You are on page 1of 100

DON’T PANIC!

Part 3

A guide to MATA21 Analysis in One Variable

Version: January 22, 2020

Jan-Fredrik Olsen
ii
Contents

III The infinite series and limits 171

4 Infinite series 173


4.1 The definition of the sum of an infinite series . . . . . . . . . . . . . . . . 173
4.2 Some important infinite series . . . . . . . . . . . . . . . . . . . . . . . . . 179
4.3 The comparison test for positive series . . . . . . . . . . . . . . . . . . . . 187
4.4 On the convergence of non-positive series . . . . . . . . . . . . . . . . . . . 194
4.5 Exam exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
4.6 Answers to selected exercises . . . . . . . . . . . . . . . . . . . . . . . . . 204

5 Limits for sequences 207


5.1 The definition for the limit of a sequence . . . . . . . . . . . . . . . . . . . 208
5.2 The rulebook for limits of sequences . . . . . . . . . . . . . . . . . . . . . 216
5.3 How to use the rulebook in practice . . . . . . . . . . . . . . . . . . . . . . 219
5.4 How to prove the rules in the rulebook . . . . . . . . . . . . . . . . . . . . 228
5.5 Convergence tests for positive series . . . . . . . . . . . . . . . . . . . . . 233
5.6 Exam exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
5.7 Answers to selected exercises . . . . . . . . . . . . . . . . . . . . . . . . . 242

6 Limits for functions 245


6.1 A first look at the limit for functions . . . . . . . . . . . . . . . . . . . . . 246
6.2 The definition of the limit for functions . . . . . . . . . . . . . . . . . . . . 250
6.3 The rulebook for limits of functions . . . . . . . . . . . . . . . . . . . . . . 255
6.4 How to use the rulebook in practice . . . . . . . . . . . . . . . . . . . . . . 258
6.5 Exam exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
6.6 Answers to selected exercises . . . . . . . . . . . . . . . . . . . . . . . . . 265

iii
iv CONTENTS
Part III

The infinite series and limits

171
Chapter 4

Infinite series

In this chapter, we investigate, and try to build a theory for, what happens when we
sum an infinite number of terms. As we shall see, infinite sums do not always behave in
the ways we expect them to!
Below, you find a few sample exercises that you should be able to solve after working
through this chapter.

Remark 4.1 (Selected problems from previous exams based on this chapter)

1. Define what we mean by a series being convergent and divergent, respectively.


1
X
2. Is it true that lim an = 0 =) an converges ?
n!1
n=1
3. Determine whether the following series converge or diverge.
1
X 1
X 1 ⇣p
X ⌘
2 2k
(a) e1/k (b) (c) k4 + 1 k2
k!
k=1 k=1 k=1

4.1 The definition of the sum of an infinite series


At the start of chapter 3, we discussed the Geometric series
1
X 1 1 1 1
k
= 1 + + + + ··· .
2 2 4 8
k=0
In particular, in Example 3.1, we computed the first few partial sums Sn of this series.
In particular, we observed that the partial sums Sn got closer to the value 2 as we add
each new term. Indeed, we can express the first few computations of Example 3.1 by
1 1
S0 = 2 1, S1 = 2 , S2 = 2 .
2 4

173
174 CHAPTER 4. INFINITE SERIES

That is, the distance between the partial sums and the value 2 is halved each time we
include one additional term.
Exercise 4.2 Prove by induction that the partial sums of the above Geometric series
satisfies the formula Sn = 2 1/2n .
So what would happen if we could sum an infinite number of terms? Well, intuitively,
we ought to get 1
X 1
= 2 0 = 2.
2k
k=0
We now try to put this into a more precise language. To this end, let us make the
following "informal definition" of what we mean by the limit of a sequence of numbers
(this should be compared to the informal definitions of limits on pages ?? and ??).

Remark 4.3 (Informal definition of the limit of a sequence) Let (cn )1


n=0 be a
sequence of numbers. We say that cn converges to L and write
lim cn = L
n!1

if the the numbers cn become arbitrarily close to L provided that n is sufficiently large.
If, on the other hand, the sequence cn does not converge to any finite number L 2 R,
we say that it diverges (or that the limit does not exist). In the special case that the
numbers cn become arbitrarily large, provided that n is sufficiently large, we say that cn
diverges to 1 and write
lim cn = 1.
n!1

Exercise 4.4 Compute the first few thousand terms of the following sequences.
What does the above informal definition say about them?
1
(a) lim 21/k (b) lim (c) lim ( 1)k
k!1 n!1 2n k!1

Exercise 4.5 Not all divergent sequences diverge to 1. Indeed:


(a) If a sequence (cn )1
n=1 diverges, but the numbers cn are bounded, i.e. satisfy
|cn |  M for some M > 0, then we say that the sequence diverges boundedly.
Give an example of such a sequence.
(b) Does there exist divergent sequences that neither diverges to 1 nor diverge
boundedly? That is, is there a third type of divergence?

Remark 4.6 A major topic of these lecture notes is to formulate a completely precise
version of this definition so that we can give proper proofs of what the limits of these
sequences are. However, since this is rather technical, we delay this until the next chapter
(which means, that, for now, we need to rely on our intuition when working with limits).
4.1. THE DEFINITION OF THE SUM OF AN INFINITE SERIES 175

The definition of the sum of a series

If we do not want to state the definition of the limit just yet, why mention limits at all?
The reason is that we want to express the connection between partial sums and infinite
series using the language of limits. Indeed, we now make the following definition (which
is completely legitimate once our definition of the limit is made precise):

P1 4.7 (The sum of an infinite series) Let Sn be the partial sums of the
Definition
series k=0 ak . Then we define the sum (or value) of this infinite series by setting
1
X
ak = lim Sn .
n!1
k=0
If the limit diverges, then we say that the series diverges (or that its sum does not exist).

With this definition, we finally have the language we need to express the following
computation:
X1
1 X n
1 ⇣ 1⌘
= lim = lim 2 = 2 0 = 2.
2k n!1 2k n!1 2n
k=0 k=0

Exercise 4.8 The following formula is very popular among students on the final
exam. But is it really true? Indicate a proof or counter-example, as appropriate.
X1
ck = lim ck .
k!1
k=0

But why do we need a definition to tell us what we mean when we sum a bunch of
numbers? Well, when "a bunch of numbers" means "an infinite sequence of numbers",
it is kind of important to be careful about what we are doing. Here is an example to
illustrate this:

Example 4.9 (Grandi’s series)


Guido Grandi famously studied the series:
1
X
( 1)k = 1 1+1 1+1 ··· .
k=0

Grandi noted that with creative use of parenthe-


ses one can do the following two computations:
X1
( 1)k = 1 1 + 1 1 + · · · Fig. 1. Guido Grandi (1671 –
k=0
1742). A mathematician, monk
= (1 1) + (1 1) + · · · and proof that religion and sci-
= 0 + 0 + ··· = 0 ence do not always mix.
176 CHAPTER 4. INFINITE SERIES

and
1
X
( 1)k = 1 (1 1) (1 1) ··· = 1 0 0 · · · = 1.
k=0

This is quite remarkable, since if both computations are correct, this implies that 0 = 1.
How did Grandi react? With skepticism? Hell, no! He proclaimed that he had just
witnessed that: “Something can be created from nothing!”, and concluded that “God
exists!”

Of course, Grandi was wrong, and it is not the case that 0 = 1. Mathematics is
quite healthy, planes will stay in the air (mostly), nuclear missiles are not flying (yet)
and Grandi did not prove that God exists. So, why is it wrong to say that
1
X
0= ( 1)k = 1 ? (4.1)
k=0
Here is the thing: Grandi cheated by using two different definitions of what the sum of
a series should mean. That is, if you use two different definitions of what the sum of a
series means, then you should not be too surprised to get two different values.

Example 4.10 (Grandi’s series, revisited) Computing the partial sums Sn of


Grandi’s series, we find that

S0 = 1, S1 = 0, S2 = 1, S3 = 0, ...

So what did Grandi do? Well, basically, in his proof that God exists, he first claimed
that since all the Sn with even n are equal to 1, then the series should sum up to 1. Next,
Grandi claimed that since all the Sn with odd n are equal to 0, then the series should
sum up to 0. It is in claiming that both of these two values can be taken simultaneously
as the definition of the sum of a series that Grandi cheats!

So, what is really the sum of Grandi’s series? We will leave it as an exercise for you
to figure out.
Exercise 4.11 Use Definition 4.7 to figure out the true "value" of Grandi’s series.

Remark 4.12 (Failure of associativity for infinite sums) Notice that Grandi’s
example shows that the axiom of associativity (recall Axiom 1.?? on page ??) can fail
for infinite sums. Indeed, as we saw above, just by moving the parentheses, we get
(1 1) + (1 1) + (1 1) + · · · 6= 1 (1 1) (1 1) ···.
| {z } | {z }
=0 =1
4.1. THE DEFINITION OF THE SUM OF AN INFINITE SERIES 177

Some initial computational rules for infinite series

Remark 4.12 is quite important. It tells us that we need to be very careful when summing
an infinite number of terms. In particular, it makes sense for us to formulate and prove
computational rules that identify conditions when everything works as expected. Here
is a first example of such a "friendly" situation:

P P1
Proposition 4.13 If 1 k=0 ak and k=0 bk are two convergent series, then the following
computational rules holds:
1
X X1 1
X
(a) (ak + bk ) = ak + bk
k=0 k=0 k=0
1
X 1
X
(b) C · ak = C ak
k=0 k=0

Proof. We prove the first part


P of this proposition,
P1 and leave the second half for the
exercises. So, suppose that 1 k=0 a k and k=0 b k are two convergent series, and let An
and Bn denote their n’th P partial sums, respectively. Moreover, let Sn denote the n’th
partial sum of the series 1 1
k=0 (ak + bk ). But for finite sums, it is clear that we can
write n n n
X X X
Sn = (ak + bk ) = ak + b k = An + B n .
k=0 k=0 k=0
Taking the limit, as n goes to 1, this gives us
1
X 1
X 1
X
(ak + bk ) = lim Sn = lim (An + Bn ) = lim An + lim Bn = ak + bk ,
n!1 n!1 n!1 n!1
k=0 k=0 k=0

and we are done!

Remark 4.14 (computational rules for the limit) Strictly speaking, the above
proof should be called a "proof". The problem is that we are using properties for the
limit that we have not yet proved (and won’t until the next chapter). For instance, notice
the step
lim (An + Bn ) = lim An + lim Bn .
n!1 n!1 n!1

Here, we are using the property that the limit of a sum is equal to the sum of the limits.
This is something that we do not yet know to be true. But to prove anything about
limits, we need a proper definition. This we don’t have yet, but is something we will
discuss at great length in the next chapter.

1
If we are to be really picky, the fact that Sn = An + Bn for finite n ought to be proved by induction.
178 CHAPTER 4. INFINITE SERIES

Exercise 4.15 Provide a similar proof for part (b) of the above proposition. What
computational rule for the limit do we need now?
Exercise 4.16 For Grandi’s series, associativity failed. However, if we, as in the
above proposition, assume from the start that we are working with a series that con-
verges, then associativity (in the sense needed by Grandi) is actually true.
(a) Formulate the result indicated at the start of this exercise as a proposition.
(b) Provide a proof of this result (similar to the one for Proposition 4.13, above).

A first test for convergence: the divergence test


In these lecture notes, we will discuss several tests for the convergence of series. The
first, strangely enough, is called the divergence test. The reason for the name is that this
is a test that can sometimes recognise that a series diverges, but it can never tell us that
a series converges.
P1
Proposition 4.17 (The divergence test) Let k=1 an be a series. Then
1
X
ak 6 ! 0 =) ak diverges
k=1

Here, by ak 6 ! 0, we mean that the terms ak do not tend to zero as k grows.

Proof of the Divergence test. We do a proof by contraposition. This means that we need
to prove: X1
ak converges =) ak ! 0.
k=1
P
As usual, put Sn = nk=1 ak . By Definition 4.7, the convergence of the series means that
Sn ! L, for some finite value L. Notice that since Sn 1 essentially represents the same
sequence of numbers as Sn , we also have Sn 1 ! L. But then we are done since now
a k = Sk Sk 1 !L L = 0.

P
Example 4.18 We consider Grandi’s series 1 k=0 ( 1) . Since the terms ak = ( 1)
k k

do not go to zero as n increases, it follows by the Divergence test that Grandi’s series
must diverge.

Exercise 4.19 What does the divergence test say about the following series?
1
X 1
X
k 1
(a) 2 (b)
2k
k=0 k=0
4.2. SOME IMPORTANT INFINITE SERIES 179

4.2 Some important infinite series


A lot of our understanding of how infinite series act is based on a good understanding
of a few important examples: the Geometric series, the Harmonic series, the Alternating
Harmonic series and the Basel series. The point of this section is to systematically
consider these fundamental examples before we study a more general theory of series.

The Geometric series


The king of all series is the Geometric series:
1
X
xk = 1 + x + x2 + x3 + · · · .
k=0

The first to make a major contribution to the


study of infinite series is probably Archimedes
who needed to compute this series in the case Fig. 2. Archimedes (287 BC – 212
x = 1/4. In particular, Archimedes obtained a BC) is perhaps best known for run-
special case of the following formula – a formula ning around naked and shouting.
that we will now try to understand.

Proposition 4.20 (Sum of the Geometric series) For all |x| < 1 it holds that
1
X 1
xk = . (4.2)
1 x
k=0

For all other x, the infinite series diverges (that is, has no finite value).

Exercise 4.21 As a warm up exercise, for each of the following values of x, use
Proposition 4.20 to determine the sum (if any) of the corresponding Geometric series:
1 1
(a) x = (b) x = (c) x = 1 (d) x = 2.
4 2
Exercise 4.22 What does the divergence test say about the Geometric series?

Now, to prove any result about infinite series, we need to consider partial sums. The
following lemma is therefore of key interest.

Lemma 4.23 (Partial sums of the Geometric series) For all x 6= 1, it holds that
n
X 1 xn+1
xk = .
1 x 1 x
k=0
180 CHAPTER 4. INFINITE SERIES

Proof of Lemma 4.23. The trick is to multiply the partial sum by one in a clever way:
n ⇣
1 x⌘
Xn X n
1 X k
xk = xk · = (x xk+1 ).
1 x
| {z } 1 x
k=0 k=0 k=0
=1

Exercise 4.24 Complete the proof by writing out the last sum above.
Our goal is now to understand why Proposition 4.20 is a consequence of Lemma
4.23. First, let us note that in the special case that x = 1/2, we obtain the formula from
exercise 4.2: Xn
1 1
k
=2 .
2 2n
k=0
Intuitively, it should be clear that the right-hand side has the limit
P1 2 as kn ! 1. By
Definition 4.7, when taking the limit, the left-hand side becomes k=0 1/2 . Hence, we
obtain the formula 1
X
1/2k = 2.
k=0

Exercise 4.25 What happens in the formula from Lemma 4.23 as n gets large for
each of the following values of x.
1 1
(a) x = (b) x = (c) x = 1 (d) x = 2.
4 2
(e) Does this match what you observed in exercise 4.21?
Exercise 4.26 Your cousin has learned of Proposition 4.20 and wants to tattoo the
formula 1
X
2k = 1
k=0
to his forehead. Can you slowly explain to him why this is a bad idea.
Ideally, after doing the above exercise, you should have an intuitive feeling of why
Lemma 4.23 implies Proposition 4.20. For a proper proof of this fact, we have to wait
until Chapter 5, where we give the proper definition of the limit of a sequence. However,
to get a sense of what a proof may look like, in the following exercise, you are asked to
combine the informal definition from Remark 4.3 with what you have discovered above.
Exercise 4.27 Let Sn be the partial sums of the Geometric series for x = 0.999 and
let L = 1/(1 0.999). Use Lemma 4.23 to figure out how large n must be so that
(a) |Sn L| < 1/10
(b) |Sn L| < 1/100
(c) |Sn L| < 1/1000.
(d) (Discussion) Explain how solving these inequalities is related to Remark 4.3.
Remark: After using the lemma, you should be able to get rid of the absolute value.
The inequalities can be solved using techniques from chapters 1 and 2.
4.2. SOME IMPORTANT INFINITE SERIES 181

Harmonic series
We now leave ancient greece, and jump to the dark
ages and the work of Nicolas Oresme. Oresme is prob-
ably as close as we get to a science superhero at the
time. He was among the first to translate the sci-
entific work of the ancient Greek into the modern
languages of the time, he was critical of astrology,
laid the foundation for Galileo’s theory of motion and
studied infinite series systematically.
Oresme knew of Archimedes work on Geometric
series (the works of Archimedes had been translated Fig. 3. Pictured is Nicolas
and made available to the scientific community about Oresme (1320 – 1382). For a bit
100 years earlier). So he definitely knew that it was of perspective, the black plague
possible to sum an infinite sequence of numbers and first came to Europe in 1348.
still get a finite answer.
In a rather modern mathematical spirit, he decided to study what happens when you
sum the infinite sequence of the so-called harmonic numbers 1, 1/2, 1/3, 1/4, . . .. That
is, he studied the harmonic series
1
X 1 1 1 1 1
= 1 + + + + + ··· .
k 2 3 4 5
k=1

In what is one of the great classical (and quite short) arguments of medieval mathematics,
Oresme proved the following result.

Proposition 4.28 (Oresme’s theorem on the harmonic series)


1
X 1
= 1.
k
k=1

To get some perspective on why both his achievement, and the result itself, should be
considered shocking, do the following exercise.
Exercise 4.29 (a) Compute the partial sum Sn in Python for n as large as your
computer can handle. How large are you able to get Sn ?
(b) Visualise the first few thousand (or more if you want) Sn in a graph. Does the
graph seem to go to infinity, or does it seem to stabilise at some finite height?
Exercise 4.30 Determine (roughly) how many terms you can sum before Python
thinks that Sn+1 = Sn ?
Hint: This question is about round-off error. To answer this exercise, you need to make
an educated guess about the size of Sn for large values of n. To make such an educated
guess, compute S100 /S10 , S1000 /S100 and S10000 /S1000 . Do you see a pattern?
182 CHAPTER 4. INFINITE SERIES

The point of the above exercises was to illustrate how numerical computations fail at
giving any hint of the behaviour of the harmonic series since the partial sums increase
so slowly.
In particular, we should realise that Oresme really does deserve some respect for
being able to figure out, in the 14th century, barely having pen and paper, that the
partial sums do diverge to infinity. His key observation was the following:

X1
1 1 1 1 1 1 1 1 1 1
= 1 + + + + + + + + + ··· + +··· . (4.3)
k 2 |3 {z 4} |5 6 {z 7 8} |9 {z 16}
k=1
1
4
+ 14 1
8
+ 18 + 18 + 18 1
16
1
+···+ 16

Exercise 4.31 (a) What does Oresme’s observation say about the size of S4 , S8 and
S16 , respectively?
(b) More generally, what does Oresme’s observation say about the size of S2m , for
m 2? (Express your answer as an inequality.)
(c) What ought the observation from (b) say about the size of Sn as n grows?

Ideally, after doing the above exercise, you should have an intuitive feeling of why
the Harmonic series diverges to 1. For a proper proof of this fact, we have to wait
until Chapter 5, where we give the proper definition of what it means for a sequence
to diverge in this way. However, to get a sense of what a proof may look like, in the
following exercise, you are asked to combine the informal definition from Remark 4.3
with what you have discovered above.

Exercise 4.32 Let Sn be the partial sums of the Harmonic series. Use Oresme’s trick
to determine, as best you can, how large n must be so that

(a) Sn > 10
(b) Sn > 100
(c) Sn > 1000.

