Analysis I

Vicky Neale

Michaelmas Term 2021

Contents
1 Introduction 4

2 Axioms for arithmetic in R 5

3 Properties of arithmetic in R 8

4 Ordering the real numbers 14

5 Inequalities and arithmetic 16

6 The modulus of a real number 19

7 The complex numbers 21

8 Upper and lower bounds 22

9 The (nearly) empty section 23

10 Supremum, infimum and completeness 23

11 Existence of roots 27

1
12 More consequences of completeness 29

13 Countability 31

14 More on countability 34

15 Introduction to sequences 37

16 Convergence of a sequence 40

17 Limits: first key results 48

18 Limits: modulus and inequalities 51

19 Bounded and unbounded sequences 54

20 Complex sequences 58

21 Subsequences 59

22 Algebra of Limits — part one 61

23 Algebra of Limits — part two 65

24 Orders of magnitude 68

25 Monotonic sequences 69

26 Convergent subsequences 74

27 Cauchy sequences 76

28 Convergence for series 79

29 Series: first results and a first test for convergence 82

2
30 Series: more results and another test for convergence 85

31 More on the Comparison Test 89

32 Ratio Test 91

33 Integral Test 95

34 Euler’s constant and rearranging series 100

35 Power series 102

36 Radius of convergence 105

37 Differentiation Theorem 110

3
1 Introduction
This course is a first introduction to real analysis. We’ll set the scene with
exploring basic properties of real numbers. Then we’ll go on to define what
it means for a sequence to converge, and for a series to converge, and we’ll
explore these ideas in detail, allowing us to give careful justifications of some
familiar results, as well as proving some that you might not have seen before.
There are several resources that will help you as you study the course:

- the lecture videos

- the slides that go with the lecture videos

- these notes

- the notes by Dr Hilary Priestley

- the problems sheets

- the additional content on Moodle

- each other

- your college tutors.

When I made the lecture videos, I used a mix of slides and handwriting,
but everything handwritten also appears on a typed slide. I’m making the
slides available alongside the videos, in case you would like a copy of these to
view or annotate. These notes (the document you’re reading now!) include
everything in the slides. You are encouraged to read them alongside the
notes by Dr Hilary Priestley.

4
Acknowledgements
These notes, and the lectures they accompany, are extremely closely based on
the notes produced by Dr Hilary Priestley. The same applies to the problems
sheets.
I would like these notes to be as useful as possible. If you (whether
student or tutor) think that you’ve noticed a typo, or mistake, or part that
is unclear, please check the current, up-to-date, notes on the website, to see
whether I’ve already fixed it. If not, please email me (vicky.neale@maths)
or put a note in the Moodle discussion, and I’ll do something about it, and
(with your permission) thank you here.
Thanks to Kendall De, Orson Hart, Aidan Strong and another (a student
who preferred not to be named) for helping to fix glitches in these notes and
the accompanying problems sheets.

2 Axioms for arithmetic in R

What are the real numbers?

We are going to work with the real numbers in this course. What is a real
number? We are not, in this course, going to define the real numbers — we
aren’t going to say what they are. Instead, we’ll focus on how they behave —
what they do. We’ll identify the key properties of the real numbers, and the
goal is to deduce everything from these properties. That leaves the task of
showing that there is a set with the required properties (that is, constructing
the real numbers), but that is not a task for this course.
So what are these key properties? They are all properties that will prob-
ably feel ‘obvious’. For us, they are assumptions. We assume that the real

5
numbers have these properties. You might feel surprised that there aren’t
more assumptions, but, as we’ll see, we can deduce the other familiar (and
not so familiar) properties from these assumptions. The goal is to avoid
assuming more than we have to.
One question to have in the back of your mind through this section is
“What other structures also have these properties?” We know that the ra-
tional numbers share some properties with the real numbers, and also the
complex numbers share some properties with the real numbers. But we also
know that in some way the rational numbers are definitely different from
the real numbers, and similarly for complex numbers, and we should see this
come up. If the rational numbers share all of our key properties of the real
numbers, then we haven’t assumed enough properties of the real numbers!
The flavour of this section will be about deducing basic properties, one
careful step at a time. Some people enjoy this way of working; others, well,
not so much. If you’re in the latter category, don’t worry: most of this
course will have quite a different flavour (although it will still be about giving
careful, rigorous proofs of results that might feel ‘obvious’). But, at its heart,
mathematics is built on the sort of axiomatic reasoning that we’ll see in this
section, so it’s good to have a sense of how that works.

Notation. We write R for the set of real numbers. We write Q for the set
of rational numbers, and C for the set of complex numbers.

Axioms for arithmetic in R

One really important feature of R is that we can do ‘arithmetic’. We have

operations of addition and multiplication. (Note that I didn’t mention sub-
traction or division. Can you think why?) Here is a careful statement of the
key properties of addition and multiplication.

6
 ˆ For every a, b ∈ R there is a unique real number a + b, called their sum.

ˆ For every a, b ∈ R there is a unique real number a · b, called their

product.

ˆ For a ∈ R there is a unique real number −a called its negative or its

additive inverse.

ˆ For a ∈ R with a ̸= 0 there is a unique real number 1

a
called its reciprocal
or its multiplicative inverse.

ˆ There is a special element 0 ∈ R called zero or the additive identity.

ˆ There is a special element 1 ∈ R called one or the multiplicative iden-

tity.

For all a, b, c ∈ R, we have

ˆ a+b=b+a (+ is commutative)

ˆ a + (b + c) = (a + b) + c (+ is associative)

ˆ a+0=a (additive identity)

ˆ a + (−a) = 0 (additive inverses)

ˆ a·b=b·a (· is commutative)

ˆ a · (b · c) = (a · b) · c (· is associative)

ˆ a·1=a (multiplicative identity)

ˆ if a ̸= 0 then a · 1
a
=1 (multiplicative inverses)

ˆ a · (b + c) = a · b + a · c (· distributes over +)

7
 ˆ 0 ̸= 1 (to avoid total collapse)

This is a long list of properties, called axioms.

I have given a name to each axiom, so that in proofs we can say which
property we are using at each stage. Some people prefer to number the ax-
ioms, but there is no standard way to do this, and I find it hard to remember
the numbers and easier to remember the names.
I think of the axioms as coming in bundles, to help me to remember
them. The first four tell us that ‘addition behaves nicely’. The next four tell
us that ‘multiplication behaves nicely’. Then there is an axiom that tells us
that ‘addition and multiplication interact nicely’, and a final technical detail
to clarify that R ̸= {0}.
As you study abstract algebra (such as linear algebra and group theory)
this year, you’ll find yourself coming across lists of axioms such as this again.

Definition. Let F be a set with operations + and · that satisfy the axioms
above. Then we say that F is a field.

Example. We’ve just said that R is a field. The rational numbers Q form a
field. The complex numbers C form a field. You’ll meet other fields too, in
other courses. The integers Z do not form a field.

In the next section, we’ll use these axioms to deduce the basic properties
of arithmetic in R. This reasoning would apply equally to any field. We’ll
need to make further assumptions about R, but we’ll postpone this till we’ve
studied basic arithmetic.

3 Properties of arithmetic in R
In the last section, we saw the arithmetic axioms for R. Now we’ll deduce
some properties of arithmetic in R.

8
Proposition 1. Let a, b, c, x, y be real numbers.

(i) If a + x = a for all a then x = 0 (uniqueness of 0).

(ii) If a + x = a + y then x = y (cancellation for +).

(iii) −0 = 0.

(iv) −(−a) = a.

(v) −(a + b) = (−a) + (−b).

(vi) If a · x = a for all a ̸= 0 then x = 1 (uniqueness of 1).

(vii) If a ̸= 0 and a · x = a · y then x = y (cancellation for ·).

1
(viii) If a ̸= 0 then 1 = a.
a

(ix) (a + b) · c = a · c + b · c.

(x) a · 0 = 0.

(xi) a · (−b) = −(a · b). In particular, (−1) · a = −a.

(xii) (−1) · (−1) = 1.

1 1
(xiii) If a · b = 0 then a = 0 or b = 0. If a ̸= 0 and b ̸= 0 then a·b
= a
· 1b .

Remark. ˆ (ii) shows the uniqueness of −a, the additive inverse of a.

ˆ (vii) shows the uniqueness of a1 , the multiplicative inverse of a (if a ̸= 0).

ˆ As we’ll see shortly, (i)–(v) can be proved using only the four axioms
about +.

ˆ Similarly, (vi)–(viii) can be proved using only the four axioms about ·.

9
 ˆ (ix)–(xiii) between them use all the axioms.

ˆ It’s worth proving results like this in a sensible order! Once we’ve
proved a property, we can add it to the list of properties we can assume
in subsequent parts. You’ll see that we prove some later parts using
earlier parts.

Proof. (i) Suppose that a + x = a for all a. Then

x = x + 0 (additive identity)

= 0 + x (+ is commutative)

= 0 (by hypothesis, with a = 0).

(ii) Suppose that a + x = a + y. Then

y = y + 0 (additive identity)

= y + (a + (−a)) (additive inverses)

= (y + a) + (−a) (+ is associative)

= (a + y) + (−a) (+ is commutative)

= (a + x) + (−a) (hypothesis)

= (x + a) + (−a) (+ is commutative)

= x + (a + (−a)) (+ is associative)

= x + 0 (additive inverses)

= x (additive identity).

(iii) We have 0 + 0 = 0 (additive identity)

and 0 + (−0) = 0 (additive inverses)

so 0 + 0 = 0 + (−0), so 0 = −0 (cancellation for + (ii)).

10
 (iv) We have

(−a) + a = a + (−a) (+ is commutative)

= 0 (additive inverses)

and (−a) + (−(−a)) = 0 (additive inverses),

so (−a) + a = (−a) + (−(−a)),

so a = −(−a) (cancellation for + (ii)).

(v) Exercise (see Sheet 1).

(vi)–(viii) Exercise — similar to (i), (ii), (iv).

(ix) (This is another form of distributivity, similar to the axiom but differ-
ent!)

We have

(a + b) · c = c · (a + b) (+ is commutative)

= c · a + c · b (· distributes over +)

= a · c + b · c (· is commutative – twice).

(x) We have a · (0 + 0) = a · 0 + a · 0 (· distributes over +),

and also

a · (0 + 0) = a · 0 (additive identity)

= a · 0 + 0 (additive identity)

so a · 0 = 0 (cancellation for + (ii))

11
 (xi) We have

a · b + a · (−b) = a · (b + (−b)) (· distributes over +)

= a · 0 (additive inverses)

and a · b + (−(a · b)) = 0 (additive inverses),

so a · (−b) = −(a · b) (cancellation for + (ii)).

(xii) We have

(−1) · (−1) = −((−1) · 1) ((xi) with a = −1, b = 1)

= −(−1) (multiplicative identity)

= 1 ((iv)).

(xiii) Suppose, for a contradiction, that a ̸= 0, b ̸= 0 but a · b = 0. Then

 
1 1
0= · · 0 ((x))
a b
 
1 1
=0· · (· is commutative)
a b
 
1 1
= (a · b) · · (hypothesis)
a b
 
1 1
= (b · a) · · (· is commutative)
a b
 
1 1
= (b · a) · · (· is associative)
a b
 
1 1
= (b · a · )· (· is associative)
a b
1
= (b · 1) · (multiplicative inverses)
b
1
=b· (multiplicative identity)
b
= 1 (multiplicative inverses)

12
 and this is a contradiction (0 ̸= 1).

So if a · b = 0 then a = 0 or b = 0.

 we showed that if a ̸= 0 and b ̸= 0 then a · b ̸= 0

Note that on the way
1 1 1 1 1
and (a · b) · · = 1 so = · (cancellation for · (vii)).
a b a·b a b

From now on, we can use all of these properties. We shan’t give such
detailed, one-axiom-at-a-time, derivations in the remainder of the course —
but we could, if we needed to!

Remark. (This remark is not part of the course!) You might be concerned
that if we don’t go back to the axioms every time then we might overlook an
unproved step, or might make a mistake. But going back to the axioms every
time is not practical: a research paper might take tens of pages to give a proof,
referring back to other results, which themselves build on results, which build
on results, . . . . This is where proof verification comes in: computers can take
care of this detailed checking, leaving humans to focus on things that humans
are good at (like having creative ideas for proofs).
One interesting project in this area (there are others, not just this one) is
Xena https://xenaproject.wordpress.com/what-is-the-xena-project/
— this is aimed at undergraduates, which is why I’m mentioning it. Some of
you might find it interesting. There’s an article about proof verification and
Xena in the London Mathematical Society Newsletter, pages 32–36 of https:
//www.lms.ac.uk/sites/lms.ac.uk/files/files/NLMS_484-forweb2.pdf.
Now back to the course . . . .

13
Notation. From now on, we use more familiar notation. We write

a − b for a + (−b)

ab for a · b
 
a 1
for a ·
b b
1
a−1 sometimes for .
a
The associativity of addition and multiplication means that we can write
expressions like a + b + c and xyz, without needing to write brackets.

Definition. Take a ∈ R \ {0}.

Define a0 = 1.
We define positive powers of a inductively: for integers k ⩾ 0, we define
ak+1 = ak · a.
1
For integers l ⩽ −1, we define al = .
a−l
Remark. Note that with this definition a1 = a and a2 = a·a (as we’d want).

Lemma 2. For a ∈ R \ {0} we have am an = am+n for m, n ∈ Z.

Proof. Exercise (see Sheet 1).

This finishes this section on arithmetic in R. In the next section, we’ll go

on to explore more key properties of R, in addition to it being a field.

4 Ordering the real numbers

When we picture R in our minds, typically it is not as a scattered collection
of numbers. Rather, we often picture them as lying along a number line,
usually running from left to right, with 0 in the ‘middle’; positive numbers
on the right, increasing as we move away from 0; and negative numbers on

14
the left, getting smaller as we move left away from 0. Even in writing that
description, I have made assumptions about R: I have assumed that we have
notions of ‘positive’ and ‘negative’, and that we can compare the sizes of two
real numbers. This section is about formalising those assumptions.
Here are our axioms for the usual ordering on R.
There is a subset P of R such that for a, b ∈ R

ˆ if a, b ∈ P then a + b ∈ P (+ and ordering)

ˆ if a, b ∈ P then a · b ∈ P (· and ordering)

ˆ exactly one of a ∈ P, a = 0 and −a ∈ P holds (positive, negative or 0).

The elements of P are called the positive numbers. The elements of P∪{0}
are called the non-negative numbers.
We write a < b, or b > a, exactly when b − a ∈ P.
We write a ⩽ b, or b ⩾ a, exactly when b − a ∈ P ∪ {0}.

Now, just as with the axioms for arithmetic, we can infer useful properties
from these axioms. First, some important properties of the ordering.

Proposition 3. Take a, b, c ∈ R. Then

(i) a ⩽ a; (reflexivity)

(ii) if a ⩽ b and b ⩽ a then a = b; (antisymmetry)

(iii) if a ⩽ b and b ⩽ c then a ⩽ c, and similarly with < in place of ⩽;

(transitivity)

(iv) exactly one of a < b, a = b and a > b holds. (trichotomy)

Proof. (i) We have a − a = 0 ∈ P ∪ {0} (additive inverses).

15
 (ii) Suppose that a ⩽ b and b ⩽ a.

If a − b = 0 or b − a = 0 then a = b (properties of +) and we are done.

If not, then b − a ∈ P and a − b ∈ P.

But b − a = −(a − b) (properties of +),

so then a − b ∈ P and −(a − b) ∈ P, contradicting ‘positive, negative

or 0’.

(iii) Note that c − a = c + (−a) = c + 0 + (−a) = c + (−b) + b + (−a) =

(c − b) + (b − a) (properties of +)

so if a < b and b < c then a < c (+ and ordering).

The cases where a = b and/or b = c are straightforward, and give the

result for ⩽.

(iv) This follows from ‘positive, negative or 0’.

In the next section, we’ll explore the interaction between the ordering
and basic arithmetic.

5 Inequalities and arithmetic

The next result has some useful results about inequalities and arithmetic,
which will save us from having to go back to the axioms every time.

Proposition 4. Take a, b, c ∈ R.

(i) 0 < 1.

(ii) a < b if and only if −b < −a. In particular, a > 0 if and only if
−a < 0.

16
(iii) If a < b then a + c < b + c.

(iv) If a < b and 0 < c then ac < bc.

(v) a2 ⩾ 0, with equality if and only if a = 0.

1
(vi) a > 0 if and only if a
> 0.
1 1
(vii) If a, b > 0 and a < b then < .
b a
Furthermore, (ii), (iii) and (iv) hold with ⩽ in place of <.

