy, z−1+x, 2−xi given that C is parametrized
R
by x = cos(t), y = sin(t), z = 1 for 0 ≤ t ≤ 2π. Note that
dx = − sin(t)dt, dy = cos(t)dt, dz = 0
Thus, for P = y = sin(t) , Q = z − 1 + x = cos(t) and R = 2 − x = 2 − cos(t) we find

I Z Z 2π Z 2π
~ 2

F d~r =
• P dx + Qdy + Rdz = − sin (t)dt + cos(t) cos(t)dt = cos(2t)dt = 0.
C C 0 0
You might wonder if the integral around a closed curve is always zero.
Example 1.3.15. Let F~ = hy, −xi and suppose x = R cos(t), y = R sin(t) parametrizes C for
0 ≤ t ≤ 2π. Calculate,
P dx + Qdy = −yR sin(t)dt − xR cos(t)dt = −R2 sin2 (t)dt − R2 cos2 (t)dt = −R2 dt
R 2π
Thus, C F~ • d~r = − 0 R2 dt = −2πR2 .
H
Apparently just because we integrate around a loop it does not mean the answer is zero. I suspect
that there are loops for which F~ (x, y, z) = hy, z − 1 + x, 2 − xi. We will
H return to that example
once more in the next section after we learn a test to determine if the C F~ • d~r = 0 without direct
calculation. In conclusion, I should mention that the properties below are easily proved by direct
calculation on the defintion,
Z Z Z Z Z Z
~ ~
(F + cG) d~r =
• ~
F d~r + c
• ~
G d~r
• & ~
F d~r =
• ~
F d~r +
• F~ • d~r.
C C C C∪C̄ C C̄
1.4 conservative vector fields

In this section we discuss how to identify a conservative vector field and how to use it. There are
about 5 equivalent ideas and our job in this section is to explore how these concepts are connected.
We also make a few connections with physics and it should be noted that part of the terminology
is certainly borrowed from classical mechanics. Let us begin with the fundamental theorem for line
integrals.
Theorem 1.4.1.
Suppose f is differentiable on some open set containing the oriented curve C from P to Q
then Z
∇f • d~r = f (Q) − f (P ).
C
Proof: let ~r : [a, b] → C ⊂ Rn parametrize C and calculate:

Z Z b
d~r
∇f • d~r = ∇f (~r(t)) • dt
C a dt
Z b
d
= f (~r(t)) dt
a dt
= f (~r(b)) − f (~r(a))
= f (Q) − f (P ).
The two critical steps above are the application of the multivariate chain-rule and then in the next
to last step we apply the FTC from single-variable calculus.
Definition 1.4.2. conservative vector field
Suppose U ⊆ Rn then we say F~ is conservative on U iff there exists a potential function

f such that F~ = ∇f on U . Moreover, if F~ is conservative on dom(F~ ) then we say F~ is a
conservative vector field.
The beauty of a conservative vector field is we trade computation of a line-integral for evaluation
at the end-points.
Example 1.4.3. Suppose F~ (x, 3

R y, z) = h2x, 2y, 3i for all (x, y, z) ∈ R . Suppose C is a curve from
(0, 0, 0) to (a, b, c). Calculate C F~ • d~r. Observe that
f (x, y, z) = x2 + y 2 + 3z ⇒ F~ = ∇f.
Therefore, Z
F~ • d~r = f (a, b, c) − f (0, 0, 0) = a2 + b2 + 3c.
C
Notice that we did not have to know where the curve C went since the FTC applies and only the
endpoints of the curve are needed. In invite the reader to check this result by explicit computation
along some path.
Why ”conservative”? Let me address that. The key is a little identity, if m is a constant,

d 1 2 d 1 1 d~v d~v
mv = m~v • ~v = m •~
v + ~v • = m~a • ~v .
dt 2 dt 2 2 dt dt
If F~ is the net-force on a mass m then Newton’s Second Law states F~ = m~a therefore, if C is a
curve from ~r1 to ~r2
Z Z t2 Z t2 Z t2
~ ~ d~r d 1 2
F • d~r = F • dt = m~a • ~v dt = mv dt = K(t2 ) − K(t1 )
C t1 dt t1 t1 dt 2
where K = 21 mv 2 is the kinetic energy. This result is known as the work-energy theorem. It does
not require that F~ be conservative. If F~ is conservative then it is traditional to choose a potential
energy function U such that F~ = −∇U . In this case we can use the FTC for line-integrals to once
more calculate the work done by the net-force,
Z Z
~
F • d~r = − ∇U • d~r = −U (~r2 ) + U (~r1 )
C C
It follows that we have, for a conservative force, K2 − K1 = −U2 + U1 hence K1 + U1 = K2 + U2 .

The quantity E = U + K is the total mechanical energy and it is a constant of the motion when
only conservative forces comprise the net-force. This is the reason I call a vector field which is a
gradient field of some pontential a conservative vector field. When viewed as a net-force it provides
the conservation of energy9 . It turns out that usually we can find portions of the domain of an
arbitrary vector field on which the vector field is conservative. The obstructions to the existence of
a global potential are the interesting part.
Definition 1.4.4. path-independence
Suppose U ⊆ Rn then we say F~ is path-independent on U iff C1 F~ • d~r = C2 F~ • d~r for

R R
each pair of curves C1 , C2 ⊂ U beginning at P and terminating at Q .
9
it is worth noticing that while physically this is most interesting to three dimensions, the math allows for more
Proposition 1.4.5.
Suppose U is an open connected subset of Rn then the following are equivalent
1. F~ is conservative; F~ = ∇f on all of U
2. F~ is path-independent on U
3. C F~ • d~r = 0 for all closed curves C in U

H
4. (add precondition n = 3 and U be simply connected) ∇ × F~ = 0 on U .
Proof: We postpone the proof of (4.) ⇒ (1.). However, we can show that (1.) ⇒ (4.). Suppose
F~ = ∇f . Note that ∇ × F~ = ∇ × ∇f = 0. I included this here since we can quickly test to see if
Curl(F~ ) 6= 0. When the curl is nontrivial then we can be certain the given vector field is not conser-
vative. On the other hand, vanishing curl is only useful if it occurs over a simply connected domain10
(1.) ⇒ (2.). Assume F~ = ∇f . Suppose C1 , C2 are two curves which both start at P and end at Q
in the set U . Apply the FTC for line-integrals in what follows:
Z Z
F~ • d~r = ∇f • d~r = f (Q) − f (P ).
C1 C1
F~ • d~r =
R R
Likewise, C2 C2 ∇f • d~r = f (Q) − f (P ). Therefore (2.) holds true.
(2.) ⇒ (1.). Assume F~ is path-independent. Pick R some point A ∈ U and let C be any curve in U
from A to B = (x, y, z). We define f (x, y, z) = C F~ • d~r. This is single-valued since we assume F~
~ ~
is path-independent. We need to show thatR ∇f = F . Denote F = hP, Q, Ri. We begin by isolating
∂ ~
the x-component. We need to show ∂x C F d~r = P (x, y, z). We can write C as curve Cx from
•
A to Bx = (xo , y, z) with xo < x pasted togther with the line-segment Lx from Bx to B. Observe
that the curve Cx has no dependence on x (of the B point)
Z Z Z Z
∂ ~ ∂ ~ ~ ∂ ~
F d~r =
• F d~r +
• F d~r =
• F d~r
•
∂x C ∂x Cx Lx ∂x Lx
The line segment Lx has parametrization ~r(t) = ht, y, zi for xo ≤ t ≤ x. We calculate that
Z Z x Z x
~
F • d~r = ~
F (t, y, z) • h1, 0, 0idt = P (t, y, z)dt
Lx xo xo
Therefore, Z Z x
∂ ∂
F~ • d~r = P (t, y, z)dt = P (x, y, z).
∂x C ∂x xo
10
a simply connected domain is a set with no holes, any loop can be smoothly shrunk to a point, it has a boundary
which is a simple curve. A simple curve is a curve with no self-intersections but perhaps one in the case it is closed.
A circle is simple a figure 8 is not.
We can give similar arguments to show that

Z Z
∂ ∂
F~ • d~r = Q & F~ • d~r = R.
∂y C ∂z C
We find F~ s conservative.
(2.) ⇒ (1.). Assume F~ is path-independent. Consider a closed curve C in U . Notice we can pick
any pair of points P, Q on C and write C1 from P to Q and C2 from Q to P such that C = C1 ∪ C2 .
Furthermore, note that −C2 also goes from P to Q. Path independence yields
Z Z Z Z Z Z I
F~ • d~r = F~ • d~r ⇒ 0 = F~ • d~r − F~ • d~r = F~ • d~r + F~ • d~r = F~ • d~r.
C1 −C2 C1 −C2 C1 C2 C
Conequently (3.) is true.
(3.) ⇒ (2.). Suppose C F~ • d~r = 0 for all closed curves C in U . Suppse C1 and C2 start at P and
H
end at Q. Observe that C = C1 ∪ (−C2 ) is a closed curve hence

I I Z Z Z Z
0= ~
F • d~r ⇒ 0 = ~
F • d~r = ~
F • d~r + ~
F • d~r = ~
F • d~r − F~ • d~r.
C C C1 −C2 C1 C2
Clearly (2.) follows.
To summarize we have shown (1.) ⇔ (2.) ⇔ (3.) and (1.) ⇒ (4.). We postpone the proof that
(3.) ⇒ (4.) and (4.) ⇒ (1.) .
The point A where f (A) = 0 is known as the zero for the potential. You should notice that the
choice for f is not unique. If we add a constant c to the potential function f then we obtain the
same gradient field; ∇f = ∇(f + c). In physics this is the freedom to set the potential energy to
be zero at whichever point is convenient.
Example 1.4.6. In electrostatics, the potential energy per unit charge is called the voltage or simply
the electric potential. For finite, localized charg distributions the electric potential is defined by
Z (x,y,z)
V (x, y, z) = − ~ • d~r
E
∞
The electric field of a charge at the origin is given by E~ = k2 ρb. We take the line from the origin to
ρ
spatial infinity11 to calculate the potential.
Z ρ
k k
V (ρ) = − 2
dρ = .
∞ ρ ρ
RQ
The notation P indicates the line integral is taken over a path from P to Q. This notation is only
unambiguous if we are working with a conservative vector field.
11
the claim implicit within such a convention is that it matters not which unbounded direction the path begins,
for convenience we usually just use a line which extends to ∞
1.5 green’s theorem

The fundamental theorem of calculus shows that integration and differentiation are inverse pro-
cesses in a certain sense. It is natural to seek out similar theorems for functions of several variables.
We begin our search by defining the flux through a simple closed planar curve12 . It is just the
scalar integral of the outward-facing normal component to the vector field. Then we examine how
a vector field flows out of a little rectangle. This gives rises us reason to define the divergence. In
some sense this little picture will derive the first form of Green’s Theorem.
Being discontent with just one interpretation, we turn to analyze how the vector field circulates
around a given CCW curve. We again look at a little rectangle and quantify how a given vector field
twists around the square loop. This leads us to another derivation of Green’s Theorem. Moreover,
it gives us the reason to define the curl of a vector field.
Finally, we offer a proof which extends the toy derivations to a general Type I & II curve. Past
that, properties of the line-integral extend our result to general regions in the plane. Applications
to the calculation of area and the analysis of conservative vector fields are given. I conclude this
section with a somewhat formal introduction to two-dimensional electrostatics, I show how Green’s
Theorem naturally supports Gauss’ Law for the plane.
1.5.1 geometry of divergence in two dimensions

A curve is said to be simple if it has no self-intersections except perhaps one. For example, a circle
is a simple curve whereas a figure 8 is not. Both circles and figure 8’s are closed curves since they
have no end points (or you could say they have the same starting and ending points). In any event,
we define the number of field lines which cut through a simple curve by the geometrically natural
definition below:
Definition 1.5.1. flux of F~ through a simple curve C.
Suppose F~ is is continuous on a open set containing the closed simple curve C. Define:
I
ΦC = (F~ • ~n)ds
C
Where ~n is the outward-facing unit-normal to C.

Recall that if ~r(s) = hx(s), y(s)i is the arclength parametrization of C then the unit-tangent vector
of the Frenet frame was defined by T~ (s) = d~ r 13
ds . For the sake of visualization suppose C is CCW
12
these are known as Jordan curves
13 ~ (s) = 1 ~
dT
yes the unit normal is defined by and N . However, this is not the ~n which we desire because
T 0 (s) ds
~
N sometimes points inward. Also, direct computation brings us to second derivatives and the geometric argument
above avoids that difficulty
1.5. GREEN’S THEOREM 23
oriented calculate ~n. Since T~ • ~n = 0 there are only two choices once we calculate ~n14 . We choose
the ~n which points outward.

~ dx dy dy dx
T = , ⇒ ~n = , −
ds ds ds ds
The picture below helps you see how the outward normal formula works:
Let’s calculate the flux given this identity. Consider a vector field F~ = hP, Qi and once more the
Jordan curve C with outward normal ~n, suppose length of C is L,
I
ΦC = F~ • ~n
C
Z L

dy dx
= P, Q • , − ds
0 ds ds
Z L
dy dx
= P −Q ds
0 ds ds
Z L

dx dy
= − Q, P • , ds
ds ds
I0
= P dy − Qdx.
C
This formula is very nice. It equally well applies to closed simple curves which are only mostly
smooth. If we have a few corners on C then we can still calculate the flux by calculating flux
through each smooth arc and adding together to find the net-flux. To summarize:
Proposition 1.5.2.
Suppose C is a piecewise-smooth, simple, closed CCW oriented curve. If F~ is continuous
on an open set containing C then the flux through C is given by
I I
ΦC = ~
F • ~n = P dy − Qdx.
C C
14
b = ha, bi then vb = hb, −ai or vb = h−b, ai are the only perpendicular unit-vectors to u
if u b
If
you were to consider the CW-oriented curve −C then the outward-normal is given by ~n =
dy dx
− ds , ds and the formula for flux is
I I
ΦC = F~ • ~n = Qdx − P dy.
−C −C
This formula is in some sense left-handed, hence evil, so we whilst not use it hence forth.
Now we turn to the task of approximating the flux by direct computation. Consider a little rectangle
R with corners at (x, y), (x + 4x, y), (x + 4x, y + 4y), (x, y + 4y).
To calculate the flux of F~ = hP, Qi through the rectangle we simply find the flux through each side
and add it up.
1. Top: (F~ • yb)4x = Q(x, y + 4y)4x
2. Base: (F~ • [− yb])4x = −Q(x, y)4x
3. Left: (F~ • [− x
b])4y = −P (x, y)4y
4. Right: (F~ • x
b)4y = P (x + 4x, y)4y
The net-flux through R is thus,

ΦR = Q(x, y + 4y) − Q(x, y) 4x + P (x + 4x, y) − P (x, y) 4y
Observe that
ΦR Q(x, y + 4y) − Q(x, y) P (x + 4x, y) − P (x, y)
= +
4x4y 4y 4x
In this limit 4x → 0 and 4y → 0 the expressions above give partial derivatives and we find that:
dΦR ∂P ∂Q
= +
dA ∂x ∂y
∂Q
This suggests we define the flux density as ∇ • F~ = ∂P
∂x + ∂y . If we integrate this density over a
finite region then we will find the net flux through the region (jump!). In any event, we at least
have good reason to suspect that
I ZZ I ZZ
∂P ∂Q
Φ∂R = F~ • ~n = ∇ • F~ dA ⇒ P dy − Qdx = + dA
∂R R ∂R R ∂x ∂y
This is for obvious reasons called the divergence form of Green’s Theorem. We prove this later in
this section. As I mentioned in lecture I found these thoughts in Thomas’ Calculus text, however,
I suspect we’ll find them in many good calculus texts at this time.
1.5.2 geometry of curl in two dimensions

Recall that C F~ • d~r calculates the work done by F~ around the loop C. This line-integral is also
H
called the circulation. Why? If we think of F~ as the velocity field of some liquid then a positive
circulation around a CCW loop suggests that a little paddle wheel placed at the center of the loop
will spin in the CCW direction. The greater the circulation the faster it spins. Let us duplicate the
little rectangle calculation of the previous section to see what meaning, if any, the circulation per
area has: Once more, consider a little rectangle R with corners at (x, y), (x + 4x, y), (x + 4x, y +
4y), (x, y + 4y).
To calculate the flow15 of F~ = hP, Qi through the rectangle we simply find the flow through each
side and add it up.
1. Top: (F~ • [− x
b])4x = −P (x, y + 4y)4x
2. Base: (F~ • x
b)4x = P (x, y)4x
3. Left: (F~ • [− yb])4y = −Q(x, y)4y
4. Right: (F~ • yb)4y = Q(x + 4x, y)4y

The net-circulation around R is thus,

WR = Q(x + 4x, y) − Q(x, y) 4y − P (x, y + 4y) − P (x, y) 4x
Observe that
WR Q(x + 4x, y) − Q(x, y) P (x, y + 4y) − P (x, y)
= −
4x4y 4x 4y
In this limit 4x → 0 and 4y → 0 the expressions above give partial derivatives and we find that:
dΦR ∂Q ∂P
= −
dA ∂x ∂y
15
the flow around a closed loop is called circulation, but flow is the term for a curve which is not closed
In the case that ∂Q ∂P