(d) (Discussion) Explain how solving these inequalities is related to Remark 4.3.
Remark: An important detail is that no one is asking you to find the smallest n so
that, say, Sn > 10 holds. It suffices to determine a sufficiently large n so that this
inequality holds.
4.2. SOME IMPORTANT INFINITE SERIES 183

Alternating harmonic series


We now move forward to the middle of the 17th
century. More specifically, the year 1644. This
is in the period called the age of disovery. It is
also the age of re-discovery. Apparently, all lot
of the knowledge obtained in previous eras seem
to have been lost. This is good news for people
who want something to discover! For instance,
that year, the Italian mathematician Pietro Men-
goli was quite pleased to re-discover that the har-
monic series diverges.
To his credit, though, this was not Mengoli’s
Fig. 4. Pietro Mengoli (1626 – 1686).
only contribution to mathematics. Indeed, he
Blissfully ignorant of our hero Nico-
seems to have been the first to compute the sum
las Oresme.
of the alternating harmonic series
1
X ( 1)k+1 1 1 1 1
=1 + + ··· ,
k 2 3 4 5
k=1
Indeed, he discovered the following result.

Proposition 4.33 (Mengoli’s formula for the alternating harmonic series)


1
X ( 1)k+1
= ln(2).
k
k=1

Exercise 4.34 To check that the above identity is credible, compute the numerical
values of the partial sums S10 , S100 , S1000 and S10000 . How many digits of ln(2) do
these match, respectively (if any)?
Mengoli’s formula can be understood visually as fol-
lows. The idea is to think of the partial sums Sn as a
walk on the real line. First, you start at the origin.
Then you add the first term (1), to obtain S1 . This
moves you one step to the right. Next, S2 is obtained
from S1 by adding the next term ( 1/2). This means
that you continue your walk by taking half a step to
the left. Next, to obtain S3 , you add the next term
(1/3). Continuing like this, you move through each
partial sum Sn in a back-and-forth walk on the real
Fig. 5. Visualisation of the par-
line, where the step sizes keep shrinking. From the fig-
tial sums of the alternating har-
ure, it seems clear that if you go on like this forever,
monic series.
you should converge to the centre of the spiral.
184 CHAPTER 4. INFINITE SERIES

As always, to truly understand an infinite series, we need to understand the behaviour


of its partial sums. This is the point of the following exercise:

Exercise 4.35 Use the Figure 4.5 to explain why the following upper bound for the
error made after n steps is intuitively true:
1
|Sn ln(2)| < .
n+1

Hint: Notice how 1/(n + 1) is exactly the length of the "next step".

Ideally, after doing the above exercises, you should have an intuitive feeling that it
makes sense for the Alternating Harmonic series to converge to the number ln(2). For
a proper proof of this fact, we have to wait until Chapter 5, where we give the proper
definition of the limit of a sequence, as well as the last chapter of these lecture notes,
where we discuss why it makes sense for ln(2) to show up (and thereby solve exercise
4.35). However, recall that in Remark 4.3, we already stated an informal definition of
the limit of a sequence. In the following exercise, you are asked to combine this informal
definition with what you have discovered above.

Exercise 4.36 Let Sn be the partial sums of the Alternating Harmonic series and let
L = ln(2). Use the inequality from exercise 4.35 to figure out how large n must be so
that

(a) |Sn L| < 1/10


(b) |Sn L| < 1/100
(c) |Sn L| < 1/1000.

(d) (Discussion) Explain how solving these inequalities is related to Remark 4.3.
4.2. SOME IMPORTANT INFINITE SERIES 185

Basel series
We now make a small jump to the year 1650 and Mengoli’s perhaps most influential
contribution to mathematics. Namely, he posed the question2 :

What is the sum of the series


1
X 1
? (4.4)
k2
k=1

This question became known as the Basel prob-


lem since several quite famous mathematicians from
Basel became obsessed with it. It took nearly 100
years for it to be solved, and it was not until 1734
that the young Leonard Euler (from Basel) proved Fig. 6. Leonard Euler (1707 –
the following result. 1783).

Theorem 4.37 (Euler’s solution to the Basel problem)


1
X 1 ⇡2
= . (4.5)
k2 6
k=1

Euler has since become known as one of the greatest mathematicians of all time, and is
certainly the most productive. In 1775, for instance, he wrote, on average, one scientific
paper per week (despite having become completely blind in 1771!).
While we will not be able to explain Euler’s proof of the Basel problem in these
lecture notes (the proof is only slightly too advanced for us – it will (or should) appear
in your first course on Fourier analysis), we will see how we can prove, using Oresme’s
ideas (see (4.3), above) that the series (4.4), which we call the Basel series, has to be
quite close to ⇡ 2 /6.

Exercise 4.38 (a) Compute the partial sum Sn in Python for n as large as your
computer can handle. How many digits of ⇡ 2 /6 does this match?
(b) Visualise the first few thousand (or more if you want) Sn in a graph. Does the
graph seem to go to infinity, or does it seem to stabilise at some finite height?
Exercise 4.39 Determine (roughly) how many terms you can sum before Python
thinks that Sn+1 = Sn ? Is your answer in any way related to part (a) of exercise
4.38?
2
Posing fruitful research questions is an extremely important ability for a mathematician. It is a bit
like pointing out a promising path through a dark jungle filled with tigers that have pointy teeth.
186 CHAPTER 4. INFINITE SERIES

Let us now try to study the Basel series a bit more systematically. First, we take a
look at how we can use a modified version of Oresme’s trick:
X1
1 1 1 1 1 1 1 1 1
2
= 1 + 2 + 2 + 2 + 2 + 2 + 2 + 2 + ··· + 2 +··· . (4.6)
k |2 {z 3 } |4 5 {z 6 7 } |8 {z 15 }
k=1
1
 + 12  1
+ 12 + 12 + 12  1
+···+ 12
22 2 42 4 4 4 82 8

Exercise 4.40 (a) What does the modified Oresme trick say about the size of S3 ,
S7 and S15 . respectively?
(b) More generally, what the modified Oresme trick say about the size of S2m 1 , for
m 2? (Express your answer as an inequality.)
(c) What ought the observation from (b) say about the size of Sn as n grows?

Ideally, by doing the above exercise, you are convinced that the size of Sn does not
diverge to 1. Moreover, by doing exercise 4.38, you have seen that by computing Sn for
large values of n, we can match quite a few decimal digits of the value Euler computed
for the Basel series.
Below, we refine the our use of Oresme’s trick in order to try to predict how large
we must choose n in order for Sn to match a given number of decimal digits of ⇡ 2 /6. To
this end, we study the tail of the Basel series. That is, we express the Basel series as
follows:
X1 Xn X1
1 1 1
2
= 2
+ .
k k k2
k=1 k=1 k=n+1
| {z } | {z }
n’th partial sum n’th tail

Exercise 4.41 Use the modified trick of Oresme to show that the (2m 1)’th tail is
less than or equal to 1/2m 1 .
Remark: We consider the (2m 1)’th tail because this fits with how we divide the Basel
series into blocks in the modified Oresme trick.

Exercise 4.42 Suppose that the Basel series converges to some number L, and let Sn
and Tn denote the n’th partial sum and tail of the Basel series. Use the upper bounds
for the tail from exercise 4.41 to determine sufficiently large n so that

(a) |Sn L| < 1/10


(b) |Sn L| < 1/100
(c) |Sn L| < 1/1000.

(d) (Discussion) Explain how solving these inequalities help us predict how large n
has to be for us to (roughly) match a given number of decimal digits of ⇡ 2 /6.
(e) (Discussion) Explain how solving these inequalities is related to Remark 4.3.
4.3. THE COMPARISON TEST FOR POSITIVE SERIES 187

4.3 The comparison test for positive series


In a sense, the series we studied above are quite exceptional. Indeed, for most series,
finding a formula for their sum is impossible. For this reason, theorems that let us
determine whether or not a series converges are valuable. We have already seen the
divergence test. In this section, we consider the comparison test for positive series.

A fundamental observation on positive series


Grandi’s series is rather annoying. It diverges, not because its partial sums become grow
to be arbitrarily large, but since the partial sums keep jumping back and forth between 0
and 1 (that is, the partial sums diverge boundedly). The observation we want to make in
this section is that this cannot happen for positive series. Here is how they are defined:
P1
Definition 4.43 (positive series) We say that an infinite series k=1 ak is positive
if all the numbers ak are positive (i.e., satisfy ak 0).

So, what makes positive


Pn series special? Well, notice that if all ak are positive, then
the partial sums Sn = k=1 ak form an increasing sequence of numbers. That is, since
Sn+1 is obtained from Sn by adding the positive number an , the sequence {S1 , S2 , S3 , . . .}
must be increasing. But this means that the following observation on balloons applies
to these partial sums:

Proposition 4.44 (Balloon lemma) Suppose that (cn )1


n=0 is an increasing sequence
of numbers. Then one of the following is true:
(i) There exists a number M so that, for all n 2 N, we have cn  M .
In this case, the sequence converges, i.e., cn ! L for some L 2 R.
(ii) There exists no such number M . In this case, cn ! 1.

Fig. 7. By the Balloon lemma, if you release a balloon filled with helium inside of a
house, then the altitude of the balloon must converge to some finite number.
188 CHAPTER 4. INFINITE SERIES

Since the proof of the Balloon lemma requires the proper definition of the limit, we
postpone it until the end of the next chapter (page 232, to be exact).
Let us introduce some terminology that makes it easier to discuss this lemma.

Definition 4.45 (Bounded sequences) We call a sequence (cn )1 n=1 bounded from
above if there exists some number M so that cn  M for all n 2 N. Similarly, we say that
n=1 is bounded from below if the upper inequality is replaced by a lower inequality.
(cn )1
A sequence that is bounded both from below and above (possibly by different numbers)
is simply called bounded. Finally, a sequence that is not bounded is called unbounded.

Exercise 4.46 Rephrase the Balloon lemma in terms of the terminology introduced
in Definition 4.45.

Exercise 4.47 Show, by finding suitable examples, that both parts of the conclusion
of the Balloon lemma fails if the sequence cn is not increasing.
Remark: That is, find examples of sequences that are bounded from above (and even
bounded) but not convergent, and that are unbounded but do not diverge to infinity.

Now, the Balloon lemma tells us that increasing sequences are well-behaved (e.g.,
bounded divergence is not possible). In the following proposition, we have just translated
the Balloon lemma into the language of infinite series.

Proposition
P1 4.48 (positive series cannot diverge boundedly) Suppose that
k=1 a k is a positive series and let Sn denote its increasing sequence of partial sums.
Then one of the following statements is true:

(i) There exists a number M so that, for all n 2 N, we have Sn  M .


P
In this case, the series converges, i.e., 1k=1 ak = L for some L 2 R.
P
(ii) There exists no such number M . In this case, 1 k=1 ak = 1.

(In particular, a positive series can never diverge boundedly – like Grandi’s series did.)

Proof. The proposition follows immediately once we let cn = Sn in the Balloon lemma.

Exercise 4.49 Use the "Balloon lemma" in combination with exercise 4.40 to prove
that the Basel series must converge to some number L.
4.3. THE COMPARISON TEST FOR POSITIVE SERIES 189

The comparison test

In light of Proposition 4.48, when discussing a positive series, instead of saying that it
converges, we sometimes just say that it is finite and write
1
X
ak < 1.
k=1

Let us now formulate the comparison test itself.

Proposition
P1 P4.50 (Comparison theorem for positive series) Suppose that
1
k=1 ak and k=1 bk are two positive series with terms satisfying ak  bk . Then
X1 X1
(i) bk < 1 =) ak < 1
k=1 k=1
1
X 1
X
(ii) ak = 1 =) bk = 1.
k=1 k=1

Instead of writing out the proof for this result, we illustrate what is going on in a
few examples, and ask you to use these examples to figure out the proof for yourself.

Example 4.51 Does the following series converge or diverge?


1
X 1
. (4.7)
k4
k=1

By Proposition 4.48, only two things can happen: either the series converges, or it
diverges to infinity. To figure out what is going on, we notice that since k 2  k 4 , for all
k 2 N, it follows that n n
X 1 X 1
4
 .
k k2
k=1 k=1
But this means that the partial sums of the series under consideration are all smaller
than the corresponding partial sums of the Basel series. But by exercise 4.49, the Basel
series is convergent. And so, by Proposition 4.48, there exists a constant M so that, for
all n 2 N, we have n n
X 1 X 1
4
  M.
k k2
k=1 k=1
P
Applying Proposition 4.48 again, but this time to the series 1 k=1 1/k , we conclude
4

that this series must converge.


190 CHAPTER 4. INFINITE SERIES

Exercise 4.52 (a) Use the above example as a blueprint to prove part (i) of Propo-
sition 4.50. (b) Use contraposition to prove part (ii).
p
Exercise 4.53 (a) Do a proof by contradiction to prove that k  k for all k 2 N.
(b) Do a comparison to prove that
X 1
p = 1.
k=1
k

Exercise 4.54 Use suitable comparisons to determine whether the following series
converge or diverge.
X1 X1 X1
1 (log k)2019 1
(a) 2
(b) (c) .
k +1 k 2 +2 k
k
k=1 k=1 k=1

Remark: These exercises were all designed to be rather friendly. Some more difficult
exercises are given after we discuss the so-called ↵-series below.

Having something to compare with: the ↵-series


In order to efficiently apply the comparison theorem for positive series, we need something
to compare with. Above, we used the Basel series, the harmonic series and the Geometric
series to make comparisons. Out of these, the Basel and Harmonic series are special cases
of what we choose to call ↵-series3 :
X1
1
.
k↵
k=1
The following result is therefore a generalisation of Proposition 4.28 and exercise 4.49.

Proposition 4.55 (convergence of ↵-series)


1
X 1
< 1 () ↵ > 1
k↵
k=1

(In particular, the ↵-series diverge to infinity if and only if ↵  1.)

In the next exercise, we ask you to prove this result.


Exercise 4.56 (a) Prove the result for ↵ > 2 by comparing to the Basel series.
(b) Prove the result for ↵ < 1 by comparing to the Harmonic series.
(c) Prove the result for 1 < ↵ < 2 by applying a suitable version of Oresme’s trick
(that is, apply either the original trick, or the modified trick).
Remark: When we study definite integrals, we will find a much easier proof for this
proposition.
3
Some textbooks call these p-series (and use p for the parameter instead of the ↵).
4.3. THE COMPARISON TEST FOR POSITIVE SERIES 191

On the importance of tails


Before we consider examples, let us formulate a little result on the convergence of the
tail of an infinite series that can often be of help when working with examples.

Proposition 4.57 (Convergence of tails) Let K 2 N be any number. Then


1
X 1
X
ak converges () ak converges.
k=0 k=K+1
| {z }
K’th tail

In other words, if we are studying whether or not some infinite series converges, then
we can always ignore the first million terms or so and instead work with one of its tails
(indeed, they will just sum up to some finite number).
P P1
Proof. Let An and Bn be the partial sums of 1 k=0 ak and k=K+1 ak , respectively,
having an as the last term.PK Then for n large enough (well, larger than K), we have
An = Bn + C, where C = k=0 ak is a constant. From this, we see that, as n ! 1, the
sequence An converges if and only if the sequence Bn converges.

In the following example, we illustrate how the result on tails can be of help in the
intuitive step when planning on doing a comparison.

Example 4.58 Let us consider the following series:


1
X k
p . (4.8)
k=1
k5 +1

Does it diverge to infinity or converge? We figure this out in two steps.


The intuitive step: We know from Proposition 4.57 that it is only the tail of the
series that matters. This means that we can think of k as being huge. But if k is huge,
then we basically have: k k 1
p ⇡ p = 3/2 .
5
k +1 k 5 k
From this, we suspect that the series will behave like an ↵-series with ↵ = 3/2 and
therefore converge.
The formal step: Inspired by the intuitive step, we try to prove that the series
(4.8) diverges by making a comparison with the alpha series for ↵ = 3/2. We do this as
follows: k k k 1
p p = 5/2 = 3/2 .
5
k +1 5
k +0 k k
By applying the comparison test (Proposition 4.50) we are done.
192 CHAPTER 4. INFINITE SERIES

A detailed example on the use of the comparison test


Let us now tie together everything we have seen in this chapter by studying an exercise
from some old exam. In particular, the ↵-series from Proposition 4.55 will come in handy.
(Please note that this example is quite long since we are determined to try out most of
the techniques we have seen so far. When we become more experienced, the example
can be worked out in a matter of a few lines.)

Example 4.59 Let us determine whether the series following series converges or di-
verges.
X1 ⇣ p ⌘3
p
k+1 k .
k=1

p intuition: We do as in Example
First attack, p 4.58,
p and use
p our intuition. For large k,
intuitively, k + 1 ought to be close to k (e.g., 1001 ⇡ 1000). Using this, we get
⇣p p ⌘ 3 ⇣p p ⌘3 ⇣p p ⌘3
k+1 k ⇡ k+0 k = k k = 0.

Now, this does not tell us that the terms are equal to zero, but it indicates that they
may become small as k increases. In particular, it indicates that the divergence test is
unlikely to give any useful information (since it only says something if an 6 ! 0).
Second attack, numerical: We use Python to compute a bunch of partial sums:

Number of terms Value of partial sum


10 0.17194728734679288...

100 0.22244145059047477...

1000 0.23941560569718923...

10000 0.24481747485506686...

Again, this doesn’t give us a definition answer on whether or not this series converges, but
it does indicate that divergence seems unlikely (unless the series diverges excruciatingly
slowly like the harmonic series).
Third attack, algebra: To learn more about the behaviour of the terms of this series,
we try to rewrite them algebraically. A trick which is often used to make square roots
easier to handle, and which is well worth remembering, is to use (a b)(a + b) = a2 b2 :
⇣p p p
p p p ⌘ k+1+ k
k+1 k= k+1 k ·p p
k+1+ k
p p
( k + 1)2 ( k)2 1
= p p =p p .
k+1+ k k+1+ k
4.3. THE COMPARISON TEST FOR POSITIVE SERIES 193

In particular, this means that we have the identity


1 ⇣
X p ⌘3 X1
p 1
k+1 k = p p
k=1 k=1
( k + 1 + k)3

Having rewritten the terms, we can now give our intuition another go.
p p
Fourth attack, intuition: Again, we use the observation that k + 1 ⇡ k for large
k. This time, we see the following:
1 1 1 1
p p ⇡ p p = 3 3/2 = 3/2 .
( k + 1 + k)3 ( k + k)3 2 k 8k
p p
Aha! Since we no longer have cancellation between the terms k + 1 and k (the
minus sign between them has been replaced by a plus), we are getting more accurate
information. Again, these terms tend to 0 as k increases, but now we have an idea about
how fast this happens: at the speed of 1/k 3/2 . By Proposition 4.55, the ↵-sum with
↵ = 3/2 converges, and we therefore expect our series converge as well. Progress!
Fifth attack, comparisonP theorem: Now that we know what we (probably) should
compare our series with 1k=1 1/k 3/2 , it is time to use the comparison theorem. The

required term-by-term comparison is obtained as follows:


1 1 1
p p  p p = 3/2 .
( k + 1 + k) 3 ( k + 0 + k) 3 8k
According to Proposition 4.55, the alpha-series with ↵ = 3/2 is convergent. By the
comparison theorem, in combination with part (b) of Proposition 4.13, we therefore
conclude that 1
X 1
p p < 1.
k=1
( k + 1 + k)3

Exercise 4.60 Determine whether the following series converge or diverge.


