FW

c Math 321, 2007/03/19 Vector Calculus Essentials

1 Curves, Surfaces, Volumes and their integrals

1.1 Curves

P Recall the parametric equation of a line: ~ r (t) = ~r 0 + t~

v0,
~
v0 −−→
where ~r (t) = OP is the position vector of a point P on the
line with respect to some ‘origin’ O, ~r 0 is the position vector
of a reference point on the line and ~ v 0 is a vector parallel
~
r0 ~
r (t) to the line. Note that this can be interpreted as the linear
motion of a particle with constant velocity ~ v 0 that was at the
point ~
r 0 at time t = 0 and ~
r (t) is the position at time t.
O
More generally, a vector function ~ r (t) of a real variable t defines
a curve C. The vector function ~ r (t) is the parametric represen- C
tation of that curve and t is the parameter. It is useful to think
d~
r /dt
of t as time and ~ r (t) as the position of a particle at time t.
The collection of all the positions for a range of t is the particle ∆~
r
~
r (t)
trajectory. The vector ∆~ r=~ r (t + h) − ~ r (t) is a secant vector
connecting two points on the curve, if we divide ∆~ r by h and
take the limit as h → 0 we obtain the vector d~ r /dt which is ~
r (t + h)
tangent to the curve at ~ r (t). If t is time, then d~ r /dt = ~
v is the
velocity. O
The parameter can have any name and does not need to corre-
spond to time. For instance ~ r (t) = ~
ex a cos θ(t)+~
ey a sin θ(t),
d~
r /dθ
with θ(t) = π cos t and ~
ex , ~
ey orthonormal, is the position of
~
r (θ) θ a particle moving around a circle of radius a, but the particle
oscillates back and forth around the circle. That same circle
is more simply described by

~
r (θ) = ~
ex a cos θ + ~
ey a sin θ,

where θ is a real parameter that can be interpreted as the

angle between the position vector and the ~
ex basis vector.
Remark: Note the common abuse of notation where ~ r (t) and ~
r (θ) are two different vector functions.
Both represent points on the same curve but not the same points, for instance ~ r (t = π/2) = a~ex
while ~
r (θ = π/2) = a~ ey . In principle, we should distinguish between the two vector functions
writing ~r = f~(t) and ~
r = ~ g (θ) with f~(t) = ~
g (θ(t)), but in applications this quickly becomes
cumbersome and a nice advantage of this abuse of notation is that it fits nicely with the chain rule

d~
r d~
r dθ
= .
dt dθ dt
Remark: A convenient parametrization, conceptually, is to use distance along the curve as the
parameter, often denoted s and called the arclength. Once a curve is specified and an s = 0
reference point has been chosen then the distance s from that reference point along the curve
uniquely defines a point on the curve, e.g. for the highway 90 curve, s could be picked as distance
from Chicago, positive westward. For the circle of radius a, this could be the parametrization
~
r (s) = ~
ex a cos(s/a) + ~
ey a sin(s/a), where s is now the distance along the circle from the x-axis and
s/a is the angle in radians.
 F.
c Waleffe, Math 321, 2007/03/19 2

1.2 Integrals along curves, or ‘line integrals’

d~
r C Line element: Given a curve C, the line element denoted d~ r
is an ‘infinitesimal’ secant vector. This is a useful shortcut
for the procedure of approximating the curve by a succession
~
r
of secant vectors ∆~ rn = ~ rn − ~r n−1 where ~ r n−1 and ~ r n are
two consecutive points on the curve, with n = 1, 2, . . . , N
integer, then taking the limit max |∆~ r n | → 0 (so N → ∞).
In that limit, the direction of the secant vector ∆~ r n becomes
∆~
rn C
identical with that of the tangent vector at that point. If an
explicit parametric representation ~ r (t) is known for the curve
~
r n−1 then
~
rn d~
r (t)
d~r= dt (1)
dt

~ · d~ ~ (~
R
The typical ‘line’ integral along a curve C has the form C F r where F r ) is a vector field, i.e.
~
a vector function of position. If F (~
r ) is a force, this integral represent the net work done by the
force on a particle as the latter moves along the curve. We can make sense of this integral as the
limit of a sum, namely breaking up the curve into a chain of N secant vectors ∆~ r n as above then

Z N
X
~ · d~
F r= lim ~ (~
F r n ) · ∆~
rn (2)
C |∆~
r n |→0 n=1

This also provides a practical procedure to estimate the integral by approximating it by a finite
sum. A better approximation would be to use the trapezoidal rule, replacing F ~ (~
r n ) in the sum

1 ~ ~ ~ ~ 1


by 2 F (~r n ) + F (~
r n−1 ) , or the midpoint rule replacing F (~ r n ) by F 2 (~
rn + ~
r n−1 ) . These finite
~ · d~
R
sums converge to the same limit, C F r , but faster for nice functions.
If an explicit representation ~ r (t) is known then we can reduce the line integral to a regular Math
221 integral: Z Z tb  
~ ~ d~
r (t)
F · d~
r= r (t)) ·
F (~ dt, (3)
C ta dt
where ~r (ta ) is the starting point of curve C and ~ r (tb ) is its end point. These may be the same point
even if ta 6= tb (e.g. integral once around a circle from θ = 0 to θ = 2π).
Likewise, we can use the limit-of-a-sum definition to make sense of many other types of line integrals
such as Z Z Z Z
r ) |d~
f (~ r |, ~
F |d~
r |, f (~
r ) d~
r, ~ × d~
F r.
C C C C
The first one gives a scalar result and the latter three give vector results.
I One important example is
Z XN
|d~
r | = lim |∆~
rn | (4)
C |∆~
r n |→0 n=1
which is the length of the curve C from its starting point ~ ra = ~
r 0 to its end point ~
rb = ~
r N . If a
parametrization ~r=~ r (t) is known then
Z Z tb  
 d~
r (t) 
|d~
r| =  dt  |dt|
  (5)
C ta
 F.
c Waleffe, Math 321, 2007/03/19 3

where ~ra = ~r (ta ) and ~ r (tb ). That’s almost a 221 integral, except for that |dt|, what does that
rb = ~
mean?! Again you can understand that from the limit-of-a-sum definition with t0 = ta , tN = tb
and ∆tn = tn − tn−1 . If ta < tb then ∆tn > 0 and dt > 0, so |dt| = dt and we’re blissfully happy.
But if tb < ta then ∆tn < 0 and dt < 0, so |dt| = −dt and
Z tb Z ta
(· · · )|dt| = (· · · )dt, if ta > tb . (6)
ta tb

~ × d~
R
I A special example of a C F r integral is
Z N Z tb  
X d~
r (t)
r × d~
~ r = lim r n × ∆~
~ rn = ~r (t) × dt (7)
C |∆~
r n |→0 n=1 ta dt
This integral yields a vector 2A~ ez whose magnitude is twice the
area A swept by the position vector ~ r (t) if and only if the curve
C lies in a plane perpendicular to ~ez and O is in that plane (recall ∆~
rn C
Kepler’s law that ‘the radius vector sweeps equal areas in equal
times’ ). This follows directly from the fact that ~ r n × ∆~ rn = ∆A
2(∆A)~ ez is the area of the parallelogram with sides ~ r n and ∆~ rn ~
r n−1
which is twice the area ∆A of the triangle ~ r n−1 , ∆~ r n . If C and
rn , ~ ~
rn
O are not coplanar then the vectors ~ r n × ∆~ r n are not all in the ~
ez
same direction and their vector sum is not the area swept by ~ r.
In that more general case, the surface is conical R and to calculate
its area S we would need to calculate S = 21 C |~ r × d~ r |.

Exercises:
1. What is the curve described by ~ r (t) = a cos ωt ~ex + a sin ωt ~
ey + bt ~
ez , where a, b and ω
are constant real numbers and ~
ex , ~
ey , ~
ez represent the orthonormal unit vectors for cartesian
coordinates? What is the tangent to the curve at ~ r (t)?
2. Consider the vector function ~ r (θ) = ~
r c + a cos θ ~
e1 + b sin θ ~
e2 , where ~
rc , ~
e1 , ~
e2 , a and b
are constants, with ~ ei · ~
ej = δij . What kind of curve is this? Next, assume that ~ rc , ~
e1
and ~ e2 are in the same plane. Consider cartesian coordinates (x, y) in that plane such that
~
r = x~ ex + y~ey . Assume that the angle between ~ e1 and ~ex is α. Derive the equation of the
curve in terms of the cartesian coordinates (x, y) (i) in parametric form, (ii) in implicit form
f (x, y) = 0. Simplify your equations as much as possible. Find a geometric interpretation for
the parameter θ.
3. Generalize the previous exercise to the case where ~ r c is not in the same plane as ~ e1 and ~
e2 .
Consider general cartesian coordinates (x, y, z) such that ~ r = x~ ex + y~ ey + z~
ez . Assume that
all the angles between ~ e2 and the basis vectors {~
e1 and ~ ex , ~ ez } are known. How many
ey , ~
independent angles is that? Specify those angles. Derive the parametric equations of the
curve for the cartesian coordinates (x, y, z) in terms of the parameter θ.
4. Derive integrals for the length and area of the planar curve in the previous exercise. Clean
up your integrals and compute them if possible.
R R
5. Calculate C d~ r and C ~r · d~
r along the curve of the preceding exercise from ~ r (0) to ~
r (−3π/2).
~ · d~ ~ × d~ ~ = (~ r |2 when C is the circle of radius
R R
6. Calculate C F r and C F r with F ez × ~ ez × ~
r )/|~
R in the x, y plane centered at the origin. How do the integrals depend on R?
 F.
c Waleffe, Math 321, 2007/03/19 4

1.3 Surfaces
Recall the parametric equation of a plane: ~r (u, v) = ~
r0 + u ~a + v ~b where ~
r 0 is a point on the plane,
~ ~
a, b are two vectors parallel to the plane (but not to each other) and u, v are real parameters.
These parameters specify coordinates for the surface. More generally, the parametric equation of a
surface prescribes the position vector ~
r as a function of two real parameters, u and v say, ~ r (u, v).
Again, the names of the parameters do not matter, ~ r (s, t) is also common. The following two
examples are fundamental.
I If the surface can be parametrized by the cartesian coordinates x and y, i.e. the surface is
described by z = h(x, y), then the position vector of a point on the surface is

~
r (x, y) = x ~
ex + y ~
ey + h(x, y) ~
ez , (8)

where ~
ex , ~
ey , ~
ez are the unit vectors in the respective coordinate directions.
I The surface of a sphere of radius R centered at ~
r c can be parametrized by

