Real Analysis I

MAT1032
2011/2012

Prof. H. Bruin

Department of Mathematics
University of Surrey

Introduction

What is Real Analysis? Real Analysis is the study of numbers such as 1, 3, 73 , 7,

2 , etc. The word real number distinguishes them from complex numbers such as

i = 1, 6 3i, etc. We wont use any complex numbers in this module. The set of real
numbers is richer than the sets of integers, or of fractions, in the sense that it contains
limit values. Limits are the central concept of this module, and while you may have
seen limits such as limn (1 + n1 )n = e before, in this module we ask more fundamental
questions such as: What exactly is a limit?, Why does limn (1 + n1 )n exist in the
first place? or Can there be more than one limit?
One reason to do this is that limits underpin many mathematical concepts. Some
you already know (such as integration, differentiation), without being aware that they
are limit procedures. Some appear in advanced mathematical objects that you will learn
more about in modules to come. For instance, limits of sequences of functions, limits of
Fourier series. The usage of limit remains the same for these objects as well, provided
you take a more abstract point of view.
Finding properties of solutions without explicit formulas. It may come as a
surprise to you, but most problems in mathematics cannot be solved by explicit formulas.
Even if you cannot find the solution explicitly, you can still say something about the
properties of the solution.
To give an example, in line with what may come at level 2, the differential equation
f 00 (t) + f 0 (t) + sin(f (t)) = 0
cannot be solved by an explicit formula. Still you can ask whether a solution exists, or
what its properties are, e.g.: Does it exist for all time t? What happens when t is very
large? Can I differentiate the solution f as often as I like? One analytic method is to
find a sequence of approximate solutions such that their limit is the exact solution to the
problem. Then we need to investigate to what extent the properties of the approximate
solutions pass over to the limit. Analytic methods require that you understand the concept
of a limit at a fundamental level, and often without having an explicit formula.
Important things to learn: In mathematics, we arrange the material into definitions,
examples, theorems, proofs, etc., some of which you will need to learn, memorise and
understand in detail.
Definitions describe important notions, for example definition of convergence.
They need to be unambiguous and free of self-contradiction, because we want to
build on them.
Axioms are basic statements which we treat as true. One has to start with something! We use axioms as building blocks of mathematical theory. Many parts of
mathematics have an axiomatic approach, e.g. geometry (Euclid), natural numbers

(Peano), real numbers (see Section 6), and quite often they have axioms in common. So dont be confused when you get axioms in this module, and then a slightly
different set of axioms for groups (in Algebra modules).
Theorems, propositions, lemmas and corollaries are a way in which mathematics
books summarise the main results. There is no big difference between the four,
except that a theorem is considered the most important. A lemma is an auxiliary
theorem, and a corollary is a consequence of a theorem or proposition. Unlike
definitions and axioms, they all require a proof!
Both definitions and theorems need to be learnt in their exact form. To help you memorise, and to improve your understanding, it is useful generate examples and counterexamples for definitions and theorems, to show why each stated condition is essential.
A proof is a logical argument that shows (starting from definitions or other theorems,
lemmas, etc.) that a particular statement is true. Several types of argument are
discussed in these Notes (e.g. proof by contradiction, proof by natural induction).
You wont be asked to memorise whole proofs word by word, but it is important to
understand them and learn its parts, so that you gradually learn to write proofs yourself.
How to study? These Notes are a condensed introduction to Real Analysis and provide
a basis for more detailed notes from lectures. Dont be surprised if you dont understand
these Notes - or any (under)graduate mathematics book - at first reading. It is quite
normal to have to re-read and possibly consult other texts as well, to form complete
understanding.
The material is more abstract than at A-level and there will be few methods-based
(i.e., with a fixed recipe to find the solution) questions. Note also that pocket calculators
are not particularly important in this module. Instead, questions will ask you to state and
apply definitions and theorems, and write proofs and variations of proofs from the Notes.
Presenting arguments logically, in well-written sentences is at least as important as the
actual answer. Sometimes there are specific formats of proof-writing (natural induction,
-N -proof template), which need to be learnt too.
The best preparation before class is to read the section to be covered. After lectures to
ensure understanding of topics, read the section again and compare with your own/fellow
students lecture notes. Also do exercises to see if your understanding of the material
is sufficient. Unassessed courseworks are a risk-free way to develop understanding (with
feedback) before the tests/final examination.
Further material: These Notes cover properties of numbers, set-theory, sequences and
series. (Further topics on functions, differentiation and integration are postponed to Real
Analysis II, at Level 2.) Some, but not all, of the exercises in these Notes have written
out solutions. Some also have the sign  to indicate that they are more challenging than
others. Details on further material, daily information for the Real Analysis I module, can

be found on-line on the modules website

http://www.maths.surrey.ac.uk/personal/st/H.Bruin/MAT1032/info.html.
The main text for back-ground reading is by John Howie [1]. The campus bookstore has
been notified to keep copies in store, and the library has copies of all the books listed.

Types of Numbers

Numbers. In mathematics we use different sets of numbers, used for various operations.
The most common are:
Name

Notation/Definition

Used for

natural numbers

N = {1, 2, 3, 4, . . . }

counting, adding, multiplying,

order (i.e., the order relation <)

integers

Z = {. . . , 2, 1, 0, 1, 2, 3, . . . }

counting, adding, subtracting,

multiplying, order

rational numbers Q = { pq : p Z, q N}
real numbers

adding, subtracting, multiplying,

dividing (except by 0), order
adding, subtracting, multiplying,
dividing (except by 0), order,
taking limits

One can question whether 0 is a natural number or not. During the development of set-theory around 1900, starting the natural numbers at 0 became more
wide-spread, especially under the influence of the French group of mathematicians
Bourbaki. Currently, there is a geographical divide: Anglosaxon countries tend
to define N = {1, 2, 3, . . . }, whereas continental countries (especially France) use
N = {0, 1, 2, 3, . . . }. For us, it is convenient to denote the set {0, 1, 2, 3, . . . } by N0 .

Rational versus irrational numbers. Every natural number is an integer but not
vice versa. Similarly every rational number is a real number but not every real number
is rational. A real number that is not rational is called irrational.
Theorem 1

2 is an irrational number.

Proof. This is proof by contradiction: We assume that 2 is a rational number,

and then try to come to a conclusion that is clearly wrong. That means then that the
assumption must be false.

Here it goes: Assume that 2 Q. Then there are integers p and q > 0 such that

p
2= .
q
3

We can assume that pq is in lowest terms, because if p and q have a common divisor, we
could divide them away. Take the square of both sides:
2=

p2
q2

or equivalently,
2q 2 = p2 .
This means that p2 is an even integer, and because odd numbers have odd squares, p itself
must be even. Let us write p = 2k for some integer k. We have
2q 2 = (2k)2 = 4k 2
or equivalently
q 2 = 2k 2 .
This shows that q 2 is even, and as before, this implies that q is even.
Hence p and q are both even, but this contradicts the fact that p and q have no
common divisor.

This
contradiction
shows
that
our
original
assumption:
2 Q is false. Therefore

2
/ Q.


Theorem 2 (Fundamental Theorem of Arithmetic) Every integer n > 1 has a unique

prime factorisation, i.e., there is one and only one way of writing
n = pe11 pe22 pel l
where p1 , . . . , pl are prime numbers, and e1 . . . el N are their powers. (Note that by
convention, 1 is not a prime; the smallest prime is 2).
Proof. Clearly every integer n > 1 is either prime itself, or can be factorised in smaller
factors. If we keep on factorising the factors, we end up with at least one prime factorisation. The tricky part is to show that there is no other prime factorisation.
Assume by contradiction that n has two prime factorisations. This means that there
are two different ways of writing n as a product of primes:
n

pe11 pe22 pekk

q1f1 q2f2 qlfl .

Assume that n is the smallest integer with this property.

Then all the pi s are different from all qj s, because if a pi and a qj were the same, then
we can divide n by it, and find a smaller integer with two different prime factorisations.
We can also assume that the primes pi and qj are listed in increasing order, and that
k, l 2 (because otherwise n is prime). In other words
n = p1 P = q 1 Q
4

where P := pe11 1 pe22 pekk > p1 and Q := q1f1 1 q2f2 qlfl > q1 .
Now subtract p1 q1 from all three parts of the formula:
n p1 q1 = p1 (P q1 ) = q1 (Q p1 ).
Both n and n p1 q1 are a multiple of p1 , so n p1 q1 p1 > 1. So n p1 q1 is a number
smaller than n that can be written in two different ways of product of integers. This
contradicts that n is the smallest such number > 1.


Example: 2 + 3 is irrational. To prove this, we first prove that 6 is irrational.

Assume by contradiction that 6 = ab for some p, q Z, q 6= 0, reduced to lowest terms,

so they dont have a common factor. Squaring this equation, and multiplying by b2 ,
we obtain 6b2 = a2 . By the Fundamental Theorem of Arithmetic, a can be written as
pe11 pe22 plel , and its square as
2el
2e2
1
a2 = p2e
1 p2 pl ,

so the same prime numbers are used, only the powers are double as those used for a. Now
a2 = 6b2 , and 6 = 2 3, so one of these primes, say p1 , is 2. Therefore 2 is a divisor of a,
or in other words a = 2k. Inserting this in the equation, we get
6b2 = (2k)2 and hence 3b2 = 2a2 .
Now the same argument shows that b is a multiple of 2 as well, and this contradicts that
a and b have no common factor.

by contradictionthat 2 + 3 = uv for some

Now to come back to 2 + 3, assume
2
integers u, v 6= 0. Then ( 2 + 3)2 = 5 + 2 6 = uv2 . Make 6 subject; this gives

But we just proved that

6=

u2 5v 2
Q.
2v 2

6
/ Q. This contradiction proves that

2+

3 6 Q.

Two other famous irrational numbers are and e (the base of the natural logarithm),
but their irrationality is much
harder to prove (for e see [1, Example 6.10]). In fact, they
are more irrational than 2 in the sense that e and cannot be the solution of any
polynomial equation
an xn + + a1 x + a0 = 0 for an , . . . , a0 Z.

(1)

We call such numbers transcendental. The transcendence of e and was proven by

Hermite (1873) and Lindemann (1882) respectively, see [2]. Numbers x that are the
solution of some equation of the form (1) are called algebraic numbers.

Example: 2 + 3 is algebraic. Indeed, if x = 2 + 3, then x2 = 5 + 2 6 and

x2 5 = 2 6. By squaring both sides of this equation once more, cancelling brackets and
making the right hand side zero, we get
x4 10x2 + 1 = 0.
5

Therefore x satisfies a polynomial equation with integer coefficients: it is algebraic.

Example: ln 2 is irrational. We argue again by contradiction, so assume that ln 2 = pq
for some p, q Z, q 6= 0. Taking the exponential function at each side, and then raising
to the q-power gives: 2q = ep , and this reduces to
ep 2q = 0.
However, this means that e is the solution of the polynomial equation with integer coefficients: xp 2q = 0. But then e is algebraic, contradicting the result of Hermite.
Exercises:

1) Prove that the following numbers are rational or irrational: 13, 3 2, 12 + 5, 3+ 5.
2) Explain why the set of rational numbers are not so useful for counting.
3 a) Given two rational numbers a < b, find a rational number x such that a < x < b.
b) Next find an irrational number x such that a < x < b.
4) Show that the sum and product of two rational numbers is rational.
5) Show that the sum and product of a rational 6= 0 and an irrational is irrational.
6) Find two irrationals x and y such that x + y is rational but xy is irrational.
7) Show that rational numbers are algebraic.
8) Prove that ln 2
/ Q. Hint: use that e is transcendental.

Sets and Set Notation

Sets. Any collection of things is called a set. In mathematics, these things are usually
numbers, but also sets of functions, sets of sets, etc. are possible. The word collection
or set are used as synonyms. It does not imply anything about the number of objects
in the set, i.e., the set can be finite/infinite, there may be none (empty set).
In this module, sets will refer to sets of numbers. Examples of sets of numbers are:
N, Z, Q, R;
the empty set ;
intervals:
[a, b] = {x R : a x b}
(a, b) = {x R : a < x < b}
[a, b) = {x R : a x < b}
(a, b] = {x R : a < x b}

closed intervals


[a, ) = {x R : a x}
(, b] = {x R : x b}

(, ) = R

open intervals
half-open intervals

unbounded intervals

A square bracket is used if the boundary point of the interval is included in the
set. The in the unbounded intervals is a convention; is not a real number.
Therefore we always use the round bracket for and . Note that R is an
open unbounded interval, [a, ) is an unbounded interval, and (a, ) is an open
unbounded interval.
In sets, order and repeated elements are irrelevant, e.g. {1, 2, 5} = {2, 1, 5, 1}. But we
try to write sets as natural as possible, so {1, 2, 5} is the preferred way of writing this set.
Note that open and closed are not opposites in set theory, but the compliment of an
open set is always closed and vice versa. A set may be both or neither, e.g. is open
(why?), therefore the complement c = R is closed. But as noted above R = (, ) is
also an open interval, so = Rc is closed as well. In short, and R are both open and
closed (such sets are called clopen).
Operations on sets. We can make statements on sets, unsing special notation too. Let
A and B be sets of (real) numbers.
x A.

x belongs to A, or in other words x is an element of A,

A=B

A and B are the same; they contain exactly the same elements.

AB
A is contained in B (or A is a subset of B); every element of A also
belongs to B.
AB
A contains B (or A is a superset of B); every element of B also belongs
to A. Note that A B or A B do not exclude the possibility that A = B.
Each of the above operations can be negated. So we get ,
/ 6, 6 for does not belong
to, is not contained in and does not contain.
AB
the intersection of A and B; this is the set of all numbers that belong
to both A and B. If A and B have no elements in common (i.e., A B = ), then
they are called disjoint.
AB
the union of A and B; this is the set of all numbers that belong to A or
B. The or in this definition is not an exclusive or as in either ... or.
A4B
Exclusive or. These are the point either in A or in B, but not in both. It
is called the symmetric difference of A and B because A4B = (A \ B) (B \ A).
A\B
the difference of A and B; this is the set of all numbers that belong to A
but not to B. NB: set difference uses the back-slash, so A \ B, not A/B.
Ac

the complement of a set; the set of all numbers that do not belong to A.

Definition 3 Let A B R be subsets of the real numbers. The set A is dense in B if

between every two numbers x, y B with x < y, there is a A such that a [x, y].
7

Examples: Q is dense in R (see Theorem 16 on page 28), Z is dense in Z, but Z is not

dense in R.
Multiple and infinite unions/intersections. We can refer to the union/intersection
of multiple sets using some further notation:
ki=1 Ai = A1 A2 Ak describes the union of sets A1 to Ak .
ki=1 Ai = A1 A2 Ak describes the intersection union of sets A1 to Ak .
Pk
1
This is in analogy to multiple sums, e.g.
i=1 i = 2 k(k + 1) or multiple products,
Qk
e.g. k! = i=1 i that you may have seen before. We can even take infinite unions and
intersections. For example

[i, i + 1) = [0, )

i=0

[0,

i=1

i+1
) = [0, 1]
i

(2)

Remark: In this section, we defined open and closed intervals, but more generally than
intervals, we say that a subset

open if it is the union of open intervals,

A R is
(3)
closed if the complement Ac is the union of open intervals.
Exercises:
1) Is the set {x} an interval? Is it open or closed?
2) Indicate the following sets in the picture:
'$
C
'$

'$
&%
&%

&%

a) A B,
b) A C;
c) B \ C;
d) (A B) \ C;
e) A (B \ C).

3 Let A and B be subsets of R. Express, using , and \, the set of numbers that belong
to either A or B (but not to both).
4) Write the following sets as union of intervals:
a) {x R : 1 < x 2 or 2.5 x}.
b) {x R : x2 7 0}.
c) {x R : 3x > 4x + 1}.
d) {x R : sin x < 0}.
5) Write the maximal domains on which the following functions are defined in set-notation:
a) f (x) = sin1x .
b) f (x) = ln(1 x2 ).
8


c) f (x) = x3 4x.
x+3
d) f (x) = 9x
2
6) Show that and satisfy the distributive law: A (B C) = (A B) (A C).
7) Write {1, 2, 3, 4} it at least four different ways, using any of the set notation in this
section.
8) Explain why the formulas in (2) are true.
9) Write in set notation (using \ if needed):
S
a) nZ (n, n + 1)
b) nN ( n1 , n1 )
c) nN ( n1 , n2 ).
10a) Let U be an open set (as defined in (3)). Show that for every x U , there is an
> 0 such that (x , x + ) U .
b) Show that if U1 and U2 are open, then U1 U2 is open and U1 U2 is open. If for every
T
k N, Uk is an open set. Does it follow that kN Uk is open?
c) Show that if C1 and C2 are closed, then C1 C2 is closed and C1 C2 is closed. If for
S
every k N, Ck is a closed set. Does it follow that kN Ck is closed?

Propositional Logic and Quantifiers

In mathematics, we often use logical relations to clarify the (causal) relations between
true or false statements (also called propositions). For example
If it rains then the roofs get wet
can be abbreviated as P Q, where P stands for it rains, Q stands for the roofs get
wet, and is the implication. Whether the implication is true or not depends not so
much on whether it rains; if it does not rain, then the implication P Q is simply not
tested. However, if it rains and the roofs remain dry, then the implication is certainly
false. We can work out all combinations of P, Q being true or false is an so-called truth
table
P Q
P
Q
true true
true
true false
false
false true
true
false false
true
Note that if the premiss P is false, then the implication is always true: From falsehood,
you can prove everything!
The most common logical relations are the following.
Name
Plain English Notation
Implication
if...then

Equivalence
if and only if

Negation
not
or
and

or

Exclusive or either ... or

XOR
9

The corresponding truth tables are as follows (where we abbreviate F = false and T =
true):
P
T
T
F
F

Q
T
F
T
F

P Q P Q P
T
T
F
F
F
F
T
F
T
T
T
T

P Q P Q P XOR Q
T
T
F
F
T
T
F
T
T
F
F
F

You can make longer expressions with multiple logical relations (using brackets helps to
be clear). For example P (P Q) has the following truth table:
P
T
T
F
F

Q
T
F
T
F

P
T
T
T
T

(P Q)
T
T
F
T

First work out the bit in the brackets, and then the final value (in bold-face). This shows
that P (Q P ) is always true, no matter what values P and Q have. Such an
always true statement is called a tautology. Another example of a tautology is (work
out the truth table yourself!)
(P Q) (Q P )
It says that if you want to prove the implication P Q, it is equally good to prove
Q P . The expression Q P is called the contrapositive of P Q. Proving
by contrapositive is akin to proving by contradiction. More precisely, to prove
(
by contrapositive means: Q P
P Q
by contradiction means: (P Q) False
Example: Show by the contrapositive that if 7n is odd, then n is odd as well. In
propositional form, this is P Q (for P = 7n is odd and Q = n is odd. The
contrapositive is Q P , i.e., if n is not odd, then 7n is not odd. This is what we will
prove:
1) Assume that n is not odd, that is: Q. Therefore n even, and we can write n = 2k.
for some k Z.
2) Then 7n = 7 2 k = 2 (7k), and this is even as well. So we have 6 P . End of proof.
Example: The expressions
P Q

and

(P Q) (Q P )

have the same truth table. (Check yourself!) What this means is that proving an equivalence P Q comes down to proving two implications: the left implication P Q and
the right implication P Q (which is just another notation for Q P ).
10

4.1

Quantifiers.

Mathematicians use two specific abbreviations called quantifiers:

The universal quantifier: for all. For example a N, a > 0 reads, for all
natural numbers a, a > 0 holds, or shorter: all natural numbers are positive.

The existential quantifier: there exists. For example a Q : 13 < a < 12
reads: there exists a rational number a which is larger than 31 and smaller than 21 ,
i.e., there exists a rational number between 13 and 12 .
The order of quantifiers is extremely important. Find this out by translating the following
statement about people into plain English:
x y: y is the mother of x.
y x: y is the mother of x.
The difference is that in the first statement y depends on x; y may change as a different
choice of x is made. In the second statement, x is independent of y. For the correct
statement, one choice of y should apply to all x.
Common phrases. Some phrases that are used very often are particularly good to
remember in quantifier language.
For n sufficiently large: N n N
For positive and sufficiently small: > 0 (0, )
Negating phrases. If you want to negate a statement with quantifiers, i.e., say the
opposite, then you can do one of two things. You can translate the statement into plain
English, write down the opposite statement in English and translate back into a statement
with quantifiers. This can be quite difficult for long statements with many quantifiers.
You can also use the following rules:
1. Write down the statement in the same order.
2. Change every into a and every into a
3. Change the last bit without quantifiers into its opposite statement (for example by
changing into ,
/ = into 6=, or < into ).
Examples:
> 0 > 0 y (x , x + ), y 2 (x2 , x2 + )
{z
} |
{z
}
|
change type of quantifier

replace by opposite

becomes
> 0 > 0 y (x , x + ),
11

y2
/ (x2 , x2 + ).

The statement:
n N p, q Z p2 + q 2 = n

(4)

translates to: every natural number is the sum of two squares. This statement is false.
To prove this we can negate (4)
n N p, q Z, p2 + q 2 6= n

(5)

and try to prove it. This statement starts with there exists n, for which a certain thing
(namely p2 + q 2 = n) is false. Such an n is a counter example to statement (4). Thus
we are going to disprove statement (5) by finding a counter example.
Try n = 3. Because p2 0 and q 2 0 for all p, q Z we only need to test a few values
of p and q:
p2
p2
p2
p2

=0
=1
=2
=3

q2
q2
q2
q2

=3
=2
=1
=0

but
but
but
but

then
then
then
then

q
/ Z.
q
/ Z.
p
/ Z.
p
/ Z.

