Section 8.

8: Improper Integrals
One of the main applications of integrals is to compute the areas under
curves, as you know. A geometric question. But there are some geometric
questions which we do not yet know how to do by calculus, even though they
appear to have the same form. Consider the curve y = 1/x2 . We can ask,
what is the area of the region under the curve and right of the line x = 1?
We have no reason to believe this area is finite, but lets ask.
Now no integral will compute thiswe have to integrate over a bounded
interval. Nonetheless, we dont want to throw up our hands.
So note that
Z b
(1/x2 )dx = (1/x)|b1 = 1 1/b.
1

In other words, as b gets larger and larger, the area under the curve and
above [1, b] gets larger and larger; but note that it gets closer and closer to
1. Thus, our intuition tells us that the area of the region were interested in
is exactly 1. More formally:
lim 1 1/b = 1.

We can rewrite that as

lim

b 1

(1/x2 )dx.

Indeed, in general, if we want to compute the area under y = f (x) and right
of the line x = a, we are computing
Z b
lim
f (x)dx.
b a

ASK: Does this limit always exist? Give some situations where it does
not exist. Theyll give something that blows up. Ask: suppose the function
is bounded. Then does the limit always exist? Then say: suppose the limit
of the function is 0. Then does the limit always exist?
An improper integral is an expression of the form
Z
f (x)dx.
a

It is said to be convergent if the limit

Z b
lim
f (x)dx.
b a

exists, and divergent otherwise. In the first case, we define

f (x)dx = lim

b a

f (x)dx.

Likewise,
Z

f (x)dx = lim

a a

f (x)dx.

If the Rintegral goes

R 0 to on both sides, write it as a sum of two improper

integrals 0 and .
If we havent already done it, give
Z
sin xdx
0

as an example of an integral that doesnt converge, even though it doesnt

blow up. Draw the picture and point out that here, fortunately, our geometric intuition concurs that it does not make sense to talk about the area
of the shaded region. Ideally, our mathematical theories should conform to
our geometric intuitionsthen in conditions of extremity the causality should
work the other way...
Main problem of the day. How do we distinguish between improper
integrals which converge and those which diverge?
Ask for a vote on the following examples. Give them 2 minutes to think.
R
1. 1 x4 dx
R
2. 1 1/x4 dx
R
3. 1 1/1 + x4 dx
R
4. 0 ex dx
R0
5. ex dx
R
6.
) ex dx.
Now lets develop some techniques to deal with these guys.

The comparison technique

This is the number one best way to show that an integral converges.
Theorem. Let f be a continuous function defined on [a, ] such that
f (x) 0 for all x [a, ]. Suppose there is another function g such that
f (x) g(x), and such that
Z
g(x)dx
a

converges. Then
Z

f (x)dx

converges, and moreover

Z

f (x)dx

g(x)dx.

This makes sense. Draw the picture. If the area under g is finite, so is
the area in the subregion under f .
Lets show how a proof would go. Define
Z b
h(b) =
f (x)dx.
a

Then h(x) is an increasing function. We want

R to show that h(b) has a limit
as b and that this limit is less than a g(x). We have
Z
Z b
Z b
g(x)dx.
g(x)dx
f (x)dx
h(b) =
a

So draw this picture, and now recall theres a theorem that an increasing
bounded function has a limit. Everything that rises must converge, as they
say. Note that the condition that h be increasingwhich is the same as the
condition that f be positiveis absolutely essential. Draw a picture.
Ex:Consider Nates integral
Z
dx/(x5 + 3x4 + 4x2 + 5x + 7)7/8 .
0

Whether or not Nate can do this integral, or even whether or not I can do
it, I can tell you that the improper integral above is convergent. How? Well,
Ill use two steps. First of all, note that I can write it as
Z
Z 1
5
4
2
7/8
dx/(x + 3x + 4x + 5x + 7) +
dx/(x5 + 3x4 + 4x2 + 5x + 7)7/8
1

where the latter integral is perfectly well-defined, so does not affect convergence. Draw the picture. Then even say, look,
Z b
Z b
Z 1
f (x)dx =
f (x)dx
f (x)dx
0

and the latter thing is a constant, so can be passed through the limit. Anyway. The point is that Nates function is smaller than 1/(x5 )7/8 = 1/x35/8 .
And we find that
Z
Z b
35/8
dx/x
= lim
dx/x35/8
b 1

lim (8/27)/x27/8 |b1

lim (8/27)/b27/8 + 8/27

= 8/27.
It follows from the theorem that Nates integral exists and is at most 8/27.
Remark: The theorem also works for integrals to , as will everything
I say today. From now on, I probably wont mention it.
Note that we can turn this theorem on its head, and use it as a criterion
for divergence.
Theorem. Given f, g as above. If
Z
f (x)dx
a

diverges, so does
Z

g(x)dx.