∂x − ∂y = 0 we say the velocity field is irrotational since it does not have
the tendency to generate rotation at the point in question. The curl of a vector field measures
how a vector field in R3 rotates about planes with normals x b, yb, zb. In particular, we defined
Curl(F~ ) = ∇ × F~ . If F~ (x, y, z) = hP (x, y), Q(x, y), 0i then

∂Q ∂P
Curl(F~ ) = 0, 0, −
∂x ∂y
and we just derived that nonzero ∂Q ∂P

∂x − ∂y will spin a little paddle wheel with axis z b. If F~ • zb 6= 0
and/or P, Q had nontrivial z-dependence then we would also find nontrivial components of the
curl in the xb or yb directions. If Curl(F~ ) • x
b or Curl(F~ ) • yb were nonzero at a point then that
suggests the vector field will spin a little paddle wheel with axis xb or yb. That is clear from simply
generalizing this calculation by replacing x, y with y, z or x, z. Another form of Green’s Theorem
follows from the curl: since dW = (∇ × F~ ) • zb dA we suspect that
I ZZ I ZZ
∂Q ∂P
WR = F~ • d~r = (∇ × F~ ) • zb dA ⇒ P dx + Qdy = − dA
∂R R ∂R R ∂x ∂y
This is the more common form found in calculus texts. Many texts simply state this formula and
offer part of the proof given in the next section. Our goal here was to understand why we would
expect such a theorem and as an added benefit we have hopefully arrived at a deeper understanding
of the differential vector calculus of curl and divergence.
1.5.3 proof of the theorem

It it is a simple exercise to show that the divergence form of Green’s theorem follows from the
curl-form we state below.
Theorem 1.5.3. Green’s Theorem for simply connected region:
Suppose ∂R is a piecewise-smooth, simple, closed CCW oriented curve which bounds the
simply connected region R ⊂ R 2 and suppose F~ is differentiable on an open set containing
R then I ZZ
∂Q ∂P
P dx + Qdy = − dA
∂R R ∂x ∂y
Proof: we begin by observing that any simply connected region can be subdivided into more ba-
sic simply connected regions which are simultaneously Type I and Type II subsets of the plane.
Sometimes, it takes a sequence of these basic regions to capture the overall region R and we will
return to this point once the theorem is settled for the basic case.
Proof of Green’s Theorem for regions which are both type I and II. We assume that
there exist constants a, b, c, d ∈ R and functions f1 , f2 , g1 , g2 which are differentiable and describe
R as follows:
R = {(x, y) | a ≤ x ≤ b, f1 (x) ≤ y ≤ f2 (x)} = {(x, y) | c ≤ y ≤ d, g1 (y) ≤ x ≤ g2 (y)}
| {z } | {z }
type I type II
Note the boundary ∂R = C1 ∪ C2 can be parametrized in the type I set-up as:

C1 : ~r1 (x) = hx, f1 (x)i, −C2 : ~r−2 (x) = hx, f2 (x)i
for a ≤ x ≤ b ( it is easier to think about parametrizing −C2 so I choose to do such). Proof of the
theorem can be split into proving two results:
I ZZ I ZZ
∂P ∂Q
(I.) P dx = − dA & (II.) Qdy = dA
∂R R ∂y ∂R R ∂x
I prove I. in these notes and I leave II. as a homework for the reader. Consider,
ZZ Z b Z f2 (x)
∂P ∂P
dA = dy dx
R ∂y a f1 (x) ∂y
Z b

= P (x, f2 (x)) − P (x, f1 (x)) dx
a
Z b Z b
= P (x, f2 (x))dx − P (x, f1 (x))dx
Z a Z a
= P dx − P dx
−C2
Z ZC1
=− P dx − P dx
C2 C1
Z
=− P dx
C
∂Q
Hence R − ∂P
RR H H RR
∂y dA = C P dx. You will show in homework that ∂R Qdy = R ∂x dA and Green’s
Theorem for regions which are both type I and II follows. ∇
If the set R is a simply connected subset of the plane then

Pit has no holes and generically the
picture is something like what follows. We intend that R = k Rk
Applying Green’s theorem to each sub-region gives us the following result.

XI X Z Z ∂Q ∂P
P dx + Qdy = − dA
k∂Rk k Rk ∂x ∂y
It is geometrically natural to suppose the rhs simply gives us the total double integral over R,
X Z Z ∂Q ∂P ZZ
∂Q ∂P

− dA = − dA.
k Rk ∂x ∂y R ∂x ∂y
I invite the reader to consider the diagram above to see that all the interior cross-cuts cancel and
only the net-boundary contributes to the line integral over ∂R. Hence,
XI I
P dx + Qdy = P dx + Qdy.
k ∂Rk ∂R
Green’s Theorem follows. It should be cautioned that the summations above need not be finite.
We neglect some analytical details in this argument. However, I hope the reader sees the big idea
here. You can find full details in some advanced calculus texts.
Theorem 1.5.4. Green’s Theorem for an annulus:
Suppose ∂R is a pair of simple, closed CCW oriented curve which bounds the connected
region R ⊂ R 2 where ∂R = Cin ∪ Cout and Cin is the CW-oriented inner-boundary of
R whereas Cout is the CCW oriented outer-boundary of R. Furthermore, suppose F~ is
differentiable on an open set containing R then
I ZZ
∂Q ∂P
P dx + Qdy = − dA
∂R R ∂x ∂y
Proof: See the picture below we can break the annulus into two simply connected regions then
apply Green’s Theorem for simply connected regions to each piece.
Observe that the cross-cuts cancel:

Z Z Z Z Z Z
F~ • d~r + F~ • d~r = F~ • d~r + F~ • d~r + F~ • d~r + F~ • d~r
CL CR Cup,out C3 Cup,in C5
Z Z Z Z
+ ~
F • d~r + ~
F • d~r + ~
F • d~r + F~ • d~r
Cdown,in C4 Cdown,out C6
Z Z Z Z
= F~ • d~r + F~ • d~r + F~ • d~r + F~ • d~r
Cup,out C3 Cup,in C5
Z Z Z Z
+ ~
F d~r −
• ~
F d~r +
• ~
F d~r −
• F~ • d~r
Cdown,in C3 Cdown,out C5
Z Z Z Z
= F~ • d~r + F~ • d~r + F~ • d~r + F~ • d~r
Cup,out Cup,in Cdown,in Cdown,out
Z Z
= F~ • d~r + F~ • d~r
Cup Cdown
Apply Green’s theorem to the regions bounded by CL and CR and the theorem follows.
Notice we can recast this theorem as follows:

I I ZZ
∂Q ∂P
P dx + Qdy − P dx + Qdy = − dA.
Cout −Cin R ∂x ∂y
Or, better yet, if C1 , C2 are two CCW oriented curves which bound R
I I ZZ
∂Q ∂P
P dx + Qdy − P dx + Qdy = − dA.
C1 C2 R ∂x ∂y
∂Q
Suppose the vector field F~ = hP, Qi passes our Clairaut Test on R then we have ∂x = ∂P
∂y and
consequently:
I I
C1 C2
I often refer to this result as the deformation theorem for irrotational vector fields in the plane.
Theorem 1.5.5. Deformation Theorem for irrotational vector field on the plane:
Suppose C1 , C2 are CCW oriented closed simple curves which bound R ⊂ R 2 and suppose
F~ = hP, Qi is differentiable on a open set containing R then
I I
C1 C2
In my view, points where ∇ × F~ 6= 0 are troublesome. This theorem says the line integral is
unchanged if we do not enclose any new troubling points as we deform C1 to C2 . On the flip-side
of this, if the integral around some loop is nonzero for a given vector field that must mean that
something interesting happens to the curl of the vector field on the interior of the loop.
Theorem 1.5.6. Green’s Theorem for a region with lots of holes.
Suppose R is a connected subset of R 2 which has boundary ∂R. We orient this boundary
curve such that the outer boundary has CCW orientation whereas all the inner-boundaries
have CW orientation. Furthermore, suppose F~ is differentiable on an open set containing
R then I ZZ
∂Q ∂P
P dx + Qdy = − dA
∂R R ∂x ∂y
Proof: follows from the picture below and a little thinking.

This is more interesting if we state it in terms of the outer loop and CCW oriented inner loops.
Denote Cout for the outside loop of ∂R and Ck for k = 1, 2, . . . , N for the inner CCW oriented
loops. Since ∂R = Cout ∪ −C1 ∪ · · · ∪ −CN it follows
I N I ZZ
X ∂Q ∂P
P dx + Qdy − P dx + Qdy = − dA
Cout Ck R ∂x ∂y
k=1
If we have ∂Q ∂P
∂x − ∂y = 0 throughout R we find the following beautiful generalization of the defor-
mation theorem:
I N I
X
Cout k=1 Ck
1.5.4 examples
Example 1.5.7. Use Green’s theorem to calculate C x3 dx + yxdy where C is the CCW boundary
H
of the oriented rectangle R: [0, 1] × [0, 1]. Identify that P = x3 and Q = xy. Applying Green’s
Theorem,
I ZZ Z 1Z 1
3 ∂Q ∂P 1
x dx + yxdy = − dA = y dx dy = .
C ∂x ∂y 0 0 2
One important application of Green’s theorem involves the calculation of areas. Note that if we
choose F~ = hP, Qi such that ∂Q ∂P
∂x − ∂y = 1 then the double integral in Green’s theorem represents
the area of R. In particular, it is common to use
F~ = h0, xi, or F~ = h−y, 0i, or F~ = h−y/2, x/2i
in Green’s theorem to obtain the identities:

I I I
1
AR = xdy = − ydx = xdy − ydx
∂R ∂R 2 ∂R
Example 1.5.8. Find the area of the ellipse bounded by x2 /a2 +y 2 /b2 = 1. Observe that the ellipse
∂R is parametrized by x = a cos(t) and y = b sin(t) hence dx = −a sin(t)dt and dy = b cos(t)dt
hence
1 2π 1 2π
I Z Z
1
AR = xdy − ydx = 1a cos(t)b cos(t)dt − b sin(t)(−a sin(t)dt) = ab dt = πab.
2 ∂R 2 0 2 0
When a = b = R we obtain the famous πR2 .
Example 1.5.9. You can show (perhaps you will in a homework) that for the line-segment L from
(x1 , y1 ) to (x2 , y2 ) we have the following excellent identity:
Z
1 1
xdy − ydx = (x1 y2 − x2 y1 ).
2 L 2
Consider that if P is a polygon with vertices (x1 , y1 ), (x2 , y2 ), . . . , (xN , yN ) with sides
L12 , L21 , . . . , LN −1,N then the area of P is given by the line-integral of F~ = h−y/2, x/2i thanks to
Green’s Theorem:
ZZ Z
1
AP = dA = xdy − ydx
P 2 ∂P
Z Z Z
1 1 1
= xdy − ydx + xdy − ydx + · · · + xdy − ydx
2 L12 2 L23 2 LN −1,N
1
= x1 y2 − x2 y1 + x2 y3 − x3 y2 + · · · + xN −1 yN − xN yN −1
2
You can calculate the area of polygon with vertices (0, 0), (−1, 1), (0, 2), (1, 3), (2, 1) is 9/2 by ap-
plying the formula above. You could just as well calculate the area of a polygon with 100 vertices.
The example below is a twist on the ellipse example already given. This time we study an annulus
with elliptical edges.
Example 1.5.10. Find the area bounded by ellipses x2 /a2 + y 2 /b2 = 1 and x2 /c2 + y 2 /d2 = 1 given
that 0 < c < a and 0 < d < b to insure that the ellipse x2 /a2 + y 2 /b2 = 1 is exterior to the ellipse
x2 /c2 + y 2 /d2 = 1. Observe that the ellpitical annulus has boundary ∂R = Cin ∪ Cout where Cout is
CCW parametrized by x = a cos(t) and y = b sin(t) and Cin is CW oriented with parametrization
x = c cos(t) and y = −d sin(t) it follows that:

I
1
AR = xdy − ydx
2 ∂R
I I
1 1
= xdy − ydx + xdy − ydx
2 Cout 2 Cin
1 2π 1 2π
Z Z
= ab dt − cd dt
2 0 2 0
= πab − πcd.
Notice the CW orientation is what caused us to subtract the inner area which is missing from the
annulus.
Example 1.5.11. Consider the CCW oriented curve C with parametrization x = (10+sin(30t)) cos(t)
and y = (10 + sin(30t)) sin(t) for 0 ≤ t ≤ 2π. This is a wiggly circle with mean radius 10. Calculate
xdy − ydx
Z
2 2
.
C x +y
Let P = −y/(x2 + y 2 ) and Q = x/(x2 + y 2 ) you can show that ∂x Q − ∂y P = 0 for (x, y) 6= (0, 0).
It follows that we can deform the given problem to the simpler task of calculating the line-integral
around the unit circle S1 : x = cos(t) and y = sin(t) hence xdy − ydx = cos2 (t)dt + sin2 (t)dt and
x2 + y 2 = 1 on S1 , calculate,
Z 2π
xdy − ydx xdy − ydx
Z Z
dt
2 2
= 2 2
= = 2π.
C x +y S1 x + y 0 1
Notice that we could still make this calculation if the specific parametrization of C was not given.
Also, generally, when faced with this sort of problem we should try to pick a deformation which
makes the integration easier. It was wise to deform to a circle here since the denominator was
greatly simplified.
1.5.5 conservative vector fields and green’s theorem

Recall that Proposition 1.4.5 gave us a list of ways of thinking about a conservative vector field:
Suppose U is an open connected subset of Rn then the following are equivalent

H
4. (add precondition U be simply connected) ∇ × F~ = 0 on U .

Proof: We finish the proof of by addressing why (4.) ⇒ (1.) in light of Green’s Theorem. Suppose
U is simply connected and ∇ × F~ = 0 on U . Let C1 be a closed loop in U and let CR be another
loop of radius R inside C1 . Since ∇ × F~ = 0 it follows we can Rapply the deformation Theorem
1.5.5 on the annulus between C1 and CR to obtain C1 F~ • d~r = CR F~ • d~r. Now, as U is simply
R
connected we can smoothly deform CR to a point C0 . You can show that

Z
lim F~ • d~r = 0
R→0 CR
since CR becomes a point in this limit. Not convinced? Consider that the integral CR F~ • d~r is at
R
most the product of max{||F~ (~r)|| | ~r ∈ CR } and the total arclength of CR . However, the magnitude
is bounded as F~ has continuous component functions and the arclength of CR clearly goes to zero
as R → 0. Perhaps the picture below helps communicate the idea of the proof:
We find (4.) ⇒ (3.) hence, by our earlier work, (3.) ⇒ (2.) ⇒ (1.).
Now, in terms of logical minimalism, to prove that 1, 2, 3, 4 are equivalent we could just prove the
string of implications (1.) ⇒ (2.) ⇒ (3.) ⇒ (4.) ⇒ (1.) then any of the reverse implications are
easily found by logic. For example, (3.) ⇒ (2.) would follow from (3.) ⇒ (4.) ⇒ (1.) ⇒ (2.). That
said, I tried to give all directions in the proof to better illustrate how the different views of the
conservative vector field are connected.
1.5.6 two-dimensional electrostatics

The fundamental equation of electrostatics is known as Gauss’ Law. In three dimensions it simply
states that the flux through a closed surface is proportional to the charge which is enclosed. We
have yet to define flux through a surface, but we do have a careful definition of flux through a
simple closed curve. If there was a Gauss’ Law in two dimensions then it ought to state that
ΦE = Qenc
In particular, if we denote σ = dQ/dA and have in mind the region R with boundary ∂R,
I ZZ
~ •n
(E b)ds = σdA
∂R R
Suppose we have an isolated charge Q at the origin and we apply Gauss law to a circle of radius r
centered at the origin then we can argue by symmetry the electric field must be entirely radial in
direction and have a magnitude which depends only on r. It follows that:
I ZZ
~ •n
(E b)ds = σdA ⇒ (2πr)E = Q
∂R R
Hence, the coulomb field in two dimensions is as follows:
~ θ) = Q rb
E(r,
2πr
Let us calculate the flux of the Coulomb field through a circle C of radius R:
I Z
~ •n Q
(E b)ds = rb rb ds
• (1.3)
C 2πr
ZC
Q
= ds
C 2πR
Z
Q
= ds
2πR C
Q
= (2πR)
2πR
= Q.
The circle is complete. In other words, the Coulomb field derived from Gauss’ Law does in fact
satisfy Gauss Law in the plane. This is good news. Let’s examine the divergence of this field. It
appears to point away from the origin and as you get very close to the origin the magnitude of E
~
is unbounded. It will be convenient to reframe this formula for the Coulomb field by E(x, y) =
Q
2 2
2π(x +y )
hx, yi. Note:
Q x2 + y 2 − 2x2 x2 + y 2 − 2y 2

