On Constructing Proofs

Richard Earl
Michaelmas 2008

1 Why do Analysis?
Each year, the analysis courses are found notoriously unpopular and/or difficult by most (if not all) new mathematics
students at Oxford (and in other universities). They are commonly felt to be pedantic, pointless and alien. You may
find it still more galling as each week your tutor rattles off a short proof, of a few lines, to an exercise you’ve been
fumbling with for the last week; still worse having seen the proof, and being convinced of its soundness, you may well
feel you could never have generated the proof yourself. Tim Gowers describes the phenomenon as: "Why easy analysis
problems are easy" on his webpage; his solution is: "Solving analysis problems on auto." In the following I hope to
likewise give you some tips on an almost algorithmic approach that should help you break down first year analysis
problems.
The analysis proofs you will see this year are littered with weird and not-so-wonderful symbols such as ∃, ∀, δ, ε, etc.
and the secret to understanding analysis, certainly to getting over one’s fear of analysis, is in making these symbols
familiar, and even helpful, rather than letting them remain alien and intimidating. They each have precise definitions
and so need showing some respect, but their individual meanings are not at all complicated.
A major point of the first year analysis courses (and one of the main points of maths degrees) is in fostering
a new, more rigorous mindset. Another aspect of the degree is in teaching students to present mathematics in an
understandable, irrefutable fashion. There is little to no focus on proof in A-levels; you may well not be explicitly
told anything wrong at A-level but that is because the syllabus deliberately avoids any weird, pathological examples.
However, mathematics (and certainly the real world) has no syllabus. At university we hope to train you so that, when
faced with a new problem, you can proceed with confidence in your arguments and not fall into traps of assuming
the seemingly obvious but wrong; even an applied mathematician modelling the real world (something quite removed
from this first term analysis course) must retain an appreciation of the assumptions and limitations of the model s/he
is using. Similarly any mathematician repeatedly running a calculation on a computer needs to know with certainty
that small initial errors won’t balloon out of all proportion in the long run.
There is much more to analysis than simply taking a rigorous mindset — for example the complex analysis and
topology courses in the second year have many genuine, novel ideas of their own. For now, though, if you find the
statements of analysis theorems this year rather obvious then can I recommend treating the subject as fine-tuning
your reasoning abilities. And if you find the exercises frustrating and intractable then please take my word for it that
by knowing your definitions precisely and carefully following "the rules of the game" you will realise that almost all
first year analysis problems have very simple calculations at their centres.

2 Quantifiers
Two important symbols that will be new to most people are ∀ and ∃. They are known as quantifiers. Their respective
meanings are "for all" and "there exists". For example, consider the following logical statements:—

∀x ∈ R x2 > 0 (1)
∃x ∈ R x2 = 2
∀x ∈ N x+1 ∈N
∃x ∈ R x4 = −1

Just reading the quantifiers as above (and reading ∈ as "in", knowing R is the set of reals and N is the set of natural
numbers) I hope it is clear that how all four statements read and that the first three are true and that the fourth is
false.

1
 Looking at these rather tame statements one might wonder what the difficulty is with analysis; well, these quantifiers
can be used together in a logical statement to make quite complicated claims. For example,
1
∀ε > 0 ∃N ∈ N ∀n > N 0< < ε. (2)
n
Let’s spend a little time trying to break down this particular statement. The heart of the statement itself, that
1
0< < ε, (3)
n
is not complicated at all. If you knew that ε = 3 and n = 2 then it is a trivial thing to check; what makes (2) somewhat
imposing is the rather alien start, heavy with quantifiers, but you should not forget that what is being claimed about
ε and n is really very simple. Hopefully, looking at (3) you are thinking "what are ε and n?". This is precisely the
information included in the quantifiers at the start of (2).
If we return to (1) it reads easily as "the square of every real number is non-negative." You were probably happy
1
to accept this at school, though if your teacher had "proven" this by checking it worked for x = 2, 10 , −2.3 and π you
would probably have seen the holes in such an "argument". Both (1) and (2) can be written in the form

∀x ∈ S1 P1 (x) ∀ε ∈ S2 P2 (ε)

where S1 = R, S2 = (0, ∞) and where P1 and P2 are just statements involving one argument (x and ε respectively).
There’s no doubt that to prove any such statement you need to prove it in every possible case (i.e. for every such x
or ε) and that you can’t do this by a case-by-case analysis if there are infinitely many cases that need checking. The
only difference in the above statements is that P2 is a rather more complicated statement in ε than P1 is in x.
There are two important things to note here: in proving a statement of the form ∀x ∈ S P (x)

• We are being asked to prove a general statement. We cannot treat the quantity x as anything but a general
element of S, which means we cannot, however convenient it might be, assign x further useful properties to
make the proof easier. Throughout the proof x is "untouchable". At best we can treat different sorts of x-value
case-by-case but even then we need to be sure to cover all possible cases.
• In the proof we need to include a statement such as "Let x ∈ S." The ∀ quantifier is a signpost for such a line —
if we don’t include such a line we are neither proving the statement in sufficient generality nor giving the reader
any idea what x is. On the other hand, seeing a ∀ quantifier in a statement is nothing to fear as we need only,
and algorithmically, introduce the line "Let x ∈ S" appropriately in the proof.

Once having introduced a general positive ε > 0 to meet the needs of the first quantifier, then the next quantifier
reads ∃N ∈ N. And this is where we need to do some work! A general ε > 0 has been introduced and is essentially
fixed for the rest of the proof, arbitrary but fixed. Given this ε we need to find an N which is allowed to depend on ε
such that
1
∀n > N 0 < < ε
n
or equivalently this can be rewritten as
1
n > N =⇒ 0 < < ε.
n
Again we can rewrite this as
1
n > N =⇒ n > . (4)
ε
This, then, is the only real work/maths in the proof. We have rephrased our problem as: we need to find N such that
any n greater than or equal to N is also guaranteed to be greater than 1/ε. I hope it is clear that taking any integer
N greater than 1/ε will work — note any such N will suffice to meet the requirements of the ∃ quantifier, N doesn’t
have to be the smallest possible or be unique in some other way.

• The work required in proving any statement involving ∀ and ∃ quantifiers is sign-posted by the ∃ quantifiers. To
meet the needs of these quantifiers you must demonstrate the existence of quantities satisfying all the subsequent
parts of the statement, but these quantities may vary with anything already introduced into the statement.

I imagine that on a first reading the previous paragraph was something of a blur, and certainly not something that
felt algorithmic. However that paragraph simply described the second quantifier — mods analysis proofs are rarely
generated in a serial line-by-line fashion. It is usually easiest to construct them simultaneously from the start and
finish as I hope you will see in the next section.
3 Constructing Proofs
If we were now to write down the full proof of (2) we found in the previous section it might read as follows:-

1. Let ε > 0.
2. Let N be a natural number greater than 1/ε.
3. Let n > N .
4. Then
1 1 1
0< 6 < = ε.
n N 1/ε
If written straight down even such a simple proof might seem to have been plucked from nowhere. If we recall how
we came to the proof, we see that:

• Line 1 is a no-brainer; we simply need to introduce ε into the mix with sufficient generality.

The next easy line to write down isn’t Line 2 though it’s Line 4!

• Line 4 is in some sense a no-brainer. The proof has to end with this line or otherwise (2) was never fully
confirmed.

So without thinking we can begin and end the proof, so far it looking like:
1
Let ε > 0. . . . . . . . . . 0 < < ε.
n
But warning bells should be ringing now as you’ve written down a quantify without introduction. What is n?

• Well we can fill in the previous line straight-forwardly as it is another "for all" clause that requires us only to
set out the generality of n.

1
Let ε > 0. . . . . . . ∀n > N 0<
< ε.
n
This new line hasn’t stopped the alarm bells though as n is still given in terms of the gatecrashing, unintroduced N .
On the other hand we have managed to write down three of the four lines of the proof without really putting our
brains in gear, instead just putting the lines where they had to go. As commented before all the work in the proof
revolves around the ∃ quantifier.
How do we fill in the middle of the proof? The proof at the moment reads like a bad short story or one missing
crucial chapters in that, out of nowhere, N has arrived without introduction to nicely tie up the conclusion. Such a
plot-line would be implausible and it remains for us to introduce N. Further N ’s existence needs to be guaranteed by
some axiom or earlier result, and if you want N to do anything remarkable for you (e.g. climb walls) you need to have
already accounted for this (e.g. N was bitten by a radioactive spider).
But by rewriting the needed properties of N as in (4) we see that any integer N greater than 1/ε will do. Not a
terribly complicated calculation ultimately, and by algorithmically constructing the proof forwards and backwards it
was fairly clear what needed to be proven.

4 Definitions, Hypotheses and Negations

All the quantified definitions which you will meet in the analysis course can be, in a systematic fashion, broken down
and placed into proofs. Statement (2) can be concisely phrased as "the sequence 1/n converges to zero" and that
is how the statement would be likely to appear in an exercise. It’s not a surprising result but without knowing the
precise definition of sequence convergence there would be little hope of proving this rigorously from only a concise
phrasing in words.

• PRECISELY KNOWING DEFINITIONS IS ESPECIALLY VITAL IN ANALYSIS

Here is another, still relatively simple proposition from later in this term’s course. We shall prove it positively
and remark on possible alternative proofs such as proof by contradiction to highlight how to negate statements. The
proposition is:
Proposition 1 Convergent sequences are bounded.
It is a hopeless task starting if we don’t know the precise definitions behind these words.
Definition 2 The real sequence xn is convergent if
∃L ∈ R ∀ε > 0 ∃N ∈ N ∀n > N |xn − L| < ε.
Definition 3 The real sequence xn is bounded if
∃M > 0 ∀n ∈ N |xn | < M.
Some intuition for these definitions is never a bad thing, but the aim of this article is to convince you that you can be
almost enitrely algorithmic in the generation of mods proofs. Later, when trying to revise these proofs, the main idea/s
of a more complicated proof will be all you need to remember if you are comfortable with the routine manipulation of
quantified statements.
But, for now, let’s set up this proof as described in the previous section. What makes this example different is that
we have a hypothesis to work with. Properly phrased the statement might read more like "all convergent sequences
are bounded" and so we need to introduce a convergent sequence xn . The hypothesis (Definition 2) introduces into
our story a limit L about which we know nothing except that it is a real number. This limit L is a given and we can
make no assumptions about it. It is true of this limit L that
∀ε > 0 ∃N ∈ N ∀n > N |xn − L| < ε.
Note that we know this is true for all ε > 0. So, if it’s helpful we can set ε to be 1 or π or 10−3 or anything positive
and know there are (probably different) N s associated with each choice of ε.
We only have one hypothesis to make use of, namely that xn converges to L. In more complicated proofs there
may well be two or three hypotheses and you may have to choose a sensible order in which to apply these. However
with only one hypothesis it seems sensible to write this down at the start of the proof as it is the only fact with which
we can argue. So our first few lines, writing down the hypothesis ought to be:
• Let xn be a convergent sequence.
• Then xn has a limit L.
• Let ε > 0.
• Let N be as guaranteed by our hypothesis.
• Let n > N
• Then |xn − L| < ε.
And at the end of the proof we know must be the line.
• Let k ∈ N.
• Then |xk | < M.
[Note as n has already been introduced as a natural number in the range n > N when a second natural number is
introduced it makes sense to use another letter.]
I hope it is clear that all the work involved in the proof is in demonstrating the existence of this M . At this stage
it is the only missing piece of our draft proof. By Line 6 of the proof we already have bounds on xN , xN +1 , xN +2 , . . .
namely that they are in the range
L − ε < xn < L + ε
and so for each such n > N we have |xn | < max (|L + ε| , |L − ε|) . However the bound M has to work for the
whole sequence, not just a tail-end of the sequence; we have no guarantee it will work for x1 , x2 , x3 , . . . , xN −1 . These
exceptions, though, are only finite in number. Even if that number N is large and the exceptions x1 , x2 , . . . , xN −1 are
huge there is a largest one and we can say
|xk | 6 max (|x1 | , |x2 | , . . . |xN−1 |) for k < N.
So the whole sequence is bounded by M = max (|x1 | , |x2 | , . . . |xN −1 | , |L + ε| , |L − ε|) .

Our hypothesis when written down provided a natural bound, just not for the whole sequence, and the only thinking
involved in the proof was in dealing with these finitely many exceptions. Our proof in full might then read:
 • Let xn be a convergent sequence.
• Then xn has a limit L.
• Let ε > 0.
• Let N be as guaranteed by our hypothesis.
• Let n > N
• Then |xn − L| < ε.
• Then L − ε < xn < L + ε.
• So |xn | < max (|L + ε| , |L − ε|) .
• Let M = max (|x1 | , |x2 | , . . . |xN−1 | , |L + ε| , |L − ε|)
• Let k ∈ N.
• Then |xk | < M.
[Stylistically some might consider it slightly better to set ε = 1 (say) as there is no real need to keep an arbitrary ε in
the argument, but this is a small point and doesn’t reflect on the validity of the above proof.]
To conclude the section we consider how we might prove this same proposition other ways. We might prove the
contrapositive statement. The contrapositive of the statement P =⇒ Q is the equivalent statement ¬Q =⇒ ¬P
where ¬P and ¬Q are the negations of P and Q and ¬ is read as "not". For this proposition this would amount to
showing that unbounded sequences are divergent. Alternatively we might try a proof by contradiction and show that
assuming P and ¬Q leads to an absurd conclusion. In this case we would need to consider a sequence which was both
convergent and unbounded and aim to find a contradiction.
However, we first need to know how to negate a logical statement. To do this we recall (the logical versions of) De
Morgan’s Laws. They state that:
¬ (∀x ∈ S P (x)) ⇐⇒ ∃x ∈ S ¬P (x)
¬ (∃x ∈ S P (x)) ⇐⇒ ∀x ∈ S ¬P (x)
If paraphrased the two might read as:
• If P does not always hold then there is a counter-example to P
• If P does not hold somewhere then it is false everywhere
So if we were to negate the formal definition of convergence by repeated applications of De Morgan’s Laws we’d
have
¬ (∃L ∈ R ∀ε > 0 ∃N ∈ N ∀n > N |xn − L| < ε)
⇐⇒ ∀L ∈ R ¬ (∀ε > 0 ∃N ∈ N ∀n > N |xn − L| < ε)
⇐⇒ ∀L ∈ R ∃ε > 0 ¬ (∃N ∈ N ∀n > N |xn − L| < ε)
⇐⇒ ∀L ∈ R ∃ε > 0 ∀N ∈ N ¬ (∀n > N |xn − L| < ε)
⇐⇒ ∀L ∈ R ∃ε > 0 ∀N ∈ N ∃n > N ¬ (|xn − L| < ε)
⇐⇒ ∀L ∈ R ∃ε > 0 ∀N ∈ N ∃n > N |xn − L| > ε
So we can define
Definition 4 The real sequence xn is divergent (non-convergent) if
∀L ∈ R ∃ε > 0 ∀N ∈ N ∃n > N |xn − L| > ε.
And by arguing similarly with De Morgan’s Laws we also have:
Definition 5 The real sequence xn is unbounded if
∀M > 0 ∃n ∈ N |xn | > M.
So, were we to try proving our proposition using contradiction we would have two hypotheses saying that a sequence
was convergent and unbounded
∃L ∈ R ∀ε > 0 ∃N ∈ N ∀n > N |xn − L| < ε,
∀M > 0 ∃n ∈ N |xn | > M,
solely with the aim of getting a contradiction.
 Analysis I — Sequences and Series
Richard Earl — Michaelmas Term 2008

1 The Real Number System — The Field Axioms

1.1 What are the reals?
What are the real numbers? For the moment this is too hard a question!
We can give various answers, but for now we will simply present a set of axioms — statements we will assume are
true about real numbers. We will base all our arguments on these axioms and on nothing else, and aim to develop all
our mathematics from these axioms alone.