~
r (u, v) = ~
rc + ~
e1 R cos u cos v + ~
e2 R sin u cos v + ~
e3 R sin v, (9)

where ~ei · ~
ej = δij . Verify that |~r−~ r c | = R for any real u and v. The parameters u and v can
be interpreted as angles. With ~ e3 as the polar axis, v is the latitude and u is the longitude. The
parametric representation ~ r (ϕ, θ) = ~
rc + ~ e1 R cos ϕ sin θ + ~
e2 R sin ϕ sin θ + ~
e3 R cos θ, is a different
but equally valid parametric representation of the same sphere. Here θ is the polar angle, or co-
latitude. It is the angle between the position vector and the polar axis ~ e3 . The angle ϕ is the
azimuthal, or longitudinal, angle.
Coordinate curves and Tangent vectors If one of the parameters is held fixed, v = v0 say, we
obtain a curve ~ r (u, v0 ). There is one such curve for every value of v. For the sphere parametrized
as in (9), ~
r (u, v0 ) is the v0 -parallel, the circle at latitude v0 . Likewise ~r (u0 , v) describes another
curve. This would be a longitude circle, or meridian, for the sphere. The set of all such curves
generates the surface. These two families of curves are parametric curves or coordinate curves for
the surface. The vectors ∂~ r /∂u and ∂~ r /∂v are tangent to their respective parametric curves and
hence to the surface. These two vectors taken at the same point ~ r (u, v) define the tangent plane
at that point. The coordinates are said to be orthogonal if the tangent vectors ∂~ r /∂u and ∂~
r /∂v
at each point ~r (u, v) are orthogonal to each other.
Normal to the surface at a point At any point ~ r (u, v) on a surface, there is an infinity of
tangent directions but there is only one normal direction. The normal to the surface at a point
~
r (u, v) is given by
~ = ∂~
N
r ∂~
×
r
. (10)
∂u ∂v
Note that the ordering (u, v) specifies an orientation for the surface, i.e. an ‘up’ and ‘down’ side.
Surface element: The surface element dS ~ at a point ~
r on a surface S is a vector perpendicular to
the surface at that point with a magnitude dS that is the area of an ‘infinitesimal’ patch of surface
at that point. Just like in the case of the line element along a curve, this is a useful shortcut for the
limit-of-a-sum interpretation of integrals. The surface element is often written dS ~ =~ ndS where
~
n is the unit normal to the surface at that point. If a parametric representation ~ r (u, v) for the
surface is known then  
~ = ∂~
dS
r ∂~
×
r
dudv, (11)
∂u ∂v
 F.
c Waleffe, Math 321, 2007/03/19 5

~
r du
since the right hand side represents the area of the parallelogram formed by the line elements ∂∂u
and ∂∂v~
r dv. Note that ~ ~ but ~
n 6= N n=N ~ /|N
~ | since ~
n is a unit vector. Although we often need to
refer to the unit normal ~n, it is usually not needed to compute it explicitly since in practice it is
the area element dS~=N ~ dudv which is required.

Exercises:
1. Compute tangent vectors and the normal to the surface z = h(x, y). Show that ∂~ r /∂x and
∂~
r /∂y are not orthogonal to each other in general. Determine the class of functions h(x, y) for
which (x, y) are orthogonal coordinates on the surface z = h(x, y) and interpret geometrically.
Derive an explicit formula for the area element dS = |dS|~ in terms of h(x, y).

2. Deduce from the implicit equation |~ r−~ r c | = R for a sphere of radius R and center ~ r c that
r −~
(~ r c )·∂~ r −~
r /∂u = (~ r c )·∂~
r /∂v = 0 for any u and v, where ~ r (u, v) is any parametrization of
a point on the sphere. Compute ∂~ r /∂u and ∂~ r /∂v for the parametrization (9) and verify that
these vectors are perpendicular to the radial vector ~ r −~r c and to each other (when evaluated
at the same point on the sphere, whatever that point is). Orthogonality of ∂~ r /∂u and ∂~r /∂v
implies that those u and v are orthogonal coordinates for the sphere. Compute the surface
element dS ~ for the sphere in terms of both the longitude-latitude parametrization (9) and
the longitude-colatitude parametrization ϕ, θ. Do the surface elements point toward or away
from the center of the sphere?

3. Explain why the surface described by ~ r (u, v) = ~

ex a cos u cos v+~
ey b sin u cos v+~
ez c sin v where
a, b and c are real constants is the surface of an ellipsoid. Are u and v orthogonal coordinates
for that surface? Consider cartesian coordinates (x, y, z) such that ~ r = x~ ex + y~ ey + z~
ez .
Derive the implicit equation f (x, y, z) = 0 satisfied by all such ~r (u, v)’s.

4. Consider the vector function ~ r (u, v) = ~rc + ~

e1 a cos u cos v + ~
e2 b sin u cos v + ~ e3 c sin v, where
~
rc , ~
e1 , ~
e2 , ~
e3 , a, b and c are constants with ~ ei · ~
ej = δij . What does the set of all such ~ r (u, v)’s
represent? Consider cartesian coordinates (x, y, z) such that ~ r = x~ ex + y~ ey + z~ ez . Assume
that all the angles between ~ e1 , ~ e3 and the basis vectors {~
e2 , ~ ex , ~ ez } are known. How
ey , ~
many independent angles is that? Specify which angles. Can you assume {~ e1 , ~ e3 } is right
e2 , ~
handed? Explain. Derive the implicit equation f (x, y, z) = 0 satisfied by all such ~ r (u, v)’s.
Express your answer in terms of the minimum independent angles that you specified earlier.

5. Explain why the surface of a torus (i.e. ‘donut’ or tire) of outer radius R and inner radius a
can be parametrized by x = (R + a cos θ) cos ϕ, y = (R + a cos θ) sin ϕ, z = a sin θ. Interpret
the geometric meaning of the parameters ϕ and θ. What are the ranges of ϕ and θ needed
to cover the entire torus? Do these parameters provide orthogonal coordinates for the torus?
~
Calculate the surface element dS.

1.4 Surface integrals

~ This represents the flux of ~
R
The typical surface integral is of the form S ~ v ·dS. v through the surface
~ is the time-rate at which
R
S. If ~
v (~
r ) is the velocity of a fluid, water or air, at point ~r , then S ~v · dS
volume of fluid is flowing through that surface per unit time. Indeed, ~ ~ = (~
v · dS v·~
n) dS where ~ n
is the unit normal to the surface and ~ v·~n is the component of fluid velocity that is perpendicular
to the surface. If that component is zero, the fluid moves tangentially to the surface, not through
the surface. Speed × area = volume per unit time, so ~ v · dS~ is the volume of fluid passing through
 F.
c Waleffe, Math 321, 2007/03/19 6

the surface element dS~ per unit time at that point at that time. The total volume passing through
~ Such integrals are often called flux integrals.
R
the entire surface S per unit time is S ~ v · dS.
We can make sense of such integrals as the limit of a sum, for instance by picking a series of points
on the surface, then forming a triangular meshing of the surface and evaluating the integral as the
limit of the sum of ~v evaluated at the center of such triangles dotted with the unit normal to the
triangle and times the area of the triangle (making sure of course that for each triangle we pick its
normal ~ n to correspond to the same side of the surface as the neighboring elements).
If a parametric representation ~r (u, v) is known for the surface then we can also write
Z Z   
~ ∂~r ∂~ r
v · dS =
~ v·
~ × dudv, (12)
S A ∂u ∂v
where A is the domain in the u, v parameter plane that corresponds to S.
As for line integrals, we can make sense of many other types of surface integrals such as
Z
p(~ ~
r ) dS,
S

which would represent the netRpressure force on S if p(~

r ) is the pressure at point ~
r . Other
R surface
~ ~
v × dS, etc. In particular the total area of surface S is S |dS|.
integrals could have the form S ~

Exercises:
1. Compute the area of the sphere and the area of a spherical cap (e.g. area of arctic cap).

2. Provide an explicit integral for the total surface area of the torus of outer radius R and inner
radius a.
~ where S is (i) the square 0 ≤ x, y ≤ a at z = b, (ii) the surface of the
R
3. Calculate S ~ r · dS
sphere of radius R centered at (0, 0, 0), (iii) the surface of the sphere of radius R centered at
x = x0 , y = z = 0.
~ where S is the surface of the sphere of radius R centered at the origin.
r /r3 ) · dS
R
4. Calculate S (~
How does the result depend on R?

5. The pressure outside the sphere of radius R centered at ~ r c is p(~ r ·~

r ) = p0 + A~ e where ~
e is a
fixed unit vector and p0 and A are constants. The pressure inside the sphere is the constant
p1 > p0 . Calculate the net force on the sphere. Calculate the net torque on the sphere about
its center ~
r c and about the origin.

1.5 Volumes and volume integrals

We have seen that ~ r (t) is the parametric equation of a curve, ~ r (u, v) represents a surface, now we
discuss ~
r (u, v, w) which is the parametric representation of a volume. Curves ~ r (t) are one dimen-
sional objects so they have only one parameter t or each point on the known curve is determined by
a single coordinate. Surfaces are two-dimensional and require two parameters u and v, which are
coordinates for points on that surface. Volumes are three-dimensional objects that require three
parameters u, v, w say. Each point is specified by three coordinates.
I A sphere of radius R centered at ~ r c has the implicit equation |~
r −~r c | ≤ R or (~
r −~ r c ) ≤ R2
r −~
r c )·(~
to avoid square roots. In cartesian coordinates this translates into the implicit equation

(x − xc )2 + (y − yc )2 + (z − zc )2 ≤ R2 . (13)
 F.
c Waleffe, Math 321, 2007/03/19 7

An explicit parametrization for that sphere is

~
r (u, v, w) = ~
rc + ~
e1 w cos u cos v + ~
e2 w sin u cos v + ~
e3 w sin v, (14)
or
~
r (r, θ, ϕ) = ~
rc + ~
e1 r sin θ cos ϕ + ~
e2 r sin θ sin ϕ + ~
e3 r cos θ, (15)
where {~e1 , ~ e3 } are any three orthonormal vectors such that ~
e2 , ~ ei ·~
ej = δij . For the parametrization
(14) we recognize u and v as the longitude and latitude parameters used earlier for spherical
surfaces (9) and note that w has units of length. From orthonormality of {~ e1 , ~ e3 } we deduce
e2 , ~
r −~
that (~ r c ) · (~
r −~ r c ) = w2 , so w can be interpreted as the distance from the center of the sphere.
To parametrize the entire sphere we need to consider all the u, v and w’s such that
π π
0 ≤ u < 2π, − ≤v≤ , 0 ≤ w ≤ R. (16)
2 2
The parametrization (15) is mathematically equivalent to this u, v, w parametrization but it is in
the standard form of spherical coordinates, with r representing the distance to the origin, ϕ the
azimuthal (or longitude) angle and θ the polar angle. To describe the full sphere we need
0 ≤ r ≤ R, 0 ≤ ϕ < 2π, 0 ≤ θ ≤ π. (17)