All other values of p and q give p2 + q 2 > 3. This proves that n = 3 is indeed the counter
example we were looking for. Let us now explain what is meant by scope of a quantifier
and dummy variables. Every quantifier introduces a new dummy variable, only one
variable for every quantifier and at most one quantifier to every variable. Therefore, if
you have x, you shouldnt have x in the same formula. Also x, y R x + y R is
just shorthand for x R y R x + y R. The scope of a dummy variable is from
where it appears at its quantifier until the end of the formula. So they should not appear
before their quantifier. For example,
n
> 0 N N n N |
1| <
n{z
+1
}
|
scope of n
|
{z
}
scope of N
|
{z
}
scope of

Here N depends on (for example, choosing N = b1/c + 2 makes the statement true),
but eps is independent of everything else.
If , n and N appear outside the formula too, then it is wise to give these dummy
variables a different name; otherwise it becomes confusing, and chances are you confuse
yourself the most). Dummy variables and scopes also appear also in other contexts, e.g.
P
sums , intergrals, unions and intersections, indices for sequences, etc. For example in
Rb 2
Pn 2
j=0 j and a x dx, the variables j and x have no meaning outside the summand and
integrand (but n, a and b have!). If j and x also appear outside these expression, then it
is better to use a different letter.

12

Wrong expression

Remark

Corrected version

x > y y

y used before its quantifier

x y < x y 2 > x

a
b

Q a, b Z

Pk
k=0

R 2x
x

y2 > x

k2 k

x2 dx

has not a single variable,

a, b have two quantifiers

x Q a Z
b Z \ {0} x = ab
Pk
2
k used as dummy and as upper
j or
j=0 j P
bound for the range of the dummy maybe kj=0 j 2 k
R 2x 2
x used as dummy and in
t dt
x
domain of integration

Exercises:
1) Make truth tables of the following expressions. Which are tautologies?
a) (P Q) (P Q);
b) (P XOR Q) (P Q).
c) P P (double negation is confirmation);
d) P Q) (Q P ).
e) P Q and P Q. What does this say about another way of proving an
equivalence?
2) Express P XOR Q) in terms of P , Q, and .
3) Let P kQ stand for neither ... nor..., with truth table
P Q P kQ
T T
F
T F
F
F T
F
F F
T
Express P , P Q and P Q in terms of P O, Q and k only?
4) Express the following in mathematical notation (using quantifiers!):
a) Every square is non-negative.
b) n is a prime. (You can use the notation m|n for m is a divisor of n.)
c) Every integer greater than 5 is the sum of three primes. (You may use the notation P
for the set of primes.)
5) Give the negation of the following statements (which of them are true?):
a) > 0 such that (0, ).
b) such that > 0 > .
c) > 0 > 0 x R
|x 3| < implies |x2 9| < .
d) > 0 > 0 x R
|x 3| < implies |x2 9| < .
6) Prove or find a counter-example to: n N p, q, r Z such that p2 + q 2 + r2 = n.

13

Figure 1: Surjective, injective and bijective functions.

Cardinality

Let us recall some terminology regarding functions first. Usually, we denote functions by
letters f, g, h, but this is not obligatory. More formally, when we write
f : A B,
it means that f assigns to every element x in the domain A, a value f (x) in the codomain B. So formally, f (x) is not the function, it is the value f takes at x. As an
example, the function f : [5, 10) R defined by f (x) = x2 , has the interval [5, 10) as
domain and R as co-domain (because we wrote so), but not every value in R is actually
assumed. The values that are assumed form the interval [0, 100) (check with the domain!),
and this is called the range of f . Let us also recall the following, see Figure 1:
Definition 4 A function f : X Y is called surjective (or onto) if y Y x X
such that f (x) = y. (In this case, the range of g is the whole co-domain Y .)
We call g : X Y injective (or one-to-one) if x, x0 X, x 6= x0 , also g(x) 6= g(x0 ).
Finally, if h : X Y is both injective and surjective, then h is bijective.
Cardinality refers to the number of elements in a set. We use the notation1 #(A),
so for example #({2, 4, 6}) = 3. However, when dealing with set with infinitely many
elements, just saying #(A) is infinite is not as precise as can be. Instead, we say:
Definition 5 Two sets A and B have the same cardinality if there is a bijection f :
A B between A and B. If f : A B is a surjection, then we say #(A) #(B). If
#(A) #(B) but #(A) 6= #(B), then we have strict inequality #(A) > #(B).
1

Other books might use the notation card(A) and |A| is used.

14

Thus the legs of a single lobster and the fingers on two hands have the same cardinality, which is less than the cardinality of the set of ants in the world. It becomes more
interesting with infinite sets. We know that
NZQR

#(N) #(Z) #(Q) #(R).

and therefore

(6)

For example, we can make a surjection f : R Q be setting f (x) = x if x Q and

f (x) = 0 if x
/ Q. But the inclusions in (6) are all strict, so does that mean that
#(N) < #(Z) < #(Q) < #(R)?
The 19th century mathematician Georg Cantor developped the theory of cardinalities
and infinite set theory. In his view, N and Z are equally big:
Theorem 6 N and Z have the same cardinality.
Proof. The function f : N Z defined by
( n
f (n) =

2
1n
2

if n is even,
if n is odd,

is a bijection. (Check!). Therefore #(N) = #(Z).

Definition 7 A set C is countable (or more precisely countably infinite) if it has

the same cardinality as N, and the notation for this is #(C) = 0 . (Read: Aleph zero,
for the Hebrew letter .) In this case, there is a bijection f : N C, so we can write
C = {cn }nN where cn = f (n). We call {cn }nN a denumeration of C.
Theorem 8 N and Q have the same cardinality.
Proof. Figure 2 gives a way to find a denumeration {pn }nN of the positive rationals. So
p1 = 1, p2 = 2, p3 = 21 , p3 = 13 , p4 = 3 (we skip 22 because 22 = 1 which we already had in
the list), p4 = 4, p5 = 32 , etc. By skipping all rationals that appeared earlier in the list,
we make sure that pi 6= pj for all i 6= j. (The denumeration becomes injective.) Convince
yourself that the denumeration is also surjective: every positive rational will eventually
appear in the list.
This shows that N has the same cardinality as the positive rationals. Finally, set

0 if n = 1,

pn/2 if n is even,
qn =

p(n1)/2 if n > 1 is odd.

Then {qn }nN is a denumeration of Q (so also the non-positive rationals), and therefore
n 7 qn gives a bijection between N and Q.

However, there are infinite sets that are uncountable i.e., sets that have cardinality
strictly larger than 0 . The main example are the real numbers, and the proof is based
on Cantors diagonal argument.
15

Figure 2: Finding a denumerable list of all positive rationals

Theorem 9 The cardinality #(R) is strictly larger than #(N ).

The notation that is sometimes used for the cardinality of the real numbers is #(R) = c.
(Here c stands for continuum.)
Proof. Since [0, 1] R, it suffices to show that #([0, 1]) > 0 . We will prove this by
contradiction.. Assume that there is a denumeration (xn )nN of [0, 1]. We write down the
decimal expansions of each xn in a countable list.
x1 = 0.x11 x12 .x13 x14 x15 x16 .x17 x18 . . .
x2 = 0.x21 x22 .x23 x24 x25 x26 .x27 x28 . . .
x1 = 0.x31 x32 .x33 x34 x35 x36 .x37 x38 . . .
x2 = 0.x41 x42 .x43 x44 x45 x46 .x47 x48 . . .
..
..
..
.
.
.
where xij {0, 1, 2, 3, 4, 5, 6, 7, 8, 9} is the j-th digit of the i-th number on the list. Now
form the diagonal number

5
if xjj 6= 5
d = 0.d1 d2 d3 d4 d5 d6 . . . where dj =
.
6
if xjj = 5
Hence the j-th digit of d is always different from the j-th digit of the j-th number on
the list. Therefore d 6= xj for each j N. In other words, d is not included in the
denumeration (xn )nN of [0, 1], which contradicts that (xn )nN is a good denumeration (it
is not surjective). From this contradiction, we conclude that [0, 1] is not countable.
However, there is a surjection f : [0, 1] N. For example,
(
n if x = n1 for some n N,
f (x) =
1 if x [0, 1] is not of that form
16

is such a surjection. Therefore #([0, 1]) #(N). We already know that #([0, 1]) 6= #(N),
and therefore #(R) #([0, 1]) > #(N) is the only possibility.

Exercises:

1) Show the following statements.

a) If A B then #(A) #(B).
b) If g : A B is an injection, then #(A) #(B). (Hint: Given g, find a surjection
f : B A.)
The following theorem (or rather its proof) is beyond these exercises or the scope of these
notes, but very relevant here:
Theorem 10 (Schr
oder-Bernstein) If there are both a surjection f : X Y and an
injection g : X Y , then #(X) = #(Y ).
In other words, if then #(X) #(Y ) and #(X) #(Y ), then #(X) = #(Y ). This may
seem trivial, but if you look closely how and are defined for cardinalities, it is not
simple at all.
2) Find a bijection between the following sets:
a) N and {n2 : n = 0, 1, 2, 3, . . . }.
b) Z and the even integers.
c) [0, 1] and [20, 10].
d) N and {0, 1, 12 , 13 , 41 , 15 , . . . }.
e) [0, 1] and [0, 1] \ { 12 }.
f) (0, 1) and R.
g) [0, 1] and R.
h) R and R2 .
3) Which of the following sets have the same cardinality?
a) Q; b) [3, 3] Z; c) [0, ); d) the prime numbers
e) {n N : 2n < 16}; f) R \ Q; g) ; h) Q [0, 1].
4) a) Let A and B be countable sets. Show that the Cartesian product A B, i.e.,
the collection of pairs (a, b), a A and b B, is also countable.
b) Conclude that Q2 = Q Q and also Qn = Q Q Q are countable for every
{z
}
|
n

n N.

times

5) Prove the following statements:

a) For every radius R, there a centre C R2 such that the circle of radius R and centred
at C R2 does not intersect any rational point (i.e., any point in Q Q).
b) For every centre C, there is radius R > 0 such that the circle of radius R and centred
at C R2 does not intersect any rational point.
17

c) You can only fit countably many disjoint round discs D in the plane (as long as all
disks have radius > 0).
d) How many disjoint circles (of radius > 0) can you fit into the plane?
NB: Make the distinction between circle and disk: A circle is the edge of a round disc,
and a disc is the interior region of a circle.
6) a) If Ai , i = 1, 2, 3, 4, 5, . . . are finite sets which are also pairwise disjoint (i.e.,
Ai Aj = whenever i 6= j). Show that i Ai is countable. 2
b) Recall from the definition of algebraic number, see (1), that an integer polynomial p is
an expression
p(x) = ad xd + ad1 xd1 + + a1 x + a0 ,
where all coefficient ai Z and the degree is d. Each polynomial of degree d has at most
d solutions. Let us call S(p) = d + |ad | + + |a0 | the size of the polynomial. Show that
for each s N, there are only finitely many polynomials p such that S(p) = s.
c) Conclude that the set of algebraic numbers is countable.
(d) Why was the term d included in the definition of S(p), i.e., wouldnt the proof of
countability work if S(p) = |ad | + + |a0 |?

The Axioms of Ordered Fields

History of axioms. Axioms are rules expressing the very basic properties of mathematical object. They are just given, so we accept them, or leave them, but we dont prove
them. The first to use axioms was Euclid (ca. 300 B.C.); his book on geometry has been
used for over 2000 years as the prime textbook when you wanted to learn about geometry
or logic. The fact that Euclids book never became obsolete is doubtlessly due to his style
of basing all his theorems rigorously on a small set of axioms.
Axiomatic approaches to other parts of mathematics were made from the 19th century onwards. In that time mathematics grew to such an extent and abstraction, that
mathematicians started run into troubles with wrong theorems and proofs, just because
their reasoning didnt have firm foundations. You might think that for simple theorems,
you could rely on your intuition to see that they are true. But gradually mathematicians
found that their intuition was not enough. For example, the Intermediate Value Theorem
(which you will learn in Real Analysis II) is intuitively true, but no rigorous proof was
ever given before the axioms of the real numbers were developed.
Purpose of axioms. In this section, we will discuss the axioms of the real numbers. Rather than constructing real numbers (as we constructed rational numbers from
integers), we give a set of axioms and say that the real numbers are precisely the set of
numbers that satisfy the axioms.
2

Although this is not evident, this result uses the Axiom of Choice, which states that from an collection
of nonempty sets, it is possible to choose an element of each set to form a new set.

18

The basic thing we can do with numbers is add and multiply them, and still have a
number in R:
x, y R, x + y R and xy R.
(A0)
The axioms of these operations are as follows:
x, y, z R (x + y) + z = x + (y + z).

(A1)

x, y R x + y = y + x.

(A2)

n1 R such that x R x + n1 = x.

(A3)

x R x R such that x + x = n1 .

(A4)

x, y, z R (xy)z = x(yz).

(A5)

x, y R xy = yx.

(A6)

n2 R, n2 6= n1 such that x R xn2 = x.

(A7)

1
1
R such that x = n2 .
(A8)
x
x
These axioms show that addition and multiplications are quite similar. Axioms (A1) and
(A5) are called the associative laws of addition and multiplication. Axioms (A2) and
(A6) are the commutative laws of addition and multiplication. Axioms (A3) and (A7)
give the neutral elements n1 of addition and n2 of multiplication. Adding or multiplying
a number with these neutral elements doesnt change the number. Usually we write 0 for
n1 and 1 for n2 . Axioms (A4) and (A8) give the additive and multiplicative inverses, also
called negative for addition and reciprocal for multiplication.
The connection between addition and multiplication is given by the distributive law:
x R with x 6= n1

x, y, z R x(y + z) = xy + xz.

(A9)

Together, these axioms are called the field axioms. In algebra modules, you will see that
field axioms are not just used for numbers. To describe numbers better, we need to give
axioms of the order relation <, saying which of two numbers is the larger.
x, y R either x < y or y < x or x = y.

(A10)

x, y, z R x < y and y < z implies x < z.

(A11)

x, y, z R,

x < y implies x + z < y + z.

x, y, z R with z > 0,

x < y implies xz < yz.

(A12)
(A13)

Axiom (A11) is called the transitivity of the order relation. A set of numbers satisfying
all these axioms is an ordered field. The rational numbers are an ordered field, but
the integers are not (because x1 is in general not an integer). The real numbers are
also an ordered field. So just from Axioms (A1)-(A13), you cannot tell Q apart from R.
19

The crucial extra axiom for the real numbers (which also uniquely determines the real
numbers) is given in Section 10
Using the axioms, you can prove more properties of numbers. For example
There exists exactly one neutral element of addition in R.

(7)

Proof. By Axiom (A3) there is at least one neutral element. Assume by contradiction
that there are at least two. Let us call them 0 and 0. Then we can compute 0 + 0 in two
different ways.
0 + 0 = 0

by Axiom (A3) applied to 0

and
0 + 0 = 0 + 0
= 0

by Axiom (A2)
by Axiom (A3) applied to 0


Therefore 0 = 0.

Exercises:
1) Use the axioms to show that
a) There is exactly one neutral element of multiplication.
b) x x 0 = 0.
c) If z 6= 0 and xz = yz then x = y.
d) If 2 is the notation for 1 + 1, then x + x = 2x.
e) 0 = 0 and 11 = 1.
f) For each x R is only one x and if x 6= 0, then there only one x1 .
g) (1) (1) = 1.
h) If x > 0 then x < 0.
i) 0 < 1.
j) If x > 0 then x1 > 0.
k) If 0 < x < y, then 0 < y1 < x1 .
l) If xy > 0 then either both x, y > 0 or both x, y < 0.
m) If x > 0, y > 0 and x2 < y 2 , then x < y.
n) Show that if z < 0 and x < y, then y z < x z.
2) Subtraction is not associative. For example (2 2) 2 6= 2 (2 2). But on the other
hand x y is just x + (y), and addition is associative. So what is going on here?
3) What is the additive inverse of the neutral element of addition? Why is it that it has
no multiplicative inverse?

Inequalities

In this section we present some inequalities and ways to keep work with them. This is
useful if we need to keep track of inequalities for longer computations.
20

The absolute value of a number x is defined

x
0
|x| =

as
if x > 0;
if x = 0;
if x < 0.

So we always have |x| 0.

Theorem 11 (Triangle Inequality) |a + b| |a| + |b| for all a, b R.
A corollary of this is |x y| = |(x z) + (z y)| |x z| + |z y|. This formula
|x y| |x z| + |z y| is also called the triangle inequality.

zs

@
@
@

@
@
@s

The name triangle inequality comes from geometry.

If is a triangle with vertices x, y and z, then the
straight path between x and y has length |x y|.
It is shorter than the path via z which has length
|x z| + |z y|.

Proof. Because |a| = a or a, we have |a| a |a|. Similarly, |b| b |b|. Add
these inequalities (using Axiom (A12)) to get
(|a| + |b|) a + b |a| + |b|.
Therefore |a + b| |a| + |b|.

Let f : D R be a function. Here D is the domain, i.e., the set of points where f is
defined. We will take D to be a subset of R.
Definition 12 The function f : D R is increasing if for all x, y D, x < y implies
f (x) f (y). It is strictly increasing if for all x, y D, x < y implies f (x) < f (y).
The function f : D R is decreasing if for all x, y D, x < y implies f (x) f (y). It
is strictly decreasing if for all x, y D, x < y implies f (x) > f (y).
The function f is (strictly) monotone if it is (strictly) increasing or (strictly) decreasing.
Note that if a function f is increaing on one part of itas domain, but decreasing on
another (such as is the case with f : R R, f (x) = x2 , for example, then f is neither
increaing nor decreasing. Is there any function that is both increasing and decreasing?

21

1
2.5

0.5
2

1.5

0.8 0.6 0.4 0.2

0.2

0.4

0.6

0.8

x
2
1

0.5

0.5
1

0.8 0.6 0.4 0.2

0.2

0.4

0.6

0.8

Figure 3: The functions f (x) = ex and f (x) = x3 are both strictly increasing; the function
f (x) = bxc is increasing, but not strictly.
Knowing whether functions are increasing or decreasing helps in dealing with inequalities.
strictly increasing
f (x) = ax + b
f (x) = xa
f (x) = ex
f (x) = ln x
f (x) = sin x
f (x) = tan x

strictly decreasing
x R, a > 0
x (0, ), a > 0
xR
x (0, )
x [ 2 , 2 ]
x ( 2 , 2 )

Example: Suppose we want to know whether

proceed in steps:

f (x) = ax + b
f (x) = xa

x R, a < 0
x (0, ), a < 0

f (x) = cos x

x [0, ]

1
1+2e3

<

1
1+2e4

or

1
1+2e3

>

1
.
1+2e4

We can

1. 3 < 4 and x 7 ex is increasing, so e3 < e4 .

2. e3 < e4 and x 7 1 + 2x is increasing, so 1 + 2e3 < 1 + 2e4 .
3. 1 + 2e3 < 1 + 2e4 (and both are positive) and x 7

1
x

is decreasing, so

1
1+2e3

>

1
.
1+2e4

Exercises:
1) Decide if the following functions are (strictly) increasing or decreasing on their domains.
a) f (x) = x2 , x R.
b) f (x) = x5 , x R.
c) f (x) = sin x, x [/2, /2).
d) f (x) = x1 , x R \ {0}.
e) f (x) = x12 , x (0, ).
f) f (x) = bxc, x R. (The function bxc stands for the largest integer x. For example,
b1 21 c = 1 and b2 23 c = 3.)
2) a) Is the sum of two increasing functions increasing or decreasing or neither?
b) What about the difference of two decreasing functions?
22

c) Is the composition of two increasing functions increasing or decreasing or neither? (A

composition of functions f and g is the function x 7 f (g(x)).)
d) What about the composition of two decreasing functions?
e) What about the composition of an increasing and a decreasing function?
3) Show from the axioms that |x| 0.
4) Show that:
a) |a b| | |a| |b| | for all a, b R.
b) |x y| | |x z| |z y| | for all x, y, z R.
5) Put the correct inequality sign in place of the :
a) sin(3/10) sin(4/10).
b) 32 42 .
1
n12 .
c) Assuming n + 10 < n2 : n+10
d) Assuming n + 10 < n2 : ln(n + 10) 2 ln n.

e) Assuming 0 < x < y: x + 1 y + 1.

2
2
f) Assuming 0 < x < y: e1/x e1/y .

g) Assuming 0 < x < y < 1: tan x tan y.

1
h) Assuming 0 < x < 1: 1 + x 1x
.
6) Given > 0, how large do you need to take N as function of to assure that
a) N 215 < .
b) |(1 + N1 )2 1| < .
c) 1 1 < .
d)
7)

N+
2
N

< .
N+ N
a
If b and dc

Q and

a
b

< dc , show that

a
b

<

a+c
c+d

< dc .