1.2 Axioms
Naively, then we assume we have a set R which we call the set of real numbers which satisfies the following axioms.
See also on the website for a full list of the axioms.
Unless otherwise made clear the quantities a, b, x, etc. discussed in the following will be real numbers.

1.3 Addition
1. For every ordered pair of real numbers a, b we can associate a real number written a + b called their sum
2. To every real number a we can associate a real number, written −a, called its negative or additive inverse
3. There is a special real number 0 called zero or the additive identity

such that

a+b=b+a (A1)
a + (b + c) = (a + b) + c (A2)
a+0=a (A3)
a + (−a) = 0 (A4)

These axioms "read" as:

A1: + is commutative;
A2: + is associative;
A3: 0 is an additive identity;
A4: −a is an additive inverse of a.

1.4 Easy properties of A1—A4

Proposition 1 If a + x = a for all real numbers a, then x = 0. This means that 0 is the only additive identity.

Proof.
x = x + 0 by A3
= 0 + x by A1
= 0 by hypothesis with a = 0

1
Proposition 2 If a + x = a + y then x = y. In particular, this means that additive inverses are unique.

Proof.
y = y+0 by A3
= y + (a + (−a)) by A4
= (y + a) + (−a) by A2
= (a + y) + (−a) by A1
= (a + x) + (−a) by hypothesis
= (x + a) + (−a) by A1
= x + (a + (−a)) by A2
= x+0 by A4
= x by A3

Proposition 3 − (−a) = a

Proof.
(−a) + a = a + (−a) by A1
= 0 by A4
(−a) + − (−a) = 0 by A4
Hence − (−a) = a as additive inverses are unique (a consequence of Proposition 2).

Proposition 4 − (a + b) = (−a) + (−b)

Proof. Left as Exercise Sheet 1, Question 1c.

Proposition 5 −0 = 0

Proof.
0+0 = 0 by A3
0 + (−0) = 0 by A4
−0 = 0 by Proposition 2

1.5 Multiplication
1. To every ordered pair of real numbers a, b we can associate a real number, written a · b, called their product
2. To every real number, except 0, we can associate a real number written 1/a called its reciprocal or multi-
plicative inverse
3. There is a special real number 1 called one or the multiplicative identity

such that

a·b=b·a (M1)
a · (b · c) = (a · b) · c (M2)
a·1=a (M3)
a · (1/a) = 1 if a 6= 0 (M4)

These axioms "read" as:

M1: · is commutative;
M2: · is associative;
M3: 1 is a multiplicative identity;
M4: 1/a is a multiplicative inverse of a.

2
1.6 Avoiding Collapse
If 1 = 0 we would have (by M3 and Proposition 11 below)
x = x · 1 = x · 0 = 0 for all x.
For any sensible arithmetic we must assume
0 6= 1. (Z)
Deducing that 0 = 1 is the only “safe” contradiction in the axioms of the real numbers, i.e. the only one that you can
be certain is a contradiction.

1.7 Easy Consequences of M1—M4

Proposition 6 If a · x = a for all real numbers a then x = 1.
Proposition 7 If a 6= 0 and a · x = a · y = 1 then x = y.
Proposition 8 If a 6= 0 then 1/ (1/a) = 1.
Proposition 9 If a 6= 0 6= b and a · b 6= 0 and 1/ (ab) = (1/a) · (1/b).
In fact we need not, in this last proposition, have the hypothesis that a · b 6= 0 as we shall see in Proposition 12.
Note that M1—M4 say the same things about · as A1—A4 say about + and that the proofs can be automatically
translated.

1.8 The Distributive Law

For all real numbers a, b, c
a · (b + c) = a · b + a · c (D)

1.9 More Consequences

Proposition 10
(a + b) · c = a · c + b · c
Proof.
(a + b) · c = c · (a + b) by M1
= c · a + c · b by D
= a · c + b · c by M1 twice

Proposition 11
a·0=0
Proof.
a·0+0 = a·0 by A3
= a · (0 + 0) by A3
= a · 0 + a · 0 by D
Hence a · 0 = 0 by Proposition 2.
Proposition 12 If a · b = 0 then either a = 0 or b = 0 (or both).
Proof. If a 6= 0 and b 6= 0 then
0 = (1/a · 1/b) · 0 by Proposition 11
= 0 · (1/a · 1/b) by M1
= (a · b) · (1/a · 1/b) by hypothesis
= (b · a) · (1/a · 1/b) by M1
= ((b · a) · 1/a) · 1/b by M2
= (b · (a · 1/a)) · 1/b by M2
= (b · 1) · 1/b by M4
= b · (1/b) by M3
= 1 by M4
This contradicts Z and hence a · b = 0 contradicts a and b both being non-zero. By De Morgan’s laws it follows that
at least one of a and b is zero.

3
Proposition 13
a · (−b) = − (a · b) .
In particular note that (−1) · a = −a.

Proof.
(a · b) + (a · (−b)) = a · (b + (−b)) by D
= a·0 by A4
= 0 by Proposition 11
(a · b) + (−(a · b)) = 0 by A4
So by Proposition 2
a · (−b) = −(a · b).

Proposition 14
(−1) · (−1) = 1.

Proof.
(−1) · (−1) = −(−1) by Proposition 13
= 1 by Proposition 3

1.10 Notation
From now on we will also write
ab for a·b
a−b for a + (−b)
a/b for a · (1/b)
a−1 for 1/a
Also we write, for a 6= 0,
a0 = 1
ak+1 = ak¡· a¢ for all k = 0, 1, 2, 3, . . .
a−l = 1/ al for all l = 1, 2, 3, . . .

1.11 Other systems

Other number systems also satisfy A1—A4, M1—M4, D, Z. Such systems are called fields.
Q, R, C are all examples of fields. Other examples involve the integers modulo a prime number.
Z is not a field as it does not meet M4 though it does satisfy the remainder of the above axioms. N fails to meet
A4 as well.
All that you do in Linear Algebra this term about real vector spaces requires only these axioms of its scalars, and
linear algebra can just as easily be defined over any field as over R.

4
2 The Real Number System — The Order Axioms
2.1 The number ‘line’
We want to capture the idea that the real numbers are ‘ordered’ with some axioms. It is easier just to give axioms for
what it is to be positive.
There is a subset P of R called the positive real numbers satisfying:

a, b ∈ P =⇒ a + b ∈ P (P1)
a, b ∈ P =⇒ ab ∈ P (P2)
Exactly one of a ∈ P, a = 0, −a ∈ P holds. (P3)

2.2 Easy consequences

Proposition 15 1 ∈ P

Proof. By P3 precisely one of

1 ∈ P, 1 = 0, −1 ∈ P
holds. Axiom Z discounts the second possibility and we will show −1 ∈ P leads to a contradiction as follows.

−1 ∈ P =⇒ (−1) (−1) ∈ P by P2
=⇒ 1∈P by Proposition 14
=⇒ −1 ∈ P and 1 ∈ P

which contradicts P3. The only remaining possibility is that 1 ∈ P.

2.3 Notation
We write

a>b for a−b∈P

a<b for b−a∈P
a>b for a − b ∈ P ∪ {0}
a6b for b − a ∈ P ∪ {0}

Note that in this notation the trichotomy axiom, P3, reads as: "precisely one of a > 0, a = 0, a < 0 holds".

2.4 Easy consequences

Proposition 16 a > b if and only if −a < −b. In particular x > 0 if and only if −x < 0.

Proof.
a > b
⇐⇒ a−b > 0 by definition
⇐⇒ − (−a) − b > 0 by Proposition 3
⇐⇒ −b − (−a) > 0 by A1
⇐⇒ −a < −b by definition

Proposition 17 For all real x, y, z

x 6 x (1)
x 6 y and y 6 x =⇒ x = y (2)
x 6 y and y 6 z =⇒ x 6 z (3)

5
Proof.
(i) By A4 x − x = 0 ∈ P ∪ {0} and so x 6 x.
(ii) If x 6 y and y 6 x then
y − x ∈ P ∪ {0} and x − y = − (y − x) ∈ P ∪ {0} .
There are then two cases to consider:
(a) y − x ∈ P and − (y − x) ∈ P. This contradicts P3.
(b) if y − x = 0 then by Proposition 2 x = y; similarly if x − y = 0.
(iii) If x = y or y = z this is trivial, so we need only consider the case where y − x ∈ P and z − y ∈ P. Using A1—A4
we can show that
z − x = (z − y) + (y − x),
and from P1 we then have that z − x ∈ P as required.

2.5 Inequalities shift

Proposition 18 Let x, y, z be reals such that x < y. Then x + z < y + z.

Proof. Simply note using A1—A4 that (y + z) − (x + z) = y − x ∈ P.

Proposition 19 Let x, y, z be reals such that x < y and 0 < z. Then zx < zy.

Proof. We have that y − x ∈ P and z ∈ P. So by P2 z (y − x) ∈ P. By D and Proposition 13

zy + z (−x) = zy − zx ∈ P

and so zx < zy as required.

Corollary 20 Let x, y, z be reals such that x < y and z < 0. Then zx > zy.

Proof. As z < 0 then 0 < −z and so by the previous proposition (−z) x < (−z) y. By Proposition 13 it follows that
− (zx) < − (zy). It follows finally that zx > zy by Proposition 16.

Corollary 21 a2 > 0 for any real a.

Proof. This follows from P2 if a > 0, from Proposition 11 if a = 0 and from the previous corollary if a < 0.

Proposition 22 If x ∈ P then 1/x ∈ P.

Proof. As x ∈ P then x 6= 0. As 1/x 6= 0 (this would contradict Z) then 1/x ∈ P or − (1/x) ∈ P. If the latter then,
by Proposition 13 and P2
− (1/x) x = − ((1/x) x) = −1 ∈ P
which contradicts Proposition 15.

Corollary 23 If x, y ∈ P and x < y then 1/y < 1/x.

Proof.
x < y assumed
=⇒ 1 < y/x by Propositions 19 and 22
=⇒ 1/y < (1/y) (y/x) by Propositions 19 and 22
=⇒ 1/y < 1/x by M2 and M4

6
2.6 Two important functions: max, min
Using the axioms we may now define the maximum and minimum of two numbers. Define max : R × R → R by
½
x if x > y
max(x, y) =
y if y > x

By the trichotomy axioms P3, this is a well-defined function.

Similarly we define min : R × R → R by
½
y if x > y
min(x, y) =
x if y > x

We can extend these to functions of many variables. For example, recursively we can define:

max (a1 , . . . , an , an+1 ) = max (max (a1 , . . . , an ) , an+1 )

Example 24 max (x, y) = − min (−x, −y)

Solution. We argue by cases:

max(x, y) − min(−x, −y)

x>y x −x < −y − (−x) = x
x=y x x=y x
x<y y −y < −x − (−y) = y

2.7 An important function: modulus

We define | | : R → R by
⎧
⎨ x if x > 0
|x| = 0 if x = 0
⎩
−x if x < 0
These cases are distinct and cover all possibilities by the trichotomy axiom P3, so we get a well-defined function.
|x| is read as "mod x" or "the modulus of x".

2.8 An easy consequence

Proposition 25 |−x| = |x|

Proof. We argue by cases:

• If x > 0 then by Proposition 16 we have −x < 0. So | − x| = −(−x) = x = |x|.

• If x = 0 then −x = −0 = 0. So | − x| = 0 = |x|.
• If x < 0 then −x > 0. So | − x| = −x = |x|.

2.9 The Triangle Law (∆)

Theorem 26 For any real numbers a, b

|a + b| 6 |a| + |b| ,
with equality if and only if (a > 0 and b > 0) or (a < 0 and b < 0) .

7
Proof. There are 8 cases possible:

a b a+b
(A) >0 >0 >0
>0 >0 <0 forbidden by P1
(B) >0 <0 >0
(C) >0 <0 <0
(B’) <0 >0 >0
(C’) <0 >0 <0
<0 <0 >0 forbidden by P1
(D) <0 <0 <0

(A) |a| = a, |b| = b, |a + b| = a + b

(B) |a| = a, |b| = −b, |a + b| = a + b
Note a + b 6 a − b ⇐⇒ b 6 −b ⇐⇒ 0 6 (−b) + (−b) which follows by P1 as −b ∈ P
(C) |a| = a, |b| = −b, |a + b| = −(a + b)
Note −(a + b) 6 a − b ⇐⇒ −a − b 6 a − b ⇐⇒ −a 6 a ⇐⇒ 0 6 a + a which follows by P1 as a ∈ P
(D) |a| = −a, |b| = −b, |a + b| = −(a + b)
(B’) As (B) with a and b swapped
(C’) As (C) with a and b swapped
It is left as an Exercise (see Sheet 1, Question 3a) to consider when equality may occur in the above cases.

2.10 The modulus of a product

Proposition 27 |ab| = |a| |b|

Proof. Again we need to treat this by cases.

• If either a = 0 or b = 0 then |ab| = |0| = 0 = |a| |b| .

• If a > 0, b > 0 then ab > 0 by P2 and so |ab| = ab = |a| |b| .
• If a > 0, b < 0 then ab < 0 by Proposition 19 and so |ab| = −ab = a (−b) = |a| |b| .
By symmetry this also deals with the case a < 0, b > 0.
• If a < 0 and b < 0 then ab > 0 by Corollary 20 and so by Propositions 3 and 13

|ab| = ab = − (−ab) = − (a (−b)) = (−a) (−b) = |a| |b| .

2.11 A useful inequality

Theorem 28 (Bernoulli’s Inequality) Let x be a real number with x > −1 and let n be a positive integer. Then

(1 + x)n > 1 + nx.

Proof. We shall prove the inequality by induction — note that the inequality is trivially true when n = 1.
Suppose it is the case that
(1 + x)N > 1 + N x
holds for all real x > −1. Then 1 + x > 0 by Proposition 18 and N x2 > 0 as N > 0 and x2 > 0 from Corollary 21.
Hence
N+1 N
(1 + x) = (1 + x) (1 + x) by definition
> (1 + x) (1 + N x) by hypothesis and Proposition 19
= 1 + (N + 1) x + N x2 by A1—A4
> 1 + (N + 1) x by Proposition 18
Hence the theorem follows by induction.

8
3 The Real Number System — The Completeness Axiom
At this stage we can surely persuade ourselves that we could write down proofs of all the usual algebraic properties of
R, and all the usual properties of “6”.
But note, many structures share these properties; in particular both Q and R.

3.1 "Gaps" in Q
So why won’t Q do? Why do mathematicians not settle for working with this rather nice field of easily understood
ratios of integers; countable, too, so that we can list the elements?
The Ancient Greeks had at least one reason — in Q we can’t find an element to measure the length of the hypotenuse
of a right angled isosceles triangle with two short sides of length 1. Here is the proof of that fact.

