Chapter 5

Orthogonal Curvilinear Coordinates

Last update: 22 Nov 2010
Syllabus section:
4. Orthogonal curvilinear coordinates; length of line element; grad, div and curl in curvilinear coordinates;
spherical and cylindrical polar coordinates as examples.
So far we have only used Cartesian x, y, z coordinates. Sometimes, because of the geometry of a given
problem, it is easier to work in some other coordinate system. Here we show how to do this, restricting the
generality only by an orthogonality condition.
5.1 Plane Polar Coordinates
In Calculus II and Chapter 2, we met the simple curvilinear coordinates in two dimensions, plane polars,
dened by
x = r cos, y = r sin.
We can easily invert these relations to get
r =
_
x
2
+y
2
, = arctan(y/x).
The Chain Rule enables us to relate partial derivatives with respect to x and y to those with respect to r and
and vice versa, e.g.
f
r
=
f
x
x
r
+
f
y
y
r
=
x
r
f
x
+
y
r
f
y
. (5.1)
In Calculus II, the rule for changing coordinates in integrals is also given. The general rule is that if we
change coordinates from x, y to u, v where x = x(u, v), y = y(u, v), then a dxdy in an area integral is replaced
by the Jacobian determinant

x
u
x
v
y
u
y
v

dudv .
If we dene r = (x(u, v), y(u, v), 0), differentiate w.r.t. u, v and take the cross-product, we will see that the
above is equal to
dS =

r
u

r
v

dudv ,
64
as we derived in section 4.2.
For plane polar coordinates, replacing u, v with r, and calcuating the determinant above gives dS =
r dr d; this can also be shown geometrically by considering an innitesimal quadrilateral with corners at
(r, ), . . . , (r +dr, +d) and working out the area from a sketch.
Example 5.1. The Gaussian integral (related to the Gaussian distribution in statistics)
Consider the integral
_

e
(x
2
+y
2
)
dxdy =
_
_

e
x
2
dx
_
2
.
Transforming to polar coordinates gives
_

0
re
r
2
dr
_
2
0
d = [
1
2
e
r
2
]

0
[]
2
0
=
and hence (according to Dr. Saha the most beautiful of all integrals)
_

e
x
2
dx =

.
For later use, we now construct the unit vectors in the directions in which r and increase at a point,
which we will denote e
r
and e

. These are tangent to the coordinate lines, where a coordinate line means a
curve on which only one of the coordinates is varying, and the other coordinates are xed. Coordinate lines
are generalizations of lines parallel to the x, y, z axes in Cartesians, but now they wont be straight lines (hence
the curvilinear in the chapter title).
We already know how to nd the tangent vectors to coordinate lines, by taking partial derivatives of r
with respect to each of r, ; then all we have to do is divide those by their lengths to get unit vectors. Thus in
plane polars we have
r = r cos i +r sinj ,
so a small change dr gives us a change
dr
r
=
r
r
dr = (cos i +sin j)dr,

r
r

= 1 e
r
=
r
r
= cos i +sin j
while a small change d gives us
dr

=
r

d = (r sin i +r cos j)d, |

r

| = r e

=sin i +cos j.
So a general small displacement becomes
r = e
r
r +r e

.
We will see the value of this later on; we are next going to consider three-dimensional versions of polar
coordinates: there are two common versions, rstly cylindrical polars and then spherical polars.
65
5.2 Cylindrical Polar Coordinates
For cylindrical polars, we turn the plane polars in the x, y plane into three-dimensional coordinates by simply
using z as the third coordinate (see Fig. 5.1). To avoid confusion with other coordinate systems, we shall for
clarity
1
rename r as and as , but beware that in other courses, books, and applications of these ideas, r
and will still be used. Thus we have
x = cos, y = sin, z = z ,
or
r = cos i + sin j +zk ,
and quantities in any plane z =constant will be as in plane polars. The gure 5.1 shows coordinate lines for
each of , and z; here the coordinate line for is a line of varying and constant ,z; and likewise for
the other two. Note that the coordinate lines for , z are straight lines, while the line is a circle around the
z axis. Thomass Fig. 15.37 shows a nice diagram of surfaces on which one of the coordinates is constant:
the constant- surface is a cylinder whose axis is the zaxis, while surfaces of constant or constant z are
planes.
Figure 5.1: Cylindrical polar coordinates relative to Cartesian, and with sample - and -curves shown.
The fact that constant gives a cylinder gives the name cylindrical polars: these coordinates are natural
ones to use whenever there is a problem involving cylindrical geometry or symmetry (for example, doing a
surface integration over a cylinder, or in physics calculating a magnetic eld around a straight wire).
To get partial derivatives in curvilinear coordinates we again use the chain rule (5.1), but now with three
terms on the right. Taking the plane polar results, changing variable names and appending e
z
= k, the unit
vectors along the coordinate lines are
e