Coordinate curves
For a curve ~ r (t), all we needed to worry about was the tangent d~ r /dt and the line element d~r=
(d~r /dt)dt. For surfaces, ~r (u, v) we have two sets of coordinates curves with tangents ∂~ r /∂u and
∂~r /∂v, a normal N ~ = (∂~ r /∂u) × (∂~
r /∂v) and a surface element dS ~=N ~ dudv. Now for volumes
~
r (u, v, w), we have three sets of coordinates curves with tangents ∂~ r /∂u, ∂~
r /∂v and ∂~r /∂w. A
u-coordinate curve for instance, corresponds to ~ r (u, v, w) with v and w fixed. There is a double
infinity of such one dimensional curves, one for each v, w pair. For the parametrization (14), the
u-coordinate curves correspond to parallels, i.e. circles of fixed radius w at fixed latitude v. The
v-coordinate curves are meridians, i.e. circles of fixed radius w through the poles. The w-coordinate
curves are radial lines out of the origin.
Coordinate surfaces
For volumes ~r (u, v, w), we also have three sets of coordinate surfaces corresponding to one parameter
fixed and the other two free. A w-isosurface for instance corresponds to ~ r (u, v, w) for a fixed w.
There is a single infinity of such two dimensional (u, v) surfaces. For the parametrization (14) such
surfaces correspond to spherical surfaces of radius w centered at ~ r c . Likewise, if we fix u but let v
and w free, we get another surface, and v fixed with u and w free is another coordinate surface.
Line Elements
Thus given a volume parametrization ~ r (u, v, w) we can define four types of line elements, one for
each of the coordinate directions (∂~r /∂u)du, (∂~ r /∂v)dv, (∂~r /∂w)dw and a general line element
corresponding to the infinitesimal displacement from coordinates (u, v, w) to the coordinates (u +
du, v + dv, w + dw). That general line element d~ r is given by (chain rule):
∂~
r ∂~
r ∂~
r
d~
r= du + dv + dw. (18)
∂u ∂v ∂w
Surface Elements
Likewise, there are three basic types of surface elements, one for each coordinate surface. The
surface element on a w-isosurface, for example, is given by
 
~ w = ∂~
dS
r ∂~
×
r
dudv, (19)
∂u ∂v
 F.
c Waleffe, Math 321, 2007/03/19 8

while the surface elements on a u-isosurface and a v-isosurface are respectively

   
~ u = ∂~
dS
r
×
∂~r
dvdw, ~ v = ∂~
dS
r
×
∂~
r
dudw. (20)
∂v ∂w ∂w ∂u
Note that surface orientations are built into the order of the coordinates.
Volume Element
Last but not least, a parametrization ~r (u, v, w) defines a volume element given by the mixed (i.e.
triple scalar) product
   
r ∂~
∂~ r ∂~r r ∂~
∂~ r ∂~r
dV = × · du dv dw ≡ det , , du dv dw. (21)
∂u ∂v ∂w ∂u ∂v ∂w
The definition of volume integrals as limit-of-a-sum should be obvious by now. If an explicit
parametrization ~ r (u, v, w) for the volume is known, we can use the volume element (21) and write
the volume integral in ~ r space as an iterated triple integral over u, v, w. Be careful that there is
an orientation implicitly built into the ordering of the parameters, as should be obvious from the
definition of the mixed product and determinants. The volume element dV is usually meant to be
positive so the sign of the mixed product and the bounds of integrations for the parameters u, v
and w must be chosen to respect that. (Recall the definition of |dt| in the line integrals section).

Exercises
1. Calculate the line, surface and volume elements for the coordinates (14). You need to calculate
4 line elements and 3 surfaces elements. One line element for each coordinate curve and the
general line element. Verify that these coordinates are orthogonal.
2. Formulate integral expressions in terms of the coordinates (14) and (15) for the surface and
volume of a sphere of radius R. Calculate those integrals.
3. A curve ~ r (t) is given in terms of the (u, v, w) coordinates (14), i.e. ~ r (t) = ~
r (u, v, w) with
(u(t), v(t), w(t)) for t = ta to t = tb . Find an explicit expression in terms of (u(t), v(t), w(t))
as a t-integral for the length of that curve.
4. Find a parametrization for the torus of outer radius R and inner radius a. Compute the
volume of that torus.

1.6 Mappings, curvilinear coordinates

Parametrizations of curves, surfaces and volumes is essentially equivalent to the concepts of ‘map-
pings’ and curvilinear coordinates.

Mappings
The parametrization (9) for the surface of a sphere of radius R provides a mapping of that surface
to the 0 ≤ u < 2π, −π/2 ≤ v ≤ π/2 rectangle in the (u, v) plane. In a mapping ~ r (u, v) a small
rectangle of sides du, dv at a point (u, v) in the (u, v) plane is mapped to a small parallelogram of
sides (∂~
r /∂u)du, (∂~
r /∂v)dv at point ~r (u, v) in the Euclidean space.
The parametrization (14) for the sphere of radius R provides a mapping from the sphere of radius
R in Euclidean space to the box 0 ≤ u < 2π, −π/2 ≤ v ≤ π/2, 0 ≤ w ≤ R in the (u, v, w)
space. In a mapping ~ r (u, v, w), the infinitesimal box of sides du, dv, dw located at point (u, v, w)
in the (u, v, w) space is mapped to a parallelepiped of sides (∂~ r /∂u)du, (∂~r /∂v)dv, (∂~
r /∂w)dw at
the point ~r (u, v, w) in the Euclidean space.
 F.
c Waleffe, Math 321, 2007/03/19 9

Curvilinear coordinates, orthogonal coordinates

The parametrizations ~ r (u, v) and ~
r (u, v, w) define coordinates for a surface or a volume, respectively.
If the coordinate curves are not straight lines one talks of curvilinear coordinates. These mappings
define good coordinates if the coordinate curves intersect transversally i.e. if the coordinate curves
are not tangent to each other. If the coordinate curves intersect transversally at a point then
the coordinates tangent vectors at that point provide linearly independent directions in the space
of ~r . Tangent intersections at a point ~ r would imply that the tangent vectors are not linearly
independent at that point and that (∂~ r /∂u) × (∂~ r /∂v) = 0 at ~r (u, v) in the surface case or that
det(∂~ r /∂u, ∂~
r /∂v, ∂~
r /∂w) = 0 at that point ~ r (u, v, w) in the volume case.
The coordinates (u, v, w) are orthogonal if the coordinate curves in ~ r -space intersect at right angles.
This is the best kind of ‘transversal’ intersection and these are the most desirable type of coor-
dinates, however non-orthogonal coordinates are sometimes more convenient for some problems.
Two fundamental examples of orthogonal curvilinear coordinates are
I Cylindrical (or polar) coordinates

x = ρ cos θ, y = ρ sin θ, z = z. (22)

I Spherical coordinates

x = r sin θ cos ϕ, y = r sin θ sin ϕ, z = r cos θ. (23)

Changing notation from (x, y, z) to (x1 , x2 , x3 ) and from (u, v, w) to (q1 , q2 , q3 ) a general change of
coordinates from cartesian (x, y, z) ≡ (x1 , x2 , x3 ) to curvilinear (q1 , q2 , q3 ) coordinates is expressed
succinctly by
xi = xi (q1 , q2 , q3 ), i = 1, 2, 3. (24)
The position vector ~
r can be expressed in terms of the qj ’s through the cartesian expression:
3
X
~
r (q1 , q2 , q3 ) = ~
ex x(q1 , q2 , q3 ) + ~
ey y(q1 , q2 , q3 ) + ~
ez z(q1 , q2 , q3 ) = ~
ei xi (q1 , q2 , q3 ), (25)
i=1

where ~ei · ~
ej = δij . The qi coordinate curve is the curve ~r (q1 , q2 , q3 ) where qi is free but the other
two variables are fixed. The qi isosurface is the surface ~r (q1 , q2 , q3 ) where qi is fixed and the other
two parameters are free.
The coordinate tangent vectors ∂~ r /∂qi are key to the coordinates. They provide a natural vector
basis for those coordinates. The coordinates are orthogonal if these tangent vectors are orthogonal
to each other at each point. In that case it is useful to define the unit vector q̂ i in the qi coordinate
direction by  
 ∂~
r  ∂~r
hi = 
 , = hi q̂ i , (26)
∂qi  ∂qi
where hi is the the magnitude of the tangent vector in the qi direction, ∂~ r /∂qi , and q̂ i · q̂ j = δij
for orthogonal coordinates. These hi ’s are called the scale factors. The distance traveled in x-
space when changing qi by dqi , keeping the other q’s fixed, is |d~
r | = hi dqi (no summation). The
distance ds travelled in x-space when the orthogonal curvilinear coordinates change from (q1 , q2 , q3 )
to (q1 + dq1 , q2 + dq2 , q3 + dq3 ) is

ds2 = d~ r = h21 dq12 + h22 dq22 + h23 dq32 .

r · d~ (27)
 F.
c Waleffe, Math 321, 2007/03/19 10

Although the cartesian unit vectors ~ ei are independent of the coordinates, the curvilinear unit vec-
tors q̂ i in general are functions of the coordinates, even if the latter are orthogonal. Hence ∂ q̂ i /∂qj
is in general non-zero. For orthogonal coordinates, those derivatives ∂ q̂ i /∂qj can be expressed in
terms of the scale factors and the unit vectors.
For instance, for spherical coordinates (q1 , q2 , q3 ) ≡ (r, ϕ, θ), the unit vector in the q1 ≡ r direction
is the vector
∂~
r (r, ϕ, θ)
=~
ex sin θ cos ϕ + ~
ey sin θ sin ϕ + ~ez cos θ, (28)
∂r
so the scale factor h1 ≡ hr = 1 and the unit vector
q̂ 1 ≡ r̂ ≡ ~
er = ~
ex sin θ cos ϕ + ~
ey sin θ sin ϕ + ~
ez cos θ. (29)
The position vector ~
r can be expressed as
3
X
~
r= ~ er r ≡ r̂ r,
ei xi = ~ (30)
i=1

so its expression in terms of spherical coordinates and their unit vectors is simpler than the
corresponding expression in cartesian coordinates, but there is a catch! The radial unit vector
r̂ ≡ ~ er = ~ er (ϕ, θ) varies in the azimuthal and polar angle directions, while the cartesian unit
vectors ~ex , ~
ey , ~
ez are independent of the coordinates!
For orthogonal coordinates, the scale factors hi ’s determine everything. In particular, the surface
and volume elements can be expressed in terms of the hi ’s. For instance, the surface element for a
q3 -isosurface is
~ 3 = q̂ 3 h1 h2 dq1 dq2 ,
dS (31)
and the volume element
dV = h1 h2 h3 dq1 dq2 dq3 , (32)
assuming that q1 , q2 , q3 is right-handed. These follow directly from (19) and (21) and (26) when
the coordinates are orthogonal.