Theorem 30 There is no element α ∈ Q such that α2 = 2

Proof. If there were such an α, then we could write α = m/n for some m, n ∈ Z, n 6= 0. Further we could assume
that this fraction is in lowest terms so that m and n are coprime. Then 2n2 = m2 . As m2 is even then m is also even
as a product of odd numbers is odd. We can then write m = 2k and hence n2 = 2k2 . But then n, too, is even by the
same reasoning and m/n wasn’t in lowest terms after all — this is the required contradiction.

So Q is lacking in some ways, certainly if we wish to discuss distances, and we look to describe the way(s) R is
different from Q.

3.2 Greatest and Least Elements

Definition 31 Let B ⊆ R.
We say that b1 is a least element or minimum of B if (i) b1 ∈ B and (ii) b1 6 b for all b ∈ B. In this case we
write b1 = min B.
We say that b2 is a greatest element or maximum of B if (i) b2 ∈ B and (ii) b 6 b2 for all b ∈ B. In this case
we write b2 = max B.

Example 32 1 is the minimum of [1, 2) but there is no maximum for this set. [If x ∈ [1, 2) were a maximum then
x < 2 and so 1 + x/2 is a greater element of the set.]

Proposition 33 A maximum (if it exists) is unique. Similarly a minimum are unique.

Proof. Suppose that b and c are both maxima of B. Then as b ∈ B and c is a maximum, b 6 c; as c ∈ B and b is a
maximum then c 6 b. Hence b = c. Similarly minima are unique if defined.

3.3 Upper and Lower Bounds

Definition 34 Let B ⊆ R.
We say that h1 is a lower bound of B if h1 6 b for all b ∈ B.
We say that h2 is an upper bound of B if b 6 h2 for all b ∈ B.

Example 35 23 and π are both upper bounds of [1, 2). And 1 is a lower bound as is −37. The set of upper bounds is
[2, ∞) and the set of lower bounds is (−∞, 1].

Definition 36 We say that a set B ⊆ R is

• bounded above if it has an upper bound;

• bounded below if it has a lower bound;
• bounded if it has upper and lower bounds.

Example 37 Q is neither bounded above nor below, N is bounded below, (−∞, e] is bounded above, ∅ is bounded.

9
3.4 The Completeness Axiom
We are now ready to give our final axiom which characterises the real numbers:

Axiom C: let E ⊆ R be a non-empty set which is bounded above. Then the set of upper bounds of E has a least
element.
Definition 38 We call this least element the least upper bound or supremum, written as sup E. (Note that we
can refer to sup E as the least upper bound as we have already shown in Proposition 33 that minima are unique.)
Example 39 2 is supremum of [1, 2)
Proof. For all x ∈ [1, 2), 1 6 x < 2 by definition, so clearly 2 is an upper bound. Now suppose that there was a
smaller upper bound, t. So t < 2, and as it is an upper bound, t > 1. Then 32 6 t+2 t+2 t+2
2 < 2. So 2 ∈ [1, 2) but t < 2
contradicting the fact that t was an upper bound.
Example 40 The set of upper bounds of ∅ is R which has no minimum element.
Proposition 41 If E ⊆ R has a maximum then max E = sup E.
Proof. Now max E > x for all x ∈ E by definition of max E being a maximum. Further if l is an upper bound for E
then l > max E by virtue of max E being an element of E. Hence max E is the least upper bound.
Proposition 42 (The Approximation Property) Let E be bounded above and non-empty and let ε > 0. Then there
exists x ∈ E such that
sup E − ε < x 6 sup E.
Proof. If not, then sup E − ε is an upper bound of E less than the least upper bound, which is a contradiction.
Corollary 43 Let E be bounded above and non-empty. There is a function a : N → R, such that for all n we have
1
sup E − < a(n) 6 sup E
n

3.5 Infima
We would like to make the symmetric definition for the maximum (if it exists) of the lower bounds of a set bounded
above. One way would be to introduce yet another axiom guaranteeing its existence. But we don’t need to do that,
we now have enough axioms: from now on proofs, and not yet more plausible assumptions, are needed.
Theorem 44 Let F be a non-empty set which is bounded below. Then the set of lower bounds of F has a greatest
element.
Proof. Let E = {−x : x ∈ F }. As F is non-empty then E is also non-empty.
In Proposition 16 we showed that x 6 y ⇐⇒ −y 6 −x. Let l be a lower bound of F . Then l 6 x for all x ∈ F . So
−x 6 −l for all x ∈ F , that is y 6 −l for all y ∈ E. So E is bounded above, and non-empty, so by the Completeness
Axiom, sup E exists.
We shall prove (i) − sup E is a lower bound of F, (ii) if l is a lower bound of F then l 6 − sup E.
(i) If x ∈ F then −x ∈ E and so −x 6 sup E. Hence by Proposition 16, x > − sup E and we see − sup E is a lower
bound of F .
(ii) If l 6 x for all x ∈ F then −l > t for all t ∈ E. Hence −l > sup E by virtue of sup E being the least upper
bound of E. Finally l 6 − sup E.
Definition 45 This element is known as the greatest lower bound or infimum of F and is written inf F.
• By an argument similar to Proposition 33 we can show easily show that infima are unique.
• Note if F has a minimum element then min F = inf F.
Example 46 sup[1, 2) = 2 and inf[1, 2) = 1. Also min[1, 2) = 1 whilst max[1, 2) does not exist.
Corollary 47 (The Approximation Property for infima) Let F be bounded below and non-empty and let ε > 0. Then
there exists x ∈ E such that
inf F 6 x < inf F + ε.
Corollary 48 Let F be bounded below and non-empty. There is a function a : N → R, such that for all n we have
1
inf F 6 a(n) < inf E + .
n

10
 √
3.6 2 exists
Theorem 49 There exists a unique positive number α such that α2 = 2.

Proof. Let E = {x ∈ R : x2 < 2}. Note that 12 = 1 < 2, so that 1 ∈ E and in particular E is non-empty. Further if
x > 2 then
x2 = xx > 2x > 4 > 2
by Proposition 19. Hence 2 is an upper bound for E and so we may define

α = sup E.

Note further that α > 1 > 0 is positive. We split the remainder of the proof into showing that α2 < 2 and α2 > 2
both lead to contradictions.

• Suppose for a contradiction that α2 < 2. Let

µ ¶
1 2 − α2
h = min α, > 0. (4)
2 3α
Then ¡ ¢
(α + h)2 = α2 + 2hα + h2 < α2 + 3hα < α2 + 2 − α2 = 2 (5)
since
2 − α2
h < α and h <
3α
Hence α + h ∈ E and since α = sup E we get α + h 6 α, a contradiction.
• Suppose instead that α2 > 2. Let µ ¶
1 α2 − 2
h= > 0. (6)
2 2α
As α − h < α there exists e ∈ E with α − h < e by the Approximation Property; then

(α − h)2 < e2 < 2 =⇒ α2 − 2hα + h2 < 2.

α2 −2
As h2 > 0 this gives α2 − 2hα < 2, and so, since α > 0, we have h > 2α which contradicts our choice of h.
2
Finally, by trichotomy, α = 2 follows as the only remaining possibility. To show uniqueness of α suppose that
β is a positive number such that β 2 = 2. Then

α2 = β 2 =⇒ α2 − β 2 = (α + β) (α − β) = 0.

By Proposition 12 it follows that β = α or β = −α. As α > 0 then −α < 0 and so α is the only positive solution
of x2 = 2.

Remark 50 Line (4) appears to come out of nowhere! To understand the need for this line we first need to look at
line (5); the idea behind our contradiction of α2 < 2 is to find a positive h small enough that (α + h)2 < 2. It is line
(5) which guides us into choosing the h in line (4). Similarly for the second contradiction, that α2 > 2 is wrong, our
2
idea is (essentially) to find a small enough positive h such that (α − h) > 2 is still true which explains our choice of
h in line (6).
© ª
Remark 51 Note that this result shows that Q does not satisfy the completeness axiom as the set x ∈ Q : x2 < 2
does not have a supremum in Q.
√
Notation 52 We write 2 for α.

3.6.1 A generalisation
√
Theorem 53 Let a be any positive real number. Then there exists a unique real number — denoted by a — whose
square is a.

Proof. This just involves a refining of the previous argument.

11
3.7 Archimedean Property
Theorem 53 (Archimedean Property of the Natural Numbers) Let x ∈ R. Then x < n for some n ∈ N. i.e. N is not
bounded above.

Proof. If not, N is bounded above and not empty. So let ξ = sup N. Then ξ − 1 < ξ, so ξ − 1 < k for some k ∈ N by
the Approximation Property. Then ξ < k + 1, but as k + 1 ∈ N, we have k + 1 6 ξ. So ξ < ξ, which is contradiction
to the trichotomy axiom.
1
Corollary 54 Let ε > 0. Then 0 < n < ε for some n ∈ N.

Proof. Apply the Archimedean Property to 1/ε.

Corollary 55 Given reals a, b with a < b then there exists a rational number q and an irrational number r such that
a < q < b and a < r < b.

Proof. Left as Sheet 2, Exercise 5c.

3.8 R is not countable

Theorem 56 R is uncountable.

Proof. If R were countable, then so too would be [0, 1]. Clearly [0, 1] is not finite as it contains all k1 where k ∈ N.
We proceed now with a proof by contradiction to show that [0, 1] is not countably infinite. Suppose θ : N → [0, 1] is
a bijection and we write xk = θ(k).

• We choose distinct a1 , b1 so that x1 ∈

/ [a1 , b1 ]. If x1 6= 1 then we can find a1 and b1 such that x1 < a1 < b1 < 1
and if x1 = 1 then we can take the interval [0, 1/2].
• Having chosen a1 , b1 we then select a2 , b2 so that a1 < a2 < b2 < b1 and x2 ∈ / [a2 , b2 ]. In a similar fashion to the
above if x2 < b1 we can find a2 and b2 so that max (a1 , x2 ) < a2 < b2 < b1 and if x2 > b1 then we can take the
interval [(2a1 + b1 ) /3, (a1 + 2b1 ) /3] , i.e. the middle third of the previous interval.
• We repeat this process producing reals ai and bj such that0 6 a1 < a2 < a3 < · · · < b3 < b2 < b1 6 1 and
xi ∈
/ [ai , bi ] for each i.

Now set E = {aj : j ∈ N} which is bounded above by 1 and F = {bj : j ∈ N} is bounded below by 0. So we may
define
λ = sup E and μ = inf F.
For all m, n we have am 6 bn . In particular, each bn is an upper bound of E and so λ 6 bn for all n as λ is the least
upper bound of E. So λ is lower bound of F which means λ 6 μ as μ is the greatest lower bound of F . Then

an 6 λ 6 μ 6 bn for all n.

For all n we have λ ∈ [an , bn ] and xn ∈

/ [an , bn ] and so λ 6= xn for all n which contradicts the fact that θ is a bijection.

We shall give the "classic" proof of the uncountability of R due to Cantor once we have discussed decimal expansions
at more length.

12
4 Complex numbers
4.1 Algebraic Properties
We can define C from R by taking the set C to be R2 , the set of real ordered pairs and defining addition + and
multiplication · by

(a1 , b1 ) + (a2 , b2 ) = (a1 + a2 , b1 + b2 )

(a1 , b1 ) · (a2 , b2 ) = (a1 a2 − b1 b2 , a2 b1 + a1 b2 ) .

So that, for example ¡ ¢

2
(0, 1) = (0, 1) · (0, 1) = 02 − 12 , 0 · 1 + 0 · 1 = (−1, 0) .
We more commonly write i for (0, 1) and write a + bi for (a, b) , so that the above equation states i2 = −1. Further we
identify each real number r with r + 0i and so can think of the reals as a subset of the complex numbers. Note this
also shows that C is uncountable.

It is not hard to check that the field axioms A1—A4, M1—M4, D, Z are all true of the complex numbers.

4.2 Order?
However, the complex numbers cannot be made into an ordered field — i.e. there is no subset of C which satisfies the
positivity axioms P1—P3. This is left to Sheet 2, Exercise 6. There is, though, a useful function, the modulus function,
that we can use to determine the "size" of complex numbers.

4.3 Definition and easy properties of | |

Let z = x + yi throughout the following.

Definition 57 The modulus of z, written |z|, is given by

p
|z| = x2 + y 2

This makes sense as x2 + y 2 > 0.

Definition 58 The conjugate of z, written z̄ (or z ∗ in some texts), is given by

z̄ = x − yi.

None of the following is at all difficult to prove — they are algebra, not analysis.

1. If z is real (i.e. z = x + 0i) then |z| = |x|; that is the definitions of real and complex modulus agree where
applicable.
2. |z| = |z̄|
3. |z|2 = z z̄
4. Re z = x, Im z = y
5. |Re z| 6 |z|, |Im z| 6 |z|
6. z + z̄ = 2 Re z
7. z − z̄ = 2i Im z
8. z1 + z2 = z1 + z2
9. z1 z2 = z1 z2

13
Theorem 59 For z, w ∈ C,
|zw| = |z| |w| .

Proof. Let z = x + iy and w = u + iv. Then all we need is the factorisation

(xv + yu)2 + (xu − yv)2 = (x2 + y 2 )(u2 + v 2 )

and the existence of unique square roots.

Theorem 60 (Triangle Law) For complex numbers z, w,

|z + w| 6 |z| + |w|

Proof. We use the above properties.

|z + w|2 = (z + w) (z + w)
= (z + w) (z̄ + w̄)
= z z̄ + (zw + z̄w) + ww
= z z̄ + 2 Re (zw) + ww
6 z z̄ + 2|zw| + ww
2 2
= |z| + 2|z| |w| + |w|
= (|z| + |w|)2

4.4 Aside — Existence of Transcendental Numbers (Off-syllabus)

Cantor proved, in 1874, that there are uncountably many real numbers. Given the simple elegance of his proof its
seminal nature can easily be missed, but it had many new and profound implications for 19th century mathematics.
The Ancient Greeks had been disturbed to discover the existence of irrational numbers would could not be written
as the ratio of two integers; Cantor’s theorem showed that most real numbers cannot be described using (integer-
coefficient) polynomials, in fact, that most real numbers cannot hope to be described by finite means alone.

Definition 61 A complex number α is said to be algebraic if there is a polynomial p (z) with integer coefficients such
that p (α) = 0. Equvalently, the coefficients may be assumed to be rational.

Definition 62 A complex number which is not algebraic is said to be transcendental.

The existence of transcendental numbers was first shown by Liouville in 1851 who showed that
∞
X
10−k!
k=1

is transcendental. Hermite showed that e is transcendental in 1873 and Lindemann showed that π is transcendental in
1882. Cantor’s remarkable result shows that "almost all" real numbers are transcendental, though without constructing
a single one. In 1873 Cantor proved:

Proposition 63 There are countably many algebraic numbers.

In order to show this we will need the following facts:

Lemma 64 If X1 , X2 , X3 , · · · are countable sets then so are

∞
[
Xk and X1 × X2 .
k=1

Proof. This was proved in the Introduction To Pure Mathematics course

Lemma 65 (Division Algorithm) Let a (x), b (x) be polynomials with complex coefficients. Then there exist polyno-
mials q (x) , r (x) such that
a (x) = q (x) b (x) + r (x)
and deg r (x) < deg b (x).

