Chapter 3.

Double and triple integrals

This material is covered in Thomas (chapter 15 in the 11th edition, or chapter 12 in the 10th
edition).
3.1 Remark. What we will do is in some ways similar to integrals in one variable, denite in-
tegrals (which evaluate to a number) rather than indenite integrals (which are essentially an-
tiderivatives, and are functions).
There are 3 ways to approach denite integrals
_
b
a
f(x) dx in one variable and we recall them
briey as we will develop something similar for functions f(x, y) of two variables and f(x, y, z)
of three variables.
(i)
_
b
a
f(x) dx = a limit of certain Riemann sums
lim
n

i=1
f(c
i
)x
i
where [a, b] is divided into n segments [a, x
1
] = [x
0
, x
1
], [x
1
, x
2
], . . . , [x
n1
, x
n
] = [x
n1
, b]
by choosing division points a < x
1
< x
2
< < x
n1
< b. By x
i
we mean the width
x
i
x
i1
of the ith interval. We also have to choose c
1
, c
2
, . . . , c
n
with x
i1
c
i
x
i
(1 i n) before we can calculate the Riemann sum. We take the limit as the number n
of intervals tends to and the widths x
i
all tend to 0.
The idea is summarised by the following picture
The Riemann sum corresponds to the sum of the widths times heights of the rectangles (and
approximates the area of the region under the curve, which represents the graph y = f(x)
for a x b). In this picture n = 8, all the widths x
i
are the same and the points c
i
are
the middle points of the allowed interval for them.
2 200607 Mathematics 2E1 (Timoney)
The idea of the limit can be explained like this. We are likely to get a better approximation
if we use more and narrower rectangles. It is a bit hard to really justify it properly, but
the integral notation we use
_
b
a
f(x) dx is supposed to represent the limiting case where
we have unbelievably narrow rectangles of width dx (and then we need an unbelievably
large number of them). The problem is we dont want to take dx to be quite 0 as then the
products f(x) dx would be all 0. There is an explanation where we take the dx to be so-
called innitesimal numbers that is smaller than any positive number you can imagine
but somehow still not 0.
(ii) We can say graphically that
_
b
a
f(x) dx means the area of the part of the plane between
the graph y = f(x) (a x b) and the x-axis.
This picture is good for case f(x) 0. When f(x) < 0 some or all of the time, we
count those parts of the region trapped between the graph and the x-axis where f(x) < 0
as negative (that is we subtract the total area where f(x) < 0 from the total area where
f(x) > 0). If you look at the Riemann sums, you see f(c
i
)x
i
and youll see that comes
out negative if f(c
i
) < 0
Double and triple integrals 3
(iii)
_
b
a
f(x) dx = [F(x)]
b
x=a
= F(b) F(a) where F(x) is an antiderivative of f(x), that is a
function where F

(x) = f(x) (a x b).