Proof. (i) By trichotomy, we have 0 < 1 or 0 = 1 or 0 > 1.

But ‘to avoid total collapse’ 0 ̸= 1. So it suffices to rule out 0 > 1.

Suppose, for a contradiction, that 0 > 1.

Then −1 ∈ P (by definition of >) so (−1) · (−1) ∈ P (· and ordering).

But (−1) · (−1) = 1 (Proposition 1 (xii)),

so 0 < 1 — but this contradicts trichotomy.

So 0 < 1.

(ii) Using properties of addition, we have

a<b⇔b−a∈P

⇔ (−a) − (−b) ∈ P

⇔ −a > −b.

(iii) Assume that a < b.

Then (b + c) − (a + c) = b − a > 0 so a + c < b + c.

(iv) Assume that a < b and 0 < c.

Then bc − ac = (b − a)c > 0 (· and ordering).

17
 (v) Certainly a2 = 0 if and only if a = 0 (Proposition 1 (x) and (xiii)).

If a ̸= 0, then exactly one of a and −a is positive, and either way

a2 = a · a = (−a) · (−a) > 0 (· and ordering).

1
(vi) Suppose, for a contradiction, that a > 0 and a
< 0, so a > 0 and
− a1 > 0.
   
1 1
Then −1 = − a · =a· − > 0. But this contradicts (i).
a a
1
Similarly if a < 0 and a
> 0.

(vii) Suppose that a, b > 0 and a < b.

Then a1 , 1
b
> 0 by (vi),
1 1 1 1
so a · · < b · · by (iv),
a b a b
1
so b
< a1 .

Now we can prove a useful inequality (you’ll have an opportunity to apply

it on Sheet 1).

Theorem 5 (Bernoulli’s Inequality). Let x be a real number with x > −1.

Let n be a positive integer. Then (1 + x)n ⩾ 1 + nx.

Proof. By induction on n. Fix x > −1.

n = 1: clear.
induction step: suppose the result holds for some n ⩾ 1, that is, (1 + x)n ⩾
1 + nx.
Note that 1 + x > 0, and nx2 ⩾ 0 (since n > 0 and x2 ⩾ 0 by Proposition
4 (v)).

18
 Then

(1 + x)n+1 = (1 + x)(1 + x)n (by definition)

⩾ (1 + x)(1 + nx) (induction hypothesis and Prop 4 (iv))

= 1 + (n + 1)x + nx2 (properties of arithmetic)

⩾ 1 + (n + 1)x (since nx2 ⩾ 0).

So, by induction, the result holds.

In the next section, we’ll move on to consider the modulus of a real

number.

6 The modulus of a real number

Definition. Let a ∈ R. The modulus |a| of a is defined to be




 a if a > 0


|a| := 0 if a = 0




−a if a < 0.


(It is also sometimes called the absolute value of a.)

Remark. The modulus is well defined (that is, this is a legitimate definition)
thanks to the ‘positive, negative or 0’ property (essentially trichotomy).

Here are some basic properties of the modulus.

Proposition 6. Take a, b, c ∈ R. Then

(i) | − a| = |a|;

(ii) |a| ⩾ 0;

19
(iii) |a|2 = a2 ;

(iv) |ab| = |a||b|;

(v) −|a| ⩽ a ⩽ |a|;

(vi) if c ⩾ 0, then |a| ⩽ c if and only if −c ⩽ a ⩽ c; and similarly with

weak inequalities (⩽, ⩾) replaced by strict (<, >).

Proof.(i), (ii) Immediate from the definition, since a > 0 if and only if −a < 0.

(iii) Check using the definition and trichotomy – go through the cases and
also use (−a)(−a) = a2 .

(iv) Check the cases using the definition and trichotomy.

(v) If a ⩾ 0, then −|a| ⩽ 0 ⩽ a = |a|.

If a < 0, then −|a| = a < 0 ⩽ |a|.

(vi) Assume that c ⩾ 0.

(⇒) Suppose that |a| ⩽ c. Then, by (v), −c ⩽ −|a| ⩽ a ⩽ |a| ⩽ c, and

we’re done by transitivity (Proposition 3).

(⇐) Suppose that −c ⩽ a ⩽ c. Then −a ⩽ c and a ⩽ c. But |a| is a

or −a, so |a| ⩽ c.

Similarly for the version with strict inequalities.

Theorem 7 (Triangle Inequality). Take a, b ∈ R. Then

(i) |a + b| ⩽ |a| + |b|;

(ii) |a + b| ⩾ ||a| − |b||.

20
Remark. (ii) is called the Reverse Triangle Inequality.

Proof. (i) We have −|a| ⩽ a ⩽ |a| and −|b| ⩽ b ⩽ |b|, by Proposition 6.

We can add these (see Sheet 1 Q2); using properties of addition, we

get − (|a| + |b|) ⩽ a + b ⩽ |a| + |b|.

By Proposition 6 (vi) (with c = |a|+|b| ⩾ 0), this gives |a+b| ⩽ |a|+|b|.

(ii) By (i), we have |a| = |a + b + (−b)| ⩽ |a + b| + | − b| = |a + b| + |b|,

so |a + b| ⩾ |a| − |b|.

Similarly (swap a and b), |a + b| ⩾ |b| − |a|.

Now ||a| − |b|| is |a| − |b| or |b| − |a|, so |a + b| ⩾ ||a| − |b||.

7 The complex numbers

You met (or renewed your acquaintance with) the set C of complex numbers
in the course Introduction to Complex Numbers ???, so we shan’t repeat much
of that material here. We’ll just focus on revisiting the complex numbers with
the new perspective of having considered axioms for R.
In particular, as we said earlier, C is a field (under the usual addition and
multiplication). If you are feeling enthusiastic, then check the axioms!
But C is fundamentally different from R because there is no ordering on
C that satisfies the ordering axioms.

Exercise. Prove this!

As you saw in the Complex Numbers course, the Triangle Inequality holds
in C, and so does the Reverse Triangle Inequality.

21
 So far, we know that R is an ordered field. But we still haven’t captured
everything that’s important about R. The rational numbers Q also form
an ordered field, but there are fundamental differences between Q and R.
√
Intuitively, in Q there are ‘gaps’ in the number line (such as at 2), whereas
this is not the case in R. In the next chunk of the course, we’ll explore this
in detail, to try to identify what additional property or properties we need
to assume hold in R.

8 Upper and lower bounds

In the next few sections of the course, we going to explore the difference
between Q and R, as discussed in the last paragraph of the previous section.
One helpful example to have in mind is the square root of two. We know
√
that 2 is not in Q (you have probably seen a proof of this elsewhere). But
there is a positive real number that squares to give 2. One goal of these
sections of the course is to prove the existence of this positive real number
√
(which we call 2).
The key property, which R has and Q does not, is called completeness.
But before we get there, we need a few preliminary definitions.

Definition. Let S ⊆ R. Take b ∈ R. We say that

ˆ b is an upper bound of S if s ⩽ b for all s ∈ S;

ˆ b is a lower bound of S if s ⩾ b for all s ∈ S;

ˆ S is bounded above if S has an upper bound;

ˆ S is bounded below if S has a lower bound;

ˆ S is bounded if S is bounded above and below.

22
Example. I could have put various examples and non-examples here, but
instead I’ve put them in a short Moodle quiz for you to try. You’ll get
immediate feedback on your answers, to help you explore and check your
understanding of the definitions. Please go to the Moodle course page for
Analysis I, and try quiz 8.1, before you read on to the next section.

9 The (nearly) empty section

Sorry, this is here just to resolve section/video numbering issues!

10 Supremum, infimum and completeness

Some upper bounds are more interesting than others. The set [0, 1] has
upper bounds including 15, 1, 1.7 and infinitely many more. Of these, 1 feels
special. This is the focus of our next definition.

Definition. Let S ⊆ R. We say that α ∈ R is the supremum of S, written

sup S, if

(i) s ⩽ α for all s ∈ S; (α is an upper bound of S)

(ii) if s ⩽ b for all s ∈ S then α ⩽ b (α is the least upper bound of S).

Remark. If S has a supremum, then sup S is unique. (Check you can show
this!)

Now that we have defined the supremum, we can state our final key
property of R (in addition to the properties that make it an ordered field).

Completeness axiom for the real numbers Let S be a non-empty subset

of R that is bounded above. Then S has a supremum.

23
Remark. There are two conditions on S here: non-empty, and bounded
above. They are both crucial!
It is easy to forget the non-empty condition, but it has to be there: the
empty set does not have a supremum, because every real number is an upper
bound for the empty set — there is no least upper bound.
The condition that S is bounded above is also necessary: a set with no
upper bound certainly has no supremum.

Example. ˆ Let S = [1, 2). Then 2 is an upper bound, and is the

least upper bound: if b < 2 then b is not an upper bound because
max(1, 1 + 2b ) ∈ S and max(1, 1 + 2b ) > b. Note that in this case
sup S ̸∈ S.

ˆ Let S = (1, 2]. Then we again have sup S = 2, and this time sup S ∈ S.

The supremum is the least upper bound of a set. There’s an analogous

definition for lower bounds.

Definition. Let S ⊆ R. We say that α ∈ R is the infimum of S, written

inf S, if

ˆ s ⩾ α for all s ∈ S; (α is a lower bound of S)

ˆ if s ⩾ b for all s ∈ S then α ⩾ b (α is the greatest lower bound of S).

Let’s explore some useful properties of sup and inf.

Proposition 8. (i) Let S, T be non-empty subsets of R, with S ⊆ T and

with T bounded above. Then S is bounded above, and sup S ⩽ sup T .

(ii) Let T ⊆ R be non-empty and bounded below. Let S = {−t : t ∈ T }.

Then S is non-empty and bounded above. Furthermore, inf T exists,
and inf T = − sup S.

24
Remark. (ii) and a similar result with sup and inf swapped essentially tell
us that we can pass between sups and infs. Any result we prove about sup
will have an analogue for inf. Also, we could have phrased the Completeness
Axiom in terms of inf instead of sup. Proposition 8(ii) tells us that we don’t
need separate axioms for sup and inf.

Proof. (i) Since T is bounded above, it has an upper bound, say b.

Then t ⩽ b for all t ∈ T , so certainly t ⩽ b for all t ∈ S, so b is an

upper bound for S.

Now S, T are non-empty and bounded above, so by completeness each

has a supremum.

Note that sup T is an upper bound for T and hence also for S, so
sup T ⩾ sup S (since sup S is the least upper bound for S).

(ii) Since T is non-empty, so is S.

Let b be a lower bound for T , so t ⩾ b for all t ∈ T .

Then −t ⩽ −b for all t ∈ T , so s ⩽ −b for all s ∈ S, so −b is an upper

bound for S.

Now S is non-empty and bounded above, so by completeness it has a

supremum.

Then s ⩽ sup S for all s ∈ S, so t ⩾ − sup S for all t ∈ T , so − sup S

is a lower bound for T .

Also, we saw before that if b is a lower bound for T then −b is an upper

bound for S.

Then −b ⩾ sup S (since sup S is the least upper bound),

so b ⩽ − sup S.

25
 So − sup S is the greatest lower bound.

So inf T exists and inf T = − sup S.

You might be wondering how all this relates to familiar notions of maxi-
mum and minimum so let’s explore that.

Definition. Let S ⊆ R be non-empty. Take M ∈ R. We say that M is the

maximum of S if

(i) M ∈ S; (M is an element of S)

(ii) s ⩽ M for all s ∈ S (M is an upper bound for S).

Remark. ˆ If S is empty or S is not bounded above then S does not

have a maximum. (Check this!)

ˆ Let S ⊆ R be non-empty and bounded above, so (by completeness)

sup S exists.

Then S has a maximum if and only if sup S ∈ S.

Also, if S has a maximum then max S = sup S.

(Check this!)

Definition. Let S ⊆ R be non-empty. Take m ∈ R. We say that m is the

minimum of S if

(i) m ∈ S; (m is an element of S)

(ii) s ⩾ m for all s ∈ S (m is a lower bound for S).

Here is a key result about the supremum, which we’ll use a lot. It is a
quick consequence of the definition, but it will be useful to have formulated
it in this way.

26
Proposition 9 (Approximation Property). Let S ⊆ R be non-empty and
bounded above. For any ε > 0, there is sε ∈ S such that sup S − ε < sε ⩽
sup S.

Proof. Take ε > 0.

Note that by definition of the supremum we have s ⩽ sup S for all s ∈ S.
Suppose, for a contradiction, that sup S − ε ⩾ s for all s ∈ S.
Then sup S − ε is an upper bound for S, but sup S − ε < sup S. Contra-
diction.
So there is sε ∈ S with sup S − ε < sε .

11 Existence of roots
Now that we have identified the completeness property of R, we are ready to
prove that R contains a square root of 2.

Theorem 10. There exists a unique positive real number α such that α2 = 2.

Proof. Existence Let S = {s ∈ R : s > 0, s2 < 2}.

Idea: argue that S has a supremum, and show that sup S has the required
properties.
Note that S is non-empty (eg 1 ∈ S)
and S is bounded above, because if x > 2 then x2 > 4 (properties of
ordering) so x ̸∈ S, so 2 is an upper bound for S.
So, by completeness, S has a supremum. Let α = sup S.
Note that certainly α > 0 (since 1 ∈ S so α ⩾ 1).
By trichotomy, we have α2 < 2 or α2 = 2 or α2 > 2.
Idea: show that if α2 < 2 or α2 > 2 then we get a contradiction.
Case 1 Suppose, for a contradiction, that α2 < 2.

27
 Then α2 = 2 − ε for some ε > 0.
Idea: consider α + h for a small h > 0. Later on, we’ll choose h small
enough that (α + h)2 < 2, and that will be a contradiction because α + h ∈ S
and α + h > sup S.
Note that α ⩽ 2 (we said earlier that 2 is an upper bound for S).
For h ∈ (0, 1) we have

(α + h)2 = α2 + 2αh + h2

= 2 − ε + 2αh + h2

⩽ 2 − ε + 4h + h

⩽ 2 − ε + 5h

ε 1
so let h = min( 10 , 2 ) and then (α + h)2 < 2.
Now α + h ∈ S and α + h > sup S. This is a contradiction.
So it is not the case that α2 < 2.
Case 2 Suppose, for a contradiction, that α2 > 2.
Then α2 = 2 + ε for some ε > 0.
Idea: consider α − h for a small h > 0. Later on, we’ll choose h small
enough that (α − h)2 > 2, and that will lead to a contradiction because
α − h < sup S.
For h ∈ (0, 1) we have

(α − h)2 = α2 − 2αh + h2

= 2 + ε − 2αh + h2

⩾ 2 + ε − 4h

so choose h = min( 8ε , 12 , α2 ) and then (α − h)2 > 2 (and also α − h > 0).
Now α − h < sup S, so by the Approximation property there is s ∈ S
with α − h < s.

28
 But then 2 < (α − h)2 < s2 < 2, which is a contradiction.
So it is not the case that α2 > 2.
Hence, by trichotomy, α2 = 2.
Uniqueness Suppose that β is also a positive real number such that β 2 = 2.
Aim: α = β.
Then 0 = α2 − β 2 = (α − β)(α + β)
and α + β > 0, so α = β.

Proposition 11. Q is not complete (with the ordering inherited from R).

Proof. If Q were complete, then the proof of Theorem 10 would work just as
well in Q. But we know that there is not an element of Q that squares to 2.
So Q is not complete.

Theorem 12. Let n be an integer with n ⩾ 2, and take a positive real number
r. Then r has a real nth root.

Proof. Exercise. (See Sheet 2 for the case of the cube root of 2.)

12 More consequences of completeness

In this course, we write N for the set of positive integers, so N = Z>0 .

Theorem 13 (Archimedean property of N). N is not bounded above.

Proof. Idea: if there’s an upper bound then we can find a natural number
just less than it, and add 1.
Suppose, for a contradiction, that N is bounded above.
Then N is non-empty and bounded above, so by completeness (of R) N
has a supremum.

29
 By the Approximation property with ε = 21 , there is a natural number
1
n ∈ N such that sup N − 2
< n ⩽ sup N.
Now n + 1 ∈ N and n + 1 > sup N. This is a contradiction.

1
Corollary 14. Let ε > 0. Then there is n ∈ N such that 0 < n
< ε.

1
Proof. If not, then ε
would be an upper bound for N. This would contradict
Theorem 13.

Theorem 15. Let S be a non-empty subset of Z.

(i) If S is bounded below, then S has a minimum.

(ii) If S is bounded above, then S has a maximum.