~ ∂ xQ ∂ yQ
∇•E = + = + = 0.
∂x 2π(x2 + y 2 ) ∂y 2π(x2 + y 2 ) 2π (x2 + y 2 )2 (x2 + y 2 )2
If we were to carelessly apply the divergence form of Green’s theorem this could be quite unsettling:
consider,
I ZZ ZZ
~
E ~n =
• • ~
∇ E dA ⇒ Q= (0)dA = 0.
∂R R R
But, Q need not be zero hence there is some contradiction? Why is there no contradiction? Can
you resolve this paradox?
Moving on, suppose we have N charges placed at source points ~r1 , ~r2 , . . . , ~rN then we can find the
total electric field by the principle of superposition.
We simply take the vector sum of all the coulomb fields. In particular,
N n
~ r) =
X
~j =
X Qj ~r − ~rj
E(~ E
2π ||~r − ~rj ||2
j=1 j=1
What is the flux through a circle which encloses just the k-th one of these charges? Suppose CR is
a circle of radius R centered at ~rk . We can calculate that
I
(E~k • n
b)ds = Qk
CR
~ j is differentiable inside all of CR for j 6= k and ∇ • E

whereas, since E ~ j = 0 we can apply the
divergence form of Green’s theorem to deduce that
I
~j • n
(E b)ds = 0.
CR
~ =E
Therefore, summing these results together we derive for E ~1 + · · · + E
~k + · · · + E
~ N that
I
~ •n
(E b)ds = Qk
CR
Notice there was nothing particularly special about Qk so we have derived this result for each
charge in the distribution. If we take a circle around a charge which contains just one charge then
Gauss’ Law applies and the flux is simply the charge enclosed. Denote C1 , C2 , . . . , CN as little
circles which enclose the charges Q1 , Q2 , . . . , QN respective. We have
I I I
Q1 = ~
(E • n
b)ds, Q2 = ~
(E • n
b)ds, . . . , QN = ~ •n
(E b)ds
C1 C2 CN
Now suppose we have a curve C which encloses all N of the charges. The electric field is differ-
entiable and has vanishing divergence at all points except the location of the charges. In fact,
the coulomb field passes Clairaut’s test everywhere. It just has the isolated singularity where the
charge is found. We can apply the general form of the deformation theorem to arrive at Gauss’
Law for the distribution of N -charges:
I I I I
~
(E • n
b)ds = ~
(E • n
b)ds + ~ b)ds + · · · +
(E • n ~ •n
(E b)ds = Q1 + Q2 + · · · + QN
C C1 C2 CN
You can calculate the divergence is zero everywhere except at the location of the source charges.
Morale of story: even one point thrown out of a domain can have dramatic and global consequences
for the behaviour of a vector field. In physics literature you might find the formula to describe
what we found by a dirac-delta function these distributions capture certain infinities and let you
work with them. For example: for the basic coulomb field with a single point charge at the origin
~ θ) = Q rb this derived from a charge density function σ which is zero everywhere except at
E(r, 2πr RR
the origin. Somehow R σdA = Q for any region R which contains (0, 0). Define σ(~r) = Qδ(~r).
Where we define: for any function f which is continuous near 0 and any region which contains the
origin Z
f (~r)δ(~r)dA = f (0)
R
RR
and if R does not contain (0, 0) then R f (~r)δ(~r)dA = 0. The dirac delta function turns integration
into evaluation. The dirac delta function is not technically a function, in some sense it is zero at
all points and infinite at the origin. However, we insist it is manageably infinity in the way just
described. Notice that it does at least capture the right idea for density of a point charge: suppose
R contains (0, 0), ZZ ZZ
σdA = Qδ(~r)dA = Q.
R R
On the other hand, we can better understand the divegence calculation by the following calcula-
tions16 :
~r
∇ • 2 = 2πδ(~r).
r
~ = Q ~r • ~
Q ~r Q ~
r
Consequentlly, if E 2π r2 then ∇ E = ∇ 2π r2 = 2π ∇ r2 = Qδ(~ r). Now once more apply
• •
Green’s theorem to the Coulomb field. Use the divergence form of the theorem and this time
appreciate that the divergence of E ~ is not strictly zero, rather, the dirac-delta function captures
the divergence: recall the RHS of this calculation followed from direct calculation of the flux of the
Coloumb field through the circle ∂R,
I ZZ ZZ
~
E • ~n ds = ~
∇ • E dA ⇒ Q= Qδ(~r)dA = Q.
∂R R R
All is well. This is the way to extend Green’s theorem for coulomb fields. You might wonder about
other types of singularities. Are there similar techniques? Probably, but that is beyond these notes.
I merely wish to sketch the way we think about these issues in electrostatics. In truth, this section
is a bit of a novelty. What really matters is three-dimensional coulomb fields whose magnitude
16
I don’t intend to explain where this 2π comes from, except to tell you that it must be there in order for the
extension of Green’s theorem to work out nicely.
1.6. SURFACE INTEGRALS 39
depends on the squared-reciprocal of the distance from source charge to field point. Perhaps I will
write an analogus section once we have developed the concepts of flux and the three-dimensional
divergence theorem.
1.6 surface integrals

We discuss how to integrate over surfaces in R3 . This section is the natural extension of the work
we have already accomplished with curves. We begin by describing the integral with respect to
the surface area, naturally this integral calculates surface area, but more generally it allows us to
continuously sum some density over a given surface. Next we discuss how to find the flux through a
surface. The concept of flux requires we give the surface a direction. The flux of a vector field is the
number17 of field lines which cut through the surface. The parametric viewpoint is primary in this
section, but we also make an effort to show how to calculate surface integrals from the graphical
or level-surface viewpoint.
1.6.1 motivations for surface area integral and flux

Let us consider a surface S parametrized by smooth ~r(u, v) = hx(u, x), y(u, v), z(u, v)i for (u, v) ∈ Ω.
If we wish to approximate the S surface area of S then we should partition the parameter space Ω
m Sm
into subregions Ωij such that i=1 j=1 Ωij = Ω where we assume that Ωij are mostly disjoint,
they Smight share an edge or point, but not an area. Naturally this partitions S into subsurfaces;
S= m
Sm
i=1 j=1 Sij where Sij = ~ r(Ωij ). Next, we replace each subsurface Sij with its tangent plane
18 ∗ ∗
based at some point ~r(ui , vj ) ∈ Sij . For convenience of this motivation we may assume that the
partition is made so that Ωij is a little rectangle which is 4u by 4v. The length of the coordinate
curves are well-approximated by ∂u ∂~r ∂~
4u and ∂vr
4v based at the point ~r(u∗i , vj∗ ). The coordinate lines
on S are not necessarily perpendicular so the area is not simply the product of length times width,
in fact, we have a little paralellogram to consider. It follows that the area Aij of the i, j-tangent
plane is given by
∂~r ∂~r ∗ ∗

Aij =
× (ui , vj )4u4v.
∂u ∂v
A good approximation to surface area is given for m, n >> 1 by
m X
n
X ∂~r ∂~r ∗ ∗

∂u × ∂v (ui , vj )4u4v

i=1 j=1
As we pass to the limit m → ∞ and n → ∞ the double finite sum becomes the double integral
taken over the parameter space Ω. For these reasons we define:
17
relative to some convention, lines drawn per unit of flux
18
when I illustrate this idea I usually take these points to the lower left of the partition region, but in principle you
could sample elsewhere.
Definition 1.6.1. scalar surface integral.
Suppose S is a surface in R3 paramterized by ~r with domain Ω and with parameters u, v.

Furthermore, suppose f is a continuous function on some open set containing S then we
define the scalar surface integral of f over S by the following integral (when it exists)
ZZ ZZ
∂~r ∂~r
f dS = f (~r(u, v))
× du dv.
S Ω ∂u ∂v
In the case we integrate f = 1 over S then we obtain the surface area RR of S. On the other hand,
if f was the mass-density f = dMdS then dM = f dS and the integral S f dS calculates the total
mass of S. Clearly it is convenient to think of dS as something on its own, however, it should be
remembered that this is just notation to package the careful definition given above,

∂~r ∂~r
dS = × du dv infinitesimal scalar surface area element
∂u ∂v
~ = ∂~r × ∂~r
Moreover, we should recall that the normal vector field to S induced by ~r was given by N ∂u ∂v
hence we can write dS = N dudv. This is nice, but for most examples it does not save us from
explicit calculation of the cross-product.
Next, consider a vector field F~ defined on some open set containing S. Suppose that S is a
regular surface and as such has a well-defined normal vector field N ~ . If we define n ~ then
b = N1 N
F~ • n
b gives the component of F~ which points in the normal direction of the surface. It is customary
to draw field-lines to illustrate both the direction and magnitude of a vector field. The number of
lines crossing a particular surface illustrates the magnitude of the vector field relative to the given
area. For example, if we had an area A1 which had 4 field lines of F~ and another area A2 which had
8 field lines of F~ then the magnitude of F~ on these areas is proportional to A41 and A82 respectively.
If the vector fields are constant, then the flux through A1 is F1 A1 whereas the flux through A2 is
F2 A2 . Generally, the flux of a vector field through a surface depends both on the size of the surface
and the magnitude of the vector field in the normal direction of the surface. This is the natural
generalization of the flux-integral we discussed previously for curves in the plane.
Definition 1.6.2. surface integral of vector field; the flux through a surface.
Suppose S is an oriented surface in R3 paramterized by ~r with domain Ω and with param-

b. Furthermore, suppose F~ is a continuous
eters u, v which induce unit-normal vector field n
vector field on some open set containing S then we define the surface integral of F~ over
S by the following integral (when it exists)
ZZ ZZ
~ ~ F~ • n

F • dS = b dS
S S
In practice, the formula we utilize for direct computation is not the one given above. Let us
calculate, ZZ ZZ ZZ
~ ~ 1 ~
F~ • N
~ du dv.

F •n
b dS = F • N N du dv =
S Ω N Ω
~ (u, v) = ∂u~r × ∂v ~r we find
Hence, recalling once more that N
ZZ ZZ
~ ~ ~ ∂~r ∂~r
F dS =
• F (~r(u, v)) • × du dv.
S Ω ∂u ∂v
The equation boxed above is how we typically calculate flux of F~ through S. I should mention that
~ separately before computation of the integral, this quantity is
is often convenient to calculate dS
called the infinitesimal vector surface area element and is defined by
~= ∂~r ∂~r ~
dS × du dv = n
bdS vector area element dS
∂u ∂v
Once more, dS is the scalar area element. Both of these are only meaningful when viewed in con-
nection with the parametric set-up described in this section.
The uninterested reader may skip to the examples, however, there is some unfinished business
theoretically here. We should demonstrate that the definitions given in this section are independent
of the parametrization. If this fails to be true then the concepts of surface area, total mass etc...
and flux are in doubt. We must consider a reparametrization of S by X ~ : D → S where a, b are
the typical parameters in D and the normal vector field induced by X is N~ ~ X (a, b) = ∂a X ~
~ × ∂b X.
Let us, in contrast denote N~ r (u, v) = ∂u~r × ∂v ~r. Since each point on S is covered smoothly by
~
both ~r = hx, y, zi and X = hX1 , X2 , X3 i there exist functions which transition between the two
parametrizations. In particular, we can find T~ : Ω → D such that T~ = hh, gi and
~
~r(u, v) = X(h(u, v), g(u, v))
We need to sort through the partial derivatives so we can understand how the normal vector fields
~ X and N
N ~ r are related, let a = h(u) and b = g(v) hence ~r(u, v) = X(a, ~ b). I’ll expand X ~ into its
component function notation to make sure we understand what we’re doing here:

∂~r ∂ ∂ ∂ ∂
= X1 (a, b), X2 (a, b), X3 (a, b) = [X1 (a, b)], [X2 (a, b)], [X3 (a, b)]
∂u ∂u ∂u ∂u ∂u
We calculate by the chain-rule, (omitting the (a, b) dependence on the lhs, technically we should
∂
write ∂u [X1 (a, b)] etc...)
∂X1 ∂X1 ∂h ∂X1 ∂g ∂X2 ∂X2 ∂h ∂X2 ∂g ∂X3 ∂X3 ∂h ∂X3 ∂g
= + , = + , = +
∂u ∂a ∂u ∂b ∂u ∂u ∂a ∂u ∂b ∂u ∂u ∂a ∂u ∂b ∂u
Likewise, for the derivative with respect to v we calculate,
∂X1 ∂X1 ∂h ∂X1 ∂g ∂X2 ∂X2 ∂h ∂X2 ∂g ∂X3 ∂X3 ∂h ∂X3 ∂g
= + , = + , = +
∂v ∂a ∂v ∂b ∂v ∂v ∂a ∂v ∂b ∂v ∂v ∂a ∂v ∂b ∂v
We find,

∂~r ∂X1 ∂h ∂X1 ∂g ∂X2 ∂h ∂X2 ∂g ∂X3 ∂h ∂X3 ∂g
= + , + , +
∂u ∂a ∂u ∂b ∂u ∂a ∂u ∂b ∂u ∂a ∂u ∂b ∂u

∂X1 ∂X2 ∂X3 ∂h ∂X1 ∂X2 ∂X3 ∂g
= , , + , ,
∂a ∂a ∂a ∂u ∂b ∂b ∂b ∂u
∂X~ ∂h ∂ X ~ ∂g
= +
∂a ∂u ∂b ∂u
Similarly,
∂~r ~ ∂h ∂ X
∂X ~ ∂g
= +
∂v ∂a ∂v ∂b ∂v
Calculate, by the antisymmetry of the cross-product,
~ ~ ∂g ∂ X ~ ∂h ∂ X ~ ∂g
∂~r ∂~r ∂ X ∂h ∂ X
× = + × +
∂u ∂v ∂a ∂u ∂b ∂u ∂a ∂v ∂b ∂v
~ ~ ∂h ∂g
~ ~ ∂g ∂h

∂X ∂X ∂X ∂X
= × + ×
∂a ∂b ∂u ∂v ∂b ∂a ∂u ∂v
~ ~ ∂h ∂g

∂X ∂X ∂g ∂h
= × −
∂a ∂b ∂u ∂v ∂u ∂v
~ X (h(u, v), g(u, v)) ∂(h,g) hence Nr (u, v) = NX (h(u, v), g(u, v)) ∂(h,g) .
~ r (u, v) = N

It follows that N ∂(u,v) ∂(u,v)
The vertical bars denote absolute value; when we pull a scalar out of a magnitude of a vector it
gets absolute value bars; ||c~v || = |c| ||~v ||. Consider the surface area integral of f over S which is
parametrized by both ~r(u, v) and X(a,~ b) as discussed above. Observe,
ZZ ZZ
f dS = f (~r(u, v))Nr (u, v)du dv
S
Z ZΩ
∂(h, g)
= ~
f (X(h(u, v), g(u, v)))NX (h(u, v), g(u, v))
du dv
Ω ∂(u, v)
ZZ
= ~
f (X(a, b))NX (a, b)da db.
D
In the last line I applied the multivariate change of variables theorem. Notice that the absolute
value bars are important to the calculation. We will see in the corresponding calculation for flux the
absolute value bars are absent, but this is tied to the orientation-dependence of the flux integral.
In the scalar surface integral the direction (outward or inward) of the normal vector field does not
figure into the calculation. We have shown
ZZ ZZ
f dS = f dS .
S −S
Next, turn to the reparametrization invariance of the flux integral. Suppose once more ~r and X ~
both parametrize S. Calculate the flux of a continuous vector field F~ defined on some open set
containing S (via the boxed equation following the definition)
ZZ ZZ
~ ~
F dS =
• ~ r (u, v) du dv
F~ (~r(u, v)) • N
S Ω
ZZ
= F~ (X(h(u,
~ v), g(u, v))) • N~ X (h(u, v), g(u, v)) ∂(h, g) du dv
Ω ∂(u, v)
∂(h,g) ~ r and N ~ X point on the same side of S
If ∂(u,v) > 0 for all (u, v) ∈ Ω then it follows that N
and we say they are consistently oriented parametrizations of S. Clearly if we wish for the
flux to be meaningful we must choose a side for S and insist that we use a consistently oriented
parametrization to calculate the flux. If ~r and X ~ are consistently oriented then ∂(h,g) > 0 for all
∂(u,v)
∂(h,g) ∂(h,g)
(u, v) ∈ Ω and hence ∂(u,v) = ∂(u,v) for all (u, v) ∈ Ω and we find
ZZ ZZ
∂(h, g)
F~ • dS
~= F~ (X(h(u,
~ ~ X (h(u, v), g(u, v))
v), g(u, v))) • N ∂(u, v) du dv

S Ω
ZZ
= F~ (X(a,
~ ~ X (a, b) da db
b) • N
D
by the change of variable theorem for double integrals. On the other hand, if ~r is oppositely oriented
from X~ then we say ~r parametrizes S whereas X ~ parametrizes −S. In the case X ~ points in the

∂(h,g) ∂(h,g)
direction opposite ~r we find the coefficient ∂(u,v) = − ∂(u,v) and it follows that:
ZZ ZZ
F~ • dS
~=− F~ • dS
~
−S S
1.6.2 standard surface examples

~ and dS for the sphere, cylinder, cone, plane and arbitrary graph.
In this section we derive dS
You can add examples past these, but these are essential. I also derive the surface area where
appropriate.
1. SR the sphere of radius R centered at the origin. We have spherical equation ρ = R which
suggests the natural parametric formulas: for 0 ≤ θ ≤ 2π and 0 ≤ φ ≤ π
x = R cos(θ) sin(φ), y = R sin(θ) sin(φ), z = R cos(φ)

~r(φ, θ) = R cos(θ) sin(φ), R sin(θ) sin(φ), R cos(φ)
We can limit the parameter space [0, π] × [0, 2π] to select subsets of the sphere if need arises.
The normal vector field is calculated from partial derivatives of ~r(φ, θ);

