BE Mathematical Extended Essay

Constructions of the real numbers

a set theoretical approach

by Lothar Sebastian Krapp

Constructions of
the real numbers
a set theoretical approach

Lothar Q. J. Sebastian Krapp

Mansfield College
University of Oxford

Supervised by
Dr Peter M. Neumann, O. B. E.
Mathematical Institute and
The Queens College
University of Oxford

College Tutors:
Derek C. Goldrei
Prof. Colin P. Please

Date of submission: 24 March 2014

Oxford, United Kingdom, 2014

 Contents

Contents
1 Introduction 1

2 Construction of the rational numbers 3

2.1 Tools for the natural numbers . . . . . . . . . . . . . . . . . . . . . . 3
2.2 From the naturals to the integers . . . . . . . . . . . . . . . . . . . . 4
2.3 From the integers to the rationals . . . . . . . . . . . . . . . . . . . . 9

3 Classical approaches to completeness 12

3.1 Dedekinds construction through cuts . . . . . . . . . . . . . . . . . 12
3.2 Cantors construction through Cauchy sequences . . . . . . . . . . . 18

4 A non-standard approach 23
4.1 Ultrafilters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.2 Hyperrationals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.3 Completeness through a quotient ring . . . . . . . . . . . . . . . . . 29
4.4 A non-typical notion of completeness . . . . . . . . . . . . . . . . . . 31

5 Contrasting methods 32
5.1 Equivalence of real number systems . . . . . . . . . . . . . . . . . . . 32
5.2 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

A Appendix 35

References 37

i
 Notation and terminology

Notation and terminology

Throughout the essay, the same symbols will be used to describe different but simi-
lar constants, operations, relations or sets. In particular this includes the numbers
0 and 1, the binary operators (+) and (), the order relation (<), the equivalence
relation () and the sets N, Z, Q and R. Usually it should be clear from the con-
text which specific constant, operation, relation or set the symbol used refers to. If,
however, confusion is likely to arise, subscripts will be used to avoid ambiguity.

The multiplication symbol () will sometimes be omitted, and the usual conven-
tions for bracketing and the order of different operations in an equation will be used.

We are only going to work with total orders denoted by (<). We will therefore
sometimes refer to totally ordered sets only as ordered sets, and to total orders sim-
ply as orders. The standard notation for the corresponding weak orders () and
converses (>) and () will be used.

We are only going to work with commutative rings and semirings with a mul-
tiplicative identity and will therefore simply refer to them as rings and semirings
respectively.

ii
 1 Introduction

1 Introduction
Please forget everything you have learnt in school; for you have not learnt it.

This advice is given by Landau in Grundlagen der Analysis (Foundations of

Analysis) [9] to students reading his book. Just like Landau, we will establish the
real numbers in a constructive way by defining what a real number is and then de-
ducing further properties, opposed to the synthetic approach which only describes
the properties of the set of real numbers and postulates the existence of an algebraic
system satisfying these properties. Hence, it is recommendable to forget everything
we believe to know about the real numbers so that we can start from scratch.

The set of real numbers R is widely introduced as the (unique) set satisfying the
axioms for a complete ordered field. While different axiomatic descriptions of ordered
fields are usually quite similar, there exist various versions of the completeness axiom
whose equivalence is non-trivial. A version frequently used is the supremum property.
It states that every non-empty subset of the real numbers which is bounded above
has a least upper bound in the real numbers.
Besides this axiomatic approach, R can also be constructed from basic principles
of set theory. In particular when it comes to completeness, there are several different
methods of filling the gaps in Q leading to those various equivalent versions of the
completeness axiom.
This essay will firstly demonstrate how Q can be founded from set theoretical
first principles and secondly describe and contrast different ways of constructing the
real numbers from the rational numbers.
We will start from the premise that the set of natural numbers and its properties
have been established set theoretically.