Proof. (i) Assume that S is bounded below.

Then, by completeness (applied to {−s : s ∈ S}), S has an infimum.

Secret aim: inf S ∈ S.

By the Approximation property (with ε = 1), there is n ∈ S such that

inf S ⩽ n < inf S + 1. Aim: inf S = n.

Suppose, for a contradiction, that inf S < n.

Write n = inf S + δ, where 0 < δ < 1.

By the Approximation property (with ε = δ), there is m ∈ S such that

inf S ⩽ m < inf S + ε = n.

Now m < n so n − m > 0

but n − m is an integer, so n − m ⩾ 1.

Now n ⩾ m + 1 ⩾ inf S + 1. This is a contradiction.

So n = inf S ∈ S so inf S = min S.

30
 (ii) Similar.

Proposition 16. Take a, b ∈ R with a < b. Then

(i) there is x ∈ Q such that a < x < b (the rationals are dense in the
reals); and

(ii) there is y ∈ R \ Q such that a < y < b (the irrationals are dense in the
reals).

Proof. Exercise (see Sheet 2).

Summary of our work so far

R is a complete ordered field.

This sums up the key properties we have identified as our assumptions

about R. From this, we shall develop the theory of real analysis.

13 Countability
The concept of countability gives us a way to distinguish between sets by
comparing their ‘sizes’. This will be a quick introduction. You can find
more information in the supplementary notes by Dr Hilary Priestley, on the
Moodle Analysis I course.
Our main tool for comparing the ‘sizes’ of two sets is to ask whether there
is a bijection between them. In this section and the next, we’ll often be using
the notions of bijection, injection and surjection. If you don’t feel confident
with the definitions of these, then I recommend you remind yourself of the
definitions before you continue with this section.

31
Definition. Let A be a set. We say that A is finite if A = ∅ or there exists
n ∈ N such that there is a bijection f : A → {1, 2, . . . , n}. We say that A is
infinite if it is not finite

Remark. ˆ A subset of a finite set is finite.

ˆ A non-empty finite subset of R is bounded above (in fact, has a maxi-

mum) and so a subset of R that is not bounded above is infinite.

ˆ N is not bounded above (by the Archimedean property) so is infinite.

Definition. Let A be a set. We say that A is

ˆ countably infinite if there is a bijection f : A → N;

ˆ countable if there is an injection f : A → N;

ˆ uncountable if A is not countable.

Remark. There are variations on the details of these definitions, so it’s worth
checking carefully if you’re looking at a book or other source. For example,
some people say ‘countable’ where we are using ‘countably infinite’.

Here are a couple of useful properties.

Proposition 17. Let A be a set.

(i) A is countable if and only if A is countably infinite or finite.

(ii) If there is an injection f : A → B and an injection g : B → A, then

there is a bijection h : A → B.

Proof. Not in this course. See Priestley’s supplementary notes on countabil-

ity.

32
Proposition 18. Each of the following sets is countably infinite.

(i) N

(ii) N ∪ {0}

(iii) {2k − 1 : k ∈ N}

(iv) Z

(v) N × N.

Remark. It might feel surprising that the set of odd natural numbers ‘has
the same size as’ the set of all natural numbers!

Proof. (i) Clear.

(ii) Define f : N ∪ {0} → N by f (n) = n + 1. This is a bijection.

(iii) Define f : N → {2k − 1 : k ∈ N} by f (n) = 2n − 1.

(iv) Idea: line up the integers as 0, 1, −1, 2, −2, 3, −3, . . . .

Define f : Z → N by

2k

if k ⩾ 1
f (k) =
1 − 2k

if k ⩽ 0.

This is a bijection.

(v) Define f : N × N → N by f ((m, n)) = 2m−1 (2n − 1).

Claim f is a bijection.

Proof of claim

injective: If f ((m1 , n1 )) = f ((m2 , n2 )) then

2m1 −1 (2n1 − 1) = 2m2 −1 (2n2 − 1),

33
 so, by uniqueness of prime factorisation in N, 2m1 −1 = 2m2 −1 and 2n1 −
1 = 2n2 − 1,

so m1 = m2 and n1 = n2 .

surjective: Take k ∈ N.

Then k = 2r (2s + 1) for some r, s ⩾ 0

k
(consider the set T = {t ∈ Z⩾0 : 2t
∈ N} — this is non-empty and
bounded above so has a maximum).

Then k = f (r + 1, s + 1).

14 More on countability
We can build new countable sets from old. This is a very helpful way to
prove that a set is countable!

Proposition 19. Let A, B be countable sets.

(i) If A and B are disjoint, then A ∪ B is countable.

(ii) A × B is countable.

Remark. In (i), we don’t need the condition that A and B are disjoint, but
it makes life easier for our proof.

Proof. Since A and B are countable, there are injections f : A → N and

g : B → N.

(i) Idea: to list elements of A ∪ B, alternate taking elements from A, B.

34
 Define h : A ∪ B → N by

2f (x) − 1

if x ∈ A
h(x) =
2g(x) if x ∈ B.


This is an injection (because f and g are).

(ii) Define h : A × B → N by h((a, b)) = 2f (a) 3g(b) .

By the uniqueness of prime factorisation in N, this is an injection.

Theorem 20. Q>0 is countable.

Proof. Define f : Q>0 → N by f ( pq ) = 2p 3q where p, q ∈ Z>0 and hcf(p, q) =

1.
This is an injection (by uniqueness of prime factorisation in N).

Corollary 21. Q is countable.

Proof. We can write Q = Q>0 ∪ {0} ∪ Q<0 . This is a disjoint union.

We have just seen that Q>0 is countable, and similarly so is Q<0 , and {0}
is finite and hence countable.
Hence, by Proposition 19, Q is countable.

Our next task will be to show that R is uncountable. We’re going to do

this using decimal expansions. That means that we need to know that deci-
mal expansions exist. Ideally we’d study sequences and series for a bit, then
look at decimal expansions, then prove that R is uncountable. But I’d like
to wrap up this section on countability now. So we’re going to assume a fact
about decimal expansions, and deduce that R is uncountable. I encourage
you to revisit the fact later in the course — I’ll let you know when we’ve cov-
ered the relevant theory. You could also (now) look at Dr Hilary Priestley’s

35
supplementary notes on decimal expansions and the uncountability of R (on
the Moodle Analysis I page).
Fact Every real number has a decimal expansion, and if we require that we
choose a non-terminating expansion (such as 0.24999 . . . for 14 ) rather than a
terminating one (such as 0.25 for 41 ) where there is a choice, then this decimal
expansion is unique.

Theorem 22. R is uncountable.

Remark. The proof strategy we are going to use is called Cantor’s diagonal
argument.

Proof. It suffices to show that (0, 1] is uncountable.

Note that certainly (0, 1] is not finite (by Corollary 14 of the Archimedean
property).
Suppose, for a contradiction, that (0, 1] is countably infinite. List the
elements as x1 , x2 , x3 , . . . . (If you like, we have a bijection N → (0, 1], and
x1 = f (1), x2 = f (2), x3 = f (3), . . . .)
Each has a non-terminating decimal expansion (where relevant choosing
the non-terminating option):

x1 = 0.a11 a12 a13 a14 . . .

x2 = 0.a21 a22 a23 a24 . . .

x3 = 0.a31 a32 a33 a34 . . .

..
.

xk = 0.ak1 ak2 ak3 ak4 . . .

..
.

Construct a real number x ∈ (0, 1] with decimal expansion 0.b1 b2 b3 . . .

36
where

5

if akk = 6
bk =
6 if akk ̸= 6.


Then x ̸= xk for all k, because x differs from xk in the k th decimal place,

so x is not on our list, which supposedly contained all elements of (0, 1]. This
is a contradiction.

Remark. The only significance of the choice of 5 and 6 as the key digits when
defining x was that we didn’t involve 0 or 9, to avoid issues with non-unique
decimal expansions.

There are many further interesting things to say about countability, but
not in this course. We need to move on to consider sequences. Before you
continue to the next section, try to come up with a collection of examples
of sequences that you can use as your personal repertoire for testing the
definitions and results we’ll look at. Try to find some ‘typical’ and some
‘extreme’ examples of sequences that you think converge, and of sequences
that you think don’t converge.

15 Introduction to sequences
Remark. You are already familiar with the sine and cosine, exponential and
logarithm functions, and with raising a number to a non-integer power. At
the moment, though, we haven’t formally defined these. We’ll do so later in
the course (using infinite series—that’s why it needs to wait till later), and
this term and later in other Analysis courses you’ll explore and prove the
familiar properties of these function.

37
 All this means that at the moment our collection of functions we’ve de-
fined is rather small, and doesn’t give the richness we’d like when exploring
examples of sequences. So for now we’ll work with these familiar functions
(trig, exponential, log)—we’ll assume that they exist and have the properties
we expect. You can do this on the problems sheets too. When we come to
defining them later on, you can watch out to see that we don’t have any
circular arguments!

Notation. When we use logarithms, these will all be to the base e. We write
log x for loge (x). We don’t write ln x.
For a > 0 and x ∈ R, we define ax = ex log a . (Of course this relies
on definitions of the exponential and logarithm functions, which will come
later.)

Remark. Examples can be really useful. I don’t mean worked examples

(although these can also be really useful), I mean examples of objects that
do or don’t have certain properties. I’ll include some examples in these
notes and the accompanying videos. You’ll find additional examples in Dr
Hilary Priestley’s lecture notes, on the Moodle page for Analysis I, and I
encourage you to work through those too. I also encourage you to try your
own examples (and non-examples), to help you to deepen your experience
and understanding of the definitions and results we’ll meet.

Example. Here are some informal examples of sequences.

3 33 333 3333
ˆ , , , , . . . are approximations to 31 , each better than the
10 100 1000 10000
previous.
14 141 1414 14142 √
ˆ , , , , . . . are approximations to 2, each better than
10 100 1000 10000
the previous.

38
 ˆ Take ε > 0. Then, by the Archimedean property, there is N ⩾ 1 such
1 1 1
that 0 < N
< ε. Now for all n ⩾ N we have 0 < n
⩽ N
< ε. We
see that apart from finitely many terms at the start, the terms of the
sequence 1, 21 , 13 , 14 , 51 , . . . all lie within distance ε of 0. This is the case
for any positive real number ε.

ˆ 1, −1, 2, −2, 3, −4, 4, −4, . . . is another sequence, and intuitively it feels

as though it does not tend to a limit.

ˆ 7, 1.2, −5, 2, 324, −9235.32, . . . is another sequence—there is no clear

pattern to the terms (I just typed them wherever my fingers landed),
but it is still a sequence.

What exactly is a sequence?

Definition. A real sequence, or sequence of real numbers, is a function

α : N → R. We call α(n) the nth term of the sequence.
We usually write an for α(n), and say that α defines the sequence (an )
with terms a1 , a2 , a3 , a4 , . . . . We might also write this as (an )n⩾1 or (an )∞
n=1 .

Similarly, a complex sequence is formally a function α : N → C, and we

write it as (an ), where now an ∈ C for n ⩾ 1.

Remark. ˆ The order of the terms in a sequence matters!

ˆ We write (an ) for the sequence, and an for a term of the sequence.

ˆ Much of the theory relating to sequences applies to both real and com-
plex sequences. Sometimes, though, we’ll need to focus only on real
sequences—for example if we’re using inequalities. In this case we’ll
carefully specify that we’re working with real sequences. If we don’t

39
 specify, and just say ‘sequences’, then it applies equally to real and com-
plex sequences. We’ll also have a section (and corresponding video) at
the end of this block concentrating on complex sequences.

Example. ˆ Let an = (−1)n . Then the first few terms of the sequence
are −1, 1, −1, 1, −1, 1, . . . .

ˆ Let an = sin n
2n+1
. Then the first few terms of the sequence are
1 1 1
sin 1, sin 2, sin 3, . . . .
3 5 7

ˆ Let

0

if n is prime
an =
1 + 1
otherwise.

n

Then the first few terms of the sequence are 2, 0, 0, 45 , 0, 76 , 0, 98 , . . . .

ˆ Let an = n. Then the first few terms of the sequence are 1, 2, 3, 4, 5, . . . .

Definition. We can make new sequences from old. Let (an ), (bn ) be se-
quences and let c be a constant. Then we can define new sequences ‘termwise’:
(an + bn ), (−an ), (an bn ), (can ), (|an |). If bn ̸= 0 for all n, then we can also
define a sequence ( abnn ).

Example. Let an = (−1)n and bn = 1 for n ⩾ 1.

Then the first few terms of (an + bn ) are 0, 2, 0, 2, 0, 2, . . . ; and (−an ) =
((−1)n+1 ); and (|an |) = (bn ).

16 Convergence of a sequence
Before we see a formal definition of convergence, let’s consider some examples
informally. Here’s one way I like to visualise a sequence. These are examples
from the previous section. Each graph plots the points (n, an ) for 1 ⩽ n ⩽ 10.

40
ˆ an = (−1)n


0

if n is prime
ˆ an =
1 + 1
otherwise

n

41
ˆ an = sin n
2n+1

ˆ an = n

42
 Here is an unofficial picture of the definition of convergence.

Definition. Let (an ) be a real sequence, let L ∈ R. We say that (an )

converges to L as n → ∞ if

∀ε > 0 ∃N ∈ N such that ∀n ⩾ N, |an − L| < ε.

In this case we write an → L as n → ∞, and we say that L is the limit of

(an ).

Remark. ˆ We might also say that (an ) tends to L as n → ∞, and we

might also write that lim an = L.
n→∞

ˆ N can depend on ε, and almost always will.

ˆ The ‘order of the quantifiers’ matters. We wrote “∀ε > 0 ∃N ∈

N . . .”. This order allows N to depend on ε. If we wrote “∃N ∈
N such that ∀ε > 0 . . .” that would be something quite different.

We could replace n ⩾ N in the definition by n > N , and |an − L| < ε

by |an − L| ⩽ ε, without changing the definition. (Check this!) But
it’s crucial that we have ε > 0 not ε ⩾ 0. (Check this!)

43
 ˆ I put ‘the’ limit in the definition. We’ll see later that if it exists then
it’s unique.

Definition. Let (an ) be a real sequence. We say that (an ) converges, or is

convergent, if there is L ∈ R such that an → L as n → ∞. If (an ) does not
converge, then we say that it diverges, or is divergent.

Intuitively, the first thousand or million terms of a sequence shouldn’t

affect whether it converges. We’ll prove a result that makes this precise, but
first we need a quick definition.

Definition. Let (an ) be a sequence. A tail of (an ) is a sequence (bn ), where

for some natural number k we have bn = an+k for n ⩾ 1. That is, (bn ) is the
sequence obtained by deleting the first k terms of (an ).

Lemma 23 (Tails Lemma). Let (an ) be a sequence.

(i) If (an ) converges to a limit L, then every tail of (an ) also converges,
and to this same limit L.

(ii) If a tail (bn ) = (an+k ) of (an ) converges, then (an ) converges.

Proof. (i) Take a tail of (an ): take k ⩾ 1 and let bn = an+k for n ⩾ 1.

Assume that (an ) converges to a limit L.

Take ε > 0.

Then there is N such that if n ⩾ N then |an − L| < ε.

Now if n ⩾ N then n + k ⩾ N so |an+k − L| < ε, that is, |bn − L| < ε.

So (bn ) converges and bn → L as n → ∞.

(ii) Assume that (bn ) = (an+k ) converges.

Then there is L ∈ R such that bn → L as n → ∞.

44
 Take ε > 0.

Then there is N such that if m ⩾ N then |bm − L| < ε, that is,

|am+k − L| < ε.

Now if n ⩾ N + k then n = m + k where m ⩾ N , and so |an − L| < ε.

So (an ) converges and an → L as n → ∞.

Example. We’ll see later (soon!) that there are other ways to prove con-
vergence, not only directly from the definition. But for now we’ve only got
the definition (and the Tails Lemma), so let’s get some practice using what
we’ve got so far.

1
Claim. n
→ 0 as n → ∞.

Proof. Take ε > 0.

1
Then there is N ∈ N such that N
< ε (by the Archimedean property).

For n ⩾ N we have | n1 − 0| = 1
n
⩽ 1
N
< ε.
1
So n
→ 0 as n → ∞.

Claim. Let an = 1 + (−1)n √1n for n ⩾ 1. Then an → 1 as n → ∞.

Proof. Take ε > 0.

Aim: want N such that if n ⩾ N then |an − 1| < ε

that is, |(1 + (−1)n √1n ) − 1| < ε

that is, √1 < ε,

n

45
 1 √
that is, ε
< n.