This connection of integrals with derivatives is so familiar that we are inclined to take it
for granted. In fact it is an important result discovered early on in the subject, called the
Fundamental Theorem of Integral Calculus that makes the connection between limits of
Riemann sums and antiderivatives.
3.2 Double integrals. We are now going to give a brief denition of what a double integral is.
We write double integrals as
__
R
f(x, y) dx dy
where f(x, y) is a function of tow variables that makes sense for (x, y) R, and R is a part
of the (x, y)-plane. We should not allow R to be too complicated and we might picture R as
something like this:
4 200607 Mathematics 2E1 (Timoney)
We could visualise the situation via the graph z = f(x, y) over the part of the horizontal plane
indicated by R.
Double and triple integrals 5
At least in the case where f(x, y) 0 always, we can imagine the graph as a roof over a oor
area R. The graphical interpretation of the double integral will be that it is the volume of the part
of space under the roof. (So think of a wall around the perimeter of the oor area R, reaching up
the the edge of the roof or graph. At least in this picture the solid part of space enclosed by the
oor, the walls and the roof would look like some sort of irregular cake.)
So this is the version for double integrals of the explanation in Remark 3.1 (ii) of ordinary
single integrals. To make it more accurate, we have to cater for functions f(x, y) that are some-
times negative, so that the roof or graph is below the oor at times! In this case we have to
subtract the volumes where the roof is below the oor, add the parts where the roof is above (that
is the parts where f(x, y) > 0 are counted positive, and the parts where f(x, y) < 0 are counted
negative).
So far what we have is a graphical or intuitive explanation, but its is not very useful if we
want to compute the double integral. It is also unsatisfactory because it assumes we know what
we mean by the volume and part of the reason for being at a loss for a way to compute the volume
of such a strange cake is that we dont have a very precise denition of what the volume is.
For single integrals this precise explanation proceeds by the notion of a Riemann sum. Even
if you did not think much of it, and indeed Im sure there was not much attempt to cross the is
and dot the ts in the explanation of what Riemann sums are, and the explanation of how to deal
with them, they are actually the basis for practical methods of calculating integrals numerically.
(Youve hear of the trapezoidal rule and Simpsons rule, which are based on rather similar ideas
to Riemann sums. As engineers, you may well end up doing numerical simulations of designs
and techniques like these are bread and butter of such simulations.)
Anyhow, to sketch how double integrals are really dened we need a two dimensional version
of a Riemann sum. What we do is think in terms of dividing up the region R (the oor in our
picture) into a grid of rectangles (or squares). You could think of tiles on the oor, maybe better
to think of very small tiles like mosaic tiles. We take the grid lines to be parallel to the x and y
axes. Here is a picture of a grid, though this one has rather large rectangles.
6 200607 Mathematics 2E1 (Timoney)
One drawback we will see no matter how small we make the grid is that there will be irregular
shapes at the edge. These do cause a bit of bother, but the idea is that if the grid is ne enough,
the total effect of these incomplete rectangles around the edge should not matter much. Lets say
we agree to omit any incomplete rectangles altogether.
No concentrate on just one of the grid rectangles (or tiles on our oor), like the one marked in
black in the picture. The part of the total volume under the roof that is above this one rectangle
will make a sort of pillar, and it will be the analogue for the three dimensional picture of the
graph z = f(x, y) of the rectangles we had in the Remark 3.1 (i) for functions of one variable.
You might look in the book to see nicely drawn artistic picture of what is going on here. We
estimate the total volume under the roof z = f(x, y) by adding up the volumes of the pillars we
have just described. We square off the top of each pillar to make it a at topped pillar of a height
f(x, y) the same as the height of the graph at some point (x, y) inside our chosen grid rectangle
(or oor tile). We get the volume of the pillar as the height f(x, y) times the area of the grid
rectangle. If the side of the rectangle are x along the x-direction and y along the y-direction,
that gives us a volume
f(x, y) x y
for that one pillar. We add all these up top get an estimate of the total volume under the graph.
To get a better estimate, we should use more smaller grid rectangles. The integral is dened
as the limit of the values we get by taking grids with smaller and smaller spacing between the
grid lines.
3.3 Fubinis theorem. What we have so far is fairly theoretical or conceptual. It is in fact
possible to make it practical with a computer program to calculate the sums weve been talking
about, but for pencil and paper calculations we need something like Remark 3.1 (ii). The fact that
works like that is called Fubinis theorem and we try to explain it rst taking R to be a rectangle
with sides parallel to the axes: say
R = {(x, y) R
2
: a x b, c x d}
Double and triple integrals 7
(where a < b and c < d are numbers).
If we go back to thinking of the picture we had above, we now have a nice regular (rectan-
gular) oor, though the roof above (the graph z = f(x, y) could be very irregular. Or maybe we
should think of a cake (or loaf of bread) baked in a rectangular cake tin. We might need to be a
bit imaginative about how the cake rose so as to allow for very strange shapes for the top of the
cake according to a graph z = f(x, y).
Now, you could explain our earlier strategy to estimate the volume of the solid object like
this. First slice the cake into thin slices one way (say along the x-direction rst) and then slice
it in the perpendicular direction. We end of with rectangular ngers of cake and we add up the
volumes of each one. But we estimate the volume of the nger by treating it as a box with a xed
height, even though it will in fact have a slightly irregular top. (Not too irregular though if we
make many cuts.)
To get some sort of explanation of why Fubinis theorem works, we think of only slicing in
one direction. Say we slice along the x-direction, perpendicular to the y-axis. That means y is
constant along each slice. Think of cutting up the bread into slices, preferably very thing slices.
(In our picture, if we cut in the x-direction we will be cutting down the length of the bread, rather
than across. We would be allowed to cut the other way, keeping x-xed, if we wanted, but lets
stick to the choice of cutting along the x-direction keeping y-xed). We can thing then of a thin
slice and we could imagine nding its volume by area and multiplying by its thickness.
Say we call the thickness dy (for a very thin slice). How about the area of the face of the
slice? Well we are looking at a prole of the slice as given by a graph z = f(x, y) with y xed
and x varying between a and b. So it has an area given by a single integral
_
x=b
x=a
f(x, y) dx
(where y is kept xed, or treated as a constant, while we do this integral). We had some similar
ideas when dealing with f/x. We kept y xed and differentiated with respect to x. Or we took
the graph of the function of one variable (x variable in fact) that arises by chopping the graph
8 200607 Mathematics 2E1 (Timoney)
z = f(x, y) in the x-direction, and looked at the slope of that. Anyhow, what we are doing now
is integrating that function of x rather than differentiating it.
The integral above is the area of one slice. To get its volume we should multiply by the
thickness dy
__
x=b
x=a
f(x, y) dx
_
dy
And then to get the total volume we should add the volumes of the thin slices. Well, in the end
we get an integral with respect to y and what Fubinis theorem says is that
__
R
f(x, y) dx dy =
_
y=d
y=c
__
x=b
x=a
f(x, y) dx
_
dy
(when R is the rectangle R = [a, b] [c, d] we chose earlier).
Fubinis theorem also allows us to do the integral with respect to y rst (keeping x xed) and
then the integral with respect to x.
__
R
f(x, y) dx dy =
_
x=b
x=a
__
y=d
y=c
f(x, y) dy
_
dx
In our explanation of why the theorem is true, this is what would happen if we sliced the other
way. It may seem overkill to have two versions of the theorem, but there are examples where the
calculations are much nicer if you do the dx integral rst that if you do the dy integral rst (and
vice versa).
That was all for the case when R is a rectangle, but a similar idea will work even when R is
more complicated. We can either integrate dy rst and then dx or vice versa. But we need to take
care that we cover all the points (x, y) in R. If we keep y constant at some value and then we
have to gure out which x-values give (x, y) inside R. That range of x-values will now depend
on which y we have xed. In the picture that range of x values is a(y) x b(y).
Double and triple integrals 9
d
R
y
a(y)
b(y)
c
__
R
f(x, y) dx dy =
_
y=d
y=c
_
_
x=b(y)
x=a(y)
f(x, y) dx
_
dy
The range c y d has to be so as to include all of R (and no more).
If we do the dy integral rst, keeping x xed, we need to gure out limits c(x) y d(x)
for y. The idea should be clear from the following picture and the formula below.
10 200607 Mathematics 2E1 (Timoney)
a
R
x
c(x)
d(x)
b
__
R
f(x, y) dx dy =
_
x=b
x=a
_
_
y=d(x)
y=c(x)
f(x, y) dy
_
dx
3.4 Example. Find
__
R
x
2
dx dy when R = {(x, y) : x
2
+ 4y
2
1}.
It could help to draw R (it is the interior of an ellipse meeting the x-axis at (1, 0) and the
y-axis at (0, 1/2)) but perhaps it is not really necessary to rely on the picture.
Double and triple integrals 11
If we x y and we want to know which x-values to integrate over (for that y) it is is not hard to
gure it out:
x
2
+ 4y
2
1
x
2
1 4y
2