1
X X1 ✓ ◆ 1 ⇣p
X ⌘
k 1 1
(a) (b) k (c) k4 + 1 k2
k3 + 1 k+1 k+3
k=1 k=1 k=1

Exercise 4.61 For which positive real numbers a does the following series converge:
X1 p
( k 2a + 1 k a )?
k=1

Exercise 4.62 (Challenge) Use comparisons, and perhaps some other tricks we have
seen, to determine whether or not the following series converges:
1
X 1
X 1
X
1 1 1
(a) (b) (c)
k! k k
ln k ln k
k=1 k=2 k=2
194 CHAPTER 4. INFINITE SERIES

4.4 On the convergence of non-positive series


Above, we considered the comparison test, which only worked when all terms are positive
(or all terms negative!). But what if the terms have varying signs? Well, we consider
two cases. The first is if the signs are alternating (such as for the alternating harmonic
series), and the second is if we don’t really know how the signs are changing.

When the signs alternate: the alternating series test


To motivate the alternating series test, let us compare the harmonic and alternating
harmonic series:
1
X 1
X
1 1 1 1 ( 1)k+1 1 1 1
= 1 + + + + ··· and =1 + + ··· .
k 2 3 4 k 2 3 4
k=1 k=1

Now, the harmonic series diverges, but the alternating harmonic series converges. In one
sense, this is slightly surprising. Indeed, the terms of both series have the same “size”.
The only difference is that the second has alternating signs. So why does this affect
convergence. Well, here is one way to visualise what is going on:

Fig. 8. The partial sums of the harmonic and alternate harmonic series interpreted
as walks on the real line. One ends up diverging to infinity, the other not.

The point is that having alternating signs allows the terms to partially "cancel" ea-
chother, and this makes it easier for the series to converge.
In fact, a careful analysis of the situation leads to the following result.

Proposition 4.63 (The alternating series test) Suppose that (ak )1 k=1 is a sequence
of positive terms that decreases monotonously to 0 as k ! 1. Then the series
X1
( 1)k+1 ak
k=1
converges to some finite value L. Moreover, if Sn is the n’th partial sum of this series,
then we have the error estimate

|Sn L|  an+1 .

Notice how the error estimate is a generalisation of the inequality from exercise 4.35.
4.4. ON THE CONVERGENCE OF NON-POSITIVE SERIES 195

P
Proof. Suppose that the series 1 k=1 ( 1)
k+1 a satisfies the conditions described in the
k
theorem. In particular, this means that the spiraling motion illustrated in Figure 8
accurately visualises the behaviour of the partial sums Sn .
The goal of the proof is to prove that the partial sums Sn converge. Our plan is
to use the "Balloon lemma" to do this. However, this plan has a flaw, the Sn do not
increase monotonously, instead they jump back and forth. But that’s not so bad. In
fact, based on the figure, we make the following claims:

• Claim 1: The subsequence S2 , S4 , S6 , ... of the partial sums is increasing, and


bounded above by, say S1 . In particular, it follows from the Balloon lemma that
this subsequence converges to some number A.
• Claim 2: The subsequence S1 , S3 , S5 , ... of the partial sums is decreasing, and
bounded below by, say 0. In particular, it follows by the Balloon lemma that
this subsequence converges to some number B.
• Claim 3: The differences S2n+1 S2n tend to 0. In particular, this means A = B.
If we denote the common value of these numbers by L, it follows that the original
sequence of partial sums S1 , S2 , S3 , . . . converges to L.
• Claim 4: We have |Sn L|  |Sn Sn+1 | = an+1 . In particular, this means that
the error estimate holds.

Notice that if we can show that all of the above claims are true, then the result holds.
We now proceed to prove claim 1 (we leave the remaining claims to the exercises).
First, we check that the subsequence (S2n )1
n=1 = (S2 , S4 , S6 , . . .) is increasing. That
is, we have to show that S2n  S2n+2 for all n. But, this follows from the computation:
S2n+2 S2n = a2n+2 + a2n+1 0.
(Here, we used the fact that the an are decreasing.)
Next, we check that the subsequence (S2n )1
n=1 is bounded above. This follows from
observing that:
S2n = a1 a2 + a3 a4 + a5 ··· a2n 2 + a2n 1 a2n

= a1 (a2 a3 ) (a4 a5 ) ··· (a2n 2 a2n 1) a2n  a1 .


| {z }
0

(Again, we used the fact that the an are decreasing.) Finally, by applying the Balloon
lemma, the last part of claim 1 follows.

Exercise 4.64 Complete the proof by proving claims 2, 3 and 4. To help you, we
split this task into three sub-tasks:
(a) Claim 2 is proved in pretty much the same way as Claim 1. However, you need
to show that the following is a consequence of the Balloon lemma: "If a sequence
is decreasing and bounded below, then it converges."
196 CHAPTER 4. INFINITE SERIES

(b) One way to prove claim 3 is to establish a chain of equalities:

A B = · · · = 0.

Hint: As an additional hint, recall Remark 4.14.


(c) One way to prove claim 4 is to divide the situation into two cases: when n is even
and when n is odd. In both cases, you know more or less how Sn , Sn+1 and L are
placed on the real line (relative to each other). Use this to justify the inequality
|Sn L|  |Sn Sn+1 | (which is just a statement about distances).
Exercise 4.65 Determine what the alternating series test says about each of the
following series.
X1 1
X 1
X
( 1)k cos(⇡k) sin(k)
(a) p (b) (c)
k k k
k=1 k=1 k=1

Exercise 4.66 Earlier in this chapter, we studied the Geometric series, and the for-
mula
1
= 1 + x + x2 + x3 + x4 + x5 + · · · .
1 x
In the last chapters of these lecture notes, we are going to study similar formulas such
as
x2 x3 x4
ln(1 + x) = x + + ···
2 3 4
x3 x5 x7
arctan(x) = x + + ··· .
3 5 7

(a) What does the divergence test say about each of the above series?
(b) What does the alternating series test say about each of the above series?
(c) Use the error estimate from the alternating series test to predict how many terms
you need from the series for the arctangent for it to approximate arctan(1/2)
to an accuracy of at least 1/1000. By making a numerical computation (using
Python or some other tool), check how close this series actually gets to the value
of arctan(1/2).
4.4. ON THE CONVERGENCE OF NON-POSITIVE SERIES 197

What to do when we have no control of the signs: Absolute convergence


We now deal with the following question: how to study the convergence of series where
we have a poor understanding of the signs of the terms? Indeed, everything we have
done so far is based on us having very strict control of the signs of these terms (in fact,
either that they are all positive or that they alternate).
As it turns out, the reasonable thing to do is
what any 2 year old would: we throw all the
signs away! That is, our strategy is to force all
the terms to be positive by taking their absolute
values. So, instead
P1 of studying the convergence
of the seriesP k=1 ak directly, we study the con-
vergence of 1 k=1 |ak |.

To this end, we make the following definition. Fig. 9. Our strategy.


P1
Definition
P1 4.67 We say that
P1 an infinite series k=1 ak is absolutely convergent if
k=1 |ak | is convergent. If k=1 ak is convergent, but not absolutely convergent, we say
that the series is conditionally convergent.

The following proposition explains why the concept of absolute convergence is useful.

Proposition 4.68 Absolute convergence implies convergence.

That is, if we have a series with signs that are hard to understand, we can throw
away the signs by taking absolute values of all the terms, and then study the positive
series obtained in this way. If this positive series converges, then so does the original
one. We will give a proof for this below, but let us first consider an example.

Example 4.69 Let us determine whether the series


1
X sin(k 2 )
(4.9)
k2
k=1

converges. The first thing to notice is that this is not a positive series. Indeed, the signs
are all over the place:

sin(1) = 0.8414... sin(42 ) = 0.2879...

sin(22 ) = 0.7568... sin(52 ) = 0.1323...

sin(32 ) = 0.4121... sin(62 ) = 0.9917...


198 CHAPTER 4. INFINITE SERIES

What does our intuition tell us? Well, the series does look a bit like the Basel series,
which is convergent. Moreover, since sin(k 2 ) is always between 1 and 1, the terms of
our series are smaller in "size" than those of the Basel series, and, finally, since the terms
are not all positive, there should be some "cancellations" going on in the series. All in
all, we expect this series to converge. But how to prove it?
Well, let us prove that it converges absolutely. But this is not so hard, indeed, observe
that when taking absolute values, we get

| sin(k 2 )| 1
2
 2.
k k
But this means, by the comparison test for positive series, that the series
1
X | sin(k 2 )|
k2
k=1

converges. By definition, this means that the series given by formula (4.9) converges
absolutely. And finally, by Proposition 4.68, this means that the series given by (4.9)
converges in the usual sense.

Exercise 4.70 Determine whether the following series diverge, converge conditionally
or converge absolutely.

X1 1
X
( 1)k ( 1)k
(a) p (b)
k k2
k=1 k=1

Proof of Proposition 4.68.


P1 The proof is basically just an exercise inP
book-keeping. Sup-
1
pose that we know that P a
k=1 k is absolutely convergent. That is, k=1 |ak | converges.
1
To deduce from this that k=1 ak converges (in the usual sense), we need to involve the
signs of the terms somehow. So, let us make the following definitions:
( (
+ ak if ak 0 0 if ak 0
ak = and ak =
0 if ak < 0 ak if ak < 0

From this definition, we can see that both a+


k and ak are positive and that ak = ak
+
ak
and |ak | = a+ + a . In particular, the inequalities a+
k  |a k | and a k  |a k | both hold.
This means that by applying the standard comparison test for series twice, we get
1
X 1
X 1
X
|ak | converges () both a+
k and ak converge.
k=1 k=1 k=1
4.4. ON THE CONVERGENCE OF NON-POSITIVE SERIES 199

Moreover, using that ak = a+


k ak , it follows by Proposition 4.13, that
1
X 1
X 1 ⇣
X ⌘ 1
X
a+
k ak = a+
k ak = ak .
k=1 k=1 k=1 k=1

SinceP
a difference between two convergent series is itself convergent, we obtain that the
sum 1 k=1 ak converges, and we are done!

Exercise 4.71 Again, let us consider the formulas

1
= 1 + x + x2 + x3 + x4 + x5 + · · ·
1 x
x2 x3 x4
ln(1 + x) = x + + ···
2 3 4
x3 x5 x7
arctan(x) = x + + ··· .
3 5 7
For all x determine whether or not these series diverge, converge conditionally, or
converge absolutely.

Remark 4.72 (The failure of commutativity P for infinite series) We end this
chapter with a rather shocking fact. If a series 1k=1 ak converges conditionally, then by
only changing the order in which we sum the terms, we can get this series to be equal
to any number that we desire. The most natural example of a conditionally convergent
series is the alternating harmonic series. Now, recall that Mengoli discovered that
1
X ( 1)k+1
= ln(2).
k
k=1

However, we now claim that it is possible, just by changing the order in which we sum
the terms, to get, say,
X1
( 1)k+1
= 2.
k
k=1

Briefly explained, the trick for doing this is to to build a "twisted" partial sum as follows:

1. Sum the positive terms of the alternating harmonic series, one by one. Stop when
your "twisted" partial sum gets above 2 for the first time (we ask you to check in
an exercise below that this will happen sooner or later).
2. Add the first negative term of the alternating harmonic series to your "twisted"
partial sum. This will make it drop below 2.
200 CHAPTER 4. INFINITE SERIES

3. Add as many positive terms that you need to the "twisted" partial sum to get it
above 2 again.
4. Add the second negative term to the sum obtained in the previous steps. This will
put you below 2 again.

Continuing in this way, all terms from the alternating harmonic series will, sooner or
later, by added to the "twisted" partial sum (and so, in this sense, we are summing all
the terms). Moreover, since the terms we are adding are getting smaller and smaller, the
"twisted" partial sum, will converge to the value 2.

The following exercises are meant to help you understand the above remark.

Exercise 4.73 Either by hand, or by using Python, plot the first 50 (or however
many you have patience for) terms of the "twisted" partial sum described in the above
remark. Do they seem to converge to the value 2?
Exercise 4.74 (a) Prove that if you sum the positive and negative terms from the
alternating harmonic series separately, then both sums diverge. In particular,
this means that it is possible to do what we claim in step 1 of Remark 4.72.
P
(b) More generally, suppose that ak is a conditionally convergent series. Prove
that if you sum of all the positive and all the negative terms separately, then
both of these sums diverge.

In contrast to conditionally convergent series, absolutely convergent series are much


more well behaved. In fact, the following is true.

P
Proposition 4.75 Suppose that 1 k=1 ak is absolutely convergent. Then all choices of
order of summation, as long as every term is included sooner or later, will give the same
value for the series.

Exercise 4.76 (Very optional) Prove the above proposition.


Exercise 4.77 Mark each of the following statements as TRUE or FALSE. Moreover,
in case the statement is true, provide a proof, and if it is false, provide a counter-
example.
P
(a) If ak ! 2, then 1 k=1 ak = 2.
P1
(b) If k=1 ak = 2, then ak ! 2.
P
(c) If 1 k=1 ak = 2 then a ! 0.
P1 k
(d) If ak ! 0, then k=1 ak converges.
P
(e) If ak ! 0 monotonously, then 1 a converges.
P1 P1 k=1 k
(f ) If k=1 ak converges, then k=1 a2k also converges.
4.5. EXAM EXERCISES 201

4.5 Exam exercises


Exercise 4.78 Determine whether the following series converge or diverge.
1
X 1
X
k2 1
(a) (b) .
k4 k2 + 1 cos(1/k)
k=1 k=1

Exercise 4.79 Determine whether the following series converge or diverge.


1
X 1
X
k 1/(ln k)2
(a) (b) e .
k2 1
k=2 k=2

Exercise 4.80 Determine whether the following series converge or diverge.


1
X 1
X
1 2
(a) p (b) e1/k .
2
k +1+k
k=1 k=1

Exercise 4.81 Determine whether the following series converge or diverge.


1 1
s
X ln k X n+2
(a) (b) 3
.
k n (n2 + 1)
2
k=1 k=1

Exercise 4.82 Determine whether the following series converge or diverge.


1 p
X 1
X p
( k2 +1 k)
(a) ( k2 + 1 k) (b) 2 .
k=1 k=1

Exercise 4.83 Determine whether the following series converges or diverges.


1
X 1
k=2
(ln k)ln(ln k)

Exercise 4.84 Determine whether the following series converges or diverges


1
X 2k
1 + 3k
k=1

Exercise 4.85 Determine whether the following series converges or diverges.


1
X 1 ⇣p
X ⌘
2
(a) e1/k (b) k4 + 1 k2 .
k=1 k=1
202 CHAPTER 4. INFINITE SERIES

P1
Exercise 4.86 (a) Prove or disprove the following. If k=1 ak diverges, then the
sequence ak must diverge.
(b) Determine whether the following seres converge or diverge.
1 p 1
X k2 + 1 X 2k
(i)
k2 1 + lnk (k)
k=1 k=1

P
Exercise 4.87 Suppose that the series 1k=1 ak converges to the value 3.2. Does this
mean that an ! 3.2 as n ! 1? Explain by referring to suitable results or examples.
P1
Exercise 4.88 (a) Define what it means for k=1 ak to be convergent and diver-
gent, respectively.
(b) Determine whether the following series converges:
!
1
X ⇣ k⌘ 1 1 + k1
1+ .
2 k/2
k=1

Exercise 4.89 (a) Does the following implication hold?


1
X
lim an = 0 =) an converges.
n!1
n=1

Motivative your answer with a proof or counter example.


(b) Determine the convergence of
1
X k!
.
3k
k=1

Exercise 4.90 Do there exist functions f that makes both of the following series
converge at the same time?
1
X 1
X
1 f (k) and 1 + f (k) .
k=1 k=1

Exercise 4.91 A disc with radius 1 is coloured according to the pattern shown in
Figure 1. At every step, the angle and radius of the area to be coloured is halved.

(a) Suppose that this process is repeated an infinite number of times. Write down a
series that expresses the total coloured area.
(b) Compute the sum of the series you found in (a).
4.5. EXAM EXERCISES 203

� �

� �

Fig. 10. Illustration Exercise 4.91.


204 CHAPTER 4. INFINITE SERIES

4.6 Answers to selected exercises


4.4 (a) 1, (b) 0, (c) diverges.

4.5 (a) Hint: use exercise 4.4, (b) hint: given two sequences (an )1 n=1 and (bn )n=1 , it
1

is possible to form a third sequence (a1 ,b1 ,a2 ,b2 ,a3 ,b3 , . . .).

4.8 Almost any series will be a counter-example.

4.11 The Grandi series has no value. Its partial sums diverge boundedly.

4.15 We need the computational rule limn!1 (Can ) = C limn!1 an .

4.19 (a) diverges, (b) no conclusion.

4.21 (a) 4/3, (b) 2, (c) diverges, (d) diverges.

4.22 Diverges for all |x| 1. No conclusion for |x| < 1.


P
4.24 The sum nk=0 (xk xk+1 ) is what we call telescoping. Indeed, only the first and
last terms survive when all cancellations are taken into account:
n
X
(xk xk+1 ) = (x0 x1 ) + (x1 x2 ) + · · · + (xn xn+1 ) = x0 xn+1 .
k=0

4.27 (a) n > 9204, (b) n > 11506, (c) n > 13807.

4.29 (a) Probably somewhere between 20 and 30, (b) hard to say.

4.30 This happens when (roughly) when n ⇡ 1014 .

4.31 (a) S4 2, S8 5/2, S16 3, (b) S2m 1 + m/2, (c) since the sequence Sn is
growing, this ought to say that Sn ! 1.

4.32 (a) n > 218 , (b) n > 2198 , (c) n > 21998 .

4.34 S10 , S100 , S1000 and S10000 match 1, 1, 2 and 3 digits of ln(2), respectively.

4.35 Basically, the n + 1’th step is longer than the distance from Sn to the true value.

4.36 (a) n > 9, (b) n > 99, (c) n > 999.

4.38 (a) My computer started getting into trouble when computing S108 . At this point,
it matched the first 8 digits of ⇡ 2 /6. (b) The graph definitely seems to stabilise at
some height.

4.39 This happens when (roughly) n ⇡ 108 .


4.6. ANSWERS TO SELECTED EXERCISES 205

4.40 (a) S3  1.5, S7  1.75, S15  1.875, (b) S2m 1  2 1/2m 1  2, (c) since Sn is
increasing this ought to say that Sn  2 for all n.

4.42 (a) m > 2, (b) m > 5, (c) m > 8.

4.49 Since Sn is increasing and bounded by 2, the conclusion follows immediately from
the Balloon lemma.
P
4.52 (a) If the series 1k=1 bk converges, then by Proposition 4.48, its partial sums (why
not denote them P Bn ) are bounded by some number M . Since ak  bk , the
by
partial sums of 1 k=1 ak (why not denote them by An ), satisfy An  Bn . But this
means that the partial sums for An are also all bounded by M . So, by Proposition
4.48, we are done.

4.53 (b) by (a), the series is larger than the harmonic series.

4.54 (a) converges due to comparison with Basel series, (b) diverges due to comparison
with harmonic series, (c) converges due to comparison to Geometric series with
x = 1/2.

4.56 (c) Since the alpha-series with ↵ 2 (1,2) are all supposed to converge, use the
modified Oresme trick more or less exactly as we used it to get an upper bound
for the Basel series.

4.60 (a) converges, (b) diverges, (c) converges.

4.61 converges if and only if a > 1.

4.62 (a) converges, (b) converges, (c) diverges.