14
Proof. This is proved in next term’s Introduction To Groups, Rings and Fields.

Corollary 66 (Factor Theorem) If a (x) is a complex polynomial and a (α) = 0 where α ∈ C, then there exists a
complex polynomial q (x) such that
a (x) = q (x) (x − α)
and deg q (x) = deg p (x) − 1.

Proof. Set b (x) = x − α in the previous lemma; note that r (x) is a constant and that r (α) = 0.

Corollary 67 A complex polynomial p (x) has at most deg p complex roots.

Proof. This follows using induction applied to the previous corollary.

Remark 68 This may seem obvious but (if you have met modular arithmetic) you might note that x2 = 1 (mod 8)
has four roots, namely 1, 3, 5, 7.

Proof. (Of Proposition 63) Let A denote the set of algebraic numbers and let An denote the set of algebraic
numbers which are the root of a degree n polynomial with integer coefficients. The polynomials of degree n with
integer coefficients can be identified with Zn , e.g. 2x2 + 3x − 5 ↔ (2, 3, −5). By Lemma 64, Zn is countable and by
Corollary 67 a integer-coefficient polynomial of degree n can have at most n roots. Hence we may define a 1-1 map
∞
[
An → Zn × {1, 2, . . . , n} and so An is countable. As A = An then A is countable.
1

Corollary 69 There are uncountably many transcendental numbers.

Proof. If C\A were countable then C = A∪ (C\A) would be countable — a contradiction.

15
5 Real and complex sequences
5.1 Real numbers in practice
How do we handle a specific real number in practice?
√ Answer: we look at successive approximations. For example,
we could have the following approximations for 2:
14 141 1414
1, , , ,...
10 100 1000
or we may find π by looking at the following:
22 333 103993
3, , , ,,...
7 106 33102
Our task is to make precise the idea that these approximations approach the real numbers that they represent.

5.2 Sequences
Definition 70 A sequence of real numbers or a real sequence is a function α : N → R.

Definition 71 A sequence of complex numbers or a complex sequence is a function α : N → C.

Definition 72 Given a natural number n, the nth term of the sequence α is α (n) and we denote this αn .

Example 73 Here are some well known sequences:

• n 7→ σ (n) = (−1)n ,
• n 7→ ζ (n) = 0,
• n 7→ ι (n) = n.

Note, in practice, we often just give the sequence’s values, and say “the sequence 1, 12 , 14 , . . . ” if it is clear what the
function “must be”. Or better we write “the sequence (an )∞ n=1 ” or “the sequence (an )”.

Note also that although n determines the nth value, the nth value does not determine n because the defining function
need not be injective. (Look at the sequences ζ and σ above.)

5.3 New sequences from old

Suppose (an ) and (bn ) are sequences of real (or complex) numbers and c ∈ R (or C). We define the sequences (an +bn ),
(can ), (an bn ), (an /bn ) in the obvious way. All are well defined except possibly the quotient, where we must insist on
all the terms of (bn ) being non-zero.
n
Example 74 an = (−1) and bn = 1 for all n.

(an + bn ) = (0, 2, 0, 2, 0, 2, 0, . . .)
(−an ) = (−1)n+1
(an bn ) = (an )
¡ 2¢
an = (bn )

5.4 Definition of convergence

Definition 75 Let (an ) be a sequence of real numbers and let L ∈ R. We say that (an ) converges to L if

∀ε > 0 ∃N ∈ N ∀n > N |an − L| < ε.

We also say that (an ) tends to L. We write this as

(an ) → L or an → L as n → ∞ or just an → L.

16
Definition 76 If (an ) → L then we say that L is a limit of (an ) and we write

L = lim an or just L = lim an .

n→∞

Definition 77 We say that (an ) converges or is convergent if there exists L ∈ R such that (an ) → L. In full then,

(an ) converges ⇐⇒ ∃L ∈ R ∀ε > 0 ∃N ∈ N ∀n > N |an − L| < ε.

Definition 78 We say that (an ) diverges or is divergent if it does not converge. In full then,

(an ) diverges ⇐⇒ ∀L ∈ R ∃ε > 0 ∀N ∈ N ∃n > N |an − L| > ε.

Remark 79 In all the above ε is an arbitrary positive number though instinctively we usually think of ε as being very
small. The smaller the value of ε the "harder" we will usually have to look to find a value of N that will suffice.

5.5 Examples
• Let
2n − 1
an = for n > 1.
2n
Then (an ) → 1.

Proof. Consider ¯ ¯
|an − 1| = ¯1 − 2−n − 1¯ = 2−n .
Given ε > 0 our task is to show there exists N such that

n > N =⇒ 2−n < ε.

But note that 2n > n for all n ∈ N and so if N > 1/ε (which we know to exist by the Archimedean Property) and
n > N we have
1 1 1
2−n = n < 6 < ε.
2 n N

• The sequence
n2 + n + 1
an = (n > 1)
3n2 + 4
is convergent.

Proof. Because the statement for convergence is

∃L ∈ R ∀ε > 0 ∃N ∈ N ∀n > N |an − L| < ε

our first work is in deciding what the limit L will be. (Note this was given to us in the previous example.) How do we
guess L? Well for large positive n,
¯
¯ NOT PART
1 + n1 + n12 1 ¯¯
an = ≈ OF
3 + n42 3 ¯¯
OUR PROOF
1
So 3 seems the obvious candidate for our limit. To begin the proof then: Let ε > 0. Note
¯ ¯ ¯ 2 ¯
¯ ¯ ¯ ¯
¯an − 1 ¯ = ¯ n + n + 1 − 1 ¯
¯ 3¯ ¯ 3n2 + 4 3¯
3n − 1
=
3(3n2 + 4)
3n
6
3(3n2 + 4)
3n
6
3 · 3n2
1
< .
n

17
By the Archimedean Property, there exists N ∈ N such that N > 1ε . Then, for any n > N, we have
¯ ¯
¯ ¯
¯an − 1 ¯ < 1 6 1 < ε
¯ 3¯ n N

to complete the proof. (Note that N is by no means the first N that will suffice to guarantee |aN − L| < ε; what
matters is that it does suffice.)

• Let
(−1)n n2
an = (n > 1) .
n2 + 1
Then (an ) diverges.

Proof. Thoughts: Looking at the sequence we can see that for large even n

(−1)n n2 1
an = 2
= ≈ 1,
n +1 1 + n−2
whilst for large odd n we have
(−1)n n2 −1
an = 2
= ≈ −1.
n +1 1 + n−2
A natural way forward seems to be to assume the existence of a limit and argue (carefully!) that this limit would need
to be both near 1 and −1; this would be our required contradiction. In fact if we look in more detail at the sequence
we see that a2n > 45 for all n and a2n−1 6 − 12 , so we will take ε in such a way that 2ε < 45 + 12 = 1310 which is the
closest the even and odd terms get. Our proof begins thus:
Suppose then that a limit L exists and set ε = 12 . Then there exists N such that for n > N
¯ ¯
¯ (−1)n n2 ¯ 1
¯ ¯
¯ n2 + 1 − L¯ < 2 .

In particular, for even n > N we have

n2 1
−L< ,
+1n2 2
1 1 1 1 3
=⇒ L > − > − = .
1 + n−2 2 5/4 2 10

Similarly, for odd n > N we have

n2 1
L+ < ,
+1n22
1 1 1 1
=⇒ L < − 6 − = 0.
2 1 + n−2 2 2
3
The inequalities L > 10 and L < 0 give us our required contradiction.

5.6 A Useful Limit

Proposition 80 Let a be a real number with a > 1, and k be a positive integer. Then there exists a positive real
number c such that
an > cnk for n = 1, 2, 3, . . .
In particular,
nk
→ 0 as n → ∞.
an
Proof. Let a = 1 + b so that b > 0. Take n > k and note that
n−k 1
n > k + 1 =⇒ > .
n k+1

18
By the Binomial Theorem
µ ¶ µ ¶ µ ¶
n n 2 n
an = 1+ b+ b + ··· + bk+1 + · · · + bn
1 2 k+1
µ ¶
n
> bk+1
k+1
n (n − 1) · · · (n − k) k+1
= b
(k + 1)!
k
(n − k) k+1
> b
(k + 1)!
µ ¶k
bk+1 n−k
= nk
(k + 1)! n
bk+1
> k
nk .
(k + 1)! (k + 1)
| {z }
c

We have found c, such that a /n > c for n > k. If we set

n k
( )
a a2 ak bk+1
c = min , ,..., k, >0
1 2k k (k + 1)! (k + 1)k

then an /nk > c holds for n > 1.

Then for any natural number k and positive a there exists c > 0 such that an /nk+1 > c for all n > 1; hence
nk 1
0< 6 .
an cn
¯ k n¯
Given ε > 0, by the Archimedean there exists N such that ¯n /a ¯ < ε for all n > N. i.e. nk /an → 0 as n → ∞.

5.7 Tails
Definition 81 Let (an ) be a sequence, and let k be a natural number. Then the kth tail of (an ) is the sequence
n 7→ an+k i.e. it equals the sequence
(ak+1 , ak+2 , ak+3 , . . .)
which we will also write as (an+k )∞ ∞
n=1 or (an )k+1 .

In practice, though, we will not be interested in a specific kth tail so much as in some (unspecified) tail or all
tails past a certain point in the sequence. The tails give a way of focusing on the long-term behaviour of a sequence
ignoring any short-term aberrant behaviour at the start of a sequence. Whether or not a sequence converges purely
depends on the long-term behaviour of the sequence as we see in the next proposition.
Proposition 82 Let (an ) be a real sequence and let L ∈ R. Then the following three statements are equivalent.
(i) (an ) converges (to L);
(ii) some tail of (an ) converges (to L);
(iii) all tails of (an ) converge (to L).
Proof. We shall demonstrate the implications as (i) implies (iii), (iii) implies (ii) and (ii) implies (i).
(a) Suppose that (an ) converges to L and let k ∈ N, ε > 0. As (an ) → L then there exists N such that
∀n > N |an − L| < ε.
For such n, we have n + k > n > N and so
∀n > N |an+k − L| < ε.
Hence, for any k ∈ N, the kth tail of (an ) converges to L.
(b) (iii) clearly implies (ii).
(c) Suppose that the kth tail of (an ) converges to L. Let ε > 0. Then there exists N such that
∀n > N |an+k − L| < ε.
Hence
∀n > N + k |an − L| < ε
and we see that (an ) converges to L.

19
5.8 Complex Sequences
Definition 83 Let (zn ) be a sequence of complex numbers and let w ∈ C. We say that (zn ) converges w and write
(zn ) → w or lim zn = w if
∀ε > 0 ∃N ∈ N ∀n > N |zn − L| < ε.
Put more succinctly this states that |zn − L| → 0 as n → ∞.

Theorem 84 Let zn = xn + iyn . Then (zn ) converges if and only if (xn ) and (yn ) both converge.

Proof. Suppose that xn and yn both converge and that ε > 0. Set x = lim xn , y = lim yn and L = x + iy. By the
Triangle Inequality
|zn − L| = |(xn − x) + i (yn − y)| 6 |xn − x| + |yn − y| .
As xn → x and yn → y then we can find N1 and N2 such that

|xn − x| < ε/2 whenever n > N1 ,

|yn − y| < ε/2 whenever n > N2 .

So if n > max (N1 , N2 ) we have |zn − L| < ε/2 + ε/2 = ε and we see that zn → L.
Conversely suppose that zn converges to L and let ε > 0. Then there exists N ∈ N such that |zn − L| < ε whenever
n > N . As |Re w| 6 |w| and |Im w| 6 |w| for any w ∈ C then

|xn − x| = |Re (zn − L)| 6 |zn − L| < ε whenever n > N,

|yn − y| = |Im (zn − L)| 6 |zn − L| < ε whenever n > N.

Hence xn → x and yn → y as required.

Example 85 Let µ ¶n
1
zn = .
1+ı
Then zn → 0.

Proof. Let ε > 0. Note ¯µ ¶n ¯ ¯ ¯ µ ¶n

¯ 1 ¯ ¯ 1 ¯n 1 1
¯
|zn − 0| = ¯ ¯ = ¯ ¯ = = √ .
1 + ı ¯ ¯1 + ı¯ |1 + ı|n 2
¡√ ¢n
We have already shown that 2−k < ε for k > 1/ε and so 2 < ε for n > 2/ε.

5.9 Uniqueness of Limits

Theorem 86 (Uniqueness of Limits) Let (an ) be a real (or complex) sequence and suppose that an → L1 and an → L2
as n → ∞. Then L1 = L2 .

Proof. Suppose not and set ε = |L1 − L2 | > 0. Then ε/2 > 0 and there exists N1 such that

n > N1 =⇒ |an − L1 | < ε/2 (7)

Likewise there exists N2 such that

n > N2 =⇒ |an − L2 | < ε/2. (8)
Then for n > max(N1 , N2 ) we have

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

6 |L1 − an | + |an − L2 | by the ∆ law
< ε/2 + ε/2 = |L1 − L2 |

which is the required contradiction.

20
5.10 Limits respect Weak Inequalities
Theorem 87 Let (an ) and (bn ) be real sequences such that (an ) → L and (bn ) → M . If an > bn for all n then
L > M.

Proof. Suppose, for a contradiction, that L < M . Set ε = (M − L) /2 > 0.

As an → L then there exists N1 such that n > N1 =⇒ |an − L| < ε;

as bn → M then there exists N2 such that n > N2 =⇒ |bn − M | < ε.

So
L+M
n > N1 =⇒ = L − ε < an
2
L+M
n > N2 =⇒ bn < M + ε = .
2
Hence for n > max(N1 , N2 ) we have
L+M
bn < < an
2
which contradicts an > bn for all n.

Remark 88 Clearly lim does not respect strict inequalities. For example, 1
n > 0 for all n > 1 but 0 = lim n1 > lim 0 = 0
is false.

Theorem 89 (Sandwich Rule) Suppose that xn 6 an 6 yn for all n and that

L = lim xn = lim yn .

Then an → L as n → ∞.

Proof. Let ε > 0. Then there exist N1 and N2 such that

xn − L > −ε for all n > N1 ,

yn − L < ε for all n > N2 .

So for n > max (N1 , N2 ) we have

−ε < xn − L 6 an − L 6 yn − L < ε,
which shows that an → L also.

5.11 Some notation

Let an , bn be sequences. We write an = O(bn ) if there exist c such that for some N

n > N =⇒ |an | < cbn .

We write an = o(bn ) if an /bn is defined and

an
→ 0.
bn
Example 90 Supposing for the moment that we can prove all the usual trigonometric results we can write (true)
statements like
¡ ¢
• n = O n2
¡ ¢
• n = o n2
• sin n = O (1)
• sin n = o (n)

21
5.12 Infinity
Definition 91 Let an be a sequence of real numbers. We say "an tends to infinity" and write an → ∞ as n → ∞ to
mean
∀M ∈ R ∃N ∈ N ∀n > N an > M.
Similarly we write bn → −∞ if
∀M ∈ R ∃N ∈ N ∀n > N bn < M.
(Here we tend to think of M as being a very large positive/negative number.)

Theorem 92 Let (an ) be a sequence of positive real numbers. The following are equivalent:
(a) an → ∞ as n → ∞;
(b) 1/an → 0 as n → ∞.