The Principle of Natural Induction

You can count the natural numbers. Of course, there are infinitely many natural numbers,
but if you start 1, 2, 3, 4, etc., then each n N gets a turn. Based on this counting, there is
a method of proving statements of the type n N A(n) holds. Here A(n) is a statement
depending on n, for example 1 + 2 + + n = 12 n(n + 1), or 7 is a divisor of 32n 2n .
The method is called:
Principle of Natural Induction Let n0 N and let A(n) be statements (depending on
n N) such that
1. A(n0 ) is true;
2. for every k n0 , if A(k) is true, then A(k + 1) is also true.
Then A(n) is true for n N .
A Proof by Induction based on this principle has the following structure:
Step 1 (Initial step) Verify that A(n0 ) is true,
23

Step 2 (Induction hypothesis) A(n) is true for some n, say n = k.

Step 3 (Inductive step) Deduce (from Steps 1 and 2) that A(n) is true for n = k + 1.
Step 4 (Conclusion) By the Principle of Natural Induction, A(n) is true for all integers
n n0 .
Example 1: We want to prove the statement 7 is a divisor of 32n 2n .
Step 1 (Initial step) First we show that the statement is true for n0 = 1. For n0 = 1, we
have 32 21 = 9 2 = 7 which is divisible by 7.
Step 2 (Induction hypothesis) 32n 2n is divisible by 7 for n = k N
Step 3 (Inductive Step) Now set n = k + 1. The assumption in Step 2 implies that there
is some integer j such that 32k 2k = 7j. Therefore 32k = 7j + 2k and
32(k+1) 2k+1 = 9 32k 2 2k
= 9(7j + 2k ) 2 2k
= 9(7j) + (9 2) 2k = 7(9j + 2k ),
which is clearly divisible by 7.
Step 4 (Conclusion) It follows by the Principle of Natural Induction, that 32n 2n is
divisible by 7 for all n 1.
We started in the base step with n0 = 1. In fact, we could already have started with
n0 = 0, because 320 20 = 1 1 = 0 is divisible by 7. Therefore 32n 2n is divisible by
7 for all n 0.
Example 2: We want to prove that for all n N
1
(A(n))
1 + 4 + 9 + + n2 = (2n3 + 3n2 + n).
6
Step 1 (Initial step) Check that the statement is true for n0 = 1. For n0 = 1, A(n0 )
means 1 = 61 (2 + 3 + 1), which is indeed true.
Step 2 (Induction hypothesis) A(n) holds for some n = k, i.e.,
1
1 + 4 + 9 + + k 2 = (2k 3 + 3k 2 + k)
for n = k
6

(IH)

Step 3 (Inductive Step) Take n = k + 1. We start with the left hand side of A(k + 1):
1
(2k 3 + 3k 2 + k) + (k + 1)2
by (IH) (8)
1 + 4 + + k 2 + (k + 1)2 =
6
1
=
(2k 3 + 3k 2 + k + 6k 2 + 12k + 6)
(9)
6
1
=
(2k 3 +6k 2 +6k +2 + 3k 2 +6k+3 + k+1) (10)
6
1
=
(2(k + 1)3 + 3(k + 1)2 + (k + 1)),
(11)
6
which is the right hand side of A(k + 1). Therefore A(k + 1) is true.
24

Step 4 (Conclusion) It follows by the Principle of Natural Induction that A(n) holds for
all n N.
A proof like this is not something you guess at once. In practice, you start with one side
of the statement A(k + 1), and try to rewrite it until you have one side of A(k), see line
(8). Then you use the induction hypothesis A(k), see line (9), and leave the obtained
expression for the moment. Then you start with the other side of A(k + 1), see line (11),
and try to work your way backwards until you get the expression that you obtained in
the first part.
Exercises:
1) Prove the following statements by induction:
a) 1 + 2 + + n = 21 n(n + 1) for all n 0.
b) 1 + 3 + 5 + 7 + + (2n 1) = n2 for all n 1.
c) 34n2 + 26n3 is divisible by 17 for all n 1.
d) 34n 1 is divisible by 80 for all n 1.
e) en > n for all n 0.
f) 3n < n! for all n 7. (The notation n! stands for n factorial: n! = 1 2 n. By
convention, 0! = 1.)
g) (1 + x)n 1 + nx for all real numbers x (1, ) and integers n 0.
Q
P
h) nj=1 (1 aj ) 1 nj=1 aj for all n N. (Here aj [0, 1] for all j N.)
P
2) We have seen that the sum of the first n squares is nj=1 j 2 = 31 n3 + 12 n2 + 16 n.
Pn
a) Show that the sum of the first even squares is j=1 (2j)2 = 34 n3 + 2n2 + 23 n.
P
b) Show that the sum of the first odd squares is nj=1 (2j 1)2 = 43 n3 13 n.
3) The n-th triangular number is defined as the number T (n) of dots necessary to
form a triangle with n dots to the side, see below (left). For example T (5) = 15, because
the pattern below has 15 dots.
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@

r
r
r
r
r

r
r

r
r

r
r

r
r

Figure 4: Left: Triangles of dots indicating triangular numbers. Right: Count the number
of smaller triangles in the big triangle.
a) Find the general formula for T (n).
P
b) Show that the sum of the first n triangular numbers A(n) := nj=1 T (j) = 16 n3 + 12 n2 +
1
n.
3

25

c) Show that the even and odd sums of triangular numbers are:
(
P
E(n) := nj=1 T (2j) = 32 n3 + 32 n2 + 65 n,
P
O(n) := nj=1 T (2j 1) = 32 n3 + 12 n2 16 n.
d) Show that the total number of subtriangles (of any size) in the big triangle of n to the
P
side, see Figure 4 (right), is A(n) + bn/2c
j=1 T (n + 1 2j), where bxc stands for the largest
integer x.
(For example, if n = 3, there are 6 triangles of size 1 (6 dots), 4 triangles of size 2 and 1
triangle of size 3, making a total of 13.)
e) Check that this number is equal to
 1 3 5 2 1
n + 8n + 4n
if n is even,
4
1 3
5 2
1
1
n + 8 n + 4 n 8 if n is odd.
4

Upper and Lower Bounds, Supremum and Infimum

Definition 13 If A is a subset of R, then x R is an upper bound if a x for all

a A. Similarly, y R is a lower bound is a y for all a A. If A has an upper
(lower) bound, then we call A bounded from above (below). A set that is both bounded
from above and below is called bounded.
Definition 14 If A is a subset of R, then x is the maximum of A, if x A and x a
for all a A. Similarly, x is the minimum of A, if x A and x a for all a A.
Definition 15 If A is a subset of R, then an upper bound x of A is called the least
upper bound or supremum if there is no upper bound of A which is smaller than x.
Similarly, a lower bound y of A is called the greatest lower bound or infimum if
there is no lower bound of A which is greater than y.
The notation for the maximum, minimum, supremum and infimum of A is max A,
min A, sup A and inf A respectively. If A has no upper bounded (i.e., A is unbounded
above) then sup A and max A do not exist. Similarly, if A has no lower bounded (i.e.,
A is unbounded below) then inf A and min A do not exist. If max A exists, then it
is always an upper bound (and in fact equal to sup A). But not every (bounded) set
has a maximum. For example, if A = (0, 1). Then 1 is the supremum, but there is no
maximum! To show this, assume by contradiction that x A is the maximum of A. Then
A and 1+x
> x. This contradicts that x is the maximum of A. Similarly, the
also 1+x
2
2
infimum of (0, 1) is 0, but there is no minimum.
Exercises:
1) Suppose that min A exists. Show that inf A = min A.
2) What are min N, max N, inf Z and sup Z?
3) Find the maximum, minimum, supremum and infimum of the following sets:
26

a) { n1 : n N};
b) (0,
1] [ 12 , 2);
2
c) [ 3 , 7] Q;
d) {tan n : n Q} (For trigonometric functions such as sin, cos and tan, we will always
compute angles in radians.)
e) {sin n : n Z}. (You may use the statement: x R, p Z, q N such that
|x pq | q12 .)
4) Find the maximum, minimum, supremum and infimum of:
2 (sin x)2
a) The function f : R R, f (x) = x 1+x
2
1+(1)n
.
b) The sequence (an )nN for an =
n
1
1
c) The set A = nZ\{0} ( n2 +1/2 , n2 1 ).
2
5) Let f : R R and g : R R be bounded functions. Show that the following statements
are true or give a counter examples:
a) sup{f (x) + g(x) : x R} = sup{f (x) : x R} + sup{g(x) : x R}
b) sup{f (x) : x R} = inf{f (x) : x R}
c) sup{f (x) : x [1, 1]} sup{f (x) : x R}
d) sup{f (x2 ) : x R} = sup{f (x) : x R}.

10

The Axiom of Completeness

The final axiom that the real numbers satisfy is the Axiom of Completeness:
Every nonempty subset of R that is bounded above has a supremum in R.

(A14)

By this axiom, you can show that numbers such as 2 exist in R. We will prove this here
using a version of proof by contradiction. Consider for example the set A of all numbers
x such that x2 2. This set is nonempty and bounded from above, for example by 2.
Axiom (A14) says that sup A exists; call it y. Then y 2 = 2, because the other possibilities
y 2 < 2 and y 2 > 2 both lead to a contradiction:
Assume that y 2 > 2, so there exists > 0 such that y 2 4 > 2. Then
(y )2 = y 2 2y + 2 > y 2 4 > 2.
Therefore y is also an upper bound of A and y cannot be the least upper bound.
If y 2 < 2, so there exists > 0 such that y 2 + 5 < 2 and 2 < . Then
(y + )2 = y 2 + 2y + 2 < y 2 + 5 < 2.
Therefore also y + A and y cannot be the least upper bound.

All real numbers are finite. One way of stating this is the Archimedean3 Property:
3

Archimedes of Syracuse (287?-212 B.C.), but in fact the property goes back to Eudoxus of Cnidos
(408-355 B.C.), as Archimedes acknowledges

27

 x R N N such that x < N .

(12)

The proof of this property is based on the Axiom of Completeness 4 .

Proof. For the Archimedean Property to make sense, it is important that N R. Let us
first show why that is. The natural numbers N can be defined as the set of those numbers
obtained from adding 1 to itself repeatedly. In other words, define N recursively as 1 N
and if n N, then also n + 1 N. Therefore, we can use induction to prove that N R:
Basic Step: 1 R. This is true because 1 = n2 is neutral element of multiplication, see
Axiom (A7).
Induction hypothesis: n R for some n N.
Induction step: n + 1 is the sum of n and 1. 1 R, see the basic step, and n R by
the induction hypothesis. Therefore n + 1 R by Axiom (A0)
Conclusion: By the Principle of Natural Induction, n R for every n N, so N R.
Now assume by contradiction that the Archimedean Property is false, i.e., by negating
(12)
x N N N N x.
Then N is bounded above, so by (A14), sup N exists in R. Call it S. Then there is n N
such that n > S 1, because if n S 1 for all n N, then S 1 is an upper bound
of N, and S was not the supremum. By Axiom (A12), also n + 1 > S. But n + 1 N, so
S 6= sup N. This is a contradiction, so the Archimedean Property is true.

Conclusion:
Infinity is NOT a real number. Every time is used in real analysis, it
is a notation for a slick way of avoiding infinity.
Theorem 16 Between every two real numbers x and y with x < y, there is a rational
number p/q such that x < p/q < y.
2
R,
Proof. Since y > x we can use Axiom (A12) to get y x > x x = 0 Therefore yx
2
so by the Archimedean Property, there exists q N such that q > yx . Now use Axioms
(A9) and (A13)
2
qy qx = q(y x) >
(y x) = 2,
yx

so qx and qy are at least 2 apart. Therefore, see Exercise 4 below, there is p Z such
that qx < p < qy. Because q > 0, also 1q > 0, see Exercise 1j) of Section 6. So we can
The Archimedean Property holds for both Q and R, but it does not follow from Axioms (A1)-(A13)
alone.
4

28

multiply with

1
q

use Axiom (A13) again to conclude

x<

p
< y.
q


Exercises:

1) Prove that 3 exists in R.

2) Show that every set which is bounded from below has an infimum in R.
3) Prove (using the axioms of R) that Q R. You can do this stepwise as follows:
a) Show that 1 R;
b) Show that N R;
c) Show that Z R;
d) Show that p, q Z, q 6= 0, then p/q R.
4) Assume that x, y R are such that y x 2. Show using the axioms of R that there
is p Z such that x < p < y.
5) For which of the following sets is the Axiom of Completeness true?
(i) N, (ii) R \ Q, (iii) [1, 1], (iv) nZ (n 13 , n].

11

Sequences

Definitions. Sequences are infinite lists of numbers. For example

1, 2, 4, 8, 16, 32, . . .

(13)

2, 2, 2, 2, 2, 2, 2, 2, 2, . . .
1 1 1 1 1
1, , , , , , . . .
2 3 4 5 6
3, 3.1, 3.14, 3.141, 3.1415, 3.14159, . . .

(14)

1, 2, 3, 5, 8, 13, 21, 34, 55, . . .

(17)

(15)
(16)

The shorthand for a sequence is (an )

n=1 or (an )nN ; n is called the index and an the nth
term of the sequence. We can write the above examples as (an )nN with an = 2n1 (in
(13)), an = (1)n 2 (in (14)), and an = n1 (in (15)), in sequence (16) the nth term is
the first n digits of . So for these sequences, we have a direct expression for each term.
Sequence (17) is can be defined recursively: each term depends on the previous terms;
here an is the sum of an1 and an2 . This particular example is called the Fibonacci
sequence.
Long term behaviour of sequences. It is important to be able to recognise (and
compare) the long term behaviour of sequences. For many sequences it is easy to see that
the terms an grow larger and larger (or smaller and smaller) as n increases. For example
29

the terms of sequence of (13) are eventually larger than those of sequence (16) which in
turn are larger than those of sequence (15). But what if the terms of two sequence (an )nN
and (bn )nN both big? For example, an = 2n and bn = n5 :
n

...

an 2 4
8
16
32
64
128
256
512
...
bn 1 32 243 1024 3125 7, 776 16, 807 32, 768 59, 049 . . .
It seems from this table that (bn )nN has larger terms than (an )nN , but in fact, this is no
longer true for n very large. You can compute that
a23 = 8, 388, 608 > 6, 436, 343 = b23
and from that moment on (i.e., for n 23), an > bn . The reason (and in time you
should immediately recognise this), is that (an )nN grows exponentially and (bn )nN only
polynomially, and exponential beats polynomial. The following theorem gives a more
extensive classification.
Theorem 17 For each fixed a > 0,
ln n

<

logarithmic

na

polynomial

en

<

<

n!

exponential

< nn < en ,

factorial

for n sufficiently large.

Proof.
Let us start with the inequality na < en . We already know that x < ex for all x R
1
(see Exercise 1e of Section 8). Take x = 2a
n. Then
1
n 1
1
n < e 2a n = (e 2 ) a ,
2a

Multiply through with 2a and take the ath power on both sides:
n

na < (2a)a e 2 .
n

For n sufficiently large, (2a)a < e 2 , so the right hand side is

n

(2a)a e 2 < e 2 e 2 = en .
Therefore na < en .
a

For the inequality ln n < na , apply the previous inequality with n 2 instead of n, so
a

n 2 < en 2 .

30

Take the ln on both sides and multiply through with a2 :

ln n =
For n sufficiently large,

2
a

a
2
2 a
ln a 2 < n 2 .
a
a

< n 2 , so the right hand side is

a
a
2 a
n 2 < n 2 n 2 = na .
a

It follows that ln n < na .

Now for the remaining inequalities en < n! < nn < en notice that each number
involves n factors:
2

en = e| e {z
e e}
n

factors

n! = 1 2 (n 1) n
|
{z
}
n

= n
n n}
| n {z
n

n2

factors

factors

= e| e {z
en en} .
n

factors

In each next line the factors get larger (except for the two first factors e and e in line
1 compared with 1 and 2 in line 2, and these two factors are insignificant compared
to the many factors where e < n), so when n is large enough, each next line is much

larger than the previous.
Comparing sequences. When comparing seuqences (an )nN and (bn )nN for large n, we
look for the part the grows fastest. We call this the leading term.
Example: Which sequence grow faster in the long run:
a n = n 3 + en

bn = n5 + (ln n)5 ?

or

Answer: bn has a logarithmic term and its polynomial term dominates the polynomial
term of an (because it has a higher exponent), but the important thing is that an has an
exponential term. Thus the leading term of an is en , the leading term of bn is n5 . By
Theorem 17, en beats n5 , and therefore an > bn for n sufficiently large.
(NB: Please dont get confused with the multiple usage of the word term. The n-th
term of the sequence (an )nN is an ; this has two terms, n3 and en , and the leading term
of these is en .)
Example: Which sequence grow faster in the long run
an = 2n + n2

bn = n5 +

or
31

1 n
e ?
1000

Answer: an and bn both have an exponential term, so the polynomial terms do not matter.
However, the exponential term of bn dominates the exponential term of an , because 2 < e.
1
The factor 1000
does not matter. Therefore an < bn for n large.

Example: Which sequence grow faster in the long run

an = n + 5
or
bn = (ln(n + 5))20 ?
Answer: The leading term of an is polynomial whereas the leading term of bn is only
logarithmic. Therefore an > bn for large n. You may think that the power in the term
(ln n + 5)20 is important, but if we take the 20th root of an and bn we get
p

1
20
20
an = (n + 5) 40
and
bn = ln(n + 5).
Now the power in bn has disappeared, and we are almost exactly in the situation of
Theorem 17.
Example: Which sequence grow faster in the long run
an = nn + en n10

bn = nn/2 + 10en en ?

or

Answer: The leading terms of an and bn are nn and en respectively. en dominates nn ,

2
but the coefficient of en is negative, so bn < 0 < an for n large. However |bn | > |an | for
n large.
2

Just as we can compare sequences with large terms, we can compare sequences with very
small terms (i.e., close to 0).
Example: Which sequence decrease faster in the long run
an =

1
en

or

bn =

n10

1
?
+ n5

Answer: The exponential and polynomial terms are in the denominator of the fractions.
As en > n10 + n5 for large n, we get e1n < n101+n5 . Both an and bn go to 0, but an goes
faster.
We can take a step further, and see what happens to an if there are interesting terms
both in the numerator and the denominator.
Example: What happens to
en + n2
ln n + n!
as n becomes large? The leading terms of numerator and denominator are en and n!
respectively. But n! dominates en , so an becomes very small (close to 0) as n becomes
large.
an =

Example: What would you think of

bn =

2n + 4n
?
4n + n
32

Now the leading terms of numerator and denominator are the same (both 4n ). The table
below suggests that bn is close to 1 as n grows large.
n

bn

5
10
15
20

1.026239067
1.000967017
1.000030504
1.000000954

The second largest term of the numerator 2n dominates the second term of the denominator n. But this is of no importance, because the leading terms dominate everything. A
way to see this is as follows:
bn =

2n + 4n
4n + n + 2n n
=
4n + n
4n + n
= 1+

2n n
,
4n + n

and now 4n dominates the leading term 2n of the numerator of the fraction in the last
line.
Estimating n factorial. To estimate n! very accurately, there is a formula, known as
Stirlings5 formula one form of which reads as follows:

1
nn en 2n n! nn en 2n (1 + ).
4n
This shows that nn /n! grows almost as fast as en , and therefore faster than 2n and slower
than 3n .
Exercises:
1) Give the general an term for the sequences:
a) 1, 31 , 15 , 17 , 19 , . . . ;
b) 1, 2, 3, 4, 5, 6, 7, . . . ;
1
1
c) 1, 12 , 16 , 24
, 120
,...;
d) 1, 2, 5, 12, 27, 58, 121, 248, 503, 1014, . . .
2) What happens to an for n very large, when the n-th term of (an )nN equals:
a)

n3 n10
;
2n9 +3n7

4n +3n
;
4n +n!
1
2n < 100 en

b)

c)

2n +10n
;
n2n +2

d)

n3 +ln n
;
4n3 +n2

e)

4n +(3)n
;
4n +3n
n

3) Prove that
for n sufficiently large. For which N is 2 <
true?
4) Fix p > 0. Show that for n sufficiently large:
p+1
c) np < 2n ;
a) (ln n)p < n;
b) nn < en ;
5

f)
1 n
e
100

n3 n9 + n
;
9
2n +ln n

for n N

James Stirling 1692 - 1770 was a Scottish mathematician and mining engineer; also a friend of Isaac
Newton.

33

12

Limits and Convergence

In Section 11 we introduced the notion of sequence (an )nN and we discussed, rather informally, what happens when n is large. In this section, the notions of limit and convergence
will be defined precisely.
Definitions. Recall the examples of sequences (an )nN from Section 11:
1, 2, 4, 8, 16, 32, . . .
2, 2, 2, 2,
1 1 1
1, , , ,
2 3 4
3, 3.1, 3.14,

2, 2, 2, 2, 2, . . .
1 1
, ,...
5 6
3.141, 3.1415, 3.14159, . . .

i.e., an = 2n1

(18)

i.e., an = 2 (1)n
1
i.e., an =
n
i.e., an = first n digits of

(19)
(20)
(21)

We are interested in what happens to an when n gets large. For sequence (18), the terms
get larger and larger, and in sequence (19), the terms are quite predictable, but still hop
around between 2 and 2. Only for sequences (20) and (21), the terms approach a certain
number (0 and respectively) more and more closely. We say therefore that sequences
(20) and (21) converge, with limits 0 and respectively. Sequences (18) and (19) on the
other hand diverge. Sometimes we say that (18) diverges to infinity.
The formal definitions are as follows:
Definition 18 A sequence (an )nN has limit L if
> 0 N N such that n > N, |an L| < .
We write an L as n , or limn an = L.
has a limit L R.