∂~r
= R cos(θ) cos(φ), R sin(θ) cos(φ), −R sin(φ)
∂φ

∂~r
= − R sin(θ) sin(φ), R cos(θ) sin(φ), 0
∂θ
In invite the reader to calculate the cross-product above and derive that

∂~r ∂~r
× = R sin(φ) cos(θ) sin(φ), sin(θ) sin(φ), cos(φ) = R2 sin(φ)b
2
ρ.
∂φ ∂θ
~ (φ, θ) = R2 sin(φ)b
We will find is useful to note that N ρ for SR . This is an outward pointing
normal vector field. To summarize, for the sphere SR with outward orientation we find
~ = R2 sin(φ)dφdθb
dS ρ and dS = R2 sin(φ)dφdθ
The surface area of the sphere SR is given by

Z 2π Z π
Area(SR ) = R2 sin(φ)dφdθ = 4πR2 .
0 0
It is interesting to note that d/dR( 34 πR3 ) = 4πR2 just like d/dR(πR2 ) = 2πR.
2. Right circular cylinder of radius R with axis along z. In cylindrical coordinates we have
the simple formulation r = R which gives the natural parametrization:
x = R cos(θ), y = R sin(θ), z=z

~r(θ, z) = R cos(θ), R sin(θ), z
∂~
r ∂~
r
for 0 ≤ θ ≤ 2π and z ∈ R. Calculate ∂θ = h−R sin(θ), R cos(θ), 0i = Rθb and ∂z = zb thus
∂~r ∂~r
× = Rθb × zb = Rb
r
∂θ ∂z
~ (θ, z) = Rb
Consequently, N r = Rhcos(θ), sin(θ), 0i and we find
~ = Rdθ dz rb
dS and dS = Rdθdz
The surface area of the cylinder for 0 ≤ z ≤ L is given by

Z 2π Z L
Area = Rdz dθ = 2πRL.
0 0
Of course, the whole cylinder with unbounded z has infinite surface area.
3. Cone at angle φo . In cylindrical coordinates r = ρ sin(φo ) thus the cartesian equation of this
cone is easily derived from r2 = ρ2 sin2 (φo ) gives x2 + y 2 = sin2 (φo )(x2 + y 2 + z 2 ) hence, for
φo 6= π/2, we find x2 + y 2 = tan2 (φo )z 2 . In cylindrical coordinates this cone has equation
r = tan(φo )z. From spherical coordinates we find a natural parametrization,
x = ρ cos(θ) sin(φo ), y = ρ sin(θ) sin(φo ), z = ρ cos(φo )

For convenience denote a = sin(φo ) and b = cos(φo ) thus

~r(θ, ρ) = aρ cos(θ), aρ sin(θ), bρ
for 0 ≤ θ ≤ 2π and ρ ∈ [0, ∞). Differentiate to see that

∂~r
∂~r

= − aρ sin(θ), aρ cos(θ), 0 & = a cos(θ), a sin(θ), b .
∂θ ∂ρ
Calculate,
∂~r ∂~r
= habρ cos(θ), abρ sin(θ), −a2 ρi = aρ b cos(θ), b sin(θ), −a

×
∂θ ∂ρ

Note that φb = cos(θ) cos(φ), cos(θ) cos(φ), − sin(φ) and a = sin(φo ) and b = cos(φo ) hence
∂~r ∂~r
× = ρ sin(φo )φb
∂θ ∂ρ
~ (θ, ρ) = ρ sin(φo )φ.
Consequently, N b We find for the cone φ = φo ,
~ = ρ sin(φo )dθ dρφb

dS and dS = ρ sin(φo )dθ dρ.

where φb = cos(θ) cos(φo ), cos(θ) cos(φo ), − sin(φo ) . The surface area of the cone φ = φo for
0 ≤ ρ ≤ R is given by
Z R Z 2π
Area = ρ sin(φo )dθ dρ = sin(φo )πR2 .
0 0
Of course, the whole cone with unbounded ρ has infinite surface area. On the other hand,
the result above is quite reasonable in the case φo = π/2.
~ and B
4. Plane containing vectors A ~ and base-point ~ro . We parametrize by
~ + vB
~r(u, v) = ~ro + uA ~
~ and ∂v ~r = B
for (u, v) ∈ R n×n . Clearly ∂u~r = A ~ hence N
~ (u, v) = A
~ × B.
~ The plane has a
constant normal vector field. We find:
~=A
dS ~ × Bdu
~ dv and ~ × B||du
dS = ||A ~ dv.
If we select a compact subset Ω of the plane then ~r(Ω) has surface area
ZZ ZZ
Area = ~ ~
||A × B||du dv = ~ × B||
||A ~ dA.
Ω Ω
In the last equation I mean to emphasize that the problem reduces to an ordinary double in-
tegral of a constant over the parameter space Ω. Usually there is some parameter dependence
in dS, but the plane is a very special case.
5. Graph z = f (x, y). Naturally we take parameters x, y and form

~r(x, y) = x, y, f (x, y)
this is the Monge patch on the surface formed by the graph. Differentiate,

∂x~r = 1, 0, ∂x f & ∂y ~r = 0, 1, ∂y f
Calculate the normal vector field,
~ (x, y) = 1, 0, ∂x f × 0, 1, ∂y f = − ∂x f, −∂y f, 1

N
We find:
q
~ = − ∂x f, −∂y f, 1 dx dy

dS and dS = 1 + (∂x f )2 + (∂y f )2 dx dy.
If Ω is a compact subset of dom(f ) then we can calculate the surface area by

ZZ q
Area = 1 + (∂x f )2 + (∂y f )2 dx dy.
Ω
1.6.3 scalar surface integrals

I examine several examples to further illustrate the construction of the scalar surface integral. We
use the surface area elements developed in the previous section.
Example 1.6.3. Calculate the total mass of a sphere of radius R which has mass-density
σ(x, y, z) = tan−1 (y/x). Identify that σ = dM/dS hence dM = σdS. Moreover, in spherical
coordinates σ = θ hence thus the integral below gives the total mass:
ZZ Z 2π Z π π 2 2π
2 2 θ
= 4π 2 R2 .

M= σ dS = R θ sin(φ)dφdθ = R − cos(φ)
SR 0 0 0 2
0
Example 1.6.4. Calculate the average of f (x, y, z) = z 2 over the circular cylinder S: x2 + y 2 = R2
for 0 ≤ z ≤ L (assume the caps are open, just find the average over the curved side). By logic,
L Z 2π
L2
ZZ Z
1 2 1 1
favg = z dS = Rz 2 dθ dz = (2πR)(L3 /3) = .
2πRL S 2πRL 0 0 2πRL 3
Generally if we wish to calculate the average

RR of a function over a surface of finite total surface area
we define favg to be the value such that S f dS = favg Area(s).
Example 1.6.5. Find the centroid of the cone φ = π/4 for 0 ≤ ρ ≤ R. The centroid is the
geometric center of the object with regard to the density. In other words, calculate the center of
mass under the assumption dM/dS = 1. However you like to think of it, the centroid (x̄, ȳ, z̄) is
given by
ZZ ZZ ZZ
1 1 1
x̄ = xdS, ȳ = y dS, z̄ = z dS
Area(S) S Area(S) S Area(S) S
For the cone S it is clear by symmetry that x̄ = ȳ = 0. Once

√ more building off (3.) of √ the
2 2
previous section we calculate: Area(S) = sin(φo )πR = πR / 2 hence as z = ρ cos(φo ) = ρ/ 2
and dS = √ρ2 dθ dρ
√ R Z 2π
√ √
2 R3 2π
Z
2 ρ ρ R 2
z̄ = √ √ dθ dρ = = .
πR2 0 0 2 2 πR2 3 2 3
We can also calculate the moment of intertia about the z-axis for the cone RR S (assume constant
mass density 1 for this example) . The moment of intertia is defined by Iz = S r2 dM and as the
equation of this cone is simply r = z we find r2 = ρ2 cos2 (π/4) = ρ2 /2 thus
R Z 2π
ρ2 ρ R4 (2π) πR4
ZZ Z
2
Iz = r δ dS = √ dθ dρ = √ = √ .
S 0 0 2 2 8 2 4 2
Example 1.6.6. Find the scalar surface integral of f (x, y, z) = xyz on the graph S: z = 6 + x + y
for 0 ≤ y ≤ x2 and 0 ≤ x ≤ 1 (this is just a portion of the total
q graph z = 6 + x + y√which is an
unbounded plane). Observe that fx = 1 and fy = 1 thus dS = 1 + fx2 + fy2 dx dy = 3 dx dy
ZZ Z 1 Z x2 √
xyz dS = xy(6 + x + y) 3 dy dx
S 0 0
√ Z 1 Z x2
6xy + 6x2 y + 6xy 2

= 3 dy dx
0 0
2
√ Z 1 x
2 2 2 3
= 3 3xy + 3x y + 2xy dx
0 0
x2
√ Z 1
3x5 + 3x6 + 2x7 dx

= 3
0 0
√

3 3 2
= 3 + +
6 7 8
√
33 3
= .
28
1.6.4 flux integrals

Once more I build off the examples from Section 1.6.2.
Example 1.6.7. Calculate the flux of F~ = h1, 2, 3i through the part of the sphere x2 + y 2 + z 2 = 4
which is above the xy-plane. Recall dS ~ = R2 sin(φ)dθ dφb ρ and note for z ≥ 0 we need no restriction
of the polar angle θ (0 ≤ θ ≤ 2π) however the azmuthial angle φ falls into the interval 0 ≤ φ ≤ π/2.
Thus, as R = 2 for this example,
ZZ Z π/2 Z 2π
~ ~

Φ= F dS =
• h1, 2, 3i • cos(θ) sin(φ), sin(θ) sin(φ), cos(φ) 4 sin(φ)dθ dφ
S 0 0
Z π/2 Z 2π
2 2
=4 cos(θ) sin (φ) + 2 sin(θ) sin (φ) + 3 cos(φ) sin(φ) dθ dφ
0 0
Z π/2 Z 2π
=6 sin(2φ) dθ dφ
0 0
π/2
−1
= 12π cos(2φ)
2 0
= 12π.
Example 1.6.8. Let n ∈ Z and calculate the flux of F~ (x, y, z) = (x2 + y 2 + z 2 )n/2−1 hx, y, zi through
the sphere SR . Observe that F~ = ρn ρb and recall

~r(φ, θ) = R cos(θ) sin(φ), R sin(θ) sin(φ), R cos(φ) = Rb ρ
Thus calculate, F~ (~r(φ, θ)) = Rn ρb

ZZ Z π Z 2π
~ ~ Rn ρb • ρb R2 sin(φ)dθ dφ

Φ= F • dS =
S 0 0
Z π Z 2π
n+2
=R sin(φ)dθ dφ
0 0
n+2
= 4πR .
Believe it!. Notice that the case n = −2 is very special. In that case the flux is independent of the
radius of the sphere. The flux spreads out evenly and is neither created nor destroyed for n = −2.
Example 1.6.9. Calculate the flux of F~ = hx2 , z, yi through the closed cylinder x2 + y 2 = R2 with
0 ≤ z ≤ L. Notice that S = S1 ∪ S2 ∪ S3 where I mean to denote the top by S1 , the base by S2 and
the side by S3 . The parametrizations and normal vectors to these faces are naturally given by
~ 1 (r, θ) = hr cos(θ), r sin(θ), Li
X ~ 1 (r, θ) = ∂r X
N ~ 1 × ∂θ X
~ 1 = rb × rθb = r zb
~ 2 (θ, r) = hr cos(θ), r sin(θ), Li

X ~ 2 (θ, r) = ∂θ X
N ~ 2 × ∂r X
~ 2 = rθb × rb = −r zb
~ 3 (θ, z) = hR cos(θ), R sin(θ), zi
X ~ 3 (θ, z) = ∂θ X
N ~ 3 × ∂z X
~ 3 = Rθb × zb = Rb
r
where 0 ≤ r ≤ R, 0 ≤ θ ≤ 2π and 0 ≤ z ≤ L. I’ll calculate the flux through each face separately.
Begin with S1 :
F~ (X
~ 1 (r, θ)) = F~ (r cos(θ), r sin(θ), L) = r2 cos2 (θ), L, r sin(θ)

~1 = rdr dθ zb and we find

note that dS
ZZ Z R Z 2π Z R Z 2π
F~ • dS
~= 2 2
r2 sin(θ)dθ dr = 0.

ΦS1 = hr cos (θ), L, r sin(θ)i • zb rdθ dr =
S1 0 0 0 0
Through a similar calculation we find ΦS2 = 0. To calculate the flux through S3 we should evaluate
the vector field on the parametrization,
F~ (X
~ 3 (r, θ)) = F~ (R cos(θ), R sin(θ), z) = R2 cos2 (θ), z, R sin(θ)

~ = Rhcos(θ), sin(θ), 0idθ dz. Thus,

also recall that rb = hcos(θ), sin(θ), 0i thus dS
Z 2π Z L
R2 cos2 (θ), z, R sin(θ) • R cos(θ), sin(θ), 0 dθ dz

ΦS3 =
0 0
Z 2π Z L
3 3
= R cos (θ) + zR sin(θ) dθ dz
0 0
Z 2π
3
= LR cos3 (θ) dθ
0
Z 2π
3
1 − sin2 (θ) cos(θ) dθ

= LR
0
Z sin(2π)
= LR3 1 − u2 du

sin(0)
= 0.
Example 1.6.10. Find the flux of F~ = C~ on the subset of the plane ~r(u, v) = ~ro + uA
~ + vB~ defined
by 1 ≤ u2 + v 2 ≤ 4. Denote Ω = dom(~r). You can calculate dS~=A ~×B ~ du dv hence
ZZ ZZ ZZ
F~ • dS
~= ~• A ~×B ~ du dv = C ~• A~×B ~ ~• A ~×B ~ .

C du dv = 15π C
S Ω Ω
Since the area of Ω is clearly π(4)2 − π(1)2 = 15π.
Finally, we conclude by developing a standard formula which is the focus of flux calculations in
texts such as Stewart’s.
Example 1.6.11. Find the flux of F~ = hP, Q, Ri through the upwards oriented graph S: z = f (x, y)
with domain Ω. We derived that dS~ = h−∂x f, −∂y f, 1idx dy relative to the Monge parametrization
~r(x, y) = hx, y, f (x, y)i for (x, y) ∈ Ω. We calculate, from the definition,
ZZ ZZ
• ~
hP, Q, Ri dS = hP, Q, Ri • h−∂x f, −∂y f, 1idx dy
S Ω
ZZ
∂f ∂f
= −P −Q + R dx dy
Ω ∂x ∂y
which is technically incorrect, we really mean the following:

ZZ ZZ
~ ∂f ∂f
hP, Q, Ri dS =
• −P (x, y, f (x, y)) − Q(x, y, f (x, y)) + R(x, y, f (x, y)) dA
S Ω ∂x ∂y
2 2
For example, to calculate the flux of F~ (x, y, z) = h−x, −y, ex +y i on z = x2 +y 2 for 0 ≤ x2 +y 2 ≤ 1
we calculate ∂x f = 2x and ∂y f = 2y
ZZ ZZ
x2 +y 2 • ~ 2 2 x2 +y 2
h−x, −y, e i dS = −2x − 2y + e dA
S Ω
In polar coordinates Ω is described by 0 ≤ r ≤ 1 and 0 ≤ θ ≤ 2π so we calculate

2π
1 2 1
Z 1 4
−r
ZZ Z
2 +y 2 2
h−x, −y, ex ~=
i • dS −2r2 + er rdr dθ = 2π + er = π(e − 2).
S 0 0 2 2 0
1.7. STOKES THEOREM 51
1.7 stokes theorem

We have already encountered a simple version of Stoke’s Theorem in the two-dimensional context.
Recall that I ZZ
∂Q ∂P
P dx + Qdy = − dA
∂R R ∂x ∂y
could be stated in terms of the z-component of the curl for F~ = hP, Q, 0i:
I ZZ
~ ∇ × F~ • zb dx dy

F • d~r =
∂R R
However, notice that the double integral above is actually the surface integral of ∇ × F~ over the
planar surface R where dS ~ = zbdx dy. Let’s generalize this idea a little. Suppose S is some simply
connected planar region with unit-normal n b which is consistently oriented19 with ∂S then we can
derive Greene’s Theorem for S and by the arguments of the earlier section we have that
I ZZ ZZ
~ ~ ∇ × F~ • dS.~

F d~r =
• ∇×F n • b du dv = ?
∂S S S
Suppose that S = S1 ∪ S2 ∪ · · · ∪ Sn is a simply connected surface where each Sj is a planar region

with unit normal n b and consistent boundary ∂Sj . The planar regions Sj are called faces and we
call such a surface a polyhedra.
Theorem 1.7.1. baby Stoke’s for piece-wise flat surfaces.
Suppose S is a polyhedra S with consistently oriented boundary ∂S and suppose F~ is

differentiable on some open set containing S then
I ZZ
~ ∇ × F~ • dS.
~

F • d~r =
∂S S
19
this means as we travel around the boundary the surface is on the left
Proof: The proof is mostly by the picture before the theorem. The key is that because S is
composed of flat faces we can apply ? to each face and obtain for j = 1, 2, . . . , n:
I ZZ
~ ∇ × F~ • dS.
~