Proof. (a) implies (b): Let ε > 0 and set M = 1/ε. As an → ∞ then there exists N such that an > M for all n > N .
But then 0 < 1/an < ε for all n > N as required.
(b) implies (a): Let M ∈ R and ε = 1/ (|M | + 1). As 1/an → 0 then there exists N such that 1/an < ε for all
n > N . But then an > 1/ε = |M | + 1 > M for all n > N as required.

6 The Algebra of limits

Most sequences can be built up from simpler ones using addition, multiplication, etc. The algebra of limits tells us
how the corresponding limits behave.

Throughout the following (an ) and (bn ) denote real or complex sequences.

Proposition 93 (AOL: Constants) If an = a for all n, then an → a.

Proof. For any ε > 0, take N = 1; n > N =⇒ |an − a| = 0 < ε.

Proposition 94 (AOL: Sums) If an → a and bn → b then an + bn → a + b.

Proof. Let ε > 0. Then ε/2 > 0 and so

∃N1 : n > N1 =⇒ |an − a| < ε/2,

∃N2 : n > N2 =⇒ |bn − b| < ε/2.

Put N3 = max(N1 , N2 ). Then

n > N3 =⇒ |(an + bn ) − (a + b)|

6 |an − a| + |bn − b| by the ∆ law
< ε/2 + ε/2
= ε

Proposition 95 (AOL: Scalar Products) If an → a as n → ∞ and λ ∈ R (or C) then λan → λa.

Proof. Let ε > 0. Then ε/ (|λ| + 1) > 0 and so there exists N such that |an − a| < ε/ (|λ| + 1) for all n > N . Hence

|λ| ε
|λan − λa| = |λ| |an − a| 6 <ε
|λ| + 1

for all n > N .

Corollary 96 (AOL: Differences) If an → a and bn → b then an − bn → a − b.

Corollary 97 (AOL: Translations) If an → a and c ∈ R (or C) then an + c → a + c.

22
Lemma 98 If xn → 0 and yn → 0 then xn yn → 0.

Proof. Given ε > 0, let ε1 = min(1, ε) > 0. Then

∃N1 : n > N1 =⇒ |xn | < ε1 ,

∃N2 : n > N2 =⇒ |yn | < ε1 .

So if n > max(N1 , N2 ) we have

|xn yn | 6 |xn | |yn | < ε21 6 ε1 6 ε,
which completes the proof.

Proposition 99 (AOL: Products) If an → a and bn → b then an bn → ab.

Proof. Note that

an bn − ab = (an − a)(bn − b) + b(an − a) + a(bn − b),
that (an − a) (bn − b) → 0 by the previous lemma, that b (an − a) → 0 and a (bn − b) → 0 by Proposition 95. Hence
an bn → ab by Proposition 94.

Proposition 100 (AOL: Reciprocals) If an → a and an 6= 0 for all n and a 6= 0 then 1/an → 1/a.

Proof. Let ε > 0. As a 6= 0 then |a| /2 > 0. So there exists N1 such that for n > N1 we have |an − a| < |a| /2. By the
Triangle Inequality
|a| 6 |an | + |a − an | = |an | + |an − a|
and so |an | > |a| /2 and |1/an | < 2/ |a| .
Further, as |a| ε/2 > 0 then there exists N2 such that for n > N2
2

2 ε
|an − a| < |a| .
2
Set N3 = max(N1 , N2 ) so that for n > N3
¯ ¯
¯ 1
¯ 1 ¯¯ |an − a| ³ 2 ε ´ 2 1
¯ an − a¯
=
|an ||a|
< |a|
2 |a| |a|
= ε.

Corollary 101 (AOL: Quotients) If an → a, bn → b, and bn 6= 0 for all n and b 6= 0, then an /bn → a/b.

Proposition 102 (AOL: Modulus) If an → a then |an | → |a| .

Proof. The inequality

||an | − |a|| 6 |an − a|
follows from the ∆ law as
|a| 6 |an − a| + |an | and |an | 6 |an − a| + |a|
from which we get (as required)
||an | − |a|| 6 |an − a| → 0.

23
6.1 Examples
n2 + n + 1 1
an = →
3n2 + 4 3
Proof. We write
n2 + n + 1 1 + n1 + n12 1+0+0 1
2
= 1 → =
3n + 4 3 + 4 n2 3+0 3
noting
1
• n → 0 by the Archimedean Property;
1
• n2 → 0 by Proposition 99;
• 1 → 1;
1 1
• 1+ n + n2 → 1 by Proposition 94;
4
• 3+ n2 → 3 by Proposition 94;
1 1
• 3+ n42
→ 3 by Corollary 101;
1
• an → 3 by Proposition 99.

Example 103 Suppose a1 = 1, a2 = 1, and we recuresively define

an+2 = an+1 + an , for n > 1.
It is easy to prove by induction on n that there is then a unique sequence of natural numbers satisfying these require-
ments. They are called the Fibonacci numbers.
Proposition 104 an+1 /an is convergent.
Proof. By induction, an > 1 for all n. So for n > 1
µ ¶ µ ¶−1
an+2 an+1
=1+ .
an+1 an
Write xn = an+1 /an for n > 1. Note that xn > 0 for all n. Then
x1 = 2 and xn+1 = 1 + 1/xn .
1 1
Suppose that we did have convergence and that xn → x. By Tails xn+1 → x and 1 + xn →1+ x by AOL. So
1
x=1+
x
√
by the Uniqueness of Limits. Hence x2 − x − 1 = 0 giving x = 1±2 5 . But xn > 0 for all n, and so x > 0 by Theorem
87 giving √
1+ 5
x= > 1.
2
√
Now we will show that xn is convergent to the real number 1+2 5 , which we will denote ϕ, and is called the golden
ratio.
1 1 1 1 1 xn − ϕ
xn+1 − ϕ = 1 + −ϕ=1+ −1− = − =
xn xn ϕ xn ϕ xn ϕ
as ϕ2 = ϕ + 1. So ¯ ¯
¯ xn+1 − ϕ ¯ 1 1 1
¯ ¯
¯ xn − ϕ ¯ = |xn ||ϕ| = ϕxn 6 ϕ
as xn > 1 for all n. By induction we get
1 1
n
6 xn − ϕ 6 n
−
ϕ ϕ
1
and are done by the Sandwich Rule, since ϕ > 1 and so ± n → 0.
ϕ

24
6.2 The Relative Orders of Terms
Our first thoughts, when needing to consider the long term behaviour of a sequence which has various components
to it, should be on which terms will dictate the sequence’s behaviour in the long term. Usually, for this, we need to
appreciate the relative magnitudes of the terms as n becomes large. As a rule of thumb, when it comes to the long
term behaviour of functions

trig functions and constants < logarithms < polynomials < exponentials.

More precisely:

• |cos n| 6 1 and |sin n| 6 1 for all n.

• For any rational q > 0, log n/nq → 0 as n → ∞.
• For any a > 1 and polynomial p then p (n) /an → 0 as n → ∞.

Example 105 Qualitatively describe the long-term behaviour of the following sequences.

• µ ¶
n6 + 7n2
(−1)n
2n
This will tend to 0 (albeit in an oscillatory way) as the dominant term is 2n .
• µ ¶
2n + 3
cos n.
3n + 8
At first glance the polynomial terms seem dominant. But being of the same magnitude, and working to counter
one another, we see (2n + 3) / (3n + 8) → 2/3. So actually it is the oscillating behaviour of cos n which stops
the sequence from converging.
• µ ¶
log n 2n − n
√ cos .
n n2 + 3n − 6
¡ 2 ¢
As |cos θ| 6 1 for all θ then√the cosine takes the sting out of n
√ the term (2 − n) / n + 3n − 6 which is just a red
herring. In the long term n dominates log n and log n/ n → 0. The messy cosine term has no crucial effect
on this behaviour.

Exercise 106 How would you make these first thoughts into rigourous proofs using the Algebra of Limits, Sandwich
Rule, etc.?

25
7 Monotone Sequences
We now turn to a crucially important sort of sequence.
Definition 107 We say that a real sequence (an ) is monotone increasing if an 6 am whenever n < m.
We say that a real sequence (an ) is monotone decreasing if an > am whenever n < m.
We say that a real sequence (an ) is strictly monotone increasing if an < am whenever n < m.
We say that a real sequence (an ) is strictly monotone decreasing if an > am whenever n < m.
We say that a real sequence (an ) is monotone if it is either monotone decreasing or monotone increasing.
Example 108 Let an = n. Then (an ) is monotone increasing. So is an = (2n + 1)2 .
Example 109 (Decimal Expansions) Let 0 < x 6 1. Then there is a unique sequence of integers a1 , a2 , a3 , . . . such
that
(i) 0 6 an 6 9 for each n;
(ii) for each n,
Xn
1 ak
x− k 6 < x;
10 10k
k=1
(iii)
n
X ak
lim = x.
n→∞ 10k
k=1
Solution. We will proceed inductively. The integer a1 needs to satisfy
1 a1
x− 6 < x =⇒ 10x − 1 6 a1 < 10x.
10 10
The interval [10x − 1, 10x) contains a unique integer a1 and further, as −1 < 10x − 1 6 a1 < 10x = 10 then 0 6 a1 6 9.
Suppose now, as our inductive hypothesis, that a1 , a2 , . . . , aN have been uniquely found satisfying (i) and (ii).
Then
N+1
X ak
1
x − N +1 6 <x
10 10k
k=1
N
X N
X
1 ak aN +1 ak
⇐⇒ x − − 6 < x −
10N +1 10k 10N +1 10k
k=1 k=1
N
X N
X
⇐⇒ 10N+1 x − 10N+1−k ak − 1 6 aN+1 < 10N +1 x − 10N +1−k ak .
k=1 k=1
There is a unique integer in this range, and we set aN +1 to be this integer. Further, by hypothesis,
à N
!
X
aN+1 > 10N+1 x − 10−k ak − 1 > −1,
k=1
à N
!
X 1
aN+1 < 10 N+1
x− 10 −k
ak 6 10N +1 × = 10.
10N
k=1
So 0 6 aN +1 6 9 as required. Applying the sandwich rule to
X ak n
1
x− k
6 <x
10 10k
k=1
we find n
X ak
lim = x.
n→∞ 10k
k=1
This sequence is called the decimal expansion of x and we write
x = 0.a1 a2 a3 . . .

Remark 110 In the sense of the above example 15 would have decimal expansion 0.1999 . . . rather than 0.200 . . .To
avoid any ambiguity for those reals with two different decimal expansions (in the usual sense) the above example chooses
decimal expansions whose terminating decimal expansions never equal the real in question.

26
7.1 Monotone Bounded Sequences
Definition 111 We say that a real sequence (an ) is bounded above if the set S = {an : n ∈ N} is bounded above.

Definition 112 We say that a real sequence (an ) is bounded below if the set S = {an : n ∈ N} is bounded below.

Theorem 113 Let (an ) be a monotone increasing, bounded above sequence. Then (an ) converges.

Proof. Let L = sup{an : n ∈ N}; this exists by the completeness axiom as the set is bounded above and non-empty.
Let ε > 0. By the Approximation Property there exists N ∈ N such that

L − ε < aN 6 L.

As the sequence is monotone increasing then for any n > N

L − ε < aN 6 an 6 L,
=⇒ ∀n > N |an − L| < ε.

That is an → L

Corollary 114 Let (an ) be a monotone decreasing, bounded below sequence. Then (an ) converges.
√
Example 115 Let x1 = 0 and for all n > 1 let xn+1 = 2 + xn .

1. (By induction if necessary, we show that) There exists a unique sequence defined so.
2. By induction: xn > 0 for all n.
3. (xn ) is monotone; more precisely we prove by induction: xn+1 > xn for n = 1, 2, 3, . . .
√
The root case is 2 > 0; the inductive step follows from the identity

x2n+2 − x2n+1 = (2 + xn+1 ) − (2 + xn ) = xn+1 − xn .

4. (an ) is bounded above: more precisely prove by induction that xn 6 2 for all n.
√ √
The root case is x1 = 0 < 2. For the inductive step, xn+1 = 2 + xn 6 2 + 2 = 2.
5. So by the previous theorem xn → L for some L. To evaluate L we note

x2n+1 = xn + 2.

Letting n → ∞ in both sides we can see

By Tails xn+1 → L
AOL x2n+1 → L2
AOL xn + 2 → L + 2
Uniqueness of limits L2 = L + 2

So L = −1 or L = 2. But xn > 0 for all n and therefore L > 0 by the preservation of weak inequalities. Hence
L = 2.
6. To further understand the sequence it is an easy check to note xn = 2 cos (π/2n ) .

27
 We can better appreciate, with a graph, just how the sequence xn emerges
√ and also appreciate how a change of
initial value would change the sequence. Below is sketched graphs of y = 2 + x and y = x.

2.5
yx

2.0
2,2
x3 ,x3 
1.5
x2 ,x2 
y 2  x
1.0

0.5

x1 ,x1 
2 1 0 1 2 3
√ √ √
We have x1 = 0 and can calculate x2 = 2 + 0 = 2 by taking the y-co-ordinate of the graph y = 2 + x
above x = x1 . This gives us the point (x1 , x2 ) and if we move horizontally √ to the point line y = x we are now at
the point (x2 , x2 ) labelled in the diagram. Moving up to the graph y = 2 + x and across to y = x we are now at
(x3 , x3 ). We see if we continue in this way our lines burrow into the tight space between the graphs at (2, 2) as the xn
rapidly converge on 2. Pictorially we can also appreciate that for any initial value in the range −2 < x1 < 2 a similar
convergence monotonically up to 2 would again occur (e.g. see left dashed lines) whilst if we started with x1 > 2 the
sequence would converge monotonically down to 2 (see right dashed lines).
Of course the diagram doesn’t prove anything! But such diagrams can be very informative qualitatively.

7.2 Reprise — The Uncountability of the Reals

Now that we have a notion of a decimal expansion we are in a position to describe Cantor’s proof of the uncountability
of R, (a fact that was already demonstrated in Theorem 56.
Theorem 116 R is uncountable.
Proof. We will show this by showing that the interval (0, 1] is uncountable. To each x in this interval corresponds
a unique decimal expansion (in the sense of Example 109); further to each non-zero decimal expansion 0.a1 a2 a3 . . .
which does not end in a string of zeros there corresponds a real number as the terminating decimal expansions form
a bounded monotone sequence between 0 and 0.999 . . . = 1.
Suppose now (for a contradiction) that f : N → (0, 1] is a bijection. Then we may uniquely write out the decimal
expansions of f (1) , f (2) , . . . . Say:
f (1) = 0.r11 r12 r13 r14 . . .
f (2) = 0.r21 r22 r23 r24 . . .
f (3) = 0.r31 r32 r33 r34 . . .
Cantor then created a real α not on the list by setting
α = 0.a1 a2 a3 . . .
where ½
6 if rkk =
6 6
ak =
7 if rkk = 6
The decimal expansion of α is allowed (in the sense of Example 109) and we see, for any k, that α 6= f (k) as α and
f (k) disagree in the kth decimal position. This contradicts the surjectivity of f .