A sequence is said to converge if it

Figure 5 illustrates a convergent sequence in a picture. If L is the limit and is given,

convergence means that the terms of (an )nN eventually (i.e., for n > N for some N ) fall
within an -band around L, and remain there.
Convergence proof template. You should be able to prove that certain sequences
converge. One important way of doing this is by first principles, i.e., by checking the
definition. As the definition consists of a list of quantifiers, we will use a template that
in successive lines verifies the use of the successive quantifiers (in the same order). For
a universal quantifier , we need to cope with arbitrary variables (here and n). For
an existential quantifier , we can choose a value for the variable (here N ), but it is only
allowed to depend on variables chosen before (so on , but not on n). In the last line,
we verify the final part of the definition (here |an L| < ).
Example: The sequence (an )nN given by an =
Proof. Take > 0 arbitrary.
6

But dont write limn an L

34

n3
n

converges to 1.

an 6
s

s
s

L+
L
L

s
s

s
s

3 ...

Figure 5: Picture of a convergent sequence.

Let N = b 3 c + 1, so N > 3 and

Take n > N arbitrary. Then
|an 1| = |

3
N

< .

n3
n3n
3
3
3
1| = |
|=| |= <
< .
n
n
n
n
N


Again note that in this proof, the variables , N and n are introduced in exactly the same
order as they appear in Definition 18, see the bold-face parts. This is important, because
this kind of proof checks whether the definition holds.
However, to find the proof of this type, you tend to work in the other direction, because
usually you cannot guess what N should be before you do the calculation of the last line.
The advice is therefore to first write the bold-face part as a template, and then work
your way backwards.
Further hints to using the template: Manipulate the expession |an L| to form a single
fraction nE (for some power , quite often equal to 1 or 12 ). Then it becomes easier to see
whether it is decreasing in n, so that you can replace n N by N . It helps to remember
inequalities such as

n<

n
+1
2

n2

1
1
1
2
< 2 but 2
< 2
+2
n
n 2
n

for n N. Finally, keep the absolute value signs until you are sure that the expression
within them is positive. Convergence proof template. In a similar fashion it is possible
to show that a limit does not exist. A specific case of this is when the terms of a sequence
grow arbitrarily large: this is divergence to infinity:
Definition 19 A sequence (an )nN diverges to if
M > 0 N N such that n > N, an > M.
The notation for this is an as n .
35

Similarly, a sequence (an )nN diverges to if

M > 0 N N such that n > N, an < M.
We write an as n . Note that although the symbols and are used,
they shouldnt be thought of as numbers.
We can prove divergence to infinity by checking the definition, using a similar template
as before. But be aware that the inqualities (here an M ) go in the other direction.
Example: The sequence (an )nN given by an = n2 10n diverges to .
Proof. Take M > 0 arbitrary.
Let N = max{M + 1, 11}.
Take n > N arbitrary. Then n > N 11, so n 10 1 and
an = n2 10n = n(n 10) n > N > M.

Again the parts that you can write down immediately as a template are in bold-face.
Having this, you can work your way backwards to find an adequate value of N .
Definition 20 A sequence (an )nN converges if it has a limit L R. If there is no limit,
then the sequence diverges. The sequence also diverges if an or an as
n .
Remark: If an > x for all n, then limn an x if the limit exists. Note that limn an =
x is possible, see for example (20).
Exercises:
1) Determine for each of the following sequences if they converge or diverge (use the
definition!):
n
;
a) an = 2n+7
n
b) an = n2 +1 ;
99999n
;
c) an = 1+n
n
d) an = sin(2/n) (Hint: | sin x| |x|);
e) an = 4 12 ;
f) an = 21/n (Hint: (1 + )N > 2 for N > 1/);
2) Show that a sequence can have no more than one limit.
3) Suppose that (an )nN is a convergent sequence such that an > x for all n. Show that
limn an x if the limit exists. Give an example where limn an = x.

4) Assume that limn an = 0. Show (using the definitions) that limn 3 an = 0.

13

Convergence Properties of Sequences

Combining convergent sequences. It is often tedious to always prove convergence

from first principles. In this section, we give some properties that can help you determine
and prove if a sequence is convergent or divergent.
36

Proposition 21 Let (an )nN and (bn )nN be convergent sequences. Then
1. limn k an = k limn an for each k R.
2. limn |an | = | limn an |.
3. limn (an + bn ) = limn an + limn bn .
4. lim (an bn ) = lim an lim bn .
n

an
limn an
=
, provided limn bn 6= 0.
n bn
limn bn

5. lim

6. If an < bn for each n N, then limn an limn bn .

Proof. We prove Statements 4 and 5. The other ones are left as exercises.
Statement 4.: Take > 0 arbitrary. The sequences (an )nN and (bn )nN converge, say
the limits are and respectively. Convergent sequences are bounded, so there are
A, B > 0 such that |an | < A and |bn | < B for all n. Note that || A and || B (Show
this yourself!).
By definition of convergence, there are and NA , NB such that
n > NA , |an | <

2B

n > NB , |bn | <

and

.
2A

Let N = max{NA , NB }, and n > N . Then n > NA and n > NB . Therefore

|an bn | = |an (bn ) + (an )|
|an ||bn | + |||an |

< A
+B
= .
2A
2B
With this computation, we have verified the definition of limn an bn = .
Statement 5.: Because limn

an
bn

= limn an

lim

1
,
bn

it suffices to prove that

1
1
=
bn
limn bn

and then use Statement 4.

Let = limn bn as above. By assumption 6= 0.
Let > 0 be arbitrary.
Take N so large that for all n > N , |bn | < 21 max{ 2 , ||}. In particular, |bn | > ||/2.
Therefore, for n N , we have
|

1
| bn |
2 /2
1
|=
< 2
= .
bn
|bn |
/2

So we have checked the definition of limn

1
bn

37

= 1 .

It may be tempting to split seqeunces and compute the limits the separate sequences,
but the separate sequences may not always be convergent, and then you cannot use
Proposition 21.
n2 + 1
n2

.
n
n
n+1
The separate sequences give

Example: Calculate lim

n2 + 1
=
n
n
lim

and

n2
= .
n n + 1
lim

Trying to subtract these two is nonsense! However, if you write the difference of the two
fractions as a single fraction, you get
n2 + 1
n2
(n2 + 1)(n + 1) n3

=
n
n+1
n(n + 1)
2
n +n+1
=
n2 + n
1
= 1+ 2
1 as n .
n +n
where we could legitimately split the fraction
and n21+n are convergent.

n2 +n+1
,
n2 +n

because the separate sequences 1

Theorem 22 (Sandwich Rule) If (an )nN and (bn )nN are convergent sequences with
the same limit L, and (cn )nN is a third sequence such that an cn bn for all n N,
then (cn )nN is also convergent and limn cn = L.


Proof. See Exercise 1.

Example: Use the Sandwich Rule to find limn

6n
.
3n2

6n
> 6n
= 2.
Answer: Since 3n 2 < 3n, we have 3n2
3n
6n
6n4+4
4
6n
Since 3n2 = 3n2 = 2 + 3n2 and 3n 2 > n, we also have 3n2
< 2 + n4 .
6n
< 2 + n4 .
Therefore we can sandwich 2 < 3n2
Both limn 2 = 2 (n does not even appear!) and limn 2 + n4 = 2.
6n
Thus the Sandwich Rule says that also limn 3n2
= 2.

Convergent, bounded and monotonic sequence. Recall from Definition 13 in Section 9 when a set is bounded. Similarly, a sequence (an )nN is bounded if there exist
m, M R such that for all n N, m an M .
Theorem 23 Every convergent sequence is bounded.
Proof. Let (an )nN be a convergent sequence, say limn an = . By Definition 18, this
means for say, = 1, that
N such that n > N, |an | < = 1.
38

Hence 1 < an < + 1 for all n N .

Now min{an : n N } is a real number, because the minimum is taken over a finite
set. Let m be the minimum of this number and 1. Then m is real, and m an for
all n N.
Similarly, if M is the maximum of + 1 and max{an : n N }, then M is a real
number, and M an for all n N.
Therefore (an )nN is bounded, with upper bounded M and lower bound m.


Definition 24 A sequence (an )nN is:

increasing if an+1 an for all n.
decreasing if an+1 an for all n.
monotonic if it is increasing or decreasing.
alternating if its terms are alternatingly positive and negative.
If an < an+1 for all n, or an > an+1 , then (an )nN is called strictly increasing and
strictly decreasing respectively.
Theorem 25 Every bounded monotonic sequence is convergent.
Proof. We give the proof for the case that (an )nN is increasing. Because (an )nN is
bounded, supnN an exists; call it L. We will show that L = limn an by checking the
definition (using the template).
Let > 0 be arbitrary.
Take N so that aN > L . This is possible because L = supnN an ; if aN L for all
N N, then L is a smaller upper bound of (an )nN .
Let n N be arbitrary.
Because (an )nN is increasing, an aN > L .
Therefore L < an L and hence |an L| < .

Example: Show the convergence of the sequence (an )nN which is defined recursively by
a1 = 0

an+1 =

2 + an

Answer: We will show that the sequence is increasing and bounded from above. Then
Theorem 25 implies that (an )nN converges.
(i) Prove by induction
(an )nN is increasing.

Step 1: a2 = 2 > 0 = a1 .
Step 2: Induction hypothesis: an+1 an for some n N.

Step 3: Induction step: an+2 = 2 + an+1 IH 2 + an = an+1 = 0.

Step 4 By the Principle of Natural Induction, an+1 an for all n N .
39

(Note also that this means that an 0 for all n N.)

(ii) Prove by induction that an < 7 for all n N. (The number 7 is a bit arbitrary, but
any upper bound is good)
Step 1: a1 = 0 < 7.
Step 2: Induction hypothesis: an < 7 for some n N.

Step 3: Induction step: an+1 = 2 + an <IH 2 + 7 = 3 < 7.

(Here we also usedthat an 0, or at least an 2, because otherwise we would have
taken the square root of a negative number.)
Step 4: By the Principle of Natural Induction, an < 7 for all n N .
As we said before, we can conclude by Theorem 25 that (an )nN is convergent, even
without knowing the limit! However, the limit is not so difficult to find. Since (an )nN is
increasing and bounded, the difference an+1 an must converge to 0. So in the limit

lim an+1 an = lim 2 + an an = 0.

n

This suggest that for = limn an , we have the equality 2 + = 0, so = 2 + .

Squaring this equation gives the quadratic equation 2 2 = 0, with = 2 as only
postive solution.
Subsequences and convergence. The alternating sequence an = (1)n n1
does not
n
converge. However, if we take every second term, say (a2k )kN , then we obtain a convergent
for the index sequence nk = 2k. Many sequences have such
subsequence ank = 2k1
2k
convergent subsequences.
Definition 26 Given a sequence (an )nN , a subsequence is obtained by leaving out
terms from (an )nN . The notation for this is (ank )kN , where (nk )kN is a strictly increasing sequence in N (i.e., nk N and nk+1 > nk for each k).
Proposition 27 If (an )nN is a convergent sequence with limit L, then every subsequence
of (an )nN is also convergent, with limit L.


Proof. See Exercise 2.

Example: We can use Proposition 27 to show that a sequence diverges. For instance,
. The subsequence along the even numbers
take the alternating sequence an = (1)n n1
n
2k 1
2k 1
1
=
=1
1 as k
2k
2k
2k
but the subsequence along the odd numbers
a2k = (1)2k

2k 1
1
2k 1
=
= 1 +
1 as k .
2k
2k
2k
So we find two convergent subsequence with thwo different limits, and therefore the
original sequence (an )n N cannot converge.
a2k1 = (1)2k1

Not every sequence has a convergent sequence, but bounded sequences do.
40

Theorem 28 (Bolzano-Weierstrass) Every bounded sequence in R has a convergent

subsequence.7
Proof. Assume that (an )nN is a sequence, and that all its terms lie in the bounded
interval B. Divide B into two equal closed intervals, B and B + . At least one of them
contains infinitely many points from (an )nN . Call this interval B1 , and let n1 be the first
index such that an1 B1 . Also let b1 be the left endpoint of B1 .
Next divide B1 into two equal closed intervals, B1 and B1+ . At least one of them
contains infinitely many points from (an )nN . Call this interval B2 , and let n2 be the first
index > n1 such that an1 B1 . Let b2 be the left endpoint of B2 .
an1

B
B

B+

B1
b1

an2

B1

B1+
an3 B2
b2 B2

B2+

..
.
Figure 6: Illustration of the proof of Bolzano-Weierstrass.

Do the same to B2 , and then to B3 etc. In the end we have (see Figure 6):
a sequence of closed intervals (Bk )
k=1 with Bk+1 Bk for each k. Moreover, the
k
length of Bk is 2 the length of B;
a subsequence (ank )kN with ank Bk ;
a sequence (bk )kN , which are the left endpoints of the Bk s. Therefore (bk )
k=1 is
monotonic and bounded, so by Theorem 28, bk L for some limit L R.
We claim that L = limk ank . To show this, we use the template:
Let > 0 be arbitrary.
Take K N so that |bK L| < /2, and such that 2K the length of B < 2 .
Take k K arbitrary.
7

Karl Weierstrass (1815-1897) and Bernhard Bolzano (1781-1848) were both major figures in the
foundations of analysis.

41

Then |bk L| < /2 because (bk )kN converges monotonically to L, and

|ank L| = |ank bk + bk L|
|ank bk | + |bk L|
by the triangle inequality

< length of Bk +
because both ank and bk Bk
2

(2k the length of B) +

2

K
(2 the length of B) +
2

<
+ = .
2 2


This proves the theorem.

Standard limits. Here are some standard limits
in Exercises 4 and 5..

1
lim an =

diverges
1
= 1.
n
n
a
lim (1 + )n = ea
n
n

to remember, some of them are proved

for
for
for
for

1 < a < 1,
a = 1,
a > 1,
a 1.

lim n sin

a R.

Exercises:
1) Prove the Sandwich Rule.
2) Let (an )nN be a convergent sequence. Show that every subsequence of (an )nN has the
same limit as (an )nN .
3) Can a sequence (an )nN be both alternating and convergent? If yes, what is its limit?
4) We can derive from the picture that
sin x x tan x,

for every x [0, /2).

x tan x
sin x

cos x
a) Derive the sandwich cos x sinx x 1.
1
b) Use a) to show that limn n sin n = 1.
1
1
5) a) Show that 1+x
1+t
1 for 0 t x.
Rx 1
b) Show that ln(1 + x) = 0 1+t dt for x 0.
x
c) Use a) and b) to show that 1+x
ln(1 + x) x.
d) Find a sandwich for bn = n ln(1 + n1 ) and compute limn bn .
e) What is limn n ln(1 + na ), for a fixed a R?

42

f) Use e) to compute the limit limn (1 + na )n .

6) Use the Sandwich Rule to find the following limits:
2
a) limn nn+1
4 ;
n
b) limn 2n+3 ;
n
2
c) limn 22n n
.
+n7
7) Determine, by any method, whether (an )nN converges:
a) an = n2 en ;
b) an = n (ln n)100 ;
n +(2)n
c) an = 33n+1
;
+2n+1
n
d) an = 1 + n+1 cos(n/12);
e) an = 10n /n!, where n! = 1 2 3 n is n factorial. (Hint: show that an+1 12 an for
n 20);

f) an = n + 2 n (Hint: multiply and divide by n + 2 + n).

8) Show that if (an )nN is a convergent sequence, and an > for every n, then limn an
. Find an example where limn an = .
9) Let an = n12 + n22 + + nn2 . What is limn an ?
10) The number is given by the formula:
=1+

1
1+

1
1+

1
1+ 1

...

a) Write as the limit of a sequence (an )nN of numbers.

b) Draw the graphs of f (x) = 1 + x1 and the line y = x. Indicate the points an on the
x-axis and y-axis for n = 1, 2, 3, 4.
c) Show that a1 < a3 < a5 < < a6 < a4 < a2 .
d) Show that (an )nN converges. Compute the limit.
11) The number is defined as:
v
s
u
r
u
q

t
= 1 + 1 + 1 + 1 + ... .
Write as the limit of a sequence, and prove that this limit indeed exists. Compute .
12) Determine whether limk liml ak,l = liml limk ak,l for the following double sequences ak,l :
a) ak,l = 2kl ;
l
b) ak,l = l+k
;
c) ak,l = (1 1l )k ;
d) ak,l = (1 lk1 )k .
What do you conclude from this exercise about swapping the order of taking limits?

43

14

Cauchy Sequences

In this section, we are going to relate sequences to the Axioms of Completeness of the
real numbers. In short, it is because of the Axiom of Completeness that we can prove
that limits of (some) sequences exist. To explain this, let us look at the sequence (bn )nN
of (16):
3, 3.1, 3.14, 3.141, 3.1415, 3.14159, . . .
that is, bn consist to the first n digits in the decimal expansion of . Note that (bn )nN is
a sequence in Q, because every an Q. (For example, b5 = 3.1415 = 31415
. But the limit
1000
limn an = 6 Q. The rational numbers are not rich enough to contain this limit.
In order to properly address this issue (and historically, it is a way to construct R,
rather than to define R by its axioms), Cauchy8 gave the following definition
Definition 29 A sequence (an )nN is called a Cauchy sequence if
> 0 N N m N n N |an am | < .
Another way of saying this is that |an am | 0 as m, n simultaneously. Note
that in this definition, no limit is mentioned. Cauchy sequences can be defined without
them converging to a limit. They are the closest you can get to a convergent sequence, but
without mentioning a limit, or proving the existence of a limit. However, if the number
system you are working in is rich or complete enough, then every Cauchy sequence
is also a convergent sequence.
Theorem 30 Every Cauchy sequence in R is convergent.
Before we are going to prove this result, lets come back to our sequence (bn )n1 of
decimals of , and show that (bn )nN is a Cauchy sequence. So we are going to check
Definition 29:
Let > 0 be arbitrary.
Take N so large that 10(N 2) = 100 10N < .
Let both m and n N be arbitrary.
That means that bm and bn coincide on their first N 1 digits, and their first difference
is on the N 2 place behind the decimal point. Therefore |bm bn | < 10(N 2) < .
This finishes the check: (bn )nN is indeed a Cauchy sequence. Theorem 30 says that it
converges, obviously to . The set R is complete as to contain a number , but Q is
not.
Lemma 31 Every convergent sequence is a Cauchy sequence.
8

Augustin Louis Cauchy,1789 - 1857, was one of the most prolific mathematicians that France ever
had. Together with Karl Weierstrass, he was one of the most important figures in bringing rigour to
mathematical analysis.

44

Proof. Let (an )nN be a convergent sequence, say with limit L. We are going to check
Definition 29 by applying Definition 18.
Let > 0 be arbitrary.
Then by definition (applied to = /2!)
> 0 N N such that n > N, |an L| < .
Take this N .
Let m N and n N be arbitrary.
Then because of the triangle inequality
|am an | = |(am L) (L an )| |am L| + |L an | < + =

+ = .
2 2


This finishes the proof.

Lemma 32 Every Cauchy sequence is bounded.

Proof. Let (an )nN be a Cauchy sequence. Apply Definition 29 for = 1, that is
N N m N n N |an am | < 1.
This means that for every n > N , aN 1 an aN + 1. Let
M = max{a1 , a2 . . . , aN , aN + 1}

and

m = min{a1 , a2 . . . , aN , aN 1}.

These are the maximum and minimum of a finite set, so they exists. It follows that for
every n N
Then for every n N, we have m an M , because if n N , then an belongs to
the set {a1 , . . . , aN }, and if n > N , then
m aN 1 an aN + 1 M.
We have now found an upper and a lower bound for (an )nN , so (an )nN is bounded.

Now we have done enough preparations to prove Theorem 30.

Proof. Let (an )nN be a Cauchy sequence. By Lemma 32, (an )nN is bounded. Therefore,
by Theorem 28, it has a convergent subsequence, say (ank )kN . Let L be its limit.
Now let us check Definition 18:
1) Let > 0 be arbitrary.
2) Take = /2.
Take N1 so large that for all m, n N1 , |am an | . (This is Definition 29 applied
to (an )nN and .)
Since ank converges to L, we can also find N2 so large that for all k N2 , |ank L| < .
Take N = max{N1 , N2 }.
45

3) Let both m and n N be arbitrary.

4) Then for k N , we know that nk N N1 . Using the triangle inequality, we find
|an L| = |(an ank ) + (ank L)|

|an ank | + |ank L|

+ = .
2 2


This finishes the check and proves Theorem 30.

Exercises:

1) Find a (not constant) sequence that converges in Q. Are there sequences in R \ Q that
converge to a number L Q?
2 Show that every subsequence of a Cauchy sequence is also Cauchy.
3 Show that every Cauchy sequence is Z is eventually constant.
4 In Exercise 5 (iv) of Section 10 we have seen that the set A = nZ (n 13 , n] satisfies
the Axiom of Completeness. Is it true that every Cauchy sequence in A converges? Why
(not)?

15

Series

Series are obtained by summing the terms of a sequence. Before explaining formally
what series are, let us give two examples.
Example: Figure 7 gives an example of a geometric series. The terms are
to 1.