Take N = ⌈ ε12 ⌉ + 1.

Here ⌈x⌉ denotes the ceiling function: it is defined to be the smallest

integer greater than or equal to x. Informally, if x is an integer then
take that value; otherwise, round up to the next integer.

If n ⩾ N , then

1
n>
ε2
√ 1
so n >
ε
1
so √ < ε
n
so |an − 1| < ε.

So an → 1 as n → ∞.

n cos(n3 + 1)
Claim. Let an = for n ⩾ 1. Then an → 0 as n → ∞.
5n2 + 1

Proof. Take ε > 0.

Aim: want N such that if n ⩾ N then |an − 0| < ε,

 n cos(n3 + 1) 
 
that is, 
 <ε
5n2 + 1 
 
 n 
but | cos(n3 + 1)| ⩽ 1 so it’s enough to ensure that  2 <ε
5n + 1 
 
and 5n2 + 1 ⩾ 5n2 so it’s enough to ensure that  5nn2  < ε
1 1
that is, 5n
< ε, that is, n > 5ε
.

Take N = ⌈ 1ε ⌉ + 1.

46
 1
If n ⩾ N , then n > 5ε
so

 n cos(n3 + 1) 
 
|an | =   ⩽ 1 < ε.
5n2 + 1  5n

So an → 0 as n → ∞.

Remark. Here are some top tips!

ˆ We don’t need the smallest possible N . It’s (almost always) not even
interesting to know what it is. So make your life easier! If an inequality
(in the right direction) helps, then go for it.

ˆ Be careful to make sure that the logic flows in the right direction, and
that you’ve set out the logic explicitly. Hopefully the examples we’ve
just seen help you to have ideas of how to do this.

ˆ The definition officially says N ∈ N, but we don’t really care whether

N is a natural number. If we have a value that works, then we can
always choose a natural number larger than it.

ˆ We think of ε as a small positive real number, but we are obliged to

prove it for all ε > 0. But if we can prove it for say 0 < ε < 1
then that’s enough—if N works for a certain ε then it works for all
larger values too. So you can work with a smaller range of ε, such as
0 < ε < 1, if that is most convenient (but it would be a good idea to
mention briefly why this is sufficient).

ˆ It’s really worth becoming comfortable with inequalities and modulus.

In the examples, it was nicer to use the absolute values to write things
like |an − L| < ε, rather than −ε < an − L < ε. If you prefer the second
at the moment, then I recommend practising to get used to the first!

47
 Working directly from the definition is often painful or impractical. Our
next goal is to prove a result that will give a more convenient strategy for
proving convergence in some circumstances.

17 Limits: first key results

The next result is extremely useful in practice! We’ll see a more general
version later, but even this version is strong enough to be useful.

Proposition 24 (Sandwiching, first version). Let (an ) and (bn ) be real se-
quences with 0 ⩽ an ⩽ bn for all n ⩾ 1. If bn → 0 as n → ∞, then an → 0
as n → ∞.

Remark. You might like to draw yourself a diagram to develop your intuition
for what this result says.

Proof. Idea: if N works for bn then it works for an too.

Assume that 0 ⩽ an ⩽ bn for all n, and that bn → 0 as n → ∞.
Take ε > 0.
Since bn → 0, there exists N such that if n ⩾ N then |bn | < ε.
Now if n ⩾ N then 0 ⩽ an ⩽ bn < ε, so |an | < ε.
So an → 0 as n → ∞.

Example. ˆ

1
Claim. 2n
→ 0 as n → ∞.

Proof. We have 2n ⩾ n for n ⩾ 1 (can prove this by induction),

1 1 1
so 0 ⩽ 2n
⩽ n
for n ⩾ 1, and n
→ 0,
1
so by Sandwiching 2n
→ 0 as n → ∞.

48
 ˆ

n cos(n3 + 1)
Claim. Let an = for n ⩾ 1 (we saw this example earlier).
5n2 + 1
Then an → 0 as n → ∞.

Proof. Idea: apply Sandwiching to (|an |).

We have
 n cos(n3 + 1) 
 
0⩽
 ⩽ 1 ⩽ 1
2
5n + 1  5n n

for n ⩾ 1,
1
and n
→ 0 as n → ∞,

so, by Sandwiching, |an | → 0 as n → ∞.

But (looking back at the definition) we see that |an | → 0 if and only if
an → 0.

Here are two key sequences; it will be useful later to have studied them.
(You can also think of them as further worked examples.)

Lemma 25. (i) Take c ∈ R with |c| < 1. Then cn → 0 as n → ∞.

n
(ii) Let an = 2n
for n ⩾ 1. Then an → 0 as n → ∞.

1
Proof. (i) Write |c| = 1+y
where y > 0.

Take ε > 0.
1
Let N = ⌈ yε ⌉ + 1. (When writing this proof, we might leave this line
blank and fill it in at the end!)

Take n ⩾ N .

49
 By Bernoulli’s inequality (since y > 0 and n ⩾ 1) we have (1 + y)n ⩾
1 + ny, so

1 1 1
|cn | = ⩽ ⩽ < ε.
(1 + y)n 1 + ny Ny

So cn → 0 as n → ∞.

n


(ii) Note that if n ⩾ 2 then 2n = (1 + 1)n ⩾ 2
(by the binomial theorem).

Take ε > 0.

Let N = ⌈2 + 2ε ⌉.

For n ⩾ N , we have

n n 2 2
|an − 0| = ⩽ n

= ⩽ < ε.
2n 2
n − 1 N − 1

So an → 0 as n → ∞.

As promised earlier, let’s show that if a sequence converges, then its limit
is unique.

Theorem 26 (Uniqueness of limits). Let (an ) be a convergent sequence.

Then the limit is unique.

Proof. Assume that an → L1 and an → L2 as n → ∞. Aim: L1 = L2 .

Idea: contradiction. If L1 ̸= L2 , then eventually all the terms are really
close to L1 , and also to L2 , and that’s not possible.

50
 Suppose, for a contradiction, that L1 ̸= L2 .
|L1 − L2 |
Let ε = > 0.
2
Since an → L1 as n → ∞, there is N1 such that if n ⩾ N1 then |an −L1 | <
ε.
Also, since an → L2 as n → ∞, there is N2 such that if n ⩾ N2 then
|an − L2 | < ε.
For n ⩾ max{N1 , N2 } we have |an − L1 | < ε and |an − L2 | < ε, so

|L1 − L2 | = |(L1 − an ) + (an − L2 )|

⩽ |L1 − an | + |an − L2 | by the triangle inequality

< 2ε = |L1 − L2 |.

This is a contradiction.
So L1 = L2 .

18 Limits: modulus and inequalities

Proposition 27. Let (an ) be a convergent sequence. Then (|an |) also con-
verges. Moreover, if an → L as n → ∞ then |an | → |L| as n → ∞.

51
Proof. Say an → L as n → ∞.
Take ε > 0.
Then there is N such that if n ⩾ N then |an − L| < ε.
Now if n ⩾ N then, by the Reverse Triangle Inequality, we have

||an | − |L|| ⩽ |an − L| < ε.

So (|an |) converges, and |an | → |L| as n → ∞.

Remark. We could instead have proved Proposition 27 using the Sandwich-

ing Lemma, since an → L as n → ∞ if and only if |an − L| → 0 as n → ∞
(check this using the definition of convergence).

Now let’s think about inequalities. If (an ) is a convergent sequence and

an > 0 for all n, then what can we say about the limit? It’s not the case that
1
the limit must be positive. For example, if an = n
then an > 0 for all n but
an → 0. But it’s hard to see how a sequence of positive terms could have a
negative limit.

Proposition 28 (Limits preserve weak inequalities). Let (an ) and (bn ) be

real sequences, and assume that an → L and bn → M as n → ∞, and that
an ⩽ bn for all n. Then L ⩽ M .

Remark. ˆ This includes the special case where an = 0 for all n: Propo-
sition 28 says that if bn ⩾ 0 for all n, and bn → M as n → ∞, then
M ⩾ 0. (This is because the constant sequence 0, 0, 0, . . . certainly
converges to 0.)

ˆ A common mistake is to use the non-result that limits preserve strict

inequalities. As we’ve seen, this is not true. Please try not to do this!

52
Proof. Suppose, for a contradiction, that it is not the case that L ⩽ M , so
(by trichotomy) L > M .

Let ε = 21 (L − M ) > 0.
Since an → L as n → ∞, there is N1 such that if n ⩾ N1 then |an −L| < ε.
Since bn → M as n → ∞, there is N2 such that if n ⩾ N2 then |bn − M | <
ε.
Now for n ⩾ max{N1 , N2 } we have an > L − ε and bn < M + ε,
so L − ε < an ⩽ bn < M + ε,
so L − M < 2ε = L − M . This is a contradiction.

We saw a sandwiching result earlier. Here is a generalisation.

Proposition 29 (Sandwiching). Let (an ), (bn ) and (cn ) be real sequences

with an ⩽ bn ⩽ cn for all n ⩾ 1. If an → L and cn → L as n → ∞, then
bn → L as n → ∞.

Proof. Take ε > 0.

Since an → L as n → ∞, there is N1 such that if n ⩾ N1 then |an −L| < ε.
Since cn → L as n → ∞, there is N2 such that if n ⩾ N2 then |cn −L| < ε.
Then for n ⩾ max{N1 , N2 } we have L − ε ⩽ an ⩽ bn ⩽ cn ⩽ L + ε,
so |bn − L| < ε.
So bn → L as n → ∞.

53
19 Bounded and unbounded sequences
Definition. Let (an ) be a sequence. We say that (an ) is bounded if the set
{an : n ⩾ 1} is bounded, that is, there is M such that |an | ⩽ M for all n ⩾ 1.
If (an ) is not bounded then we say that it is unbounded.

Proposition 30 (A convergent sequence is bounded). Let (an ) be a conver-

gent sequence. Then (an ) is bounded.

Remark. Proposition 30 tells us that if (an ) is unbounded then (an ) diverges.

Proof.
Assume that an → L as n → ∞.
Then (taking ε = 1) there is N such that if n ⩾ N then |an − L| < 1 so

|an | = |(an − L) + L| ⩽ |an − L| + |L| < 1 + |L|.

Let M = max{|a1 |, |a2 |, . . . , |aN |, |L| + 1}.

Then |an | ⩽ M for all n ⩾ 1.

Remark. ˆ As remarked earlier, if (an ) is unbounded then (an ) diverges.

So, for example, (2n ) diverges.

ˆ Unboundedness is not the same as divergence. The converse of Propo-

sition 30 is not true. A bounded sequence can diverge. For example,
let an = (−1)n . Then |an | ⩽ 1 for all n ⩾ 1, so (an ) is bounded.

Claim. ((−1)n ) does not converge.

54
 Proof. Suppose, for a contradiction, that (−1)n → L as n → ∞.

Then (taking ε = 1) there is N such that if n ⩾ N then |(−1)n −L| < 1.

In particular (n = 2N ) we have |L − 1| < 1 so L > 0,

and (n = 2N + 1) we have |L + 1| < 1 so L < 0.

This is a contradiction.

What would it mean to say that a sequence tends to infinity?

Definition. Let (an ) be a real sequence. We say that (an ) tends to infinity
as n → ∞ if

∀M ∈ R ∃N ∈ N such that ∀n ⩾ N, an > M.

In this case we write an → ∞ as n → ∞.

Similarly, we say that (an ) tends to negative infinity as n → ∞ if

∀M ∈ R ∃N ∈ N such that ∀n ⩾ N, an < M.

In this case we write an → −∞ as n → ∞.

55
Remark. This is a separate definition from our earlier definition of conver-
gence, and ∞ is definitely not a real number. Results about convergence to
a real number L cannot just be applied by ‘taking L = ∞’—this would be
highly illegal!

Example. ˆ Let an = n2 − 6n for n ⩾ 1.

Claim. an → ∞ as n → ∞.

Proof. Fix M > 0. (It suffices to prove the result for M > 0.)

We want N such that if n ⩾ N then n2 − 6n ⩾ M

but n2 − 6n = (n − 3)2 − 9

so we are done if (n − 3)2 ⩾ M + 9

√
that is, we are done if n − 3 ⩾ M + 9
√
Let N = ⌈4 + M + 9⌉.
√
If n ⩾ N , then n − 3 ⩾ M + 9 > 0,

so (n − 3)2 ⩾ M + 9,

so n2 − 6n ⩾ M .

So an → ∞ as n → ∞.

0 if n prime

ˆ Let an =
n otherwise.


Then (an ) does not tend to infinity, because there are infinitely many
primes: for any N ∈ N, there is a prime n with n > N , and then
an = 0.

Lemma 31. (i) If α < 0, then nα → 0 as n → ∞.

56
 (ii) If α > 0, then nα → ∞ as n → ∞.

Proof. (i) Take ε ∈ (0, 1). We have

nα < ε

⇔ eα log n < ε

⇔ α log n < log ε

1
⇔ log n > log ε (note α < 0)
α
1
⇔ n > e α log ε

1
so we can take N = 1 + ⌈e α log ε ⌉.

(ii) Take M > 0. We have

nα > M

⇔ eα log n > M

⇔ α log n > log M

1
⇔ log n > log M (note α > 0)
α
1
⇔ n > e α log M

1
so we can take N = 1 + ⌈e α log M ⌉.

Lemma 32. Let c ∈ R>0 .

(i) If c < 1, then cn → 0 as n → ∞.

(ii) If c = 1, then cn → 1 as n → ∞.

(iii) If c > 1, then cn → ∞ as n → ∞.

Proof. (i) This was Lemma 25.

57
 (ii) This is clear from the definition of convergence.

(iii) Exercise. (You could adapt the argument from (i), or use logarithms.)

20 Complex sequences
A lot of the theory we have just seen applies equally to complex sequences,
but there are some differences. Let’s spell out the definition of convergence
explicitly.

Definition. Let (zn ) be a complex sequence, let L ∈ C. We say that (zn )

converges to L as n → ∞ if

∀ε > 0 ∃N ∈ N such that ∀n ⩾ N, |zn − L| < ε.

Remark. ˆ If (zn ) tends to a limit, then this limit is unique, exactly as

in Theorem 26.

ˆ We can have a sort of sandwiching for complex sequences, if we use the

modulus. If (zn ) and (wn ) are complex sequences, and |wn | ⩽ |zn | for
all n ⩾ 1, and zn → 0 as n → ∞, then wn → 0 as n → ∞.

Given a complex sequence (zn ), there are two associated real sequences
(Re(zn )) and (Im(zn )). The next result relates convergence of (zn ) to con-
vergence of (Re(zn )) and (Im(zn )).

Theorem 33 (Convergence of complex sequences). Let (zn ) be a complex

sequence. Write zn = xn + iyn with xn , yn ∈ R, so that (xn ) and (yn )
are real sequences. Then (zn ) converges if and only if both (xn ) and (yn )
converge. Moreover, in the case where (zn ) converges, we have lim zn =
n→∞
lim xn + i lim yn .
n→∞ n→∞

58
Proof. Exercise.

in
Example. ˆ Let zn = n
. Then |zn | = 1
n
→ 0 as n → ∞ so zn → 0 as
n → ∞.

ˆ Let zn = (1 + i)n . The sequence is

1 + i, 2i, −2 + 2i, −4, −4 − 4i, −8i, 8 − 8i, 16, . . . .

The real parts are 1, 0, −2, −4, −4, 0, 8, 16, . . . —this sequence doesn’t
converge, and hence neither does (zn ).

21 Subsequences
We can make a good informal guess as to what we mean by a subsequence.
Let (an )n⩾1 be a sequence. Then a subsequence is a sequence (br )r⩾1 ,
where each br is in (an ), and the terms are in the right order.

Example. Let an = n for n ⩾ 1. The following are subsequences of (an ).

ˆ 2, 4, 6, 8, . . . — the subsequence (a2n )

ˆ 2, 4, 8, 16, . . . — the subsequence (a2n )

The following are not subsequences of (an ).

ˆ 6, 4, 8, . . . — the terms are not in the right order

ˆ 2, 4, 0, . . . — not all the terms are in (an )

ˆ 1, 2, 3, . . . , 2020 — finite so not a sequence.

Now let’s give a formal definition of a subsequence.

59
Definition. Let (an )n⩾1 be a sequence. A subsequence (br )r⩾1 of (an )n⩾1 is
defined by a function f : N → N such that f is strictly increasing (if p < q
then f (p) < f (q)), and br = af (r) for r ⩾ 1.
We often write f (r) as nr . Then n1 < n2 < n3 < · · · is a strictly
increasing sequence of natural numbers, and br = anr so the sequence (br )
has terms an1 , an2 , an3 , . . . .