F • d~r =
∂Sj Sj
Add these equations together and identify the surface integral,

n I
X n ZZ
X ZZ
F~ • d~r = ∇ × F~ ~= ∇ × F~ ~

• dS • dS.
j=1 ∂Sj j=1 Sj S1 ∪S2 ∪···∪Sn
To simplify the sum of the circulations we need to realize that all the edges of faces which are
interior cancel against oppositely oriented adjacent face edges. The only edge which leads to an
uncancelled flow are those outer edges which are not common to two faces. This is best seen in the
picture. It follows that
X n I I
~
F • d~r = F~ • d~r
j=1 ∂Sj ∂S
which completes the proof of the theorem.
The theorem above naturally extends to a theorem for mostly regular surfaces. I say mostly regular
since we do allow for surfaces which have edges and corners. The normal vector field may vanish
at such edges, however, it is assumed to be nonvanishing elsewhere. There are surfaces where the
normal vector field vanishes at points other than the edge or corner. For example, the mobius
band. Such a surface is non-orientable. Generally, we only wish to consider oriented surfaces. I
implicitly assume S is oriented by stating it has consistently oriented boundary ∂S.
Theorem 1.7.2. Stoke’s Theorem for simply connected surface.
Suppose S is a simply connected surface S with consistently oriented boundary ∂S and

suppose F~ is differentiable on some open set containing S then
I ZZ
~ ∇ × F~ • dS.
~

F d~r =
•
∂S S
Proof: If S is a mostly regular, simply connected, surface then it can be approximately modeled
by a simply connected polyhedra with n-faces. As we take n → ∞ this approximation becomes
exact and we obtain Stoke’s Theorem. .
The reader should find the limit above geometrically obvious, but analytically daunting. We will not
pursue the full analysis of the limit imiplicit within the proof above. However, we will offer another
proof of Stoke’s Theorem for a curved surface of a simple type at the end of this section. This
should help convince the reader of the generality of the theorem. That said, the reader probably
just wants to see some examples at this point:
Example 1.7.3. Let F~ = h−y, x, zi. Find the flux of ∇ × F~ over the half of the outward-oriented
sphere ρ = R with z ≥ 0. Denote the hemisphere SR+ . The hemisphere is simply connected and
the boundary of the outward-oriented hemisphere is given by x = R cos(θ), y = R sin(θ) and z = 0.
Apply Stoke’s Theorem:
ZZ I
~ ~ F~ • d~r

∇ × F dS =
•
SR+ ∂SR+
Z 2π

= − R sin(θ), R cos(θ), 0 • − R sin(θ), R cos(θ), 0 dθ
0
Z 2π
= R2 dθ
0
= 2πR2 .
Example 1.7.4. Let F~ = h−y, x, zi. Find the flux of ∇ × F~ over the half of the outward-oriented
sphere ρ = R with z < 0. Denote the lower hemisphere by SR− . To solve this we can use the result
of the previous problem. Notice that SR+ and SR− share the same set of points as a boundary,
however, ∂SR+ = −∂SR− . Apply Stoke’s Theorem and the orientation-swapping identity for line-
integrals: ZZ I I
~ ~ ~ F~ • d~r = −2πR2 .

∇ × F • dS = F • d~r = −
SR− ∂SR− ∂SR+
Example 1.7.5. Once more think about the vector field F~ = h−y, x, zi. Notice that F~ is differen-
tiable on R3 . We can apply Stoke’s Theorem to any simply connected surface. If the consistently-
oriented boundary of that surface is ∂SR+ then the flux of ∇ × F~ is 2πR2 .
Stoke’s Theorem allows us to deform the flux integral of ∇× F~ over a family of surfaces which share
a common boundary. What about a closed surface? A sphere, ellipsoid, or the faces comprising
a cube are all examples of closed surfaces. If S is an closed surface then ∂S = ∅. Does Stoke’s
Theorem hold in this case?
Example 1.7.6. Suppose S is a simply connected closed surface S and suppose F~ is differentiable
on some open set containing S then I claim that
ZZ
∇ × F~ • dS
~ = 0.

S
To see why this is true simply cut S in halves S1 and S2 . Notice that to consistently oriented S1
and S2 we must have that ∂S1 = −∂S2 . Apply Stoke’s Theorem to each half to obtain:
I ZZ I ZZ
F~ • d~r = ∇ × F~ • dS
~ F~ • d~r = ∇ × F~ • dS~

& (?).
∂S1 S1 ∂S2 S2
Note that, ZZ ZZ ZZ
∇ × F~ ~+ ∇ × F~ ~= ∇ × F~ ~

• dS • dS • dS
S1 S2 S
F~ • d~r = − F~ • d~r. Therefore, addiing the eq. in ? yields that ∇ × F~ ~ = 0.

H H RR
But, ∂S2 ∂S1 S
• dS
We can easily include the result of the example above by defining the integral over the empty set
to be zero. Another interesting extension of the basic version of the theorem is the case that the
surface has a few holes. The justification for this theorem will be a simple extension of the argument
we already offered in the case of Greene’s Theorem.
Theorem 1.7.7. Stoke’s Theorem for connected surface possibly including holes.
Suppose S is a connected surface S with consistently oriented boundary ∂S. If S has holes
then we insist that the boundaries of the holes be oriented such that S is toward the left
of the curve as we travel along the edge of the hole. Likewise, the outer boundary curve
must also be oriented such that the surface if on the left as we traverse the boundary in the
direction of its orientation. Suppose F~ is differentiable on some open set containing S then
I ZZ
F~ • d~r = ∇ × F~ • dS.
~

∂S S
Proof: Mostly by a picture. If S is connected with holes then we can cut S into pieces which are
simply connected. We then apply Stoke’s Theorem to each simply connected component surface.
Finally, sum these equations together to obtain the surface integral of the flux over the whole
surface and the line-integral around the boundary.
The key feature revealed by the picture is that all the interior cuts will cancel in this sum since
any edge which is shared by two simply connected components must be oppositely oriented when
viewed as the consistent boundary of the simply connected components. .
Example 1.7.8. Suppose S is a pyramid with square-base on the xy-plane (do not include the
square-base in the surface S so the boundary of S is the square at the base of the pyramid). Find
the flux of ∇ × F~ through the pyramid if F~ (x, y, z) = h1, 3, z 3 i. Apply Stoke’s Theorem,
ZZ I
∇ × F~ • dS ~= F~ • d~r.
S ∂S
Note that F~ |∂S = h1, 3, 0i since z = 0 on the boundary of the pyramid. Define g(x, y, z) = x + 3y
and note that ∇g = h1, 3, 0i thus F~ |∂S = h1, 3, 0i is conservative on the xy-plane and it follows that
the integral around the closed square loop ∂S is zero. Thus,
ZZ
∇ × F~ • dS ~ = 0.
S
This is not terribly surprising since direct computation easily shows that ∇ × F~ = 0.
Suppose S is a connected surface which has outer boundary Cout and inner boundary Cin where we
have consistently oriented Cout but oppositely oriented Cin ; ∂S = Cout ∪ (−Cin ). Applying Stoke’s
Theorem with holes to a vector field F~ which is differentiable on an open set containing S 20 ,
I I ZZ
~ ~ ∇ × F~ • dS
~

F d~r −
• F d~r =
•
Cout Cin S
This is a very interesting formula in the case ∇ × F~ = 0 on some connected annular surface:
Theorem 1.7.9. Deformation Theorem for connected surfaces
Suppose S is connected with inner boundary Cin (oriented such that the surface’s normal
side is to the right of Cin ) and outer boundary Cout oriented such that S is on the left of
the curve then if F~ is differentiable on an open set containing S and has ∇ × F~ = 0 on S,
I I
F~ • d~r = F~ • d~r.
Cout Cin
Proof: follows from the formula above the theorem .
20
if S were a donut this does not necessitate that F~ be differentiable in the center of the big-circle of the donut, it
merely means F ~ is differentiable near where the actual donut is found
Example 1.7.10. How does Stoke’s Theorem help us understand Conservative Vector Fields in
R3 ? Recall we have a list of equivalent characterizations for a simply connected space U as given in
Proposition 1.4.5: Suppose U is an open connected subset of Rn then the following are equivalent

H
4. (add precondition n = 3 and U be simply connected) ∇ × F~ = 0 on U .

We argued before how ∇ × hP, Q, 0i • zb = ∂x Q − ∂y P = 0 paired with the deformation version of
Green’s Theorem allowed us to shrink loop integrals to a point hence estabishing that F~ = hP, Q, 0i
is conservative if it passed Clairaut’s test (∂y P = ∂x Q on a simple connected two-dimensional do-
main. Let us continue to three dimensions now that we have the needed technology.
Suppose F~ = hP, Q, Ri has vanishing curl (∇ × F~ = 0) on some simply connected subset U of R3 .

Suppose C1 and C2 are any two paths from P to Q in U . Observe that C1 ∪ (−C2 ) bounds a simply
connected surface S on which ∇ × F~ = 0 (since S sits inside U , and U has no holes). Apply Stoke’s
Theorem to S: I ZZ
~ ∇ × F~ • dS ~=0

F • d~r =
C1 ∪(−C2 ) S
F~ • d~r = F~ • d~r and path-independence on U follows.

R R
Therefore, C1 C2
Discussion: criteria 1,2 and 3 are n-dimensional results since the arguments we gave apply
equally well in higher dimensions. However, item 4 is only worthile in it’s application to n = 3 or
with proper specialization n = 2 because we have no cross-product and hence no curl in dimension
n = 4, 5... etc.... You might wonder what is the generalization of (4.) for vector fields in Rn .
The answer involves differential forms. The exterior derivative allows us to properly extend vector
differentiation to n-dimensions. This is not just an academic21 comment, in the study of differential
equations we enjoy solving exact differential equation. If P dx + Qdy + Rdz = 0 then we can solve
by f (x, y, z) = 0 if we can find f with ∂x f = P , ∂y f = Q and ∂z f = R. But, this problem is one
we have already solved:
P dx + Qdy + Rdz = 0 is exact F~ = hP, Q, Ri = ∇f
Thus ∇ × hP, Q, Ri = 0 on simply connected U ⊂ R3 implies existence of solutions for the given
differential equation P dx+Qdy+Rdz = 0. What about the case of additional indpendent variables;
suppose w, x, y, z are variables
Idw + P dx + Qdy + Rdz = 0 is exact F~ = hI, P, Q, Ri = ∇f

21
sad comment on our culture that this is an insult!
If we could find the condition analogue to ∇ × F~ = 0 then we would have a nice test for the
existence of solutions to the given DEqn. It turns out that the test is simply given by the exterior
derivative of the given DEqn; 22 If P1 dx1 + P2 dx2 + · · · + Pn dxn = 0 then this DEQn is exact on
a simply connected domain in Rn iff d[P1 dx1 + P2 dx2 + · · · + Pn dxn ] = 0. Consult my advanced
calculus notes and/or ask me for details about what ”d” means in the context above.
Naturally, we can extend the annular result to the more general case that the surface as finitely-
many holes:
Theorem 1.7.11. Stoke’s Theorem for connected surface possibly including holes.
Suppose S is a connected surface S with consistently oriented boundary ∂S. If S has k holes
H1 , H2 , . . . , Hk then ∂S = Cout ∪ (−∂H1 ) ∪ (−∂H2 ) · · · (−∂Hk ). Suppose F~ is differentiable
on some open set containing S then
I I I I ZZ
~ ~ ~ ~ ∇ × F~ • dS. ~

F • d~r − F • d~r − F • d~r − · · · − F • d~r =
Cout ∂H1 ∂H2 ∂Hk S
Remark 1.7.12.
Electrostatics for a curved surface is an interesting problem. Imagine that the electric field
was confined to the surface. We considered this problem in some depth for the plane. It is
worth mentioning that we could just as well repeat those arguments here if we wished to
model some field which is bound to flow along a surface. The details of the theory would
depend on the particulars of the surface. I leave this as an open problem for the interested
reader. You might even think about what surface you could pick to force the field to have
certain properties... this is a prototype for the idea used in string theory; the geometry of
the underlying space derives the physics. At least, this is one goal, sometimes realized...
22
example: x2 dy − ydx = 0 is not exact since d[x2 dy − ydx] = 2xdx ∧ dy − dy ∧ dx 6= 0. In contrast ydx + xdy = 0 is
exact since d[ydx + xdy] = dy ∧ dx + dx ∧ dy = 0. We discussed the wedge product and exterior derivative in lecture,
ask if interested and missed it...
1.7.1 proof of Stoke’s theorem for a graph

Our goal is to show that ∂S F~ • d~r = S (∇ × F~ ) • dS ~ for a simply connected surface S which can
H RR
be expressed as a graph. Suppose z = f (x, y) for (x, y) ∈ Ω In particular, as a starting point, let
Ω = [a, b] × [c, d]23 . It is easily calculated that ~r(x, y) = hx, y, f (x, y)i induces normal vector field
~ (x, y) = h−∂x f, −∂y f, 1i. The boundary of S consists of 4 line segments: C1 ∪ C2 ∪ (−C3 ) ∪ (−C4 )
N
where
1. C1 : for a ≤ t ≤ b we set x = t, y = c, z = f (t, c) hence dx = dt, dy = 0, dz = ∂x f (t, c)dt
2. C2 : for c ≤ t ≤ d we set x = a, y = t, z = f (a, t) hence dx = 0, dy = dt, dz = ∂y f (a, t)dt
3. C3 : for a ≤ t ≤ b we set x = t, y = d, z = f (t, d) hence dx = dt, dy = 0, dz = ∂x f (t, d)dt
4. C4 : for c ≤ t ≤ d we set x = b, y = t, z = f (b, t) hence dx = 0, dy = dt, dz = ∂y f (b, t)dt.
We could visualize it as follows:
Consider a vector field F~ = hP, Q, Ri which is differentiable on some open set containing S. Cal-
culate, for reference in the calculations below,
∇ × F~ =

∂y R − ∂z Q, ∂z P − ∂x R, ∂x Q − ∂y P
To calculate the flux of ∇ × F~ we need to carefully compute (∇ × F~ ) • N

~;
(∇ × F~ ) • N
~ =

∂y R − ∂z Q, ∂z P − ∂x R, ∂x Q − ∂y P • h−∂x f, −∂y f, 1i

= ∂z Q∂x f + ∂x Q − ∂z P ∂y f + ∂y P + ∂x R∂y f − ∂y R∂x f
23
In my first attempt I tried Ω as a type-I region given by functions f1 , f2 such that Ω = {(x, y) | f1 (x) ≤ y ≤
f2 (x), a ≤ x ≤ b}, however, this is too technical, it is clearer to show how this works for a rectangular domain.
To proceed we break the problem into three. In particular F~ = F~1 + F~2 + F~3 where we let
F~1 = hP, 0, 0i, F~2 = h0, Q, 0i and F~3 = h0, 0, Ri. For F~1 = hP, 0, 0i we calculate:
ZZ Z bZ d
∇ × F~1 ~=

• dS −∂z P ∂y f − ∂y P dy dx
S a c
Z bZ d
=− ∂y P (x, y, f (x, y)) dy dx (chain-rule)
a c
Z b

= P (x, c, f (x, c)) − P (x, d, f (x, d)) dx. (?)
a
On the other hand, we calculate the circulation of F~1 = P x

b around ∂S note that dx = 0 for C2
and C4 hence only C1 and C3 give nontrivial results,
Z Z Z Z
F~1 • d~r = P dx = P dx − P dx
∂S ∂S C1 C3
Z b Z b
= P (t, c, f (t, c))dt − P (t, d, f (t, d))dt
a a
Z b
= P (x, c, f (x, c)) − P (x, d, f (x, d)) dx.
a
Consquently we have established Stoke’s Theorem for F~1 over our rather simple choice of surface.
Continuing, consider F~2 = Q yb. Calculate, given our experience with the P dx integrals we need not
meet in the middle this time, I offer a direct computation:
ZZ Z bZ d
∇ × F~2 ~=

• dS ∂z Q∂x f − ∂x Q dy dx
S a c
Z dZ b
= ∂x Q(x, y, f (x, y)) dy dx (chain-rule, & swapped bounds)
c a
Z d Z d
= Q(b, y, f (b, y))dy − Q(a, y, f (a, y))dy
c c
Z d Z d
= Q(b, t, f (b, t))dt − Q(a, t, f (a, t))dt
Zc Z c
= Qdy − Qdy
C2 −C4
I
= Qdy (integrals along C1 and C3 are zero)
I∂S
= F~2 • d~r.
∂S
Next, we work out Stoke’s Theorem for F~3 = R zb. I’ll begin with the circulation this time,
I Z Z Z Z
~
F3 d~r =
• Rdz − Rdz + Rdz − Rdz
∂S C1 C3 C2 C4
Zb Z b
∂f ∂f
= R(t, c, f (t, c)) (t, c) dt − R(t, d, f (t, d)) (t, d) dt
a ∂x a ∂x
Z d Z d
∂f ∂f
+ R(b, t, f (b, t)) (b, t)dt − R(a, t, f (a, t)) (a, t)dt
c ∂y c ∂y
Z b
∂f ∂f
=− R(x, d, f (x, d)) (x, d) − R(x, c, f (x, c)) (x, c) dx
a ∂x ∂x
Z d
∂f ∂f
+ R(b, y, f (b, y)) (b, y) − R(a, y, f (a, y)) (a, y) dy
c ∂y ∂y
Z b d Z d b
∂f ∂f
=− R(x, y, f (x, y)) (x, y) dx + R(x, y, f (x, y)) (x, y) dy
a ∂x c c ∂y a
Z bZ d Z dZ b
∂R ∂f ∂R ∂f
=− dy dx + dx dy
a c ∂y ∂x c a ∂x ∂y
Z b Z d
∂R ∂f ∂R ∂f
= − dy dx
a c ∂x ∂y ∂y ∂x
ZZ
∇ × F~3 • dS. ~

