Integration on manifolds 1

Chapter 11 Integration on manifolds

We are now almost ready for our concluding chapter on the great theorems of classical
vector calculus, the theorems of Green and Gauss and Stokes.
The nal thing we need to understand is the correct procedure for integrating over a
manifold. Of course, manifolds are typically curved objects, so there are signicant issues
here that we did not have to face in dealing with integration over (at) Euclidean space. Even
though we have been able to use integration to compute such things as the volume of a ball
in R
n
, yet the integration involved is essentially at. This is reected in a formula such as
dvol
n
= dx
1
dx
2
. . . dx
n
.
As a specic example, we have for the ball B(0, r) in R
3
,
vol
3
(B(0, r)) =
4
3
r
3
,
but we have not yet even dened what we should mean by the area of the sphere,
vol
2
(S(0, r)).
It will in fact turn out that there is a very reasonable way to accomplish this task, and it will
be based on our knowledge of at integration.
We choose to employ the parametric presentation of manifolds as discussed thoroughly
in Section 6E. It is in this context that we shall see how to dene volume and integration.
Though our denition is to be based on a particular choice of parameters, we shall be able to
prove that the integration we dene is actually invariant under a change of parameters and is
thus intrinsic to the manifold.
We denote by M an m-dimensional manifold contained in R
n
. In what follows we shall
rst investigate how to dene the m-dimensional volume of subsets of M, and after that we
shall easily dene integrals of real-valued functions dened on M.
We begin with sort of a warm up case, that of one-dimensional manifolds in R
n
.
A. The one-dimensional case
We assume in this section that m = 1, so we are dealing essentially with a curve M R
n
.
We eectively already know exactly how to deal with this case, thanks to Section 2B.
If M is represented parametrically as the image of a one-to-one function of class C
1
,
(a, b)
F
M,
2 Chapter 11
then the length of M is given as
vol
1
(M) =
_
b
a
F

(t)dt.
Of course, M R
n
and F

(t) R
n
and F

(t) is the norm of the vector F

(t).
We really need to say no more about this denition except to remark that if we imagine the
interval (a, b) to be partitioned into small pieces (, ), then F sends (, ) to an approximate
interval in R
n
whose length should be approximately F

()(). Thus F

() represents
a scale factor relating the parameter length to a length in R
n
. A useful way to represent
this denition is
dvol
1
= F

(t)dt.
As we discussed in Section 2B, the denition of vol
1
(M) is independent of the choice of
parametric representation of M.
PROBLEM 111. Suppose M R
2
is a curve described in polar coordinates by an
equation r = g(), where a b. Show that the length of M is
_
b
a
_
g

()
2
+ g()
2
d.
(This formula is usually written as
_
b
a

_
dr
d
_
2
+ r
2
d
or even as
_
b
a
_
(dr)
2
+ r
2
(d)
2
.
_
Integration on manifolds 3
PROBLEM 112. A certain curve in R
2
is described in polar coordinates by the
equation
r =

cos 2.
This gure eight curve is called a lemniscate. (See Problem 26.) Show that its length
is equal to

_
1
4
_

_
3
4
_ =

2
_
1
4
_

2
.
PROBLEM 113. Show that
_
/2
0
d

1 + sin
2

_
1
4
_

_
3
4
_.
(HINT: substitute x = sin and then x = t
1/4
.)
PROBLEM 114. Show that
_
/2
0
sin
2

1 + sin
2

d =

_
3
4
_

_
1
4
_.
PROBLEM 115. Combine the preceding two problems to show that
_
/2
0
_
1 + sin
2
d =

_
1
4
_

_
3
4
_ +

_
3
4
_

_
1
4
_.
PROBLEM 116. The two cylinders x
2
+z
2
= 1 and x
2
+y
2
= 1 intersect in a certain
curve. Find its length.
(Answer: 8 times the answer to the preceding problem.)
Finally, we easily extend the denition of the length of M to the more general denition
4 Chapter 11
of integration of a function over M. Suppose then that M
f
R. Then we write
_
M
fdvol
1
=
_
b
a
f(F(t))F

(t)dt.
In this particular case this is often called an arc-length integral, and is written
_
M
fds or
_
M
fd.
A frequent application of these integrals is the calculation of the average value of f over
M, dened to be

f =
1
vol
1
(M)
_
M
fdvol
1
.
Notice the easy facts
f + g =

f + g,
cf = c

f if c is a constant,

1 = 1.
DEFINITION. The centroid of the curve M is the point x R
n
whose coordinates are
given by
x
i
= average of x
i
over M
=
1
vol
1
(M)
_
M
x
i
dvol
1
.
In other words, x is the average position in R
n
of points in M. Note that it is not to be
expected that x necessarily belongs to M.
PROBLEM 117. Calculate the centroid of a semicircle.
PROBLEM 118. Calculate the centroid of the right-half of the lemniscate of
Problem 112.
(Answer: x=4