= cos i +sin j , e

=sin i +cos j , e
z
= k
respectively. We can write this in matrix form as
_
_
e

e
z
_
_
=
_
_
cos sin 0
sin cos 0
0 0 1
_
_
_
_
i
j
k
_
_
. (5.2)
1
Unfortunately, for the same reasons of clarity, Thomas adopts the alternative solution of renaming two of the spherical polar coor-
dinates. To avoid confusion with past years exam papers I have kept to the choice used there, which is also the one used in most books.
Thomas chooses (, , ) for the usual (r, , ). The swap of and is particularly likely to be confusing.
66
It is easy to see from the above that the dot-product of any two es gives 1 (if they are the same) or 0 (for
any two different ones), like the rules for i, j, k. This implies that the three es are an orthogonal triple of
unit vectors, and also implies geometrically that the cross-product of any two different es will be the third
one.
We can also express this property in matrix notation: the 3 3 matrix above, call it R, is a rotation matrix,
i.e. one such that R
1
= R
T
, where the R
T
denotes transpose. This comes about because the dot-product
of any two es is given by one element of the matrix RR
T
, and the es are an orthogonal triple if and only
if RR
T
= I, the identity matrix.
2
Also note that if we want i, j, k in terms of the es, we can just multiply
Eq. 5.2 by R
1
= R
T
.
The lengths of r/, r/ and r/z are respectively 1, and 1; we can use these together with the
es to nd innitesimal area elements: e.g. taking a surface =constant (a cylinder), we can treat this as a
2-parameter surface with , z as the parameters, so the vector area element for small changes d, dz is given
by
dS =
r

r
z
d dz
= e

e
z
d dz
= e

d dz ;
this will be useful when doing surface integrals over a cylinder. (As usual, there is a potentially ambiguous
choice of sign with vector areas, due to the sign-ip in changing order of a cross product; take care with this,
e.g. when doing a problem check that your vector area matches the desired direction).
When doing volume integrals, we may need the volume element which is
dV = d ddz
from the scalar triple product.
5.3 Spherical Polar Coordinates
These are coordinates (r, , ), where r measures distance from the origin, measures angle from some
chosen axis, called the polar axis, and measures angle around that axis (see Fig 5.2.) To relate them to
Cartesian coordinates we usually assume that the z-axis is the polar axis. Then, let P be our chosen point
at (r, , ), and drop a perpendicular from P to the zaxis meeting it at Q. The line OP is at angle to the
positive z-axis, so clearly OQ = z = r cos and PQ = r sin. Dropping another perpendicular from P to the
xy plane, we get a point in the xy plane at distance r sin from the origin; then inserting = r sin into the
cylindrical polars in Sec. 5.2 gives us:
x = r sin cos, y = r sin sin, z = r cos.
or, as a position vector
r = r sin cos i +r sin sin j +r cos k
Here the is the same as that of cylindrical polars, which explains why we chose the same letter. The inverse
of these relations is
r =
_
x
2
+y
2
+z
2
, = arctan
_
_
x
2
+y
2
z
_
, = arctan
_
y
x
_
.
2
Rotation matrices are special because they preserve lengths and angles; e.g. if we take two vectors a, b, write them as column
vectors, then their scalar product in matrix notation is a
T
b. The two vectors rotated by matrix R are Ra and Rb. To conserve scalar
product, we must have (Ra)
T
(Rb) =a
T
b, and using the transpose rule this becomes a
T
R
T
Rb =a
T
b. For this to apply for any two a, b
we must have R
T
R =I, the identity matrix.
67
Coordinate lines of r, (i.e. lines of constant and ), are straight radial lines from the origin; coordinate
Figure 5.2: Spherical polar coordinates relative to Cartesian, and with sample r-, and -curves shown.
lines of (constant r and ) are meridional semicircles, i.e. semicircles centred at the origin and in a plane
containing the polar axis; and coordinate lines of (constant r and ) are latitudinal circles, i.e. circles
centred at a point on the polar axis and in a plane perpendicular to it. Note however that while r runs from
0 to (like the r of plane polars and of cylindrical polars) and runs from 0 to 2 (like the of plane
polars), only runs from 0 to , since for any point P the angle between OP and the z-axis wont exceed
180 degrees = radians.
The coordinate lines of are strictly semi-circles, rather than circles. To make a circle we have to take
the coordinate lines of for two different , say
0
and
0
+. Thomass Fig. 15.42 shows a nice diagram
of surfaces on which one of the coordinates is constant.
You should beware of the fact that some authors, including Thomas, use different notation, in particular
swapping the meanings of and in the denition of spherical polars. We shall consistently use the above
notation for spherical polar coordinates, which is the most common one, throughout this course.
Note that these again generalize the plane polar coordinates, but this time the polars r, are in planes
containing the z (or polar) axis, rather than in planes perpendicular to it. The spherical polar coordinates are
of course the natural ones to use when we have a spherical geometry, or part of a sphere.
Now we construct the e vectors as before: taking partial derivatives of r above with respect to each of the
coordinates in turn, we get
r/r = sin cos i +sin sin j +cos k,
r/ = r cos cos i +r cos sin j r sin k
r/ = r sin sin i +r sin cos j.
The lengths of these, by simple applications of cos
2
+sin
2
= 1, are respectively 1, r, and r sin. Dividing
these derivatives by their lengths gives us the unit vectors e
r
, e