28
8 Subsequences
1
Example 117 Let an = n2 i.e. µ ¶
1 1 1 1
(an ) = 1, , , , , . . .
4 9 16 25
We can get new sequences by looking at
¡1 1 1 ¢
everything after second place ¡ 9 , 116 ,125 , . . ¢.
all odd terms ¡1, 91 , 25 ,... ¢
all prime terms 1, 4 , 19 , 25
1
, 49 1
,...
etc.

However, when we are talking about ‘subsequences’ we usually want to keep a grip on not just the sequence we extract
in such a way, but also the way we extract it. Our definition will do that.
¡ ¢
Definition 118 Let (an ) be a sequence. A subsequence of (an ) is a sequence af (n) where (f (n)) is a strictly
monotone increasing sequence of natural numbers together with the sequence (f (n)) . Often we write nr for f (r)
∞
and write a subsequence as (anr ) or (anr )r=1 .

So a subsequence is both a sequence, and a memory of how the subsequence was extracted; there may be more than
one way to extract the same sequence from the original sequence.

Example 119 Let

¡ ¢
(an ) = n2 = (1, 4, 9, 16, . . .)
(bn ) = (0) = (0, 0, 0, 0, . . .)
(f (n)) = (2n) = (2, 4, 6, 8, . . .)
(g (n)) = (2n − 1) = (1, 3, 5, 7, . . .)

Then
¡ ¢
af (n) = (a2n ) = (4, 16, 36, 64, . . .)
¡ ¢
ag(n) = (a2n+1 ) = (1, 9, 25, 49, . . .)
¡ ¢
bf (n) = (b2n ) = (0, 0, 0, 0, . . .)
¡ ¢
bg(n) = (b2n+1 ) = (0, 0, 0, 0, . . .)

So whilst (b2n ) and (b2n+1 ) are the same sequence we would consider them different subsequences of (bn ).

Proposition 120 Suppose that the sequence (an ) converges to L. Then every subsequence (anr ) also converges to L.

Proof. Let ε > 0. Then there exist N such that

n > N =⇒ |an − L| < ε

As r 7→ nr is increasing then nr > r for all r and so

r > N =⇒ nr > N =⇒ |anr − L| < ε.

The converse in the form "if all subsequences of (an ) converge to L then (an ) → L" is true because the whole sequence
is a subsequence of itself. However that just one subsequence converges is not enough to guarantee convergence of the
whole sequence. e.g. an = (−1)n which is divergent despite a2n → 1.

29
8.1 Bolzano—Weierstrass Theorem
Theorem 121 Let (an ) be a real sequence. Then (an ) has a montone subsequence.

Proof. We consider the set V = {k ∈ N : m > k =⇒ am < ak }. This is the set of "scenic viewpoints” — were we to
plot the points (k, ak ) then from a scenic viewpoint we can see all the way to ∞ with no greater an getting in the way.
There are two cases to consider: the set V is either finite or infinite.

1. |V | is infinite. If we list the elements of V in increasing order: k1 < k2 < . . . then (akr ) is subsequence and

r > s =⇒ kr > ks =⇒ akr < aks

i.e. (akr ) is monotone decreasing.

2. |V | is finite. Let m1 be the last viewpoint and consider am1 +1 .
As m1 + 1 is not a viewpoint then there exists m2 > m1 + 1 such that am2 > am1 .
As m2 is not a viewpoint then there exists m3 > m2 such that am3 > am2 .
...
Continuing in this way and we can generate a monotone increasing sequence (amk ) .

Theorem 122 (Bolzano-Weierstrass Theorem) Let (an ) be a real bounded sequence. Then (an ) has a convergent
subsequence.

Proof. By the previous theorem (an ) has a monotone sequence which is also bounded. By Theorem 113 this
subsequence converges.

Theorem 123 (Bolzano-Weierstrass Theorem in C) A bounded sequence in C has a convergent subsequence.

Proof. Let (zn ) be a bounded sequence in C — let’s say |zn | < M for all n. If we write zn = xn + iyn then we have
|xn | < M and |yn | < M for all n. So (xn ) and (yn ) are bounded sequences. By the Bolzano-Weierstrass Theorem (xn )
has a convergent
¡ subsequence
¢ (xnk ) which converges to L1 , say. As (ynk ) is also bounded then it has a convergent
subsequence
¡ y
¢ nkr which converges to L2 , say. ¡ ¢
As xnkr is a subsequence of (xnk ) then it too converges to L1 by Proposition 120. We then have that znkr
converges to L1 + iL2 as its real and imaginary parts converge — see Theorem 84.

8.2 Limit points

Here is alternative way of phrasing the Bolzano-Weierstrass Theorem.

Definition 124 Let S ⊆ R We say that x is a limit point or accumulation point of S if for every ε > 0 there
exists y ∈ S, such that
0 < |y − x| < ε.

Note that x itself need not be in the set.

Example 125 The set of limit points of (0, 1) is [0, 1]

The set of limit points of Q is R.
The set of limit points of Z is ∅.

Remark 126 The Bolzano-Weierstrass Theorem can be rephrased as: "An infinite bounded subset has a limit point."
Given such a set, S, then we can select a sequence (xn ) of points of S and by the Bolzano-Weierstrass Theorm this
sequence converges to a limit L. It is not hard to show that L is then a limit point of {x1 , x2 , x3 , . . .} and so of the set
S.

30
9 The Cauchy Criterion
A first difficulty in proving a sequence converges is deciding upon a suitable candidate for the limit. Cauchy saw that
it was enough to show that if the terms of the sequence got sufficiently close to each other, then completeness will
guarantee convergence. In fact, Cauchy’s insight can be used to construct the reals from the rationals so that we could
show the existence of a complete ordered field rather than assuming that a field satisfying all our axioms exists — but
we shall leave this for later foundational courses.
Definition 127 Let (an ) be a real or complex sequence. We say that (an ) is a Cauchy sequence, or simply is
Cauchy, if
∀ε > 0 ∃N ∈ N ∀m, n > N |am − an | < ε.
Note that the definition makes no mention of a limit, but we shall see that this criterion is equivalent to convergence
in R or C (but not in Q!).
Proposition 128 A (real or complex) Cauchy sequence is bounded.
Proof. Let (an ) be a real or complex Cauchy sequence. Taking ε = 1, we know there exists N such that
|an − aN | < 1 whenever n > N.
Hence, by the Triangle Inequality
|an | < |aN | + 1 for all n > N.
So for all m (including those m < N ) we have
|am | 6 max {|a1 | , |a2 | , . . . , |aN −1 | , |aN | + 1}
and we see that the sequence is bounded.
Proposition 129 A convergent sequence is Cauchy.
Proof. Let (an ) be a convergent sequence with limit L, and let ε > 0. Then there exists a natural number N such
that
|ak − L| < ε/2 for all k > N.
So for all m, n > N we have, by the Triangle Inequality
ε ε
|am − an | 6 |am − L| + |L − an | < + = ε,
2 2
and we see that (an ) is Cauchy.
Lemma 130 If (an ) is a real or complex Cauchy and a subsequence (ank ) converges to L then (an ) converges to L.
Proof. Let ε > 0. So there exists K ∈ N such that
|ank − L| < ε/2 whenever k > K.
As the sequence (an ) is Cauchy then there exists N ∈ N such that
|an − am | < ε/2 whenever m, n > N.
If we select take k > max (K, N ) so that nk > N then we have, by the Triangle Inequality
ε ε
|an − L| 6 |an − ank | + |ank − L| < + = ε for all n > N
2 2
and the proof is complete.
Theorem 131 A real or complex Cauchy sequence is convergent.
Proof. Let (an ) be a real or complex Cauchy sequence. By Proposition 128 (an ) is bounded, and so by the Bolzano-
Weierstrass Theorem (Theorems 122 and 123) (an ) has a convergent subsequence (ank ). By the previous lemma (an )
converges to the same limit.

We have then establishded the Cauchy Convergence Criterion for real and complex sequences:—
(an ) is convergent ⇐⇒ (an ) is Cauchy.

31
 √
Example 132 The terminating decimal expansions of 2, namely the sequence (qn ):

1, 1.4, 1.41, 1.414, . . .

is a sequence of rational numbers which is Cauchy (for example, because it is a convergent real sequence) but it is not
convergent in the rationals — i.e. it does not satisfy

∃L ∈ Q ∀ε > 0 ∃N ∈ N ∀n > N |qn − L| < ε.

Example 133 The log 2 sequence. For n ∈ N let

1 1 1
an = 1 − + + · · · + (−1)n+1 .
2 3 n
Then with m > n > 0, and m − n even we have
¯z }| {z }| { z }| {¯¯
¯
¯ 1 1 1 1 1 1¯
|am − an | = ¯¯ − + − +...+ − ¯¯ [grouped in positive pairs]
¯n + 1 n + 2 n + 3 n + 4 m − 1 m¯
1 1 1 1 1 1
= − + −...− + − [one initial term then grouped in negative pairs]
n+1 n+2 n+3 m − 2 m − 1 |{z}
m
| {z } | {z } | {z }
1
6 .
n+1
If m − n is odd, we write
¯z }| {z }| { z }| { ¯
¯ ¯
¯ 1 1 1 1 1 1 1¯
|am − an | = ¯¯ − + − +...+ − + ¯¯ [grouped in positive pairs and one final term]
¯n + 1 n + 2 n + 3 n + 4 m − 2 m − 1 m¯
1 1 1 1 1
= − + −...− + [one initial term then grouped in negative pairs]
n+1 n+2 n+3 m−1 m
| {z } | {z } | {z }
1
6
n+1

Let ε > 0 and take N > 2ε1

. Then |an − am | < ε whenever m, n > N and we see that (an ) is Cauchy. This shows that
the sequence is convergent even though we currently have no idea of its limit. In due course we shall see that the limit
is log 2.

32
10 Series
This is another of those places where to make real progress we have to stand back a little, put out of our minds
pre-conceptions, then make clear definitions and see what follows logically from them. Looking back at our axioms,
we see that given any pairP of real numbers a, b we can form the sum a + b. By mathematical induction, we can, for
n
any n ∈ N, form the sum 1 ak of any n-tuple of natural numbers. (The associative law means we don’t even have
to worry about where the parentheses go.)
What our axioms don’t do is licence usPto start writing down infinite sums, and behaving as though the mere
act of writing down similar looking signs ( ∞1 , say) entitles us to assume that all the properties of finite sums hold.

Definition 134 Let (an )∞

n=1 be a sequence of (real or complex) numbers. For n ∈ N, the nth partial sum is the
finite sum is
Xn
sn = ak = a1 + a2 + · · · + ak .
k=1

By the series
∞
X X
ak or just ak
k=1

we mean the sequence of partial sums (sn ).

P
Example 135 (i) The Geometric Series. Let x ∈ C, and let an = xn Then xn is

(1, 1 + x, 1 + x + x2 , . . . , 1 + x + x2 + · · · + xn , . . . )
P1
(ii) The Harmonic Series. Let an = n1 . Then n is
µ ¶
1 1 1
1, 1 + , 1 + + , . . .
2 2 3
P n
(iii) The Exponential Series. Let x ∈ C and let an = xn /n!. Then x /n! is
µ ¶
x2
1, 1 + x, 1 + x, 1 + x + , . . .
2!

(iv) The Cosine Series. Let x ∈ C and set

(
x2m m
an = (2m)! (−1) if n = 2m
0 otherwise
P
Then an is µ ¶
x2 x2 x2 x4
1, 1, 1 − , 1 − , 1 − + , ...
2! 2! 2! 4!
P
Definition 136 Let (an ) be a (real or complex) sequence. We say that the series ∞ 1 ak converges (resp. diverges)
if the sequence (sn ) of partial sums converges (resp. diverges). If sn → L as n → ∞ then we write
∞
X
ak = L.
k=1

We refer to L as the sum of the series.

P∞
P∞ 137 Our earlier results regarding the tails of sequences still apply — it follows that 1 ak converges
Remark P if and
only K ak for some K. Consequently it makes sense to discuss the convergence (or otherwise) of ak without
needing to identify the initial value. But to calculate the sum of a convergent series we do need to specify the initial
value.

Example 138 Let an = xn for n > 0 where x ∈ C.¡ ¢

(a) If x 6= 1 then snP= 1 + x + x2 + · · · + xn = 1 − xn+1 / (1 − x) .
(b) If |x| < 1 then P xn is convergent noting xn → 0 and using the algebra of limits.
(d) If |x| > 1 then
n
xn is divergent as |sn − sn−1 | = |x| 9 0 and so (sn ) is not Cauchy.

33
 1
P 1
Example 139 Let an = n. Then n is divergent. To prove this note

1 1
s2n = 1 +
+ ··· + n =
2 2
¡1¢ ¡1 1¢ ¡1 1 1 1¢ 1
= 1 + 2 + 3 + 4 + 5 + 6 + 7 + 8 +... + (· · · + 2n )
2n−1
> 1 + 1
2 + 2
4 + 4
8 + ... + 2n
> 1+ n
2

so (sn ) has a subsequence which is divergent and so is itself divergent.

P 1
Example 140 Let an = n12 . Then n2 is convergent.

Proof. Clearly the partial sums form an increasing sequence. The trick here is to note
n
X n
X n ½
X ¾ n−1
X1 X n
1 1 1 1 1 1
sn = 61+ =1+ − =1+ − = 1 + 1 − 6 2.
k2 k(k − 1) k−1 k k k n
k=1 k=2 k=2 k=1 k=2

Hence (sn ) is a bounded-above increasing sequence and so convergent. [In due course we will meet, with the Integral
Test, a systematic way of dealing with such series and won’t have to resort to tricks.]
Applying Cauchy’s Criterion for convergence for sequences to series (which, recall, is just a sequence of partial
sums) we have
P
Theorem 141 (Cauchy’s Criterion for Series) The series ∞ 0 ak converges if and only if for all ε > 0 there exists
N ∈ N such that for all m, n > N we have
¯ n ¯
¯X ¯
¯ ¯
|sn − sm | = ¯ ak ¯ < ε.
¯ ¯
m+1

10.1 Conditional and Absolute Convergence

P
Definition
P 142 Let (an ) be a real or complex sequence. Then we say that an is absolutely convergent if the series
|an | converges. A series which is convergent, but not absolutely convergent, is called conditionally convergent.

Theorem 143 An absolutely convergent series is convergent

P
Proof. Suppose that an converges absolutely and let ε > 0. By Cauchy’s Criterion there exists N ∈ N such that
¯ l ¯
¯X ¯
¯ ¯
l > k > N =⇒ ¯ |an |¯ < ε.
¯ ¯
k+1

By the Triangle Inequality ¯ l ¯ ¯ l ¯

¯X ¯ X l ¯X ¯
¯ ¯ ¯ ¯
l > k > N =⇒ ¯ an ¯ 6 |an | = ¯ |an |¯ < ε
¯ ¯ ¯ ¯
k+1 k+1 k+1

P∞
Example 144 (i) 0 xn absolutely convergent if |x| < 1.
P∞ (−1)n
(ii) 1 n2 is absolutely convergent.
P sin n
(iii) ∞
1 n3 is absolutely convergent.
P∞ (−1) n+1
(iv) 1 n is conditionally convergent. (See Examples 133 and 139).
P
Definition 145P Let p : N −→ N be a bijection and set bn = ap(n) and consider bn , which we call a rearrangement
of the series an .