/
2
(1/4), y=0.)
B. The general case
Integration on manifolds 5
Amazingly, what we have accomplished in the one-dimensional case generalizes almost
immediately to m dimensions. We thus suppose M R
n
is an m-dimensional manifold. We
also suppose that M is given by a parametric presentation of class C
1
,
A
F
R
n
, F(A) = M,
where A is a contented subset of R
m
. For short we may write x = F(t), so that the coordinates
of x are given real-valued functions of t = (t
1
, . . . , t
m
). In our discussion we must have
1 m n; the case m = n is tantamount to a discussion of integration over R
n
itself.
In every situation our goal is the following: to use the calculus associated with F to nd
a scale factor J(t) which converts the m-dimensional innitesimal volume dt = dt
1
dt
m
of the parameter space to the corresponding m-dimensional volume of M. We shall write this
correspondence in the form
dvol
m
= J(t)dt.
We have chosen the notation J(t) to remind ourselves that this scale factor is like a Jacobian
determinant. In this and other formulas of a similar nature we shall be very insistent upon
gaining an understanding of the Jacobian J(t).
As in the one-dimensional case, what we really need to do is see how F transforms small
rectangles in A and in particular what scale factor to multiply by to compute the approximate
m-dimensional volume of the image of the rectangle. As usual, let e
1
, . . . , e
m
denote the unit
coordinate vectors in R
m
. Then consider a special rectangle
[t
1
, t
1
+
1
] [t
m
, t
m
+
m
],
where t is a vertex and the edge lengths are
1
, . . . ,
m
. We think of the
i
s as small positive
numbers.
The corresponding image of this rectangle with respect to F is approximately an m-
dimensional parallelogram with one vertex F(t) and edges given by

1
F
t
1
(t), . . . ,
m
F
t
m
(t):
6 Chapter 11
Of course this at parallelogram does not lie in the manifold M. However, it is tangent to M
at the point F(t).
We know how to compute the m-dimensional volume of this parallelogram: from Section 8C
it is
_
Gram(
1
F/t
1
, . . . ,
m
F/t
m
) =
1
. . .
m
_
Gram(F/t
1
, . . . , F/t
m
).
Since the factor
1
. . .
m
is the volume of the rectangle in the parameter space, we see imme-
diately that the scale factor for computing the volume of M is precisely
_
Gram(F/t
1
, . . . , F/t
m
).
Since we think of
i
as dt
i
in some sense, and thus
1

m
is like dt
1
dt
m
= dt, we are thus
led to the
DEFINITION. Given the C
1
parametric presentation of the m-dimensional manifold M
R
n
A
F
M,
where A is a contented subset of R
m
,
vol
m
(M) =
_
A
_
Gram(F/t
1
, . . . , F/t
m
)dt
1
. . . dt
m
.
More generally, if f is a real-valued function dened on M, its integral over M is given by
_
M
fdvol
m
=
_
A
f(F(t))
_
Gram(F/t
1
, . . . , F/t
m
)dt.
Integration on manifolds 7
We summarize both these equations by simply writing
dvol
m
=
_
Gram(F/t
1
, . . . , F/t
m
) dt.
This equation gives the scale factor J(t) explicitly in this general setting.
There is a nice alternate description of the scale factor described above. Namely, use the
Jacobian matrix
DF =
_
_
_
F
1
t
1
. . .
F
1
t
m
.
.
.
.
.
.
F
n
t
1
. . .
F
n
t
m
_
_
_
(see Section 2H). This Jacobian matrix is an n m matrix, and the matrix product
(DF)
t
DF
is precisely the m m Gram matrix (F/t
i
F/t
j
). Thus the scale factor in the above
denition can also be written in the form
J(t) =
_
det(DF)
t
DF.
EXAMPLES. For the case of curves, m = 1, this is the same denition we used before, as
Gram(F/t) = Gram(F

(t)) = F

(t) F

(t), so the scale factor is F

(t).
For the case of 2-dimensional manifolds, we have
vol
2
(M) =
_
A
_
F/t
1

2
F/t
2

2
(F/t
1
F/t
2
)
2
dt.
Let us examine the sphere S(0, a) of radius a in R
3
. Let us use spherical coordinates in the
usual way, so that
F(, ) = (a sin cos , a sin sin , a cos ).
Then
F/ = a(cos cos , cos sin , sin ),
F/ = a sin (sin , cos , 0).
The scale factor is therefore
_
a
2
(a sin )
2
0 = a
2
sin .
8 Chapter 11
Thus for this sphere we have
dvol
2
= a
2
sin dd.
In particular, the area of the sphere is
vol
2
(S(0, a)) =
_
2
0
_