Exercises
1. Find the scale factors hi and the unit vectors q̂ i for cylindrical and spherical coordinates.
Express the 3 surface elements and the volume element in terms of those scale factors and
unit vectors. Sketch the unit vector q̂ i in the (x, y, z) space (use several ‘views’ rather than
trying to make an ugly 3D sketch!). Express the position vector ~ r in terms of the unit vectors
q̂ i . Calculate the derivatives ∂ q̂ i /∂qj for all i, j and express these derivatives in terms of the
scale factors hk and the unit vectors q̂ k , k = 1, 2, 3.
2. A curve in the (x, y) plane is given in terms of polar coordinates as ρ = ρ(θ). Deduce θ-integral
expressions for the length of the curve and for the area swept by the radial vector.
3. Consider elliptical coordinates (u, v, w) defined by x = α cosh u cos v, y = α sinh u sin v , z = z
for some α > 0, where x, y and z are standard cartesian coordinates in 3D Euclidean space.
What do the coordinate curves correspond to in the (x, y, z) space? Are these orthogonal
coordinates? What is the volume element in terms of elliptical coordinates?
4. For general curvilinear coordinates, not necessarily orthogonal, is the qi -isosurface perpendic-
ular to ∂~ r /∂qj ) × (∂~
r /∂qi ? is it orthogonal to (∂~ r /∂qk ) where i, j, k are all distinct? What
about for orthogonal coordinates?
 F.
c Waleffe, Math 321, 2007/03/19 11

1.7 Change of variables

Parametrizations of surfaces and volumes and curvilinear coordinates are geometric examples of a
change of variables. These change of variables and the associated formula and geometric concepts
can occur in other non-geometric contexts. The fundamental relationship is the formula for a
volume element (21). In the context of a general change of variables from (x1 , x2 , x3 ) to (q1 , q2 , q3 )
that formula (21) reads
dx1 dx2 dx3 = dVx = J dq1 dq2 dq3 = J dVq (33)
where  
 ∂x1 /∂q1 ∂x1 /∂q2 ∂x1 /∂q3   
  ∂xi
J =  ∂x2 /∂q1 ∂x2 /∂q2 ∂x2 /∂q3  = det (34)
 ∂x3 /∂q1 ∂x3 /∂q2 ∂x3 /∂q3  ∂qj

is the Jacobian determinant and dVx is a volume element in the x-space while dVq is the corre-
sponding volume element in q-space. The Jacobian is the determinant of the Jacobian matrix
 
∂x1 /∂q1 ∂x1 /∂q2 ∂x1 /∂q3
∂xi
J =  ∂x2 /∂q1 ∂x2 /∂q2 ∂x2 /∂q3  ⇔ Jij = . (35)
∂qj
∂x3 /∂q1 ∂x3 /∂q2 ∂x3 /∂q3

If q̂ 1 dq1 , q̂ 2 dq2 and q̂ 3 dq3 are three orthogonal infinitesimal vectors at point (q1 , q2 , q3 ) in q-space
then these vectors are mapped to the vectors (∂~ r /∂q1 )dq1 , (∂~
r /∂q2 )dq2 , (∂~
r /∂q3 )dq3 . In component
form this is  (1)    
dx1 ∂x1 /∂q1 ∂x1 /∂q2 ∂x1 /∂q3 dq1
 (1)  
dx2  = ∂x2 /∂q1 ∂x2 /∂q2 ∂x2 /∂q3   0 , (36)
(1) ∂x /∂q ∂x /∂q ∂x /∂q 0
3 dx 3 1 3 2 3 3

(1) (1) (1)

where (dx1 , dx2 , dx3 ) are the x-components of the vector (∂~ r /∂q1 )dq1 . Similar relations hold
for the other basis vectors. Note that the rectangular box in q-space is in general mapped to a
non-rectangular parallelepiped in x-space so the notation dx1 dx2 dx3 for the volume element in (33)
is a (common) abuse of notation.
The formulas (33), (34) tells us how to change variables in multiple integrals. This formula gen-
eralizes to higher dimension, and also to lower dimension. In the 2 variable case, we have a
2-by-2 determinant that can also be understood as a special case of the surface element formula
(11) for a mapping ~ r (q1 , q2 ) from a 2D space (q1 , q2 ) to another 2D-space (x1 , x2 ). In that case
~
r (q1 , q2 ) = ~
e1 x1 (q1 , q2 ) + ~ r /∂q1 ) × (∂~
e2 x2 (q1 , q2 ) so (∂~ r /∂q2 )dq1 dq2 = ~
e3 dAx and

dx1 dx2 = dAx = J dq1 dq2 = J dAq (37)

where the Jacobian determinant is now

 
 ∂x1 /∂q1 ∂x1 /∂q2 
J =  . (38)
∂x2 /∂q1 ∂x2 /∂q2 

1.7.1 Change of variables example

Consider a Carnot cycle for a perfect gas. The equation of state is P V = nRT = N kT , where P
is pressure, V is the volume of gas and T is the temperature in Kelvins. The volume V contains
n moles of gas corresponding to N molecules, R is the gas constant and k is Boltzmann’s constant
with nR = N k. A Carnot cycle is an idealized thermodynamic cycle in which a gas goes through (1)
a heated isothermal expansion at temperature T1 , (2) an adiabatic expansion at constant entropy
 F.
c Waleffe, Math 321, 2007/03/19 12

S2 , (3) an isothermal compression releasing heat at temperature T3 < T1 and (4) an adiabatic
compression at constant entropy S4 < S2 . For a perfect monoatomic gas, constant entropy means
constant P V γ where γ = CP /CV = 5/3 with CP and CV the heat capacity at constant pressure P
or constant volume V , respectively. Thus let S = P V γ (this S is not the physical entropy but it is
constant whenever entropy is constant, we can call S a ‘pseudo-entropy’).

T1
T1
S4
P T S4 S2
S2

T3 T3

V S
RV
Now the work done by the gas when its volume changes from Va to Vb is Vab P dV (since work =
Force × displacement, P = force/area and V =area × displacement). Thus the (yellow) area inside
the cycle in the (P, V ) plane is the net work performed by the gas during one cycle. Although we
can calculate that area by working in the (P, V ) plane, it is easier to calculate it by using a change
of variables from P , V to T , S. The area inside the cycle in the (P, V ) plane is not the same as the
area inside the cycle in the (S, T ) plane. There is a distortion. An element of area in the (P, V )
plane aligned with the T and S coordinates (i.e. with the dashed curves in the (P, V ) plane) is
    
 ∂P ∂V ∂P ∂V 
dA =   ~
eP + eV dS ×
~ ~
eP + ~
eV dT  (39)
∂S ∂S ∂T ∂T
where ~eP and ~ eV are the unit vectors in the P and V directions in the (P, V ) plane, respectively.
~ = ∂~
r ∂~
r
This is entirely similar to the area element of surface ~
r (u, v) being equal to dS ∂u du × ∂v dv (but
don’t confuse the surface element dS ~ with the pseudo-entropy differential dS used in the present
example!). Calculating out the cross product, we obtain
  
 ∂P ∂V ∂V ∂P  P,V
dA =  − dSdT  = ± JS,T dSdT (40)
∂S ∂T ∂S ∂T
where the ± sign will be chosen to get a positive area (this depends on the bounds of integrations)
and  
 ∂P ∂P 
 
P,V P,V
 ∂S ∂T 
JS,T = det(J S,T ) =  (41)
 

 ∂V ∂V 
 
∂S ∂T
 
is the Jacobian determinant (here the vertical bars are the common notation for determinants). It
is the determinant of the Jacobian matrix J P,VS,T that corresponds to the mapping from (S, T ) to
(P, V ). The cycle area AP,V in the (P, V ) plane is thus
Z T1 Z S2  
 P,V 
AP,V = JS,T  dSdT. (42)
T3 S4
 F.
c Waleffe, Math 321, 2007/03/19 13

P,V
Note that the vertical bars in this formula are for absolute value of JS,T and the bounds have
been selected so that dS > 0 and dT > 0 (in the limit-of-a-sum sense). This expression for the
(P, V ) area expressed
  in terms of (S, T ) coordinates is simpler than if we used (P, V ) coordinates,
 P,V 
except for that JS,T  since we do not have explicit expression for P (S, T ) and V (S, T ). What we
have in fact are the inverse functions T = P V /(N k) and S = P V γ . To find the partial derivatives
that we need we could (1) find the inverse functions by solving for P and V in terms of T and
S then compute the partial derivatives and the Jacobian, or (2) use implicit differentiation e.g.
N k ∂T /∂T = N k = V ∂P/∂T + P ∂V /∂T and ∂S/∂T = 0 = V γ ∂P/∂T + P γV γ−1 ∂V /∂T etc. and
solve for the partial derivatives we need. But there is a simpler way that makes use of an important
property of Jacobians.

1.7.2 Geometric meaning of the Jacobian determinant and its inverse

P,V
The Jacobian JS,T represents the stretching factor of area elements when moving from the (S, T )
plane to the (P, V ) plane. If dAS,T is an area element centered at point (S, T ) in the (S, T ) plane
then that area element gets mapped to an area element dAP,V centered at the corresponding point
in the (P, V ) plane. That’s what equation (40) represents. In that equation we have in mind the
mapping of a rectangular element of area dSdT in the (S, T ) plane to a parallelogram element
P,V
in the (P, V ) plane. The stretching factor is |JS,T | (as we saw earlier, the meaning of the sign is
related to orientation, but here we are worrying only about areas, so we take absolute values). That
relationship is valid for area elements of any shape, not just rectangles to parallelogram since the
differential relationships implies an implicit limit-of-a-sum and in that limit, the ‘pointwise’ area
stretching is independent of the shape of the area elements. A disk element in the (S, T ) plane
would be mapped to an ellipse element in the (P, V ) plane but the pointwise area stretching would
be the same as for a rectangular element (this is not true for finite size areas). So equation (40)
can be written in the more general form
P,V
dAP,V = |JS,T | dAS,T (43)

which is locally valid for area elements of any shape. The key point is that if we consider the
inverse map, back from (P, V ) to (S, T ) then there is an are stretching give by the Jacobian
S,T
JP,V = (∂S/∂P )(∂T /∂V ) − (∂S/∂V )(∂T /∂P ) such that
S,T
dAS,T = |JP,V | dAP,V (44)

but since we are coming back to the original dAS,T element we must have

P,V S,T
JS,T JP,V = 1, (45)

so the Jacobian determinant are inverses of one another. This inverse relationship actually holds
for the Jacobian matrices also  
P,V S,T 1 0
J S,T J P,V = . (46)
0 1

The latter can be derived from the chain rule since (note the consistent ordering of the partials)
   
P,V ∂P/∂S ∂P/∂T S,T ∂S/∂P ∂S/∂V
J S,T = , J P,V = (47)
∂V /∂S ∂V /∂T ∂T /∂P ∂T /∂V
 F.
c Waleffe, Math 321, 2007/03/19 14

and the matrix product of those two Jacobian matrices yields the identity matrix. For instance,
the first row times the first column gives
         
∂P ∂S ∂P ∂T ∂P
+ = = 1.
∂S T ∂P V ∂T S ∂P V ∂P V

A subscript has been added to remind which other variable is held fixed during the partial differ-
entiation. The inverse relationship between the Jacobian determinants (45) then follows from the
inverse relationship between the Jacobian matrices (46) since the determinant of a product is the
product of the determinants. This important property of determinants can be verified directly by
explicit calculation for this 2-by-2 case.
So what’s the work done by the gas during one Carnot cycle? Well,
−1  −1
P,V
 S,T ∂S ∂T ∂S ∂T
−1 Nk
JS,T = JP,V = − = N k V γ P − γV γ−1 P V = (48)
∂P ∂V ∂V ∂P (1 − γ)S
so
T1 S2 T1 S2
N k(T1 − T3 ) S2
Z Z Z Z
 P,V  Nk
AP,V = JS,T  dSdT = dSdT = ln , (49)
 
T3 S4 T3 S4 (γ − 1)S (γ − 1) S4
since γ > 1 and other quantities are positive.