4.64 (a) To prove the statement on the Balloon lemma, notice that if the sequence dn is
decreasing and bounded below by some constant M1 , then the sequence cn = dn
is increasing and bounded above by the constant M = M1 . So, by the Balloon
lemma... (b) Let us give the first step in the chain of equalities:

A B = lim S2n lim S2n+1 = ...


n!1 n!1

4.65 (a) converges, (b) converges, (c) no conclusion.

4.66 (a) by the divergence test, both diverge for |x| > 1, (b) by the alternating series
test, the series for the logarithm converges for x 2 [0,1], and the series for the
arctangent converges for x 2 [ 1,1]. For x 2 [ 1,0), the alternating series test
gives no information about the series for the logarithm (it does not alternate).

4.70 (a) converges conditionally, (b) converges absolutely.


206 CHAPTER 4. INFINITE SERIES

4.71 All three series diverge for |x| > 1 (divergence test). They all converge absolutely
for |x| < 1 (comparison with Geometric series). For |x| = 1, the first series diverges,
while the third converges conditionally. The second converges conditionally for
x = 1 and diverges for x = 1.

4.74 (a) The point is to show that


1
X X
1 1
= 1 = 1.
2k 2k + 1
k=1 k=1

We can do both by comparing to the harmonic series. (b) In the proof of Proposi-
tion 4.68, it is observed that
1
X 1
X 1
X
|ak | converges () both a+
k and ak converge.
k=1 k=1 k=1

In particular, if the series on the left-hand side is conditionally convergent, then


at least one of the two series on the right-hand side have to diverge. However, if
only one of them diverges, then the series on the left-hand side has to be divergent
(why?). So, for a conditionally convergent series, both the positive terms and the
negative terms, when summed separately, have to diverge.

4.77 (a) false, (b) false, (c) true, (d) false, (e) false, (f) false.
Chapter 5

Limits for sequences

Introduction
Up until this point, we have mostly used limits
without defining exactly what they are1 . This
approach matches the historical development.
Indeed, scientist working in the 18th century,
and earlier, such as Archimedes, Oresme, Men-
goli, Euler and even Newton, worked with limits
without ever giving a proper definition for what
this meant.
In this chapter we follow the historical devel-
opment into the 19th century when we define pre-
Fig. 1. The definition of the limit was
cisely what we mean by the limit of a sequence,
actually first widely introduced into
and use it to develop a proper theory that puts
mainstream mathematics to more ef-
our intuition on a firm mathematical ground.
ficiently teach French soldiers calcu-
1
Well, this is not really true, in Chapter 4.2, we tried lus in the early 1800s.
to trick you into taking the definition of the limit out for
a spin – see exercises 4.27, 4.32, 4.36 and 4.42

Remark 5.1 (Selected problems from previous exams based on this chapter)

1. Give the definition for limk!1 ak = 0, and use this definition to show that
sin k
lim p = 0.
k!1 k
2. Give the definition for limk!1 ak = 1, and use this definition to show that
n!
lim = 1.
k!1 20n

207
208 CHAPTER 5. LIMITS FOR SEQUENCES

5.1 The definition for the limit of a sequence


Step 1 of 2: What it means for positive sequences to converge to zero
We begin by formulating the following special case of the definition of the limit in order
to capture the main idea of what is going on in this chapter, and perhaps the main idea
of what is going on in these lecture notes.

Definition 5.2 The positive sequence (an )1


n=1 is said to converge to 0 if:

For every ✏ > 0, there exists N > 0 so that for all n 2 N it holds that

n>N =) an < ✏.
If this holds, we write

lim an = 0 or an ! 0 as n ! 1.
n!1

If (an )1
n=1 does not converge to 0, then we write an 6 ! 0.

The Czech mathematician Bernard Bolzano was


the first to properly state this rather abstract
definition of what we ought to mean by a limit.
However, living in the mathematical backwater
of Prague, it took around 50 years for anyone
to notice (which also gave others – such as the
French – the time to discover it independently).
When you read the example below, notice how
Definition 5.2 can be thought of as a game of Fig. 2. Bernard Bolzano (1781 –
"challenges and responses". 1848). Never afraid of a challenge.

Example 5.3 We prove that


1
lim
= 0.
2n
n!1

To do this, we need to show that the conditions of Definition 5.2 are satisfied when
an = 1/2n . That is, we must prove that given any ✏ > 0, we can find a number N , so
that
1
n > N =) n < ✏.
2
(The hard part is usually to understand that this is what the definition asks of us.)
While this proof can be done in one line, let us take our time to figure out what is
going on. We begin by taking a specific epsilon, say ✏ = 1/5. Since

21 = 2, 22 = 4, 23 = 8, 24 = 16,
5.1. THE DEFINITION FOR THE LIMIT OF A SEQUENCE 209

it seems that if n > 2, then we have 1/2n < ✏. But what if ✏ is something else, such as
✏ = 1/10? Well, in this case, by the same computation, n > 3 implies 1/2n < ✏.

Fig. 3. In the language of the definition, for ✏ = 1/5, we choose N = 2, and for
✏ = 1/10, we choose N = 3. Notice how we can think of being challenged with an ✏
and then having to respond by giving a suitable N .

The point is this: to be able to say that an ! 0, we need to show that for all "✏-
challenges" (in the above figures, the challenge is represented by the height of the green
line), we need to be able to respond by saying how far to the right do we have to go
before every an is smaller than ✏. That is, we have to figure out how large N has to be
so that an < ✏ whenever n > N .
But how to do this for every ✏ > 0? Let us make two observations:
• It is hopeless to work with specific ✏, like we did above (that was just for fun).
However, to simplify life, we like to pretend that we can. Indeed, we usually
assume that ✏ is a fixed, but unknown, number given to us by some person we
respect, but are a little afraid of (like, say, my grandmother). This is supposed to
remind us that we can use ✏ in our formulas – as if it was a concrete number – but
that we are not allowed to change ✏ or assuming anything about it (except that it
is strictly positive), since that would make grandmother angry.
• We cannot expect a specific N to work for every ✏. Indeed, the typical situation is
that the smaller ✏ becomes, the larger N has to be. In other words, N is going to
be a function of ✏ (sometimes we even write N (✏) to emphasise this).
So, we assume that we have our concrete, but unknown, challenge ✏. Our goal is the
same as for the concrete values we considered above: to respond with an N . To do this,
we need to understand how large n needs to be for the inequality 1/2n < ✏ to hold. That
is, we need to solve this inequality for n. This is exactly the type of thing we did in
Chapter 0:
1 1
n
< ✏ () 2n >
2 ✏
1
() log 2n > log

210 CHAPTER 5. LIMITS FOR SEQUENCES

() n log 2 > log ✏


log ✏
() n > .
log 2

That is, given ✏, a suitable choice for N is log(✏)/ log(2). Since this works for every
✏ > 0, we have proved that by Definition 5.2 it holds that limn!1 (1/2n ) = 0.

After reading the above example, you may be excused for thinking that the formula
we found for N (as a function of ✏) is rather ugly and uninformative. But nothing could
be farther from the truth. Indeed, let us see what happens for ✏ = 1/5 and ✏ = 1/10.
This gives us:
log ✏ n2N
✏ = 1/5 =) n = 2,32... =) n > 2
log 2
log ✏ n2N
✏ = 1/10 =) n = 3,32... =) n > 3.
log 2
But this is exactly what we got when we did this "by hand"! Moreover, the formula will
tell us exactly which N we are supposed to choose no matter the ✏. Sort of nice, no?

Exercise 5.4 Answer the following questions on the above example.

(a) Use the formula to find suitable N when ✏ = 1/1000 and 1/10000.
(b) Use a while-loop to write a program in Python that for any given ✏ produces the
first n so that an < ✏. Verify that it matches what you found in (a).
(c) As we are challenged with smaller and smaller ✏, does N become bigger or smaller?
Does this match the formula for ✏ from the above example? Can you explain this
graphically?
(d) We seem to be focused on small ✏. What is a valid choice for N for large ✏?
Say, ✏ = 100? Try to justify your answer using both your Python program, the
inequality obtained above, and by thinking graphically.

Exercise 5.5 (a) Use the definition of the limit (as in Example 5.3) to prove
5
lim = 0.
n!1 2n + 3

(b) Make a program using a while-loop in Python to verify that your formula for N
in part (a) makes sense.

Exercise 5.6 Here is an exercise that at first may seem awkward. Prove that
1
lim = 0.
n!1 n!
5.1. THE DEFINITION FOR THE LIMIT OF A SEQUENCE 211

In the next example, now consider a sequence that has no limit.

Example 5.7 We show that the sequence an = 1 + ( 1)n does not converge to zero.
This formula may look a bit strange, but if we check the first few numbers, we notice
that
(an )1
n=1 = (0, 2, 0, 2, 0, 2, 0, . . .).

Intuitively, it should be quite clear that the


sequence does not converge to zero. But how
to prove this? Well, suppose that we are
challenged with ✏ = 1. Then the point is
that no matter how large we choose N , we
can always find a number n > N so that
an > ✏. For instance, this holds if we choose
n = 2N . Therefore, we have found one value
of ✏ for which the definition cannot hold, and
so it is impossible that an ! 0. (Notice, we
Fig. 4. Here, ✏ = 1. How far do we have
have not proven that this sequence diverges
to move to the right so that the an < ✏?
– which it does – but only that it does not
converge to 0.)

Exercise 5.8 We are given two sequences an


(red) and bn (blue) as in Figure 5.

(a) For what value of N does it seem that


n > N =) an + bn < ✏ for the indicated
epsilons?
(b) Suppose that both an and bn converge to 0.
Also, suppose that we are challenged with
some unknown, but fixed, ✏ > 0. Explain
why there must exist some number N so
that an + bn < ✏ whenever n > N . Try
to be as concrete as possible so that your
explanation counts as a proof.

Exercise 5.9 Here is a warm up for the next


section:

(a) Give a definition for limn!1 an = 0 when


all an are negative.
(b) Give a definition for limn!1 an = 0 when Fig. 5.
the an may be both positive and negative.
212 CHAPTER 5. LIMITS FOR SEQUENCES

Step 2 of 2: The general definition of the limit for sequences

We now explain how to pass from the spe-


cial case of the definition of limits to the
general case. In the general definition of
the limit, what we really want to capture
mathematically is what we really mean by

an ! L as n ! 1.

But this should be the same as saying that


an L approaches the value 0 as n in-
creases. Since we do not care about the
sign of an L, we introduce absolute val-
ues, and this becomes

|an L| ! 0 as n ! 1.

But now a nice happens: if you apply the Fig. 6. Top: Here, some sequence an that
definition of what it means for the positive appears to be approaching the limit L as n
sequence |an L| to approach 0, you get grows. Bottom: That |an L| approaches
exactly the general definition for the limit. zero means that the length of the green lines
That is, we get the following definition: on the figure to the left go to zero as n grows.

Definition 5.10 (The limit of a sequence) The sequence (an )1


n=1 is said to converge
to the finite value L if the following holds

For every ✏ > 0, there exists N > 0 so that for all n 2 N it holds that

n>N =) |an L| < ✏.

If this holds, we write

lim an = L or an ! L as n ! 1.
n!1

If an does not converge to any finite value L, then we say that the sequence diverges.

Just like like in the special case (Definition 5.2), the general definition tells us that to
prove that an ! L, we must be able to win a game of challenges and responses. Indeed,
for every ✏ > 0 we are challenged with, we need to be able to respond with a number
5.1. THE DEFINITION FOR THE LIMIT OF A SEQUENCE 213

N (that depends on ✏) telling us how far we have to go to the right before the distances
|an L| become less than ✏.

Fig. 7. Left: The sequence from Figure 6 is challenged by an epsilon. Right: Choos-
ing N = 6, it seems we successfully responded to this challenge.

As we noted in Example 5.3, please keep in mind that N is a function of ✏. When we


get challenged with smaller and smaller ✏, we usually have to respond with larger and
larger N to compensate. In a sense, it really is a game of cat and mouse.
Let us consider an example. Notice that the only difference from what we did in
Example 5.3 is that we need to involve absolute values.

Example 5.11 We prove that


n
lim = 1.
n!1 n + 1

According to the definition above, we need to


show no matter which ✏ > 0 we are challenged
by, then by choosing n large enough, we get

n
1 < ✏.
n+1
Notice that the situation is exactly the same as in
Example 5.3, except that the sequence 1/2n has
Fig. 8. The sequence an = n/(n + 1).
been replaced by the sequence |(n/n + 1) 1|.

So, what do we do? Well, the first thing is to try to simplify the expression for the
terms:
n n n+1 1 1
1 = = = .
n+1 n+1 n+1 n+1 n+1

This means that the question has become: given ✏ > 0, fixed but unknown, find a formula
for how large n must be for us to have 1/(n + 1) < ✏. Well, this is something we (should)
214 CHAPTER 5. LIMITS FOR SEQUENCES

know how to do:


1 1 1
< ✏ () n + 1 > () n > 1.
n+1 ✏ ✏
Again, we have found a formula for how to choose n, and again, we are done!

Exercise 5.12 Prove that the sequence of partial sums Sn from the Geometric series
P 1
k=0 1/2 converges to 2.
k

Hint: Why not try to get rid of the absolute value in the expression |Sn L|?

Let us now consider a slightly more complicated example.

Example 5.13 We prove that

2n2 3n + 5
lim =2
n!1 n2 9
According to the definition above, we need to
show no matter which ✏ > 0 we are challenged
by, then by choosing n large enough, we get

2n2 3n + 5
2 < ✏.
n2 9
Fig. 9. The sequence an =
So, what do we do? Well, we compute! The first (2n2 3n + 5)/(n2 9). Notice that
thing is to simplify the expression inside of the something weird happens at n = 3.
absolute value: Do you see why this is?

2n2 3n + 5 2n2 3n + 5 2n2 + 18 | 3n + 23|


2
2 = 2
= .
n 9 n 9 |n2 9|
Keep in mind that our goal is to convince ourselves that for n large enough, this expres-
sion is smaller than ✏. In particular, if n 8, then | 3n + 23| = 3n 23. Moreover,
for n 78, it also holds that |n2 9| = n2 9. This allows us to simplify the fraction
as follows:
| 3n + 23| 3n 23
2
= .
|n 9| (n 3)(n + 3)
The next natural step would be to try to solve the inequality we obtain by putting this
less than ✏. However, this would lead to rather annoying computations. To avoid this, let
us try to simplify this expression further. We can do this as follows: for the denominator,
we observe that 3n 23  3n. Moreover, for the numerator, we observe that n + 3 n
(which gives 1/(n + 3)  1/n). In combination, this gives
3n 23 3n 3
 = .
(n 3)(n + 3) (n 3)n n 3
5.1. THE DEFINITION FOR THE LIMIT OF A SEQUENCE 215

Finally, this expression is friendly enough for us to solve:


3 3
< ✏ () + 3 < n. (5.1)
n 3 ✏
In summary, what we did above shows that
8
<n 8 2n2 3n + 5 3
=) 2  < ✏.
:n > 3 + 3 n2 9 n 3

In particular, this means that whatever ✏ > 0 we are challenged with, we beat this
epsilon-challenge by choosing n larger than the maximum of 8 and 3/✏ + 3. We can
formulate this by saying that given ✏ > 0, we choose N = max{8, 3/✏ + 3}.

Exercise 5.14 (a) In the previous example, we see in the figure that something
strange happens at n = 3. Why is this, and is this a problem for our computations
and/or our final answer?
(b) Write a program in Python that checks whether or not the formula for N in
the above example makes sense. That is, the program should, given a value of
epsilon, compute N according to the formula in the example, and then check how
close the sequence is to the value 2 for a couple of n larger than N. Do you need
to take the observation from (a) into account?
Remark: For this exercise, the Python function np.ceil() can be useful. Given
a floating point number, this function returns the smallest integer larger than it.

Exercise 5.15 (Discussion) Is it possible to write a Python program to prove that


some limit holds? What about writing a program to show that a limit does not hold?
Exercise 5.16 Prove that
6n n2 n + 1
(a) lim =3 (b) lim =1
n!1 2n + 5 n!1 n2 + n 1

Hint: For (a), you can more or less do as in the first example above, but for (b), you
need to follow the technique of the second example.

Exercise 5.17 Prove that


5n
lim = 0.
n!1 n!

Hint: Here, the key is to write out what the denominator and numerators really mean.
216 CHAPTER 5. LIMITS FOR SEQUENCES

5.2 The rulebook for limits of sequences


You may be happy to hear that when com-
puting limits in practice, we rarely need
to work the actual definition. Instead, we
usually rely on softer arguments that we
can justify thanks to a "rulebook" for lim-
its that we establish in this chapter. In
fact, the real reason for why we have been
using the definition to check limits above,
was merely that we wanted to take our new
piece of mathematical technology out for a Fig. 10. Only after taking your new toy
spin. out for a test-drive would you trust it to
take you where no man has gone before!
To get a feeling for what the rulebook for
limits will look like, we start by doing some intuitive computations below. We will see a
situation where our intuition works, and one where it fails.

An example where a naive intuitive approach works


In the following example, we illustrate the typical intuitive approach used by mathe-
maticians when computing limits. The main idea of this section is to both justify this
approach and explain its limitations.

Example 5.18 In exercise 5.16 we used the definition of the limit to check that
6n
lim = 3.
n!1 2n + 5

However, to actually figure out that the limit is equal to 3, you may agree that the
following intuitive computation seems quite convincing:
6n n large 6n
lim ⇡ lim = lim 3 = 3.
n!1 2n + 5 n!1 2n n!1
Here, we simply say that when n is of the size, let us say a few gazillions (to use
a technical term from Donald Duck), then the +5 in the denominator can safely be
ignored. To support this conclusion, we can use Python either to plot a few thousand
terms, or to simply compute what happen for a few large values of n by using the
command print(6⇤n⇤(2⇤n+5)⇤⇤( 1)).

Exercise 5.19 The same approach used in the above example should work to deter-
mine the following limits.
p
(n + 10)(n + 100) 9n6 + 1 + n3 n2 + sin(3n + 1)
(a) lim (b) lim (c) lim
n!1 100(n3 + 1) n!1 2n3 + 5 n!1 3n2 + cos(3n + 2)
5.2. THE RULEBOOK FOR LIMITS OF SEQUENCES 217

An example where the same naive intuitive approach fails


While the method of example 5.18 is convenient, it is easy to slip up when you do not
actually know what you are doing2 .

Example 5.20 We consider the limit


✓ ◆
n2
lim n .
n!1 n+2
Arguing as in Example 5.18, it is intuitively clear that for large n, we can ignore the +2
in the denominator. This implies that
n2 n2
⇡ = n,
n+2 n
which again leads to
✓ ◆
n2
lim n ⇡ lim (n n) = lim 0 = 0.
n!1 n+2 n!1 n!1

Case closed! No? Ok, just to be sure, let us use Python to visualise what is going on:

Fig. 11. To the left we have the first 20 terms. They do not seem to go to zero. To
the right, we include the first 100 terms. Still no sign of tending to zero.

Whops! It seems that 2, and not 0, is the correct limit for the sequence. Let us see what
happens if we rewrite the expression before computing the limit:
n2 n(n + 2) n2 2n
n = = .
n+2 n+2 n+2
We now use the same intuition as above, namely that n + 2 ⇡ n for large n. This gives:
✓ ◆
n2 2n 2n
lim n = lim ⇡ lim = 2.
n!1 n+2 n!1 n+2 n!1 n
Well, that is convenient. This time the answer matches Python’s suggestion.