and e

tangent to the coordinate lines, which

we can write as
_
_
e
r
e

_
_
=
_
_
sin cos sin sin cos
cos cos cos sin sin
sin cos 0
_
_
_
_
i
j
k
_
_
. (5.3)
It is straightforward to show that again the dot-product of any two es is 1 (if they are the same) or 0 (if
different); therefore the cross-product of any two es is the third one and the matrix above is again a
rotation matrix.
68
It is also worth noting that e
r
=r/r, as expected by symmetry since e
r
is a unit vector pointing away from
the origin at point r.
For doing integrals later on, the volume element is given by the scalar triple product
dV = (r/r) (r/).(r/) dr d d = r
2
sin dr d d .
The innitesimal area element on a sphere (i.e. a surface of constant r) is given by
dS = (r/) (r/) d d = r
2
sin e
r
dd .
Similar results hold for surfaces of constant and of constant , but are not so common in practice; note
that the above area element on a sphere turns up in many examples and exam questions, and is well worth
memorising.
Example 5.2. Earth polar coordinates
To dene spherical polars on the Earth, let the polar axis be the Earths rotation axis, with z increasing
to the North, let the equator dene the x, y plane, and let the prime meridian (the one through Greenwich)
be = 0. Then any point on the Earths surface can be referred to by the spherical polar angles (, ). In
navigation people use latitude and logitude. Longitude is measured East or West from the prime meridian
and is in the range (0, 180

) so to get for a place with Westerly longitude we just subtract from 2 = 360

.
Latitude is dened to be 0 at the equator (whereas =90

=/2 there). Given a latitude, we need to subtract

it from 90

if it is North and add it to 90

if it is South.
For example Buenos Aires, which has latitude 34

36

S, and longitude 58

22

W, will have Earth polar

coordinates = 125

, = 302

to the nearest degree.

5.4 Some applications of these polar coordinates
Using polar (or cylindrical) coordinates the area within a circle of radius R,
_
R
0
_
2
0
r d dr, comes out imme-
diately as R
2
.
Using spherical polar coordinates the volume of a sphere of radius R is
_
R
0
_

0
_
2
0
r
2
sin d d dr
which evaluates to
4
3
R
3
. (Remember that for a full sphere, the ranges of integration are 0 , 0
2).
Example 5.3. Area of a cone:
Consider the conical surface =
1
cut in a sphere of radius s. The area is given by integrating
_
2
0
d
_
s
0
sin
1
r dr = s
2
sin
1
.
Here s is the slant height of the cone. The cones base (say b) will be ssin
1
. Hence we can express the
sloping area of a cone neatly as sb.
Example 5.4. We now reconsider Example 4.5.
Find the ux of the eld F = zk across the portion of the sphere x
2
+y
2
+z
2
= a
2
in the rst octant with
normal taken in the direction away from the origin.
69
Because of the geometry of the surface, it is easiest to work in spherical polar coordinates (r, , ), so the
sphere has r = a. The unit normal n to the sphere that points away from the origin is just e
r
, the outward
radial vector of unit length. Now
F.e
r
= zk e
r
= zcos = r cos
2
.
using 5.3 to evaluate k.e
r
= cos. An area element on the surface of a sphere of radius r is
(rd)(r sin d) =r
2
sin d d. For our given sphere r = a, so
_
S
F ndS =
_
/2
0
_
/2
0
acos
2
a
2
sin d d
= a
3
_
/2
0
_
/2
0
cos
2
sin dd
=

2
a
3
_
1
3
cos
3

_
/2
0
=

6
a
3
.
Note that the integrand didnt depend on , so we just replaced the d integral with a multiplication by the
range, here (/2 0). This is a common short-cut to note.
Example 5.5. Cutting an apple
In his book, Matthews poses a good problem for illustrating integration using curved coordinates: A
cylindrical apple corer of radius a cuts through a spherical apple of radius b. How much of the apple does it
remove?
We can reformulate the problemslightly, without losing generality, by letting the radius of the apple equal
unity and introducing sin
1
= a/b (i.e. we scale the problem by b). In our restated problem the corer cuts
through the peel at =
1
and =
1
2