Exercises
1. Calculate the area between the curves xy = α1 , xy = α2 and y = β1 x, y = β2 x in the (x, y)
plane. Sketch the area. (α1 , α2 , β1 , β2 > 0.)

2. Calculate the area between the curves xy = α1 , xy = α2 and y 2 = 2β1 x, y 2 = 2β2 x in the
(x, y) plane. Sketch the area. (α1 , α2 , β1 , β2 > 0.)

3. Calculate the area between the curves x2 + y 2 = 2α1 x, x2 + y 2 = 2α2 x and x2 + y 2 = 2β1 y,
x2 + y 2 = 2β2 y. Sketch the area. (α1 , α2 , β1 , β2 > 0.)

4. Calculate the area of the ellipse x2 /a2 + y 2 /b2 = 1 and the volume of the ellipsoid x2 /a2 +
y 2 /b2 + z 2 /c2 = 1 by transforming them to a disk and a sphere, respectively, using a change
of variables. [Hint: consider the change of variables x = au, y = bv, z = cw.]
R∞ R∞ 2 2
5. Calculate the integral −∞ −∞ e−(x +y ) dxdy. Deduce the value of the Poisson integral
R ∞ −x2
−∞ e dx. [Hint: switch to polar coordinates].
R∞ R∞
6. Calculate −∞ −∞ (a2 + x2 + y 2 )α dxdy. Where a 6= 0 and α is real. Discuss the values of α
for which the integral exists.
 F.
c Waleffe, Math 321, 2007/03/19 15

2 Grad, div, curl

Consider a scalar function of a vector variable: f (~ r ), for instance the pressure p(~ r ) as a function
of position, or the temperature T (~ r ) at point ~
r , etc. One way to visualize such functions is to
consider isosurfaces or level sets, these are the set of all ~ r ’s for which f (~
r ) = C0 , for some constant
C0 . In cartesian coordinates ~ r = x~ e + y~
ey + z ~ ez and the scalar function is a function of the
three coordinates f (~r ) ≡ f (x, y, z), hence we can interpret an isosurface f (x, y, z) = C0 as a single
equation for the 3 unknowns x, y, z. In general, we are free to pick two of those variables, x and y
say, then solve the equation f (x, y, z) = C0 for z. For example, the isosurfaces √ of f (~ r ) = x2 +y 2 +z 2
are determined by the equation f (~ r ) = C0 . This is the sphere or radius C0 , if C0 ≥ 0.

2.1 Geometric definition of the Gradient

The value of f (~ r ) at a point ~
r 0 defines an isosurface f (~
r ) = f (~
r 0 ) through that point ~r 0 . The
gradient of f (~ r 0 can be defined geometrically as the vector, denoted ∇f
r ) at ~ ~ (~
r 0 ), that

(i) is perpendicular to the isosurface f (~

r ) = f (~
r 0 ) at the point ~
r 0 and points in the direction of
increase of f (~
r ) and

(ii) has a magnitude equal to the rate of change of f (~

r ) with distance from the isosurface.

From this geometric definition, we deduce that the plane tangent to the isosurface f (~
r ) = f (~
r 0 ) at
~
r 0 has the equation
r−~
(~ ~ (~
r 0 ) · ∇f r0 ) = 0 (50)
Likewise, the plane (~ r−~ ~ (~
r 0 ) · ∇f r 0 ) =  is parallel to the tangent plane but further ‘up’ in the
direction of the gradient if  > 0 or ‘down’ if  < 0. More generally, the function f (~r ) can be locally
(i.e. for ~
r near ~
r 0 ) approximated by

r ) ≈ f (~
f (~ r−~
r 0 ) + (~ ~ (~
r 0 ) · ∇f r 0 ). (51)

Indeed since ∇f ~ points in the direction of fastest increase of f (~ r ) and has a magnitude equal to
the rate of change of f with distance in that direction, the change in f as we move slightly away
from ~ r 0 only depends on how much we have moved in the direction of the gradient. The actual
change in f is the distance times the rate of change with distance, hence it is (~ r−~ ~ (~
r 0 ) · ∇f r 0 ).
Equation (51) states that the isosurfaces of f (~ r ) look like planes perpendicular to ∇f (~ ~ r 0 ) for ~
r in
the neighborhood for ~ r 0 . Equation (51) is a linear approximation of f (~ r ) in the neighborhood of
r 0 , just like f (x) ≈ f (x0 ) + (x − x0 )f 0 (x0 ) for functions of one variable. This is not exact, there is
~
a small error that goes to zero faster than |~ r−~ r 0 | as ~
r→~ r 0 . The limit as ~r→~ r 0 can be written
as the exact differential relation
df (~
r ) = d~ ~ (~
r · ∇f r ), (52)
where df (~r ) = f (~ r ) − f (~
r + d~ r ) is the differential change in f (~ r ) when ~
r changes from ~ r to ~
r + d~
r.
This differential relationship holds for any ~ r and d~ r . It is analogous to the differential relationship
df = f 0 (x) dx for functions of a single variable.
I It follows immediately from the geometric definition of the gradient that if r = |~ r | is the distance
~ = r̂ and more generally ∇f
r /r is the unit radial vector, then ∇r
to the origin and r̂ = ~ ~ (r) = r̂df /dr.
~
For instance, ∇(1/r) = −r̂/r2 . 
 F.
c Waleffe, Math 321, 2007/03/19 16

2.2 ~ operator
Directional derivative, gradient and the ∇
The rate of change of f (~
r ) in the direction of the unit vector ~ n at point ~ r , denoted ∂f (~
r )/∂n is
defined as
∂f (~r) f (~
r + h~n) − f (~ r)
= lim . (53)
∂n h→0 h
From (51) and (52),
∂f (~
r) ~ (~
=~n · ∇f r ). (54)
∂n
This is exact because the error of the linear approximation (51), f (~ r + h~n) ≈ f (~
r ) + h~ ~ (~
n · ∇f r ),
goes to zero faster than h as h → 0.
In particular,
∂f ~ ∂f ~ ∂f ~
=~ ex · ∇f, =~ey · ∇f, =~ ez · ∇f, (55)
∂x ∂y ∂z
hence
~ =~ ∂f ∂f ∂f
∇f ex +~
ey +~
ez , (56)
∂x ∂y ∂z
for cartesian coordinates x, y, z. We can also deduce this important result from the Chain rule
for functions of several variables. In cartesian coordinates, written (x1 , x2 , x3 ) in place of (x, y, z),
the position vector reads ~r = x1~e1 + x2~e2 + x3~e3 and the scalar function f (~ r ) ≡ f (x1 , x2 , x3 ) is a
function of the 3 coordinates. The directional derivative
∂f (~
r) f (~ n) − f (~
r + h~ r) ∂f ∂f ∂f ~
= lim = n1 + n2 + n3 n · ∇f.
=~ (57)
∂n h→0 h ∂x1 ∂x2 ∂x3
This follows from rewriting the difference f (~ n) − f (~
r + h~ r ) as the telescoping sum

[f (x1 + hn1 , x2 + hn2 , x3 + hn1 )−f (x1 , x2 + hn2 , x3 + hn3 )]

+[f (x1 , x2 + hn2 , x3 + hn3 )−f (x1 , x2 , x3 + hn3 )]
+[f (x1 , x2 , x3 + hn3 )−f (x1 , x2 , x3 )] (58)

then using continuity and the definition of the partial derivatives, e.g.
f (~ e3 ) − f (~
r + hn3~ r) f (~ e3 ) − f (~
r + δ~ r) ∂f
lim = n3 lim = n3 .
h→0 h δ→0 δ ∂x3
Either way, we establish that in cartesian coordinates

~ =~ ∂ ∂ ∂
∇ e1 +~
e2 +~
e3 ≡ ~
ei ∂i (59)
∂x1 ∂x2 ∂x3

where ∂i is short for ∂/∂xi and we have used the convention of summation over all values of the
repeated index i.

2.3 Div and Curl

We’ll depart from our geometric point of view to first define divergence and curl computationally
based on their cartesian representation. Here we consider vector fields ~ v (~
r ) which are vector
functions of a vector variable, for example the velocity ~
v (~
r ) of a fluid at point ~r , or the electric
~ r ) at point ~
field E(~ r , etc. For the geometric meaning of divergence and curl, see the sections on
divergence and Stokes’ theorems.
 F.
c Waleffe, Math 321, 2007/03/19 17

~ v . Now since the unit vectors

The divergence of a vector field ~
v (~
r ) is defined as the dot
∇ ·~

product
~ ~
ei are constant for cartesian coordinates, ∇ · ~ v= ~ ei ∂i · ~ ei · ~
ej vj = (~ ej ) ∂i vj = δij ∂i vj hence

~ ·~
∇ v = ∂i vi = ∂1 v1 + ∂2 v2 + ∂3 v3 . (60)

Likewise, the curl of a vector field ~ r ) is the cross product ∇

v (~ ~ ×~ v . In cartesian coordinates,
~ ×~
∇ ej ∂j ) × (~
v = (~ ej × ~
ek vk ) = (~ ek ) ∂j vk . Recall that ijk = ~ei · (~
ej × ~
ek ), or in other words ijk is
the i component of the vector ~ ej × ~ek , thus ~ ej × ~
ek = ~
ei ijk and

~ ×~
∇ v=~
ei ijk ∂j vk . (61)

(Recall that ~a × ~b = ~
ei ijk aj bk .) The right hand side is a triple sum over all values of the repeated
indices i, j and k! But that triple sum is not too bad since ijk = ±1 depending on whether (i, j, k)
is a cyclic (=even) permutation or an acyclic (=odd) permutation of (1, 2, 3) and vanishes in all
other instances. Thus (61) expands to
~ ×~
∇ e1 (∂2 v3 − ∂3 v2 ) + ~
v=~ e2 (∂3 v1 − ∂1 v3 ) + ~
e3 (∂1 v2 − ∂2 v1 ) . (62)

We can also write that the i-th cartesian component of the curl is

~ ×~ ~ ×~



ei · ∇
~ v = ∇ v i = ijk ∂j vk . (63)

Note that the divergence is a scalar but the curl is a vector.

2.4 Vector identities

In the following f = f (~r ) is an arbitrary scalar field while ~v (~
r ) and w(~
~ r ) are vector fields. Two
crucial relationships that follow easily from ~ a × ~b) = 0 and ~
a · (~ a × (α~
a) = 0 are

~ · (∇
∇ ~ ×~
v ) = 0, (64)

and
~ × (∇f
∇ ~ ) = 0. (65)
The divergence of a curl and the curl of a gradient vanish identically (assuming all those derivatives
exist). These can be verified using indicial notation. Other useful identities are

~ · (f~
∇ ~ )·~
v ) = (∇f ~ ·~
v + f (∇ v ), (66)
~ × (f~
∇ ~ )×~
v ) = (∇f ~ ×~
v + f (∇ v ), (67)
~ × (∇
∇ ~ ×~ ~ ∇
v ) = ∇( ~ ·~
v ) − ∇2 ~
v, (68)
where ∇2 = ∇ ~ ·∇
~ = ∂ 2 + ∂ 2 + ∂ 2 is the Laplacian operator.
1 2 3
The identity (66) is verified easily using indicial notation
~ · (f~
∇ ~ )·~
v ) = ∂i (f vi ) = (∂i f )vi + f (∂i vi ) = (∇f ~ ·~
v + f (∇ v ).