2
Which is why, at the end of the day, we need a formal definition for the limit to separate between
what is correct and what is not.
218 CHAPTER 5. LIMITS FOR SEQUENCES

A rulebook for computing limits


Our goal now is to figure out what are allowed to do when computing limits. Our plan
is as follows. First, we state the entire "rulebook" for computing limits. Second, we play
around with these rules in order to understand how to use them. And finally, we prove
them.

Proposition 5.21 (Rulebook for the limit of sequences) Suppose that the limits
limn!1 an and limn!1 bn exist (and are finite). Then the following hold:
⇣ ⌘
(i) lim an ± bn = lim an ± lim bn
n!1 n!1 n!1

(ii) lim an · bn = lim an · lim bn


n!1 n!1

If, in addition, limn!1 bn 6= 0, then

(iii) lim an /bn = lim an / lim bn


n!1 n!1 n!1

The following rule is rather famous, and is often called the Squeeze theorem:

(iv) If an  bn  cn , and lim an = lim cn = L, then lim bn = L.

The following rule cannot really be understood until we reach Chapter 6:


(
If f is continuous and both the an and its limit
(v)
are contained in Df , then lim f (an ) = f ( lim an )
n!1 n!1

Moreover, the following concrete limits are useful enough to be included here:
1
(vi) lim C = C (vii) lim = 0.
n!1 n!1 n
Here C is a constant.

Exercise 5.22 (a) Prove rule (vi), and (b) prove rule (vii), above.
Exercise 5.23 Prove that an ! 0, bn ! 0 =) an + bn ! 0.
Remark: This just asks you to write the solution of exercise 5.8(b) properly.
The reason why rule (v) cannot really understood right now is that we do not yet
know what it means for a function f to be continuous. We also mention that some
textbooks state this rules in a slightly different way which avoids mentioning continuity:
(v’) lim an = A and lim f (x) = B =) lim f (an ) = B.
n!1 x!A n!1

(We remark that to make this rule look slightly nicer than it actually is, we are skipping
one little technicality. We point it out in the next chapter.)
5.3. HOW TO USE THE RULEBOOK IN PRACTICE 219

5.3 How to use the rulebook in practice


Indeterminate forms
The first step to understanding how to use the rulebook for limits is to notice that every
rule requires that all the limits involved exist and are finite numbers. For this reason,
it is important to be able to know when this is not the case. To keep track of this, we
introduce a notation for so-called indeterminate forms. To explain what this is, we recall
the following two expression that we considered in Example 5.20:
⇣ n2 ⌘ 2n
lim n and lim .
n!1 n+2 n!1 n + 2
If we consider the second expression first, then notice that as n grows, then both the
numerator and denominator become infinitely large. We say that this expression is of
the type [1/1] and write
2n h1i
lim = .
n!1 n + 2 1
Here, we use square brackets around the "1/1" to remind us that 1 is not an actual
number (if it was, we could write 1/1 = 1 and be done – but notice that in the example,
this quotient should be equal to 2).
In the first expression above, the main "struggle" is between the term n and the
quotient n2 /(n + 1), since, as n grows, both become "infinitely" large. We express this
by writing
⇣ n2 ⌘
lim n = [1 1].
n!1 n+2
Again, we use square brackets to remind us that 1 is not an actual number, and there
is no reason to expect that 1 1 = 0 (this is the mistake we made in the example!).

Remark 5.24 (indeterminate forms) Limit expression to which the rulebook (or any
reasonable extension of it) does not immediately apply are called indeterminate forms.
In addition to what we saw above, these can be of the form [0/0], [0 · 1], [10 ], [11 ] or
even [00 ].

Exercise 5.25 Here is an example of a limit of the indeterminate form [11 ]:


⇣ 1 ⌘n
lim 1 + .
n!1 n
Use Python to figure out what the limit of this expression is (we will revisit it below.)
Exercise 5.26 (discussion) (a) Define what we mean when we say that a sequence
tends to infinity. That is, define what we mean by an ! 1 as n ! 1. (b) Suppose
that an ! 1 and bn ! 1. Use your definition to prove that then also an + bn ! 1.
Remark: You have now proved that [1 + 1] is not an indeterminate form (since it is
always equal to 1).
220 CHAPTER 5. LIMITS FOR SEQUENCES

Rule of thumb 1: Rewrite your expression as a quotient

In practice, most limits we consider are indeterminate


forms of some type. So how to deal with them? Well,
if you look back at Example 5.20, notice how we rewrote
an expression on the form [1 1] so that it become on
the form [1/1]. As a rule of thumb, rewriting your ex-
pression on the form of a quotient usually makes it easier
to solve.
Let us revisit Example 5.20 to illustrate how this rule Fig. 12. Quotients good!
of thumb helps us out.

Example 5.27 We now break the computation of the limit from Example 5.20 into
steps that can be justified. First, we notice that
⇣ n2 ⌘
lim n = [1 1].
n!1 n+2
Following the rule of thumb, we begin by rewriting this expression on the form [1/1]:
⇣ n2 ⌘ n(n + 2) n2
lim n = lim
n!1 n+2 n!1 n+2
2n h1i
= lim = .
n!1 n + 2 1
So, what next? Notice that by rules (vi) and (vii) in the rulebook, we know the limits
of constants and the expression 1/n. To get these into play, let us multiply by one:
2n 2n (1/n) 2
lim = lim · = lim 2 . (5.2)
n!1 n+2 n!1 n + 2 (1/n) n!1 1+ n

Finally, we have arrived at an expression that we can apply our rulebook to:

2 lim 2
n!1
lim 2 =
n!1 1 +
n lim (1 + n2 )
n!1
(5.3)
2 2
= = = 2.
1 + lim 2 1+0
n!1 n

Notice the following, rather peculiar, detail in the above example. It is only after
making the very last step of the computation (5.3), that we know that the preceding
steps in this computation are ok. Only then do we know that all limits involved exist.
5.3. HOW TO USE THE RULEBOOK IN PRACTICE 221

Rule of thumb 2: In a quotient, let the strongest terms fight eachother


In the above example, we multiplied by one to get
into a situation where we could use the rules from
the rulebook. Another way of thinking about this,
which is more flexible, and therefore more useful, is
in terms of factoring out the strongest terms. That
is, figure out who is the strongest in the numera-
tor and denominator, respectively, and factor these
terms out. Chances are that this will leave you with
a computable expression.
Fig. 13. Think of this as choosing
Here is an example of how this works. champions who are to fight.

Example 5.28 Let us compute the limit


n2 + 1
lim .
n!1 3n3 + 2n + 1

This limit is already on the form [1/1]. Here, we see that the strongest term in
the numerator is n2 , while the strongest term in the denominator is n3 (we ignore the
constant). By factoring out these terms, we arrive at the following computation:

n2 + 1 n2 1 + n12 1 1 + n12 1+0


lim = lim 3 · = lim · =0· = 0.
n!1 3n3 + 2n + 1 n!1 n 3 + n22 + 1
n3
n!1 n 3 + 22 +
n
1
n3
3+0+0

In the last step, we merely indicate how to use the rulebook to finish the computation.

Notice that if we flipped things around, we would get the computation


3n3 + 2n + 1 3 + n22 + 1
n3
lim = lim n · = [1 · 3].
n!1 n2 + 1 n!1 1 + n12
The problem is that the rulebook does not actually tell us how to deal with this situation.
However, it ought to be intuitively clear that it is safe to say that the above expression
is not really indeterminate since [3 · 1] = 1. The following exercise indicates several
formulas that we can add to our rulebook:

Exercise 5.29 Combine the two first rules of thumb to quickly verify all the limits
in exercise 5.16.
Exercise 5.30 Interpret the following formulas in a suitable way, and prove them:
(a) [C + 1] = 1 (b) [C · 1] = 1
Remark: This is a follow-up to exercise 5.26. Moreover, in (b), you need to add some
conditions on what C is allowed to be.
222 CHAPTER 5. LIMITS FOR SEQUENCES

Rule of thumb 3: The limit loves smooth functions


This rule of thumb is just our way of trying to
make sense of computational rule (v), which we
do not really have the background to understand
yet. Specifically, the rule says that the formula
⇣ ⌘
lim f (an ) = f lim an
n!1 n!1

holds if the function f is continuous. Being con-


tinuous essentially means that the graph of the
function does not have any "abrupt" jumps on Fig. 14. Hey! Is this allowed? As we
its domain. In particular, the graphs of all func- will see in Chapter 6, this rule is even
tions we saw in Chapter 2 are nice and smooth on more forgiving than what we formu-
their domains and so satisfy this rule of thumb. late here.

Example 5.31 Since the graph of the square root function is nice and smooth, we can
use the rule of thumbs as follows:
r⇣ r
n2 ⌘ ⇣ n2 ⌘ p
lim n = lim n = 2,
n!1 n+2 n!1 n+2
where, in the last step, we used the result of Example 5.27.

Exercise 5.32 Compute the exact value of both of the following limits:
⇣ 1⌘
(a) lim cos sin (b) lim cos(arctan n)
n!1 n n!1

Exercise 5.33 Use the trick a2 b2 = (a + b)(a b) to compute the following limits.

1 ⇣p ⌘
(a) lim p p (b) lim n2 + n n
n!1 n2 + n n2 + 2 n!1

Exercise 5.34 Does the following computation work? Compare with exercise 5.25.
⇣ 1 ⌘n ⇣ 1 ⌘
lim 1 + = lim exp n · ln 1 +
n!1 n n!1 n
⇣ 1 ⌘ ⇣ ⌘ ⇣ ⌘
= exp lim n · ln 1 + = exp lim n · ln 1 + 0 = exp lim 0 = 1.
n!1 n n!1 | {z } n!1
=0

p p
Exercise 5.35 (Challenge) Use Definition 5.10 to prove an ! A =) an ! A.
5.3. HOW TO USE THE RULEBOOK IN PRACTICE 223

Rule of thumb 4: When all else fails, give your limit a good squeeze
This rule of thumb is our way of emphasising that
rule (iv), called the Squeeze theorem, turns out
to be extremely useful. Another indication of its
usefulness is that it has many nicknames. For
instance, it has been called the "Sandwich the-
orem", the "Policeman theorem" and the "KGB
theorem" (apparently the last one was popular
among soviet mathematicians). In Sweden, it is
usually called "Instängningssatsen".
The squeeze theorem is a rather important
rule, since it allows us to compute limits that
we, for some reason, find difficult by replacing Fig. 15. Or... call social services?
them with a pair of simpler limits.
Here is an example where the Squeeze theorem helps us out.

Example 5.36 We want to compute the limit


sin n
lim .
n!1 n

Since sin(n) never settles as n increases, this limit is not like any of the indeterminate
forms we have seen before.
To understand what is going on, we try to
use the Squeeze theorem. Indeed, taking the in-
equality
1  sin n  1,
and dividing it by n yields
1 sin n 1
  .
n n n
Since both expressions on the "outside" tend to
0 as n ! 1, it follows, by the Squeeze theorem,
that the expression in the middle is "escorted"
Fig. 16. Our original sequence is in
to the same limit (see figure to the right). That
blue. The red dots represent the up-
is, by rule rule (iv), we conclude that
per and lower bounds.
sin n
lim = 0.
n!1 n
224 CHAPTER 5. LIMITS FOR SEQUENCES

In Chapter 3, we mentioned that computers are not always to be trusted. For in-
stance, below, we use Python to help us visually determine the limit. Let us now consider
the limit
lim 2n log(1 + 2 n ). (5.4)
n!1

Here is how Python visualises the behaviour of the sequence in question:

Fig. 17. To the left, we see the first 20 terms, and visually, it appears that the limit
is 1. However, when we visualise the first 80 terms on the right, we see a sudden
drop to 0.
In the following exercises you are to study the limit (5.4), above.

Exercise 5.37 The sudden drop to 0 in Figure 17 is due to a round-off error. Explain
why it seems to happen almost exactly at n = 53. (Hint: Recall exercise 3.36)
Exercise 5.38 In this exercise, we apply the double-inequality

x 1
 log x  x 1. 8x > 0.
x
(recall Proposition ??.52) to better understand the limit (5.4).

(a) Plot both curves y = log x and y = x 1 in the same coordinate system. Does
the inequality seem reasonable?
(b) Apply the above double-inequality for the logarithm to the expression 2n log(1 +
2 n ) in order to squeeze it between two more friendly sequences. What does this
say about the limit?
(c) Use Python to visualise the three sequences from (c) in the same plot (as in
Figure 16). Does this confirm your conclusion?

Exercise 5.39 Use the above inequalities for the logarithm to determine:
⇣ 1 ⌘n
lim 1+ .
n!1 n
Hint: The connection to the logarithm is given in exercise 5.34.
5.3. HOW TO USE THE RULEBOOK IN PRACTICE 225

A special notation for the squeeze theorem: The Big-oh


As another indication of how useful the squeeze theorem is, we briefly mention the
following notation designed to be used by this theorem:

Definition 5.40 (Big-oh notation) Let (an )1 n=1 , (bn )n=1 be sequences. Then we say
1

that an = O (bn ) if there exists a constant C > 0 so that

|an |  C|bn |, 8n 2 N.

Note that we can read the above definition as saying that the sequence an is dom-
inated by the sequence bn (notice that we do not really care about the value of the
constant C, since we are usually only interested in knowing if an is forced down to
zero.).
We will only really care about the Big-oh notation in the last chapter of these lecture
notes (and then we will really care about it). So for the moment, we only give the
following exercise as a small taste of things to come:

Exercise 5.41 Suppose that you have a sequence (an )1n=1 where an = O (1/n).
(a) Explain why this means that an is squeezed between two inequalities.
(b) What is the only reasonable value of the limit
lim an ?
n!1

Exercise 5.42 Suppose that an = O n2 . For which values of ↵ > 0 can we deter-
mine the limit
an
lim ↵ ?
n!1 n

Exercise 5.43 Is it possible for one and the same sequence an to satisfy both, say,
an = O (n) and an = O n2 at the same time? If so, give an example where this
happens, if not, prove that it leads to an absurdity.
Exercise 5.44 (a) Explain why
n
X n2
k= + O (n) .
2
k=1

Remark: Here, you are allowed to use the result of Example 1.77.
(b) Use the Big-oh notation to compute the following limits in as few steps as possible.
n n n
1 X 1 X 1X
(i) lim k (ii) lim k (iii) lim k
n!1 n3 n!1 n2 n!1 n
k=1 k=1 k=1
226 CHAPTER 5. LIMITS FOR SEQUENCES

A word of advice on how to choose your champion: the table of growth


Rule of Thumb 2 is all about identifying the strongest term in the denominator and
numerator in order to understand the indeterminate forms [1/1] and [0/0]. This is
important since most indeterminate forms we meet may be rewritten on this form.

Example 5.45 To rewrite [1 · 0], we essentially use that [1/0+ ] = [1]. For example:
⇣ ⌘
⇣ ⌘ 1 h0i
1 log 1 + n
lim n · log 1 + = [1 · 0] = lim 1 = .
n!1 n n!1
n
0

Notice that while you are not allowed to write 1/0 = 1 (division by zero is not
defined!), it is perfectly fine to write [1/0+ ] = 1. Indeed, by this we mean that if we
divide 1 by a sequence that goes to zero (from the positive side) this gives us something
that goes to (positive) infinity.
The following notation helps us identify the strongest term more efficiently.

Definition 5.46 We write an ⌧ bn as n ! 1 to indicate that limn!1 an /bn = 0.

The relative strengths of the most common sequences may now be stated as follows.

Proposition 5.47 (table of growth) For any 0 < ↵ < and 1 < a < b, the following
holds:
log n ⌧ n↵ ⌧ n ⌧ an ⌧ bn ⌧ n! ⌧ nn

Fig. 18. Here you can compare the sequences from the table of growth. Note that
n2 is represented in both plots as the red dots.

Exercise 5.48 Use the table of growth, in combination with the rules of thumbs
above, to compute the following limit.
en + 2 ln n + 3n10 arctan(n) ln(n10 ) + 4 ln(n100 )
(a) lim (b) lim .
n!1 5en + 7 ln n + 11n100 n!1 cos(1/n2 ) ln(n100 ) + 5 ln(n)
5.3. HOW TO USE THE RULEBOOK IN PRACTICE 227

We now justify the table of growth in the following example and exercises.

Example 5.49 (Exponential functions versus factorials) We use the squeeze


theorem to show that 3n ⌧ n! as n ! 1. That is, we show that
3n
lim = 0.
n!1 n!

To apply the Squeeze theorem to the sequence bn = 3n /n!, we first notice that the
expression is positive. This means that we always have the lower bound 0  3n /n! (and
that we can work without absolute values). All that remains is to find a suitable upper
bound cn . To better see what is going on, we make the following computation:
3n 3 3 3 3 3 3 3 3 3 27
= · · · ··· ·  · ·1· = .
n! 1 2 |3 4 {z n 1} n 1 2 n 2n
1

We conclude by applying the Squeeze theorem with an = 0, bn = 3n /n! and cn = 27/2n.

Exercise 5.50 Show that the following is true as n ! 1.

(a) Given any fixed, but unknown, b > 0, show that bn ⌧ n!.
(b) Show that n! ⌧ nn .
(c) Use (a) and (b) to deduce we also have bn ⌧ nn .

Hint: For (a) and (b), modify the above example.

Exercise 5.51 In this exercise, we prove that for all ↵ > 0, we have

log n ⌧ n↵ as n ! 1. (5.5)

(a) Combine the squeeze theorem with a suitable inequality for the logarithm from
exercise 5.38 to convince yourself that (5.5) holds for ↵ = 2. For what other
values of ↵ does the argument work?
(b) Combine the inequality you used from exercise 5.38 with the identity log(n1/2 ) =
(log n)/2 to show that (5.5) also holds for ↵ = 1.
(c) Modify your argument from (b) to show that (5.5) holds for all 0 < ↵ < 1.

Exercise 5.52 Use the limit from the previous exercise to show that for all >0
and a > 1, we have n ⌧ an as n ! 1.
Hint: You need to rewrite the limit expression so that the logarithm appears.
228 CHAPTER 5. LIMITS FOR SEQUENCES

5.4 How to prove the rules in the rulebook


In this section we prove all the remaining rules in the rulebook, except rule (v). For the
latter rule, we need to wait for the next chapter where we discuss limits for functions.
Note that every proof ends with an exercise that you need to complete. This is to force
you to try to actively think about the proofs and discuss them with others!

First things first: a proof that the limit is "well-defined"


Recall that Grandi was able to show that 1 = 0 by using a "bad" definition for the sum
of an infinite series. What if our definition of the limit has a similar weakness? Well, it
does not. The following result says that our definition of the limit is "well-defined":

Proposition 5.53 If the sequence (an )1


n=1 has both the limits L and M , then L = M .

Proof. To prove this by contradiction, we suppose that we have two numbers L 6= M


such that both
an ! L and an ! M
at the same time. According to Definition 5.10, this means that for every ✏1 > 0 and
✏2 > 0, there exist numbers N1 and N2 so that
n > N1 =) |an L| < ✏1 , and (5.6)

n > N2 =) |an M | < ✏2 . (5.7)


That is, we can get the distance between an and L, and an and M , to be as small as we
want, provided that we make n large enough. Keep in mind, that the distance between
L and M is fixed.
Here is an illustration of this (absurd) situation:

Fig. 19. We can get the distances labeled I and II as small as we want, but the
distance III is of fixed length. This makes no sense at all! In the exercise below,
you are asked to verify this.