_
1 4y
2

_
1 4y
2
So the rst integral to work out should be
_
x=

14y
2
x=

14y
2
x
2
dx.
Then we have to integrate that over all y values for which there are any points. But it is fairly
clear that there are points only when that square root is a square root of something that is positive.
So we need
1 4y
2
0
1 4y
2
1
4
y
2

1
2
y
1
2
12 200607 Mathematics 2E1 (Timoney)
We get
__
R
x
2
dx dy =
_
y=1/2
y=1/2
_
_
x=

14y
2
x=

14y
2
x
2
dx
_
dy
=
_
y=1/2
y=1/2
_
x
3
3
_
x=

14y
2
x=

14y
2
dy
=
_
y=1/2
y=1/2
2(1 4y
2
)
_
1 4y
2
3
dy
Let 2y = sin , 2 dy = cos d. For y = 1/2, we get 1 = sin and so = /2. For
y = 1/2, we get 1 = sin and so = /2.
So, after the substitution, the integral we want becomes
_
=/2
=/2
2
3
(1 sin
2
)
_
1 sin
2

1
2
cos d
=
_
/2
/2
1
3
cos
4
d
=
_
/2
/2
1
3
_
1
2
(1 + cos 2
_
2
d
=
1
12
_
/2
/2
(1 + 2 cos 2 + cos
2
2) d
=
1
12
_
/2
/2
_
(1 + 2 cos 2 +
1
2
(1 + cos 4)
_
d
=
1
12
_
/2
/2
_
3
2
+ 2 cos 2 +
1
2
(1 + cos 4)
_
d
=
1
12
_
3
2
+ sin 2 +
1
8
sin 4
_
/2
/2
=