Ex:The integral
Z

x2 /(x3 1)dx

diverges, because the function is greater than 1/x, and we have already seen
that this integral diverges.
ASK: Why did I have 2 as my lower bound there? Why not 1 or 0?
This leads us to:
4

Unbounded functions
Draw the graph of x2 /(x3 1). We might as well ask: what is the area
between 0 and 2? This makes geometric sense, even though we havent
defined integrals over regions where the function is unbounded. Well, what
does make sense? We can write
Z b
x2 /(x3 1)dx
0

and
2

x2 /(x3 1)dx

(draw the picture) and the intuition is that, as b gets close to 1 from below,
and c gets close to 1 from above, the sum of the two integrals above should
get closer and closer to the actual area, if it exists. To solve this, note that
the expression has an indefinite integral (1/3) log |x3 1|. So the expressions
above turn into
(1/3) log | 1| (1/3) log |b3 1| = (1/3) log |b3 1|
while the latter becomes
(1/3) log 7 (1/3) log |c3 1|.
Neither of these approaches a limit as b, c nears 1. We conclude that the
integral is divergent. However, it is possible for the integral of a function
over an interval to converge even when the function is not bounded on that
interval. See example 5 in your book.

Absolute value test

In the comparison test, we deal entirely with functions which take on positive
values. In life, that isnt always the way of it. What if our function takes
on both positive and negative values?
Thoughts: On the one hand, wed think convergence was more likely,
because the positive and negative parts would cancel each other out. On
the other hand, we can have problems with oscillation, as weve seen, so
maybe its harder. We at least have
Theorem. Let f be a continuous function defined on [a, ]. Suppose
that
Z
|f (x)|dx = L
a

is a convergent indefinite integral. Then

Z
f (x)dx
a

converges and its value lies between L and L.

Proof by picture, and previous theorem. We show by the previous theorem that the positive part A and the negative part B both converge,
and moreover that A < L and B > L, and since A + B is the desired
integral, were done.
Remark: Weve implicitly used the fact that if an integral splits into
the sum of two integrals, each of which converge, then the whole integral
converges. Why?

Rational functions
If theres time (which there wont be), ask: for which P (x) is it true that
Z
dx/P (x)
inf ty

converges? What general results can we come up with?

Prompt them to come up with some examples from todays class. See if
we can get to a general idea of what must happen. Especially see if I can
get anyone to suggest a general class of polynomials making this converge.
Leave it as a contemplation.

Section 10.2: Sequences.

Check out http://forum.swarthmore.edu/dr.math/.
Suggest that they read Section 10.1, as itll motivate what were doing
more. For now, hit them with the following.
e = 1 + 1/2 + 1/3! + 1/4! + 1/5! + . . .
This kind of infinite sum will be of crucial importance for us in the
weeks to come and we had better get ourselves ready to understand what it
means.
But weve got to walk before we run.
An infinite sequence is just a list
a1 , a2 , a3 , . . .
6