1
in spherical polars, i.e. in cylindrical polars at
= sin
1
, z = cos
1
,
and, of course, at z =cos
1
.
We can now complete the solution of this problem in (at least) four different ways: three of these are
relegated to an appendix, not given in lectures.
3
The rst way is to integrate over z and then
4
_
sin
1
0
d
_

1
2
0
dz = 4
_
sin
1
0
(1
2
)
1
2
d =
4
3
(1 cos
3

1
).
5.5 General Orthogonal Curvilinear Coordinates
The two sets of polar coordinates above have a feature in common: the three sets of coordinate lines are
orthogonal to one another at all points, because their tangent vectors and corresponding unit vectors es are
orthogonal. (This is where the orthogonal in the chapter title comes from).
3
I give only the key steps. Some algebraic lling-in is needed. In each version we can shorten the calculations by replacing the
integration with multiplication by 2 (since the integrand doesnt depend on ) , and also doing the integrals only for z 0, and then
doubling using symmetry.
70
General orthogonal coordinates are coordinates for which these properties are true, i.e. the coordinate
lines are always mutually perpendicular at a given point, though they are generally curved. In general,
coordinates need not be orthogonal. However, we shall be concerned only with orthogonal curvilinear co-
ordinates. Cylindrical polars and spherical polars are the only non-Cartesian coordinate systems in which
you will be expected to perform explicit calculations in this course, apart from simple substitutions into the
general formulae.
Suppose (u
1
, u
2
, u
3
) are a general set of coordinates, dened by some given function r(u
1
, u
2
, u
3
). As
before, we calculate r/u
1
which is the tangent vector to a u
1
line (varying u
1
, constant u
2
, u
3
). Next we
dene the arc-length h
1
and unit vector e
1
as
h
1
=

r
u
1

, e
1
=
r
u
1
/h
1
therefore
r
u
1
h
1
e
1
.
It is easy to calculate that
h
2
1
=
_
x
u
1
_
2
+
_
y
u
1
_
2
+
_
z
u
1
_
2
.
Likewise differentiating r by u
2
, u
3
, we dene two more unit vectors e
2
, e
3
, along the coordinate lines of
u
2
and u
3
, and associated arc-length parameters h
2
and h
3
. This is useful for several reasons: rstly, e
1
tells
us in which direction r moves with a small change in u
1
, while h
1
du
1
is the distance moved along e
1
, and
likewise for changes du
2
, du
3
.
We dene a coordinate system to be orthogonal iff e
1
, e
2
and e
3
are mutually orthogonal everywhere:
Coordinates (u
1
, u
2
, u
3
) are orthogonal
r
u
1
.
r
u
2
=
r
u
2
.
r
u
3
=
r
u
3
.
r
u
1
= 0
For orthogonal coordinates, a general small change (du
1
, du
2
, du
3
) in the coordinates means a displacement
dr = h
1
du
1
e
1
+h
2
du
2
e
2
+h
3
du
3
e
3
, (5.4)
which corresponds to a distance
_
h
2
1
du
2
1
+h
2
2
du
2
2
+h
2
3
du
2
3
_
1/2
.
Also, for orthogonal coordinates the dot and cross products of any two es will obey the same rules we met
before: therefore the matrix R relating (e
1
, e
2
, e
3
) to (i, j, k) will be a rotation matrix (from above) and have
the property that R
T
= R
1
.
Cartesian coordinates are of course a special simple case of orthogonal curvilinear coordinates, in which
all the coordinate lines are straight lines and all of h
1
= h
2
= h
3
= 1.
Sometimes it is convenient to replace the 1,2,3 with the letters of the coordinates, e.g. in cylindrical polar
coordinates, we wrote e

, e

, e
z
. There we already found h

= 1 and h
z
= 1, but h

= , so a change d
corresponds to moving a distance d along a circle around the zaxis.
In spherical polar coordinates, h
r
= 1 again, and h

= r. A change d in corresponds to moving a

distance r sin d (because r sin is the radius of the particular latitudinal circle around the zaxis), so
h

= r sin.
One reason that orthogonal coordinates are so useful is that in any orthogonal coordinate system (u
1
, u
2
, u
3
),
small displacements along u
1
and u
2
dene small rectangles, while small displacements along u
1
, u
2
, u
3
de-
ne small cuboids. In other words, h
1
h
2
du
1
du
2
is an area element normal to e
3
on a surface of constant u
3
,
and h
1
h
2
h
3
du
1
du
2
du
3
is a volume element.
71
5.6 Vector elds and vector algebra in curvilinear coordinates
Scalar elds can of course be expressed in (orthogonal) curvilinear coordinates: they are simply written as
functions f (u
1
, u
2
, u
3
) or for brevity f (u
i
).
As you will know from Linear Algebra, vectors can be expressed using any basis of the vector space
concerned. The same is true, at each point, of vector elds. Up to now we have always chosen i, j, k as our
basis vectors: however, when using curvilinear coordinates we will normally use the orthogonal unit vectors
along the coordinate lines as our basis vectors, and write
F = F
1
e
1
+F
2
e
2
+F
3
e
3
.
For clarity, we can use the coordinate names instead of 1,2,3 as subscripts for the three components. Thus we
may write
F = F
x
i +F
y
j +F
z
k
= F