8
3.5 Applications of Double integrals.
(i) If we take the case of the constant function 1, f(x, y) = 1,
__
R
f(x, y) dx dy =
__
R
1 dx dy
= volume between the graph z = 1 and
the region R in the x-y (horizontal) plane
= (area of R) (height = 1)
= area(R)
In short, we have
Double and triple integrals 13
__
R
1 dx dy = area(R)
(ii) If R represents the shape of a thin (at) plate, it will have a surface density (x, y). By
surface density we mean mass per unit area and if the plate is the same thickness and the
same material throughout that will be a constant
=
mass
area
.
But if it is not constant, we nd the density at (x, y) by taking a very tiny sample of the
plate at (x, y) and taking the mass per unit area of that. So if the sample is a rectangular
piece of (innitesimally short) sides dx in the x-direction times dy in the y-direction, then
it has area dx dy. Say we call its mass dm (for a tiny bit of mass), then the surface density
is
(x, y) =
dm
dx dy
To get the total mass we should add up the little small masses dm, or actually integrate
them. We get
total Mass =
__
R
dm =
__
R
(x, y) dx dy
The centre of mass of such a plate is at a point ( x, y) which is also computed with double
integrals. The formulae are:
x =
__
R
x dm
mass
=
__
R
x(x, y) dx dy
__
R
(x, y) dx dy
, y =
__
R
y dm
mass
=
__
R
y(x, y) dx dy
__
R
(x, y) dx dy
where (x, y) denotes the surface density as above. These formulae are arrived at by start-
ing with the condition for the centre of mass to be along the y-axis (so x = 0). The
condition is a moment condition, that the total moment of the body around the axis should
be 0. The total moment about the y-axis is
__
R
x dm. We will not go into detail on this
derivation of the formula for x.
3.6 Triple integrals. We now give a brief explanation of what a triple integral
___
D
f(x, y, z) dx dy dz
14 200607 Mathematics 2E1 (Timoney)
of a function f(x, y, z) of 3 variables over a region D in R
3
(space) means.
As we already know we cannot visualise the graph of a function of 3 variables in any realistic
way (as it would require looking at 4 dimensional space) we cannot be guided by the same sort
of graphical pictures as we had in the case of single integrals and double integrals. (Remember
Remark 3.1 (ii) and the volume under the graph idea of 3.2.)
We can proceed to think of Riemann sums, without the graphical back up. We think of D
as a shape (solid shape) in space and we imagine chopping D up into a lot of small boxes with
sides parallel to the axes. Think of slicing D rst parallel to the x-y plane, then parallel to the
y-z plane and nally parallel to the x-z plane. Preferably we should use very thin separations
between the places where we make the slices, but we have to balance that against the amount of
work we have to do if we end up with many tiny boxes.
Look at one of the boxes we get from D. Say it has a corner at (x, y, , z) and side lengths
x, y and z. We take the product
f(x, y, z) x y z = f(x, y, z) Volume.
We do this for each little box and add up the results to get a Riemann sum

f(x, y, z) x y z
(We should agree some policy about what to do with the edges. If there are boxes that are not
fully inside D we could agree to exclude them from the sum, say.) We now dene the integral
___
D
f(x, y, z) dx dy dz = lim(Riemann sums)
where we take the limit as the Riemann sums get ner and ner (in the sense that the maximum
side length of the little boxes is becoming closer and closer to 0). If we are lucky, this limit
makes sense and there is a theorem that says the limit will make sense if D is bounded, not
too badly behaved and f is a continuous function. So we will take it that the limit makes sense.
As for a picture to think about, we can think maybe of the example we used of mass. If
f(x, y, z) is the density (= mass per unit volume) of a solid object occupying the region D of
space, then
f(x, y, z) dx dy dz = density volume = mass = dm
(mass of a little tiny piece of the solid) and we get the total mass by adding these up. So
___
D
(density function) dx dy dz = total mass.
At least in the case where f(x, y, z) 0 always, we can interpret triple integrals this way.
An even simpler situation is where we take f(x, y, z) 1 to be the constant function 1. Then
f(x, y, z) dx dy dz = dx dy dz = volume = dV
Double and triple integrals 15
is just the volume of the little tiny piece. When we add these up (or integrate the constant function
1) we get
___
D
1 dx dy dz = total volume.
As for double integrals, there is a Fubini Theorem for triple integrals that allows us to
work out
___
D
f(x, y, z) dx dy dz by working out three single integrals. The rst integral (or
inner integral) should be with respect to one of the variables, keeping the other two constant.
Say we integrate dz rst, keeping (x, y) xed. We should integrate over all z that give points
(x, y, z) D. At least in simple cases, that will be a range of z from a smallest we might call
z
0
(x, y) to a largest z
1
(x, y).
In the following picture, there is an attempt to explain this. The curved surface is in fact the
part of the ellipsoid
(x 4)
2
3
2
+
(y 2)
2
2
2
+
z
2
3
2
= 1
where z 0 and the at part at the base is where z = 0. A quarter is cut away to allow us to see
inside the object
D =
_
(x, y, z) :
(x 4)
2
3
2
+
(y 2)
2
2
2
+
z
2
3
2
1, z 0
_
16 200607 Mathematics 2E1 (Timoney)
For each xed (x, y) (like the one shown) we have to nd the limits for z so that (x, y, z) D
and it is easy enough to see that the smallest z is z
0
(x, y) = 0, while the largest is
z
1
(x, y) = 3
_
1
(x 4)
2
3
2