Exercise 5.54 Based on the intuition given by the above figure, try to deduce an
absurdity by choosing ✏1 and ✏2 sufficiently small.
Hint: Since you are comparing distances between three points, the triangle inequality
|x + y|  |x| + |y| will be of critical importance.
5.4. HOW TO PROVE THE RULES IN THE RULEBOOK 229

Proof of Proposition 5.21(i): The addition rule


We consider the proof of the summation rule for the limit. Notice that this proof is
essentially a careful solution to exercise 5.8 carefully.
We begin by recalling the rule we seek to prove:

Proposition 5.21(i) (slightly reformulated)


lim an = L and lim bn = M =) lim (an + bn ) = L + M.
n!1 n!1 n!1

Exercise 5.55 The biggest challenge here may actually be to understand why there
is something to prove here. To shed a little light on this, consider the sequences
an = 1/n + ( 1)n and bn = 1/n ( 1)n .
(a) Determine whether or not the following limits exist:
(i) lim an (ii) lim bn (iii) lim (an + bn ).
n!1 n!1 n!1

(b) What does this example tell you about computational rule (i) for limits?

Proof of Proposition 5.21(i) To prove that an + bn ! L + M , we begin assuming we are


given an ✏ > 0 that is fixed but unknown (this philosophy was explained in, for instance,
Example 5.3). According to the definition of the limit, we must respond to this epsilon
by showing that there exists some number N so that the following implication is true:

n > N =) |(an + bn ) (L + M )| < ✏. (5.8)


To do this, we need to use the hypothesis of the computational rule we are trying to
prove. According to the definition of the limit, the statements an ! L and bn ! M
mean the following: for all ✏1 , ✏2 > 0, there exist numbers N1 , N2 > 0, so that

n > N1 =) |an L| < ✏1


(5.9)
n > N2 =) |bn M | < ✏2 .

(Here, we use subscripts to keep these epsilons apart from each other and from the epsilon
we are given above.) Having lined up all these facts, we are almost done. We ask you to
end the proof in the following exercise.

Exercise 5.56 (a) Use the triangle inequality to show that |(an + bn ) (L + M )| 
|an L| + |bn M |.
(b) We know that (5.9) holds for all choices of ✏1 , ✏2 . For which choice does this, in
combination with (a), imply (5.8)? (Now is the time to use the given ✏!)
(c) What is a suitable choice for N ? (Here it may be of help to recall exercise 5.8.)
230 CHAPTER 5. LIMITS FOR SEQUENCES

Proof of Proposition 5.21(iv): The squeeze theorem


We now turn to the proof of the Squeeze theorem.

Proposition 5.21(v) If an  bn  cn , then the following holds:

lim an = L and lim cn = L =) lim bn = L.


n!1 n!1 n!1

(Note that we do not need to assume that the limit of the sequence bn exists.)

Proof of the Squeeze theorem As usual, we begin by assuming that we are given a number
✏ > 0 which is fixed but unknown. Our goal is to respond to this epsilon by showing
that there exists some number N so that the following implication is true:

n > N =) |bn L| < ✏. (5.10)

According to the hypothesis, what we know is that for all ✏1 , ✏2 > 0, there exist numbers
N1 , N2 , respectively, such that

n > N1 =) |an L| < ✏1 ,


(5.11)
n > N2 =) |cn L| < ✏2 .

Moreover, by the hypothesis, we have the following connection between the three se-
quences an , bn , cn . Namely, for all n it holds that

a n  bn  c n . (5.12)

To take advantage of this connection, we use the result of exercise 2.48 to “open up” the
inequalities in (5.11) and rewrite them as follows:

|an L| < ✏1 () L ✏ 1 < an < L + ✏ 1


(5.13)
|cn L| < ✏2 () L ✏ 2 < cn < L + ✏ 2 .

As in the previous proof, it is now the case that having lined up all of these facts, we
are almost done. Again, we ask you to finish the proof in the following exercise.

Exercise 5.57 (a) Combine (5.12) and (5.13) to obtain a double inequality for bn
L.
(b) We know that (5.11) holds for all choices of ✏1 , ✏2 . For which choice does this, in
combination with (a), imply (5.10)?
(c) What is a suitable choice for N ?
5.4. HOW TO PROVE THE RULES IN THE RULEBOOK 231

Proof of Proposition 5.21(ii),(iii): The multiplication and quotient rules


We start out by considering the multiplication rule in a slightly reformulated form:

Proposition 5.21(ii) (slightly reformulated)


lim an = L and lim bn = M =) lim (an bn ) = LM.
n!1 n!1 n!1

The proof begins exactly as in that of the two previous ones:


Exercise 5.58 As in the previous two proofs, write out what the hypothesis and
conclusion of Proposition 5.21(ii) means in terms of the definition of the limit.
Next, we need to somehow use the information that an ! L and bn ! M to conclude
that an bn ! LM . That is, we need to relate the distances |an L| and |bn M | to the
distance |an bn LM |. (This step is analog to part (a) of exercises 5.56 and 5.57.)
We do this by using the trick of adding by 0 followed by the triangle inequality:
|an bn LM | = |an bn an M + an M LM |

= |an bn an M + an M LM |
(5.14)
= |an bn M + M an L |

 |an ||bn M | + |M ||an L|.


This seems rather promising! Indeed, it would seem that we could now end the proof by
choosing ✏1 = ✏/2|M | and ✏2 = ✏/2|an |. However, life is not this good. By the definition
of the limit, the epsilons have to be constants, and in particular, are not allowed to
depend on n. Therefore our choice of ✏2 is not valid. We invite you to fix this in the
following exercise. (We also have a problem if M = 0. Do you see how to fix this?)
Exercise 5.59 (a) Show that there exists an N3 such that for all n > N3 then |an | 
L + 1. (Hint: Try to understand this by making an appropriate illustration.)
(b) Combine your inequality in (a) with the computation (5.14). Use the resulting
inequality to choose suitable ✏1 and ✏2 so that you end up with |an bn LM | < ✏.
(c) What is a suitable choice for N ?
Exercise 5.60 (Challenging) After some preparation, the quotient rule follows from
the multiplication rule. In this exercise, we ask you to go through the details.
(a) Prove the reverse triangle inequality: |a + b| |a| |b|.
(b) Prove that bn ! M implies that |bn | |M |/2 for sufficiently large n. (This is
analog to part (a) of the previous exercise.)
(c) Prove that bn ! M implies 1/bn ! 1/M . (You need part (b) here.)
(d) Use the multiplication rule for the limit to conclude.
232 CHAPTER 5. LIMITS FOR SEQUENCES

Proof of Proposition 4.44: the Balloon lemma

To prove part (i), suppose that cn is growing and


bounded from above. That the sequence is grow-
ing means exactly that cn  cn+1 for all n 2 N,
and that the sequence is bounded from above
means exactly that there exists some number M
so that cn  M for all n 2 N (see Figure 20).
Our first job is to find something for the se-
quence to converge to. As should be clear from
the figure, there is no reason to expect the se-
quence to converge to M . However, from the
supremum axiom, we know that since the set

C = {c1 , c2 , c3 , . . .}

is bounded above, it also has a smallest upper


bound sup M . We denote it by L. The idea is
now to try to prove that cn ! L.
As in the above proofs, we start out by letting
✏ > 0 be a fixed but unknown number. Opening
up the absolute values, we can state that our goal
is to find a number N so that Fig. 20. Top: The sequence cn with
its upper bound M , and supremum
n > N =) L ✏ < cn < L + ✏. L. Botom: In green, the line y =
L ✏.
By the bottom part of Figure 20, it seems clear that there should exists some index
n0 so that cn0 > L ✏. This is great, since the existence of this number is the crucial
ingredient which allows us to wrap us this proof (as we ask you to do in the following
exercise).

Exercise 5.61 (a) Explain why there exists an index n0 so that cn0 > L ✏.
(b) Explain why, for all sufficiently large n, we have cn > L ✏.
(c) Explain why, for all n, we have cn < L + ✏.
(d) Combine the above steps to explain a suitable choice for N which makes the
implication n > N =) |cn L| < ✏ true?

Exercise 5.62 Adapt the above argument into a proof of part (ii) of the proposition.
5.5. CONVERGENCE TESTS FOR POSITIVE SERIES 233

5.5 Three convergence tests for positive series using limits


We end this chapter with a section aimed at motivating, formulating and finally prov-
ing three additional convergence tests for series that, in some sense, are slightly more
advanced than the divergence and comparison tests that we saw in Chapter 4.

The limit comparison test


To motivate the test, we consider the following example.

Example 5.63 (Part 1 of 2) Let us check whether the following series converges or
diverges: 1 p 2
X k +k+1
.
k2 + 1
k=1
As we noted in Chapter 4, what is important is the behaviour of these terms for large
values of k. Intuitively, we see that
p p
k 2 + k + 1 large k k 2 + 0 + 0 1
2
⇡ 2
= . (5.15)
k +1 k +0 k
This means that for large k, the series behaves like the harmonic series (which is di-
vergent). For this reason, we expect this series to diverge as well. However, to justify
this using the usual comparison test (Proposition 4.50), we would now need to establish
an inequality relating the terms of these series with each other. The point of the limit
comparison test is that we can avoid the inequalities. We now put this example on hold
to discuss how to do this.

Exercise 5.64 Complete the above example using the usual comparison test (Propo-
sition 4.50).

Let us now take a moment to consider what limit could be relevant for the above
example. The intuitive computation (5.15) is essentially a naive version of the following
computation (which is justified, step by step, by the rulebook for limits):
⇣ p k 2 + k + 1⌘
p
k 2+1 k( k 2 + k + 1) h 1 i
lim ⇣ 1⌘ = lim =
k!1 k!1 k2 + 1 1
k (5.16)
q
1+ 1
+ 1 p
k2 k k2 1+0+0
= lim 1 = = 1.
k!1 k 2 1+ 1+0
|{z} k2
=1
234 CHAPTER 5. LIMITS FOR SEQUENCES

In Figure 21, we visualise the above computa-


tion. We see that the terms of the original series
(blue) and the terms of the the harmonic series
(red) are quite close to eachother. In green, we
visualise the quotient between these terms. As
we verified in the computation (5.16), these quo-
tients tend to the value 1.
In particular, this means that if we take
k large enough, then the quotient between the
terms of the original series and the harmonic se-
ries (the green dots) will lie as close to 1 as we
want. In particular, for ✏ = 0.1, there exists an Fig. 21.
N so that p
⇣ k 2 + k + 1⌘

k > N =) 0.9  k2 + 1  1.1


⇣ 1⌘
k
Multiplying up the factor (1/k), this becomes
p
0.9 k 2 + k + 1 1.1
k > N =)   .
k k2 + 1 k
But this gives us exactly what we need to apply the usual comparison test (Prop. 4.50).
In this case, we want to show divergence, so by the left-most of these two inequalities,
we conclude that our original series diverges. (Recall that a series converges if and only
if its tail converges. This also means that a series diverges if and only if the same is true
for its tail.)
We have now, more or less, indicated how to prove the following result.

Proposition 5.65 (Limit comparison theorem for infinite series)


Suppose that ak , bk are two positive sequences, and that there exists L 0 so that
ak
lim = L.
k!1 bk
If 0 < L < 1, then
1
X 1
X
ak converges () bk converges.
k=1 k=1

For each of the cases L = 0 and L = 1, only one implication holds (do you see which?).

We now return to Example 5.63. Notice how the intuitive and formal steps are now
almost identical.
5.5. CONVERGENCE TESTS FOR POSITIVE SERIES 235

Example 5.63 (Part 2 of 2) We study whether the following series converges or not:

X1 p 2
k +k+1
.
k2 + 1
k=1

Intuitive step: Do as in Example 5.63, and observe


P that for large k, the terms in
this series behave like those of the harmonic series 1
k=1 1/k. This means we expect
divergence.
Formal step: We repeat, word by word, the computation (5.16) above. This shows
that the limit comparison test with LP= 1 applies. We conclude that our series is
divergent since the same holds true for 1
k=1 1/k.

Exercise 5.66 Solve exercise 4.60 using the limit comparison theorem.

Exercise 5.67 Solve exercise 4.61 using the limit comparison theorem.

Exercise 5.68 Use the limit comparison test to determine if the following series con-
verge or diverge.
1
X X1
n + log n 3n + n 4
(a) (b) p .
n2 4 n + n3
n=1 n=1

Hint: It may be a good idea to use the table of growth here.

Exercise 5.69 Turn the discussion leading to Proposition 5.65 into a proper proof
for the case that 0 < L < 1.
Remark: The proof is essentially given in the text immediately before the result. Note
that you will have to bring out the definition of the limit at some point.

Exercise 5.70 Explain intuitively what should happen when L = 0 and L = 1 in


Proposition 5.65, respectively. Explain how to modify the proof in each of these cases.

Exercise 5.71 (Challenge) Determine for all ↵ 2 R whether the series


1
X 1
Pn ↵
( k=1 k)
n=1

converges or diverges.
236 CHAPTER 5. LIMITS FOR SEQUENCES

The root test

We explain the basic idea of the root test in the following example.

P1
Example 5.72 Suppose that k=1 ak is a positive series where, for all k 1, we have
the estimate
1/k 1
ak  .
2
But this is equivalent to
1
ak  .
2k
P
Since 1 k=1 1/2 is P
k a convergent geometric series, it follows by the standard comparison
test that the series 1k=1 ak is convergent.

The general formulation of the root test is as follows.

Proposition 5.73 (Root test)


Suppose that ak is a sequence of positive numbers, and that
p
lim k
ak = Q.
k!1

Then the following hold:


P1
(i) Q < 1 =) k=1 ak converges
P1
(ii) Q > 1 =) k=1 ak diverges

(iii) Q = 1 =) no information

Proof. We consider the case when Q < 1, and


need to show that if ak 1/k ! Q, then the series
is absolutely convergent. The idea of the proof
is essentially to mimic Example 5.72, with Q in
place of 1/2.
However, it turns that out we cannot use Q
directly. The problem is that we could have the
situation shown in the figure to the right. That
is, even if ak 1/k converges to Q, this does not
mean that ak 1/k  Q has to hold for any value Fig. 22. Uh oh...
of k.
5.5. CONVERGENCE TESTS FOR POSITIVE SERIES 237

Fortunately, it is not too hard to get


around this. If you choose your favourite
number q so that Q < q < 1, we get the
situation shown to the right. Indeed, since
ak 1/k converges to Q, it follows that for k
large enough, then these terms will be as
close to y = Q as we want. In particular,
sooner or later, the terms will have to fall
below the line y = q. For such k, the in-
Fig. 23. Yay!
equality ak 1/k  q will hold.
So, suppose that K is a number so that for all k K we have ak 1/k  P q. But this is
the same as having ak  q k for all k K. Since 0  q < 1, it follows that 1 k=K q is a
k

convergent geometric series. By the comparison test, and the fact that the it is enough
P to
show that the tail of a series converges (recall Proposition 4.57), it follows that 1 a
k=1 k
is absolutely convergent.
We leave the case when Q > 1 to the reader (note that the divergence test could be
useful).

Example 5.74 We use the root test to investigate whether the following series converges
or not:
X1
k2
.
2k
k=1

To use the root test, we do the following computation:


r
k k
2 k 2/k e2 ln k/k e2 limk!1 ln k/k e0 1
Q = lim k
= lim = = = = .
k!1 2 k!1 2 2 2 2 2
We conclude that the series converges since Q < 1.

Exercise 5.75 Use the root test to check the convergence of


1
X 1 ✓
X ◆k 2 1
X
k 1 1
(a) k (b) 1 (c)
k k k
ln
k=1 k=1 k=2

Exercise 5.76 In this exercise, we complete the proof of the root test for convergence.

(a) Fill in the details for the case Q > 1.


(b) Show that there exist both divergent and convergent series for which Q = 1.
Hint: Consider ↵-series.
238 CHAPTER 5. LIMITS FOR SEQUENCES

The quotient test


The quotient test is sort of the twin sibling of the root test. As with the root test, we
start out by considering an example to convey the general idea of what is going on.

P1
Example 5.77 Suppose that k=1 ak is a positive series so that for all k 1 we have

ak+1 1
 .
ak 2
But this is the same as
ak
ak+1 .
2
Unlike for the root test, this does not immediately lead to a useful comparison. To see
how to proceed, let us see what this means for the first few terms:
a1
a2 
2
a2 a1
a3   2
2 2
a3 a2 a1
a4   2  3
2 2 2
Aha! In other words, we have
a1
ak  .
2k
P
SinceP 1
k=1 a1 /2 is a convergent geometric series, the standard comparison test implies
k
1
that k=1 ak is convergent.

The general formulation of the quotient test is as follows.

Proposition 5.78 (Quotient test for absolute convergence)


Suppose that ak is a sequence of positive numbers, and that
ak+1
lim = Q.
k!1 ak

Then the following hold:


P
(i) Q < 1 =) ak converges
P
(ii) Q > 1 =) ak diverges

(iii) Q = 1 =) no information
5.5. CONVERGENCE TESTS FOR POSITIVE SERIES 239

Proof. The proof follows by modifying the example in the same way as for the root test.
This time, let us go through the details in the case when Q > 1.
As in the case when Q < 1, the limit ak+1 /ak ! Q does not imply any inequality
between the terms ak+1 /ak and Q. However, since Q > 1, it follows that eventually we
will have ak+1 /ak 1. But this is the same as having ak+1 ak for k large enough. But
this means that for k large enough, the sequence ak will be increasing. In particular, this
implies that ak 6 ! 0. Hence, by the divergence test, the original series diverges.

Example 5.79 We now use the quotient test to investigate the series
1
X k2
2k
k=1

To this end, we make the following computation:


2
( (k+1)
2k+1
) (k + 1)2 2k
Q = lim 2 = lim · k+1
k!1 ( 2kk ) k!1 k2 2
⇣ k + 1 ⌘2 1 1 1
= lim · =1· = .
k!1 k 2 2 2
Since Q < 1, we conclude that the series is convergent.

Exercise 5.80 Use the quotient rule to check whether the following series converge.
X1 X1 X1
2k 1 1
(a) (b) (c)
k! k2 + 1 ek + e k
k=0 k=0 k=1

Exercise 5.81 In exercise 4.66, we briefly studied some infinite series representations
for the functions ln(1 + x) and arctan(x). Here, we consider the following formulas:
x2 x3 x4
ex = 1 + x + + + + ···
2 3! 4!
x3 x5 x7
sin(x) = x + + ···
3! 5! 7!
x2 x4 x6
cos(x) = 1 + + ··· .
2 4! 6!
(a) What does the quotient test say about each of the above series?
(b) Verify numerically that each of the above formulas approximate their functions for
x = 1. For which of these series does the error estimate from the alternating series
test allow you to predict how many terms to include for these approximations to
have an error of at most 1/1000?
240 CHAPTER 5. LIMITS FOR SEQUENCES

5.6 Exam exercises


Exercises on the limit of sequences
Many exam exercises on limits of sequences also involve techniques that require knowl-
edge on Taylor expansions. These exercises are delayed until the relevant chapter.