+F

+F
z
e
z
= F
r
e
r
+F

+F

.
to express the same vector in Cartesian, cylindrical polar and spherical polar coordinates (of course e
x
=i and
so on in Cartesians). Note that the same vector F will have different components depending on our choice
of basis vectors: suppose we are given an F with dened F
x
, F
y
, F
z
above, but we want to nd F
r
, F

, F

, then
we need to use the matrix as in Eq. 5.3 to express i, j, k in terms of the es, multiply out and collect into one
term in each e. (This effectively turns into a matrix multiplication).
In any orthogonal coordinate system, the scalar (dot) and vector (cross) products work just as in Cartesian
coordinates:
w.v = w
1
v
1
+w
2
v
2
+w
3
v
3
(5.5)
and
wv =

e
1
e
2
e
3
w
1
w
3
w
3
v
1
v
2
v
3

, (5.6)
but note this only works if the vectors are dened at the same point , such as a dot product F dr or F dS
in a line or surface integral. We cannot use these for two position vectors at widely separate points, because
the es vary with position.
Vector differentiation is more complicated, because the unit vectors are no longer constant: when we
differentiated a vector in Cartesians
F = F
1
i +F
2
j +F
3
k
we just differentiated the components (F
1
, F
2
, F
3
) because the unit vectors are constant; but in general coor-
dinates the es depend on position, so we have to use the product rule and differentiate the e vectors as well
as the components F
i
.
Differentiation of these vectors with respect to a variable other than position (like the derivatives in Sec-
tion 3.1) is straightforward. For example if position r depends on time, and is given in cylindrical polars so
r = e

+ze
z
, we just use the product rule to get the time derivative
r = e

+ e

+ ze
z
+z e
z
.
(where the over-dots are shorthand for time derivative, as is common). Then since e

= cosi +sinj from

(5.2),
e

=

(sini +cosj) =

e

.
72
Similarly e
z
= 0. Substituting into the previous result, we get
r = e

+

e

+ ze
z
for a velocity in cylindrical polar coordinates.
When differentiating scalar or vector elds with respect to position, the key operations are always grad of a
scalar, and div and curl of a vector eld (this is because these are the only combinations that behave sensibly
after rotations). In the next sections, we will show how to calculate the grad, div and curl operators in general
orthogonal coordinates; then we apply those general formulae to the most common cases of cylindrical polars
and spherical polars.
5.7 The Gradient Operator in curvilinear coordinates
To calculate the gradient of a scalar eld V(u
1
, u
2
, u
3
) in orthogonal curvilinear coordinates (u
1
, u
2
, u
3
), we
go back to the denition
dV = V dr . ()
for the change dV caused by an innitesimal position change dr.
(Note: here dV is the innitesimal change in scalar eld V resulting from a small change dr; it is not a
volume element. )
We dene V (V)
1
e
1
+(V)
2
e
2
+(V)
3
e
3
, and we want to nd the three components (V)
1
etc.
From the denitions of the unit vectors previously, we have dr = e
1
h
1
du
1
+e
2
h
2
du
2
+e
3
h
3
du
3
, so the
right-hand side of () becomes
((V)
1
e
1
+(V)
2
e
2
+(V)
3
e
3
) (e
1
h
1
du
1
+e
2
h
2
du
2
+e
3
h
3
du
3
)
= (V)
1
h
1
du
1
+(V)
2
h
2
du
2
+(V)
3
h
3
du
3
using the orthogonality of the es.
Now turning to the left-hand side of of (), using Taylors theorem (in 3 dimensions), and discarding
terms of second and higher derivatives, we get
dV =
V
u
1
du
1
+
V
u
2
du
2
+
V
u
3
du
3
These two expressions above must be equal for any arbitrary changes du
1
, du
2
and du
3
. Hence we must have
(V)
1
h
1
=
V
u
1
; (V)
2
h
2
=
V
u
2
; (V)
3
h
3
=
V
u
3
.
Dividing by the hs and substituting back into the original denition, in orthogonal curvilinear coordinates
we have
V =
1
h
1
V
u
1
e
1
+
1
h
2
V
u
2
e
2
+
1
h
3
V
u
3
e
3
. (5.7)
Clearly in Cartesian coordinates, we have u
1
= x, e
1
= i etc and all three hs are 1, so this simplies to
the well-known formula from Chapter 1.
For a geometrical explanation, the 1/h
i
terms take care of the arc-length effects, i.e. how far r moves for
a small change in each coordinate. So the 1-component of V represents the change dV per small distance
73
ds in the direction e
1
; but, moving a distance ds in direction e
1
requires a change u
1
= ds/h
1
in coordinate
u
1
; therefore the 1/h
i
terms appear in grad V above.
Example 5.6. What is V in spherical polar coordinates ? Evaluate V where V = r sin cos.
In spherical polars, (u
1
, u
2
, u
3
) = (r, , ) and h
1
= 1, h
2
= r, h
3
= r sin. Putting those into 5.7 we have
V =
V
r
e
r
+
1
r
V