0
a
2
sin dd
= 4a
2
,
a familiar result to be sure.
The example of a right circular cylinder in R
3
is even easier, but it is so important we
should pause to consider it. Let us use the cylinder of radius a with axis given as the z-axis:
x
2
+ y
2
= a
2
.
It is convenient to use cylindrical coordinates (no surprise!), which we call , t:
_

_
x = a cos ,
y = a sin ,
z = t.
We thus have
F(, t) = (a cos , a sin , t);
F

= a(sin , cos , 0);

F
t
= (0, 0, 1);
J(, t) = a;
dvol
2
= addt.
Integration on manifolds 9
PROBLEM 119. Consider the right circular cone as in Problem 629:
z = cot
_
x
2
+ y
2
.
z
In terms of cylindrical coordinates , z show that for this manifold
dvol
2
= z tan sec ddz.
Then show that the area of a right circular cone whose base has circumference c and
whose slant height is equals
1
2
c.
Just as we have done in the one-dimensional case of the preceding section, we need to check
that our denition is independent of the choice of parametrization of M. In other words, we
must verify that if we change parameters we do not change the value of the integral. To be
sure, our geometric intuition tells us that all is well, but we should do the work to certify the
formulas.
Thus we suppose that M is parametrized in two manners:
10 Chapter 11
There is a parameter change involved here, namely
h(s) = F
1
(G(s)).
That is, h = F
1
G; we may represent
t = h(s),
and h is a C
1
function from B onto A and has a C
1
inverse.
THEOREM. In the above situation the integral
_
M
fdvol
m
has the same value whether computed using F or G.
PROOF. By now we realize that this proof will probably amount to no more than an
automatic calculation. And so it does. The scale factor in using G is the square root of
det(DG)
t
DG = det D(F h)
t
D(F h)
= det (((DF) h)Dh)
t
((DF) h)Dh (chain rule)
= det(Dh)
t
((DF) h)
t
((DF) h)Dh
= det(Dh)
t
det[((DF) h)
t
((DF) h)] det Dh
= (det(Dh))
2
[det(DF)
t
DF] h.
Integration on manifolds 11
Thus we obtain
_
B
f(G(s))
_
det(DG(s))
t
DG(s)ds
=
_
B
f(F(h(s)))| det Dh(s)|
_
det(DF)
t
DF(h(s))ds
=
_
A
f(F(t))
_
det DF(t)
t
DF(t)dt.
This is the desired equation. Notice that the last step used the change of variables formula
for integrals over R
m
, t = h(s) and dt = | det h

(s)|ds, which we introduced in Section 10G.

QED
In summary, it should be stressed that the thing which shows that our denition of inte-
gration over manifolds makes sense is precisely the change of variables formula of Section 10G.
PROBLEM 11-10. The at torus in R
4
is a nice example of a two-dimensional
manifold. See Example 3 of Section 5G for the denition. Calculate its two-dimensional
volume (its area).
C. Hypermanifolds
We now examine the important special case of a manifold M R
n
of dimension n1. The
scale factor for volume calculations has a particularly nice form. We rst give a calculation:
LEMMA. det(
ij
+ a
i
a
j
) = 1 +

j
a
2
j
.
PROOF. This is a special case of Problem 341; nevertheless we provide a proof here.
Suppose we are dealing with an n n matrix. Then its j
th
column is e
j
+a
j
a, where a is the
column vector with coordinates a
1
, . . . , a
n
. The linearity property of det (Section 3F), implies
that the determinant we want,
det ( e
1
+ a
1
a e
2
+ a
2
a e
n
+ a
n
a) ,
is equal to the sum of 2
n
terms, each term being the determinant of a matrix with either
e
j
or a
j
a in the j
th
column. If a scalar multiple of a appears in at least two columns, the
corresponding determinant is zero. Thus only n of the terms of this sum survive, and we
12 Chapter 11
obtain for the above determinant
det I +
n

j=1
det ( e
1
a
j
a e
n
)
= 1 +
n

j=1
a
j
det ( e
1
a e
n
)
= 1 +
n

j=1
a
j
det ( e
1
a
j
e
j
e
n
)
= 1 +
n

j=1
a
2
j
.
QED
PROBLEM 1111. For the vectors e
j
+ a
j
e
n
, 1 j n 1, show that
Gram( e
1
+ a
1
e
n
, . . . , e
n1
+ a
n1
e
n
) = 1 +
n1

j=1
a
2
j
.
Now suppose that the hypermanifold M R
n
is presented as the graph of a function
A

R, where A R
n1
. Then we obtain an explicit parametrization of M with A
F
R
n
,
where
F(x
1
, . . . , x
n1
) = (x
1
, . . . , x
n1
, (x
1
, . . . , x
n1
)).
Therefore
F
x
j
= e
j
+

x
j
e
n
, 1 j n 1.
Problem 1111 immediately gives
Gram(F/x
1
, . . . , F/x
n1
) = 1 +
n1

j=1
(/x
j
)
2
.
Thus we can compute volume integrals over M in terms of the coordinates x
1
, . . . , x
n1
, by
dvol
n1
=