=
S
Therefore, by linearity of the curl and line and surface integrals we find that
I ZZ
~ ∇ × F~ • dS.
~

F d~r =
•
∂S S
Notice that the choice of rectangular bounds for Ω allowed us to freely exchange the order of
integration since x and y bounds were independent. If Ω was a less trivial type I or type II region,
then the arguments given in this section need some modification since swapping bounds is in general
a somewhat involved process. That said, the result just proved is quite robust when paired with the
earlier polyhedral proof to make a general argument. If surface consists of a graph with a curved
domain then we can break it into rectangular subdomains and apply the result of this section to
each piece. Once more when we sum those results together the nature of the adjoining regions is to
cancel all line integrals modulo the boundary of the overall surface.24 If the surface does not admit
presentation as a graph z = f (x, y) then generally we can patch it together with several graphs25 .
We apply the result of this section to each such patch and the sum the results to obtain Stoke’s
Theorem for a general simply connected surface.
24
Technically, we’ll have to form a sequence of such regions for some graphs and then take the limit as the
rectangular net goes to infinitely many sub-divisions, however, the details of such analysis are beyond the scope of
these notes. If this seems similar to the proof we presented for Greene’s Theorem then your intuition may serve you
well in the remainder of this course.
25
could be z = f (x, y) type, or y = g(x, z) or x = h(y, z), the implicit function theorem of advanced calcululus
will give a general answer to how this is done for a level surface
1.8. GAUSS THEOREM 61
1.8 gauss theorem

Warning: the argument that follows is an infinitesimal argument. To properly unde-
stand the meaning of this discussion we should remember this is simply notation for
a finite approximation where a certain limit is taken. That said, I will not clutter this
argument with limits. I leave those to the reader here. Also... I will offer another less
heuristic argument towards the end of this section, I prefer this one since it connects
with our previous discussion about the divergence of a two-dimensional vector field
and Greene’s Theorem
Greene’s Theorem in the plane quantifies the divergence of the vector field P x
b + Q yb through the
curve ∂D; Z ZZ
∂P ∂Q
(P x
b + Q yb) • n
b ds = + dx dy
∂D D ∂x ∂y
Suppose we consider a three-dimensional vector field F~ = P x
b + Q yb + R zb. Furthermore, suppose
we consider an infinitesimal cylinder E = D × dz.
What is the flux of F~ out of the cylinder? Apply Greene’s Theorem to see that the flux through
the vertical faces of the cylinder are simply given by either of the expressions below:
ZZ ZZZ
∂P ∂Q
Φhor = (P x
b + Q yb) • n
b ds dz = + dx dy dz
∂D×[z,z+dz] D×[z,z+dz] ∂x ∂y
~=n
Since dS bds dz is clearly the vector area element of the vertical face(s) of the cylinder it is clear
that the double integral above is simply the surface integral over the vertical faces of the cylinder.
We identify,
ZZ ZZZ
~ ~ ∂P ∂Q
Φhor = F dS =
• + dx dy dz.
∂D×[z,z+dz] D×[z,z+dz] ∂x ∂y
On the other hand, the flux through the horizontal caps of the cylinder D × {z} and D × {z + dz}
involve only the z-component of F~ since dS~ = zbdx dy for the upper cap and dS~ = − zbdx dy for the
lower cap hence the fluxes are
ZZ ZZ
Φup = R(x, y, z + dz) dx dy & Φdown = −R(x, y, z) dx dy
D D
The sum of these gives the net vertical flux:

ZZ ZZZ
∂R
Φvert = R(x, y, z + dz) − R(x, y, z) dx dy = dx dy dz.
D D×[z,z+dz] ∂z
where in the last step we used the FTC to rewrite the difference as an integral. To summarize,
ZZ ZZZ
~ ~ ∂R
Φvert = F dS =
• dx dy dz.
caps D×[z,z+dz] ∂z
The net-flux through the cylinder is the sum Φvert + Φhor . We find that,
ZZ ZZZ

~ ∂P ∂Q ∂R
P, Q, R • dS = + + dx dy dz.
∂E E ∂x ∂y ∂z
I prefer to write the result as follows:

ZZ ZZZ
F~ • dS
~= ∇ • F~ dV.
∂E E
This is the celebrated theorem of Gauss. We often refer to it as the divergence theorem. It simply
says that the net flux through a surface is portional to the continuous sum of the divergence
throughout the solid. In other words, the divergence of a vector field F~ measures the number of
field lines flowing from a particular volume. We found the two-dimensional analogue of this in our
analysis of Greene’s Theorem and this is the natural three-dimensional extension of that discussion.
For future reference: (this is also called Gauss’ Theorem)
Theorem 1.8.1. divergence theorem for a simple solid
Suppose E a simple solid (has no interior holes) with consistently oriented outward facing
boundary ∂E. If F~ is differentiable on an open set containing E then,
ZZ ZZZ
F~ • dS
~= ∇ • F~ dV.
∂E E
Discussion: But, I only proved it for a cylinder? Is this needed? Does it apply to other shapes?
Yes. Consider the case that D is a rectangular region. We can use the argument offered above to
obtain Gauss’ theorem for a rectangular solid. Take any other simple solid ( one with no holes ) and
note that you can obtain the solid as a union (possibly infinite!) of rectangular solids. Positively
orient each rectangular solid and apply Gauss’ Theorem to each member of the partition. Next,
add the results together. On the one side we obtain the volume integral of the divergence. On the
other side we get a sum of flux over many rectangular solids, some with adjacent faces. Think about
this, any interior face of a particular rectangular solid will share a face with another member of the
partition. Moreover, the common faces must be oppositely oriented in the distinct, but adjacent,
rectangular solids. Thus, the interior flux all cancels leaving only the outside faces. The sum of the
flux over all outside faces is simply the surface integral over the boundary of the simple solid. In
this way we extend Gauss’ Theorem to any solid without holes. Naturally, this leaves something to
be desired analytically, but you can also appreciate this argument is very much the same we gave
for Greene’s Theorem. This would seem to be part of some larger story...but, that is a story for
another day26 .
Example 1.8.2. Suppose F~ (x, y, z) = hx + y, y + x, z + yi and you wish to calculate the flux of
F~ through a set of stairs which has width 3 and 10 steps which are each height and depth 1. Let
E be the set of stairs and ∂E the outward-oriented surface. Clearly the calculation of the flux
over the surface of the stairs would be a lengthy and tedious computation. However, note that
∇ • F~ = ∂(x+y)
∂x + ∂(y+x)
∂y + ∂(z+y)
∂z = 3 hence we find by Gauss’ Theorem:
ZZ ZZZ ZZZ
F~ • dS
~= ∇ • F~ dV = 3 dV = 3V ol(E).
∂E E E
Elementary school math shows:
V ol(E) = 3(1 + 2 + 3 + 4 + 5 + 6 + 7 + 8 + 9 + 10) = 165

ZZ
Hence, F~ • dS
~ = 495.
∂E
Challenge: work the previous example for n-steps.
Example 1.8.3. Suppose F~ is a differentiable at all points near a simple solid E. Calculate the
flux of the curl of F~ through ∂E:
ZZ ZZZ ZZZ
~ ~ ~

∇ × F • dS = ∇ • ∇ × F dV = (0)dV = 0.
∂E E E
I used the identity ∇ • (∇ × F~ ) = 0. You can contrast this argument with the one given in Example
1.7.6. Both examples are worth study.
Example 1.8.4. Problem: Consider the cube E with side-length L and one corner at
the origin. Calculate the flux of F~ = hx, y, zi through the upper face of the cube.
Solution: Note that we cannot use a simple symmetry argument to see it is 1/6 of the given cube
since the face in question differs from the base face (for example) in its relation to the vector field
F~ . On the other hand, if we imagine a larger cube of side-length 2L which is centered at the origin
26
look-up a proof of the generalized Stoke’s Theorem in an advanced calculus text if you are interested. The
key construction involves generalizing the polyhedral decomposition to something called an n-chain or perhaps an
n-simplex depending what you read. Basically, you need some way of breaking the space into oriented parts with
nice oriented boundaries, you prove the theorem for one such item and extrapolate via face-cancelling arguments as
we have seen here in this case
then the vector field is symmetric with respect to the faces of [−L, L]3 . Call this larger cube E 0 and
observe that we can easily calculate the net-flux through ∂E 0 by the divergence theorem.
ZZ ZZZ ZZZ
Φ∂E 0 = ~ ~
F • dS = ~
∇ • F dV = 3 dV = 3V ol(E 0 ) = 3(2L)3 = 24L3 .
E0 E0
Notice that the face [0, L] × [0, L] × {L} is 1/4 of the upper face of E 0 and it is symmetric with
respect to the other 3/4 of the face [−L, L] × [−L, L] × {L} with regard to F~ . It follows the flux
through the upper face of E is 1/4 of the flux through the upper face of E 0 . Moreover, since the
faces of E 0 are symmetric with regard to F~ we find that 1/6 of the total flux through ∂E 0 passes
through that upper face of E 0 . In summary, the flux through the face in question is simply 1/24 of
the total flux through ∂E 0 and the flux through the upper face of E is L3 .
It should come as no suprise that there is a simple argument to extend the divergence theorem to
a solid with a hole(s) in it. Suppose E is a solid which has a hole H in it. Denote the boundary
of E by ∂E = Sout ∪ Sin where these surfaces are oriented to point out of E. Notice we can do
surgery on E and slice it in half so that the remaining parts are simple solids (with no holes). The
picture below illustrates this basic cut.
More exotic holes require more cutting, but the idea remains the same, we can cut any solid with
a finite number of holes into a finite number of simple solids. Apply the divergence theorem, for
an appropriately differentiable vector field, to each piece. Then add these together, note that the
adjacent face’s flux cancel leaving us the simple theorem below:
Theorem 1.8.5. divergence theorem for a solid with k-interior holes.
Suppose E a solid with interior holes H1 , H2 , . . . , Hk . Orient the surfaces S1 , S2 , . . . , Sk of

the holes such that the normals point into the holes and orient the outer surface Sout of
E to point outward; hence ∂E = Sout ∪ S1 ∪ S2 ∪ · · · ∪ Sk gives E an outward oriented
boundary. If F~ is differentiable on an open set containing E then,
ZZ ZZZ
~ ~
F • dS = ∇ • F~ dV.
∂E E
This is perhaps more interesting if we take the holes as solids on their own right with outward
oriented surfaces ∂H1 , ∂H2 , . . . , ∂Hk (this makes ∂Hi = −Si for i = 1, 2, . . . , k. It follows that:
ZZ ZZ ZZ ZZ ZZZ
~ ~
F • dS − ~ ~
F • dS − ~ ~
F • dS − · · · − ~ ~
F • dS = ∇ • F~ dV.
∂Sout ∂H1 ∂H2 ∂Hk E
Note that we have an interesting result if ∇ • F~ = 0 on the E described above. In that case we
obtain a deformation theorem for the flux spreading between surfaces:
Theorem 1.8.6. deformation theorem for a solid with k-interior holes.
Suppose E a solid with interior holes H1 , H2 , . . . , Hk . Let the outer surface Sout of E be
oriented to point outward and give the hole surfaces an orientation which points out of the
hole; If F~ is differentiable on an open set containing E and ∇ • F~ = 0 then,
ZZ ZZ ZZ ZZ
~ ~
F • dS = ~ ~
F • dS + ~ ~
F • dS + · · · + F~ • dS.
~
∂Sout ∂H1 ∂H2 ∂Hk
This theorem forms the basis for the three-dimensional electrostatics and much more. Basically
it says that if a field is mostly divergence free then the flux comes from only those places where
the divergence is non-vanishing. Everywhere else the field lines just spread out. Given F~ with
∇ • F~ = 0 most places, I think of the holes as where the charge for the field is, either sinks or
sources. From these mysterious holes the field lines flow in and out.
1.8.1 three-dimensional electrostatics

The fundamental equation of electrostatics is known as Gauss’ Law. In three dimensions it simply
states that the flux through a closed surface is proportional to the charge which is enclosed.
ΦE = Qenc
In particular, if we denote σ = dQ/dV and have in mind the solid E with boundary ∂E,
I ZZZ
~ • dS
(E ~= σdV
∂E E
Suppose we have an isolated charge Q at the origin and we apply Gauss law to a sphere of radius
ρ centered at the origin then we can argue by symmetry the electric field must be entirely radial
in direction and have a magnitude which depends only on ρ. It follows that:
I ZZZ
~ • dS
(E ~= δdV ⇒ (4πρ2 )E = Q
∂E E
Hence, the coulomb field in three dimensions is as follows:
~ φ, θ) = Q
E(ρ, ρb
4πρ2
Let us calculate the flux of the Coulomb field through a sphere SR of radius R:
I Z
~ • dS
~= Q
(E 2
ρb • ρbdS (1.4)
SR SR 4πρ
Z
Q
= 2
dS
SR 4πR
Z
Q
= dS
4πR2 SR
Q
= (4πR2 )
4πR2
= Q.
The sphere is complete. In other words, the Coulomb field derived from Gauss’ Law does in fact
satisfy Gauss Law in the plane. This is good news. Let’s examine the divergence of this field. It
appears to point away from the origin and as you get very close to the origin the magnitude of E
is unbounded. It will be convenient to reframe this formula for the Coulomb field by
~ Q Q
E(x, y, z) = hx, y, zi = hx, y, zi.
4π(x2 2 2
+y +z ) 3/2 4πρ3
p
Note that as ρ = x2 + y 2 + z 2 it follows that ∂x ρ = x/ρ and ∂y ρ = y/ρ and ∂z ρ = z/ρ. Conse-
quently:

~ Q ∂ x ∂ y ∂ z
∇•E = + +
4π ∂x ρ3 ∂y ρ3 ∂z ρ3
Q ρ − 3xρ2 ∂x ρ ρ3 − 3yρ2 ∂y ρ ρ3 − 3zρ2 ∂z ρ
3
= + +
4π ρ6 ρ6 ρ6
Q ρ − 3x2 ρ ρ3 − 3y 2 ρ ρ3 − 3z 2 ρ
3
= + +
4π ρ6 ρ6 ρ6
3 2 2 2

Q 3ρ − 3ρ(x + y + z )
=
4π ρ6
= 0.
If we were to carelessly apply the divergence theorem this could be quite unsettling: consider,
ZZ ZZZ ZZZ
~ ~
E dS =
• • ~
∇ E dV ⇒ Q= (0)dV = 0.
∂E E E
But, Q need not be zero hence there is some contradiction? Why is there no contradiction? Can
you resolve this paradox?
Moving on, suppose we have N charges placed at source points ~r1 , ~r2 , . . . , ~rN then we can find the
total electric field by the principle of superposition.
We simply take the vector sum of all the coulomb fields. In particular,
N n
~ r) =
X
~j =
X Qj ~r − ~rj
E(~ E
4π ||~r − ~rj ||3
j=1 j=1
What is the flux through a sphere which encloses just the k-th one of these charges? Suppose SR
is a sphere of radius R centered at ~rk . We can calculate that
ZZ
E ~ = Qk
~ k • dS
SR
~ j is differentiable inside all of SR for j 6= k and ∇ • E

whereas, since E ~ j = 0 we can apply the
divergence theorem to deduce that ZZ
~ j • dS
E ~ = 0.
SR
~ =E
Therefore, summing these results together we derive for E ~1 + · · · + E
~k + · · · + E
~ N that
ZZ
E~ • dS
~ = Qk
SR
Notice there was nothing particularly special about Qk so we have derived this result for each
charge in the distribution. If we take a sphere around a charge which contains just one charge
then Gauss’ Law applies and the flux is simply the charge enclosed. Denote S1 , S2 , . . . , SN as little
spheres which enclose the charges Q1 , Q2 , . . . , QN respective. We have
ZZ ZZ ZZ
Q1 = ~ ~
E • dS, Q2 = ~ ~
E • dS, . . . , QN = ~ • dS
E ~
S1 S2 SN
Now suppose we have a surface S which encloses all N of the charges. The electric field is dif-
ferentiable and has vanishing divergence at all points except the location of the charges. In fact,
the superposition of the coulomb fields has vanishing divergence (∇ • E ~ = 0) everywhere except
the location of the charges. It just has the isolated singularities where the charge is found. We
can apply deformation theorem version of the divergence theorem to arrive at Gauss’ Law for the
distribution of N -charges:
ZZ ZZ ZZ ZZ
~ ~
E dS =
• ~ ~
E dS +
• ~ ~
E dS + · · · +
• E~ • dS
~ = Q1 + Q2 + · · · + QN
S S1 S2 SN
You can calculate the divergence is zero everywhere except at the location of the source charges.
Morale of story: even one point thrown out of a domain can have dramatic and global consequences
for the behaviour of a vector field. In physics literature you might find the formula to describe
what we found by a dirac-delta function these distributions capture certain infinities and let you
work with them. For example: for the basic coulomb field with a single point charge at the origin
~ φ, θ) = Q 2 ρb this derived from a charge density function σ which is zero everywhere except at
E(ρ, 4πρ RRR
the origin. Somehow E σdV = Q for any region R which contains (0, 0, 0). Define σ(~r) = Qδ(~r).
Where we define: for any function f which is continuous near 0 and any solid region E which
contains the origin Z
f (~r)δ(~r)dV = f (0)
E
RR
and if E does not contain (0, 0, 0) then E f (~r)δ(~r)dV = 0. The dirac delta function turns
integration into evaluation. The dirac delta function is not technically a function, in some sense it
is zero at all points and infinite at the origin. However, we insist it is manageably infinity in the
way just described. Notice that it does at least capture the right idea for density of a point charge:
suppose E contains (0, 0, 0), ZZ ZZ
σdV = Qδ(~r)dV = Q.
E E
On the other hand, we can better understand the divegence calculation by the following calcula-
tions27 :
ρb
∇ • 2 = 4πδ(~r).
ρ
~ = Q ρ ~ = ∇ • Q ρb2 = Q ∇ • ρb2 = Qδ(~r). Now once more apply