Exercise 5.82 (Lund, January 2016) Define what we mean by the symbol limn!1 an
and use this definition to show that limn!1 ( 1)n does not exist.
Exercise 5.83 (Lund, August 2015)
(a) Give the definition for
lim ak = 0.
k!1

(b) Use this definition to show that


sin k
lim p = 0.
k!1 k
Exercise 5.84 (Lund, May 2015) Give the definition of limn!1 an = +1 and use
it to show that
n!
lim = 1.
n!1 20n

Exercises on the root and quotient tests for convergence


Exercise 5.85 (Exam 2015-08-17) Determine whether the following series con-
verges. X1
2k
.
k=2
1 + lnk (k)

Exercise 5.86 (Exam 2015-05-27, part of exercise) Determine whether the fol-
low series converges or diverges.
1
X k!
.
k 20 20k
k=1

Exercise 5.87 (Exam 2014-12-18, part of exercise) Determine the convergence


of
1 ⇣
X 1 ⌘k 2
1 .
k
k=1

Exercise 5.88 (Exam 2014-08-18, part of exercise) Determine the convergence


of
X1
k!
.
3k
k=1
5.6. EXAM EXERCISES 241

Exercise 5.89 (Exam 2013-05-29, part of exercise) Determine the convergence


of
X1
n3/2
.
5n ln(n)
n=2
242 CHAPTER 5. LIMITS FOR SEQUENCES

5.7 Answers to selected exercises


5.4 (a) N = 9 and N = 13, respectively. (b) Use a while-loop that runs n = n +1 as
long as 1/2⇤⇤n > epsilon is true and then prints the final value of n. (c) Smaller
✏ leads to larger N . (d) For all ✏ 1/2, you can choose N = 0.

5.5 (a) N = (5/✏ 3)/2, (b) N = 23, N = 234, (e) For all ✏ 1, you can choose
N = 0.

5.6 The key observation is that n! n.

5.8 (a) In the top figure, it seems clear that this is N = 4, in the bottom, it is less
clear, but N = 10 seems like a reasonable guess.

5.9 (a) For example, it works to apply the definition for positive sequences to an .
(b) Apply the definition for positive sequences to |an |.

5.12 The key is to use the formula for partial sums of the Geometric series (see Chapter
4).

5.14 (a) an is not defined for n = 3. This is not a problem as long as N is taken to be
larger than or equal to 3.

5.16 (a) Given ✏, a suitable choice is N = (5/2)(3/✏ 1). (b) Here, the computation
2n 1 2n 2n 2
 2  = <✏
n2 +n 1 n +n 1 n 2 n
works for n > N = max{1, 2/✏} (but other choices for N are also possible).

5.17 Use that


5n 5 5 5 5
= · · ··· .
n! 1 2 3 n
5.19 The limits are (a) 0, (b) 2, (c) 1/3.

5.22 (a) Given ✏, any N works. (b) Given ✏, then N = 1/✏ works.

5.23 The key observation is that for N large enough, n > N =) |an | < ✏/2 and
|bn | < ✏/2. Combine this with the triangle inequality, and you are done.

5.25 The limit is close to 2.71824....

5.26 (a) For instance, 8⌦ > 0, 9N > 0 such that for all n 2 N we have n > N =)
an > ⌦. (b) This is very similar to 5.23.

5.30 Here are the interpretations: (a) an ! 1 =) C + an ! 1. (b) C > 0 and


an ! 1 =) C · an ! 1.

5.32 (a) 1, (b) 0.


5.7. ANSWERS TO SELECTED EXERCISES 243

5.33 (a) 2, (b) 1/2.

5.34 The third step is wrong.

5.35 It may be smart to divide this proof into two cases: (i) A = 0 and (ii) A > 0. In
the second case, the key observation is that
p p x A
x A= p p .
x+ A

5.38 (c) The limit should go to 1. (d) Yes.

5.39 The key is to observe that


⇣ 1 ⌘n 1
1+ = en log(1+ n ) .
n

5.41 (a) |an |  C/n () C/n  an  C/n. (b) 0.

5.42 For ↵ > 2 (then the limit is 0).

5.43 Yes (I will let you figure out an example).

5.44 In part (b), the limits are (i) 0, (ii) 1/2, (iii) 1.

5.48 (a) 1/5, (b) (5⇡ + 400)/105

5.50 In (a), (b) follow Example 5.49. To quickly solve (c), you can observe that
bn bn n!
= · .
nn n! nn

5.51 (a) The inequality log n  n 1 does the trick for all ↵ > 1. (c) The key is to
use, say, log(n) = 2↵
1 1
log(n2↵ )  2↵ (n2↵ 1).

5.52 Here, the key is to observe that

n e log n log n n log a


= =e .
an en log a

5.54 Choose ✏1 = ✏2 = |L M |/2 (or anything smaller than this).

5.55 In part (a) the answers are (i) no, (ii) no, (iii) yes.

5.56 (b) ✏1 = ✏2 = ✏/2.

5.57 (b) ✏1 = ✏2 = ✏.

5.59 (a) Use the fact that for large enough n then |an L|  1, and then apply triangle
inequality to |an | after adding by 0 in a clever way.
244 CHAPTER 5. LIMITS FOR SEQUENCES

5.60 (a) This follows by realising that |a|  |a + b| + |b| is just the ordinary triangle
inequality in disguise. (b) Use the fact that for large enough n, then |bn M | 
|M |/2. After this, apply triangle inequality to |bn | after adding by 0 in a clever
way.

5.61 (a) Well, L is the least upper bound of M , so...


p p
5.64 Use k 2 + 1 k 2 + 0 and k 2 + k + 1  k 2 + k 2 + k.

5.68 (a) Diverges (compare to harmonic series), (b) diverges (compare to a geometric
series).

5.70 For L = 0, (= is valid, and for L = 1, =) is valid.

5.71 Converges if and only if ↵ > 1/2.

5.75 (a) converges, (b) converges, (c) no conclusion.

5.76 (a) This case is easier than Q < 1. The reason is that from |ak |1/k ! Q > 1, you
can deduce that ak 1 for large enough k, and therefore the series diverges by the
divergence test.
(b) Use the harmonic series and the Basel series as examples.

5.80 (a) converges, (b) no conclusion, (c) converges.


Chapter 6

Limits for functions

Introduction

In this chapter we study limits for functions. For sequences, we considered what hap-
pened to a sequence an as n ! 1. Now, we mainly consider what happens to a function
f (x) as x ! a. Even though these two situations are quite similar, we need to be a bit
more nuanced when it comes to limits of functions.

Remark 6.1 (Selected problems from previous exams based on this chapter)

1. (a) Formulate the epsilon-delta definition of limx!a f (x) = L.


(b) Explain what it means that f (x) is continuous at x = a.
(c) Use this definition to show that

lim (3x3 9x + 1) = 7.
x!2

2. We consider the function


8
< x sin( 1 ) x 6= 0
f (x) = x
:
C x=0

(a) For what value of C is f (x) continuous at x = 0?


(b) Use the epsilon-delta definition of the limit to prove your answer in (a).
(c) What is the derivative of f for x 2 R\{0}?
(d) Is f differentiable at x = 0?

245
246 CHAPTER 6. LIMITS FOR FUNCTIONS

6.1 A first look at the limit for functions


In this section, we begin with a quick tour of some basic uses of the limit in the context
of functions.

First, an informal definition of the limit for functions


We take the limit of f (x) as x approaches a to understand how the function f acts near
a point x = a. Here is an informal definition of what we mean by this.

Remark 6.2 (Informal definition of the limit) If f (x) gets "as close as we want"
to some value L as x approaches a (from both sides), we write

lim f (x) = L or f (x) ! L.


x!a x!a

If this does not happen, we say that the function diverges as x ! a.

One reason to use the limit is to check what happens near a point where a function
is not defined:

Example 6.3 Consider sin x


f (x) = .
x
The natural domain of f is R\{0}. That is, we are allowed to plug in all x except x = 0.
To see what happens as x approaches 0, we consider the following plots made in Python:

Fig. 1. Two visualisations of f (x) where we indicated the computed values of f (x)
with red dots. Notice that x = 0 is not in the domain of f .

From the above figures, we get a good idea what happens as x ! 0. Indeed, the values
of f (x) seem to approach 1. It is therefore reasonable to believe that

sin x
lim = 1.
x!0 x
6.1. A FIRST LOOK AT THE LIMIT FOR FUNCTIONS 247

Let us summarise the insight from the above example in a rule of thumb:

Remark 6.4 (Rule of thumb) To visually de-


termine a limit, follow your finger!

That is, to visually determine the limit of some


function as, say, x ! 0, just let your finger trace
the graph as x approaches 0. If the finger be-
comes confused along the way, then the limit
(most likely) does not exist!
Fig. 2. The finger knows :-D
Notice the following technical point about the
finger.

Remark 6.5 (The finger does not care about the point itself ) When discussing
if limx!a f (x) = L, we are not allowed to use any information about f (x) at the point
x = a itself. This is a feature of the formal definition of the limit which is necessary
since we typically want to investigate what happens as we approach points just outside
the domain of f . For this reason, the finger is blind to any information at x = a itself.

Exercise 6.6 What appears to be the limit for each of the functions in Figure 3 as
x approaches 0?

Fig. 3. What is the limit as x ! 0?


Exercise 6.7 For another way to confuse your finger, use Python to plot f (x) =
(|x + 1| 1)/x for x 2 (0,10 14 ]. What does your finger tell you? Determine the limit
of f (x) as x ! 0 algebraically. Do you get the same result?
248 CHAPTER 6. LIMITS FOR FUNCTIONS

One-sided limits versus two sided limits

We now take a closer look at the plot in the


bottom-right corner of Figure 3 (shown again to
the right).
This function has no limit as x ! 0. Indeed, the
finger gets confused since it arrives at a different
value when coming in from the left or from the
right (the limit exists if and only if our finger
approaches the same value when coming in from
the right and left). Fig. 4.

To better be able to discuss such a situation, we sometimes need to talk about one-
sided limits (the usual limit, considered above, is sometimes called the two-sided limit).

Remark 6.8 (one-sided limits) Suppose that the values of the function f (x) ap-
proaches L as x approaches a from the right or right, then we write, respectively,

lim f (x) or lim f (x).


x!a+ x!a

As we mentioned above, in order for the finger not to become confused, the limit
lim f (x) = L
x!a
can be said to exist if and only the one-sided limits from both sides exist and are equal
to L. We formulate this in the following proposition.

Proposition 6.9 8
> lim f (x) = L
< x!a +
lim f (x) = L ()
x!a >
: lim f (x) = L
x!a

This proposition, which we prove later in this chapter, is rather useful since it some-
times reduces checking an annoying limit (involving, say, absolute values) by two hope-
fully friendlier one-sided limits.

Exercise 6.10 Use Python to investigate the following limits. Do they exist?
1 1 1
(a) lim 2 e 1/x (b) lim 2 e 1/x (c) lim 2 e 1/x
x!0+ x x!0 x x!0 x

Exercise 6.11 Does the following limit exist?


x2 x 2
lim
x!2 |x 2|
6.1. A FIRST LOOK AT THE LIMIT FOR FUNCTIONS 249

Continuity
Above, we remarked that when computing limx!a f (x) our finger does not care about
about what happens at x = a itself. But what if we do care?

Definition 6.12 (Continuity) We say that f is continuous at x = a if


lim f (x) = f (a).
x!a

Moreover, a function is continuous on an interval I if it is continuous at all a 2 I. Finally,


we call a function continuous if it is continuous on all points in its domain.

Fig. 5. Guess which one of these is continuous at x = 0? :-)

Notice that for a function not to be continuous at a point x = a in its domain, then
either limit has to not exist or the limit has to be different than the function value. That
is, to not be continuous at x = a, then your finger must, somehow, be confused by what
happens as you approach x = a. In particular, this happens if the graph "jumps" exactly
at x = a. For this reason, the following remark is tempting to make:

Remark 6.13 (A usual, but slightly inaccurate, description of continuity) A


function is continuous at all points on its domain if its graph has no jumps there.
8
Exercise 6.14 Consider > sin x
< if x 6= 0
f (x) = x
>
:
C if x = 0
Based on Example 6.3, for what value of C is this function continuous at x = 0?
Exercise 6.15 (a) Are any of the graphs shown in Figure 3 continuous at x = 0?
(b) Suppose that you are allowed to change the value of each of these graphs at x = 0
(if needed). Can you make any of them continuous by doing this?
Remark: Hopefully, by doing this exercise, you will recognize why we call Remark 6.13
"slightly inaccurate".
Exercise 6.16 Does the graph of f (x) = 1/x have any jumps? Is this a continuous
function? Is this function continuous at x = 0?
250 CHAPTER 6. LIMITS FOR FUNCTIONS

6.2 The definition of the limit for functions


We now give the proper definition of what we mean by the limit of a function. To
formulate it, we need to introduce a bit of vocabulary on intervals.
First, an interval may contain two types of points: endpoints and inner points.

Fig. 6. The interval, [a,b) has endpoints a and b (but note that only the endpoint a
is contained in the interval). All other points in the interval are inner points.

We say that a function f (x) is defined in a neighbourhood of x = c if it is (at


least) defined on an interval that contains x = c as an inner point. If f (x) is defined
on a neighbourhood of x = c except at the point x = c itself, then we say that f (x) is
defined in a punctured neighbourhood of x = c.

Fig. 7. To the left, (a, b) is a neighbourhood of x = c, and to the right, [a, c) [ (c,b)
is a punctured neighbourhood of x = c.
Note that if a < c < b, we can write

(a,c) [ (c,b) = (a,b)\{c}.

We are now ready to formulate and study the definition of the limit for functions.

Definition 6.17 Suppose that f (x) is defined in a punctured neighbourhood of x = a.


Then f (x) has L as its limit as x tends to a if, for every ✏ > 0, there exists a > 0 so
that (
|x a| <
=) |f (x) L| < ✏.
x 6= a
If this holds, we write f (x) ! L as x ! a, or, which is the same,
lim f (x) = L.
x!a

Over the next few pages, we try this definition out on a bunch of examples in order
to understand what it says.
6.2. THE DEFINITION OF THE LIMIT FOR FUNCTIONS 251

How to use the definition on specific examples


We begin by considering the following example.

Example 6.18 We prove that


lim (2x 1) = 3. (6.1)
x!2

As we did for sequences, we begin by letting ✏ > 0 be some unknown, but fixed, number.

Fig. 8. Here, y = 2x 1 is illustrated along with the epsilon challenge ✏ = 1.

Our goal is now to show that we get (2x 1) closer than ✏ to 3 by moving x sufficiently
close to 2. That is, we want to show that

|x 2| small =) |(2x 1) 3| < ✏.

The first step towards achieving this is to simplify the expression we are considering:

|(2x 1) 3| = |2x 4| = 2|x 2|.

As by a miracle, the expression |x 2| suddenly appears, and we can observe that



|x 2| < =) 2|x 2| < ✏,
2
In particular, this means that the requirement of Definition 6.17 is satisfied by choosing
= ✏/2. We have found our -response to the ✏-challenge, and we are done!

Fig. 9. Here we see appropriate responses to the challenges ✏ = 1 and ✏ = 1/2.


252 CHAPTER 6. LIMITS FOR FUNCTIONS

Exercise 6.19 What is the appropriate delta response to ✏ = 1 and ✏ = 1/2, respec-
tively, according to the computations in the above example? Does this match what
we see in Figure 9?
Exercise 6.20 In this exercise you are to use the definition of the limit to show that
lim (4x 3) = 9.
x!3

(a) How small does |x 3| have to be to beat the challenge ✏ = 1?


(b) How small does |x 3| have to be to beat the challenge ✏ = 1/10?
(b) How small does |x 3| have to be to beat the a general challenge ✏ > 0?

Example 6.21 Let us do a more difficult example. We use the epsilon-delta definition
of the limit to show that
f (x) = x3 2x2 5x + 8
is continuous at x = 1. That is, we need to use the epsilon-delta definition to show that
lim (x3 2x2 5x + 8) = f (1).
x!1

Again, we begin by supposing that we are given fixed but unknown ✏ > 0.

Fig. 10. Here, we illustrate the function f (x) = x3 2x2 5x + 8 with two possible
epsilon challenges. The question is, how do we respond?

The point is now to prove that


|x 1| small =) |(x3 2x2 5x + 8) f (1)| < ✏.
The first step towards achieving this is to simplify the expression we are considering:
|(x3 2x2 5x + 8) f (1)| = |(x3 2x2 5x + 8) 2| = |x3 2x2 5x + 6|.
To have any hope of succeeding, we need the factor x 1 to appear. But this is exactly
what happens since x = 1 is a root of this expression (check this!). Using what we
learned in Chapter 1, we obtain
|x3 2x2 5x + 6| = |x2 x 6||x 1|.
6.2. THE DEFINITION OF THE LIMIT FOR FUNCTIONS 253

This is all good, but how to deal with |x2 x 6|? We now use a trick: observe that
while we do not yet know how small we need to make |x 1|, let us at least agree that we
will make this distance smaller than 1 (because why not?). Opening the absolute value,
we see that
|x 1| < 1 () 1 < x 1 < 1 () 0 < x < 2.
Next, we see that, under the condition 0 < x < 2, the triangle inequality gives us that
|x2 x 6|  x2 + |x| + 6 < 22 + 2 + 6 = 12.
That is, by combining what we have done so far, we get

|x 1| < 1 =) |(x3 2x2 5x + 8) 2| < 12|x 1|.


But now, we are in exactly the same situation as in the previous example. Here, the
final observation is that

|x 1| < =) 12|x 1| < ✏.
12
In conclusion, we have shown that
(
|x 1| < 1
=) |(x3 2x2 5x + 8) 2| < ✏.
|x 1| < ✏/12

But wait! Does this satisfy the definition of the limit? Yes, what this means is that given
a challenge ✏ > 0, then the proper response is to be the smallest of the two numbers 1
and ✏/12, whichever that may be. We express this as choosing = min{1, ✏/12}. Done!

The first three exercises below are meant to help you understand the above example.

Exercise 6.22 Draw the graph of the function (✏) = min{1, ✏/12}.
Exercise 6.23 In Example 6.21, what seems like appropriate delta responses based
on Figure 10 (where epsilon is equal to 4 and 2, respectively)? What delta response is
suggested by the computations in the example? Does it matter that are not the same?
Exercise 6.24 Write a Python program that checks, for the above example, how
close x has to be to the point x = 1 before |(x3 2x2 5x + 8) f (1)| < ✏, where ✏
is some variable specified at the start of the program. Keep in mind that you need to
check what happens both as x ! 1+ and x ! 1 .
Hint: Here, a good start is to start at x = 1 and them jump towards the origin at steps
of, say, 1/100. .
254 CHAPTER 6. LIMITS FOR FUNCTIONS

Exercise 6.25 In this exercise you are to use the definition of the limit to show that

lim (x3 4x2 + 10x 1) = 11.


x!2

In particular, make sure that you answer the following:

(a) How small does |x 2| have to be to beat the challenge ✏ = 1?


(b) How small does |x 2| have to be to beat the challenge ✏ = 1/10?
(c) How small does |x 2| have to be to beat the a general challenge ✏ > 0?

If you are able to solve the following exercise, then you have an excellent understand-
ing of the epsilon-delta type proofs and the techniques involved. Most students will need
more than one attempt to solve it.

Exercise 6.26 (Challenging) In this exercise, we investigate how to prove that


1
f (x) =
x 1
is continuous at x = 2.

(a) According to the definition of continuity, what limit do we need to prove?


(b) Draw the graph of f (x), and visually insert the epsilon challenges ✏ = 1 and
✏ = 1/2. What seem like appropriate delta responses?
(c) Is there any epsilon we can beat by choosing = 2? (You are also supposed to
answer this question by inspecting the graph visually.)
(d) Use the definition of the limit to verify that f (x) is continuous at x = 2.

Remark: Part (c) is included to help you notice an added difficulty when solving (d).
The point of this exercise is to figure out how to successfully deal with this in part (d).
6.3. THE RULEBOOK FOR LIMITS OF FUNCTIONS 255

6.3 The rulebook for limits of functions


We now state the “rulebook” for how to deal with limits for functions. As you see, this
rulebook is more or less identical to that of sequences.