Likewise the second identity (67) follows from ijk ∂j (f vk ) = ijk (∂j f )vk + f ijk (∂j vk ). The identity
(68) is easily remembered from the double cross product formula ~ a × (~b × ~ c) = ~b(~ a·~ c) − ~ a · ~b)
c(~
~ ~
but note that the ∇ in the first term must appear first since ∇(∇ · ~ ~ ~
v ) 6= (∇ · ~ ~
v )∇. That identity
 F.
c Waleffe, Math 321, 2007/03/19 18

can be verified using indicial notation if one knows the double cross product identity in terms of
the permutation tensor (see earlier notes on ijk and index notation) namely

ijk klm = kij klm = δil δjm − δim δjl . (69)

A slightly trickier identity is

~ × (~
∇ v × w) ~ v − (∇
~ · ∇)~
~ = (w ~ ·~ ~ · w)~
~ + (∇
v )w ~ v − (~ ~ w,
v · ∇) ~ (70)
where (w ~ v = (wj ∂j )vi in indicial notation. This can be verified using (69) and can be
~ · ∇)~
remembered from the double cross-product identity with the additional input that ∇ ~ is a vector
~
operator, not just a regular vector, hence ∇ × (~ v × w)
~ represents derivatives of a product and this
doubles the number of terms of the resulting expression. The first two terms are the double cross
product ~a × (~b × ~
c) = (~ c)~b − (~
a ·~ a · ~b)~
c for derivatives of ~
v while the last two terms are the double
cross products for derivatives of w.~
Another similar identity is
~ ~ × w)
v × (∇ ~ w)
~ = (∇ ~ ·~
v − (~ ~ w.
v · ∇) ~ (71)
In indicial notation this reads
ijk vj (klm ∂l wm ) = (∂i wj )vj − (vj ∂j )wi . (72)
~ w.
Note that this last identity involves the gradient of a vector field ∇ ~ This makes sense and is a
tensor, i.e. a geometric object whose components with respect to a basis form a matrix. In indicial
notation, the components of ∇ ~w~ are ∂i wj and there are 9 of them. This is very different from
~ ·w
∇ ~ = ∂i wi which is a scalar.
The bottom line is that these identities can be reconstructed relatively easily from our knowledge
of regular vector identities for dot, cross and double cross products, however ∇ ~ is a vector of
derivatives and one needs to watch out more carefully for the order and the product rules. If in
doubt, jump to indicial notation.

Exercises
1. Verify (64) and (65).
2. Digest and verify the identity (69) ab initio.
3. Verify the double cross product identities ~a × (~b × ~
c) = ~b(~
a·~
c) − ~ a · ~b) and (~
c(~ a × ~b) × ~
c=
~b(~
a·~
c) − ~ ~
a(b · ~
c) in indicial notation using (69).
4. Verify (70) and (71) using indicial notation and (69).
a × ~b) · (~
5. Use (69) to derive vector identities for (~ ~ and (∇
c × d) ~ ×~ ~ × w).
v ) · (∇ ~
6. Show that ∇ ~ · (~
v × w)
~ =w ~ ×~
~ · (∇ v) − ~ ~ × w)
v · (∇ ~ and explain how to reconstruct this from
the rules for the mixed (or box) product ~a · (~b × ~
c) of three regular vectors.
~ ·~
7. Find the fastest way to show that ∇ ~ ×~
r = 3 and ∇ r = 0.
~ · (r̂/r2 ) = 0 and ∇
8. Find the fastest way to show that ∇ ~ × (r̂/r2 ) = 0 for all ~
r 6= 0.
9. Quickly calculate ∇~ ·B~ and ∇~ ×B ~ when B ~ = (~ez × ~ r |2 (cf. Biot-Savart law for a
ez × ~
r )/|~
line current) [Hint: use both vector identities and cartesian coordinates where convenient].
 F.
c Waleffe, Math 321, 2007/03/19 19

3 Fundamental theorems of vector calculus

3.1 Integration in R2 and R3
of two variables f (x, y) over a domain A of R2 denoted A f (x, y)dA can be
R
The integral of a function P
defined as a the limit of a n f (xn , yn )∆An where the An ’s, n = 1, . . . , N provide an (approximate)
partition of A that breaks up A into a set of small area elements, squares or triangles for instance.
∆An is the area of those element n and (xn , yn ) is a point inside that element, for instance the
center of area of the triangle. The integral would be the limit of such sums when the area of the
triangles goes to zero and their number N must then go to infinity. This limit should be such that
the aspect ratios of the triangles remain bounded away from 0 so we get a finer and finer sampling
of A. This definition also provides a way to approximate the integral by such a finite sum.

If we imagine breaking up A into small squares

yT aligned with the x and y axes then the sum over all
A squares inside A can be performed row by row. Each
row-sum then tends to an integral in the x-direction,
y this leads to the conclusion that the integral can be
calculated as iterated integrals
Z Z yT Z xr (y)
yB f (x, y)dA = dy f (x, y)dx (73)
A yB x` (y)

x` (y) xr (y)

We can also imagine summing up column by col-

umn instead and each column-sum then tends to
an integral in the y-direction, this leads to the yt (x) A
iterated integrals
Z Z xR Z yt (x)
f (x, y)dA = dx f (x, y)dy. (74)
A xL yb (x)
yb (x)
Note of course that the limits of integrations dif-
fer from those of the previous iterated integrals.
xL x xR
This iterated integral approach readily extends to integrals over three-dimensional domains in R3
and more generally to integrals in Rn .

3.2 Fundamental theorem of Calculus

The fundamental theorem of calculus can be written
Z b
dF
dx = F (b) − F (a). (75)
a dx

Once again we can interpret this in terms of limits of finite differences. The derivative is defined as
dF ∆F
= lim (76)
dx ∆x→0 ∆x
 F.
c Waleffe, Math 321, 2007/03/19 20

where ∆F = F (x + ∆x) − F (x), while the integral

Z b N
X
f (x)dx = lim f (x̃n )∆xn (77)
a ∆xn →0
n=1

where ∆xn = xn − xn−1 and xn−1 ≤ x̃n ≤ xn , with n = 1, . . . , N and x0 = a, xN = b, so

the set of xn ’s provides a partition of the interval [a, b]. The best choice for x̃n is the midpoint
x̃n = (xn + xn−1 )/2. This is the midpoint scheme in numerical integration methods. Putting these
two limits together and setting ∆Fn = F (xn ) − F (xn−1 ) we can write
Z b N N
dF X ∆Fn X
dx = lim ∆xn = lim ∆Fn = F (b) − F (a). (78)
a dx ∆xn →0 ∆xn ∆xn →0
n=1 n=1

We can also write this in the integral form

Z b Z F (b)
dF
dx = dF = F (b) − F (a). (79)
a dx F (a)

3.3 Fundamental theorem in R2

From the fundamental theorem of calculus and the reduction of integrals on a domain A of R2 to
iterated integrals on intervals in R we obtain for a function F (x, y)
Z Z yT Z xr (y) Z yT
∂F ∂F  
dA = dy dx = F (xr (y), y) − F (x` (y), y) dy. (80)
A ∂x yB x` (y) ∂x yB

This looks nice enough but we can rewrite the integral on the right-hand side as a line integral over
the boundary of A. The boundary of A is a closed curve C often denoted ∂A (not to be confused
with a partial derivative). The boundary C has two parts C1 and C2 .
The curve C1 can be parametrized in terms of y as
yT ~
r (y) = ~ ey y with y = yB → yT , hence
ex xr (y) + ~
Z Z yT
ey · d~
F (x, y)~ r= F (xr (y), y)dy.
A C y
C2 1 B
y
C1 Likewise, the curve C2 can be parametrized using y as
~
r (y) = ~ ey y with y = yT → yB , hence
ex x` (y) + ~
yB Z Z yB
ey · d~
F (x, y)~ r= F (x` (y), y)dy.
C2 yT
x` (y) xr (y)
Putting these two results together the right hand side of (80) becomes
Z yT Z Z I
 
F (xr (y), y) − F (x` (y), y) dy = ey · d~
F~ r+ F~ey · d~ ey · d~
r = F~ r,
yB C1 C2 C

where C = C1 + C2 is the closed curve bounding A. Then (80) becomes

Z I
∂F
dA = F ~ ey · d~
r. (81)
A ∂x C
 F.
c Waleffe, Math 321, 2007/03/19 21

H
The symbol is used to emphasize that the integral is over a closed curve. Note that the curve C
has been oriented counter-clockwise such that the interior is to the left of the curve.
Similarly the fundamental theorem of calculus and iterated integrals lead to the result that
Z Z xR Z yt (x) Z xR
∂F ∂F  
dA = dx dy = F (x, yt (x)) − F (x, yb (x)) dx, (82)
A ∂y xL yb (x) ∂y xL

and the integral on the right hand side can be rewritten as a line integral around the boundary
curve C = C3 + C4 .
The curve C3 can be parametrized in terms of x as
~
r (x) = ~ ey yb (x) with x = xL → xR , hence
ex x + ~
C4 Z Z xR
yt (x)
ex · d~
F (x, y)~ r= F (x, yb (x))dx.
A C3 xL

Likewise, the curve C4 can be parametrized using x

as ~
r (x) = ~ ey yt (x) with x = xR → xL , hence
ex x + ~
yb (x) Z xL
C3
Z
ex · d~
F (x, y)~ r= F (x, yt (x))dx.
C4 xR
xL x xR
The right hand side of (82) becomes
Z xR Z Z I
 
F (x, yt (x)) − F (x, yb (x)) dx = − ex · d~
F~ r− ex · d~
F~ r=− ex · d~
F~ r,
xL C4 C3 C

where C = C3 + C4 is the closed curve bounding A oriented counter-clockwise as before. Then (82)
becomes Z I
∂F
dA = − F ~ ex · d~
r. (83)
A ∂y C

3.4 Green and Stokes’ theorems

The two results (81) and (83) can be combined into a single important formula. Write (81) for a
function G(x, y) in place of F (x, y) and subtract from it (83) to deduce the curl form of Green’s
theorem Z   I I
∂G ∂F
− dA = (F ~ ex + G~ey ) · d~
r = (F dx + Gdy). (84)
A ∂x ∂y C C