Consequentlly, if E 4π ρ
b
2 then ∇ •E
4π ρ 4π ρ
Gauss’ theorem to the Coulomb field. This time appreciate that the divergence of E ~ is not strictly
zero, rather, the dirac-delta function captures the divergence: recall the RHS of this calculation
followed from direct calculation of the flux of the Coloumb field through the circle ∂R,
ZZ ZZZ ZZZ
~
E • dS =~ ~
∇ • E dV ⇒ Q= Qδ(~r)dV = Q.
∂E E E
All is well. This is the way to extend Gauss’ theorem for coulomb fields. The fields discussed in
this section are the ones found in nature for the most part. Electric fields do propagate in three-
dimensions and that means that isolated charges establish a coulomb field. In a later section of
this chapter we seek to describe how a continuous distribution of charge can generate a field. At
the base of that discussion are the ideas presented here, although, we will not have need of the
dirac-delta for the continuous smeared out charge. Some physicists argue that there is no such
thing as a point charge because the existence of such a charge comes with some nasty baggage. For
example, if you calculate the total energy of the Coulomb field for a single point charge you find
there is infinite energy in the field. Slightly unsettling.
27
I don’t intend to explain where this 4π comes from, except to tell you that it must be there in order for the
extension of Gauss’ theorem to work out nicely.
1.8.2 proof of divergence theorem for a rectangular solid

Suppose F~ is differentiable near the solid E = [x1 , x2 ] × [y1 , y2 ] × [z1 , z2 ]. Denote the faces of the
solid as follows:
S1 : ~ = −x
x = x1 , (z, y) ∈ [z1 , z2 ] × [y1 , y2 ] has dS b dz dy
S2 : ~= x
x = x2 , (y, z) ∈ [y1 , y2 ] × [z1 , z2 ] has dS b dy dz
S3 : ~ = − yb dx dz
y = y1 , (x, z) ∈ [x1 , x2 ] × [z1 , z2 ] has dS
S4 : ~ = yb dz dx
y = y2 , (z, x) ∈ [z1 , z2 ] × [x1 , x2 ] has dS
S5 : ~ = − zb dy dx
z = z1 , (y, x) ∈ [y1 , y2 ] × [x1 , x2 ] has dS
S6 : ~ = zb dx dy
z = z2 , (x, y) ∈ [x1 , x2 ] × [y1 , y2 ] has dS
The nice thing about the rectangular solid is that only one component of F~ = hP, Q, Ri cuts through
a given face of the solid.
Observe that:
Z Z
Φ12 = ~ ~
F • dS + F~ • dS~ (this defines Φ12 for future reference)
S1 S2
Z y2 Z z 2 Z z 2 Z y2
= ~
F (x1 , y, z) (− x
• bdz dy) + F~ (x2 , y, z) • ( x
bdz dy)
y1 z1 z1 y1
Z y2 Z z 2 Z z 2 Z y2

= −P (x1 , y, z) dz dy + P (x2 , y, z) dy dz
y z z1 y1
Z 1z2 Z 1y2

= P (x2 , y, z) − P (x1 , y, z) dy dz
z1 y1
Z z 2 Z y2 Z x2
∂P
= dx dy dz by the FTC.
z1 y1 x1 ∂x
Likewise,
Z Z
Φ34 = F~ • dS
~+ F~ • dS~ (this defines Φ34 for future reference)
S3 S4
Z z 2 Z x2 Z x2 Z z 2
= ~
F (x, y1 , z) • (− ybdx dz) + F~ (x, y2 , z) • ( ybdz dx)
z1 x1 x1 z1
Z z2 Z x2 Z x2 Z z 2

= −Q(x, y1 , z) dz dy + Q(x, y2 , z) dy dz
Zz1x2 Zx1z2 x1 z1

= Q(x, y2 , z) − Q(x, y1 , z) dy dz
x1 z1
Z x2 Z z 2 Z y2
∂Q
= dx dy dz.
x1 z1 y1 ∂y
Repeating the same argument once more we derive:

Z Z
Φ56 = F~ • dS
~+ F~ • dS
~ (this defines Φ12 for future reference)
S5 S6
Z x2 Z y2 Z z2
∂R
= dz dy dx.
x1 y1 z1 ∂z
The flux of F~ over the entire boundary ∂E is found by summing the flux through each face.
Therefore, by linearity of the triple integral for the second line,
ZZ
F~ • dS
~ = Φ12 + Φ34 + Φ56
∂E
Z x2 Z y2 Z z 2
∂P ∂Q ∂R
= + + dz dy dx.
x1 y1 z1 ∂x ∂y ∂z
Which proves the divergence theorem for an arbitrary rectangular solid.
In contrast to the earlier argument in this section the third dimension of the cylinder was not take
as infinitesimal. That said, it wouldn’t take much to modify the earlier argument for a finite height.
The result just proved extends to more general solids in the way discussed earlier in this section
following the cylindrical proof.
1.9 green’s identities and helmholtz theorem

This section is mostly developed in your homework, well, except for Green’s Third identity and
its applications. I think you’ll be happy I did not relagate that to your homework. The purpose
of this section is severalfold: (1.) we get to see how to use the divergence theorem yields further
identities, in some sense these are the generalization of integration by parts to our current context,
(2.) we lay some foundational mathematics which is important for the logical consistency of the
potential formulation of electromagnetism (3.) we see new and fun calculations.
In your homework I asked you to show the following identities:
Proposition 1.9.1. Green’s First and Second Identities
Suppose f, g ∈ C 2 (D0 ) where E 0 is an open set containing the simple solid E which has
piecewise smooth boundary ∂E. Then,
ZZZ ZZZ ZZ
f ∇2 g dV = ~

(1.) ∇f • ∇g dV + f ∇g • dS
ZZZ E E ZZ ∂E
2 2 ~

(2.) f ∇ g − g∇ f dV = f ∇g − g∇f • dS
E ∂E
The identities above yield important results about harmonic functions. A function f is called
harmonic on E if ∇2 f = 0 on E. You are also asked to show in the homework that:
Proposition 1.9.2. properties of harmonic functions on a simple solid E
Suppose the simple solid E which has piecewise smooth boundary ∂E and suppose f satisfies
∇2 f = 0 throughout E. Then,
ZZ
(1.) ~=0
∇f • dS
ZZZ∂E ZZ
~

(2.) ∇f • ∇f dV = f ∇f • dS
E ∂E
(3.) f (x, y, z) = 0 for all (x, y, z) ∈ ∂E ⇒ f (x, y, z) = 0 for all (x, y, z) ∈ E.
2 2
(4.) If ∇ V1 = b and ∇ V2 = b throughout E and V1 = V2 on ∂E then V1 = V2 throughout E.
In words, (3.) states that if the restriction of f to ∂E is identically zero then f is zero throughout
E. Whereas, (4.) states the solution to the Poisson Equation ∇2 V = b is uniquely determined
by its values on the boundary of a simple solid region.
1.9. GREEN’S IDENTITIES AND HELMHOLTZ THEOREM 73
This picture should help make sense of the notation used in the Third Identity of Green:
Proposition 1.9.3. Green’s Third Identity.

Suppose E is a simple solid, with piecewise smooth boundary ∂E, and assume f is twice
differentiable throughout E. We denote ~r = hx, y, zi as a fixed, but arbitrary, point in E
and denote the variables of the integration by ~r 0 = hx0 , y 0 , z 0 i, so dV 0 = dx0 dy 0 dz 0 and
∇0 = x b0 ∂x
∂
b0 ∂y
0 + y
∂
b0 ∂z∂ 0 . With this notation in mind,
0 + z
∇0 2 f (~r 0 ) 0 ∇0 f (~r 0 )

−1
ZZZ ZZ
1 0 0 1 ~
f (~r) = 0
dV + −f (~r )∇ + • dS
4π E ||~r − ~r || 4π ∂E ||~r − ~r 0 || ||~r − ~r 0 ||
We partly follow the proof in Colley. Begin with a Lemma,

Lemma 1.9.4. Let h be a continuous function and SR is a sphere of radius R centered at ~r then
h(~r 0 )
ZZ
lim dS = 0.
R→0+ r − ~r 0 ||
SR ||~
The proof relies on the fact that if ~r 0 ∈ SR then, by the definition of a sphere of radius R centered
at ~r, we have ||~r − ~r 0 || = R. Thus,
h(~r 0 ) h(~r 0 )
ZZ ZZ ZZ
1
dS = dS = h(~r 0 ) dS
r − ~r 0 ||
SR ||~ r − ~r 0 ||
SR ||~ R SR
Since h is continuous28 it follows that there exist ~a, ~b ∈ SR such that h(~a) ≤ h(~r 0 ) ≤ h(~b) for all
~r 0 ∈ SR . Consequently,
h(~r 0 ) h(~b) h(~r 0 )
ZZ
h(~a)
≤ ≤ ⇒ 4πRh(~
a ) ≤ dS ≤ 4πRh(~b)
||~r − ~r 0 || ||~r − ~r 0 || ||~r − ~r 0 || r − ~r 0 ||
SR ||~
As R → 0+ is clear that ~a, ~b → ~r hence h(~b) → h(~r) and h(~a) → h(~r). Observe, as R → 0+ we
obtain 4πRh(~a) → 4πRh(~r) → 0 and 4πRh(~b) → 4πRh(~r) → 0. The Lemma above follows by the
squeeze theorem O
28
I use the extreme value theorem: any continuous, real-valued, image of a compact domain attains its extrema;
Lemma 1.9.5. Let h be a continuous function and SR is a sphere of radius R centered at ~r then
ZZ
0 1 0 ~ = −4πh(~r).
lim h(~r )∇ • dS
R→0+ SR ||~
r − ~r 0 ||
The proof of this Lemma is similar to the previous. We begin by simplifying the integral. Note,
~r − ~r 0 = hx − x0 , y − y 0 , z − z 0 i
Let L = ||~r − ~r 0 || thus L2 = (x − x0 )2 + (y − y 0 )2 + (z − z 0 )2 . Calculate,
∂ ∂L ∂L x0 − x
0
[L2 ] = 2L 0 = −2(x − x0 ) ⇒ = .
∂x ∂x ∂x0 L
∂L y 0 −y ∂L z 0 −z
Likewise, a similar calculation shows, ∂y 0 = L and ∂z 0 = L . Thus,

0 1 0 1 1 1
1
= − 2 ∇0 L = 3 x − x0 , y − y 0 , z − z 0 = 3 ~r − ~r 0

∇ 0
=∇
||~r − ~r || L L L L
~ on SR points in the ~r − ~r 0 direction at ~r 0 thus

Continuing, note the normal vector field N
~r − ~r 0 1
~=
dS 0
dS = (~r − ~r 0 )dS.
||~r − ~r || L
To calculate the surface integral, note L = R on SR thus,

ZZ ZZ
1 0 1 0 • 1
h(~r 0 )∇0 ~= (~r − ~r 0 )dS

0 ||
• dS h(~r ) 3
~
r − ~
r
SR ||~
r − ~
r SR R R
||~r − ~r 0 ||2
ZZ
= h(~r 0 ) dS
SR R4
ZZ
1
= 2 h(~r 0 ) dS
R SR
We once again use the squeeze theorem argument of the previous Lemma, since h is continuous it
follows that there exist ~a, ~b ∈ SR such that h(~a) ≤ h(~r 0 ) ≤ h(~b) for all ~r 0 ∈ SR . Consequently,
ZZ ZZ ZZ
0
h(~a)dS ≤ h(~r )dS ≤ h(~b)dS
SR SR SR
But, the integrals on the edges are easily calculated since h(~a), h(~b) are just constants and we
deduce:
ZZ ZZ
0 1
2
4πR h(~a) ≤ 2 ~
h(~r )dS ≤ 4πR h(b) ⇒ h(~a) ≤ h(~r 0 )dS ≤ h(~b).
SR 4πR2 SR
As R → 0+ is clear that ~a, ~b → ~r hence h(~b) → h(~r) and h(~a) → h(~r) and the lemma follows by
the squeeze theorem. O.
Green’s Second Identity applies to solid regions with holes provided we give the boundary the
standard outward orientation. With that in mind, consider E 0 = E − BR where BR is the closed-
ball of radius R which takes boundary SR ; ∂BR = SR . However, we insist that ∂E 0 = ∂E ∪ (−SR )
so the hole at ~r has inward-pointing normals. Apply Green’s Second Identity with g(~r 0 ) = ||~r−~
1
r 0 || :
∇0 2 f (~r 0 ) ∇0 f (~r 0 )
ZZZ ZZ
2 1 1
f (~r 0 )∇0 − dV 0
= f (~
r 0
)∇0
− ~
• dS
E0 ||~r − ~r 0 || ||~r − ~r 0 || ∂E 0 ||~r − ~r 0 || ||~r − ~r 0 ||
Once more, use the notation L = ||~r − ~r 0 ||. Observe that
0
L − (x0 − x) x L−x

∂2L ∂ x0 − x L2 − (x0 − x)2

= = =
∂x02 ∂x0 L L L2
similar formulas hold for y and z hence:
2 1 ∂2L ∂2L ∂2L L2 − (x0 − x)2 − (y 0 − y)2 − (z 0 − z)2
∇0 = + + = = 0.
||~r − ~r 0 || ∂x02 ∂y 02 ∂z 02 L2
Therefore, Green’s Second Identity simplifies slightly: (in the second line we use ∂E 0 = ∂E ∪ (−SR )
∇0 2 f (~r 0 ) ∇0 f (~r 0 )
ZZZ ZZ
0 0 0 1 ~
− dV = f (~r )∇ − • dS
E0 ||~r − ~r 0 || ∂E 0 ||~r − ~r 0 || ||~r − ~r 0 ||
∇0 f (~r 0 )
ZZ
1
= f (~r 0 )∇0 − • dS~
∂E ||~r − ~r 0 || ||~r − ~r 0 ||
∇0 f (~r 0 )
ZZ
0 0 1 ~
− f (~r )∇ − • dS
SR ||~r − ~r 0 || ||~r − ~r 0 ||
Observe that ∇0 f (~r 0 ) is continuous as required by the Lemma 1.9.5. Suppose R → 0+ and apply
Lemma 1.9.5 and Lemma 1.9.4 to simplify the surface integrals over SR . Moreover, as R → 0+ we
see E 0 → E − {~r} and it follows:
∇0 2 f (~r 0 ) ∇0 f (~r 0 )
ZZZ ZZ
0 0 0 1 ~ + 4πf (~r).
− dV = f (~r )∇ − • dS
E ||~r − ~r 0 || ∂E ||~r − ~r 0 || ||~r − ~r 0 ||
Green’s Third Identity follows by algebra, just solve for f (~r).
I like the proof of this proposition because it is little more than careful calculation paired with a
few natural limits. If you study the Coulomb field and the way it escapes the divergence theorem29
due to the singularity at the origin then you might be led to these calculations. In any event, we’ve
proved it so we can use it now.
29
I discussed how many physics students are taught to escape the difficulty in Section 1.8.1
Problem: Solve ∇2 f = b for f on E.
Take the Laplacian of Green’s Third Identity with respect to ~r ∈ E. It can be shown through
relatively straight-forward differentiation the surface integral over ∂E are trivial hence we find the
beautiful formula:
−1 2 ∇0 2 f (~r 0 ) 0
ZZZ
∇2 f (~r) = ∇ dV
4π r − ~r 0 ||
E ||~
Let h(~r) = ∇2 f (~r) to see what the formula above really says:
−h(~r 0 )
ZZZ
h(~r) = ∇ 2
dV 0
E 4π||~r − ~r 0 ||
We have only provided evidence it is true if h is the Laplacian of another function f , but it is true
in more generality30 The formula boxed above shows how a particular modified triple integration
in an inverse process to taking the Laplacian. It’s like a second-order FTC for volume integrals.
Returning to the problem, and placing faith in the generality of the formula31 , think of h = f and
assume ∇2 f = b: ( I leave the details of why ∇2 can be pulled into the integral and changed to
∇0 2 )
−f (~r 0 ) −∇0 2 f (~r 0 ) −b(~r 0 )