1
2n

add up

1
16
1
128

1
2

1
32

1
8

1
n=1 2n

1
64

1
1
+
+ +
2
4
|{z}
|{z}
first term

1
4

Figure 7: Rectangles of area

second term

1
+
2n
|{z}
general term

= 1

1
,
2n

n = 1, 2, 3, . . . can be fitted into a unit square.

46

Example: On the other hand, when summing the terms

harmonic series), we get no finite answer.

1
n=1 n

= |{z}
1 +
first term

1
,
n

n N (this is called the

1
+ +
2
|{z}
second term

1
+
n
|{z}
general term

diverges to infinity.

Figure 8: Rectangles of areaR n1 , n = 1, 2, 3, . . . , cover the area unther the graph y = x1 ,

1 x < , and this area is 1 x1 dx = [ln x]

1 = .
In both examples, we add infinitely many small positive numbers, called terms. If
these terms dont converge to 0, then the series diverges. But in these examples, the
terms do converge to 0. The question whether the infinite sum of such terms sums to a
finite answer or diverges, depends on how fast the terms converge to 0. The first example
shows that the terms 21n tend to zero exponentially fast, cf. Section 11, Theorem 17. This
is sufficiently fast for the series to convergence. In the second example, the terms n1 do
not tend to zero linearly, but that is not sufficiently fast for the series to convergence.
In fact, this series is more or less on te dividing line. In Theorem 39 we will see that
P 1
n=1 np converges if p > 1 but diverges if p 1. In short,
Infinite many terms can add up to a finite sum, but only if the terms converge
to 0 sufficiently fast.
Definitions. Now we turn to the formal definitions.
P
Let us call the sequence of terms (an )nN , the series
n=1 is obtained by adding the
terms; it is therefore an infinite sum. To deal with this infinity correctly, the series is
defined as the limit of the partial sums, where the kth partial sum of the series is the sum
of the first k terms of (an )nN . The formal definition is as follows:
Definition 33 Suppose we are given sequence (an )nN of real numbers. The sum of the
first k terms is the kth partial sum:
Sk =

k
X
n=1

47

an .

The partial sums form a sequence (Sk )kN . The (infinite) series9 is the limit of this
sequence of partial sums:

X
an = lim Sk .
k

n=1

The numbers an are the terms of the series.

Although in a series the terms are added, the limit is used exactly as for sequences,
namely as the sequence of partial sums.
Definition 34 A series is
convergent if the limit of the partial sums Sk exists;
absolutely convergent if the series of the absolute values of the terms converges,
P
that is
n=1 |an | converges;
divergent if it is not convergent.
Example: Take an =

1
n2

for n N as our sequence. Then the partial sums are

S1 = a 1 = 1
S2 = a1 + a2 = 1 + 14 = 45
S3 = a1 + a2 + a3 = 1 + 14 + 19 = 49
36
..
..
..
.
.
.
2
limk Sk = a1 + a2 + a3 + a4 + = 6
The infinite sum

1
n=1 n2

k=1
k=2
k=3

is difficult to compute, but in this case it is equal to

2
.
6

Properties of convergent series:

P
The terms of a convergent series have to converge to zero: If nN an is convergent,
P
then an 0. In other words, if an 6 0 then
n=1 an diverges, see Exercise 1.
P
The other direction is not true, i.e., an 0 does not necessarily mean that n an
P
1
converges. An example is the harmonic series
n=1 n mentioned above.
Every absolutely convergent series is convergent.
The other direction is not true, i.e., not every convergent series is absolutely converP
n+1 1
gent. An example is the alternating harmonic series
is convern=1 (1)
n
gent (and equal to ln 2), but absolutely divergent, see the Leibniz test in Section 16.
Geometric series. A geometric series is a series in which the terms are powers of a fixed
number r.
9

the plural of series is series

48

Theorem 35 (Geometric Series) The geometric series

1
1. In this case, the limit is equal to 1r
.

P
n=0

rn converges iff |r| <

Proof. Let us multiply the kth partial sum with 1 r and cancel the brackets:
(1 r)Sk = (1 r)(1 + r + r2 + + rk )
= (1 r + r r2 + r2 r3 + . . . rk rk+1 )
= 1 rk+1 .
Since |r| < 1 we can take the limit:
lim (1 r)Sk = lim 1 rk+1 = 1.

Since r 6= 1, we can divide both sides by 1 r.

rn = lim Sk =
k

n=0

NB. If |r| 1, then the terms of the series

Exercise 1) the series diverges.

P
n=0

1
.
1r

xn dont converge to 0. Therefore (see



Example:
32
32
32
+
+
+ ...
2
100 100
1003

X
32
=
100n
n=1

0.3232323232 . . . =

32
1
32
.
1 =
100 1 100
99

From A-level you may remember the formula

P
n=N

arn = a

rN
1r

for |r| < 1.

Example (Decimal Expansions): An example of a series is the decimal expansions of

numbers in [0, 1]. We are used to writing 0.d1 d2 d3 d4 . . . , with dn {0, 1, . . . , 9} the nth
digit of the number. This is shorthand for the infinite series

X
dn
.
0.d1 d2 d3 =
n
10
n=1

Does this series converge? Yes, because the partial sums Sk =

increasing (because dn 0 for each n);
49

Pk

dn
n=1 10n

are

 bounded (because dn 9 for all n, we have Sk =

1
P
( 1 )k+1
1 n
9 kn=1 ( 10
) = 9 10 1101
< 1).

Pk

dn
n=1 10n

Pk

9
n=1 10n

10

By Theorem 25, (Sn )nN converges, and the limit is some number in R. This is the number
that we indicate by 0.d1 d2 d3 . . .
Example (Binary Expansions): We use the decimal system because we count on ten
fingers. If you have only two fingers (such as computers usually have), youd probably
use the binary system, which has just two digits: 0 and 1. Therefore, as binary number
1101binary = 1 23 + 1 22 + 0 21 + 1 20 = 8 + 4 + 1 = 13.
For numbers between 0 and 1, we use the decimal point, for example:
0.101binary = 1 21 + 0 22 + 1 23 =
In general,

X
bi
0.b1 b2 b3 =
2i
i=1

1 1
5
+ = .
2 8
8

bi {0, 1},

and the same argument as in the previous example shows that this infinite sum converges
P
(the partial sums Sk = ki=1 2bii form an increasing and buonded sequence).
Exercises:

P
P
1) Show that if n an converges, then limn an = 0. Hint: If Sn = nk=1 ak are the
partial sums and limn Sn = L, then > 0 N N n N |Sn L| < /2. Show that
also |Sn+1 Sn | < . What does this mean for the sequence (an )nN ?
2) Prove from the definitions and Theorem 35 that 0.9999999 = 1.
3) Show that every rational number has a periodic binary expansion. (Hint: if x = p/q
Q, what happens if you try long division to find its decimal expansion?)
5
.
4a) Find the binary expansions of 7, 0.875 and 6 16
1
b) Find the binary expansion of 3 .
c) Prove from the definitions and Theorem 35 that 0.1111111 . . .binary = 1.
5) Find the sum of the following geometric series:
P
P
P
1 n
n+1
n
; c)
; b)
a)
n=10 ( 4 ) ;
n=0 (0.45)
n=0 (2)
P
P
P
(1)5n +e3n
1 2n
1+2n
d)
e)
f)
.
n=0 ( 9 ) ;
n=0 3n .
n=1
21n

16

Convergence Criteria for Series

It is not always easy to see from first principles if a series converges or not. In this section
we give some tests which can tell you series converges, but not what the limit is. In
fact, for many convergent series, the precise value limit is not known. Dont mix up
converge tests for sequences with convergence test for series!
50

Theorem 36 (Comparison Test) Let (an )nN and (bn )nN be non-negative sequences,
and suppose that bn Kan for all n. Then
P
P
if n an converges, then n bn converges;
P
P
if n bn diverges to , then n an diverges to .


Proof. See Exercise 1.

Example: The terms of the series

3n+1
n=1 n3 +1

satisfy:

X 1
3n + 1
3n
=
+
n3 + 1
n3 + 1 n=1 n3 + 1

1
3n
+ 3
3
n
n
3
1
4

+ 2 = 2.
2
n
n
n
P 4
P 3n+1
The series
n=1 n2 converges, so
n=1 n3 +1 converges too.
P 4
P

1
2 2
n=1 n2 = 4
n=1 n2 = 3 .)

(In fact,

3n+1
n=1 n3 +1

The idea behind the next two theorems comparing the terms with a geometric series
P n
n
n L . We need to find the best L so that an L , and then Theorem 35 tells us that
the series converges if |L| < 1.
Theorem 37 (Ratio Test) Let (an )nN be a sequence such that the limit limn
L. Then

X
absolutely convergent if L < 1;
the series
an is
divergent
if L > 1.

|an+1 |
|an |

If L = 1, then this test is inconclusive.

Proof. First assume that L < 1. Let = (1 L)/2, so L + < 1. By definition
|
of convergence, there exists N such that for all n > N , | |a|an+1
L| < . Therefore
n|
|an+1 |
|an |

< L + < 1 for n N . It follows that

|an | =

|an | |an1 |
|aN +1 |
|aN | (L + )nN |aN |
...
|an1 | |an2 |
|aN |
|
{z
}
nN factor, each L+

Therefore, using Theorem 35,

|an | =

n=1

N
1
X
n=1

N
1
X

|an | +

X
n=N

|an | + |aN |

n=1

N
1
X

|an |

(L + )nN

n=N

|an | + |aN |

n=1

51

1
< .
1 (L + )

If L > 1, then take > 0 such that L > 1. By definition of convergence, there exists
|
|
N such that for all n > N , | |a|an+1
L| < , and therefore |a|an+1
> L > 1. As before,
n|
n|
nN
we get that |an | > (L)
|aN |, so |an | 6 0 as n . Therefore the series diverges. 

Theorem 38 (Root Test) Let (an )nN be a sequence such that the limit limn
L. Then

X
absolutely convergent if L < 1;
the series
an is
divergent
if L > 1.

p
n
|an | =

If L = 1, then this test is inconclusive.

Proof. First assume that L < 1. Let = (1 L)/2, so
L + < 1. By definition of
p
n
convergence, there exists N such that for all n N , | |an | L| < , and therefore
|an |n < (L + )n . Therefore, using Theorem 35
X
X
X
|an | =
|an | +
|an |
n

n<N

nN

|an | +

n<N

(L + )n

nN

|an | +

n<N

(L + )N
< .
1 (L + )

Now if L > 1, then |an | 6 0 as n . Therefore the series diverges.

Theorem 39 (p-test) The series

1
n=1 np

is convergent if p > 1 and divergent if p 1.

Proof. The proof of Theorem 39 that we give is based on the Integral Test, and is best
understood with a graph. Imagine the graph f (x) = x1p for p > 1. The number n1p is the
2

the area of the n-th block is

1.8

1
np

1.6

1
np

1.4

1.2

0.8

0.6

0.4

0.2
1

n1
3

4
x

Figure 9: The graph of f (x) = 1/xp for p > 1.

area of a rectangle of width 1 (between n 1 and n on the horizontal axis) and height
52

1
,
np

Rn
see Figure 9. The area of this rectangle is smaller than n1 1/xp dx, which is the area
between n 1 and n under the graph of 1/xp . Putting all these rectangles together, we
get

Z n
X
X
X
1
1
1
=
1
+

1
+
dx
np
np
xp
n=1
n=2
n=2 n1
Z y
1
= 1 + lim
dx
y 1 xp
1
= 1 + lim
(1 y 1p ) < .
y p 1
For p < 1, we have a similar argument, but the inequalities work in the other direction.
The number 1/np is the area of a rectangle of width 1 (between n and n + 1 on the
R n+1
horizontal axis) and height 1/np . The area of this rectangle is larger than n 1/xp dx,
which is the area between n and n + 1 under the graph of 1/xp . Putting these rectangles
together, we get
Z m
Z n+1

X
X
1
1
1

dx = lim
dx.
p
p
m 1
n
x
xp
n=1 n
n=1
Rm
1
The integral 1 x1p dx = 1p
(m1p 1) as m .
For p = 1, the Integral Test gives
Z y
Z n+1

X
X
1
1
1

dx
=
lim
dx = lim ln y 0.
p
y 1 x
y
n
x
n
n=1
n=1
This diverges too.
P
Some special cases without proof:
n=1

1
n2

2
6

and

1
n=1 n4


=

4
.
90

Example: The harmonic series

X
1
1 1 1 1
= 1 + + + + ....
n
2 3 4 5
n=1

diverges. A nice way of seeing this as follows: We can group the terms into groups that
add up to at least 21 like this:
1 1
1 1 1 1
1
1
1
(1) + ( ) + ( + ) + ( + + + ) + ( + + ) + . . . .
2
3 4
5 6 7 8
9
16
The kth group consists of the numbers 2k11 1 down to 21k , which are 2k1 numbers. The
P
P 1
1
sum of this group is at least 2k1 21k = 12 . Therefore
n=1 n
k=1 2 and this series
diverges.
Alternating Series.
P
called alternating if the terms an are alternately posDefinition 40 A series
n=1 an isP
P
P
P
itive and negative. So we can write n1 an = n (1)n |an | or n1 an = n (1)n1 |an |
depending on whether the first term is negative or positive.
53

Example: The following series is alternating:

X
(1)n
n=1

n!

=1

1 1
1
1
+
+
+ ...
2 6 24 120

In this case, the limit is 1 1e .

Theorem 41 (Leibniz Test) If

P
n

an is

an alternating series, such that

the terms an 0, and
the absolute values |an | form a decreasing sequence,
P
then n an is convergent.
P
n+1
Proof. Assume that a1 > 0 and write the alternating sum as
bn where
n=1 (1)
bn = |an | form a decreasing sequence. Then the even partial sum are
S2n = (b1 b2 ) + (b3 b4 ) + + (b2n1 b2n ).
The sums in the brackets are non-negative, so (S2n ) is increasing. Furthermore (by grouping the pairs differently)
S2n = b1 (b2 b3 ) (b4 b5 ) (b2n2 b2n1 ) b2n .
As the numbers in brackets are non-negative, we get S2n b1 . Therefore (S2n )nN is
bounded as well. By Theorem 25 (every bounded monotonic sequence is convergent),
limn S2n exists.
A similar argument shows that (S2n1 )nN (i.e., the odd partial sums) is a bounded
decreasing sequence, so also limn S2n1 exists. To prove that both limits are the same,
observe that
S2n1 b2n = S2n S2n+1 .
By the Sandwich Rule
lim S2n1 b2n lim S2n lim S2n+1 .

This proves the theorem, because limn S2n1 b2n = limn S2n1 limn b2n =
limn S2n+1 .

P
n+1 1
is convergent - check that
Example: The series 1 12 + 13 14 =
n=1 (1)
n
P
n+1 1
10
the Leibniz Test applies. In fact, the limit is ln 2.
Note that
is not
n=1 (1)
n
10

See Exercise 5a for a proof. Exercise 5b in Section 17 gives another approach to this result (but you
need Abels Theorem as well).

54

P
P 1
n+1 1
absolutely convergent, because
|(1)
|
=
n=1
n=1 n is divergent, as we saw in
n
Theorem 39.
The 1 12 + 31 14 . . . is very strange in the sense that you can rearrange it, and get a
different sum! For example
1 1 1 1 1 1
1
1
+ + + +
+ ...
(22)
3 2 5 7 4 9 11 6
has exactly the same terms, only in a different order. There is a trick to show that (22)
sums up to 23 ln 2. You can use a variation of an integral test to show that
1+

lim (1 +

1
1 1 1
+ + + + ln n)
2 3 4
n

(23)

exists. This limit is called the constant of Euler-Mascheroni 11 and 0.5772156649 . . . ,

but nobody has been able to show if is rational or not. Now let Sn be the nth partial
sum of formula (22). If n is a multiple of 3, say n = 3k, then Sn consists of 2k positive
and k negative terms.
S3k

2k
X

X 1
1
=

2i 1 i=1 2i
i=1
!
4k
2i
k
X
X
1 X 1
1
=

i
2i
2i
i=1
i=1
i=1
=

4k
X
1
i=1

1X1 1X1

.
i
2 i=1 i 2 i=1 i
2k

The sums in the last line can be replaced by ln(ak) + + hak , where a = 4 or 2 or 1, and
hak is an error term that converges to 0 as k . Therefore we get
S3k

1
1
= ln(4k) + + h4k (ln(2k) + + h2k ) (ln(k) + + hk )
2
2
4k
1
1
= ln + h4k h2k hk
2
2
2k k
3

ln 2
as k .
2

Because S3k S3k+1 S3k+2 S3k+3 + k2 , you can use the Sandwich Rule to show that
Sn 23 ln 2 as well.
You can take your favourite number and find a rearrangement of 1 12 + 13 41 . . . that
sum up to it. This is explained by the following theorem:
P
Theorem 42 (Riemanns Theorem on Rearranging Alternating Series) If n an
is an alternating series which is convergent, but not absolutely convergent, then for each
x R, there is a rearrangement of the terms so that the series converges to x. You can
also rearrange the terms such that the series diverges to or to .
11

The Swiss mathematician Leonhard Euler (1707-1783) worked in Germany and Russia, and was
an absolute genius with series. Lorenzo Mascheroni (1750-1803) wrote a treatise on Eulers work and
computed up to 32 decimal places.

55

Exercises:
1) Prove the Comparison Test.
2) Find out if the following series are convergent:
P
n2 +1
a)
,
n=1
P nn4
b) n=1 2n+3 ,
P
2
c)
(recall that 0! = 1),
n!
Pn=0

1
d) n=1 n(n+1)
(can you find the limit?).
3) Suppose that (an )nN and (bn )nN are sequences such that limn an /bn = 2. Show
P
P
that n an converges iff n bn converges.
4) Are the following series convergent or divergent:
P
n
,
a)
(1)n n+2
Pn=0

n
1

b) n=1 n cos 2 ,

P
n n
c)
n=0 (1) n+7 .
P
p
5) Use an integral test to find out for which p R the series
n=2 1/(n(ln n) ) converges.
6) Adjust the arguments following (23) to show that:
a) 1 21 + 13 14 + 15 = ln 2;
1
1
b) 1 12 41 + 13 16 18 + 51 10
12
= 12 ln 2.
7) Let (an )nN be a sequence of real number such that there is > 0 such that an for
all n N. Show that

X
an
= 1,
(24)
(1 + a1 )(1 + a2 ) (1 + an )
n=1
using the following steps.
a) Show that (24) works if an = for all n N (where 6= 1).
P
b) Write Pn = (1 + a1 )(1 + a2 ) (1 + an ) and denote the partial sums as Sk = kn=1 Pann .
Prove by induction that Sk = PkP1
.
k
c) Show that limk Pk = .
d) Conclude that limk Sk = limk PkP1
= 1.
k
P
e) Show that for step c) to work, it is sufficient that
n=1 an = . (Hint: ln Pk =
Pk
n=1 ln(1 + an ) and estimate ln(1 + an ) from below.)
P
1
f) Apply (24) to an = n1 . Show that this is another way to prove that
n=1 n(n+1) = 1.

17

Power Series

A motivating example. Suppose you want to solve the differential equation:

f 0 (x) = f (x),

f (0) = 1.

You remember from A-level that if f (x) = ex , then the derivative of f is the same as f
itself. Furthermore, e0 = 1, therefore f (x) = ex is a solution to the problem. But suppose
you didnt know this, then you could try to use a formal power series
f (x) =

X
n=0

56

an xn .

and try to find all the coefficients an R. The condition f (0) = 1 gives (because x0 = 1
also when x = 0)

X
1 = f (0) =
an 0n = a0 .
n=0

Therefore a0 = 1. Let us compute the derivative of f term-wise:

X
X
X
f 0 (x) = (
an xn )0 =
(an xn )0 =
nan xn1
n=0

n=0

n=1

The second equality is the dubious part (which can be justified by Theorem 46 later on,
but we will not prove that theorem in these notes), but we continue undeterred with the
P
P
n1
equation f 0 (x) = f (x). This gives
= n0 an xn , or, relabelling the terms,
n=1 nan x

(n + 1)an+1 xn =

n=0

an xn .

n0

It is plausible that this equation is only satisfied for all x if the coefficients left and right
are the same for each power. This gives
n=0

a1 = a0

n=1

2a2 = a1

n=1

3a3 = a2

..
.

..
.

a1 = 1.
1
a2 = .
2
1
a3 = .
6
..
.

(n + 1)an+1 = an

an+1 =

general n

1
.
(n + 1)!

P 1 n
x
Therefore the power series solution to our problem f (x) =
n=0 n! x . Indeed e =
P 1 n
n=0 n! x is a true statement.
The above arguments require formal justification. In particular, we want to know:
P
For which x does the formal power series n0 an xn converge, and to what?
The key ingredient to answer this question is the radius of convergence.
Definition 43 Given a power series
convergence, such that

P
n0

an xn , there exists R, called the radius of

the power series converges for |x| < R, and

diverges for |x| > R.
If R = , this is a shorthand for saying that the power series converges for all x R.
The region of convergence is the set of all x such that the power series converges.
57

Example 1: What is the radius and region of convergence for the series

X
n=0

xn
?
3n + 1

If |x| > 3, then the terms dont converge to 0, and hence the series diverges. If |x| < 3,
then the term converge to 0 exponentially fast. Apparently, the radius of convergence is
R = 3. It follows that (3, 3) belongs to the region of convergence. To find the entire
region of convergence, we need to look specifically at the boundary points x = 3:
n

x = 3: In this case limn 3nx+1 = limn 3n3+1 = 1. Hence the terms still dont
converge to 0, and the power series diverges.
n

doesnt exist. Again the terms dont

x = 3: Now limn 3nx+1 = limn (3)
3n +1
converge to 0, and the power series diverges.
We conclude that the region of convergence is (3, 3).
Example 2: What is the radius and region of convergence for the series

X
2n
n=1

Let bn =

n2

xn ?