Example 146 (See Exercise Sheet 6) If we rearrange the log 2 series from Example 133 then we can change the sum:—
1 1 1
1−
+ − + · · · = log 2
2 3 4
1 1 1 1 1 3
1 + − + + − + ··· = log 2.
3 2 5 7 4 2

34
 P P
Theorem 147 If an is absolutely convergent then ap(n) is absolutely convergent for any rearrangement p and
∞
X ∞
X
an = ap(n)
1 1
P∞
Theorem 148 [Beyond syllabus] (Riemann) If 1 an is a real conditionally convergent series and −∞ 6 L 6 ∞
then there exists a bijection p : N → N such that
∞
X
ap(n) = L.
1

10.2 Multiplication of series

P∞ P∞
Theorem 149 Suppose 0 an and 0 bn are absolutely convergent. For each n ∈ N we set
n
X
cn = ak bn−k .
k=0
P∞
Then 1 cn is absolutely convergent and
∞
Ã∞ !à ∞ !
X X X
cn = an bn
0 0 0
Pn
Proof. Let N ∈ N and set Cn = |ak | |bn−k |. By the Triangle Inequality and simple algebra
k=0

N N N N
Ã∞ !à ∞ !
X X X X X X
|cr | 6 |Cr | 6 |an | |bn | 6 |an | |bn |
0 0 0 0 0 0
P P P
So the montonic series |cn | and |Cn | are bounded above and so convergent. Hence cn is convergent.
Given ε > 0 there exist N such that
Xl
l > k > N =⇒ |Cn | < ε.
k+1

Then for k > N

¯ 2k ¯ ¯¯ ¯
¯ X
¯X k
X k
X ¯ ¯ X ¯ 2k
X
¯ ¯
¯ cn − an bn ¯ = ¯¯ ar bs ¯¯ 6 |ar bs | 6 |Cn | < ε.
¯ ¯ ¯ ¯ ditto
n=0 n=0 n=0 r+s6 2k, r> k or s> k k
P P P
We can see that the sum cn and product of sums an bn differ my at most ε for any ε > 0, and so are equal.

Example 150 For |x| < 1,

∞
X 1
(n + 1)xn =
0
(1 − x)2

Proof. Set an = xn , bn = xn , so that X

cn = xr xs = (n + 1)xn .
r+s=n

Example 151 For x, y ∈ C Ã∞ !à ∞ !

X xn X yn X (x + y)n
=
0
n! 0
n! n!
n n P P
Proof. Let an = xn! , bn = yn! . Then the series an and bn are absolutely convergent (check — or wait for later
discussion of the Ratio Test). Then
X xr y s (x + y)n
cn = =
r+s=n
r! s! n!
by the Binomial Theorem.

35
11 Some Tests for Convergence
Here we discuss some classic tests for convergence and divergence. The idea that there are ‘tests’ is very attractive,
but in practice (for problems arising from real-word situations) these tests may not apply. However the tests do give
us clues, suggest ways of thinking about series, what sort of estimates need to be made, and a sense of the relative
magnitude of terms.
P
Proposition 152 (A Simple Test for Divergence)PIf an converges then an → 0. In practice, the contrapositive is
used more, namely: if an does not tend to 0 then an diverges.
P
Proof. If ∞ 1 an = L then
an = sn − sn−1 → L − L = 0
by the Algebra of Limits.
P 1
Remark 153 The converse is far from true. For example, n diverges yet 1/n → 0 as n → ∞.

Theorem 154 (The Comparison Test) Suppose (an ) and (bn ) are real sequences and 0 6 an 6 bn . Then
P P
• bn is convergent =⇒ an is convergent;
P P
• an is divergent =⇒ bn is divergent.

Proof. Note that theP second statement is just the contrapositive

P of the first, and so it is enough to just prove the
first. Suppose that bk converges. Then the partial sums n1 ak satisfy
n
X n
X ∞
X
ak 6 bk 6 bk
1 1 1

and hence form an increasing bounded sequence which converges.

Example 155 The following sequences

∞
X ∞
X ∞
X
1 xn
n−5/2 , , where |x| < 1,
1 1
n (n + 1) (2 + cos n) 1
n

all converge.
P −2
Proof. (i) This converges by comparison with
P −2 n .
(ii) This converges by comparison with n . P n
(iii) This is absolutely convergent by comparison with |x| and hence is convergent.

Theorem 156 (The Ratio Test) Let (an ) be a real or complex sequence with an 6= 0 for all n. Suppose that
¯ ¯
¯ an+1 ¯
lim ¯ ¯=L
n→∞ ¯ an ¯

exists.
P
• If |L| < 1 then an converges absolutely;
P
• If |L| > 1 then an diverges;
P
• If |L| = 1 then an may converge or diverge (i.e. the test is inconclusive).

36
Proof. (i) Choose K such that |L| < K < 1. As ε = K − |L| > 0 there exists N such that
¯¯ ¯ ¯
¯¯ an+1 ¯ ¯
n > N =⇒ ¯¯¯¯ ¯ − L¯ < ε,
¯ ¯
an

i.e.for n > N ¯ ¯
¯ an+1 ¯
¯ ¯
¯ an ¯ 6 ε + |L| = K.
P k P∞
So by induction, |aN +k | 6 |aN | K k for
Pk > 0. Now K is convergent, and so N an is absolutely convergent by
the Comparison Test. Finally we see an is convergent as it has a convergent tail.
(ii) Choose K such that 1 < K < |L| . Then there exists N such that
¯ ¯
¯ an+1 ¯
n > N =⇒ ¯¯ ¯ > K.
an ¯
P
Then |aN +k | > K k |aN | and hence
P −1 we see P
an does not tend to 0. So an is divergent by Proposition 152.
(iii) For each of the series n and n−2 we have L = 1 yet the former diverges and the latter converges.
X
Remark 157 If an > 0 for all n and an converges this does not mean that lim |an+1 /an | exists; for example

1 1 1 1 1 1 1
1+ + + + + + + + ···
3 2 9 4 27 8 81
converges whilst |an+1 /an | does not have a limit.

Example 158 For all x ∈ C, the exponential series

∞
X xn
0
n!

converges absolutely.

Solution. The case x = 0 is trivial. If x 6= 0 then

|an+1 | |x|
= → 0 < 1 as n → ∞
|an | n+1

and apply the ratio test.

Example 159 The series

∞
X
(sinh n) xn
1

converges absolutely for |x| < e−1

and diverges for |x| > e−1 .

Solution. By definition sinh n = (en − e−n ) /2 and so

¯ ¯
¯ an+1 ¯ sinh (n + 1)
¯ ¯
¯ an ¯ = sinh n
|x|

en+1 − e−n−1
= |x|
en − e−n
e − e−2n−1
= |x|
1 − e−2n
→ e |x|

as n → ∞. If |x| = e−1 then the ratio test is inconclusive but

|an | = sinh n × e−n → 1 6= 0

and so the series does not converge.by the Algebra of Limits. Hence sk converges to L. (See Exercise Sheet 4, Question
4).

37
Theorem 160 (Leibniz Alternating Series Test) Let (an ) be a non-negative decreasing series which tends to 0. Then
∞
X
(−1)n an
n=1

converges.
Pn
Proof. If we consider the partial sums sn = k=1 (−1)k ak we see that

s2k = (−a1 + a2 ) + (−a3 + a4 ) + · · · + (−a2k−1 + a2k )

| {z } | {z } | {z }
60 60 60
= −a1 + (a2 − a3 ) + (a4 − a5 ) + · · · + a2k
| {z } | {z }
>0 >0
> −a1 .

Hence s2k is a decreasing sequence bounded below by −a1 and so s2k converges to a limit L. We also have

s2k+1 = s2k − a2k+1 → L − 0 = L

P∞ P∞ n+1
Remark 161 Nothing we have done so far lets us tackle series like 2 n(log1 n)2 or to evaluate 1 (−1)n . In the
remainder of this section we deal with these: but in order to do so we need to make use of the properties of “integral”
and “logarithm” which are not established until the HT and TT courses. At the end of the year you will be able to
persuade yourself that these properties which we are to now use do not depend on any of the results of this section,
and that no circular arguments have been made. Basically, it is just impatience that forces us to deal with this test
now and not wait until Trinity Term.

Theorem 162 Let K ∈ N and let f : [K, ∞) → [0, ∞) be continuous and decreasing. For n > K we define
n−1
X Z n
δn = f (k) − f (x) dx.
K K

Then for n > K

0 6 δ n 6 δ n+1 6 f (K)
and hence δ n converges.
P∞ Rn
Corollary 163 (The Integral Test) With f as above, the series K f (k) is convergent if and only if K
f (x) dx is
convergent.

We postpone the proof for now and first look to apply the Integral Test to a few series.
P
Example 164 an = 1/nα where α ∈ R. If α 6 0 then an does not tend to 0 and so there is no hope of an
converging. So suppose that α > 0. To apply the Integral Test we need to consider the function f (x) = x−α . As
f 0 (x) = (−α) x−α−1 < 0 we have (by a HT theorem) that f is decreasing. We take K = 1 and note that if α 6= 1 then
Z n ∙ 1−α ¸n
t 1 ¡ 1−α ¢
f (t) dt = = n −1
1 1 − α 1 1 − α

which converges as n → ∞ if α > 1 and diverges if α < 1. If α = 1 then

Z n
f (t) dt = [log t]n1 = log n
1

which diverges. Hence

∞
X 1
1
nα
converges when α > 1 and diverges for α 6 1.

38
Example 165 an = (n log n)−1 for n > 2. Hence we take f (x) = 1
x log x , and
−1 1 1 1
f 0 (x) = − =− 2 [log x + 1] < 0 for x > 1.
x2 log x x(log x)2 x x (log x)2
Then Z n
1
dx = log log n − log log 2 → ∞ as n → ∞.
2 x log x
P 1
Therefore n log n is divergent.
Proof. (Of Theorem 162) We set
n−1
X Z n
δn = f (k) − f (x) dx.
K K

In the diagram below, which includes a graph of y = 1/x for x > K = 1 we can see δ 4 as the "excess area" above the
graph between 1 6 x 6 4.
1.0

0.8 Δ4

0.6

0.4 Δ4
Δ4 y1x
0.2

0.0
1 2 3 4 5 6

As f is decreasing,
f (k + 1) 6 f (x) 6 f (k), if k 6 x 6 k + 1
We use the following properties of integration:
R
• preserves weak inequalities;
R n+1
• n 1 dx = 1;
R Rb Rc Rb
• is additive: a = a + c ;
R
• is a linear map on the space of integrable functions.
So we get:
Z k+1
f (k + 1) 6 f (x) dx 6 f (k)
k
and we can add such equations to get
Z n
f (K + 1) + f (K + 2) + · · · + f (n) 6 f (t) dt 6 f (K) + f (K + 1) + · · · + f (n − 1)
K
So using the second inequality above we have
n−1
X Z n
06 f (r) − f (t) dt
r=K K

which shows 0 6 δ n . Also, using the first inequality we have

n−1
X Z n n−1
X n
X
δn = f (r) − f (t) dt 6 f (r) − f (r) = f (K) − f (n) 6 f (K).
r=K K r=K r=K+1

We also have Z n+1

δ n+1 − δ n = f (n) − f (t) dt > 0.
n
Now (δ n ) bounded above and increasing and so convergent.

39
11.1 Euler’s Constant (γ)
If we apply Theorem 162 to f (x) = 1/x we get
Z n
1 1 dx
γn = 1+ + ··· + −
2 n 1 x
1 1
= 1 + + · · · + − log n
2 n
1
= δn +
n
is convergent. This limit is called Euler’s constant, and often denoted as γ:
à n !
X1
γ = lim − log n .
n→∞
1
k
The approximate numerical value of γ is
0.57721566490153286060 . . .
Relatively little is known about γ — for example it is an open problem still on whether γ is irrational.
Example 166 We make use of γ in Exercise Sheet 6, Question 5 to show that
1 1 1 1 1
1− + − + − + · · · = log 2
2 3 4 5 6

11.2 Euler’s Number (e)

In Exercise Sheet 5, Question 4, we showed
X∞
1 1 1 1
e=1+1+ + + + ··· =
2! 3! 4! r=0
r!
converges to an irrational number e. Its approximate numerical value is
2.7182818284590452353 . . .
Proposition 167 µ ¶n
1
e = lim 1+ .
n→∞ n
Proof. Let µ ¶n n
X
1 1
αn = 1+ and β n = .
n k!
k=0
It was shown in Exercise Sheet 5, Question 4 that lim β n exists and we defined e as this limit. It was also shown in
Exercise Sheet 1, Question 6, that αn is an increasing sequence bounded above and so converges also converges. By
the Binomial Theorem
µ ¶ µ ¶2 µ ¶3
1 n (n − 1) 1 n (n − 1) (n − 2) 1 1
αn = 1 + n + + + ··· + n
n 2! n 3! n n
µ ¶ µ ¶µ ¶ µ ¶µ ¶
1 1 1 1 2 1 1 2 1
= 1+1+ 1− + 1− 1− + 1− 1− ···
2! n 3! n n n! n n n
1 1 1
6 1 + 1 + + + ··· + = βn.
2! 3! n!
From this we have lim αn 6 e. On the other hand if m, n are natural numbers with m < n focusing on the first m + 1
terms of αn we see
µ ¶ µ ¶µ ¶ µ ¶
1 1 1 2 m−1 1
1+1+ 1− + ··· + 1 − 1− ··· 1 − 6 αn .
n 2! n n n m!
If we fix m and let n → ∞ then we have, using the Algebra of Limits and recalling that limits respect weak inequalities,
1 1 1
1+1+ + + ··· + 6 lim αn .
2! 3! m!
Finally letting m → ∞ we have e 6 lim αn and the result follows.

40
11.3 More Complicated Examples
Whilst the tests are useful series are not usually met in such a straightforward way that a single convergence test can
be employed. If they can be employed at all, some combination of the tests may be needed.

Example 168 Discuss the convergence or divergence of the following series.

• ¡ ¢
X cos n2 + 1
.
n2 + log n
We note that ¯ ¯
¯ cos ¡n2 + 1¢ ¯ 1 1
¯ ¯
06¯ 2 ¯6 6 2
¯ n + log n ¯ n2 + log n n
X
and so the series is absolutely convergent by comparison with n−2 .
• ¡ ¢
X log n2 + 1
n
(−1) √
n+2
¡ ¢
If y (x) = log x2 + 1 (x + 3)−1/2 then

1 2x 1 ¡ ¢
y 0 (x) = √ 2 + 1)
− 3/2
log x2 + 1
x+2 (x 2 (x + 2)
∙ ¸
1 2x (x + 2) 1 ¡ 2 ¢
= − log x + 1
(x + 2)
3/2 x2 + 1 2
∙ ¸
1 1 ¡ 2 ¢
< 3/2
4 − log x + 1
(x + 2) 2
< 0 for x > e4 .