ZZZ ZZZ ZZZ
f (~r) = ∇ 2
dV 0 = dV 0 = dV 0
E 4π||~r − ~r 0 || E 4π||~r − ~r 0 || E 4π||~r − ~r 0 ||
Therefore, we find the following theorem.
Theorem 1.9.6. Integral solution to Poisson’s Equation:
If ∇2 f = b for some continuous function b on a simple solid region E then
−b(~r 0 )
ZZZ
f (~r) = dV 0 .
E 4π||~r − ~r 0 ||
If we are given that b tends to zero fast enough as we let ||~r|| → ∞ then the domain of
integration E may be extended to R3 and the boxed equation serves to define a global solution to
Poisson’s Equation. Helmholtz’ Theorem is related to this discussion. Let me state the Theorem
for reference:
30
Colley points to Kellog’s Foundations of Potential Theory from 1928, I’d wager you could find dozens of texts
to support this point. For example, Flander’s text develops these ideas in blinding generality via differential form
calculations.
31
sorry folks, I’d like to fill this gap, but time’s up
Theorem 1.9.7. Helmholtz
Suppose F~ is a vector field for which ∇ • F~ = D

~ and ∇ × F~ = C. ~ Furthermore, suppose
F~ → 0 as ||~r|| → ∞ and C, D tend to zero faster than 1/||~r||2 then F~ is uniquely given by:
F~ = −∇U + ∇ × W
~
where
D(~r 0 ) C(~r 0 )
ZZZ ZZZ
1 1
U (~r) = dV 0 & W (~r) = dV 0
4π R3 ||~r − ~r 0 || 4π R3 ||~r − ~r 0 ||
For the interested reader, this does not contradict the multivariate Taylor theorem. In Taylor’s
theorem we are given all derivatives at a particular point and that data allows us to reconstruct
the function. In Helmholtz Theorem we are given two globally defined vector fields and some first
order derivative data which is sufficient to reconstruct the vector field F~ . The difference is in the
domain of the givens. At a point vs. on all points.
1.10 maxwell’s equations and the theory of potentials

The central equations of electromagnetism are known as Maxwell’s Equations in honor of James
Clerk Maxwell who completed these equations around the time of the Civil War. Parts of these
were known before Maxwell’s work, however, Maxwell’s addition of the term µo o ∂E ∂t was crucial in
the overal consistency and eventual success of the theory in explaining electromagnetic phenomena.
In this section we will examine how these equations can either be stated locally as a system of PDEs
or as integrals which yield the fields. Potentials for the fields are also analyzed, we see how Greene’s
identities help connect the integral formulations of the potentials and the corresponding PDE which
are, in a particular gauge, Poisson-type equations. Please understand I am not attempting to ex-
plain the physics here! That would take a course or two, our focus is on the mathematical backdrop
for electromagnetism. I’ll leave the physics for our junior-level electromagnetism course.
Let me set the stage here: E ~ is the electric field, B

~ is the magnetic field, both depend on time t
and space x, y, z generally. In principle the particular field configuration is due to a given charge
~ The electric and magnetic fields are solutions to the following set
density ρ and current density J.
32
of PDEs :
ρ

∂E~ ~
~
∇•E = , ~ ~
∇ × B = µo J + o , ∇•B~ = 0, ∇×E ~ = − ∂B
o ∂t ∂t
These are Gauss’ Law, Ampere’s Law with Maxwell’s correction, no magnetic monopoles and
Faraday’s Law written as locally. In most elementary electromagnetics courses33 these Laws are
presented as integral relations. In those elementary treatments the integral relations are taken as
primary, or basic, experimentally verified results. In constrast, to Colley’s Vector Calculus text,
or most introductory physics texts, I take the PDE form of Maxwell’s equations as basic. In my
opinion, these are the nexus from which all else flows in E & M. For me, Maxwell’s equations define
electromagnetism34 . Let’s see how Stoke’s and Gauss’ theorems allow us to translate Maxwell’s
equations in PDE-form to Maxwell’s equations in integral form.
1.10.1 Gauss’ Law

Suppose M is a simple solid with closed surface S = ∂M where and apply the divergence theorem
to Gauss’ Law:
ZZZ ZZZ ZZ
~ ρ ~
1 ~ = Qenc
~ • dS
∇•E = ⇒ ∇ • E dV = ρ dV ⇒ E
o M o M ∂M o
The equation boxed above is what we call Gauss’ Law in the freshman35 E & M course. It simply
states that the flux through a closed ”gaussian” surface is given by the charge enclosed in the
surface divided by the permitivity of space.
32
I use SI units and o and µo are the permitivity and permeability of empty space
33
for example, Physics 232 at LU
34
ok, in truth there is a case that escapes Maxwell’s equations, but that nonlinear case is considerably more
sophisticated than these notes...
35
because there are three more years of physics past this course... it is to be done in the Freshman year of university.
1.10. MAXWELL’S EQUATIONS AND THE THEORY OF POTENTIALS 79
Example 1.10.1. Suppose you have a very long line of charge along the z-axis with constant
density λ = dQ/dz. Imagine a Gaussian cylinder S length L centered about the z-axis: only the
~ = Eb
curved side of the cylinder gets flux and by the geometry it is clear that E r. Thus,
ZZ
~ = Qenc
~ • dS
E ⇒ 2πrLE =
λL
S o o
~ = λ
and we find E rb.
2πo r
It is interesting this is the dependence on distance we saw in two-dimensional electrostatics.
1.10.2 Ampere’s Law

Suppose we have steady currents. It turns out that ∂E
∂t = 0 in this case so we need to solve
~ ~
∇ × B = µo J in this magnetostatic case. Suppose S is a simply connected surface and use
Stoke’s Theorem in the calculation that follows:
ZZ ZZ I
~ = µo J~ ⇒ ~ ~ = µo J~ • dS
~ ~ • d~r = µo Ienc .

∇×B ∇×B • dS ⇒ B
S S ∂S
The boxed equation above is called Ampere’s Law in the basic course. It states that the circulation
of the magnetic field around an ”amperian” loop is the product of the current that cuts through
the loop and the permeability of space.
Example 1.10.2. Suppose we are given a constant current I in the zb-direction. We can apply
Ampere’s Law to a circle centered on the z-axis:
I
~ • d~r = µo I µo I
B ⇒ B(2πr) = µo I ⇒ B=
C 2πr
To make the calculation above we used the symmetry of the given current. There was no way to
have a z or θ dependence on the magnitude. To anticipate the direction of B ~ I used the right-hand-
rule for current which states that the magnetic field induced from a current wraps around the wire.
If you point your right thumb in the direction of current then your fingers wrap around like the
magnetic field. In short, B~ = µo I θ. ~ from direct analysis
b We could derive the directionality of B
2πr
~ ~ ~
of ∇ × B = µo J, this says that B is changing in a direction which is perpendicular to J. ~ The
cross-product of the curl builds in the right-handedness.
If the electric field has a time-dependence we can use a modified form of Ampere’s Law where the
~
psuedo-current J~d = o ∂∂tE is added to the real current. Otherwise the mathematics is the same, we
just have to throw in Maxwell’s correction to the enclosed current.
1.10.3 Faraday’s Law

Suppose S is a simply connected surface with consistently oriented boundary ∂S.
~
∂B
ZZ ZZ
∂B~
~ =− ~ ~=− ~

∇×E ⇒ ∇×E dS • • dS
∂t S S ∂t
I ZZ
~ ∂ ~ • dS
~
⇒ E d~r = −
• B
∂S ∂t S
∂ΦB
⇒ E =− .
∂t
The boxed equation is called Faraday’s Law in the basic E & M course. The script E is the voltage
around the loop ∂S and ΦB is the magnetic flux through the loop. In words, a changing magnetic
flux induces a voltage in the opposite direction. The minus in this law expresses the fact that
nature abhors a change in flux. Of course, we have yet to properly introduce voltage in this section.
~ = 0 equation.
That is our next project, right after we deal with the ∇ • B
1.10.4 no magnetic monopoles?

In the case of the electric field we have ∇ • E~ = ρ and the charge density can be nonzero since
o
electric charge does in fact exist. On the other hand, if there was such a thing as magnetic charge36
it would have a density ρm and, depending on our choice of units, some k such that
~ = kρm
∇•B
However, since we are taking ∇ • B ~ = 0 as basic in this section we can conclude that ρm = 0
throughout all space. There is no isolated magnetic charge. North and South poles always come
in pairs it seems. For these reasons the equation ∇ • B ~ = 0 is called the no magnetic monopoles
equation. It is a silent participant in the story of basic electromagnetism. Its presence is felt in
what is not considered. All of this said, the existence of magnetic monopoles is an ongoing and
interesting question in theoretical physics.
36
if you’d like to read more on magnetic charge there are many sources. One nice discussion is found in Gravitation
by Misner, Thorne and Wheeler where possible theories of electromagnetism are selected by something called the
complexion
1.10. MAXWELL’S EQUATIONS AND THE THEORY OF POTENTIALS 81
1.10.5 potentials
Both the electric and magnetic fields can be derived from potentials. This is part of a much
~ and B
more general result which I intend to have you explore in a homework. In the case of E ~
it is customary to derive these from the scalar potential φ and the vector potential A. ~ In
~ ~
particular: for a given E and B we insist the potentials satisfy:
~ = −∇φ + ∂A
E & ~ = ∇ × A.
B ~
∂t
Notice this means that the choice of φ and A ~ are far from unique. Why? Check it for yourself, if
~ ~ 0 ~ 0
A, φ are as above then A = A + ∇g and φ = φ + ∂t g will yield the same E, ~ B.
~ For convenience in
~ = 0 this is called the Coulomb gauge condition. Typically we
this section we suppose that ∇ • A
~
call φ the voltage whereas A is the vector potential. Only φ is seen in the basic E & M course.
I want to connect with some earlier discussion of Laplace’s and Poisson’s Equations. Let’s examine
how Gauss’ Law translates when put into the potential formulation:
~ = ρ ρ −ρ
∇•E ⇒ ∇ • (−∇φ) = ⇒ ∇2 φ = .
o o o
Therefore, the potential must be a solution to Poisson’s equation. In the case that ρ = 0 on some
simply connected volume we find the potential must be a solution to Laplace’s equation ∇2 φ = 0.
This means in the absence of charge we can state some very interesting results about voltage:
1. if the voltage is constant along the boundary of a charge-free simply connected region then
the voltage is constant throughout the region.
2. by Theorem 1.9.6 we can solve for φ by calculating the appropriate integral of the charge
density! Identify that b = −ρ/o hence
ρ(~r 0 )
ZZZ
φ(~r) = dV 0 .
M 4πo ||~r − ~r 0 ||
In electrostatics we arrive at the same formula by applying the superposition principle to the po-
dQ
tential due to a continuum of charge ρ = dV . Each infinitesimal charge dQ at ~r 0 gives a voltage of
dQ
dφ = 4πo ||~ r 0 || then we integrate over the whole region where ρ 6= 0 and obtain the net-voltage
r−~
due to the charge distribution ρ. We have shown that this intuitive approach is correct, it yields the
same solution to Gauss’ Law for the potential. Moreover, since this is a Poisson equation we even
know that once we are given ρ and specify the voltage along a particular boundary then the voltage
is uniquely specified elsewhere. I highly reccomend Griffith’s Introduction to Electrodynamics if
you’d like to read the physics which leads to the same conclusion.
Continuing, what about Ampere’s Law? How does it translate into potential formalism?
~
∂E

∂∇φ

~ ~
∇ × B = µo J + o ⇒ ~ ~
∇ × (∇ × A) = µo J − o
∂t ∂t

~ ~ ~ ∂φ
⇒ ∇(∇ • A) − ∇ • ∇A = µo J − o ∇
∂t

⇒ ∇2 A ~ = −µo J~ + o ∇ ∂φ .
∂t
The process of choosing the Coulomb gauge is an example of gauge fixing. These partially adhoc
conditions placed on the potentials greatly simplify the derivation of the potentials. Although, here,
the Coulomb gauge actually makes the Poisson Equation for the vector potential a bit cluttered.
However, if the currents considered are all steady and constant then there is no time-dependence
to the electric field and the Coulomb gauge turns Ampere’s Law into a set of three Poisson-type
equations:
∇2 A1 = −µo J1 , ∇2 A2 = −µo J2 , ∇2 A3 = −µo J3 .
Perhaps you can also appreciate that the distinction between φ and A ~ is less than you might think,
in fact, these are just two parts of what is properly viewed as the four-potential in relativistic
electrodynamics, but, I think I should probably leave that for another course37 . This much we
can say, in the absence of current the vector potential must have component functions which solve
Laplace’s equation ∇2 Ai = 0 for i = 1, 2, 3. This means:
1. if the vector potential is constant along the boundary of a current-free simply connected
region then the vector-potential is constant throughout the region.
2. by Theorem 1.9.6 we can solve for A ~ by calculating the appropriate integral of the current
density! (in the magnetostatic case where ∂φ ∂t = 0). Identify that b = −µo Jk hence for
k = 1, 2, 3 hence, applying the Theorem three times:
µo J1 (~r 0 )
ZZZ
A1 (~r) = 0 ||
dV 0
M 4π||~
r − ~
r
µo J2 (~r 0 )
ZZZ
A2 (~r) = dV 0
M 4π||~r − ~r 0 ||
µo J3 (~r 0 )
ZZZ
A3 (~r) = dV 0
M 4π||~r − ~r 0 ||
Consequently, we can write the single vector integral to capture the vector potential:
ZZZ ~ r 0)
µo J(~
~ r) =
A(~ dV 0
M 4π||~r − ~r 0 ||.
37
those notes are already posted, see my Math 430 notes, the whole goal of that course is to analyze Maxwell’s
Equations in differential forms

Calculus IIIvectorcalculus 2011

Uploaded by

Copyright:

Available Formats

Calculus IIIvectorcalculus 2011

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Calculus IIIvectorcalculus 2011

Uploaded by

Copyright:

Available Formats

Chapter 1

1.1 vector fields

We have already encountered examples of vector fields earlier in this course.

Example 1.1.3. Let F~ = x ~ = yb or H

Similarly, the spherical coodinate frame ρb, φ,

Example 1.1.5. Suppose F~ = x b then obviously x(t) = xo + t, y = yo , z = zo is the streamline

1.2 grad, curl and div

Suppose f is a scalar function on R 3 then we defined the gradient vector field

We studied this before, recall that we can compactly express this by

where ∂i = ∂/∂xi and x1 = x, x2 = y and x3 = z. Moreover, we have also shown previously in

∇(f + g) = ∇f + ∇g, & ∇(cf ) = c∇f, & ∇(f g) = (∇f )g + f (∇g)

Together these say that ∇ is a derivation of differentiable functions on Rn .

Suppose F~ = hF1 , F2 , F3 i is a vector field. We define:

∂F1 ∂F2 ∂F3

More compactly, we can express the divergence by

More compactly, using the antisymmetric symbol ijk 4 ,

Proof: Consider (i.), let F~ = Fi ei and G

where the sums

If f is a differentiable R-valued function and F~ is a differentiable vector field then

If we consider the curl of the curl equations we derive,

1.3 line integrals

1.3.1 curves and paths

~γreverse (0) = ~γ (1) & ~γreverse (1) = ~γ (0)

Generally, ~γreverse (t) = ~γ (1 − t).

~ = ~γ (3π − t) = hcos(3π − t), sin(3π − t)i

~ = h− cos(t), − sin(t)i for π ≤ t ≤ 2π. You can easily verify

Let ~γ1 : [a1 , b1 ] → R3 be a path. We say another path ~γ2 : [a2 , b2 ] → R3 is a

Reparametrizations with du/dt > 0 are said to be orientation preserving.

1.3.2 line-integral of scalar function

Often the centroid is found off the curve.

Example 1.3.10. Suppose x = RR cos(t), yR = R sin(t), z = h for 0 ≤ t ≤ π for a curve with δ = 1.

Of course there are many other applications, but I believe

1.3.3 line-integral of vector field

u : [a1 , b1 ] → [a2 , b2 ] where we have γ~1 (t) = γ~2 (u(t)). Consider,

first understand how to parametrize the ellipse:

Now put it together,

by x = cos(t), y = sin(t), z = 1 for 0 ≤ t ≤ 2π. Note that

Thus, for P = y = sin(t) , Q = z − 1 + x = cos(t) and R = 2 − x = 2 − cos(t) we find

1.4 conservative vector fields

Proof: let ~r : [a, b] → C ⊂ Rn parametrize C and calculate:

Definition 1.4.2. conservative vector field

Suppose U ⊆ Rn then we say F~ is conservative on U iff there exists a potential function

Example 1.4.3. Suppose F~ (x, 3

It follows that we have, for a conservative force, K2 − K1 = −U2 + U1 hence K1 + U1 = K2 + U2 .

Definition 1.4.4. path-independence

Suppose U ⊆ Rn then we say F~ is path-independent on U iff C1 F~ • d~r = C2 F~ • d~r for

each pair of curves C1 , C2 ⊂ U beginning at P and terminating at Q .

Suppose U is an open connected subset of Rn then the following are equivalent

3. C F~ • d~r = 0 for all closed curves C in U

4. (add precondition n = 3 and U be simply connected) ∇ × F~ = 0 on U .

We can give similar arguments to show that

Conequently (3.) is true.

end at Q. Observe that C = C1 ∪ (−C2 ) is a closed curve hence

Clearly (2.) follows.

1.5 green’s theorem

1.5.1 geometry of divergence in two dimensions

Definition 1.5.1. flux of F~ through a simple curve C.

Where ~n is the outward-facing unit-normal to C.

More compactly, using the antisymmetric symbol ijk 4 ,