+
1
r sin
V

.
For the given V, V/r = sin cos, V/ = r cos cos and V/ =r sin sin. Hence, using
the result above,
V = sin cos e
r
+cos cos e

sin e

.
(In this case we can observe that V = x and V = i, using the matrix from Eq. 5.3, so this example is a lot
easier in Cartesians; however, many problems involving circular or spherical symmetry do get easier in polar
coordinates).
Exercise 5.1. What is V in cylindrical polar coordinates (, , z) ? 2
Exercise 5.2. Let (r, , ) be spherical polar coordinates. Evaluate f where
(a) f = ; (b) f = ; (c) f = (r
n
sinm) .
2
5.8 The Divergence Operator in curvilinear coordinates
Next we want to compute F in orthogonal curvilinear coordinates. Although we could directly calculate
the divergence in any coordinates, using the Cartesian denition, the matrix relating basis unit vectors, and the
chain rule, the results can be found with less effort from the Divergence Theorem. The Divergence Theorem
is true in all coordinates (since it equates scalars, whose value must be independent of the coordinates). Thus
_
V
FdV =
_
S
F.dS ,
where S is the closed surface enclosing volume V.
Now, we apply this to an innitesimal cuboid with one corner at (u
1
, u
2
, u
3
) and edges corresponding to
changes u
1
, u
2
, u
3
in each coordinate; so this has eight corners at (u
1
, u
2
, u
3
), (u
1
+u
1
, u
2
, u
3
), . . . (u
1
+
u
1
, u
2
+u
2
, u
3
+u
3
). From before, the volume of the cuboid is V = (h
1
u
1
)(h
2
u
2
)(h
3
u
3
). For a
sufciently small volume, we can approximate F as constant across V, so the left-hand side becomes
( F)V = ( F)(h
1
h
2
h
3
u
1
u
2
u
3
) .
Next we consider the right-hand side of the Divergence Theorem: we need to take the surface integral
over the six faces of our cuboid, and add results. First consider the integral of F.n over the face of the
cuboid where the rst coordinate has value u
1
+u
1
. This face is a rectangle with unit normal +e
1
and area
(h
2
u
2
)(h
3
u
3
), so the surface integral is approximately
(h
2
h
3
u
2
u
3
F
1
)
u
1
+u
1
,
74
where the subscript shows it is evaluated at u
1
+u
1
. On the opposite face at u
1
we have unit normal e
1
(pointing outwards i.e. away from the rst face), so the surface integral gives us
(h
2
h
3
u
2
u
3
F
1
)
u
1
.
Repeating the above for the other four faces we get symmetrical results; nally summing the six terms and
then taking the limit as V 0, we obtain
F = lim
u
1
, u
2
, u
3
0
1
V
_
(h
2
u
2
h
3
u
3
F
1
)
u
1
+u
1
(h
2
u
2
h
3
u
3
F
1
)
u
1
+(h
3
u
3
h
1
u
1
F
2
)
u
2
+u
2
(h
3
u
3
h
1
u
1
F
2
)
u
2
+(h
1
u
1
h
2
u
2
F
3
)
u
3
+u
3
(h
1
u
1
h
2
u
2
F
3
)
u
3
_
.
Though each pair of brackets looks the same, this is not zero because the hs and Fs are different on opposite
faces of the cuboid; the rst two terms give us u
1
times the partial derivative /u
1
of the bracket, and so
on for the next pairs, so this gets us the result
F =
1
h
1
h
2
h
3
_
(h
2
h
3
F
1
)
u
1
+
(h
3
h
1
F
2
)
u
2
+
(h
1
h
2
F
3
)
u
3
_
. (5.8)
Note: In this last step, we have taken some us outside the brackets and cancelled them with the ones
in V, but we must leave the hs inside the differentiation since the hs generally vary with position. This
comes about because our cuboid may be slightly tapering, so the areas of opposite faces are not exactly
equal; and differentiating the h
i
s takes care of that.
Example 5.7. What is F in cylindrical polar coordinates, where F = F

+F

+F
z
e
z
?
In cylindrical polars, (u
1
, u
2
, u
3
) = (, , z) and h
1
= 1, h
2
= , h
3
= 1. Hence
F =
1

_
(F

+
F

+
(F
z
)
z
_
.
Note that we can apply the product rule, and since /z = 0, / = 1 we get
F =
1

+
F

+
1

+
F
z
z
.
Note: It is important to note that an F

term has appeared here, which is not a derivative of F. This has

appeared because the coordinate lines for have a built in divergence, they all radiate outwards from the
z-axis, so a eld with constant F

has a positive divergence term due to this.