Remark. ˆ Formally, (an ) corresponds to a function α : N → R or

α : N → C. Then a subsequence of (an ) corresponds to a function
α ◦ f , where f : N → N is strictly increasing.

ˆ Subscripts are ‘dummy variables’. We can write (an ) as (ar ) or (am )

or (aα ) or (ax ). It is conventional to use a letter close to n in the
alphabet, to help us remember that it is a natural number. We can
use any letter for the subscripts in the subsequence (br ), except that if
we write our original sequence as (an ) then we should avoid using n for
the subsequence too.

ˆ It’s sometimes useful to know that nr ⩾ r for r ⩾ 1. (Exercise: prove

this inequality, using induction.)

Proposition 34 (Subsequences of a convergent sequence). Let (an ) be a

sequence. If (an ) converges, then every subsequence (anr ) of (an ) converges.
Moreover, if an → L as n → ∞ then every subsequence also converges to L.

Remark. So if (an ) is a sequence, and it has two subsequences that tend

to different limits, then (an ) does not converge. This follows from Proposi-
tion 34, and can be a useful strategy for showing that a sequence does not
converge.

Proof. Assume that (an ) converges to L.

60
 Let (anr ) be a subsequence of (an ).
Take ε > 0.
Since an → L, there is N such that if n ⩾ N then |an − L| < ε.
If r ⩾ N , then nr ⩾ r ⩾ N (see remark before this result),
so |anr − L| < ε.
So anr → L as r → ∞.

0

if n is prime
Example. Let an = .
1 + 1
otherwise

n

Claim. (an ) does not converge.

Proof. Idea: the subsequence of terms with prime subscripts tends to 0,

and the subsequence of terms with non-prime subscripts tends to 1, so (an )
doesn’t converge.
Let the primes be p1 < p2 < p3 < · · · . Let P = {p1 , p2 , p3 , . . . }.
Note that there are infinitely many primes, so (apr )r⩾1 is a subsequence.
We have apr = 0 for all r ⩾ 1, so apr → 0 as r → ∞.
Let the elements of N \ P be n1 < n2 < n3 < · · · .
Note that there are infinitely many non-primes, so (anr )r⩾1 is a subse-
quence.
1
We have anr = 1 + nr
for r ⩾ 1, and so we see that anr → 1 as r → ∞.
So (an ) has subsequences that converge to different limits, so, by Propo-
sition 34, (an ) does not converge.

22 Algebra of Limits — part one

Example. This is an unofficial example. We’ll return to it once we’ve proved
some results.

61
 7n5 − n sin(n2 + 5n) + 3
Let an = .
4n5 − 3n2 + n + 2
What can we say about (an )?
Intuitively...

- the numerator grows like 7n5 — the other terms are much smaller for
large n, which is all we care about;

- the denominator grows like 4n5

7
so we might conjecture that an → 4
as n → ∞.

To prove this (and lots more!), we’ll prove a bunch of results that are
extremely useful in practice. Collectively, these are known as the ‘Algebra of
Limits’, and we’ll quote “by AOL” in arguments.

Theorem 35 (Algebra of Limits, part 1). Let (an ) and (bn ) be sequences
with an → L and bn → M as n → ∞. Let c be a constant.

(i) (constant) If an = c, so (an ) is a constant sequence, then an → c as

n → ∞.

(ii) (scalar multiplication) The sequence (can ) converges, and can → cL as

n → ∞.

(iii) (addition) The sequence (an + bn ) converges, and an + bn → L + M as

n → ∞.

(iv) (subtraction) The sequence (an − bn ) converges, and an − bn → L − M

as n → ∞.

(v) (modulus) The sequence (|an |) converges, and |an | → |L| as n → ∞.

Proof. (i) This is immediate from the definition.

62
(ii) If c = 0, then we’re done by (i). So assume that c ̸= 0.

Take ε > 0.

Since an → L, there is N such that if n ⩾ N then |an − L| < ε.

Now if n ⩾ N then |can − cL| = |c||an − L| < |c|ε.

So (can ) converges to cL.

OR...

Take ε > 0.
ε
Since an → L, there is N such that if n ⩾ N then |an − L| < |c|
.

Now if n ⩾ N then |can − cL| = |c||an − L| < ε.

So (can ) converges to cL.

(iii) Take ε > 0.

Since an → L as n → ∞ there is N1 such that if n ⩾ N1 then |an −L| <

ε.

Since bn → M as n → ∞ there is N2 such that if n ⩾ N2 then

|bn − M | < ε.

Let N = max{N1 , N2 }. If n ⩾ N , then |an − L| < ε and |bn − M | < ε,

so

|(an + bn ) − (L + M )| ⩽ |an − L| + |bn − M | (by triangle inequality)

< 2ε.

So (an + bn ) converges to L + M .

(iv) This follows from (ii) and (iii).

(v) This was Proposition 27.

63
Remark. In (iii), I ended up showing that we can make |(an +bn )−(L+M )|
less than 2ε by going far enough along the sequence. But the definition says
ε, not 2ε, so isn’t this a problem?
Well, no, it’s not a problem. We need to show that we can make |(an +
bn ) − (L + M )| less than any positive real number — and that’s what we’ve
done. The important thing is that 2 was a (positive) constant: it didn’t
depend on n.
ε
We could instead have chosen N1 and N2 corresponding to 2
(so if n ⩾ N1
ε
then |an − L| < 2
and similarly for bn ), and then we’d have got ε at the end.
But if I’d done that then it might have seemed more mysterious: you might
have wondered “how would I have known to choose 2ε ?”
In practice, sometimes I doodle on scrap paper and consequently know
what to choose at the start, and sometimes I just work through and see
what happens, and if I get 2ε or 1000ε at the end then it doesn’t matter. I
illustrated these two alternative approaches in (ii) — but really they’re the
same, and both are fine.

Example.
n cos(n3 + 1)
 
1 n 1
Claim. Let an = n + 1 + (−1) √ + . Then an → 1 as
2 n 5n2 + 1
n → ∞.
1 1
Proof. We showed earlier that n
→ 0 and 1 + (−1)n √ → 1 and also
2 n
n cos(n3 + 1)
→ 0 as n → ∞ (see Section 16).
5n2 + 1
So, by AOL, (an ) converges, and an → 0 + 1 + 0 = 1 as n → ∞.

Example.

n
Claim. Let an = (−1)n + 2n
for n ⩾ 1. Then (an ) does not converge.

64
Proof. Suppose, for a contradiction, that (an ) converges.
Note that 2nn converges (this was an earlier example).

n
So, by AOL, the sequence with nth term (−1)n = an − 2n
converges.
But we showed earlier that ((−1)n ) does not converge (or we could now
note that it has subsequences tending to different limits 1 and −1). This is
a contradiction.

23 Algebra of Limits — part two

Theorem 36 (Algebra of Limits, part 2). Let (an ) and (bn ) be sequences
with an → L and bn → M as n → ∞.

(vi) (product) The sequence (an bn ) converges, and an bn → LM as n → ∞.

 
1 1 1
(vii) (reciprocal) If M ̸= 0, then the sequence bn
converges, and bn
→ M

as n → ∞.
 
an an L
(viii) (quotient) If M ̸= 0, then the sequence bn
converges, and bn
→ M

as n → ∞.
   
1 an
Remark. You might wonder whether the sequences bn
and bn
in (vii)
and (viii) are defined. This is a good question. The answer is that — as we’ll
show in the proof — if M ̸= 0 then a tail of (bn ) has all its terms nonzero,
   
and hence there’s a tail of b1n that exists, and similarly for abnn . When
we talk about convergence of these sequences, it’s enough to consider a tail.

Proof. (vi) We’re going to want to study

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

⩽ |an ||bn − M | + |M ||an − L|

65
 — this use of the triangle inequality can help us to see how to proceed.

Take ε > 0. We may assume that ε < 1.

Since an → L, there is N1 such that if n ⩾ N1 then |an − L| < ε.

Since bn → M , there is N2 such that if n ⩾ N2 then |bn − M | < ε.

Let N = max{N1 , N2 }.

If n ⩾ N , then |an − L| < ε and |bn − M | < ε and |an | < |L| + ε, so

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

⩽ |an ||bn − M | + |M ||an − L|

< (|L| + ε) · ε + |M | · ε

< ε(1 + |L| + |M |).

Since 1 + |L| + |M | is constant, this is enough to show that (an bn )

converges, and the limit is LM .

(vii) Assume that M ̸= 0.

 
1 1 
Idea: (1) eventually bn is close to M , so can’t be 0. (2)  −  =
bn M
|bn − M |
– eventually the numerator is small, and |bn | is close to |M |.
|M ||bn |
Take ε > 0.

Since bn → M and |M | > 0, there is N1 such that if n ⩾ N1 then

|M |
|bn − M | < 2
, so (by the Reverse Triangle Inequality)

|M |
|bn | ⩾ ||bn + (M − bn )| − |M − bn || > > 0.
2

So the tail (bn )n⩾N1 has all terms nonzero, so we can consider the se-
 
quence b1n .
n⩾N1

Also, there is N2 such that if n ⩾ N2 then |bn − M | < ε.

66
 Let N = max{N1 , N2 }. If n ⩾ N , then
 
 − 1  = |M − bn | < ε · 2 .
1 
 bn M  |M ||bn | |M | |M |
 
Since |M2|2 is a positive constant, this shows that b1n converges,
n⩾N1
1
and the limit is M
.

(viii) This follows from (vi) and (vii).

7n5 − n sin(n2 + 5n) + 3

Example. Let an = (we saw this example at the
4n5 − 3n2 + n + 2
start of Section 22).
7
Claim. an → 4
as n → ∞.

Proof. Idea: the important terms (for large n) are 7n5 and 4n5 .
We have
1 3
7− n4
sin(n2 + 5n) + n5
an = .
4 − n33 + n14 + n25
Now
 
1 2
 1 1
0 ⩽  4 sin(n + 5n) ⩽ 4 ⩽
n n n
1 1
and n
→ 0, so by Sandwiching n4
sin(n2 + 5n) → 0, and several other terms
also tend to 0 (eg by Sandwiching),
so, by AOL, (an ) converges, and
7−0+0 7
an → =
4−0+0+0 4
as n → ∞.

Proposition 37 (Reciprocals and infinite/zero limits). Let (an ) be a se-

quence of positive real numbers. Then an → ∞ as n → ∞ if and only if
1
an
→ 0 as n → ∞.

Proof. Exercise (using the definitions).

67
24 Orders of magnitude
When we’re studying a sequence (an ), it can be really useful to develop some
intuition about the behaviour of an for large n, in order to make a conjecture
about the convergence (or otherwise) of the sequence, and to select a proof
strategy. (This is what we did in the example at the end of the last section,
for example.)
8n2 + 1000000n + 1000000
Example. ˆ Let an = .
14n6 + n3 + n
Intuitively, the key term in the numerator is 8n2 , and the key term in
the denominator is 14n6 . Even with the amusingly large coefficients in
the numerator, when n is large these terms will be much smaller than
8n2 .
8
So it feels like the sequence grows roughly like 14n4
, so should tend to
0.

We can formalise this using AOL. Dividing through top and bottom by
n6 (since this is the key term), we get
8
n4
+ 1000000
n5
+ 1000000
n6 0+0+0
an = 1 1 → =0
14 + n3 + n5
14 + 0 + 0

as n → ∞.

ˆ We showed in Lemma 25 that n

2n
→ 0 as n → ∞.

This is an example of the idea that ‘exponentials beat polynomials’.

But while ‘exponentials beat polynomials’ is a useful slogan for intu-
ition, it is not suitable for rigorous proofs!

ˆ We’ve seen a couple of examples where we used that | cos x| ⩽ 1 and

| sin x| ⩽ 1 for all x — this can be useful.

68
 ˆ We’ll show in the next section that log n
n
→ 0 as n → ∞. Intuitively,
polynomials grow faster than logarithms.

Definition. Let (an ) and (bn ) be sequences. We write an = O(bn ) as n → ∞

if there is a constant C ∈ R>0 and there is N such that if n ⩾ N then
|an | ⩽ C|bn |. This is ‘big O’ notation.
If bn ̸= 0 for all n (or all sufficiently large n), then we write an = o(bn ) as
an
n → ∞ if bn
→ 0 as n → ∞. This is ‘little o’ notation.

Remark. ˆ Sandwiching tells us that if an = O(bn ) and bn → 0 as

n → ∞ then an → 0 as n → ∞.

ˆ Big O and little o notation give us precise ways to make precise state-
ments about comparative rates of growth of sequences. Please use them
precisely!

Example. This example is in a Moodle quiz. Before you read on to the next
section, please go to the Moodle course page for Analysis I, and try the quiz
for section 24 (it’s a short multiple choice quiz).

25 Monotonic sequences
Definition. Let (an ) be a real sequence.

ˆ We say that (an ) is monotonic increasing, or monotone increasing, or

increasing, if an ⩽ an+1 for all n.

ˆ We say that (an ) is strictly increasing if an < an+1 for all n.

ˆ We say that (an ) is monotonic decreasing, or monotone decreasing, or

decreasing, if an ⩾ an+1 for all n.

69
 ˆ We say that (an ) is strictly decreasing if an > an+1 for all n.

ˆ We say that (an ) is monotonic, or monotone, if it is increasing or de-

creasing.

Example. Notice that a constant sequence is both increasing and decreasing.

This might seem counterintuitive!

We know that a convergent sequence is bounded. What can we say about

a bounded monotone sequence?

Theorem 38 (Monotone Sequences Theorem). Let (an ) be a real sequence.

(i) If (an ) is increasing and bounded above, then (an ) converges.

(ii) If (an ) is decreasing and bounded below, then (an ) converges.

Remark. ˆ So ‘a bounded monotone sequence converges’.

ˆ The result applies to tails of sequences too: if (an ) has a tail that is
monotone and bounded, then it converges.

Proof. (i) Assume that (an ) is increasing and bounded above.

70
 Idea: {an : n ⩾ 1} has a supremum, and (an ) converges to this.

The set S = {an : n ⩾ 1} is non-empty and bounded above, so, by

Completeness, it has a supremum.

Take ε > 0.

By the Approximation Property, there is N such that sup S −ε < aN ⩽

sup S.

If n ⩾ N , then sup S − ε < aN ⩽ an ⩽ sup S,

so |an − sup S| < ε.

So (an ) converges, and an → sup S as n → ∞.

(ii) If (an ) is decreasing and bounded below, then (−an ) is increasing and
bounded above, so (ii) follows from (i).

Lemma 39. Let (an ) be a real sequence that is increasing and not bounded
above. Then an → ∞ as n → ∞.

Proof. Take M ∈ R.
Since (an ) is not bounded above, there is N such that aN > M .
Then, since (an ) is increasing, if n ⩾ N then an ⩾ aN > M .

71
  n
1
Example. Let an = 1 + .
n
On Sheet 1, you proved that (an ) is increasing and that (an ) is bounded
above (by 3). So, by the Monotone Sequences Theorem, (an ) converges. Say
an → L as n → ∞. Then, since limits preserve weak inequalities, we see that
2 ⩽ L ⩽ 3.
(Secretly, we know more about L, but that’s strictly unofficial for now.)
√
Example. Let c ⩾ 0. In this example, we’ll show that c exists. (This
√
generalises earlier work on 2, and usesa different  strategy.)
1 c
Define (an ) by a1 = 1 and an+1 = an + for n ⩾ 1.
2 an
This is a legitimate definition, since (by induction) an ̸= 0 for n ⩾ 1.

Claim. (an ) converges, and if an → L then L2 = c.

Proof. ˆ (an ) bounded below:

by a straightforward induction argument, we have an > 0 for all n.

ˆ study a2n − c:

for n ⩾ 1, we have
 2
1 c
a2n+1 −c= an + −c
4 an
c2
 
1 2
= an + 2c + 2 − c
4 an
c2
 
1 2
= an − 2c + 2
4 an
 2
1 c
= an −
4 an
⩾ 0,

so a2n+1 ⩾ c for n ⩾ 1.