(y 2)
2
2
2
obtained by solving for z in terms of (x, y) when (x, y, z) is on the upper (curved) surface.
We can see that the values of (x, y) for which there are any possible z to worry about are
those (x, y) where the square root is a square root of something positive. So, those (x, y) where
(x 4)
2
3
2
+
(y 2)
2
2
2
1.
Another way to think of it is that this is the outline of the object when viewed along the z-
direction (from far away).
Anyhow our rst integral (if we integrate dz rst) is
_
z=3

1(x4)
2
/9(y2)
2
/4
z=0
f(x, y, z) dz
Double and triple integrals 17
Say we next integrate dx, keeping y xed. The inequality to be satised by all the (x, y) we need
to worry about is above and it can be expressed as
3
_
1
(y 2)
2
2
2
x 4 3
_
1
(y 2)
2
2
2
so that our next integral should be
_
x=4+3

1(y2)
2
/4
x=43

1(y2)
2
/4
_
_
z=3

1(x4)
2
/9(y2)
2
/4
z=0
f(x, y, z) dz
_
dx
Finally we have to integrate this dy. The limits for y are those corresponding to the extreme
values of y for points in D. In this case the restriction on y is
1
(y 2)
2
2
2
0
and that turns out to be the same as
2 y 2 2
or
0 y 4
So in this case
___
D
f(x, y, z) dx dy dz is
_
y=4
y=0
_
_
x=4+3

1(y2)
2
/4
x=43

1(y2)
2
/4
_
_
z=3

1(x4)
2
/9(y2)
2
/4
z=0
f(x, y, z) dz
_
dx
_
dy
It is possible to do the integrals in a different order. Say dy rst, then dx and nally dz. All
the limits will be changed if we do that and we would get
_
z=3
z=0
_
_
x=4+3

1z
2
/9
x=43

1z
2
/9
_
_
y=2+2

1(x4)
2
/9z
2
/9
y=22

1(x4)
2
/9z
2
/9
f(x, y, z) dy
_
dx
_
dz
An advantage of being able to choose the order is that sometimes the calculations are easier
in one order than another. A trick that is sometimes useful for working out an iterated integral
like the one we have just written is this:- gure out which D R
3
it corresponds to, write the
integral as
___
D
f(x, y, z) dx dy dz, that is use Fubinis theorem in reverse rst, and then work
out
___
D
f(x, y, z) dx dy dz with Fubinis theorem using a different order for the single integrals.
Sometimes it turns a hard problem into one that is easier.
3.7 Change of variables in multiple integrals. We now come to a topic that goes by the name
substitution in the case of ordinary single integrals. You may recall that substitution for func-
tions of one variable can be justied using the chain rule
dy
dx
=
dy
du
du
dx
when y = y(u), u = u(x), y = y(u(x)).
18 200607 Mathematics 2E1 (Timoney)
For indenite integrals it allows us to say that
_
g(u(x)) dx =
_
g(u)
dx
du
du
if we interpret the right hand side, a function of u, as a function of x by u = u(x). For denite
integrals we can change limits as well as variables and get an equation that says two numbers are
equal.
_
x=b
x=a
g(u(x)) dx =
_
u=u(b)
u=u(a)
g(u)
dx
du
du.
The point to remember from this is that when we change from integrating over the interval
[a, b] in x, we must not only change the range of integration to the corresponding range in the u
variable, we must also multiply the integrand by a factor dx/du.
For integrals in two variables (and similarly in three variables) we have to explain what that
factor is that works in a similar way. It is the absolute value of a certain determinant of partial
derivatives. Suppose we change from (x, y) to (u, v) say, we have to change dx dy into a multiple
of dudv and the multiple is

det
_
x
u
x
v
y
u
y
v
_

x
u
y
v

x
v
y
u
y
v

The matrix inside the determinant is called a Jacobian matrix. Its rows are the gradient vectors
of x and y with respect to the u and v variables. The determinant itself is called a Jacobian
determinant.
The rule is then that we must change dx dy to
dx dy =