As a further example we can note that in cylindrical polars, r = e

+0e

+ze
z
. Plugging in components
(, 0, z) to the above, we get
r = 1 +1 +0 +1 = 3
which agrees with the result in Cartesians, as it must.
(If we had just taken / +z/z we would have got r = 2; clearly wrong) .
Example 5.8. What is F in spherical polar coordinates, where F = F
r
e
r
+F

+F

?
In spherical polars, (u
1
, u
2
, u
3
) = (r, , ) and h
1
= 1, h
2
= r, h
3
= r sin. Hence
F =
1
r
2
sin
_
(r
2
sinF
r
)
r
+
(r sinF

+
(rF

_
.
75
5.9 The Curl Operator in curvilinear coordinates
Finally we want curl: as before we have curvilinear coordinates (u
1
, u
2
, u
3
), and a vector eld F = F
1
e
1
+
F
2
e
2
+F
3
e
3
; we want to calculate
F (F)
1
e
1
+(F)
2
e
2
+(F)
3
e
3
,
so we want the 1,2,3 components of the above.
In analogy with the previous section, we use Stokess theorem to provide a coordinate-independent de-
nition of F:
_
S
(F) dS =
_
C
F dr,
where S is a surface spanning the closed curve C.
To calculate the 1-component (F)
1
, consider a planar curve around a small rectangle" on a surface
of constant u
1
, with sides given by small changes u
2
and u
3
. From previous results, the vector area of this
rectangle dS = h
2
u
2
h
3
u
3
e
1
; now taking (F) dS, the 2 and 3 components of F disappear so the
LHS of Stokess theorem is approximately
(F)
1
h
2
u
2
h
3
u
3
.
Now looking at the RHS of Stokess theorem, the line integral around the edge of the same rectangle is given
by adding the line integrals along the four sides: this is approximately
(h
2
u
2
F
2
)
u
3
+(h
3
u
3
F
3
)
u
2
+u
2
(h
2
u
2
F
2
)
u
3
+u
3
(h
3
u
3
F
3
)
u
2
,
where the subscripts denote that the term is evaluated at that value, and two minus signs appear because oppo-
site sides are traversed in opposite directions around the closed rectangle. Equating the last two expressions,
and taking the limit as u
2
, u
3
0, we have
(F)
1
=
1
h
2
h
3
lim
u
2
,u
3
0
_
(h
3
F
3
)
u
2
+u
2
(h
3
F
3
)
u
2
u
2

(h
2
F
2
)
u
3
+u
3
(h
2
F
2
)
u
3
u
3
_
=
1
h
2
h
3
_
(h
3
F
3
)
u
2

(h
2
F
2
)
u
3
_
.
This is just the 1-component of F. To get the 2- and 3- components, we just repeat all the above
for two more small rectangles in surfaces of constant u
2
, u
3
respectively; this looks the same but cycling the
1/2/3s, and we get
(F)
2
=
1
h
3
h
1
_
(h
1
F
1
)
u
3

(h
3
F
3
)
u
1
_
,
(F)
3
=
1
h
1
h
2
_
(h
2
F
2
)
u
1

(h
1
F
1
)
u
2
_
.
These results can be written in a compact (and more memorable) form as a determinant:
F =
1
h
1
h
2
h
3

h
1
e
1
h
2
e
2
h
3
e
3
/u
1
/u
2
/u
3
h
1
F
1
h
2
F
2
h
3
F
3

. (5.9)
Once again, in Cartesian coordinates this simplies to the well-known expression from Chapter 3.4.
Example 5.9. What is F in spherical polar coordinates?
76
In spherical polar coordinates (r, , ) we have h
1
= 1, h
2
= r, h
3
= r sin. Hence, using the determinant
form:
F =
1
r
2
sin

e
r
re

r sine

/r / /
F
r
rF

r sinF

.
or in expanded form
F =
1
r
2
sin
_
(r sinF

(rF

_
e
r
+
1
r sin
_
F
r

(r sinF

)
r
_
e

+
1
r
_
(rF

)
r

F
r

_
e

.
Note that since r is independent of and , etc., we can for instance take the r outside the differentiations in
the e
r
component and cancel it with an r in the denominator. Remember the answer is a curl so its a vector
eld. Do not add all the components together, forgetting the vectors e
r
etc (this is a common error).
Note: the full expression above looks quite daunting. However in many problems this may simplify
considerably using symmetry: for example, if a given problem is symmetrical around the z-axis, then we will
have F

= 0 and F
r
/ = 0 and F

/ = 0, so four of the six derivatives will vanish.