_
1 +
n1

j=1
(/x
j
)
2
dx
1
. . . dx
n1
.
Integration on manifolds 13
PROBLEM 1112. Compute the area of the portion of the surface z = xy contained
inside the cylinder x
2
+ y
2
3.
(Answer: 14/3.)
PROBLEM 1113. Compute the area of the portion of the paraboloid 2z = x
2
+ y
2
contained inside the cylinder x
2
+ y
2
3.
We can now immediately generalize this result to hypermanifolds presented implicity.
THEOREM. Suppose M R
n
is a hypermanifold presented as the level set
g(x) = 0,
where g is a function on R
n
of class C
1
satisfying g = 0 on M. Suppose for instance that
g/x
n
= 0. Then integration over M can be calculated by means of the formula
dvol
n1
=
g
|g/x
n
|
dx
1
. . . dx
n1
.
PROOF. The implicit function theorem guarantees the local existence of a function R
n1

R such that
g(x) = 0 x
n
= (x
1
, . . . , x
n1
).
Furthermore, dierentiating the equation
g(x
1
, . . . , x
n1
, (x
1
, . . . , x
n1
)) = 1
gives the formula
/x
j
=
g/x
j
g/x
n
.
Thus the scale factor found above in terms of can be manipulated to give

_
1 +
n1

j=1
(/x
j
)
2
=

_
1 +
n1

j=1
(g/x
j
)
2
/(g/x
n
)
2
=
1
|g/x
n
|

_
(g/x
n
)
2
+
n1

j=1
(g/x
j
)
2
.
14 Chapter 11
QED
EXAMPLE. Let the manifold be the sphere S(0, a) R
n
. This is an (n 1)-dimensional
manifold and is described implicitly by the equation
x
2
a
2
= 0.
For the function we have
g = 2x, g/x
n
= 2x
n
.
In order to cope with g/x
n
= 0 we shall nd the volume of the upper hemisphere x
n
> 0
and then double the answer. Thus
vol
n1
(S(0, a)) = 2
_
2x
2x
n
dx
1
. . . dx
n1
= 2
_
a
x
n
dx
1
. . . dx
n1
= 2
_
a
_
a
2
x
2
1
x
2
n1
dx
1
. . . dx
n1
,
where the integration region is the ball x
2
1
+ +x
2
n1
< a
2
in R
n1
. From Problem 1049 we
therefore obtain
vol
n1
(S(0, a)) = 2(n 1)
n1
_
a
0
a

a
2
r
2
r
n2
dr.
Substitute r = a

t to obtain
vol
n1
(S(0, a)) = (n 1)
n1
_
1
0
1

1 t
a
n2
t
n2
2
at

1
2
dt
= (n 1)
n1
a
n1
_
1
0
1

1 t
t
n3
2
dt
= (n 1)
n1
a
n1
B
_
n 1
2
,
1
2
_
= (n 1)

(n1)/2

_
n+1
2
_ a
n1

_
n1
2
_

_
n
2
_
=
2
n/2

_
n
2
_ a
n1
.
Integration on manifolds 15
We record this result for later reference:
vol
n1
(S(0, 1)) =
2
n/2
(n/2)
= n
n
.
For instance,
vol
2
(S(0, 1)) = 4,
vol
3
(S(0, 1)) = 2
2
,
vol
4
(S(0, 1)) =
8
2
3
.
EXAMPLE. Here we simply reconsider the sphere S(0, a) in R
3
. In Section B we calculated
the area formula
dvol
2
= a
2
sin dd
in terms of the usual spherical coordinates. It is quite interesting to think of this result also
in terms of cylindrical coordinates for the sphere: x
2
+ y
2
= a
2
z
2
, so that
_

_
x =

a
2
z
2
cos ,
y =

a
2
z
2
sin ,
z = z.
Of course we assume here that a < z < a. Then we can calculate J(z, ) from scratch, or
we can instead observe that in terms of the angle we have z = a cos ; thus dz = a sin d
(weve included the absolute value). The result is that
dvol
2
= adzd.
This is a truly fascinating result. The expression for this area is exactly that of the circum-
scribed cylinder
x
2
+ y
2
= a
2
!
An interesting application of this result has to do with constructing maps of the sphere
onto planes. If we project the sphere onto the cylinder by moving in horizontal straight lines
from the z-axis, then area is preserved! The precise formula for this map is
F(x, y, z) =
_
ax