72
 ˆ (an )n⩾2 decreasing:

for n ⩾ 2, we have
   
1 c 1 c 1
an+1 − an = an + − an = − an = (c − a2n ) ⩽ 0,
2 an 2 an 2an

so an+1 ⩽ an for n ⩾ 2.
So, by the Monotone Sequences Theorem, (an ) converges.
Say an → L as n → ∞.
Then also an+1 → L as n → ∞ (it’s a tail of the sequence).
But if L ̸= 0 then
 
1 c 1 c
an+1 = an + → L+
2 an 2 L

by AOL.
1 c


Since limits are unique, we have L = 2
L+ L
,
so, rearranging, L2 = c.
Also, we have an > 0 for all n, and limits preserve weak inequalities, so
L ⩾ 0.
√ √
So c exists (L = c).
In the case that L = 0, since limits preserve weak inequalities and a2n ⩾ c
for n ⩾ 2 we have c ⩽ 0, so c = 0 and L2 = c.

log n
Lemma 40. We have n
→ 0 as n → ∞.

log n
Proof. Let an = n
.
Then an ⩾ 0 for all n, so (an ) is bounded below.
Also, by properties of log we see that (an )n⩾100 is decreasing.
log n
So, by the Monotone Sequences Theorem, (an ) converges. Say n
→L
as n → ∞.
Since limits preserve weak inequalities, we have L ⩾ 0.

73
 Now
log(2n) log 2 + log n L
a2n = = →0+
2n 2n 2
by AOL,
but also (a2n ) is a subsequence of (an ) so a2n → L as n → ∞.
L
So, by uniqueness of limits, 2
= L, so L = 0.

26 Convergent subsequences
Theorem 41 (Scenic Viewpoints Theorem). Let (an ) be a real sequence.
Then (an ) has a monotone subsequence.

Proof. Idea: consider the ‘peaks’ of the sequence.

Let V = {k ∈ N : if m > k then am < ak }. (The elements of V are

‘peaks’ or ‘scenic viewpoints’: if k ∈ V then ak is higher than all subsequent
terms.)
Case 1: V is infinite.
Say the elements of V are k1 < k2 < · · · .
Then (akr )r is a subsequence of (an )
and it is monotone decreasing (if r < s then kr < ks so akr > aks ).
Case 2: V is finite.
Then there is N such that if k ∈ V then k < N .

74
 Let m1 = N . Then m1 ̸∈ V so there is m2 > m1 with am2 ⩾ am1 .
Also, m2 ̸∈ V so there is m3 > m2 with am3 ⩾ am2 .
Continuing inductively, we construct m1 < m2 < m3 < · · · such that
am1 ⩽ am2 ⩽ am3 ⩽ · · · .
Then (amr )r is an increasing subsequence of (an ).

Theorem 42 (Bolzano-Weierstrass Theorem). Let (an ) be a bounded real

sequence. Then (an ) has a convergent subsequence.

Proof. By the Scenic Viewpoints Theorem, (an ) has a monotone subsequence.

This monotone subsequence is bounded (because the whole sequence is),
so by the Monotone Sequences Theorem (Theorem 38) it converges.

Remark. ˆ This proof of the Bolzano-Weierstrass Theorem was very

short, because we did all the work in the Monotone Sequences Theorem
and Scenic Viewpoints Theorem! I have another favourite proof of
Bolzano-Weierstrass. I’ve turned it into a quiz ‘proof sorter’ activity
on Moodle.

ˆ The Monotone Sequences Theorem and Scenic Viewpoints Theorem

don’t make sense for complex sequences. But Bolzano-Weierstrass po-
tentially could . . .

Corollary 43 (Bolzano-Weierstrass Theorem for complex sequences). Let

(zn ) be a bounded complex sequence. Then (zn ) has a convergent subsequence.

Proof. Study real and imaginary parts, and repeatedly pass to subsequences
Write zn = xn + iyn where xn , yn ∈ R.
Say (zn ) is bounded by M , so |zn | ⩽ M for all n.
Then (xn ) and (yn ) are also bounded by M , and they are real sequences.
By Bolzano-Weierstrass, (xn ) has a convergent subsequence, say (xnr )r .

75
 Now (ynr )r is a bounded real sequence, so by Bolzano-Weierstrass it has
a convergent subsequence, say (ynrs )s .
Note that (xnrs )s is a subsequence of the convergent sequence (xnr )r and
hence converges.
So, by Theorem 33, (znrs )s converges (since its real and imaginary parts
converge).

27 Cauchy sequences
Example. Let (an ) be a convergent sequence.
Then an+1 − an → 0 as n → ∞.
We can prove this directly from the definition (with the triangle inequal-
ity), or using tails and the Algebra of Limits.
But it is not the case that if an+1 −an → 0 as n → ∞ then (an ) converges.
√
For example, consider an = n. Certainly (an ) does not converge. But
√ √ (n + 1) − n 1
an+1 − an = n+1− n= √ √ =√ √ →0
n+1+ n n+1+ n
as n → ∞.
Nonetheless, intuitively it seems that if eventually all the terms of a se-
quence are bunched up close together then the sequence might converge.

Definition. Let (an ) be a sequence. We say that (an ) is a Cauchy sequence

if

∀ε > 0 ∃N ∈ N such that ∀m, n ⩾ N |an − am | < ε.

Remark. Note that this definition makes sense for complex sequences as
well as for real sequences.

Proposition 44. Let (an ) be a convergent sequence. Then (an ) is Cauchy.

76
Proof. Say an → L as n → ∞.
Take ε > 0.

Since an → L, there is N such that if n ⩾ N then |an − L| < 2ε .

ε
Take m, n ⩾ N . Then |am − L| < 2
and |an − L| < 2ε ,
so, by the triangle inequality,

|am − an | = |(am − L) + (L − an )|

⩽ |am − L| + |an − L| < ε.

So (an ) is Cauchy.

Proposition 45. Let (an ) be a Cauchy sequence. Then (an ) is bounded.

Proof. Idea: use a similar strategy to Proposition 30, where we showed that
a convergent sequence is bounded.
Since (an ) is Cauchy, there is (applying the definition with ε = 1) N such
that if m, n ⩾ N then |am − an | < 1.
Now for n ⩾ N we have |an − aN | < 1,
so |an | = |(an − aN ) + aN | ⩽ 1 + |aN |.
Let K = max{|a1 |, |a2 |, . . . , |aN −1 |, 1 + |aN |}.
Then |an | ⩽ K for all n ⩾ 1.
So (an ) is bounded.

77
Proposition 46. Let (an ) be a Cauchy sequence. Suppose that the subse-
quence (anr )r converges. Then (an ) converges.

Proof. Idea: eventually all the terms of (anr ) are really close to the limit L,
and eventually all the terms of (an ) are really close to terms in the subse-
quence and hence also really close to L.
Say that anr → L as r → ∞.
Take ε > 0.
Then there is N1 such that if r ⩾ N1 then |anr − L| < 2ε .
Also, since (an ) is Cauchy there is N2 such that if m, n ⩾ N2 then |am −
an | < 2ε .
Let N = max{N1 , N2 }.
ε
Let r = N . Then nr ⩾ r ⩾ N1 so |anr − L| < 2

and if n ⩾ N then n, nr ⩾ N2 so |anr − an | < 2ε ,

so

|an − L| = |(an − anr ) + (anr − L)|

⩽ |an − anr | + |anr − L| < ε.

So an → L as n → ∞.

The following result is really useful! We’ll use it in later sections.

Theorem 47 (Cauchy Convergence Criterion). Let (an ) be a sequence. Then

(an ) converges if and only if (an ) is Cauchy.

Proof. (⇒) This was Proposition 44.

(⇐) Assume that (an ) is Cauchy.
Then (an ) is bounded, by Proposition 45,
so by the Bolzano-Weierstrass Theorem (Theorem 42), (an ) has a conver-
gent subsequence, say (anr ).

78
 Then, by Proposition 46, (an ) converges.

Remark. One reason this is so useful is that it gives us a way to show that
a sequence converges without needing to know in advance what the limit is.

28 Convergence for series

Example. Here are some informal examples of series to set the scene.
∞
X
ˆ For suitable r, we can consider the geometric series rn (you might
n=0
already have some ideas about this series).

ˆ Decimal expansions. When we write 1

9
= 0.111 . . . or 1
9
= 0.1̇, we mean
∞
X 1
.
n=1
10n
∞
X 1
ˆ We’ll define e = .
n=0
n!
∞
X xn
ˆ We’ll define e =
x
.
n=0
n!
We’ll revisit these examples once we’ve explored some theory.

Definition. Let (ak ) be a sequence. For n ⩾ 1, let

n
X
s n = a1 + a2 + · · · + an = ak .
k=1
∞
X
This is called a partial sum of the series ak .
k=1
∞
X
We say that the series ak converges if the sequence (sn ) of partial
k=1
∞
X
sums converges. If sn → s as n → ∞, then we write ak = s.
k=1
∞
X
If (sn ) does not converge, then we say that ak diverges.
k=1

79
Remark. ˆ So convergence of series is really a special case of convergence
of sequences, rather than a new concept.

ˆ A series is a limit.
X ∞
X
ˆ We might sometimes write
P
ak or even ak instead of ak .
k⩾1 k=1

n
X
ˆ It would be highly illegal to write something like an — we need to
n=1
use different letters for quantities that can be different. That’s why
I’ve put k as the dummy variable in the sums, because it isn’t n (and
is still a good letter for a natural number).

ˆ It’s sometimes helpful to note that (with the notation above) ak =

sk − sk−1 for k ⩾ 2.

Example. Geometric series. Take z ∈ C. Let ak = z k for k ⩾ 0, and let

n
X
sn = z k . Then for n ⩾ 0 we have
k=0

 1−zn+1

if z ̸= 1
1−z
sn =
n + 1

if z = 1.

∞
1 X 1
If |z| < 1, then sn → as n → ∞, so z n exists and equals .
1−z n=0
1−z
If |z| ⩾ 1, then (sn ) does not converge and so the series diverges. (One way
to see that (sn ) does not converge is to note that if |z| ⩾ 1 then sn − sn−1 =
an = z n does not tend to 0 as n → ∞.)

Remark. Notice how we worked with partial sums, and determined that the
P n
limit exists before writing down z .
1
Example. A telescoping series. Let ak = for k ⩾ 1.
k(k + 1)

80
 n
X 1
Let sn = .
k=1
k(k + 1)
Then
n  
X 1 1
sn = −
k=1
k k+1
       
1 1 1 1 1 1 1
= 1− + − + − + ··· + −
2 2 3 3 4 n n+1
1
=1− → 1 as n → ∞
n+1
∞
X 1
so exists and equals 1.
k=1
k(k + 1)

Remark. Notice how we worked with partial sums, and determined that the
P 1
limit exists before writing down k(k+1)
.
n
X
k
Example. Let ak = (−1) , let sn = (−1)k .
k=1
Then

−1

if n odd
sn =
0

if n even.

∞
X
So (sn ) does not converge, that is, (−1)k diverges.
k=1

Remark. Notice how we worked with partial sums, not the series, and in
fact the limit doesn’t exist. We definitely didn’t write anything dodgy like
∞
X
(−1)k = (−1 + 1) + (−1 + 1) + · · · = 0,
k=1

because this would be wrong.

81
29 Series: first results and a first test for con-
vergence
∞
X ∞
X
Proposition 48. Consider the series ak . If ak converges, then ak →
k=1 k=1
0 as k → ∞.

Remark. So one way to show that a series diverges is to show that ak ̸→ 0.

This is disproportionately useful!
n
X
Proof. Let sn = ak . Then (sn ) converges by assumption. Say sn → s as
k=1
n → ∞.
Then also sn−1 → s as n → ∞,
so by AOL an = sn − sn−1 → s − s = 0 as n → ∞.
P
Remark. Proposition 48 does not say that if ak → 0 as k → ∞ then ak
converges. That’s because this is false. For example . . .
n ∞
X 1 X 1
Example. For n ⩾ 1, let sn = . The series is called the harmonic
k=1
k k=1
k
series.

Claim. The harmonic series diverges.

Proof. Idea

82
Show that (sn ) is not Cauchy.
Consider |s2n+1 − s2n |. We have
n+1
2X
1 1 1
|s2n+1 − s2n | = ⩾ 2n · n+1 = .
k=2n +1
k 2 2

So (sn ) is not Cauchy, so (sn ) does not converge.

Remark. It is interesting to study the partial sums of the harmonic series.

We’ll do this in more detail in a future section.

Proposition 49. Let (ak ) be a sequence of non-negative real numbers, and let
n
X X∞
sn = ak . Suppose that (sn ) is bounded. Then the series ak converges.
k=1 k=1

Proof. Since ak ⩾ 0 for all k, we see that (sn ) is increasing.

So (sn ) is monotone and bounded, so by the Monotone Sequences Theo-
∞
X
rem (Theorem 38) it converges, that is, ak converges.
k=1

Remark. Proposition 49 is a result that can be useful in practice for showing

that a series converges. One particularly frequent way to apply it is to show
that the partial sums are bounded by comparing with another series that we
already know converges. We’ll record that as a separate result, but really it’s
just a special case of Proposition 49, which is in turn just a special case of
the Monotone Sequences Theorem.

Theorem 50 (Comparison Test). Let (ak ) and (bk ) be real sequences. As-
∞
X X∞
sume that 0 ⩽ ak ⩽ bk for all k ⩾ 1, and that bk converges. Then ak
k=1 k=1
converges.
n
X
Proof. Let sn = ak .
k=1
Then (sn ) is increasing, since ak ⩾ 0 for all k ⩾ 1.

83
 Also,
n
X n
X ∞
X
sn = ak ⩽ bk ⩽ bk
k=1 k=1 k=1

(since this last series converges),

so (sn ) is bounded.
∞
X
Hence, by the Monotone Sequences Theorem (or Proposition 49), ak
k=1
converges.

Remark. ˆ More generally, if there is a positive constant C such that

∞
X ∞
X
0 ⩽ ak ⩽ Cbk for k ⩾ 1, and if bk converges, then ak converges,
k=1 k=1
by a small generalisation of the argument.

ˆ The Comparison Test can also be used to show that a series diverges.
∞
X ∞
X
If 0 ⩽ ak ⩽ bk for all k and ak diverges, then bk diverges.
k=1 k=1

ˆ We don’t need to know the value of

P
bk to use the Comparison Test,
just that it exists.

ˆ Please check the conditions of the Comparison Test very carefully be-
∞
X
fore applying it. Please do not do this by writing things like ak ⩽
k=1
∞
X P
bk . We can’t write down ak (which is, remember, a limit) until
k=1
we know that the limit exists. So either check the precise conditions of
the Comparison Test, or work with partial sums as in Proposition 49.

ˆ The Comparison Test is great!

Example.
∞
X 1
Claim. converges.
k=1
k2

84
Proof. For k ⩾ 2, we have

1 1
0⩽ 2
⩽ ,
k k(k − 1)
∞
X 1
and converges (we saw this previously),
k=2
k(k − 1)
∞
X 1
so by the Comparison Test we have that 2
converges.
k=1
k
∞
X 1
Remark. Note that this tells us nothing about the value of 2
! That is
k=1
k
an interesting, but more challenging, problem for another time (not in this
P 1
course). But we can still use k2
in future applications of the Comparison
Test, even without knowing the value.
∞
X 1
Example. The series converges. (As usual, we define 0! = 1.) This
k=0
k!
is an exercise on Sheet 5.
∞
X 1
We can then define e = .
k=0
k!

Example. Decimal expansions. I’m not going to go through this example,

but now is a good time to revisit it. You’ll find the details in Hilary Priestley’s
supplementary notes on the uncountability of the reals, on Moodle.

30 Series: more results and another test for

convergence
In Section 27, we met the Cauchy criterion for convergence (Theorem 47):
a sequence (an ) converges if and only if it is Cauchy. That immediately
translates to a useful result for series.

85
Theorem 51 (Cauchy Convergence Criterion for series). Let (ak ) be a se-
n
X ∞
X
quence, and write sn = ak . Then ak converges if and only if
k=1 k=1
 
n
 X 
∀ε > 0 ∃N ∈ N such that ∀n > m ⩾ N |sn − sm | =  ak  < ε.
 
 
k=m+1

Proof. Immediate from the Cauchy Convergence Criterion (Theorem 47).

So far, we have mostly considered series where all the terms are non-
negative real numbers. We are interested in other series too, though. When
a series can have negative terms (or even non-real), it might be that they
enable enough cancellation that a series converges, or it might be that the
P
convergence is so robust that even |ak | converges. That’s what the next
definition and result are about.
∞
X
Definition. Let (ak ) be a sequence. We say that ak converges absolutely
k=1
∞
X
if |ak | converges.
k=1

Remark. ˆ This makes sense for real and complex series.

ˆ The series
P
|ak | is a series where all the terms are (real and) non-
negative. Such series are particularly nice!

Theorem 52 (Absolute convergence implies convergence). Let (ak ) be a

∞
X ∞
X
sequence. If |ak | converges, then ak converges.
k=1 k=1

Proof. Idea: use partial sums and the Cauchy criterion.