det
_
x
u
x
v
y
u
y
v
_

dudv
To change an integral
__
R
f(x, y) dx dy
to an integral in (u, v) = (u(x, y), v(x, y)), we have to change R to the same set described in the
(u, v) variables and change dx dy as above.
For the case of triple integrals, if we change from (x, y, z) coordinates to
(u, v, w) = (u(x, y, z), v(x, y, z), w(x, y, z))
we have to make a similar change
dx dy dz =

det
_
_
x
u
x
v
x
w
y
u
y
v
y
w
z
u
z
v
z
w
_
_

dudv dw
While this theory can be applied to any change of coordinates, the ones that appear most
often are polar coordinates in the plane, and cylindrical and spherical coordinates in space. So
Double and triple integrals 19
we work out what these Jacobian factors are in polar coordinates to start with. We can relate
cartesian (x, y) coordinates in R
2
to polar coordinates (r, ) via
x = r cos , y = r sin
and so we can work out the 4 partial derivatives we need for the Jacobian.
x
r
= cos
x

= r sin
y
r
= sin
y

= r cos
Then the Jacobian matrix is
_
x
r
x

y
r
y

_
=
_
cos r sin
sin r cos
_
.
The determinant is
r cos
2
+ r sin
2
= r(cos
2
+ sin
2
) = r.
The absolute value of the determinant is also r (as long as we keep r > 0 as we usually do for
polar coordinates). This gives us the relation
dx dy = r dr d
For the change from cartesian (x, y, z) coordinates in space to cylindrical (r, , z) coordinates
the calculation is not really much harder that what we have just done. We wont give the details
but it comes to
dx dy dz = r dr d dz
For spherical coordinates (, , ) in space, the calculation is a little longer. Again we will
not work it out but just give the result:
dx dy dz =
2
sin d d d
As we have not given any details on how these Jacobian factors are justied, we look at a
picture relating to polar coordinates in the plane. The inner circle has radius 1, the next has radius
1.1, the next 4 and the outer one 4.1. The rays are at /4 and /4 + 0.1.
20 200607 Mathematics 2E1 (Timoney)
You can see that the area of the polar rectangle at (r, ) = (1, /4) is smaller than the area
of the polar rectangle at (4, /4). Both rectangles of squares of side 0.1 in polar coordinates,
but it is reasonably clear that the outer one has 4 times the area of the inner one. In fact a polar
rectangle with one corner at (r, ) and opposite corner at (r + dr, + d) is a (slightly bent)
rectangle in the plane with side lengths dr and r d. This is a way to see that the r dr d formula
is at least plausible.
3.8 Example. Find
__
R
_
8
(x
2
+ y
2
)
2
2
_
dx dy
where R = {(x, y) : x
2
+ y
2
2}.
This is a problem that works out rather more easily in polar coordinates. We can describe R
in polar coordinates as the points (r, ) with 0 r

2 and 0 < 2. In this way we can

express the integral as
__ _
8
(r
2
)
2
2
_
r dr d
Double and triple integrals 21
Putting in the limits
_
=2
=0
_
_
r=

2
r=0
_
8
(r
2
)
2
2
_
r dr
_
d =
_
=2
=0
_
_
r=

2
r=0
8r
r
5
2
dr
_
d
=
_
=2
=0
_
4r
2

r
6
12
_
r=

2
r=0
d
=
_
=2
=0
_
8
8
12
_
d
=
23
3
(2) =
46
3
3.9 Applications of triple integrals. In fact we have already mentioned two applications (mass
and volume) but we repeat them here so as to have them on one place. Then there is centre of
mass (the formulae look similar to the ones we had for thin plates, but the 3 variable ones are
more realistic since massive objects are usually 3 dimensional). Finally we mention moments of
inertia.
(i) If D R
3
, then
___
D
1 dx dy dz = volume(D)
(ii) If D R
3
and f(x, y, z) makes sense for (x, y, z) D, then the average value of the
function f over D is dened as follows:
average
D
(f) =
___
D
f(x, y, z) dx dy dz
volume(D)
=
___
D
f(x, y, z) dx dy dz
___
D
1 dx dy dz
(iii) If a solid object occupies a region D R
3
and has (possibly variable) density (x, y, z) at
(x, y, z) D, then
___
D
(x, y, z) dx dy dz =
___
D
dm = mass of the object
The centre of mass of the solid is the point ( x, y, z) given by
x =
___
D
x(x, y, z) dx dy dz
mass
, y =
___
D
y(x, y, z) dx dy dz
mass
,
z =
___
D
z(x, y, z) dx dy dz
mass
.
22 200607 Mathematics 2E1 (Timoney)
The moment of inertia I of a solid object about a xed axis is a quantity that takes the place
of mass when we deal with rotating the object about a xed axis as opposed to dealing
with unconstrained motion. It measures the resistance of the body to being set in rotational
motion about the axis in the same way that mass measure the resistance of a body to being
pushed along. One manifestation of that is in the formula for kinetic energy. For arbitrary
motion of a mass m with velocity vector v the kinetic energy is
1
2
mv
2
. For rotational
motion of a body around an axis, the different particles making up the body are rotating
in unison and all have a common angular velocity (= rate of change of angle around the
axis, =
d
dt
). Those close to the axis have a smaller actual velocity than those farther away
(since the farther away ones go around a big circle in the same time the closer in points go
around a small circle). It comes down to a relation
v = r