a
2
z
2
,
ay

a
2
z
2
, z
_
.
16 Chapter 11
Its inverse is
F
1
(x, y, z) =
_

a
2
z
2
x
a
,

a
2
z
2
y
a
, z
_
.
this is an amoeba on the sphere
this is its projection onto a cylinder
Then the map onto a plane is obtained by cutting the cylinder along a generator and unrolling
it.
An interesting and curious consequence comes from the above observation. Namely, a zone
on the sphere is dened as a set of the form
{(x, y, z) | x
2
+ y
2
+ z
2
= a
2
, z
1
z z
2
}.
Here a z
1
< z
2
a. The area of this zone is precisely 2a(z
2
z
1
), as that is the
corresponding area of the circumscribed cylinder. Thus the area of a zone depends only on
its height z
2
z
1
, and not on its location.
EQUAL
AREAS
Integration on manifolds 17
PROBLEM 1114. Let H be the unit hemisphere in R
3
described as x
2
+y
2
+z
2
= 1,
0 z 1. Let C be the boundary equator x
2
+ y
2
= 1, z = 0. Let 0

2
and let M
be the portion of H which lies over the triangle shown in the gure. That is, in standard
spherical coordinates M is described by
_

_
0

2
,
,
sin cos cos .
Show that the area of M equals

y
x
C
vol
2
(M) = 2 + cos .
(HINT: do no integration.)
PROBLEM 1115. (This problem is based on Problem B3 from the William Lowell
Putnam Mathematical Competition for 1998.) Continue with the hemisphere H of the
preceding problem. Let P be a regular n-gon inscribed in the equator C. Determine the
surface area of that portion of H lying over the planar region inside P.
(Answer: 2n+n cos /n)
PROBLEM 1116. Find the area of the portion of the surface of the cylinder x
2
+y
2
=
1 in R
3
which is contained inside the cylinder x
2
+ z
2
1.
(Answer: 8)
PROBLEM 1117. Find the area of the portion of the surface of the cylinder x
2
+y
2
=
1 in R
3
which is contained in the intersection of the two cylinders x
2
+z
2
1, y
2
+z
2
1.
(Answer: 168

2)
18 Chapter 11
PROBLEM 1118

. A surface M R
3
is described implicitly by the equation
(x
2
+ y
2
+ z
2
)
2
= x
2
y
2
.
Prove that this is actually a two-dimensional manifold at each point except the origin.
Compute its area.
(Answer:
2
/2)
PROBLEM 1119. Consider these two surfaces in R
3
: the unit sphere x
2
+y
2
+z
2
= 1
and the right circular cylinder x
2
+ y
2
= y.
a. Find the area of the portion of the unit sphere that is inside the cylinder.
b. Find the area of the portion of the cylinder that is inside the sphere.
PROBLEM 1120. Let M be the elliptical cone in R
3
dened by the equation
z
2
c
2
=
x
2
a
2
+
y
2
b
2
and the inequality 0 z h. Show that the centroid of M is (0, 0, 2h/3). (You will not
be able to calculate the area of M.)
D. The Cauchy-Binet determinant theorem
This section is not very important for our development of integration, but is so fascinating
that it should be included.
Integration on manifolds 19
We start with an interpretation of the formula from Section 7B: for any two vectors a,
b R
n
Gram(a, b) =

i<j
(a
i
b
j
a
j
b
i
)
2
.
We know of course that the left side equals the square of the area of the parallelogram P with
vertices 0, a, b, and a +b (Section 8C). Now consider the orthogonal projection of P onto the
coordinate space R
2
, namely the x
i
x
j
plane. This projection P(i, j) is the parallelogram
with vertices
(0, 0), (a
i
, a
j
), (b
i
, b
j
), and (a
i
+ b
i
, a
j
+ b
j
).
x
i
x
j
(b
i
, b
j
)
(a
i
, a
j
)
The square of its area is of course the corresponding Gram determinant,
det
_
a
2
i
+ a
2
j
a
i
b
i
+ a
j
b
j
a
i
b
i
+ a
j
b
j
b
2
i
+ b
2
j
_
= (a
i
b
j
a
j
b
i
)
2
.
Thus we have the interesting fact that the square of the area of P is equal to the sum of
the squares of the areas of all the projections P(i, j), for 1 i j n. This is sort of a
Pythagorean theorem for two-dimensional parallelograms in R
n
.
The above result is a special case of the theorem of this section. Before discussing it let us
see what our special case implies about the change of variables formula for two-dimensional
manifolds. We use the parametric presentation of M R
n
, but instead of naming the presen-
tation function we simply denote its components as x
1
, . . . , x
n
regarded as functions of t
1
, t
2
.
Then we have from the above equation
(J(t
1
, t
2
))
2
=

i<j
_
det
_
x
i
/t
1
x
i
/t
2
x
j
/t
1
x
j
/t
2
__
2
.
In terms of the notation mentioned in Section 10I, this says
(J(t
1
, t
2
))
2
=

i<j
_
(x
i
, x
j
)
(t
1
, t
2
)
_
2
.
20 Chapter 11
This makes the volume formula look like a rather natural generalization of the formula for
changing variables in an ordinary integral over R
2
:
dvol
2
=