Let
n
X n
X
sn = ak and Sn = |ak |.
k=1 k=1

86
 So we are assuming that (Sn ) converges, and want to deduce that (sn )
converges.
For n > m, we have
 n  n
 X  X
|sn − sm | =  ak  ⩽ |ak | = |Sn − Sm |.
 
 
k=m+1 k=m+1
P
Now |ak | converges by assumption,
so (Sn ) is Cauchy by the Cauchy Convergence Criterion,
so (sn ) is Cauchy by the inequality above,
P
so ak converges by the Cauchy Convergence Criterion.
sin3 (k 2 )
Example. Let ak = (−1)3k .
k2 + 1
Then 0 ⩽ |ak | ⩽ k12 for k ⩾ 1,
P 1
and k2
converges,
P
so by the Comparison Test |ak | converges,
P
so ak converges since absolute convergence implies convergence.
∞
X
Lemma 53. Take p ∈ R. Then k −p diverges for p ⩽ 1, and converges if
k=1
p > 1.

Proof. Case 1 p ⩽ 0. Then k −p ̸→ 0 as k → ∞, so the series does not

converge (by Proposition 48).
Case 2 p = 1. This is the harmonic series (see an example in Section 29).
Case 3 0 < p < 1. Note that then k −p > k −1 > 0, and we know that
P −1
k
P −p
diverges, so by the Comparison Test k diverges.
P 1
Case 4 p ⩾ 2. We already know that k2
converges (this was an example
near the end of Section 29), and 0 ⩽ k −p ⩽ k −2 , so, by the Comparison Test,
P −p
k converges.
Case 5 1 < p < 2. We’ll do this later, once we’ve developed some more
theory.

87
 P 1
P (−1)n
Example. We know that n
diverges, and so n
does not converge
absolutely. But does it converge? The next result will give us a way to show
that it does.

Theorem 54 (Alternating Series Test). Let (uk ) be a real sequence, and

∞
X
consider the series (−1)k−1 uk . If
k=1

ˆ uk ⩾ 0 for k ⩾ 1; and

ˆ (uk ) is decreasing, that is, uk+1 ⩽ uk for k ⩾ 1; and

ˆ uk → 0 as k → ∞,
∞
X
then (−1)k−1 uk converges.
k=1

Proof. Idea: consider partial sums, get subsequences that are monotone and
bounded.
n
X
Let sn = (−1)k−1 uk .
k=1

ˆ (s2n ) bounded above: We have

s2n = u1 − (u2 − u3 ) − (u4 − u5 ) − · · · − (u2n−2 − u2n−1 ) − u2n ⩽ u1 ,

so u1 is an upper bound for (s2n ).

ˆ (s2n ) is increasing: We have

s2n+2 − s2n = u2n+1 − u2n+2 ⩾ 0.

So, by the Monotone Sequences Theorem, (s2n ) converges. Say s2n → s

as n → ∞.
Now s2n+1 = s2n + u2n+1 → s + 0 = s as n → ∞, by AOL.
So (s2n+1 ) also converges to s.
Then (by Sheet 4 Q2) (sn ) converges.

88
Example.
∞
X (−1)n
Claim. converges.
n=1
n

Proof. We have n1 ⩾ 0 for all n,

and n1 n is decreasing,

1
and n
→ 0 as n → ∞.
∞
X 1
Hence, by the Alternating Series Test, (−1)n−1 converges,
n=1
n
∞
X (−1) n
and so (by AOL) converges.
n=1
n

Example.
∞
X (−1)n
Claim. √ converges.
n=1
n

Proof. Exercise.
P (−1)n
Remark. This remark is not part of the course. A series such as n
that
converges but does not converge absolutely is said to converge conditionally.
Such series are delicate, when compared to more robust series that converge
absolutely!

31 More on the Comparison Test

k2 + k + 1 X k2 + k + 1
Example. Let ak = and consider .
4k 4 − k 2 − 1 k
4k 4 − k 2 − 1
For large enough k, the denominator is positive, so ak exists and ak ⩾ 0.
Can we apply the Comparison Test?
1
Idea: ak grows roughly like 4k2
, so try comparing with that.

89
 For sufficiently large k, we have
ak 4k 2 (k 2 + k + 1)
1 =
4k2
4k 4 − k 2 − 1
4k 4 + 4k 3 + 4k 2
=
4k 4 − k 2 − 1
1 + k1 + k12
= → 1 as k → ∞,
1 − 4k12 − 4k14
so there is K such that if k ⩾ K then
ak 3
0⩽ 1 ⩽
4k2
2
so
3 1
0 ⩽ ak ⩽ · .
2 4k 2
X 3
Now converges,
k
8k 2
P
so, by the Comparison Test, k ak converges.
(It doesn’t matter that we have the inequalities only for large enough k
— the first finitely many terms don’t affect convergence.)

Remark. Some people like to summarise this strategy in the following result.

Theorem 55 (Limit form of Comparison Test). Let (ak ), (bk ) be real se-
ak
quences of positive terms, and assume that there is L > 0 such that bk
→L
P P
as k → ∞. Then ak converges if and only if bk converges.

Proof. Since abkk → L as k → ∞ and L2 > 0, there is K such that if k ⩾ K

 
 ak  L L ak 3L
then  − L < , and so <
 < .
bk 2 2 bk 2
(⇐) Then for k ⩾ K we have 0 < ak < 3L
P
2
b k , so if bk converges then
P 3L P
so does b and hence, by the Comparison Test,
2 k
ak converges.
(⇒) Also, for k ⩾ K we have 0 < bk < L2 ak (noting that L ̸= 0), so
P P2
if ak converges then so does a and hence, by the Comparison Test,
L k
P
bk converges.

90
Remark. It was important that, at least for sufficiently large k, the terms
ak
ak and bk are positive, and it was important that bk
converges to a positive
real number.
k2 + k + 1
Example. Let ak = .
4k 4 − k 2 − 1
Then (as before)

ak
1 → 1 as k → ∞,
4k2

and ak > 0 for sufficiently large k

1
and 4k2
> 0 for k ⩾ 1
1
P
and 4k2
converges
P
so, by the limit form of the Comparison Test, ak converges.

32 Ratio Test
Our next test for convergence is really useful. It sometimes gives a self-
contained way to decide whether a series converges, rather than having to
have an idea already (as is necessary for the Comparison Test, for example).
The idea of the next test is essentially comparison with a geometric series.
Let’s do an example before we state the general result.
∞
k X k
Example. Let ak = k and consider .
2 k=1
2k
P 1
We can’t directly compare with 2k
.
1
Idea: the terms ak decrease nearly like 2k
.

91
 More precisely,
ak+1 k+1 k
= k+1 / k
ak 2 2
k + 1 2k
= k+1 ·
2 k
k+1 1
= ·
k 2

1 1 1
= 1+ · → as k → ∞,
k 2 2
where we used AOL at the end.  
 ak+1 1  1 ak+1 3
So there is K such that if k ⩾ K then  −  < , so < .
ak 2 4 ak 4
3 k−K


Then for k ⩾ K we have 0 < ak ⩽ 4 aK
∞  k−K
X 3
and aK converges (geometric series with common ratio 34 ,
4
 3  k=K
and  4  < 1)
P
so, by the Comparison Test, ak converges.

Theorem 56 (Ratio Test). Let (ak ) be a real sequence of positive terms.

ak+1
Assume that converges as k → ∞, say to limit L.
ak
P
(i) If 0 ⩽ L < 1, then ak converges.
P
(ii) If L > 1, then ak diverges.

Remark. ˆ Here, exceptionally, we allow L = ∞, and this is covered by

the L > 1 case.

ˆ If L = 1, then the Ratio Test tells us nothing.

ak+1
ˆ If ak
does not tend to a limit as k → ∞, then the Ratio Test tells us
nothing.

Proof. (i) Assume that 0 ⩽ L < 1.

1+L
Let α = 2
, so that L < α < 1. Let ε = α − L > 0.

92
  
ak+1
Since → L, there is N such that if k ⩾ N then  ak+1 − L < ε,
 
ak ak

ak+1
so ak
< L + ε = α.

Now for k ⩾ N we have 0 < ak ⩽ αk−N aN .

X
But αk−N aN converges (constant times a geometric series with com-
k⩾N
mon ratio α, where |α| < 1).
P
So, by the Comparison Test, ak converges (the first N terms do not
affect convergence).

(ii) Assume that L > 1.

Case 1 L ∈ R.
1+L
Let α = 2
, so 1 < α < L. Let ε = L − α > 0.
ak+1 
 ak+1

Since → L, there is N such that if k ⩾ N then  ak − L < ε,

ak
ak+1
so ak
> L − ε = α.

Now for k ⩾ N we have ak ⩾ αk−N aN > 0,

P
and so ak ̸→ 0 as k → ∞, so ak diverges.

Case 2 L = ∞.

Let α = 2.
ak+1 ak+1
Since → ∞, there is N such that if k ⩾ N then an
> α.
ak
Then finish as in Case 1.

k
Example. Let ak = (we did this before!).
2k
Then ak > 0 for all k, and

k + 1 2k
 
ak+1 1 1 1
= k+1 · = 1+ · → < 1 as k → ∞,
ak 2 k k 2 2

93
by AOL.
P
So, by the Ratio Test, ak converges.

Example. Let ak = k1 .
Then ak > 0 for all k, and
ak+1 k
= → 1 as k → ∞,
ak k+1
so the Ratio Test tells us nothing.
ak+1
Notice how we really had to consider the limit. We have ak
< 1 for
all k, but that’s not enough to determine convergence — remember that we
already know that this series diverges.

Example. Let

 1k

if k = 2m for some m ⩾ 1
2
ak =
0

otherwise.

As it stands, we can’t apply the Ratio Test, because the terms aren’t all
positive.
But we can omit the zero terms (which do not affect the convergence of
1 X
the series): let bm = 2m for m ⩾ 1, and consider bm .
2 m
Now bm > 0 for all m, and
m
bm+1 22 1
= 2m+1 = 2m → 0 < 1 as m → ∞,
bm 2 2
P P
so by the Ratio Test bm converges and hence ak converges.

Remark. ˆ The Ratio Test is brilliant, but please make sure you apply
it carefully. Check the conditions!

ak+1
ˆ It’s not always the case that ak
converges, so that’s why we stated it
as a condition in the Ratio Test. Try to avoid assuming that the limit
exists.

94
 ˆ We proved the Ratio Test by comparing with a geometric series. So
we shouldn’t use the Ratio Test to decide whether a geometric series
converges!

We can adapt the Ratio Test to study absolute convergence of series even
when the terms are not all real and positive.

Corollary 57. Let(ak ) be a sequence of non-zero (real or complex) numbers.

 ak+1 
Assume that   converges as k → ∞, say to limit L.
ak 
P
(i) If 0 ⩽ L < 1, then ak converges absolutely and hence converges.
P
(ii) If L > 1, then ak diverges.

Remark. ˆ As before, we allow L = ∞ and include this in the case

L > 1.

ˆ If L = 1 then the Ratio Test tells us nothing.

Proof. (i) Apply the Ratio Test to (|ak |).

(ii) If L > 1, then the proof of the Ratio Test as applied to (|ak |) shows
P
that |ak | ̸→ 0, so ak ̸→ 0, and so ak diverges.

Remark. We’ll see later in the course that the Ratio Test (especially in this
form) is extremely helpful for studying power series.

33 Integral Test
In this section, we’ll study certain series by considering corresponding inte-
grals. This is a bit surprising, since we currently don’t know what integration
is. But it’s nice to see the link to convergence of series now, so we’ll pretend

95
that we know what integration is, and that we know some basic facts about
integration. In Analysis III, you’ll fill in the details of this — you might like
to revisit this section/video after studying Analysis III.
Some (for now unofficial) facts we’ll assume:

ˆ Suitably nice functions are integrable (in this section we’ll consider only
suitably nice functions).
R k+1
ˆ We can integrate constants: if c ∈ R then k
c dx = c.

ˆ Integration preserves weak inequalities: if f, g : [a, b] → R are suitably

Rb Rb
nice, and f (x) ⩽ g(x) for all x ∈ [a, b], then a f ⩽ a g.
Rc Rb Rc
ˆ If a < b < c and f : [a, c] → R is suitably nice, then a
f= a
f+ b
f.

Theorem 58 (Integral Test). Let f : [1, ∞) → R be a function. Assume

that

ˆ f is non-negative (f (x) ⩾ 0 for all x ∈ [1, ∞));

ˆ f is decreasing (if x < y then f (x) ⩾ f (y));

R k+1
ˆ k
f (x)dx exists for each k ⩾ 1.
Pn Rn
Let sn = k=1 f (k) and In = 1
f (x)dx.

(i) Let σn = sn − In . Then (σn ) converges, and if we let σ be the limit of

(σn ), then 0 ⩽ σ ⩽ f (1).
P
(ii) f (k) converges if and only if (In ) converges.

Remark. ˆ The main part of the Integral Test is (ii), and (i) is mostly
interesting for helping us to prove (ii), but (as we’ll see) (i) is also useful
in its own right.

96
 R k+1
ˆ If f is continuous then k
f (x)dx exists for each k ⩾ 1.

Proof. (i) Idea: show that (σn ) is bounded below and decreasing.

Since f is decreasing, for x ∈ [k, k + 1], we have

f (k + 1) ⩽ f (x) ⩽ f (k),

and so
Z k+1 Z k+1 Z k+1
f (k + 1) = f (k + 1)dx ⩽ f (x)dx ⩽ f (k)dx = f (k).
k k k

Now
Z 2
f (2) ⩽ f (x)dx ⩽ f (1)
1
Z 3
and f (3) ⩽ f (x)dx ⩽ f (2)
2
..
and .
Z n
and f (n) ⩽ f (x)dx ⩽ f (n − 1).
n−1

Adding these (finitely many) inequalities gives

sn − f (1) ⩽ In ⩽ sn − f (n)

97
 so

0 ⩽ f (n) ⩽ sn − In ⩽ f (1)

so

0 ⩽ σn ⩽ f (1) for all n ⩾ 1.

Also,

σn+1 − σn = sn+1 − In+1 − sn + In

Z n+1
= f (n + 1) − f (x)dx ⩽ 0
n

as above.

So (σn ) is decreasing and bounded below,

so, by the Monotone Sequences Theorem, it converges.

Say σn → σ as n → ∞.

Then, since limits preserve weak inequalities, and 0 ⩽ σn ⩽ f (1) for all
n ⩾ 1, we have 0 ⩽ σ ⩽ f (1).

(ii) If (sn ) converges, then by AOL so does (In ), since In = sn − σn .

And if (In ) converges, then by AOL so does (sn ), since sn = In + σn .

Example. This is Lemma 53 revisited.

k −p diverges, and if p > 1 then k −p con-

P P
Claim. If 0 < p ⩽ 1, then
verges.

Proof. Fix p > 0. Define f : [1, ∞) → R by f (x) = x−p .

Then f is non-negative, and decreasing on [1, ∞), and continuous.

98
 Now for p ̸= 1 we have
Z n
In = x−p dx
1 n
1 1−p
= x
1−p 1
1
= (n1−p − 1),
1−p

so for p < 1 we see that (In ) does not converge,

and for p > 1 we see that (In ) does converge.
Also, for p = 1 we have
Z n
In = x−p dx = [log x]n1 = log n,
1

so (In ) does not converge.

k −p converges for p > 1 and diverges for
P
Hence, by the Integral Test,
0 < p ⩽ 1.

Remark. The Integral Test handles p ⩾ 0, but not p < 0 because in this
case the function is not decreasing. Fortunately we can handle p < 0 directly,
because in this case k −p ̸→ 0 and so
P −p
k diverges.

Example.
X 1
Claim. diverges.
k⩾2
k log k

Proof. Exercise — use the Integral Test.

Remark. This series can be useful for counterexamples, because it feels like
it ‘only just’ diverges.