2n n
x .
n2

In order to use the root test to the sequence (bn )nN , we compute
r
r
r
n
p
2
1
1
n
n
n
n
|bn | =
|xn | = 2n |xn | 2 = 2|x|
2|x| as n .
2
n
n
n2

According to the root test

X


bn is

absolutely convergent if 2|x| < 1;

divergent
if 2|x| > 1.

Therefore the radius of convergence is 12 .

We can also apply the ratio test.
1
)
|bn+1 |
(2n+1 + 1) n2 |xn+1 |
n2 2n (2 + 2n+1
=
=
|x|
2|x|
1
2
n
n
2
n
|bn |
(n + 1) (2 + 1)|x |
(n + 1) 2 (1 + 2n )

as n .

According to the ratio test

X
n


bn is

absolutely convergent if 2|x| < 1;

divergent
if 2|x| > 1.

The conclusion is the same: the radius of convergence is 12 .

We know now that the power series converges for x ( 21 , 21 ), and diverges for x
(, 12 ) ( 12 , ). To find the entire region of convergences, we need to check the
x = 12 and x = 12 separatedly.
58

If x = 21 , then 2n2 xn = n22 2n n22 and according to the Comparison test combined
P
1
n
n
with the p-test,
n=1 (2 + 1)x converges for x = 2 .
n

2
n
n 2 +1
If x = 21 , then 2 n+1
2 x = (1) n2 2n . We cant say that this is n2 , but in absolute
P
n

2
n
n
n
value, we do have | 2 n+1
2 x | n2 . So in fact,
n=1 (2 + 1)x converges absolutely
for x = 12 , and so it converges as well, for x = 21 .

Therefore the region of convergence is [ 12 , 12 ].

Example 3: What is the radius of convergence for the series

X
n! n
x ?
n
4
n=0

Again, let bn =

n! n
x .
4n

Now if we try the root test on bn , we get

r
p
|x|
n
n n!
n
n| =
|bn | =
|x
n!,
n
4
4

and it is unclear how to proceed from here. However, for factorial n!, the ratio test works
better:
|bn+1 |
|xn+1 |4n (n + 1)!
|x| (n + 1)!
|x|
=
=
=
(n + 1)
n+1
n
|bn |
4 |x |n!
4
n!
4

as n

Therefore the series diverges for every x 6= 0. The radius of convergence is 0, and in this
case, the region of convergence is just {0}.
Example 4: A more difficult example is the series

X
(x 2)n
n=1

P
yn
In this case, it helps to substitute y = x 2, so the series becomes
n=1 n . The terms
diverge for |y| > 1 and converge to 0 exponentially fast for |y| < 1. Therefore the radius
of convergence is R = 1. To find the region of convergence, we need to check y = 1.
P
1
y = 1: In this case the series becomes
n=1 n ; this is the harmonic series, and it
diverges.
P
(1)n
y = 1: The series becomes
; this is the alternating harmonic series,
n=1
n
which converges by the Leibniz criterion.
Thus the region of convergence is [1, 1), but this is for y. We must not forget to transform
back to x. As x = y + 2, the true region of convergence is [1, 3).
General approach to finding R. The arguments in this example were a bit ad hoc.
There are more general ways of finding the radius of convergence.
59

p
Theorem 44 Let (an )nN be a sequence such that R = limn 1/ n |an |. Then R is the
P
radius of convergence of the power series n an xn .
Proof. Assume that R (0, ); for R = or R = 0, see Exercise 3. Let bn = an xn .
If |x| < R, then
lim

p
n
|bn | = lim

Therefore, by the Root Test,

If |x| > R, then
lim

Therefore

P
n

limn |x|
R
|x|
p
p
=
<
= 1.
n
n
R
1/ |an |
limn 1/ |an |

P
n

an xn is absolutely convergent.

p
n
|bn | = lim

limn |x|
R
|x|
p
p
=
>
= 1.
n 1/ n |a |
R
limn 1/ n |an |
n


an xn is divergent.

an
Theorem 45 Let (an )nN be a sequence such that R = limn | an+1
|. Then R is the
P
n
radius of convergence of the power series n an x .

Proof. Assume that R (0, ); for R = or R = 0, see Exercise 4. Let bn = an xn .

If |x| < R, then
lim |

By the Ratio Test,

If |x| > R, then

P
n

an xn is absolutely convergent.

lim |

Therefore

P
n

bn+1
|x|
limn |x|
R
| = lim an =
<
= 1.
a
n
n |
bn
|
limn | an+1 |
R
an+1

bn+1
limn |x|
R
|x|
<
| = lim an =
= 1.
an
n
bn
| an+1 |
limn | an+1 |
R


an xn is divergent.

Some power series to remember.

1
1x

1 + x + x 2 + x3 + + x n +

ln(1 + x) = x 12 x2 + 13 x3 + (1)n+1 n1 xn +
ex

1+x+

x2
2

x2
2

x4
24

sin x = x

x3
6

x5
120

arctan x = x

x3
3

x5
5

cos x =

x3
6

+ +

xn
n!

+
2n

x
+ + (1)n (2n)!
+
2n+1

x
+ + (1)n (2n+1)!
+
2n+1

+ (1)n x2n+1 +
60

|x| < 1
x (1, 1]
xR
xR
xR
x (1, 1]

A very useful theorem, which we wont prove in these notes, is the following:
P
n
Theorem 46 If f (x) =
n=0 an x is a power series with radius of convergences R, then
for |x| < R, you can find the derivative and integral of f by differentiating and integrating
the power series term-wise:

X
X
d
0
n
f (x) =
an x =
nan xn1
dx
n=0
n=1
and

Z
f (x)dx =

Z
X
n=0

X
an n+1
an x dx =
x
+ C.
n+1
n=0
n

In Exercises 5 and 6 below, we will discuss some applications of this theorem.

Exercises:
1) Find the radii and regions of convergence of the following power series:
P
(5)n
a)
xn ;
n=1
P n21
b) n=1 2n (n+3) x2n ;
P
c)
sin n xn ;
Pn=1
n n
d)
n=1 n x ;
P
n
e) n=1 nn! xn . For this part, try to find the radius of convergence only. Hint: you may
use that limn (1 + n1 )n = e.
2) Pocket calculators use power series to compute standard functions. Some examples
are given below. Find the radius of convergence, and the region in which the power series
equals the function.
3
n
2
a) ex = 1 + x + x2 + x6 + + xn! . . . .
2
4
x2n
b) cos x = 1 x2 + x24 + (1)n (2n)!
....
c) sin x = x

x3
6

x5
120

2n+1

x
+ (1)n (2n+1)!
....
p
p
3) Prove Theorem 44 if limn n |an | = and if limn n |an | = 0.
an
an
| = and if limn | an+1
| = 0.
4) Prove Theorem 45 if limn | an+1
P
2

1
n
5) a) Apply Theorem 46 to f (x) = 1x = n=0 x to show that ln(1 x) = x x2
n
x3
xn . . . .
3
2
3
n
b) Prove the relation ln(1 + x) = x x2 + x3 + (1)n xn . . . . For which x does this
hold?
c) We have seen in Exercise 5 of Section 16 that ln 2 = 1 21 + 13 14 + . . . . Why is this
P
1 1 n
not such a good way of computing ln 2 in practice? Show that ln 2 =
n=1 n ( 2 ) , and
explain why this is a more useful series to compute ln 2 in practice.
6) a) Show that
x3 x5
x2n+1
arctan x = x
+
+ (1)n
....
3
5
2n + 1
d
1
(Hint: Use that dx
arctan x = 1+x
2 .)
b) Since arctan 1 = /4, we could conclude that
1 1 1 1
= 4(1 + + . . . ).
3 5 7 9

61

Check this formula. Is it justified by Theorem 46?12 Why (not)?

c) Use the relation 4 = arctan 12 + arctan 13 to show that

4
=
(1)
2n + 1
n=1


1 2n+1
1 2n+1
( )
+( )
.
2
3

d) Cleverer still than the formula in part c) is Machins13 formula:

= 16 arctan

1
1
4 arctan
.
5
239

Indicate the infinite series that results from this formula. Why is this formula better to
approximate the digits of than the formulas in b) and c)?

12
13

Abels Theorem (which is not covered in these notes) does justify the formula.
John Machin (1680-1752) taught at Cambridge.

62

18
18.1

Answers to Selected Exercises:

Answers to Exercises of Section 2: Types of Numbers

1) Hint for 3 + 5
/ Q: Suppose by contradiction that 3 + 5 =
squaring both sides, we get

p
q

Q. Then

p2
1 p2
p2 8q 2
3 + 2 15 + 5 = 2 , hence 15 = ( 2 8) =
Q.
q
2 q
2q 2

/ Q, we have the contradiction we need.

Therefore, if we can show that 15
2) There is no natural next largest number to any x Q.
fits the bill. Verify why that is true!
3) The number x = a + ba
2
p
4) Let x = q and y = m
be rational, so p, q, m, n Z and q 6= 0 and n 6= 0. Then
n
p
pn+qm
m
x + y = q + n = qn Q, because pn + qm, qn Z and qn 6= 0. Similarly xy = pm
Q,
qn
because pm, qn Z and qn 6= 0.
7) If x = pq Q, then x satisfies qx p = 0 with p, q Z. Therefore x is algebraic.
p

8) Hint: argue by contradcition: suppose ln 2 = pq Q. take e-power: 2 = e q which can

be re-written to 2q = ep . Why does this give a contradiction?

18.2

Answers to Exercises of Section 3: Sets and Set Notation

1) The set {x} is a closed interval; it is often called a degenerate interval, because it
consists of only
one point.

4) b) (, 7] [ 7, ).
S
d) . . . (, 0) (, 2) (3, 4) = nZ ((2n + 1), 2n).
1
5) b) (1, 1), d) R \ {3, 3}, but if you simplify the function to 3x
, then the domain is
R \ {3}. So this simplification does actually change the function (!).

7) Write {1, 2, 3, 4} = [1, 4] N = (0, 5) Z = {n Z : 0 < n 4} = {n N : n 2} =

4n=1 {n} and there are many more.
S
9) a) nZ (n, n + 1) = R \ Z b) nN ( n1 , n1 ) = {0} c) nN ( n1 , n2 ) = .
S
10) a) Since U is the union of open intervals, say (ak , bk ), we can write U = k (ak , bk ).
If x U , there must be at least one k such that x (ak , bk ). Therefore ak < x < bk , and
both bk x and x ak are positive. Take = min(bk x, x ak ). Then (x , x + )
(ak , bk ) U .

18.3

Answers to Exercises of Section 4: Propositional Logic and

Quantifiers

1) a) is a tautology but b) is not because:

63

P
T
T
F
F

Q
T
F
T
F

a) (P Q)
F
F
F
T

T
T
T
T

(P Q) b)
F
F
F
T

(P XOR Q)
F
T
T
F

T
T
F
T

(P Q)
F
T
F
F

2) P XOR Q is equivalent to P Q and also to (P Q) (Q P )

3) P is equivalent to P kP . P Q is equivalent to (P kP )k(QkQ). P Q is equivalent
to (P kQ)k(P kQ).
4) b) n N \ {1} and m N \ {1, n} m 6 | n.
5) b) The negation is: > 0 > 0 such that . The statement is true (and the
negation is false) for real numbers.
d) The negation is: > 0 > 0 x R such that |x 3| < but |x2 9| . This
negation is true. For example, given > 0, take = and x = 3 + /2. Then
|x2 9| = |(3 + /2)2 9| = |(9 + 3 + 2 /4) 9| = 3 + 2 /4 > = .

18.4

Answers to Exercises of Section 5: Cardinality

1a) If B = , then #(B) = 0 and #(A) #(B) automatically. If B 6= and A B,

then we can construct a surjection f : A B by picking an element b B and setting
f (x) = x if x B and f (x) = b if x
/ B.
2 b) f (n) = 2n.
1
d) f (n) = n1
if n 2 and f (1) = 0.
e) f (x) = x if x
/ { 21 , 13 , 14 , 51 , . . . } and f ( n1 ) =
f) f (x) = tan((x 21 )).

1
n+1

if x =

1
n

{ 12 , 13 , 14 , 51 , . . . }.

3 Q, the prime numbers and Q [0, 1] have the same cardinality 0 . [0, ) and R \ Q have
the same cardinality c. [3, 3] Z and {n : 2n < 16} have the same cardinality 7.
5b) Fix centre C. Assume by contradiction that for every R > 0, the circle centred at C
and radius R contains a rational point. Call this point P (R). For different values of R
we have different points P (R). Therefore P : [0, ) {rational points in R2 } = Q2 is an
injection. This would imply that #([0, )) #(Q2 ). But there are uncountably many
values of R and only countably many points in Q2 . So this is a contradiction. Hence, there
must be values of R so that the circle centred at C and radius R contains no rational
point.

18.5

Answers to Exercises of Section 6: The Axioms of Ordered

Fields

1) Use the axioms to show that

a) There is exactly one neutral element of multiplication.
By axiom (A7), there is at least one. Suppose that both 1 and 1 are neutral elements of
multiplication. Then 1 = 1 1 =by A6 1 1 = 1, so they are the same.
64

b) x x 0 = 0.
x 0 =by A3 x 0 + 0 =by A4 x 0 + (x 0 + (x 0))
=by A1 (x 0 + x 0) + (x 0)) =by A9 (x (0 + 0) + (x 0))
=by A3 x 0 + (x 0) =by A4 0
d) If 2 is the notation for 1 + 1, then x + x = 2x.
2x = (1 + 1)x =by A9 = 1 x + 1 x =by A6 = x 1 + x 1 =by A7 = x + x.
g) (1) (1) = 1.
(1) (1) =by A3 (1) (1) + 0 =by A4 (1) (1) + (1 + (1))
=by A2 (1) (1) + ((1) + 1) =by A1 ((1) (1) + (1)) + 1
=by A7 ((1) (1) + (1) 1) + 1 =by A9 (1) ((1) + 1) + 1
=by A2 (1) (1 + (1)) + 1 =by A4 (1) 0) + 1
=by b) 0 + 1 =by A2 1 + 0 =by A3 1.
h) If x > 0 then x < 0.
Start with 0 < x, add x on both sides to get (unsing A12) 0 + (x) < x + (x). By A2
and A4, this becomes x + 0 < 0. Finally, by A3, x < 0.
i) 0 < 1.
Assume by contradiction that 1 < 0, then by adding 1 on both sides (using A12) also
0 =by A4 1 + (1) <by assumption and A12 0 + (1) =by A2 (1) + 0 =by A4 1
Then
0 =by b) (1) 0 =by A2 0 (1) <by A13 (1) (1) =by g) 1
So both 0 < 1 and 1 < 0, and this contradicts A10.
n) Show that if z < 0 and x < y, then y z < x z.
This goes in several steps:
Step 1) (x) = 1 x.
0 =by b x 0 =by A4 x (1 + (1)) =by A9 x 1 + x (1) =by A7 and A2 x + (1) x.
Now add x at the left of both sides:
(x) + 0 = (x) + (x + (1) x) =by A1 ((x) + x) + (1) x =by A4 0 + (1) x =by A2 and A3
LHS is (x) + 0 =by A3 = (x) and RHS is (x) + (x + (1) x) =by A1 ((x) + x) + (1)
x =by A4 0 + (1) x =by A2 and A3 (1) x. Therefore (x) = (1) x.
Step 2) z < 0 implies (z) > 0.
0 =by A4 z + (z) =by A11 0 + (z) =by A2 and A3 (z).
65

Step 3) x < y implies (y) < (x).

Add (x) + (y) on both of x < y sides using A11:
x + ((x) + (y)) < y + ((x) + (y))
Now use A1 and A2 repeatedly to work this back to
(y) + (x + (x)) < (x) + (y + (y))
Finally use A4 and A3 on both sides to get (x) < (y).
Step 4) (a) = a for all a R.
a =by A3 a + 0 =by A4 a + (a + (a)) =by A1 (a + (a)) + (a)
=by A4 0 + (a) =by A2 (a) + 0 =by A3 (a).
Now we can do the last bit. z < 0, so by Step 2, (z) > 0. Therefore x < y together
with A12 give
x (z) < y (z)
Step 1 then gives x (1) z < y (1) y and by A6 used repeatedly,
(1) (xz) < (1) (yz)
Step 1 once more (applied to both sides) gives
(xz) < (yz).
Apply Step 3 to get
(yz) < (xz)
Finally, Step 4 gives yz < xz.

18.6

Answers to Exercises of Section 7: Inequalities

1) a) neither, b) strictly increasing, c) strictly increasing, d) neither! e) strictly decreasing,

f) increasing (but not strictly)
2) a) increasing, b) you cant tell, c) increasing)
5) a) <, b) >, c) >, d) <, e) <, f) >, g) <
1
e
6) a) N = d 1

d)

N+ N

18.7

<

<

N
+ N
N

N + 1 N > ( 1 1)2 , therefore e.g. N = d 12 e suffices.

Answers to Exercises of Section 8: The Principle of Natural

Induction

1e) For n = 1, en = e > 1 = n.

(IH) en > n for some n N.
66

(IS) First note that n+1

= 1 + n1 2 for all n N. Then en+1 = e en >IH e n > 2 n
n
n+1
n = n + 1.
n
By the Principle of Natural Induction, en > n for all n N.
f ) For n = 7, we find 3n = 2187 < 5040 = 7!.
(IH) 3n < n! for some n N.
(IS) 3n+1 = 3 3n <IH 3 n! < (n + 1) n! = (n + 1)!.
By the Principle of Natural Induction, 3n < n! for all integers n 7.
g) Fix x (1, ). For n = 1, we find (1 + x)n = 1 + x = (1 + nx).
(IH) (1 + x)n (1 + nx) for some n N.
(IS) Because x > 1, 1 + x is positive. Therefore (1 + x)n+1 = (1 + x) (1 + x)n IH
(1 + x) (1 + nx) = (1 + (n + 1)x + nx2 ) (1 + (n + 1)x).
By the Principle of Natural Induction, (1 + x)n (1 + nx) for all n N.

18.8

Answers to Exercises of Section 9: Upper and Lower Bounds,

Infimum and Supremum

1) If a = min A exists, then a a0 for all a0 A by definition of min. Therefore a is a

lower bound of A. It must also be the smallest lower bound, because any largest number
is larger than a and a A. Therefore a = inf A.
3)
a)
b)
c)
d)
e)

max A
1
6
6
6
6

min A
6
6
2
3

6
6

sup A
1
2
7
6
1

inf A
0
0
2
3

6
1

4)
max
a) 6
b) 1
c) 6

min
0
0
6

sup
1
1
2

inf
0
0
0

5a) False. Take e.g. f (x) = sin2 (x and g(x) = cos2 9x). Then f (x) + g(x) = 1 so
sup{f (x) + g(x) : x R} = inf{f (x) + g(x) : x R} = 1, but sup{f (x) : x R} +
sup{g(x) : x R} = 1 + 1 = 2.
b) This is false in general; only if f is a consant function, then sup{f (x) : x R} =
inf{f (x) : x R}.
c) This is true. If you make the set over which you take the supremum larger (here the
f -image of [1, 1] versus the f -image of R), then the supremum can only become larger.
d) False. For example, try f (x) = arctan(x).

67

18.9

Answers to Exercises of Section 10: Axiom of Completeness

2) If A is bounded below (say K is a lower bound), then A = {a : a A} is

bounded above (with K as upper bound). By the Axiom of Completeness, A contains
a supremum, say M . Then M A and M = inf A.
5) (i) true, (ii) false, (iii) true, (iv) true (but since not every set that is bounded from
below has a infimum in nZ (n 31 , n], there are still Cauchy sequences without limit in
this case!)

18.10

Answers to Exercises of Section 11: Sequences

You can check limits of sequences by the Maple command:

limit(an , n = infinity)
For example, typing (1 + 1/n)n when an = (1 + n1 )n gives the answer e. Note that
Maple uses the convention n = in this command. Dont do this yourself on written
coursework etc.: Write an instead!
2) b) limn an = 0, d) limn an = 14 , f) limn an = 21
n

1
3) 2e < 1, so 2en becomes small as n . So for n sufficiently large, 2en < 100
, and
1 n
n
n
multiplying this with e , we have 2 < 100 e . In fact, N = 16 is the first integer for which
1 N
e .
2N < 100

18.11

Answers to Exercises of Section 12: Limits and Convergence

1) b) limn an = 0. Proof:
Let > 0 be arbitrary.
Take N = d 1 e + 1, so N1 < .
Let n N be arbitrary.
Then |an 0| = n2n+1 < nn2 =

1
n

d) limn an = 0. Proof:
Let > 0 be arbitrary.
e + 1, so 2
< .
Take N = d 2

N
Let n N be arbitrary.
Then |an 0| = sin 2
by the hint
n
f) limn an = 1. Proof:
Let > 0 be arbitrary.
Take N = d 1 e + 1, so N >
1 + > 21/N .