i<j
_
_
(x
i
, x
j
)
(t
1
, t
2
)
_
2
_
1/2
dt
1
dt
2
.
The above result does not appear to be particularly useful for calculations, but its elegance
is very interesting. The algebraic generalization we now present is likewise not something we
shall be using, but we present it here to display a beautiful piece of algebra.
Instead of m = 2, we now consider the general case of 1 m n. For the present
discussion we shall use the notation u to stand for a generic m-tuple of increasing integers in
the range [1, n]:
u = (u
1
, . . . , u
m
), 1 u
1
< u
2
< < u
m
n.
Given an m n matrix A and an index u, we can construct a (square) m m matrix A(u)
by using only the columns of A corresponding to u:
A(u) =
_
_
_
a
1u
1
. . . a
1u
m
.
.
.
.
.
.
a
mu
1
. . . a
1u
m
_
_
_
.
Likewise, given an n m matrix B we construct B(u) by selecting only the rows of B corre-
sponding to u:
B(u) =
_
_
_
b
u
1
1
. . . b
u
m
m
.
.
.
.
.
.
b
u
m
1
. . . b
u
m
m
_
_
_
.
Here is the amazing result:
THEOREM. If 1 m n and A is an mn matrix and B is an n m matrix, then
det AB =

u
det A(u) det B(u).
PROOF. The ij entry of AB is of course

k
a
ik
b
kj
,
Integration on manifolds 21
where the sum extends for 1 k n. Therefore we can write the j
th
column of AB in the
form
(AB)
column j
=

k
b
kj
(A)
column k
.
Now we use the linearity of det with respect to its columns, the multilinear property of
Section 3F. For each column of AB we need a distinct summation index; so in the above
formula for the j
th
column we change k to k
j
. Then we have
det AB =

k
1
,...,k
m
b
k
1
1
b
k
m
m
det ((A)
col k
1
(A)
col k
m
) .
In this huge sum each k
j
runs from 1 to n.
Now think about the determinants in the sum. If two of the k
j
s are equal, the corre-
sponding matrix has two identical columns so the determinant is zero. Thus we may restrict
the sum to distinct k
1
, . . . , k
m
. These distinct choices, if reordered, correspond to the various
increasing sets of indices u. For a given u, there are m! distinct orderings k
1
, . . . , k
m
, and each
determinant in the sum will be det A(u). Thus we can rewrite the above formula as
det AB =

u
f(B, u) det A(u),
where f(B, u) just stands for some number that depends only on B and u (and not on A).
Now its easy to compute each coecient by just picking a particular A. Given a particular
u choose A to be that matrix dened by
_
a
iu
j
=
ij
,
a
ik
= 0 if k = u
1
, . . . , u
m
.
That is, the u
th
j
column of A is the coordinate vector e
j
R
n
, and all other columns of A are
0. Thus all det A(u

) except u

= u are zero, and

det AB = f(B, u) det A(u)
= f(B, u) det I
= f(B, u).
On the other hand the ij entry of AB is

k
a
ik
b
kj
= a
iu
1
b
u
1
j
+ + a
iu
m
b
u
m
j
=
i1
b
u
1
j
+ +
im
b
u
m
j
= b
u
i
j
.
22 Chapter 11
That is, AB = B(u). This proves that f(B, u) = det B(u).
QED
REMARKS. 1. This result in case m = n just asserts that for square matrices det AB =
det Adet B. Thus we have given a third proof of this crucial property of determinants!
2. This proof should remind you of the proof of the lemma of Section C. In fact, in that
notation
(
ij
+ a
i
a
j
) = B
t
B,
where
B =
_
_
_
_
_
_
_
1 0 . . . 0
0 1 0
.
.
.
.
.
.
.
.
.
0 0 1
a
1
a
2
a
n
_
_
_
_
_
_
_
is an (n + 1) n matrix. Note that
B
_
(1, . . . , n)
_
= I,
B
_
(1, . . . , k 1, k + 1, . . . , n + 1)
_
=
_
_
_
_
_
1 0 . . . 0
.
.
.
0 0 . . . 1
a
1
a
2
. . . a
n
_
_
_
_
_
_
k
th
row omitted
,
so
det B
_
(1, . . . , n)
_
= 1,
det B
_
(1, . . . , k 1, k + 1, . . . , n + 1)
_
= a
k
.
Thus the lemma is a special case of the theorem.
3. In general, in case A = B
t
, we have for mn matrices with 1 m n,
det AA
t
=