99
34 Euler’s constant and rearranging series
X1
Example. We know that the harmonic series diverges. But the Inte-
k
k
gral Test can give us additional information.
Let γn = 1 + 12 + 13 + · · · + 1
n
− log n.
Define f : [1, ∞) → R by f (x) = x1 .
Then f is non-negative, decreasing and continuous, and
n Z n
X 1 1
γn = − dx,
k=1
k 1 x

so (i) of Theorem 58 tells us that (γn ) converges as n → ∞, and the limit is

in [0, 1].
Let γ be this limit (this is standard notation), so
1 1 1
γn = 1 + + + · · · + − log n → γ as n → ∞,
2 3 n
and 0 ⩽ γ ⩽ 1.
So, roughly speaking, the partial sums of the harmonic series grow like
log n, and hence tend to infinity rather slowly.
The number γ is known as Euler’s constant.
It is not known whether γ is rational or irrational.
n
X 1
Example. Let sn = (−1)k−1 .
k=1
k
Then
1 1 1 1
s2n = 1 − + + · · · + −
 2 3 2n −1 2n 
1 1 1 1 1 1
= 1 + + + ··· + −2 + + ··· +
2 3 2n 2 4 2n
= (γ2n + log(2n)) − (γn + log n)

= log 2 + γ2n − γn

→ log 2 as n → ∞,

100
 1
and s2n+1 = s2n + 2n+1
→ log 2 as n → ∞,
so (by a result on a problems sheet) (sn ) converges to log 2, that is,
∞
X (−1)k−1
= log 2.
k=1
k

Remark. The order in which we sum the terms in this series really matters.
It turns out that if we regroup to have the same terms but in another
order, with three positive terms followed by one negative, so

1 1 1 1 1 1 1
1+ + − + + + − + ···
3 5 2 7 9 11 4

then we instead get log 2 + 12 log 3 (exercise: show this!). There’s yet another
version of the series, with yet another value, on Sheet 7.

Definition. Let g : N → N be a bijection (a permutation of N). Given a

P P P
series ak , write bk = ag(k) . Then bk is a rearrangement of ak .

ˆ It turns out (no proof in this course!) that if

P
Remark. ak is abso-
P
lutely convergent, then any rearrangement of ak also converges, to
the same limit. In this sense absolutely convergent series are ‘robust’.

ˆ As we have seen, a series that converges but not absolutely (that is,
a series that converges conditionally) is less robust. A rearrangement
might give a series that converges to a different value, or even that does
not converge at all.

Remark. In this course we’ve seen several tests for convergence of a series:

ˆ the Comparison Test;

ˆ the Alternating Series Test;

ˆ the Ratio Test;

101
 ˆ the Integral Test.

We also saw that absolute convergence implies convergence.

These are the main tools for studying convergence of a series, but they
are not the only ones: not every series is susceptible to one of these tests, and
there are other convergence tests that can be useful — but they are beyond
the scope of the course.

35 Power series
∞
X
Definition. A real power series is a series of the form ck xk , where ck ∈ R
k=0
for all k ⩾ 0 and x is a real variable.
∞
X
A complex power series is a series of the form ck z k , where ck ∈ C for
k=0
all k ⩾ 0 and z is a complex variable.

Remark. ˆ Much of the theory applies equally to real and complex

power series, and of course every real power series is also a complex
power series. Our focus in this course is mostly on real power series,
but sometimes it is at least as convenient, or even more convenient, to
work in the more general complex setting and then specialise later.

ˆ We typically want to define a function using a power series. This is

why we think of x or z as a variable.
∞
X
ˆ By convention, when we consider the series ck z k at z = 0, we mean
k=0
just c0 . There are no issues about what 00 might mean! Every power
series converges at z = 0, so we do not need to consider this case when
studying convergence.

102
 ∞
X zk
Example. Consider . We use the Ratio Test: for z ̸= 0, we have
k=0
k!
 k+1
z k 

 z k! |z|
 (k + 1)! / k!  = (k + 1)! |z| = k + 1 → 0 as k → ∞


and 0 < 1, so by the Ratio Test the series converges absolutely, and hence
converges, for all z ∈ C.

Definition. We define the exponential function exp : C → C by exp(z) =

∞
X zk
. We also write ez for exp(z).
k=0
k!

Example. Consider
∞ ∞ ∞ ∞
X
k z 2k X
k z
2k+1 X z 2k+1 X z 2k
(−1) and (−1) and and .
k=0
(2k)! k=0
(2k + 1)! k=0
(2k + 1)! k=0
(2k)!

Each of these converges for all z ∈ C. (Exercise: use the Ratio Test to
prove this for z ̸= 0.)

Definition. We define the sine function sin : C → C by

∞
X z 2k+1
sin(z) = (−1)k ,
k=0
(2k + 1)!

and the cosine function cos : C → C by

∞
X z 2k
cos(z) = (−1)k .
k=0
(2k)!

Definition. We define the hyperbolic sine function sinh : C → C by

∞
X z 2k+1
sinh(z) =
k=0
(2k + 1)!

and the hyperbolic cosine function cosh : C → C by

∞
X z 2k
cosh(z) = .
k=0
(2k)!

103
Remark. ˆ We can go on to define other trig functions such as tan, cosec,
sec and cot using these, on suitable domains. We wouldn’t expect these
further functions to have power series that converge on the whole of C.

ˆ We have defined sin and cos by power series, not by right-angled trian-
gles.

ˆ We need to go on to deduce the usual properties of exp, sin and cos,

working from the power series definitions. We’ll make a start on that
in this course, and you will continue in Analysis II next term.

Remark. We previously proved (as part of AOL) that if (sn ) and (tn ) are
convergent sequences, with sn → L and tn → M , then (sn +tn ) also converges,
and sn + tn → L + M .
We can apply this to sequences of partial sums, which gives us a way to
consider the sum of two series.
P P
To put that more explicitly, let ak and bk be convergent series, and
write
n
X n
X
sn = ak and tn = bk .
k=1 k=1

∞
X
Then (sn ) and (tn ) converge. Say sn → s and tn → t (that is, ak = s and
k=1
∞
X
bk = t). Then
k=1

n
X n
X n
X
sn + t n = ak + bk = (ak + bk ),
k=1 k=1 k=1

∞
X
so by AOL (sn + tn ) converges, and (ak + bk ) = s + t.
k=1
So we can sum two convergent series, and we can also use AOL to show
that we can multiply a series by a (real or complex) number.

104
Remark. The above remark gives a useful way to show that a series diverges.
P P P
If ak converges and bk diverges, then (ak +bk ) diverges too. That’s
P P
because if (ak + bk ) converges, then also ((ak + bk ) − ak ) converges, by
the remark above.
P P
Exercise: show, through suitable examples, that if ak and bk both
P
diverge, then it might be that (ak + bk ) converges and it might be that it
diverges.

Example. From the power series definitions earlier, and this remark about
AOL applied to series, we can see that for z ∈ C we have
1
cos z = (eiz + e−iz )
2
1
and sin z = (eiz − e−iz )
2i
1 z
and cosh z = (e + e−z )
2
1 z
and sinh z = (e − e−z )
2
and eiz = cos z + i sin z.

We can also see from the power series definitions that for z ∈ C we have
cos(iz) = cosh z, and other similar relationships between cos and cosh, and
between sin and sinh. (Exercise: think about all of these!)

36 Radius of convergence
In this section, it will be more natural to study power series in C. The
main goal will be to determine the subset of C on which a given power series
converges. As we’ll see, this subset must have a rather specific form. You
might already have ideas about this, having tackled Sheet 6 Q4.
We expect that a power series will be more likely to converge for small
|z| than for large |z|. In principle, the subset of C on which a given power

105
 ck z k converges might be a blob of convergence, or an ink splat of
P
series
convergence, or an even more complicated region.

It turns out, though, that in fact the relevant shape is not a blob or an
ink splat, but a disc. There are some details to be addressed, but that is the
secret reason for the following definition — we’ll go into the details after the
definition.

ck z k be a power series. We define its radius of convergence

P
Definition. Let
to be

sup{|z| ∈ R : |ck z k | converges} if the sup exists
 P
R :=
∞

otherwise.

ˆ We certainly have 0 ∈ {|z| ∈ R : |ck z k | converges}, so

P
Remark.
the set is non-empty. So this subset of R has a sup if and only if it is
bounded.

ˆ There are other equivalent ways to define the radius of convergence,

so if you look at another source then you might see a slightly different
definition.

This definition is interesting, but without the following proposition it

doesn’t get us far. The proposition is more important than it might at first
sight appear!

106
 ck z k be a power series with
P
Proposition 59 (Radius of convergence). Let
radius of convergence R.

ck z k converges absolutely and hence

P
(i) If R > 0 and |z| < R, then
converges.

ck z k diverges.
P
(ii) If |z| > R, then

Remark. This proposition says nothing about what happens if |z| = R.

This is deliberate!

Proof. (i) Case 1: R ∈ R.

Assume that R > 0, and take z ∈ C with |z| < R.

Then there is S with |z| < S < R. Let ε = R − S > 0.

|ck wk | converges}, by the Approximation

P
Since R = sup{|w| ∈ R :
|ck ρk |
P
Property there is ρ such that S = R − ε < ρ ⩽ R and
converges.

|ck ρk | converges, so by the Comparison Test

P
Then 0 ⩽ |z| < ρ and
|ck z k | converges.
P

ck z k
P
Since absolute convergence implies convergence, this shows that
converges.

Case 2: R = ∞.

Very similar to Case 1.

(ii) Take z ∈ C with |z| > R

|ck z k | diverges, by definition of R, but we don’t

P
Then we know that
ck z k .
P
know about

ck z k converges.
P
Suppose, for a contradiction, that

107
 |ck ρk | converges.
P
Find ρ with ρ > R such that

Then ck z k → 0 as k → ∞, so (ck z k ) is bounded, so there is M such

that |ck z k | ⩽ M for all k.

Take ρ with R < ρ < |z|.

Then
 ρ k  ρ k
0 ⩽ |ck ρk | ⩽ |ck z k |   ⩽ M  
   
z z
P  ρ k ρ
and z
converges (geometric series with common ratio  , and
z
ρ
  < 1),
z

|ck ρk | converges,
P
so, by the Comparison Test,

contradicting the definition of R.

Remark. ˆ We call {z ∈ C : |z| < R} the disc of convergence for the

power series. Proposition 59 shows that this is a useful concept. For
a real power series, the corresponding concept is an interval of conver-
gence.

ˆ Anything at all can happen on the circle {z ∈ C : |z| = R}! The series
might converge everywhere on the circle, or diverge everywhere on the
circle, or converge at some points and diverge at others.

ˆ You might like to revisit Sheet 6 Q4 briefly having seen the theory, to
see the connections.

Example. ˆ We have already seen that the exponential, sine and cosine,
hyperbolic sine and hyperbolic cosine series have radius of convergence
∞ (using the Ratio Test).

108
 ˆ The geometric series z k has R = 1 (from an example in Section 28).
P

∞
X k! k
Example. Consider k
x .
k=0
k
For x ̸= 0, we have
k
 (k + 1)! k+1 k k  kk
  
k! k
 (k + 1)k+1 x
 ·  = |x| = |x|
k!xk  (k + 1)k k! k+1
 −k
1 1
= 1+ |x| → |x| as k → ∞,
k e
P  k! k 
so by the Ratio Test the series  k x  converges for |x| < e (so R ⩾ e)
k

and diverges for |x| > e (so R ⩽ e).

So R = e.

Remark. Note that it was not enough to use the Ratio Test to show that
the series converges (absolutely) for |x| < e — this shows that R ⩾ e, not
that R = e.

1 if k prime

ck xk where ck =
P
Example. Consider
0 otherwise.


So the Ratio Test won’t work!

For x = 1, we see that ck xk ̸→ 0 as k → ∞ (because there are infinitely
many primes), so R ⩽ 1.
If |x| < 1, then 0 ⩽ |ck xk | ⩽ |xk |,
P k
and |x | is a convergent geometric series,
ck xk converges absolutely and hence con-
P
so, by the Comparison Test,
verges. So R ⩾ 1.
So R = 1.

Remark. The Ratio Test is often useful for finding the radius of convergence
of a power series, but does not always work. There are more sophisticated
strategies that work in other situations, but it is easy to apply them incor-
rectly, and they are not needed for Prelims.

109
37 Differentiation Theorem
We have seen that we can define a function using a power series. For ex-
ample, we defined the exponential, sine and cosine functions in this way. In
this section, we’re going to explore the derivative of a function defined by a
power series, as this will enable us to study some key properties of familiar
functions. This will not be an in-depth exploration, since you haven’t stud-
ied differentiability yet. We’ll state a theorem, and then see how extremely
useful it can be in practice. We shan’t prove it in this course; you’ll see a
proof in a future analysis course.

ck xk be
P
Theorem 60 (Differentiation Theorem for real power series). Let
a real power series with radius of convergence R. Assume that 0 < R ⩽ ∞.
X∞
For |x| < R, define f (x) = ck x k .
k=0
Then f (x) is well defined whenever |x| < R. Moreover, if |x| < R then
the derivative f ′ (x) exists, and
∞
! ∞ ∞
d X X d X
f ′ (x) = ck x k
= k
(ck x ) = kck xk−1 .
dx k=0 k=0
dx k=1

Remark. ˆ The slogan is that “on the disc of convergence, we can dif-
ferentiate term-by-term”.

ˆ The theorem is definitely not obvious! It involves exchanging the order

of limiting processes, and that is a delicate business.

Example. We saw that the power series defining the exponential, sine, co-
sine, sinh and cosh functions have R = ∞, so the series converge on R (and
on C), and by the Differentiation Theorem they are differentiable on all of
R. Moreover, by the Differentiation Theorem we can differentiate term by
term on R.

110
 For example, for x ∈ R we have
∞
!
k
d x d X x
e =
dx dx k=0 k!
∞
d xk
X  
= by the Differentiation Theorem
k=0
dx k!
∞
X kxk−1
=
k=1
k!
∞
X xk−1
=
k=1
(k − 1)!
∞
X xk
= = ex .
k=0
k!

To summarise, for all x ∈ R we have

d x
e = ex
dx
d
sin x = cos x
dx
d
cos x = − sin x
dx
d
sinh x = cosh x
dx
d
cosh x = sinh x.
dx

Example.

Claim. sin2 x + cos2 x = 1 for all x ∈ R.

Proof. Define h : R → R by h(x) = sin2 x + cos2 x.

Then (using properties of differentiability that you’ll study in Analysis II
next term) h is differentiable on R, and

h′ (x) = 2 cos x sin x − 2 sin x cos x = 0 for all x ∈ R.

This means (using a result you’ll see in Analysis II) that h is constant.

111
 But we know from the power series that sin 0 = 0 and cos 0 = 1, so
h(0) = 1.
So h(x) = 1 for all x ∈ R.

Remark. It would not be a good plan to try to do this by squaring power

series and manipulating terms — this would need a lot of justification.

Example.

Claim. ea+b = ea eb for all a, b ∈ R.

Proof. Fix c ∈ R, and define g : R → R by g(x) = ex ec−x .

Then (Analysis II) g is differentiable on R, and

g ′ (x) = ex ec−x − ex ec−x = 0 for all x ∈ R.

This means (Analysis II) that g is constant.

But we know from the power series that e0 = 1, so g(0) = ec .
So g(x) = ec for all x ∈ R.
This argument works for all c ∈ R. Take a, b ∈ R, and apply it with
x = a, c = a + b to get ea+b = ea eb .

Remark. This shows that for all x ∈ R we have ex e−x = e0 = 1. From the
power series, we see that ex > 0 for x ⩾ 0, and hence in fact ex > 0 for all
x ∈ R.

Remark. These examples illustrate a really useful strategy, which can also
be used to prove results like trig identities. Watch out for more on this in
Analysis II next term!

What is π? We have defined sine and cosine using power series, without
mentioning right-angled triangles. We can then define π to be the smallest

112
 π
positive x such that sin x = 0, or 2
as the smallest positive x such that
cos x = 0. It is not obvious that smallest such values exist; you’ll look at this
in more detail in Analysis II. You’ll then be able to go on and prove that
sine and cosine are 2π-periodic, for example.

Example. We see that if x, y ∈ R then

ex+iy = ex (cos y + i sin y).

We can then use properties of π to see that e2πi = 1.

You’ll study differentiability in C as part of the Part A Complex Anal-
ysis course, when you’ll go on to explore many interesting (and surprising)
properties of complex functions.

There are some further examples at the end of Hilary Priestley’s notes on
Moodle, which you’ll also see in Analysis II/III.
This brings us to the end of Analysis I, but definitely not the end of
analysis.
Building on your knowledge of analysis so far, you might like to consider
the following questions, as a warm up for Analysis II.

ˆ Given a function f : R → R and a, L ∈ R, what does it mean to say

that f (x) → L as x → a?

ˆ What does it mean to say that f : R → R is continuous at a point

x ∈ R?

x if x ∈ Q

ˆ Define f : R → R by f (x) = At which points (if any)
0

otherwise.
is f continuous?

113
ˆ What does it mean to say that f : R → R is differentiable at a point
x ∈ R?

Well done and thank you for making it to the end of Analysis I!

To be continued. . . (in Analysis II)

114