Note that dx and dy in the last integral are not independent quantities, they are the projection
of the line element d~r onto the basis vectors ~ ex and ~ ey as written in the middle line integral. If
(x(t), y(t)) is a parametrization for the curve then dx = (dx/dt)dt and dy = (dy/dt)dt and the
t-bounds of integration should be picked to correspond to counter-clockwise orientation. Note also
that Green’s theorem (84) is the formula to remember since it includes both (81) when F = 0 and
(83) when G = 0.
Green’s theorem can be written in several equivalent forms. Define the vector field ~ v = F (x, y)~
e+
G(x, y)~ey . A simple calculation verifies that its curl is purely in the ~
ez direction indeed (62) gives
~ ×~
∇ v=~ ez (∂G/∂x − ∂F/∂y) thus (84) can be rewritten in the form
Z   I
~
∇×~ v ·~ ez dA = ~ v · d~
r, (85)
A C
 F.
c Waleffe, Math 321, 2007/03/19 22

This result also applies to any 3D vector field ~ v (x, y, z) = F (x, y, z)~ex + G(x, y, z)~ey + H(x, y, z)~ez
and any planar surface A perpendicular to ~ ez since ~ ~
ez · (∇ × ~ v ) still equals ∂G/∂x − ∂F/∂y for
such 3D vector fields and the line element d~ r of the boundary curve C of such planar area is
perpendicular to ~ v · d~
ez so ~ r is still equal to F (x, y, z)dx + G(x, y, z)dy. The extra z coordinate is a
mere parameter for the integrals and (85) applies equally well to 3D vector field ~ v (x, y, z) provided
A is a planar area perpendicular to ~ ez .
In fact that last restriction on A itself can be removed. This is Stokes’ theorem which reads
Z   I
~
∇×~ ~
v · dS = ~ v · d~
r, (86)
S C

where S is a bounded orientable surface in 3D space, not necessarily planar, and C is its closed
curve boundary. The orientation of the surface as determined by the direction of its normal ~ n,
~
where dS = ~ ndS, and the orientation of the boundary curve C must obey the right-hand rule.
Thus a corkscrew turning in the direction of C would go through S in the direction of its normal
~
n. That restriction is a direct consequence of the fact that the right hand rule enters the definition
of the curl as the cross product ∇~ ×~ v.
Stokes’ theorem (86) provides a geometric interpretation for the curl. At any point ~ r , consider a
small disk of area A perpendicular to an arbitrary unit vector ~ n, then Stokes’ theorem states that
I
~ 1
(∇ × ~v) · ~
n = lim v · d~
~ r (87)
A→0 A C
H
where C is the circle bounding the disk A oriented with ~ n. The line integral ~ v · d~
r is called the
v around the closed curve C.
circulation of the vector field ~

Partial proof of Stokes’ Theorem

I We’ll sketch a proof of Stokes’ theorem and do a non-trivial exercise in vector calculus in the
process. Imagine a bounded surface S in 3D space that can be parametrized in terms of cartesian
coordinates (x, y), that is S : z = h(x, y) with x, y in some domain A in the (x, y)-plane then
~
r (x, y) = ~
ex x + ~
ey y + ~
ez h(x, y) and

∂~
r ∂h ∂~
r ∂h
=~
ex + ~
ez , =~
ey + ~
ez , (88)
∂x ∂x ∂y ∂y
so    
~= r ∂~
∂~ r ∂h ∂h
dS × ez − ~
dxdy = ~ ex −~
ey dxdy. (89)
∂x ∂y ∂x ∂y
Now if we write ~
v=~ ex u + ~
ey v + ~
ez w then from (62)
     
~ ×~ ∂w ∂v ∂u ∂w ∂v ∂u
∇ v=~ ex − +~ey − +~
ez − ,
∂y ∂z ∂z ∂x ∂x ∂y

and from (89)

       
~ ×~ ∂~
r ∂~r ∂v ∂u ∂h ∂w ∂v ∂h ∂u ∂w
(∇ v) · × = − − − − −
∂x ∂y ∂x ∂y ∂x ∂y ∂z ∂y ∂z ∂x
     
∂v ∂v ∂h ∂u ∂u ∂h ∂h ∂w ∂h ∂w
= + − + + − . (90)
∂x ∂z ∂x ∂y ∂z ∂y ∂y ∂x ∂x ∂y
 F.
c Waleffe, Math 321, 2007/03/19 23

Now each of these expressions are partial derivatives of functions of three variables, thus ∂v/∂x for
instance means derivative with respect to x with y and z fixed. However in the surface integral
(86), all these expressions must be evaluated on the surface S, meaning at z = h(x, y) with x, y in
A. The chain rule then says that
   
∂v ∂v ∂h ∂v̄ ∂u ∂u ∂h ∂ ū
+ = and + = (91)
∂x ∂z ∂x S ∂x ∂y ∂z ∂y S ∂y
where
ū(x, y) = u(x, y, h(x, y)), v̄(x, y) = v(x, y, h(x, y)). (92)
and the subscript S indicates that the expression must be evaluated on S meaning at z = h(x, y)
with x, y in A. We can massage the last term in parentheses on the right hand side of (90)
   
∂h ∂w ∂h ∂w ∂h ∂w ∂w ∂h ∂h ∂w ∂w ∂h
− = + − +
∂y ∂x ∂x ∂y ∂y ∂x ∂z ∂x ∂x ∂y ∂z ∂y
∂h ∂ w̄ ∂h ∂ w̄
= −
∂y ∂x ∂x ∂y
   
∂ ∂h ∂ ∂h
= w̄ − w̄ . (93)
∂x ∂y ∂y ∂x
where w̄(x, y) = w(x, y, h(x, y)) as in (92) and the last step follows from the cancellation of the
mixed partials ∂ 2 h/∂y∂x = ∂ 2 h/∂x∂y. Substituting (91) and (93) into (90) yields
     
~ ∂~
r ∂~ r ∂ ∂h ∂ ∂h
(∇ × ~
v) · × = v̄ + w̄ − ū + w̄
∂x ∂y ∂x ∂y ∂y ∂x
∂G(x, y) ∂F (x, y)
≡ − . (94)
∂x ∂y
Hurrah! we have managed to write
Z  Z  
~

~ ∂G ∂F
∇×~v · dS = − dA. (95)
S A ∂x ∂y
Now for any point ~
r (x, y) on the surface S, we have from (88)
 
∂~
r ∂~
r ∂h ∂h
d~r= dx + dy = ~
ex dx + ~
ey dy + ~
ez dx + dy
∂x ∂y ∂x ∂y
where dx and dy would be independent for any point on S except on its boundary C where dx and
r is tangent to C. In any case, at any point on the surface S
dy must be such that d~
   
∂h ∂h
v · d~
~ r = ū + w̄ dx + v̄ + w̄ dy = F (x, y)dx + G(x, y)dy.
∂x ∂y
r is on the boundary C of S then (x, y) is on the boundary CA of A and we
Hurrah again! finally if ~
have I I
v · d~
~ r= F dx + Gdy. (96)
C CA

The results (95) and (96) reduce Stokes’ theorem (86) to Green’s theorem (84), thereby proving it
in this particular case where S can be parametrized by x and y. 
 F.
c Waleffe, Math 321, 2007/03/19 24

More general proof of Stokes’ theorem

Indicial notation enables a fairly straightforward proof of Stokes’ theorem for the more general case
of a surface S in 3D space that can be parametrized by a ‘good’ function ~ r (s, t) (differentiable and
integrability as needed). Such a surface S can fold and twist (it could even intersect itself!) and is
therefore of a more general kind than those that can be parametrized by the cartesian coordinates
x and y. The main restriction on S is that it must be ‘orientable’. This means that it must have
an ‘up’ and a ‘down’ as defined by the direction of the normal ∂~ r /∂s × ∂~ r /∂t. The famous Mobius
strip only has one side and is the classical example of a non-orientable surface. The boundary of
the Mobius strip forms a knot.
I Let xi (s, t) represent the i component of the position vector ~ r (s, t) in the surface S with i = 1, 2, 3.
Then
 
~ ×~ ∂~
r ∂~
r ∂vk ∂xl ∂xm ∂vk ∂xl ∂xm
(∇ v) · × =ijk ilm = ijk ilm
∂s ∂t ∂xj ∂s ∂t ∂xj ∂s ∂t
∂vk ∂xl ∂xm
= (δjl δkm − δjm δkl )
∂xj ∂s ∂t
∂vk ∂xj ∂xk ∂vk ∂xk ∂xj
= −
∂xj ∂s ∂t ∂xj ∂s ∂t
∂v̄k ∂xk ∂v̄k ∂xk
= −
∂s  ∂t ∂t ∂s  
∂ ∂xk ∂ ∂xk
= v̄k − v̄k
∂s ∂t ∂t ∂s
∂G(s, t) ∂F (s, t)
= − , (97)
∂s ∂t
where we have used (69), then the chain rule
∂vk ∂xj ∂vk ∂x1 ∂vk ∂x2 ∂vk ∂x3 ∂v̄k
= + + = ,
∂xj ∂s ∂x1 ∂s ∂x2 ∂s ∂x3 ∂s ∂s
and equality of mixed partials ∂ 2 xk /∂s∂t = ∂ 2 xk /∂t∂s with v̄k (s, t) = vk x1 (s, t), x2 (s, t), x3 (s, t) .

The result (97) is equivalent to (94) with s, t in place of x, y. It is not only more general but
also easier to obtain using indicial notation. That’s because the cartesian parametrization actually
breaks the symmetry of the problem. This proof demonstrates the power of the indicial notation
and the summation convention. The first line of (97) is a quintuple sum over all values of the indices
i, j, k, l and m! This would be unmanageable without the compact notation.
Now for any point on the surface S with position vector ~ r (s, t)
     
∂xi ∂xi ∂xi ∂xi
v · d~
~ r = v̄i ds + dt = v̄i ds + v̄i dt = F (s, t)ds + G(s, t)dt, (98)
∂s ∂t ∂s ∂t
and we have again reduced Stokes’ theorem to Green’s theorem (84) but expressed in terms of s
and t instead of x and y. In details we have shown that
Z Z   Z  
~ ~ ~ ∂~
r ∂~ r ∂G(s, t) ∂F (s, t)
(∇ × ~
v ) · dS = (∇ × ~ v) · × dsdt = − dsdt, (99)
S A ∂s ∂t A ∂s ∂t
I I
v · d~
~ r= F (s, t)ds + G(s, t)dt. (100)
C CA
The right hand sides of (99) and (100) are equal by Green’s theorem (84). 
 F.
c Waleffe, Math 321, 2007/03/19 25