(x, y, z)
where r

(x, y, z) stands for the perpendicular distance from the point (x, y, z) in the body
to the axis. The formula for I is then
moment of inertia about axis
= I =
___
D
(r

(x, y, z))
2
dm =
___
D
(r

(x, y, z))
2
(x, y, z) dm
As a special case we can work out the moments of inertia about the x-, y- and z-axes. We
get
I
x
=
___
D
(y
2
+ z
2
) dm =
___
D
(y
2
+ z
2
)(x, y, z) dm
I
y
=
___
D
(x
2
+ z
2
) dm =
___
D
(y
2
+ z
2
)(x, y, z) dm
I
z
=
___
D
(x
2
+ y
2
) dm =
___
D
(y
2
+ z
2
)(x, y, z) dm
There is a theorem that allows one to work out the moment of inertia about any axis though
the centre of mass in terms of I
x
, I
y
, I
z
and 3 other numbers that are also given by integrals.
These other numbers are denoted I
xy
, I
xz
and I
xy
and are given by
Double and triple integrals 23
I
xy
=
___
D
xy dm =
___
D
xy(x, y, z) dm
I
xz
=
___
D
xz dm =
___
D
xz(x, y, z) dm
I
yz
=
___
D
yz dm =
___
D
yz(x, y, z) dm
The 6 numbers are usually arranged as a 3 3 matrix
_
_
I
x
I
xy
I
xz
I
xy
I
y
I
yz
I
xz
I
yz
I
z
_
_
which is called the inertia matrix or inertia tensor of the body.
We need to assume that the origin is the centre of mass. Then we can work out the moment
of inertia of the body about any axis using the inertia matrix. The formula will not concern
us, but here it is anyhow. If u = (u
1
, u
2
, u
3
) is a unit vector parallel to the axis then the
moment of inertia is
(u
1
u
2
u
3
)
_
_
I
x
I
xy
I
xz
I
xy
I
y
I
yz
I
xz
I
yz
I
z
_
_
_
_
u
1
u
2
u
3
_
_
This matrix product works out as a number (or 1 1 matrix)
I
x
u
2
1
+ I
y
u
2
2
+ I
z
u
2
3
+ 2I
xy
u
1
u
2
+ 2I
yz
u
2
u
3
+ 2I
xz
u
1
u
3
This seems to link in to linear algebra and it does. The eigenvectors of the inertia matrix
are called principal axes. Except in cases where the body is exceptionally symmetric,
there will be just 3 such axes. They have the following physical signicance: if the body
is rotated about any axis other than a principal axis, then it will tend to vibrate at higher
rates of rotation. This applies to wheels, shafts, rotors in electric motors, propellers, fan
blades and anything that rotates. It is important that the wheel (or whatever it is) should
be designed so that the axis of rotation is one of the principal axes. If youve been at a
tyre shop when they change a tyre, youll notice that they will clip on small lead weights
around the rim to balance the wheel. These weights are actually changing the inertia
matrix slightly to compensate for irregularities in the wheel rim, or the tyre, or the fact that
the valve stem has a small effect. A wheel rim (or a propeller) is not fully symmetric under
rotation about its axle. There are usually air holes or other shaping to the design. These
have to be designed so as to keep the axle as a principal axis.
So Mathematical things have their uses!
24 200607 Mathematics 2E1 (Timoney)
3.10 Examples. (i) Find the volume of a solid ball of radius a.
This is a problem that is well suited to an integral in spherical coordinates. We can take the
ball to be centered at the origin, so that it is
D = {(x, y, z) R
3
:
_
x
2
+ y
2
+ z
2
a}.
We know
volume(D) =
___
D
1 dx dy dz.
Although this integral is not quite impossible if we do it in x-y-z coordinates via Fubinis
theorem, the calculations are fairly long. Using spherical coordinates we can describe D as
{(, , ) : 0 a, 0 < 2, 0 }
and the integral for the volume becomes
volume =
_
=
=0
__
=2
=0
__
=a
=0