u
_
det A(u)
_
2
.
4. Consider vectors x, y, u, v in R
3
and dene
A =
_
x
1
x
2
x
3
y
1
y
2
y
3
_
,
B =
_
_
u
1
v
1
u
2
v
2
u
3
v
3
_
_
.
Integration on manifolds 23
Then
AB =
_
x u x v
y u y v
_
,
so the theorem gives
det
_
x u x v
y u y v
_
= det
_
x
1
x
2
y
1
y
2
_
det
_
u
1
v
1
u
2
v
2
_
+ det
_
x
1
x
3
y
1
y
3
_
det
_
u
1
v
1
u
3
v
3
_
+ det
_
x
2
x
3
y
2
y
3
_
det
_
u
2
v
2
u
3
v
3
_
.
This is precisely the result of Problem 73, the generalization of Lagranges identity.
E. Miscellaneous applications
1. Surfaces of revolution
We are going to examine a special class of surfaces in R
3
obtained by revolving about an
axis. We may as well use the z-axis as the axis of revolution. We then start with a curve of
class C
1
lying in the x z plane, where all points of have positive x-coordinates.
x
z

We revolve the points of about the z-axis through a full angle of 360

and thus obtain a

2-dimensional manifold (a surface) M. We want to investigate integration on M.
Let = (t) be a parametrization, where a t b. We assume does not intersect itself,
except that perhaps is a closed curve ((a) = (b)). We shall designate the coordinates of
as
(t) = (x(t), z(t)).
Thus, x(t) > 0.
24 Chapter 11
Let denote the angle of rotation about the z-axis. Then M can be parametrized as
follows:
F(t, ) =
_
x(t) cos , x(t) sin , z(t)
_
.
This is a representation using cylindrical coordinates, where the distance from the axis is x(t).
Then we compute easily
F
t
=
_
x

(t) cos , x

(t) sin , z

(t)
_
,
F

= x(t)(sin , cos , 0).

These tangent vectors are orthogonal, so the corresponding Gram determinant is
_
_
_
F
t
_
_
_
2
_
_
_
F

_
_
_
2
=
_
x

(t)
2
+ z

(t)
2
_
x(t)
2
.
Thus the Jacobian factor for dvol
2
is
J(t, ) = x(t)
_
x

(t)
2
+ z

(t)
2
.
Hence we obtain for instance
area(M) = vol
2
(M)
=
_
b
a
_
2
0
x(t)
_
x

(t)
2
+ z

(t)
2
ddt
= 2
_
b
a
x(t)
_
x

(t)
2
+ z

(t)
2
dt.
Another way of saying this is
area(M) = 2
_

xds.
Integration on manifolds 25
A surface of revolution.
PROBLEM 1121. Instead of revolving a curve, we can revolve a two-dimensional
region around an axis. Suppose that R is such a contented subset of the x z plane and
again suppose that for all (x, z) R, x > 0. Then revolve R around the z-axis through
360

, obtaining the solid of revolution D:

D = {(x cos , x sin , z) | (x, z) R, 0 2}.
Prove that
vol
3
(D) = 2
_
R
xdxdz.
2. Pappus theorems
Pappus of Alexandria lived about 290350, and has been described as the last of the great
Greek geometers. Here is one of his theorems, which is known as the basis of modern projective
geometry:
26 Chapter 11
If the vertices of a hexagon lie
meets of opposite sides are
collinear.
alternately on two lines, then the
Pappuss theorem states:
Just for fun, you might try proving this for yourself.
The theorems of Pappus we want to discuss are about areas and volumes of gures of
revolution in R
3
. (We shall of course prove them with calculus).
First consider the case of a surface of revolution, using the notation introduced above.
Then
area(M) = 2
_

xds.
But also the centroid of has its x-coordinate dened by
x =
_

xds
_

ds
=
_

xds
L
,
where L = the length of the curve . Thus we obtain Pappus theorem,
area(M) = 2 xL.
Notice that 2 x is the distance the centroid of travels in one rotation. Thus Pappus theorem
can be stated in words as
The area of a surface of revolution equals the product of the length of the
initial curve and the distance traveled around the axis by its centroid.
Integration on manifolds 27
A great example is provided by a torus of revolution. Using the notation of Section 5G, where
the center of the initial circle of radius a was located at distance b > a from the axis, the
resulting area is given by 2a 2b = 4
2
ab.
Sometimes the Pappus theorem can be used backwards. For example, suppose we revolve
a semicircle around the z-axis as shown:
a
z
The resulting surface is a sphere of radius a. Since we already know the area of the sphere,
we obtain from Pappus,
4a
2
= a 2 x,
and we have found the centroid of a semicircle:
x =
2a

.
PROBLEM 1122. Prove another Pappus theorem for a solid of revolution, and
express the result in words as we did above.
PROBLEM 1123. Use Pappus to nd the centroid of a semiball in R
3
.
3. Tubes
There is an interesting extension of Pappus theorems to gures in R
3
we might describe
as tubes. Start with a nonselntersecting C
1
curve in R
3
, parametrized as (t), a t b.
Then imagine constructing a circle of radius r centered at each point (t) of the curve, and
orthogonal to the curve. To make sense of the orthogonality, we of course require the tangent
vector