3.5 Divergence form of Green’s theorem

The fundamental theorems (81) and (83) in R2 can be rewritten in a more palatable form.
The line element d~r at a point on the curve C is in the direction of
the unit tangent ~
t at that point, so d~r =~t dr, where ~t points in the
~
t ~
n counterclockwise direction of the curve. Then ~
t × ~
e z = ~
n is the unit
outward normal ~n to the curve at that point and ~ ez × ~n=~ t so
~
ey
ex · ~
~ ex · (~
t=~ ez × ~ ex × ~
n) = (~ ez ) · ~
n = −~
ey · ~
n (101)

ey · ~
~ ey · (~
t=~ ez × ~ ey × ~
n) = (~ ez ) · ~ ex · ~
n=~ n. (102)
~
ez ~
ex
Hence since d~ r =~ t dr, the fundamental theorems (81) and (83) can
be rewritten
Z I I
∂F
dA = F ~ ey · d~
r = F~ ex · ~
n dr, (103)
A ∂x C C
Z I I
∂F
dA = − F ~ ex · d~
r = F~ ey · ~
n dr. (104)
A ∂y C C
The right hand side of these equations is easier to remember since they have ~ ex going with ∂/∂x
and ~ ey with ∂/∂y and both equations have positive signs. But there are hidden subtleties. The
arclength element dr = |d~ r | is positive by definition and ~ n must be the unit outward normal to
the boundary, so if an explicit parametrization
H of the boundary curve is known, the bounds of
integration should be picked so that C dr would be the length of the curve with a positive sign.
For the d~r line integrals, the bounds of integration must correspond to counter-clockwise orientation
of C.
Formulas (103) and (104) can be combined in more useful forms. First, if s is the signed distance
in a direction of the unit vector ŝ, then the derivative in the direction ŝ is ∂F/∂s = ŝ · ∇F ~ =
sx ∂F/∂x + sy ∂F/∂y, where ŝ = sx~ ex + sy ~
ey , therefore combining (103) and (104) accordingly we
obtain Z I
∂F
dA = F ŝ · ~ n dr. (105)
A ∂s C

This result is written in a coordinate-free form. It applies to any direction ŝ in the x,y plane.
Another useful combination is to add (103) to (104) written for a function G(x, y) in place of F (x, y)
yielding the divergence-form of Green’s theorem
Z   I
∂F ∂G
+ dA = (F ~ ex + G~ey ) · ~
n dr. (106)
A ∂x ∂y C

~ ·~
The left hand side integrand is easily recognized as the divergence ∇ v of the vector field ~
v (x, y) =
F~
ex + G~ ey .

3.6 Gauss’ theorem

Gauss’ theorem is the 3D version of the divergence form of Green’s theorem. It is proved by first
extending the fundamental theorem of calculus to 3D.
If V is a bounded volume in 3D space and F (x, y, z) is a scalar function of the cartesian coordinates
(x, y, z), then we have Z I
∂F
dV = F~ ez · ~
ndS, (107)
V ∂z S
 F.
c Waleffe, Math 321, 2007/03/19 26

where S is the closed surface enclosing V and ~ n is the unit outward normal to S.
The proof of this result is similar to that for (81). Assume that the surface can be parametrized
using x and y in two pieces: an upper ‘hemisphere’ at z = zu (x, y) and a lower ‘hemisphere’ at
z = zl (x, y) with x, y in a domain A, the projection of S onto the x, y plane, that is the same
domain for both the upper and lower surfaces. The closed surface S is not a sphere in general but
we used the word ‘hemisphere’ to help visualize the problem. For a sphere, A is the equatorial disk,
z = zu (x, y) is the northern hemisphere and z = zl (x, y) is the southern hemisphere.
Iterated integrals with dV = dA dz and the fundamental theorem of calculus give
Z Z Z zu (x,y) Z
∂F ∂F  
dV = dA dz = F (x, y, zu (x, y)) − F (x, y, zl (x, y)) dA. (108)
V ∂z A zl (x,y) ∂z A

We can interpret that integral over A as an integral over the entire

closed surface S that bounds V . All we need for that is a bit of
α ~ =~
geometry. If dA is the projection of the surface element dS ndS
~
ez
~
n onto the x, y plane then we have dA = cos α dS = ±~ ez · ~
n dS. The
dS + sign applies to the upper surface for which ~ n is pointing up (i.e.
α in the direction of ~
ez ) and the - sign for the bottom surface where
~
n points down (and would be opposite to the ~ n on the side figure).
Thus we obtain Z I
∂F
dV = F~ez · ~
n dS, (109)
dA V ∂z S
where ~n is the unit outward normal to S.
We can obtain similar results for the volume integrals of ∂F/∂x and ∂F/∂y and combine those to
obtain Z I
∂F
dV = F ŝ · ~
n dS, (110)
V ∂s S

for arbitrary but fixed direction ŝ. This is the 3D version of (105) and of the fundamental theorem
of calculus.
We can combine this theorem into many useful forms. Writing it for F (x, y, z) in the ~ ex direction,
with the ~ey version for a function G(x, y, z) and the ~ez version for a function H(x, y, z) we obtain
Gauss’s theorem Z I
~ ·~
∇ v dV = v·~
~ n dS, (111)
V S

where ~v = F~ex + G~ey + H~ez . This is the 3D version of (106). Note that both H (110) and (111) are
expressed in coordinate-free forms. These are general results. The integral S ~ v·~ndS is the flux of
~
v through the surface S. If ~v (~
r ) is the velocity of a fluid at point ~
r then that integral represents
the time-rate at which volume of fluid flows through the surface S.
Gauss’ theorem provides a coordinate-free interpretation for the divergence. Consider a small
sphere of volume V and surface S centered at a point ~ r , then Gauss’ theorem states that
I
~ ·~ 1
∇ v = lim ~v·~ n dS. (112)
V →0 V S

Note that (110) and (111) are equivalent. We deduced (111) from (110), but we can also deduce
(110) from (111) by considering the special ~ v (~
r ) = F (~
r )ŝ where ŝ is a unit vector independent of
~ · (F ŝ) = ŝ · ∇F
r . Then from our vector identities ∇
~ ~ = ∂F/∂s.
 F.
c Waleffe, Math 321, 2007/03/19 27

3.7 Other useful forms of the fundamental theorem in 3D

A useful form of (110) is to write it in indicial form as
Z I
∂F
dV = nj F dS. (113)
V ∂xj S

Then with f (~
r ) in place of F (~
r ) we deduce that
Z I
∂f
~
ej dV = ~
ej nj f dS (114)
V ∂xj S

since the cartesian unit vectors ~

ej are independent of position. This result can be written in
coordinate-free form as Z I
~
∇f dV = f~
n dS. (115)
V S
One application of this form of the fundamental theorem is to prove Archimedes’ principle.
Next, writing (113) for vk in place of F yields
Z I
∂vk
dV = nj vk dS, (116)
V ∂xj S

which can be multiplied by the position-independent ijk and summed over all j and k to give
Z I
∂vk
ijk dV = ijk nj vk dS. (117)
V ∂xj S

The coordinate-free form of this is

Z I
~ ×~
∇ v dV = n×~
~ v dS. (118)
V S

The integral theorems (115) and (118) provide yet other geometric interpretations for the gradient
I
~ 1
∇f = lim f~n dS, (119)
V →0 V S

and the curl I

~ ×~ 1
∇ v = lim n×~
~ v dS. (120)
V →0 V S
These are similar to the result (112) for the divergence. However, the geometric definition of the
gradient as the vector pointing in the direction of greatest rate of change (sect. 2.1) and of the ~ n
component of the curl as the limit of the local circulation per unit area as given by Stokes’ theorem
(87) are perhaps more fundamental.
In applications we use the fundamental theorem as we do in 1D, namely to reduce a 3D integral
to a 2D integral, for instance. However we also use them the other way, to evaluate a complicated
surface integral as a simpler volume integral for instance.
Exercises
~ · d~
H H
1. If C is any closed curve in 3D space (i) calculate C ~
r · d~
r in two ways, (ii) calculate C ∇f r
in two ways, where f (~ r ) is a scalar function. [Hint: by direct calculation and by Stokes
theorem].
 F.
c Waleffe, Math 321, 2007/03/19 28

r−3~
H
2. If C is any closed curve in 3D space not passing through the origin calculate C r · d~
r in
two ways.
~ = (~
3. Calculate the circulation of the vector field B ez × ~ ez × ~
r )/|~ r |2 (i) about a circle of radius
R centered at the origin in a plane perpendicular to ~ ez ; (ii) about any closed curve C in 3D
that does not go around the z-axis; (ii) about any closed curve C0 that does go around the
~
ez axis. What’s wrong with the z-axis anyway?

4. Consider ~ v = ω ~ ×~ r where ω ~ is a constant vector, independent of ~ r . (i) Evaluate the

circulation of ~v about the circle of radius R centered at the origin in the plane perpendicular
to the direction ~ n by direct calculation of the line integral; (ii) Calculate the curl of ~v using
vector identities; (iii) calculate the circulation of ~ v about a circle of radius R centered at ~
r0
in the plane perpendicular to ~ n.
H
5. If C is any closed curve in 2D, calculate C ~ n dr where ~
n(~r ) is the unit outside normal to C
at the point ~ r of C.
H
6. If S is any closed surface in 3D, calculate ~ ndS where ~n(~
r ) is the unit outside normal to S
at a point ~ r on S.
H
7. If S is any closed surface in 3D, calculate p(~ r )~
ndS where ~ n is the unit outside normal to
S and p(~ r ) = (p0 − ρg~ ez · ~
r ) where p0 , ρ and g are constants (This is Archimedes’ principle
with ρ as fluid density and g as the acceleration of gravity.)

8. Calculate the flux of ~

r through (i) the surface of a sphere of radius R centered at the origin
in two ways; (ii) through the surface of the sphere of radius R centered at ~r 0 ; (iii) through
the surface of a cube of side L with one corner at the origin in two ways.

9. (i) Calculate the flux of ~

v = ~r /r3 through the surface of a sphere of radius  centered at
the origin. (ii) Calculate the flux of that vector field through a closed surface that does
not enclose the origin [Hint: use the divergence theorem] (iii) Calculate the flux through an
arbitrary closed surface that encloses the origin [Hint: use divergence theorem and (i) to
isolate the origin. What’s wrong with the origin anyway?]
~ r−~
10. Calculate ∇|~ ~ ·~
r 0 |−1 and ∇ v with ~ r−~
v = (~ r 0 |3 where ~
r−~
r 0 )/|~ r 0 is a constant vector.

11. Calculate the flux of ~ v through the surface of a sphere of radius R centered at the origin when
~
v = α1 (~ r −~ r −~
r 1 )/|~ r 1 |3 +α2 (~
r −~ r −~
r 2 )/|~ r 2 |3 where α1 and α2 are scalar PNconstants and (i) |~ r1 |
and |~r 2 | are both less than R; (ii) |~ r 1 | < R < |~ r 2 |. Generalize to ~v = i=1 αi (~ r −~ r i |3 .
r −~
r i )/|~
~ (~r ) = V0 f~dV0 where V0 is the inside of a sphere of radius R centered at O,
R
12. Calculate F
f~ = (~r−~ r−~
r 0 )/|~ r 0 |3 with |~r | > R and the integral is over ~ r 0 . Warning: this is essentially
the gravity field or force at ~ r due to a sphere of uniform mass density. Supposedly, it took
Newton 20 years to figure it out... of course he was never taught calculus, he had to invent
it, and vector calculus came much after him and he never knew about Gauss’ theorem. The
integral can be cranked out if you’re good at analytic integration. But the smart solution
is to realize that the integral over ~ r 0 is essentially a sum over ~ r i as in the previous exercise
so we can figure out the flux of F ~ through any closed surface enclosing all of V0 . Now by
symmetry F ~ (~r ) = F (r)r̂, so knowing the flux is enough to figure out F (r). Newton would
be impressed!