For instance,
1, 2, 3, 4, 5, . . .
or
1, 1/4, 1/9, 1/16, . . .
or
1, 1/2, 1/6, 1/120, . . .
which we will write for shorthand as
n, 1/n2 , 1/n!.
Definition: We say that L is the limit of the sequence a0 , a1 , . . . if, for
all any interval [a, b] containing L, the terms of the sequence eventually lie
entirely within [a, b].
Ex:For the second series above, 0 is a limit. Because if you draw any
interval, say, [0.01, 0.01]. Then dont you agree that after a given point,
all the elements of the sequence lie inside the interval? In particular, a1 0
lies in this interval, and so do all further an .
Ex:For the third series above, 0 is also a limit.
Ex:The first series above does not have a limit.
Theorem: No sequence can have two different limits.
Proof. Suppose L and L0 were limits for the sequence. Then choose
intervals [a, b] containing L and [c, d] containing L0 which are so small that
they do not intersect. (Draw this.) Then eventually all the terms of the
sequence have to be in [a, b]. But then they cant be in [c, d]! Contradiction.
This fact was obvious anyway from your intuitive notion of limit. But
then again, if I asked you to prove it, youd throw up your hands, and
say, its obvious! You have to have a precise definition in order to prove
anything. Note that even here we relied on obviousness when choosing
our small intervals. Exercise: finish the proof by proving that such small
intervals really do exist.
A sequence that has a limit is called convergent and a sequence without
a limit is called divergent.
Vote on:
1. (0.5)n ;
2. 5n ;
7

3. 1n ;
4. (1)n ;
5. (1n )/n!;
6. (2n )/n!;
If theres some controversy, any at all, divide into twos, discuss 3 minutes,
vote again.
Theorem: (Monotone Convergence) Suppose an is a non-decreasing
sequence, and is bounded by B; that is, an < B for all n. Then an has a
limit, which is less than B.
This is a deep, deep theorem. First of all, it seems obvious. Then you try
to prove it and you get very confused. Its sort of like this. You want to prove
something about limits. But you just throw up your hands and say, its
obvious! Whats the problem? That you havent precisely defined what
a limit is-you just know. Not good enough. When trying to do a problem
like this, you find that you havent precisely defined what real numbers are.
And in fact this had to wait until Dedekind in 1858 to be precise on 24
November 1858.
Likewise, this works upside down; if an is a non-increasing sequence
bounded below, it has a limit. So for instance, for the last guy above, show
that it eventually decreases, and since it is bounded below by 0, it has a
limit. In fact, the limit is 0.
To see this, write
2n /(n!) = 2 2 2 2 . . . 2/1 2 3 4 . . . n
Now we see that this is less than
2n /(n!) = 2 2 2 2 . . . 2/1 2 3 3 . . . 3 = 2/1 2/2 (2/3)n2
which evidently approaches 0. Draw the squeeze in picture. Note the
similarity to the comparison theorem picture.

Section 10.3: Infinite series

Tell them to put their section time and my name on paper. And
STAPLE.

An infinite series is a special kind of infinite sequence. Namely, it is an

infinite sum.
Examples:
1 + 1/3 + 1/9 + 1/27 + 1/81 + . . .
1 1 + 1 1 + 1 1 + ...
1 + 1 + 1 + 1 + 1 + ...
1 + 1/4 + 1/9 + 1/16 + 1/25 + . . .
The first case, for instance, refers to the infinite sequence
1, 4/3, 13/9, 40/27, . . .
and the third refers to
1, 2, 3, 4, 5, . . .
So
a1 + a2 + . . . =

ai

i=1

is defined to be
lim

n
X

an

i=1

P
if this exists. The ni=1 an are called partial sums.
As always, the MAIN QUESTION is:
How do we tell a convergent series from a divergent one?
So let me make a point. Its not whether the things were summing
converge. Consider the third series above. The things were summing are
always 1. But the sum does not converge.
Ex:Geometric series.
These are series of the form
a + ar + ar2 + ar3 + ar4 + . . .
for some a, r. ASK: Which of the above series are geometric?
9

FACT:
a + ar + ar2 + ar3 + ar4 + . . . + arn = a(1 rn )/(1 r), r 6= 1
We can check this by multiplying. Example: 1+x+x2 = (1x3 )/(1x),
which you probably know as the factorization (1 x3 ) = (1 x)(1 + x + x2 ).
So if we sum n terms of the sequence we get the above expression. Question:
does this converge? Sit and let them think about it.
FACT: It converges if and only if |r| < 1, in which case it converges to
a/(1 r). So in particular, the first series above converges to 1/(1 1/3) =
3/2.
BASIC IDEAS.
(Monotone convergence theorem.) Suppose all the terms a1 , a2 , a3 , . . .
are positive, and that there is some upper bound B such that a1 +
a2 + . . . + an < B for all n. Then a1 + a2 + . . . converges.
P
( 2
Ex:: 1 + 1/10 + 1/104 + 1/109 + 1/101 6 + . . . =
i=1 1/10 i )
Write the partial sums out as decimals:

a1 = 1
a1 + a2 = 1.1
a1 + a2 + a3 = 1.1001
a1 + a2 + a3 + a4 = 1.100100001

Clearly these are always less than 2. So by the theorem this series
converges.
Natural complaint: But we already know this sequence converges! Because it converges to 1.1001000010000001.... But then you have to
ask yourself, why do we believe that such a number exists? Now we
are dipping perilously close to foundations, something that confused
mathematicians for a long time.
In this connection, Liz asked me last time Is this improper integral
we get really the area or is it just something that gets closer and closer
to the area?
Similarly, one might ask, are we saying that 4/3 is really the value of
the sum above, or only that the sum gets closer and closer to it?
In this connection, quote Hardy, from Divergent Series (1948).
10

...it does not occur to a modern mathematician that a collection of mathematical symbols should have a meaning until
one has been assigned to it by definition. It was not a triviality even to the greatest mathematicians of the eighteenth
century. They had not the habit of definition: it was not
natural to them to say, in so many words, by X we mean
Y. ... it is broadly true to say that mathematicians before
Cauchy asked not How shall we define 1 1 + 1 1 + . . . but
What is 1 1 + 1 1 + . . . ?, and that this habit of mind led
them into unnecessary perplexities and controversies which
were often really verbal.
Suppose the series a1 + a2 + a3 + . . . converges. Then the sequence
a1 , a2 , a3 , . . . converges to 0.
Well not prove this, though the book gives a slick proof. For us,
consider the contrapositive.
Suppose the sequence a1 , a2 , a3 , . . . does not converge to 0. Then the
sequence a1 + a2 + a3 . . . converges.
Well do the case where a1 , a2 , a3 , . . . converges to L 6= 0. Then after
a while the sequence
a1 + a2 + a3 + . . . + a1 00 + a1 01 + . . .
is well-approximated by
a1 + a2 + a3 + . . . + L + L + . . .
and evidently this sequence does not converge.
The converse of this statement would be
Suppose the sequence a1 , a2 , a3 , . . . converges to 0. Then the series
a1 + a2 + a3 + a4 + . . . converges.
True or false? Vote: then split into pairs, discuss, come back to me.
Ex:The harmonic series.
1 + 1/2 + 1/3 + 1/4 + 1/5 + 1/6 + . . .
The terms converge to 0. Nonetheless, the sum diverges.

11

To prove this, observe that we can split the terms up like

(1 + 1/2) + (1/3 + 1/4) + (1/5 + 1/6 + 1/7 + 1/8)+

(1/9 + 1/10 + 1/11 + 1/12 + 1/13 + 1/14 + 1

and the terms within each parenthesis add up to at least 1/2. So the partial
sums are unbounded.
Again, remember that the sequence of partial sums is
1, 1 + 1/2, 1 + 1/2 + 1/3, 1 + 1/2 + 1/3 + 1/4
Now note that this series indeed diverges, but it diverges much more
slowly than something like 1 + 1 + 1 + 1 + . . . . In higher math, once we determine that a sequence converges or diverges, we like to ask, HOW QUICKLY?
According to our calculation, it takes at most
2 terms to get to 1/2;
4 terms to get to 1;
8 terms to get to 3/2;
16 terms to get to 2;
32 terms to get to 5/2;
64 terms to get to 3;
4k terms to get to k;
Actual calculation shows it takes
1 term to get to 1;
4 terms to get to 2;
11 terms to get to 3;
31 terms to get to 4;
83 terms to get to 5;
227 terms to get to 6;
616 terms to get to 7;
12

 1674 terms to get to 8...

In other words, our argument shows that number of terms necessary to
get to k is 4k . This suggests that maybe that number grows exponentially!
And indeed, we find that by experimental data, it is very close to the truth
that number of terms necessary to get to k is very close to ek .
Another way to say it is that
1 + 1/2 + 1/3 + . . . + 1/ek k.
or
1 + 1/2 + 1/3 + . . . + 1/n log n
Does this look like anything? Draw the curve, the step function, the integral.
And this prepares them for next time.

13