Exercise 5.3. Show by expanding it that the determinant denition is equivalent to the full expressions
for the individual components given above. 2
Exercise 5.4. What is F in cylindrical polar coordinates?
Note that if and z have dimensions of length and is dimensionless (because its an angle), then all the
terms in the expression for F should have the same dimensions, namely the dimensions of F divided by
length. This is a simple check that you should make. 2
Exercise 5.5. Use spherical polar coordinates to evaluate the divergence and curl of r/r
3
. [Hint: dont
forget that in spherical polar coordinates, the position vector r is equal to re
r
.] 2
Exercise 5.6. State Stokess theorem, and verify it for the hemispherical surface r = 1, z 0, with the
vector eld A(r) = (y, x, z). 2
Exercise 5.7. The vector eld B() = (0,
1
, 0) in cylindrical polar coordinates (, , z). Evaluate
B. Evaluate the line integral
_
C
B.dr, where C is the unit circle z = 0, = 1, 0 2. Does Stokess
theorem apply? 2
Note: To conclude this chapter, we will note that many applied maths or Physics problems involve an
expression like
2
V, where V is a scalar eld and
2
is the Laplacian operator, in cylindrical or spherical
polar coordinates. We can get the expressions for
2
V in polar coordinates using rstly the denition Eq. 3.10
(recall this was
2
V div(gradV)) , and then using Eq. 5.7 for grad V, then taking div of that with Eq. 5.8.
The results are available in most textbooks; you will not be expected to memorise those, but you might
be given them in an exam question and asked to calculate something, so its worth taking a look especially if
you are taking applied maths courses later.
77
Appendix
Other ways of doing Example 5.5 are as follows
The second method is to divide the volume removed into two parts: (i) a cylinder with radius sin
1
and
height cos
1
, and (ii) a top-slice. Volume (i), the cylinder, is easy: 2 sin
2

1
cos
1
. To get volume (ii) we
integrate over and then z
4
_
1
cos
1
dz
_

1z
2
0
d = 2
_
1
cos
1
(1 z
2
)dz =
2
3
(2 +cos
3

1
3cos
1
).
The sum of volumes (i) and (ii) is
4
3
(1 cos
3

1
) as expected.
A third way also divides the volume removed into two parts: (i) an ice-cream cone or cone with a
spherical top, and (ii) a cylinder minus cone. The volume (i) is
4
_

1
0
sin d
_
1
0
r
2
dr =
4
3
(1 cos
1
).
Volume (ii), a cylinder with cone removed, is a bit harder:
4
_
cos
1
0
dz
_
sin
1
ztan
1
d = 2
_
cos
1
0
(sin
2

1
z
2
tan
2

1
)dz =
4
3
sin
2

1
cos
1
(which notice is
2
3
of the volume of the cylinder). Again the sum of the volumes integrated is
4
3
(1cos
3

1
).
Finally, a fourth possibility is to integrate for the volume remaining after coring, which is
4
_
cos
1
0
dz
_

1z
2
sin
1
d = 2
_
cos
1
0
(1 z
2
sin
2

1
)dz =
4
3
cos
3

1
.
78
SUMMARY OF ORTHOGONAL CURVILINEAR COORDINATES
In orthogonal curvilinear coordinates (u
1
, u
2
, u
3
), with corresponding unit vectors e
1
, e
2
, e
3
and arc-
length parameters h
1
, h
2
, h
3
, the gradient of a scalar eld V is given by
V =
1
h
1
V
u
1
e
1
+
1
h
2
V
u
2
e
2
+
1
h
3
V
u
3
e
3
;
the divergence of a vector eld F = F
1
e
1
+ F
2
e
2
+ F
3
e
3
is given by
F =
1
h
1
h
2
h
3
_

u
1
(h
2
h
3
F
1
) +

u
2
(h
3
h
1
F
2
) +

u
3
(h
1
h
2
F
3
)
_
;
and the curl of the same vector eld is given by
F =
1
h
1
h
2
h
3

h
1
e
1
h
2
e
2
h
3
e
3
/u
1
/u
2
/u
3
h
1
F
1
h
2
F
2
h
3
F
3

.
Cartesian coordinates:
(u
1
, u
2
, u
3
) (x, y, z) ; arc-length parameters h
1
= 1, h
2
= 1, h
3
= 1 .
Cylindrical polar coordinates:
(u
1
, u
2
, u
3
) (, , z) ; arc-length parameters h
1
= 1, h
2
= , h
3
= 1 .
Spherical polar coordinates:
(u
1
, u
2
, u
3
) (r, , ) ; arc-length parameters h
1
= 1, h
2
= r, h
3
= r sin .
79