Proposition 6.27 (Rulebook for the limit of functions) Suppose that the limits
limx!a f (x) and limx!a g(x) exist (and are finite). Then the following hold:
⇣ ⌘
(i) lim f (x) ± g(x) = lim f (x) ± lim g(x)
x!a x!a x!a

(ii) lim f (x) · g(x) = lim f (x) · lim g(x)


x!a x!a x!a

If, in addition, limx!a g(x) 6= 0, then

(iii) lim f (x)/g(x) = lim f (x) / lim g(x)


x!a x!a x!a

The following rule is rather famous, and is often called the Squeeze theorem:
8
> f (x)  g(x)  h(x) for all x in a
>
<
(iv) punctured neighbourhood of x = a, =) lim g(x) = L.
>
> x!a
: and lim f (x) = lim h(x) = L
x!a x!a

The following rule is sometimes called the "change of variables" rule. It says that if f g
is defined in a punctured neighbourhood of x = a, then
(
f is continuous, and
(v) =) lim f g(x) = f lim g(x)
lim g(x) exists and is in Df x!a x!a
x!a

Finally, the following concrete limits are useful enough to be included here:

(vi) lim C = C for all constants C 2 R, (vii) lim x = a


x!a x!a

These rules also apply if we replace a by +1 or 1.

We remark that some textbooks also formulate rule (v) as follows:


(v’) lim g(x) = b and lim f (u) exists =) lim f g(x) = lim f (u).
x!a u!b x!a u!b

Exercise 6.28 Show that if f is continuous at u = b, then (v’) implies (v).


Remark: Strictly speaking, we ought to add one additional assumption in (v’), can you
see what that is? This comment also applies to the corresponding rule on page 218.
256 CHAPTER 6. LIMITS FOR FUNCTIONS

Proof of Proposition 6.27(i): The sum rule for limits


To see how to prove a result on limits for functions, we consider the following rule.

Proposition 6.29
⇣ ⌘
lim f (x) = L and lim g(x) = M =) lim f (x) + g(x) = L + M.
x!a x!a x!a

Proof of Proposition 6.29. To prove that f (x) + g(x) ! L + M , we begin assuming,


just as for sequences, that we are given a fixed but unknown ✏ > 0. According to the
definition of the limit, we must respond to this challenge by showing that the following
implication is true:

|x a| small =) |(f (x) + g(x)) (L + M )| < ✏. (6.2)

Note that the statement “|x a| is small” is exactly mean when we write |x a| < .
In particular, the hypothesis of the proposition, i.e., the statements f (x) ! L and
g(x) ! M , mean the following: for all ✏1 , ✏2 > 0, there exist numbers 1 , 2 > 0, so that

|x a| < 1 =) |f (x) L| < ✏1


(6.3)
|x a| < 2 =) |g(x) M | < ✏2 .

As in the corresponding proof for sequences, we are almost done with the proof at this
point. Recall that our goal is to somehow use the information in (6.3) to obtain (6.2).
So, what we do is to connect these expressions by using the triangle inequality as follows:

|(f (x) + g(x)) (L + M )| = |(f (x) L) + (g(x) M )|


 |f (x) L| + |g(x) M |.

Next, observe that if we choose ✏1 = ✏/2 and ✏2 = ✏/2, then we are guaranteed that there
exist numbers 1 , 2 so that (6.3) holds. Combining this with the above, we find that
8
< |x a| < 1
>
|x a| < 2 =) |(f (x) + g(x)) (L + M )|  ✏1 + ✏2 = ✏.
>
:
x 6= a

In other words, the definition of the limit limx!a (f (x) + g(x)) = L + M is satisfied if we
choose = min{ 1 , 2 }.

Exercise 6.30 Modify the proof of the product rule for limits of sequences so that it
applies to limits for functions.

Exercise 6.31 Modify the proof of the squeeze theorem for limits of sequences so
that it applies to limits for functions.
6.3. THE RULEBOOK FOR LIMITS OF FUNCTIONS 257

Proof of Proposition 6.27(v’): Change of variables formula


We begin by essentially reformulating the change of variables formula. However, note
that this formulation is more forgiving than the one in the rulebook.

Proposition 6.32 If f g(x) is defined in a punctured neighbourhood of x = a, then

lim g(x) = b and lim f (u) = L =) lim f (g(x)) = L


x!a u!b x!a

Proof of Proposition 6.32. In a slightly sloppy language, what we know is the following:
8✏1 : |x a| small =) |g(x) b| < ✏1
(6.4)
8✏2 : |u b| small =) |f (u) L| < ✏2
What we need to prove is that given some unknown, but fixed, ✏ > 0, then
|x a| small =) |f (g(x)) L| < ✏.
But this is rather reasonable. Indeed, note by using the connection u = g(x), we ought to
be able to combine the two lines in (6.4) to arrive at the following chain of implications:
u=g(x)
|x a| small =) |u b| small =) |f (u) L| < ✏

In the figure, below, we illustrate the basic objective of the proof. You start out with
an epsilon target around L and are supposed to find a some delta on the x-axis that you
know answers this challenge. To do this, you need to take into account what happens
on the u-axis in the middle.
Exercise 6.33 In this exercise you are
asked to complete the proof.

(a) First identify a suitable choice for ✏2 ,


and write out the condition this gives
on |u b|.
(b) Next, since we have put u = g(x), we
can use the condition on |u b| to make
a suitable choice for ✏1 . This in turn
gives a condition on |x a|. Write out
this condition.
(c) Finally, make a suitable choice for a
number so that the following impli-
cation holds: Fig. 11. The basic idea of the proof.
|x a| < and x 6= a =) |f (g(x)) L| < ✏.
258 CHAPTER 6. LIMITS FOR FUNCTIONS

6.4 How to use the rulebook in practice


The intuitions and rules of thumb we gained when working with limits of sequences also
hold for limits of functions. However, at the start of this chapter we formulated a new
rule of thumb (the finger!). Here, we formulate yet another one (an algebraic finger!).
We also take another look at how to use the change of variables formula and at how to
compute asymptotes.

A final rule of thumb


One difference when working with limits of functions as opposed to limits of sequences is
that we are no longer letting n tend to 1, but rather we are letting x tend to some specific
value. For this reason, the following rule of thumb is usually helpful in determining when
an expression is no longer on an indeterminate form.

Remark 6.34 (A final rule of thumb – an algebraic finger) If you can get away
with replacing x by its limit, then go for it!

We illustrate what we mean by this in the following example.

Example 6.35 We compute the limit


x2 1
lim .
x!1 x 1
Since this limit is of the form [0/0], we must begin by rewriting the expression:
x2 1 h 0 i (x 1)(x + 1)
lim = = lim = lim (x + 1)
x!1 x 1 0 x!1 x 1 x!1

Here, we end up with an expression where it makes complete sense to replace x by 1.


Following our rule of thumb, we conclude that

x2 1
lim = lim (x + 1) = 2.
x!1 x 1 x!1

Exercise 6.36 Use the rulebook to justify the steps in the above example.
Exercise 6.37 Determine the following limits.
✓ ◆
x 2 1 2
(a) lim (b) lim + .
x!2 x2 + x 6 x! 1 x + 1 x2 1
6.4. HOW TO USE THE RULEBOOK IN PRACTICE 259

Using the change of variables rule in practice


We now discuss the change of variables rule. Recall that we already discussed it in the
context of limits for sequences (indeed, recall our Rule of thumb 3). Here, since both
formulations (v) and (v’) are useful, we illustrate how to use both. Keep in mind that
all functions we introduced in Chapter 2 are continuous (as we prove in a later chapter).

Example 6.38 Suppose we want to compute the limit

lim (4x + 1)2 .


x!1

Using the computational rules we have above, one way to deal with this would be as
follows:
lim (4x + 1)2 = lim (16x2 + 8x + 1) = 16 + 8 + 1 = 25.
x!1 x!1

Using rule (v) we can avoid some of the computations:


⇣ ⌘2
lim (4x + 1)2 = lim (4x + 1) = 52 = 25.
x!1 x!1

Basically the same happens if we use rule (v’) with u = g(x) = 4x + 1:

lim (4x + 1)2 = lim u2 = 25.


x!1 u!5

The change of variable formula allows us to reduce more complicated limits to simpler
ones. In particular, the following “standard limits” tend to show up all over the place
(we prove them in later chapters).

Proposition 6.39 (Standard limits) The following limits hold:


sin x ln(1 + x) ln x
(i) lim =1 (ii) lim =1 (iii) lim = 0,
x!0 x x!0 x x!1 x↵

where, in the last limit, we assume that ↵ > 0.

Fig. 12. From the left to right, we have graphs illustrating the standard
limits (i), (ii) and (iii) (with ↵ = 1), respectively.
260 CHAPTER 6. LIMITS FOR FUNCTIONS

Example 6.40 We use the standard limits in combination with a change of variables
to show that for > 0, we have
x
lim = 0.
x!1 ex

Notice that this limit is of the form [1/1]. So, this is just another example of the
exponential function winning essentially every fight he is involved in.
Let us try the change of variables ex = u, which is the same as x = ln u. Our hope
is that this will transform this expression into standard limit (iii), above. We note that
x ! +1 implies that u = ex ! +1. This allows the following computation:
x u=ex (ln u) ⇣ ln u ⌘
lim x = lim = lim
x!1 e u!1 u u!1 u1/
⇣ ln u ⌘
= lim 1/ = 0.
u!1 u

In the last line, we first used the fact that y = x is continuous. This is something we
will prove later in the course. Then, we used standard limit (iii) with ↵ = 1/ .

Exercise 6.41 Justify each step in the above example by referring to the relevant
rule.
Exercise 6.42 Compute the limits

sin(x2 1) ex 1
(a) lim (b) lim
x!1 x 1 x!0 x

Exercise 6.43 Compute 1 cos x


lim
x!0 x
Exercise 6.44 For ↵ > 0, compute the limit

lim x↵ ln x.
x!0+

Exercise 6.45 Compute the following two limits.


⇣ 1 ⌘2x ⇣ 1 ⌘2x
(a) lim 1+ (b) lim 1+
x!1 3x x!0+ 3x
6.4. HOW TO USE THE RULEBOOK IN PRACTICE 261

Computations involving asymptotes


In Chapter 2, we gave "informal" definitions of the various types of asymptotes that we
meet in this course (vertical, horisontal and skew). The reason these definitions were
called "informal" is that we did not really have a definition for what we meant by
lim f (x) = L. (6.5)
x!1

You are to close this gap in the following exercise.

Exercise 6.46 (a) Formulate a suitable definition for what (6.5) ought to mean.
(b) Use the above definition to prove that
2x2
lim = 2.
x!1 x2 + 1

In Chapter 2, we only considered asymptotes of rational functions. Let us now


consider an example where we check the vertical and horisontal asymptotes of a non-
rational function.

Example 6.47 (Vertical asymptotes) Let us determine any vertical and horisontal
asymptotes of the expression
1
f (x) = p .
x2 2x x
To find the vertical asymptotes, we need to identify all points in R where the function
can tend to infinity. First, note that when x approaches points where the denominator
is defined and non-zero, the "additional rule thumb" applies and we get a finite limit.
p
p To figure out where we can have an infinite limit, we investigate x2 2x =
x(x 2). We see that this root is defined for on ( 1,0] [ [2, + 1), and, moreover,
that it is zero when x 2 {0, 2}. We check the following limits:
1 1 h 1 i h 1 i
lim f (x) = = and lim f (x) = = = +1.
x!2+ 0 2 2 x!0 0+ 0 0+

That is, we have a vertical asymptote at x = 0. Here, we used the symbols 0+ and 0
to indicate whether or not the zeroes are approached from the positive or negative sides.
Moreover, we checked one-sided limits since we cannot approach from inside (0,2).
The
p final possibility for vertical asymptotes are at points where the root is defined,
but x2 2x x = 0. That is, at points where x satisfies:
p
x2 x = x =) x2 x = x2 () x = 0.

But this is the point we already detected, so we have found all vertical asymptotes.
262 CHAPTER 6. LIMITS FOR FUNCTIONS

To check for horisontal asymptotes, we need to figure out if expression approaches


some constant as x approaches +1 or 1. In this case, the computations more or less
act as if we were computing with sequences and letting n ! 1.
Let us revisit the above example.

Example 6.48 (Horisontal asymptotes) We check whether the function f (x) in the
previous example has any horisontal asymptotes. First, we check if there is a horisontal
asymptote as x ! +1:
h 1 i p
1 1 x2 x + x
lim p = = lim p ·p
x!1 x2 x x 1 1 x!1 x2 x x x2 x + x
p
x2 x + x
= lim
x!1 x
q
1 p
x 1 x +1 1+0+1
= lim = = 2.
x!1 x 1 1
(Notice how we used practically every rule of thumb here.)
We leave it as an exercise to check for a horisontal asymptote as x ! 1.

Exercise 6.49 (a) Determine whether the function in the previous example has a
horisontal asymptote as x ! 1.
(b) Use some visualisation tool to draw the graph of f (x) to verify your answer in
(a).

Exercise 6.50 (a) Determine any vertical and horisontal asymptotes of


⇣1⌘ 1 ⇣1⌘
(i) f (x) = arctan . (ii) f (x) = sin .
x x2 x

(b) Use some visualisation tool to draw the graphs of the functions in (a) to verify
your answers.

We now turn to the question of how to compute skew asymptotes (which, we remind
you, are sometimes also called oblique asymptotes). We already saw in Chapter 2, how
we can identify skew asymptotes of rational functions by using polynomial division. Here
is a recipe for finding skew asymptotes that also works for functions that are not rational.
6.4. HOW TO USE THE RULEBOOK IN PRACTICE 263

Method 6.51 (Finding skew asymptotes) To determine the oblique asymptote


y = kx + m of a function as x ! 1 one can use the following algorithm:

Step 1. Compute A = limx!1 f (x)/x.

Step 2. Compute B = limx!1 (f (x) Ax).

Step 3. Conclude: If both limits A and B exist, y = Ax + B is the oblique


asymptote as x ! +1. If either A or B do not exist, then f has no oblique
asymptote as x ! 1.

By replacing 1 by 1, the same method allows us to find skew asymptotes as x ! 1.

Exercise 6.52 Use the above method to find the skew asymptotes as x ! 1 of the
following functions.
x3 + 2x2 5x
(a) f (x) = (b) f (x) = xe1/x .
x2 + 1

Hint: The skew asymptote in part (a) can also be found using polynomial division, as
we did in Chapter 2. Try both methods to see if they match up.

Exercise 6.53 In this exercise, we prove that Method 6.51 will always give the correct
answer.

(a) Prove that if f (x) has y = kx + m as an oblique asymptote as x ! 1, then the


above method produces gives A = k and B = m.
(b) Prove that if the above method indicates that f (x) has y = Ax + B as an oblique
asymptote, then y = Ax+B satisfies the definition of being an oblique asymptote
of f (x).
264 CHAPTER 6. LIMITS FOR FUNCTIONS

6.5 Exam exercises


Exercise 6.54 (Lund, January 2016) On this exercise you could get a maximum
of 5 points.

(a) (1 point) Define what we mean by limx!c f (x) = L.


(b) Use the definition you gave in (a) to verify one of the following limits.

(i) (2 points) lim (3x2 9x + 1) = 5


x!2

(ii) (3 points) lim (3x3 9x + 1) = 7


x!2

x3 3x + 1
(iii) (4 points) lim =3
x!2 2x 3

Exercise 6.55 (Lund, May 2015) Do one of the following exercises (both are worth
the same number of points).

(a) Give the definition of limx!a f (x) = L and use it to show that

lim (x2 + 3x + 1) = 5.
x!1

(b) Give the definition of limn!1 an = +1 and use it to show that

lim n!/20n = +1.


n!1

Remark: Part (b) actually belongs to Chapter 4.

Exercise 6.56 (Lund, August 2014)

(a) Formulate the epsilon-delta definition of the limit of a function at a point.


(b) Explain what it means for a function to be continuous at a point.
(c) Prove, using the epsilon-delta definition, that f (x) = 2x2 3x + 1 is continuous
at x = 1.

Exercise 6.57 (Lund, May 2014) Explain briefly what it means for a function f
to be continuous at a point x. Then use the epsilon-delta definition of the limit to
show that f (x) = x3 + 2x2 + 3x + 1 is continuous at x = 2.
6.6. ANSWERS TO SELECTED EXERCISES 265

6.6 Answers to selected exercises


6.6 Clockwise, starting at top-left: diverges to 1, 0, diverges, diverges.

6.7 The limit is 1, but Python cannot see this at the given scale due to round-off errors.

6.10 (a) 0, (b) 1, (c) diverges.

6.11 No.

6.14 C = 1.

6.15 No.

6.16 The function is continuous, but not at x = 0 (how can this be?).

6.19 = 1/2 and = 1/4, respectively (or any smaller than this).

6.20 (a) |x 3| < 1/4, (b) |x 3| < 1/40, (c) |x 3| < ✏/4.

6.23 It is a bit hard to see, but based on the first figure = 1/2 (or even something
slightly larger than this) seems to work, and in the second figure, = 1/4 seems to
work. The formula from the example gives = 4/12 = 1/3 and = 2/12 = 1/6,
respectively. It does not matter that these are not the same (since if one works,
then all smaller also automatically work – the point is to find some that is
small enough).

6.24 For example, do a for-loop that at step n evaluates |(x3 2x2 5x + 8) f (1)| at
x = 1 n/100. Break the loop when the expression is smaller than your given ✏.

6.25 (c) Following the steps of Example 6.21, we arrive at = min{1, ✏/21}. (But
just changing the steps slightly may lead to other choices for which are also
acceptable).

6.28 Using the continuity of f and (v’), then each of the following steps can be justified:

lim f (g(x)) = lim f (u) = f (b) = f ( lim g(x)).


x!a u!b x!a

6.30 Basically, the proofs are the same except conditions of the type n > N are replaced
by conditions of the type |x a| < .

6.33 (a) ✏2 = ✏, (b) ✏1 = 2, (c) = 1.

6.36 You need rules (i), (vi) and (vii).

6.37 (a) 1/5, (b) 1/2.

6.41 Rule (v’), (v) and then finally standard limit (iii).
266 CHAPTER 6. LIMITS FOR FUNCTIONS

6.42 (a) 2, (b) 1.

6.43 The limit is equal to 0. (Use trigonometric identities to relate this to the standard
limit for the sine function.)

6.44 The limit is equal to 0. Do a change of variables to make it into standard limit
(iii).

6.45 Rewrite the expressions using ax = ex log a , then you will find the answers (a) e2/3 ,
(b) 1.

6.46 (a) The correct definition is found by mimicking the definition for limn!1 an = L.
That is, you are looking to define what "x large =) f (x) close to L" means.

6.49 (a) y = 0 is a horistonal asymptote when x ! 1.

6.50 (a) Neither (i) nor (ii) have vertical asymptotes, but both have y = 0 as a ho-
risontal asymptote as x ! ±1. (ii).

6.52 (a) y = x + 2, (b) y = x + 1.

6.53 In this exercise, it is important to keep in mind that y = kx+m is a skew asymptote
for f (x) if limx!1 (f (x) kx m) = 0. In (a), the strategy is to add by 0 in both
the expressions for A and B to make kx + m appear, and then to use the definition
of the skew asymptote. In (b), it is enough to rewrite the computation that gives
you B in order to verify that y = Ax + B satisfies the definition of being a skew
asymptote.

You might also like