(t) = 0. Then it can be shown that if r is suciently small, self-intersections do

not occur, and the resulting set is an actual two-dimensional manifold. We want to nd its
area. In fact, it equals the length of times 2r.
28 Chapter 11
In order to prove this with calculus, we need to construct an orthogonal frame at each
point (t). For the rst vector of the frame we can use

(t). The other two vectors may be

chosen rather arbitrarily. Let us call such a choice u(t), v(t), where u(t) and v(t) are unit
vectors in R
3
, and the three vectors,

(t), u(t), v(t),

are mutually orthogonal. We also need to know that u(t) and v(t) are of class C
1
. (One way to
insure this is to assume is of class C
2
and take u(t) as the unit vector in the same direction
as
d
dt

(t)

(t)
,
and then
v(t) =

(t)
(t)
u(t).)
Now notice that dierentiating the equation u u = 1 produces u

u = 0. Thus we may
express u

as a linear combination of

, u, and v, and the result looks like this:

u

(t) = a(t)

(t) + c(t)v(t).
Likewise,
v

(t) = b(t)

(t) + d(t)u(t).
Notice also that u v = 0 implies u

v +u v

= 0, so that c(t) +d(t) = 0. Thus we have the

formulas
_
u

= a

+ cv,
v

= b

cu.
Now we present a parametrization of our tube:
F(t, ) = (t) + r cos u(t) + r sin v(t).
We calculate as follows:
F
t
=

+ r cos (a

+ cv) + r sin (b

cu)
= (1 + ar cos + br sin )

cr sin u + cr cos v,
F

= r sin u + r cos v.
Integration on manifolds 29
Thus we have
_
_
_
F

_
_
_
2
= r
2
and
F
t

F

= cr
2
sin
2
+ cr
2
cos
2
= cr
2
.
Thus the Gram determinant is therefore
J(t, )
2
=
_
_
_
F
t
_
_
_
2
r
2
(cr
2
)
2
= {(1 + ar cos + br sin )
2

2
+ c
2
r
2
}r
2
c
2
r
4
= (1 + ar cos + br sin )
2

2
r
2
.
How nice! If r is suciently small that 1 + ar cos + br sin is always positive, then
J(t, ) = (1 + ar cos + br sin )

r.
We conclude that
area(M) = r
_
b
a
_
2
0
(1 + a(t)r cos + b(t)r sin )

(t)dt
= r
_
b
a
2

(t)dt
= 2r length of .
PROBLEM 1124. Perform the analogous computation for the 3-dimensional volume
of a solid tube in R
3
, using the parametrization given above written now as
F(t, , r) = (t) + r cos u(t) + r sin v(t),
where the new variable r lies in the interval 0 < r < r
0
.
4. Ellipsoids
You probably know that the arc length of an ellipse cannot be expressed in elementary
terms, unless of course the ellipse is merely a circle. In fact, it is for this very reason that a
class of special functions has been dened, the elliptic integrals. Here is one example:
DEFINITION. The complete elliptic integral of the second kind is the function E given by
E(k) =
_
/2
0
_
1 k
2
sin
2
t dt.
30 Chapter 11
For our purposes, we restrict attention to 0 k 1. Of course we have the elementary values
E(0) =

2
, E(1) = 1.
PROBLEM 1125. Find one of the few known elementary values of E:
E
_
1

2
_
=
_

2
_

_
1
4
_
4
_
3
4
_ +

_
3
4
_

_
1
4
_
_
(see Problem 115).
PROBLEM 1126. Prove that the length of an ellipse is equal to
4 (semimajor axis) E(eccentricity).
We should expect that calculating the area of an ellipsoid in R
3
would lead to even more
complicated integrals. Indeed it does. However, there is quite a surprise: the case of ellipsoids
of revolution is completely elementary. The results are given in
PROBLEM 1127. Consider the ellipsoid of revolution in R
3
,
x
2
a
2
+
y
2
a
2
+
z
2
b
2
= 1.
Oblate spheroid: Show that if a > b, then the area is equal to
2a
2
+
2ab
2

a
2
b
2
log
_
a +

a
2
b
2
b
_
.
Prolate spheroid: Find the area in case a < b (notice that the above formula doesnt make
sense in this case).
Although the length of an ellipse in R
2
and the area of an ellipsoid in R
3
are not elementary
integrals, there happen to be associated integrals that are. Here is an example:
Integration on manifolds 31
PROBLEM 1128. Let M R
n
be the ellipsoid described implicitly as
x
2
1
a
2
1
+ +
x
2
n
a
2
n
= 1.
Through any x M construct the hyperplane
which is tangent to M at x. Let D(x) = the
distance from this hyperplane to the origin.
a. Prove that
D(x)
D(x) =
1
_
x
2
1
a
4
1
+ +
x
2
n
a
4
n
b. Calculate explicitly
_
M
Ddvol
n1
.