2
sin d
_
d
_
d
=
_
=
=0
_
_
=2
=0
_

3
3
sin
_
=a
=0
d
_
d
=
_
=
=0
__
=2
=0
a
3
3
sin d
_
d
=
_
=
=0
_
2
a
3
3
sin
_
d
=
_
2
a
3
3
cos
_
=
=0
= 2
a
3
3
cos (2
a
3
3
cos 0)
= 2
a
3
3
(1) + 2
a
3
3
=
4
3
a
3
Of course this is the answer we know all along.
(ii) Find the volume of an ellipsoid
x
2
a
2
+
y
2
b
2
+
z
2
c
2
1
We want the triple integral of the constant function 1 over the part of space dened above.
This is a case where an unusual change of variables saves a lot of pain. Say we dene
u =
x
a
, v =
y
b
, w =
z
c
.
Double and triple integrals 25
In (u, v, w) coordinates we describe the ellipse as
u
2
+ v
2
+ w
2
1
(that is a unit ball centre the origin).
To change the integral for the volume
volume(ellipsoid) =
___
ellipsoid
1 dx dy dz
into an integral in dudv dw we need to work out the Jacobian determinant. Using x = au,
v = ay, and z = cw we get
det
_
_
x
u
x
v
x
w
y
u
y
v
y
w
z
u
z
v
z
w
_
_
=
_
_
a 0 0
0 b 0
0 0 c
_
_
.
The determinant of this matrix is abc and the absolute value of the determinant is also abc.
So
___
ellipsoid
1 dx dy dz =
___
u
2
+v
2
+w
2
1
abc dudv dw
= abc
___
u
2
+v
2
+w
2
1
1 dudv dw.
We can bring the factor abc outside the integral because it is a constant. This latter integral
is the volume of the unit ball, and so it is
4
3
. So the volume of the ellipse is
4
3
abc
(iii) Find the centre of mass of a uniform solid half ball.
We can take the half ball to be the top hemisphere of a ball of some radius a centered at the
origin. So we can take the region of space occupied by the solid to be
D = {(x, y, z) R
3
: x
2
+ y
2
+ z
2
a
2
, z 0}
The radius a will have to be part of the answer. Since we are told that the object is uniform,
it means that the density (x, y, z) is a constant throughout the object. That tell us that
26 200607 Mathematics 2E1 (Timoney)
the value of will not come into the answer. For example
z =
___
D
z(x, y, z) dx dy dz
mass
.
=

___
D
z dx dy dz
___
D
dx dy dz
=

___
D
z dx dy dz
___
D
dx dy dz
=
___
D
z dx dy dz
___
D
1 dx dy dz
=
___
D
z dx dy dz
volume
This fact (that the density does not come in to the position of the centre of mass in the case
of constant density) is always true, but we dont really need it. We could just work out z
and nd that the value of cancels out in the end.
In the case of x and y, we have a similar fact that does not come into their values. So
the position of ( x, y, z) depends only on the shape of the object if it is made of a uniform
material. But the symmetry of the object around the z-axis then makes it fairly obvious that
the centre of mass is along the z-axis. In other words x = y = 0 in this case. We could go
about working out the integrals for x and y, but if we do it right we will get 0 both times.
So really the question is to nd z = the distance the centre of mass is away from the centre
along the central axis of the half-ball. Well do this in spherical coordinates (, , ) and
we need the ranges 0 a, 0 < 2 and 0 /2 to cover the half-ball. Recall
z = cos , dx dy dz =
2
sin d d d
Double and triple integrals 27
and so we get
___
D
z dx dy dz =
__
( cos )
2
sin d d d
=
_
=a
=0
_
_
=2
=0
_
_
=/2
=0

3
cos sin d
_
d
_
d
=
_
=a
=0
_
_
=2
=0
_
_
=/2
=0

3
2
sin 2d
_
d
_
d
=
_
=a
=0
_
_
=2
=0
_
_

3
4
cos 2
_
=/2
=0
_
d
_
d
=
_
=a
=0
__
=2
=0
_

3
2
_
d
_
d
=
_
=a
=0

3
d
=
_

4
4
_
=a
=0
=
a
4
4
The volume of the half ball is 2a
3
/3 and so we get
z =
a
4
/4
2a
3
/3
=
3
8
a
The centre of mass is at (0, 0, (3/8)a), or 3/8 of the way along the central radius from the
centre.
TO BE CHECKED
Richard M. Timoney (February 13, 2007)