Hence by Leibniz’s Test a tail of the series converges and so the whole series converges.
• X 1
√
2
n +n
We see
1 1 1
√ >√ = √
2
n +n 2n 2 n 2
and so the series diverges by comparison with the harmonic series.
• √ √
X n+1− n
√
n2 + n
Note √ √
n+1− n 1 1
√ =√ −√
2
n +n n n+1
and hence √
N
X √
n+1− n 1
√ =1− √ → 1 as N → ∞.
1
n2 + n N +1

41
12 Power Series
Definition 169 By a power series we will mean a series of the form
X
an z n

where (an ) is a complex sequence and z ∈ C. We consider (an ) as fixed for this series, and z as a variable. Clearly
the series might converge for some values of z and not for others.
P n
Example 170 • an = 1 : z : Geometric Series : as we have already seen, this series is convergent when
|z| < 1 and divergent when |z| > 1.
P n
• an = 1/n! : z /n! : Exponential Series : we have shown (using the Ratio Test) that this series is convergent
for all z ∈ C.
P∞ n
• an = 1/n : n=1 z /n : Logarithmic Series : convergent for |z| < 1 and divergent for |z| = 1. Converges at
z = −1 and diverges at z = 1. What about for other values where |z| = 1? Well at z = i we have
2N n
à ! à !
X i 1 1 1 (−1)N 1 1 (−1)N−1
= − + − + ··· + + i 1 − + − ··· +
1
n 2 4 6 2N 3 5 2N − 1

and we see that both real and imaginary parts converge by the Leibniz Test. In fact, the Logarithmic Series
converges on the circle |z| = 1 except at z = −1.
¾
a2n = (−1)n / (2n)! P (−1)n 2n
• : z : Cosine Series : convergent for all z by the Ratio Test.
a2n+1 = 0 (2n)!
¾
a2n = 0 P (−1)n 2n+1
• n : z : Sine Series : convergent for all z by the Ratio Test
a2n+1 = (−1) /(2n + 1)! (2n + 1)!
again.
P
Definition 171 Given a power series an z n the set
n X o
S= z∈C: an z n converges ⊆ C

is either bounded or unbounded. Note also, that S is non-empty as 0 ∈ S. We define the power series’ radius of
convergence R as ½
sup {|z| : z ∈ S} when S is bounded,
R=
∞ when S is unbounded.
P n P
Lemma 172 Suppose that the power series an (z0 ) converges. Then an z n converges absolutely when |z| < |z0 | .
P
Proof. As an (z0 )n converges then an (z0 )n → 0 and in particular the sequence an (z0 )n is bounded; say |an (z0 )n | <
M . Then, for |z| < |z0 | , ¯ ¯n ¯ ¯n
n ¯¯ z ¯¯ ¯z¯
|an z | = |an (z0 ) | ¯ ¯ < M ¯¯ ¯¯
n
z0 z0
P n
P n
and so |an z | converges by comparison with the convergent geometric series M |z/z0 | .
P
Theorem 173 Given a power series an z n with radius of convergence R,
P
• an z n converges absolutely when |z| < R,
P
• an z n diverges when |z| > R.
P
In particular note that when R = ∞ then an z n converges absolutely for all z ∈ C.
P
Proof. If |z| < R then there exists z0 ∈ S such that |z| < |z0 | < R P and hence an z n converges absolutely by the
previous lemma. On the other hand if |z| > R then z 6∈ S and hence an z n diverges.

Definition 174 The set S is called the disc of convergence.

42
 P∞
• 1 z n /n2 . If we set an = z n /n2 then
¯ ¯ µ ¶2
¯ an+1 ¯ |z|n+1 / (n + 1)2 1
¯ ¯= = 1 + |z| → |z| .
¯ an ¯ |z|n /n2 n

Hence, by the Ratio

P Test the series converges absolutely when |z| < 1 but diverges when |z| > 1. In fact, by
comparison with n−2 we see that the series also converges when |z| = 1.
P n
• z /n. If we set an = z n /n we can argue as above to see R = 1. The series converges at some points on the
disc’s boundary (e.g. z = −1) and diverges at others (e.g. z = 1)
P p
• z where the sum is taken P over all primes p. Then R = 1. To see this we can note z p does not tend to 0 when
|z| > 1. On the other hand
P z p converges absolutely when |z| < 1 by comparison with the geometric series
n
z .
P
• Cosine Series: ∞ n 2n n 2n
0 (−1) z / (2n)! If we set an = (−1) z / (2n)! then
¯ ¯
¯ an+1 ¯ |z|2n+2 / (2n + 2)! |z|2
¯ ¯
¯ an ¯ = 2n
|z| / (2n)!
=
(2n + 2) (2n + 1)
→ 0 as n → ∞.

Hence by the Ratio Test the cosine series converges absolutely for all z.
P
• Sine Series: ∞ n 2n+1
0 (−1) z / (2n + 1)! If we set an = (−1)n z 2n+1 / (2n + 1)! then
¯ ¯
¯ an+1 ¯ |z|2n+3 / (2n + 3)! |z|2
¯ ¯= = → 0 as n → ∞.
¯ an ¯ |z|2n+1 / (2n + 1)! (2n + 2) (2n + 3)

Hence by the Ratio Test the sine series converges absolutely for all z.

The following theorem is beyond the scope of this course, but will be proved in Trinity Term.
P∞
Theorem P∞175 Suppose the (real or complex) power series 0 an z n has radius of convergence R. Then the derived
n
series 0 (n + 1) an+1 z has radius of convergence R and for |z| < R the derived series converges to the derivative
of the original: ̰ !
X∞
n d X n
(n + 1) an+1 z = an z .
0
dz 0

13 The Elementary Functions

13.1 The exponential function
P
1. For all z ∈ C, z n /n! is convergent: so R = ∞.
2. The exponential function is defined as
∞
X zn
exp(z) =
0
n!

3. exp(0) = 1
4. exp (1) = e
5. exp0 (z) = exp (z) .
Proof. We need Theorem 175 for this. Note
Ã∞ ! ∞ ∞
d d X zn X (n + 1) z n X z n
exp z = = = = exp z.
dz dz 0
n! 0
(n + 1)! 0
n!

43
 6. exp(x + y) = exp(x) exp(y).
Proof. We proved this in Example 151. We can also use Theorem 175 to show this: for fixed c ∈ C we define

F (z) = exp (z + c) exp (−z) .

Then, by the product rule

F 0 (z) = exp (z + c) exp (−z) − exp (z + c) exp (−z) = 0.

Hence F (z) is constant — as F (0) = exp (c) then

exp (z + c) exp (−z) = exp (c) .

Set c = x + y and z = −y for the required result.

7. exp(z) 6= 0.
Proof. For any z ∈ C we have
exp (z) exp (−z) = exp (0) = 1.

8. exp (q) = eq for rational q.

Proof. Say q = m/n then
³ ³ m ´´n ³ m´
exp = exp n = exp (m · 1) = (exp 1)m = em .
n n

13.2 Powers and Logarithms

It seems appropriate to make the following definitions here, though some of what follows requires theory from Hilary
Term. We now restrict our attention to the real exponential exp : R → R. It is clear from the power series definition
of exp that exp x > 0 if x > 0. Further if x < 0 then
1
exp (x) = >0
exp (−x)
also. So exp (R) ⊆ (0, ∞).
Now exp0 x = exp x > 0 and so exp is an increasing function (a result from HT); in particular this means that
exp : R → (0, ∞) is injective. Also for x > 0, exp x > x and so exp takes arbitrarily large values of x and similarly
exp (−x) = 1/ exp (x) takes arbitrarily small values. So, by the Intermediate Value Theorem (proved in HT), we have

• exp : R → (0, ∞) is a bijection and hence invertible.

• The inverse is denoted as log : (0, ∞) → R.

Definition 176 Given a > 0 and x ∈ R we define

ax = exp (x log a) .

Note, with this definition,

ex = exp x for x ∈ R.

Proposition 177 Let a, b > 0 and x ∈ R. Then

log (ab) = log a + log b, log (ax ) = x log a.

Proof. Note
exp (log a + log b) = exp (log a) exp (log b) = ab = exp (log (ab))
and then take the log of both sides. Also

log (ax ) = log (exp (x log a)) = x log a.

44
Proposition 178 Let a > 0 and x, y ∈ R. Then

ax+y = ax ay , (ax )y = a(xy) .

Proof. Note

ax+y = exp ((x + y) log a)

= exp ((x log a) + (y log a))
= exp (x log a) exp (y log a)
= ax ay ,

and ³ ´
y
log (ax ) = y log (ax ) = y (x log a) = (xy) log a = log a(xy) .

Proposition 179 For x > 0,

1 d
log0 x = , (xa ) = axa−1 .
x dx
Proof. As exp (log x) = x then, by the chain rule,

log0 (x) × exp (log x) = 1

and the first result follows. Also by the chain rule

d d a
(xa ) = (exp (a log x)) = exp (a log x) = ax−1 xa = axa−1 .
dx dx x

13.3 The Trigonometric Functions

1. For all z ∈ C we define
exp (iz) + exp (−iz) exp (iz) − exp (−iz)
cos z = , sin z =
2 2i

2. Then
∞
X ∞
X
(−1)n z 2n (−1)n z 2n+1
cos z = and sin z =
0
(2n)! 0
(2n + 1)!
with these series converging for all z ∈ C.
Proof.
∞ ∞ ∞
1 X (in + (−i) ) n 1 X (−1) + (−1) 2k X (−1) z 2k
n k k k
exp (iz) + exp (−iz)
= z = z = ;
2 2 0 n! 2 (2k)! 0
(2k)!
k=0
∞ ∞ ∞
exp (iz) − exp (−iz) 1 X (in − (−i)n ) n 1 X (−1)k i + (−1)k i 2k+1 X (−1)k z 2k+1
= z = z = .
2i 2i 0 n! 2i (2k + 1)! 0
(2k + 1)!
k=0

3. cos 0 = 1, sin 0 = 0
4.
exp (iz) = cos z + i sin z

Proof. µ ¶ µ ¶
exp (iz) + exp (−iz) exp (iz) − exp (−iz)
cos z + i sin z = +i = exp (iz) .
2 2i

45
5.
cos z = cos(−z), sin z = − sin(−z).

6.

sin(z + w) = sin z cos w + cos z sin w,

cos (z + w) = cos z cos w − sin z sin w.

Proof.

sin z cos w + cos z sin w

µ ¶µ ¶ µ ¶µ ¶
exp (iz) − exp (−iz) exp (iw) + exp (−iw) exp (iz) + exp (−iz) exp (iw) − exp (−iw)
= +
2i 2 2 2i
1
= [2 exp (iz) exp (iw) − 2 exp (−iz) exp (−iw)]
4i
exp (iz + iw) − exp (−iz − iw)
=
2i
= sin (z + w) .
cos z cos w − sin z sin w
µ ¶µ ¶ µ ¶µ ¶
exp (iz) + exp (−iz) exp (iw) + exp (−iw) exp (iz) − exp (−iz) exp (iw) − exp (−iw)
= −
2 2 2i 2i
1
= [2 exp (iz) exp (iw) + 2 exp (−iz) exp (−iw)]
4i
exp (iz + iw) − exp (−iz − iw)
=
2
= cos (z + w) .

7.
cos2 z + sin2 z = 1

Proof.
µ ¶2 µ ¶2
exp (iz) + exp (−iz) exp (iz) − exp (−iz)
cos2 z + sin2 z = +
2 2i
1³ 2 2 2 2
´
= exp (iz) + 2 + exp (−iz) − exp (iz) + 2 − exp (−iz)
4
= 1.

8.

cos0 (z) = − sin z,

sin0 (z) = cos z.

Proof. Using exp0 z = exp z then

µ ¶ µ ¶
d d exp (iz) + exp (−iz) d i exp (iz) − i exp (−iz)
(cos z) = = = − sin z,
dz dz 2 dz 2
µ ¶ µ ¶
d d exp (iz) − exp (−iz) d exp (iz) + exp (−iz)
(sin z) = = = cos z.
dz dz 2i dz 2

46
 9. It is easy to note that cos 0 = 1 and that
∞
X n
(−1) 22n
cos 2 =
0
(2n)!
2
2 24 26 28
= 1− + − + − ···
2! 4! 6!
µ 8! ¶ µ ¶
2 26 22 210 22
= 1−2+ − 1− − 1− − ···
| {z 3} |6! {z
7×8
} |
10! 11 × 12
{z }
<0 >0 >0
< 0.

It follows from theorems we will meet in Hilary Term that there exists a smallest positive root to the equation
cos x = 0. We will define π/2 as the smallest root of cosine.
10. As cos2 z + sin2 z = 1 then sin (π/2) = ±1 (in fact it equals 1 as we know) and

exp (πi/2) = cos (π/2) + i sin (π/2) = ±i.

Then µ ¶4
πi 4
exp (z + 2πi) = exp z exp = (exp z) (±i) = exp z.
2
Hence exp has period 2πi and cosine and sine have period 2π — i.e.

cos(z + 2π) = cos(z) and sin(z + 2π) = sin(z).

13.4 Hyperbolic Functions

Define
∞
exp (z) + exp (−z) X z 2n
cosh z = = ,
2 0
(2n)!
∞
exp (z) − exp (−z) X z 2n+1
sinh z = = ,
2 0
(2n + 1)!

where these series converge for all z ∈ C. Note that

cos iz = cosh z, cosh iz = cos z,

sin iz = i sinh z, sinh iz = i sin z,
cosh (−z) = cosh z, sinh (−z) = − sinh z,
cosh0 z = sinh z, sinh0 z = cosh z,
sin (x + iy) = sin x cosh y + i cos x sinh y,
cos (x + iy) = cos x cosh y − i sin x sinh y,
cosh2 z − sinh2 z = 1.

13.5 The other trigonometric functions

Define
1 sin x
sec x = , tan x = ,
cos x cos x
1 cos x
csc x = , cot x = .
sin x sin x

47
13.6 Logarithmic Series
Consider the power series
∞
X xn
L (x) = .
1
n

The radius of convergence is 1. (this follows easily from the Ratio Test) and so converges for |z| < 1.
Now consider real x. Note that, by Theorem 175,
∞
X ∞
X
nxn−1 1
L0 (x) = = xn−1 =
1
n 1
1−x

from our knowledge of the geometric series. Let −1 < x < 1 and set

M (x) = (1 − x) exp (L (x)) .

Then, by the chain and product rules,

M 0 (x) = −1 × exp L (x) + (1 − x) L0 (x) exp L (x) = 0.

It follows that from the Mean-value Theorem (HT) that L (x) is constant and equals M (0) = exp L (0) = exp 0 = 1.
Hence
1
exp (L (x)) = for − 1 < x < 1
1−x
and, with using the definition of log from earlier
µ ¶
1
L (x) = log = − log (1 − x) .
1−x

13.7 Binomial Series

Let α ∈ C. The series
∞
X α(α − 1) . . . (α − k + 1)
B (x, α) = xk
k!
k=0

has radius of convergence 1 (unless α ∈ N in which case the RHS is only a finite sum). We note two aspects of this
series: by Theorem 149 we can show with a little working that

(1 + x) B (x, α − 1) = B (x, α) ,

and by Theorem 175

B 0 (x, α) = αB (x, α − 1) .
If we set C (x, α) = B (x, α) (1 + x)−α then
−α −α−1
C 0 (x, α) = B 0 (x, α) (1 + x) − αB (x, α) (1 + x)
−α −α
= αB (x, α − 1) (1 + x) − αB (x, α − 1) (1 + x)
= 0.

Hence C (x, α) is constant and we note

C (0, α) = B (0, α) 1−α = 1.
Finally
B (x, α) = (1 + x)α for α ∈ R and − 1 < x < 1.

48