1
N

2
n

< .

2
N

< .

and (see the hint) (1 + )N 1 + N > 2, and therefore

68

Let n N be arbitrary.
Then |an 1| = |21/n 1| 21/N 1 < 1 + 1 = .
2) Assume by contradiction that (an )nN , has at least two limits, say A and B. Take
= |B A|/3 > 0.
Using the definition of an A, we can find NA such that for all n NA , |an A| < .
Using the definition of an B, we can find NB such that for all n NA , |an B| < .
Take n max{NA ., NB }.
Then by the triangle inequality: |A B| = |A an + an B| |A an | + |an B| <
+ = 2 = 32 |A B|.
This is impossible for |A B| > 0. This contradiction proves that A = B.

18.12

Answers to Exercises of Section 13: Convergence Properties of Sequences

1) To prove the Sandwich Rule, start with > 0 arbitrary.

Since an L and bn L, there exist Na and Nb N such that
(a) if n Na , then |an L| <
(b) and if n Nb , then |bn L| < .
Take N = max{Na , Nb }.
Let n N be arbitrary.
Recall that an cn bn . There are now two cases:
Case 1: If L cn , then |cn L| |bn L| < .
Case 2: If L > cn , then |cn L| |an L| < .
So in either case |cn L| < . This completes the verification of the definition of cn L.
2

2n
= n22 . Because n22 0 as n , also nn+1
0.
6) a) 0 nn+1
4
4
n4
2n
2n

b) n+ n n = 2 works for the upper bound. The lower bound is trickier. One way is
splitting off 2 first:

2n
2n + 2 n 2 n
2 n
2 n
2
=

2
=2
= 2 2 as n .
2n
n+ n
n+ n
n+ n
n
So

2n

n+ n

2 as n .

n
0 because 2e < 1. By the Sandwich Rule
7) a) 0 n2 en 102n en 10 2e
n2 en 0 as well.
d) Hint: Try subsequences a24k and a24k+12 . If you can show that along these subsequences,
you get different limits, then the whole sequence (an )nN cannot converge.

10) a) Take a1 = 1 and continue recursively: an+1 = 1 + an . So a2 = 1 + 11 = 2,

a3 = 1 + a12 = 1 + 12 , a4 = 1 + a13 = 1 + 23 , etc.
b) See Figure 10, left panel.
c) We have a1 < a3 < a4 < a2 as computed in a). Assume by induction that an < an+2 <
an+3 < an+1 for some odd n N.
69

2.5

2.5
2

1.5

1.5

0.5

0.5

0.6

0.8

1.2

1.4

1.6

1.8

2.2

2.4

0.6

0.8

1.2

1.4

1.6

1.8

2.2

2.4

Figure 10: Graphs of 1 +

1
x

(left) and

Then
an+4 = 1 +
an+5 = 1 +

1
1+

1+

<1+

1
an+3

1
1+

>1+

an+2

and also
an+5 = 1 +

1 + x (right) together with the diagonal x = y.

>1+

1
an+3

1
= an+2 ,
1 + a1n
1

1+

1
an+1

1
1+

= an+3 ,

= an+4 .

an+2

Therefore an+2 < an+4 < an+5 < an+3 , and by the Principle of Induction, an < an+2 <
an+3 < an+1 for all odd n N.
d) Since (a2k1 )kN is increasing and bounded above by 2, (a2k1 )kN . Let odd be the
limit.
Since (a2k )kN is increasing and bounded below by 1, (a2k )kN . Let even be the limit.
Because a2k1 < a2k for every k N, we also have odd even .
Both odd and even should satisfy
1+
Solve this equation, we find = 12 (1

= odd = even = 12 (1 + 5).

1
1+

= .

5). But the sign gives an answer < 1, so

11) This time we take a1 = 1 and an+1 = 1 + an . The graph is in Figure 10, right
panel. The sequence (an )nN is increasing (use induction) and bounded by2 (use induc
tion again!), so there is a limit, say Solve 1 + = to find = 21 (1 + 5).

18.13

Answers to Exercises of Section 14: Cauchy Sequences

1) Take an = n1 , then n N an Q, and also limn an = 0 Q. For the second part,

take bn = n , then n N bn R \ Q, but limn bn = 0 Q.
3) Assume that (an )nN is a Cauchy sequence in Z. Let = 21 .
By definition, there is N N such that for all m, n N , |am an | < = 12 .
70

Take m = N . We know aN Z.
If an 6= aN , and at the same time an Z, |an aN | 1, contrary to the choice of N .
Therefore n N an = aN , so the sequence is eventually constant.
4) See Exercise 5 (iv) of Section 10

18.14

Answers to Exercises of Section 15: Series

You can check sums of series by the Maple command:

sum(an , n = 1 . . infinity)
2

For example, typing 1/n2 when an = 1/n2 gives the answer 6 . When you try this for
1/n3 (i.e., an = 1/n3 ), then Maple gives the answer (3). This is because Maple has
P
1
a special function, called Riemanns -function14 , which is defined as (s) =
n=1 ns
for every s > 1. There is no nice surd form for (3). If you want a numerical answer
nonetheless, type
evalf (sum(1/n3 , n = 1 . . infinity))
and Maple will give you 1.202056903.
5
= 110.1001binary .
4a) 7 = 111binary , 0.875 = 0.111binary and 6 16
1
b) The binary expansion of 3 is 0.010101010101 . . .
c) Prove from the definitions and Theorem 35 that As limit of partial sums,

0.1111111 . . .binary

k
1
1
X
2k+1
1
1
2
= lim
=
lim
= lim 1 k = 1.
1
i
k
k 1
k
2
2
2
i=1

4) P
P
9
n
n+1
a)
= 23 ; b)
= 29
;
n=0 (2)
n=0 (0.45)
P 1+2n
P 1 2n 81
e) n=0 3n = 29 .
d) n=0 ( 9 ) 80 ;

18.15

c)
f)

1 n
n=10 ( 4 )

n=1

1
;
349

(1)5n +e3n
21n

21 e3 1
.
22 e3 21

Answers to Exercises of Section 16: Convergence Criteria

for Series

2) Find out if the following series are convergent:

P
n2 +1
converges. (Majorise the terms by n22 .)
a)
n4
Pn=1

n
b) n=1 2n+3 diverges. The terms 6 0.
P
2
c)
is convergent. (Use ratio test.)
n!
Pn=0

1
= 1. (Use telescoping series.)
d) n=1 n(n+1)
P
n
4) a) n=0 (1)n n+2
diverges. The terms 6 0.
P 1
n
b) n=1 n cos 2 is convergent, but not absolutely convergent. (Note that cos n
takes
2
14

Bernhard Riemann, 1826 - 1866, German mathematician who did groundbreaking work in number
theory, complex analysis and in geometry.

71

values 0, 1, 0,1, 0, 1, 0, 1, 0, . . . , so we have an alternating series here.)

P
n n
c)
n=0 (1) n+7 is convergent, but not absolutely convergent.
6) a)
1 1 1
1
1
+ +

2 3 4
2k 1 2k
2k
k
k
X1 X 1 X 1
=

i
2i i=1 2i
i=1
i=1

S2k = 1

2k
X
1
i=1

k
X
1
i=1

= ln 2k + + h2k (ln k + + hk )
= ln 2 + ln k + h2k ln k hk
= ln 2 + h2k hk ln 2

18.16

as

k .

Answers to Exercises of Section 17: Power Series

1 a) R = 51 , region of convergence is [ 15 , 51 ].
c) R = 1, region of convergence is (1, 1): for x = 1, the terms sin n and (1)n sin n
dont converge to 0, so at the boundary points x = , the series diverges.


(n+1)n+1 n!
|
n+1 n
=
=
= (1 + n1 )n e.
e) R = 1/e. Use the ration test: |a|an+1
n
n (n+1)!
n
n|
2) All the power series in this question converge on the entire R.
p
3) The proper statement is: If n |an | as n , then the radius of convergence is
0, and the region of convergence is {0}.
p
p
To prove this, assume x 6= 0, and call bn = an xn . Then n |bn | = |x| n |an | ,
so the series diverges by Theorem 38. Since this is true for every x 6= 0, the radius of
convergence is T = 0. However, if x = 0, every term but the first (with x0 ) is 0, so the
series converges. Hence the region ofpconvergence is {0}.
As for the other case, namely if n |an | 0 as n , then the region of convergence
is R.
p
p
To prove this, take x R arbitrary and call bn = an xn . Then n |bn | = |x| n |an |
0 < 1, so the series converges by Theorem 38. Since this is true for every x R, the
region of convergence is R.

72

Z
ln(1 x) =
0

xn is 1. For |x| < 1, we get

!
Z x X

1
dt =
tn dt
1t
0
n=0
Z

X x
=
tn dt
by Theorem 46

5) a) The radius of convergence for

n=0

n=0

X
n=0

1
xn+1
n+1

1
1
1
= x x2 x3 x4 . . .
2
3
4
b) Use the above with x replaced by x:
1
1
1
ln(1 + x) = ln(1 (x)) = (x) (x)2 (x)3 (x)4 . . .
2
3
4
1 2 1 3 1 4
= x x + x x + ...
2
3
4
c) For x = 1, this suggests that ln(2) = ln(1 + 1) = 1 12 + 13 14 + . . . , which is true, see
Exercise 6a of Section 16. However, Theorem 46 does not cover the case x = R, so you
cannot use parts a) and b) of this exercise to proof it. (The missing step is supplied by
Abels Theorem, see [1, Theorem 7.34] in this case.)

19

Greek Letters

The roman letters are just too few for mathematics, so Greek letters are used. Here is
the full alphabet 15 .
lower case upper case name

lower case upper case name

A
B

E
Z
H

I
K

alpha
beta
gamma
delta
epsilon
zeta
eta
theta
iota
kappa
lambda
mu

N
X
O

nu
xi
omicron
pi
rho
sigma
tau
upsilon
phi
chi
psi
omega

Some Greek letters have standard usage in mathematics or physics; for example:
15

I dont know why the book [1] puts them in the wrong order.

73

 = 3.1415927 . . . Jones (1675-1749) was the first to use for the o (=

circumference) of a circle of diameter 1.
and for small positive numbers.
P
1
2
for a sum, for example
k=1 k2 = 6 .
Q
for a product, for example nk=1 k = n!

References
[1] J. Howie, Real Analysis, Springer 2003, ISBN 1-85233-314-6
[2] I. Stewart, Galois Theory, Chapman & Hall 2004, ISBN 1-58488-393-6
An excellent source on the history of mathematics and its key players can be found on
http://www-groups.dcs.st-and.ac.uk/history/ of the School of Mathematical and
Computational Sciences University of St Andrews. For more history, including the origin
of mathemical words and notation, see http://members.aol.com/jeff570/mathword.
html.

74

20
20.1

Summaries and Extra Exercises.

Section 1: Introduction

Read the Introduction once more, to familiarise yourself with differences between definitions, axioms, theorems, etc.

20.2

Section 2: Types of Numbers

To know by heart:
1. What are rational, irrational and real numbers.
2. What are algebraic and transcendental numbers (including examples and the definition).
3. Structure of proof by contradiction.

4. The proof of 2
/ Q.
Test exercises:

1. For which n Z is n Q?
/
Q.
2. Prove that 6
3. Prove that 2 + 3
/ Q.
4. Give the definition of an algebraic number.

5. If x is algebraic, show that x is also algebraic.

Answers to test exercises:

1. n Q for n = 0, 1, 4, 9, 16 and all the other squares. Otherwise,

n
/ Q.

/ Q.
2. Proof by contradiction, basicallythe same
as the proof
for
2
3. Assume by contradiction that 2 + 3 Q, so 2 + 3 = pq for some p Z and

2
2
q N. Square the equation to find 2 + 2 2 3 + 3 = pq2 which simplies to 6 = p 2q10
2 .

a
2
2
This means that 6 = b for a = p 10 Z and b = 2q N. Now use exercise 2.
4. x is algebraic if there are d 1 and a0 , . . . , ad Z such that ad xd + + a1 x + a0 = 0.
5. Since x is algebraic, there are d 1 and a0 , . . . , ad Z such that ad xd + +a1 x+a0 = 0.

But then also ad ( x)2d + + a1 ( x)2 + a0 = 0. In other words, if we take a2i = ai for

i = 0, . . . , d and a
2i1 = 0 for i = 1, . . . , d, then a
2d y 2d + + a
1 y + a
0 = 0 for y = x.
Hence y satisfies the definition of an algebraic number.

20.3

Section 3: Sets and Set Notation

To know by heart:
1. Interval notation [x, y] (closed), [x, y) (half-open), (x, y) (open), etc.
2. When the symbols are used in intervals, they always get a round bracket; e.g.
[1, ).
75

3. Subset , superset , is element of .

4. Intersection , union and set difference \.
5. The definition of dense. Main example: Q is dense in R.
Test exercises:
1. Write the following sets in interval notation. Are they open, closed or neither?
(i) {x R : 0 < x 5}, (ii) [6, 8] \ [7, 9], (iii) [6, 8] [7, 9].
2. Write the following sets in interval notation. Are they open, closed or neither (or both)?
(i) nN ( n1 , n1 ), (ii) nN ( n1 , n1 ), (iii) nN (0, n1 ), (iv) nN [0, n1 ), (v) nN [n, n)
3. Show that Q is dense in R.
4. Show that the transcendental numbers are dense in R.
Answer to test exercises:
1. (i) {x R : 0 < x 5} = (0, 5] (neither open nor closed), (ii) [6, 8] \ [7, 9] = [6, 7)
(neither open nor closed), [6, 8] [7, 9] = [7, 8] (closed).
2. (i) nN ( n1 , n1 ) = {0} (closed)
(ii) nN ( n1 , n1 ) = (1, 1) (open)
(iii) nN (0, n1 ) = , (both open and closed, this is by convention)
(iv) nN [0, n) = [0, ) (closed, this is a bit confusing maybe, but the part has no
effect on whether an interval is open or closed, and since the boundary point 0 belongs
to the interval, it is closed)
(v) nN [n, n) = (, ) (both open and closed)
3. First look up the definition of what it means that Q is dense in R.
Let x, y R be arbitrary such that x < y. Take q N so large that 1q < 21 (y x). Then
the points . . . , 3
, 2
, 1
, 0, 1q , 2q , 3q , . . . are all rational and 1q apart. Since 1q < 12 (y x),
q
q
q
there is p Z such that x < pq < y. This completes the proof.
4. Let x, y R be arbitrary such that x < y. Take q N so large that 1q < 31 (y x). Then
the points . . . , 3
, 2
, 1
, 0, 1q , 2q , 3q , . . . are all rational and 1q apart. Since 1q < 12 (y x),
q
q
q
< y. Now take z = p+/4
, then x < pq < z < p+1
< y,
there is p Z such that x < pq < p+1
q
q
q
and also z is transcendental. (If z was not transcendental, then 4(qz p) = would not
be transcendental either, but is transcendental.) So we have found a transcendental
number z between x and y. This finishes the proof.

20.4

Section 4: Propositional Logic and Quantifiers

To know by heart:
1. (not), (or) and (and).
2. and in words, and their truth tables.
3. and in words.
4. The order of quantifiers is important: x y is different from y x. In the former
y can depend on x, in the latter not.
76

5. Standard phrases in quantifiers:

For all sufficiently large n: m N n m.
For all sufficiently small positive : 0 > 0 (0, 0 ).
6. Negating statements with quantifiers: replace with and vice versa. Then negate
the final statement.
Test exercises:
1. Work out the truth table for the expression (P Q) (P Q).
2. Express in quantifiers:

a) n > 100 for all sufficiently large integers.

b) The square root of every rational number is real.
c) Every square is positive.
d) 0 < 10x2 < x for every sufficiently small positive real number.
e) Q is dense in R.
3. Negate the expressions in 1. Which is true, the statement or its negation?
4. Express in plain English:
(i) n Z m = 2n. (ii) A R B R such that A B = and A B = R.
Answers to Test exercises:
1. P Q (P Q) (P Q)
T T T F
F
T
F F F
T F T T
T
T
F T T
F T F T
F
T
T T F
F F F T
T
T
T T T
2. Expressed in quantifiers:

a) m n m n > 100.

b) x Q x R.
c) x R x2 > 0.
d) x0 > 0 x (0, x0 ) 0 < 10x2 < x
e) x, y R with x < y z Q z [x, y].

3. a) Negation: m n m n 100. The statement itself is true, for example, we can

take m = 10001.

b) Negation: x Q x
/ R. The negation is true; for example x = 1 has no real
square root.
c) Negation: x R x2 0. The negation is true; for x = 0 we have x2 = 0
d) Negation: x0 > 0 x (0, x0 ) 0 10x2 or 10x2 x. The statement is true, for
example we can take x0 = 1/10.
e) Negation: x, y R with x < y z Q z
/ [x, z]. The statement is true, see previous
section.
4. (i) There is an integer such that m is twice as much. In other words: m is even.
(ii) For every subset of the reals, there is another subset that is disjoint from the first
such that their union is R. In other words: Every set has a complement.
77

20.5

Section 5: Cardinality

To know by heart:
1. What is injective (i.e., one-to-one), surjective (i.e., onto) and bijective (i.e.,
both one-to-one and onto). A bijection is also called a one-to-one correspondence.
2. Definition: A and B have the same cardinality if there is a bijection between A
and B.
3. For finite sets A, the cardinality #(A) equal the number of elements.
4. Infinite sets A can be countable, denoted #(A) = 0 , or uncountable.
5. The sets N, Z and Q are countable; R is uncountable, its cardinality denoted #(R) =
c.
Test exercises:
1. What is the cardinality of [0, ), of the primes, and of R \ Q.
2. What is the cardinality of N N? (Hint: the argument is similar to the proof of the
cardinality of Q.)
3. Give one-to-one correspondence between the integers and the positive cubes. Give
one-to-one correspondence between R and (0, ).
4. Are there real numbers x such that x 6= tan n2 for every n Z?
Answer to test exercises:
1. #([0, )) = c, #({the primes}) = 0 , #(R \ Q) = c.
2. #(N N) = 0 . The reason is that you can put all the points {(m, n) : m, n N}
in a square grid, and then draw a zig-zag path through all of them (in the same way as
the proof of why Q is countable). This gives a denumeration of {(m, n) : m, n N}, so it
must be countable.
3. n (2n)3 if n > 0 and n 7 (2|n| + 1)3 if n 0 gives a one-to-one correspondence
between the integers and the positive cubes. The map x 7 ex gives one-to-one correspondence between R and (0, ).
4. The set {tan n2 : n Z} is at most countable, and therefore contains too few points
to cover the uncountable set R. Therefore x R n Z x 6= tan n2 .

78

20.6

Section 6: Axioms of Ordered Fields

A field is closed

Addition
x, y R x + y R

Multiplication
x, y R xy R

Associative law

x, y, z R (x + y) + z = x + (y + z)

x, y, z R (xy)z = x(yz)

Commutative law

x, y R x + y = y + x

x, y, z R xy = yx

Neutral element

n1 R x R x + n1 = x

n2 R n2 6= n1 x R xn2 = x

Negative and
reciprocal

x R (x) R x + (x) = n1

x R x 6= n1 x1 R x

Distributive law

x, y, z R x(y + z) = xy + xz

Order is complete

x, y R either x < y or x > y or x = y

Transitivity

x, y, z R x < y and y < z implies x < z

x, y, z R
x < y implies x + z < y + z

1
x

= n2

x, y, z R with z > 0
x < y implies xz < yz

To know by heart:
1. What the associative law, commutative law, distributive law mean.
2. What neutral elements are. (In the table above n1 = 0, n2 = 1.)
3. Proof of uniqueness of neutral elements.
4. What is meant by an ordering <
Test exercises:
1. Is subtraction commutative? Is it associative?
2. Which of the following are ordered fields. If not, why?
(i) : Z, (ii) : Q, (iii) : R, (iv) : { 2pn : p Z, n N.}
3. Show that 1 is the only neutral element for multiplication.
4. Formulate the associative, commutative and distributive law for taking unions and
intersections.
5. Are the complex numbers C an ordered field?
Answer to test exercises:
1. 2 1 = 1 6= 1 = 1 2, so subtraction is not commutative.
(1 2) 3 = 4 6= 2 = 1 (2 3), so subtraction is not associative.
2. Which of the following are ordered fields. If not, why?
(i) Z is not an ordered field: Not every n Z has n1 Z, so axiom (A8) fails.
(ii) Q is an ordered field.
(iii) R is an ordered field.
(iv) The dyadic rationals { 2pn : p Z, n N.} is not an ordered field. For example,
1
4
3 = 3 doe snot belong to this set.
4
3. By Axiom (A7) there is at least one neutral element 1. Assume by contradiction that 1
was another neutral element of multiplication. Then we can compute 1 1 in two different
79

ways:
1 1 = 1 by Axiom (A7) applied to 1, and
1 1 = 1 1( by Axiom (A6) ) = 1( by Axiom (A3) applied to 1)
Therefore 1 = 1.
4. For all A, B, C R we have:
(A B) C = A (B C) and (A B) C = A (B C) (associative laws),
A B = B A and A B = B A (commutative laws),
(A B) C = (A C) (B C) and (A B) C = (A C) (B C) (distributive laws).
5. The complex numbers C an not ordered field, because there is no way you can define
a proper order < on it.

20.7

Section 7: Inequalities

To know by heart:

|a + b| |a| + |b|

|x y| |x z| + |z y|
The triangle inequality, in various forms:
|a b| ||a| |b||

|x y| ||x z| |z y||

increasing x < y f (x) f (y), or
Monotone means
decreasing x < y f (x) f (y).

For strictly monotone, replace the and by < and >.

if x > 0;
x
0
if x = 0;
Absolute value |x| =

x if x < 0.
The floor function x 7 bxc (rounding downwards to nearest integer).
Test exercises:
1. Assume that 0 < x < y. Which is larger:

(a) tan ex or tan ey , (b) e x2 +1 or e x2 +1 , (c) bln xc or bln yc.

2. Fix > 0 and M N. What is the least N N such that
2
1
n < ,
< ,
(b) n N sin
(c) n N en 1 > M .
(a) n N n+57
n
Answer to test exercises:
1
1
1. (a) tan ex or tan ey . (b) e x2 +1 < e x2 +1 .
(c) bln xc < bln yc. Since the floor function is not strictly increasing, you cannot be sure
that bln xc < bln yc; possibly bln xc = bln yc.
p
2. (a) N = b 1 c 56. (b) N = b 12 c + 1, (c) N = b ln(M ) + 1c + 1.

20.8

Section 8: The Principle of Natural Induction

To know by heart:

80

 The Principle of Natural Induction states that if

(i) a statement A(n0 ) is true for some n0 ;
(ii) if A(n) is true then A(n + 1) is true as well,
then A(n) is true for all n N, n n0 .
A proof by induction has four steps:
1. Initiation step: A(n) is true for n = n0 .
2. Induction hypothesis: A(n) is true for some n.
3. Induction step: Show that if A(n) is true for n = k, then A(n) is also true for
n = k + 1.
4. By the Principle of Natural Induction, A(n) is true for all n n0 .
Strong natural induction: Sometimes you might need a stronger induction hypothesis: A(k) is true for all k n.
Test exercises:
1. Prove by natural induction:
(a) xn y n is a multiple of (x y) for every even n N.
Hint: ln(1 + x) x for all x > 1.
(b) ln(n + 1) 1 + 21 + 13 + . . . n1 for all n N.
(c) Inequality of Bernoulli: For every a 1, we have (1 + a)n 1 + an for all n 0.
Answer to test exercises:
(a) Since this is only about even n, we should prove: x2m y 2m is a multiple of (x y)
for every m N.
For m = 1, we have x2 y 2 = (x + y)(x y) is a multiple of x y.
(IH) x2m y 2m is a multiple of (x y) for some m N.
(IS) x2(m+1) y 2(m+1) = x2m+2 y 2m+2 = (x2m y 2m )x2 + y 2m (x2 y 2 ) = (x2m y 2m )x2 +
y 2m (x + y)(x y). By (IH), (x2m y 2m ) is a multiple of x y, and therefore the whole
expression is a multiple of x y. This finishes the induction step.
By the Principle of Natural Induction, x y divides xn y n for all even n N.
(b) For n = 1, ln(1 + 1) = 0.693147... < 11 is true.
(IH): ln(n + 1) 1 + 12 + 13 + . . . n1 for some n 1.
(IS):
ln(n + 2) = ln(n + 1 + 1) = ln((n + 1)(1 +
= ln(n + 1) + ln(1 +

1
))
n+1

1
)
n+1

by (IH)

1 1
1
1
+ + . . . + ln(1 +
)
2 3
n
n+1
1 1
1
1
1 + + + ... +
,
2 3
n n+1
1+

where in the last step we used the inequality ln(1 + x) x. This finishes the induction
step.
81

By the Principle of Natural Induction, ln(n + 1) 1 + 12 + 13 + . . . n1 for all n N.

(c) Fix a 1.
For n = 0, we have (1 + a)0 = 1 = 1 + a 0.
(IH) (1 + a)n 1 + an for some n 0.
(IS) (1+a)n+1 = (1+a)n (1+a) by (IH) and because 1+a0 (1+an)(1+a) = 1+a(n+1)+a2 n
1 + a(n + 1). This finishes the induction step.
By the Principle of Natural Induction, (1 + a)n 1 + an for all n N.

20.9

Section 9: Upper and Lower Bounds, Supremum and Infimum

To know by heart:
An upper bound of A is any number U R such that a A a U . A lower
bound of A is any number L R such that a A a L. The set A is bounded
if it has an upper and lower bound.
The supremum sup(A) is the smallest upper bound, and the infimum inf(A) is
the largest lower bound of a set A. The supremum/infimum of A need not belong
to A.
The maximum max(A) is the largest element and The minimum min(A) is the
smallest element of A. So the maximum/minimum must belog to A.
If max(A) exists, then max(A) = sup(A). If min(A) exists, then min(A) = inf(A).
However, the existence of sup(A) or inf(A) does not guarantee that max(A) or
min(A) exist.
Test exercises:
1. Is an upper/lower bound of a set A unique? Are inf(A) and sup(A) unique? Justify
your answer.
2.
A
min A
max A
inf A
sup A
a)
[2, 6] (2, 2)

b)
{x Z : x > 9}
c)

{x R : ex cos 2x 1}

d)

{e2xx : x R}
2

Answer to test exercises:

1. If U R is an upper bound of A, then U + 1 is as well. In fact, there are infinitely
many upper bounds. Similar for lower bounds. However, if S = sup(A) R, then any
x > S is an upper bound, but not the smallest upper boud, and every y < S is not an
upper bound. Therefore, sup(A) is unique. Similar for inf(A).
2.
82

min A
6

max A
6

inf A
2

sup A
6

b)

A
[2, 6] (2, 2)

{x Z : x > 9}

82

82

c)

{x R : ex cos 2x 1}

a)

{e

2xx2

d)

20.10

: x R}

Section 10: Axiom of Completeness

To know by heart:
The Axiom of Completeness reads: Every nonempty subset of R that is bounded
above has a supremum in R.
The Axiom of Completeness is the crucial in distinguishing Q from R. It ensures
that the set R is rich enough to contain limits of sequences when they are supposed
to converge.
The Axiom of Completeness is used to prove the Property of Archimedes:
x R N N such that x < N .
In other words, there is no largest real number; is not a real number.
Test exercises:
1. For which of the following sets is the Axiom of Completeness true?
(i) [0, ), (ii) (2, 2], (iii) { n1 : n Z \ {0}}, (iv) the transcendental numbers.
Are they ordered fields? Why (not)?
Answer to test exercises:
1. (i) Axiom of Completeness holds, but [0, ) is not an ordered field, because e.g. 1
doesnt exists.
(ii) Axiom of Completeness holds, but (2, 2] is not an ordered field, because e.g. 11 = 3
3
doesnt belong to (2, 2].
(iii) Axiom of Completeness does not hold, because { n1 : n Z, n < 0} is bounded but has
no supremum: 0 is not transcendental. For the same reason, it is not an ordered field.
(iv) Axiom of Completeness does not hold, because {x transcendental : x < 0} is bounded
but has no supremum: 0 doesnt exist. For the same reason, it is not an ordered field.

20.11

Section 11: Sequences

To know by heart:
A sequence (an )nN is an ordered list of numbers, and n is called the index.

83

 For each fixed a > 0, and all n sufficiently large:

ln n

na

<

logarithmic

<

polynomial

en

<

exponential

Test Exercises:
1) Put in increasing order when n is very large:

3n
3
n5
nn
n(ln n)7
10n3 + 7

< nn < en ,

n!

factorial

n!

10

What happens for n very large to the numbers:

2
5
n
n
n
n
4
ln n
a) n2n120n
b) 23n+3
;
c) 2n2+10
d) 6n3n+12
9 +3n3 ;
n +2 ;
4 +n2
+n!

1
3) Stirlings formula states that: nn en 2n n! nn en 2n (1 + 4n
).
n n
Use this to decide if ( 3 ) is smaller or greater than n!.
Answers to Test
Exercises:

n
3
7
5
1) n(ln n) < n < 10 n < 10n33 +7 < n! < nn .
2
5
9
2) a) n2n120n
9 +3n3 0 as leading term n is in denominator.
n
n
b) 23n+3
1. Leading terms in numerator and denominator are the same.
+n!
2n +10n
c) n2n +2 . The leading term 10n is in the numerator.
4
4
2
2 +12 ln n
4
ln n
ln n
2, because we can set aside 6n3n+12
= 6n +2n3n2n
= 2+
d) 6n3n+12
4 +n2
4 +n2
4 +n2
n2 +12 ln n
2 + 0.
3n4 +n2

3) Since 3 > e, we have ( n3 )n < nn en < nn en 2n < n! by Stirlings formula.

20.12

Section 12: Limits and Convergence

To know by heart:
The -N definition of an L.
> 0 N N such that n > N, |an L| < .
and its proof template:
Take > 0 arbitrary.
Let N =
Take n > N arbitrary.
Then |an L| =

< . (Estimate from above: <)

The M -N definition of an .
M > 0 N N such that n > N, an > M.
and its proof template:
Take M > 0 arbitrary.
Let N =
Take n > N arbitrary.
Then an =

> M . (Estimate from below: >)

84

 Standard limits:

1
lim an =

diverges
lim n sin

lim

1 < a < 1,
a = 1,
a > 1,
a 1.

1
= 1.
n

p = 1 p > 0.

lim (1 +

for
for
for
for

a n
) = ea
n

a R.

Test Exercises:
1) What are the limits as n of:

n
n+7
3n

(a) n+
(b)
(c)
n2 + 10 (d) (xn + y n )1/n for fixed x, y > 0.
n!
n
2) Give -N -proofs (or M -N -proofs) for the convergence/divergence of the following sequences:
6n
(d) cos n.
(a) n23n+2 ,
(b) 2n7
,
(c) 5+nn ,
3) Compute the limits:
1
1
(a) limn (1 + n5 )3n (b) limn n n (c) limn (2n) n2 +1 .
Answers to Test Exercises:

n
n+7
= 1
1) (a) limn n+
(b) limn 3n! = 0 (c) limn n n2 + 10 = 1
n
(d) limn (xn + y n )1/n = max(x, y) for fixed x, y > 0.
2) (a) limn n23n+2 = 0. We prove:
Let > 0 be arbitrary.
Take N = b3/c + 1, so 1/3N < .
Let n N be arbitrary.
1
1
Then |an 0| = 3n2n+2 3n
3N
< .
6n
(b) limn 2n7 = 3. Proof:
Let > 0 be arbitrary.
c + 14, so 21
< and 2N 7 N .
Take N = b 21

N
Let n N be arbitrary.
6n
21
21
Then | 2n7
3| = | 6n(6n21)
| = | 2n7
| = 2n7
2N217 21
< .
2n7
N
n
(c) limn 5+ n = . Proof:
Let M > 0 be arbitrary.

Take N = b2M 2 c + 25, so 2N > M and 5 + N 2 N .

Let n N be arbitrary.
Then 5+nn 5+nN = 2N > M .
(d) The sequence cos n diverges. We prove this by contradiction, assuming that L =
limn cos n, and separating in the cases L 0 and L > 0.
L 0. Take = 12 cos 12 > 0 and let N N be arbitrary. There are infinitely many
pairs n, k N so that n [2k 21 , 2k + 21 ], simply because these intervals have
85

length 1, so they must contain an integer. Take such pair n, k with n N . Then
| cos n L| cos n = cos(n 2k) cos 12 > . This shows that no L 0 can be the
limit.
For L > 0 we apply the same argument, but now using that there are infinitely many
pairs n, k N so that n [2k + 12 , 2k + + 12 ].
3) (a) limn (1 + n5 )3n = e15

20.13

(b) limn n

1
n

=1

(c) limn (2n) n2 +1 = 1.

Section 13: Convergence Properties of Sequences

To know by heart:
Let (an )nN and (bn )nN be convergent sequences. Then
1. limn (an + bn ) = limn an + limn bn .
Only split the limit of sums if limn an and limn bn exist!
2. lim (an bn ) = lim an lim bn .
n

an
limn an
=
, provided limn bn 6= 0.
n bn
limn bn

3. lim

If an < bn for each n N, then limn an limn bn .

NB: you can lose the strict inequality in the limit! E.g. n N
limn n1 = limn n2 = 0.

1
n

<

2
n

but

Sandwich Rule If (an )nN and (bn )nN are convergent sequences with the same
limit L, and an cn bn , then (cn )nN is also convergent and limn cn = L.
Definitions of monotone, increasing and decreasing sequences.
Definition of subsequence.
Every convergent sequence is bounded.
Every bounded monotone sequence is convergent.
Every bounded sequence has a convergent subsequence. (Bolzano-Weierstrass).
Test Exercises:
1) What are the limits as n of the following sequences:

2
4 n3
) , (b) ( n + 2( n + 4 n))nN , (c) n 1 + 2 + 3 + + n.
(a) ( nn3 +n
n4 +n2 nN
R1
1
1
2) We 1 ex e n for x [0, n1 ]. Also 0n ex dx = e n 1. Use this to find a good
1
sandwich for n(e n 1) and compute its limit as n .
Answers to Test Exercises:
1
1
+1 n
2
4 n3
n2
1) (a) limn nn3 +n
=
lim
= 1.
1
1
n
4
2
n +n
1+ 2
n
n

n)( n+4+ n)

4 n+2

(b) limn n + 2( n + 4 n) = limn n + 2 ( n+4

=
lim
=
n
( n+4+ n)
( n+4+ n)

n 1+ 2
1+ 2
limn 4 4 n = limn 4 n 4 = 2.
n( 1+ n +1)
1+ 1+ n

n
n
(c) 1 1 + 2 + 3 + + n n2 . Therefore, the Sandwich Rule gives

n
1 = lim n 1 + 2 + 3 + + n lim n2 = 1.
n

86

R1
R1
R1 1
1
1
2) Because 1 ex e n for x [0, n1 ], also n1 = 0n 1 dx 0n ex dx 0n e n dx = n1 e n .
R1
R1
1
1
Multiply by n: 1 n 0n ex dx e n . But 0n ex dx = (e n 1), so by the Sandwich Rule:
1
1
1 limn n(e n 1) limn e n = 1.

20.14

Section 14: Cauchy Sequences

To know by heart:
1. The definition of Cauchy sequence: > 0 N N m, n N |am an | < .
2. In a complete ordered field (such as R), every Cauchy sequence convergence.
If the ordered field is not complete (such as Q), then a Cauchy sequence need not
have a limit.
3. Every convergent sequence is Cauchy. Every Cauchy sequence is bounded.
Test exercises:
1. Which of the following sequences are Cauchy? Do they have a limit in the algebraic
numbers? (a) an is the first n digits of e, (b) an = ln n, (c) an = Fn+1 /Fn ,
where Fn are the Fibonacci numbers.
2) Let (an )nN be a sequence such that |an an+1 | < 2n . Show that (an )nN is Cauchy.
Answers to test exercises:
1. (a) Cauchy, but limt e is not algebraic (b) not Cauchy (not bounded),
(c) an = Fn /Fn+1 is Cauchy, with algebraic limit 1+2 5 .
2) Let > 0 be abitrary.
Take N = bln( 2 )c + 1, so 2 2N < 2eN < .
Let n m N be arbitrary.
Then (by the triangle inequality applied several times, and geometric series)
|am an | |am+1 am | + |am+2 am+1 | + + |an an1 |
2m
2
m
2m + 2(m+1) + + 2(n1)
N < .
1 = 22
2
1 2

20.15

Section 15: Series

To know by heart:
Pk
P
1. The sum of a series
n=1 an .
n=1 an is the limit of it partial sums Sk =
P
2. For a series to
n=1 an converges, it is necessary that the terms an 0, but this is
not sufficient. It depends on how fast the terms tend to zero.
P
1
3. The harmonic series
n=1 n diverges.
P
n+1 1
converges.
4. The alternating harmonic series
n=1 (1)
n
87

P
P
5. A series
a
is
absolutely
convergent
if
n
n=1
n=1 |an | converges. The alternating
harmonic series is an example of a convergent, but not absolutely convergent series.
P
a
rN
6. Geometric series:
for |r| < 1.
n=N r n = a 1r
7. Finite geometric series:

PM

a
n=N r n

=a

r N r M +1
1r

8. You can split a series

(an + bn ) into two series
n=1
P
P
know that
n=1 an and
n=1 bn both converge.
Some special series:
P
2
1

2 = 6

n=1
n

P 14 = 4
n=1 n
90
P
n+1 1

= ln 2

n=1 (1)
n

1
=1
n=1 n(n+1)

for any r 6= 1.

X
n=1

an +

bn

alternating harmonic series

use partial fractions & telescoping series

Test exercises:
1. Compute the following geometric series:
P
P 23n+1
P cos n
P
2n +(3)n
3n
(b)
(c)
(d)
(a)
n
n
n=0
n=1
n=1 32n1
n=4 5n
7
13
2. Compute the following telescoping series:
P
P
P
3
1
1
(a)
(b)
(c)
n=1 n2 +3n
n=1 n2 +n3
n=4 n2 +1 .
Answers to test exercises:
P
P
2n +(3)n
3n
81
7
1
(b)
= 75 + 10
= 2 10
1. (a)
n = 125
n=0
n=4
5
7n
P 23n+1
P
cos n
27
(c) n=1 13n = 16
(d)
n=1 32n1 = 82 .
5
2. Compute the following telescoping series:
P
P
P
3
5
1
2
1
7
(a)
(b)
(c)
n=1 n2 +3n = 1 6
n=1 n2 +n3 = 6 1
n=4 n2 +1 = 24 .

20.16

Section 16: Convergence Criteria for Series

To know by heart:

X
1
1. p-test:
converges if and only if p > 1.
np
n=1

2. Leibniz Criterion: If

n=1

only

n=1

an satisfies

(an )nN is alternating;

an 0;
|an | is decreasing,
P
then
n=1 an converges.
88

if you

Test exercises:
1. Are the following series convergent or divergent?
P
P (1)n
P 1
P
n+1 n+2
1

(a)
(c)
(b)
(d)
n=1 n
n=1
n=1 (1)
n
2n+1
P n=1 n+7
2. Assume that n an is an alternating series and a1 > 0. What can you say about the
odd partial sums (S2k+1 ) and the even partial sums (S2k )?
Answers to test exercises:
1. (a) divergent (b) divergent (c) convergent (d) divergent
2. The odd partial sums (S2k+1 ) are a decreasing sequence, the even partial sums (S2k )
are an increasing sequence, and if also an 0, then both have the same limit.

20.17

Section 17: Power Series

To know by heart:
1. A power series is a series of the form:

an xn .

n=0

2. It has a radius of convergence R. The power series converges for |x| < R and
diverges for |x| > R.
3. Compute the radius of convergence by the ratio test or the root test.
4. The region of convergence includes at least (R, R). What happens for the
boundary points R requires further analysis.
5. If R = 0, then

an x converges only for x = 0. If R is infinite, then

n=0

X
n=0

converges for all x R.

6. Special power series to remember:

1
1x

= 1 + x + x2 + x3 + + xn +

ln(1 + x) =
ex

x 21 x2 + 13 x3 + (1)n+1 n1 xn +

= 1+x+

x2
2

cos x

= 1

x2
2

x4
24

sin x

= x

x3
6

x5
120

arctan x

= x

x3
3

x5
5

x3
6

+ +

xn
n!

+
2n

x
+ + (1)n (2n)!
+
2n+1

x
+ + (1)n (2n+1)!
+
2n+1

+ (1)n x2n+1 +

Test Exercises:
1) Give the radii and region of convergence for the following series:


P
P
9n x3 2
1
n
(b)
(a)
n=1 n x+1 .
n=1 ln n+3 x ,
89

|x| < 1
x (1, 1]
xR
xR
xR
x (1, 1]

an x n

1
1
1
2) What is 12 16 + 24
120
+ 720
...?
Answers to Test Exercises:
P
1) (a) The radius of convergence of
n=1
1
ln(n+3)(1+ n+3
)
limn
ln n+3

1
xn is (using the
ln n+3
1
ln(n+3)+ln(1+ n+3
)
limn
= 1,
ln n+3

ratio test): R =

n+4
=
=
so the series conlimn ln
ln n+3
verges at least for x (1, 1).
Now the boundary points:
P
P 1
1
x = 1 gives

n=1 ln n+3
n=1 n which diverges (harmonic series).
P

1
1
x = 1 gives n=1 ln n+3 (1)n . This is an alternating series, the term ln n+3
(1)n 0
1
as n and the absolute values of the terms ln(n+3) form a decreasing sequence. So by
Leibniz criterion, the series converges.
Hence the region of convergence is [1, 1).

2
P 9n n 1
.
The
radius
of
convergence
of
(b) First simplify y = x3
n=1 n y is 9 .
x+1
For the right boundary point:

X
9n n X 1
y =
, which diverges (harmonic series).
y = 91 gives
n
n
n=1
n=1
The left boundary point y = 91 is of no consequence, because y 0 (it is the square of
something).
= 13 . Solving x3
= 13 gives x = 5, and x3
= 13 gives x = 2.
y = 19 translates into x3
x+1
x+1
x+1
Neither boundary point belongs to the region of convergence, so the region of convergence
in (2, 5).
P
(1)n
1
1
1
1
1
1
2) 21 16 + 24
120
+ 720
= 1 1 + 12 16 + 24
120
+ 720
=
= e1 .
n=0 n!

90