In the first chapter, a standard way of constructing the integers from the natural
numbers and thereupon the rational numbers from the integers will be demonstrated.
Both the integers and the rational numbers form algebraic constructs which can be
introduced axiomatically. Therefore, in each section the axiomatic approach will
briefly be explained, then the set including operations and order relations will be
defined, and finally we will give a proof that the set does indeed satisfy the axioms
given before.
In the second chapter, two classical approaches to completeness, namely Dedekinds
and Cantors constructions, will be explained in detail. The versions of the com-
pleteness axiom corresponding to each construction will be stated.
A non-standard approach via hyperrational numbers is described in the third

1
 1 Introduction

chapter. Since it is a rather uncommon way of completing the rational numbers, a

version of the completeness axiom is suggested but not considered thereafter.
The last chapter will contain a theorem establishing uniqueness of the real num-
bers and finally discuss how different concepts of completeness are useful, but how
they all result in the same object: the complete ordered field of real numbers.
Throughout the essay, once we have shown that a set we have established forms
a certain algebraic construct, we will use the standard notations and results for it,
e. g. uniqueness of identities and inverses etc.

2
 2 Construction of the rational numbers

2 Construction of the rational numbers

Both Goldrei [7] and Landau [9] describe in detail how to construct the real num-
bers starting from the natural numbers. However, they use different approaches:
While Goldrei constructs the integers from the naturals, thereupon the rationals,
and ultimately the reals, Landau first constructs the positive rationals from the nat-
urals, then the the positive reals and finally gives a notion of negative real numbers.1
Since the methods used in [7] consist entirely of set theoretical constructions, we will
mainly follow this one. Endertons construction in [4] is similar. All three of these
references will be used for justification of certain results.

2.1 Tools for the natural numbers

The set of natural numbers N, as introduced in Pila [12], is the unique inductive set
contained in every inductive set. Furthermore, it is equipped with the two binary
operators of addition (+) and multiplication (), which are recursively defined on N.
A strict total order on N is given by

n < m : n m.

Multiplication and addition are associative and commutative, multiplication is dis-

tributive over addition, and 0 and 1 are the additive and multiplicative identity
respectively (see [4] p. 79 ff.). That is, for all m, n, k N, the following hold:

(m + n) + k = m + (n + k), (2.1)
m + n = n + m, (2.2)
n + 0 = n, (2.3)
(m n) k = m (n k), (2.4)
m n = n m, (2.5)
n 1 = n, (2.6)
m (n + k) = (m n) + (m k). (2.7)

The order relation (<) has the following properties (see [4] p. 84 f.):
For all m, n, k N:

If m < n, then m + k < n + k. (2.8)

If m < n and 0 < k, then k m < k n. (2.9)
If m < n and n < k, then m < k. (2.10)

1
We will see in section 3.1 why avoiding negative real numbers at first makes it easier to define
multiplication on the real numbers.

3
 2 Construction of the rational numbers

Exactly one of the following holds: m < n, m = n, or n < m. (2.11)

In fact, these properties of N directly imply that (N, +, , <) is an ordered semir-
ing (see Definition A.1). These tools for the natural numbers are enough to define
the integers with corresponding operations and order relation.

2.2 From the naturals to the integers

Axiomatically, the set of integers Z can be introduced as the smallest ordered ring.
This means that every ordered ring R contains a natural copy of Z, namely the
additive subgroup of R generated by 1R . The set N can be considered as the subset
of non-negative integers.
However, in this way the negative integers are obtained synthetically by defining
them as additive inverses of corresponding natural numbers. We do not obtain a set
theoretical definition!
The property of the ring Z which distinguishes it from the semiring N is the
existence of additive inverses. To define negative integers rigorously, we therefore
need, in some sense, a notion of subtraction.
The trick is to consider ordered pairs (m, n) N N. Using our traditional
notion of subtraction on the integers, the ordered pair (m, n) corresponds to the
number m n in Z. The representation of an integer by a pair of naturals is not
unique, as different pairs of natural numbers can correspond to the same integer.
For instance, (0, 1) and (2, 3) correspond to 0 1 and 2 3 respectively, which
both equal 1. Hence, we must define an equivalence relation by determining when
two pairs of natural numbers correspond to the same integer. The idea behind this
equivalence relation will be that m n = k ` if and only if m + ` = k + n.

Definition 2.1. We define a relation () on N N by

(m, n) (k, `) : m + ` = k + n.

Notation 2.2. The following proofs might become clearer if we use a notation which
is more familiar to the reader. We will therefore denote the equivalence class [(m, n)]
of (m, n) by
[m n],

without specifying what the new symbol () means.

Proposition 2.3. () defines an equivalence relation on N N.

Proof. Reflexivity and symmetry of () directly follow from the reflexivity and sym-
metry of equality (=).

4
 2 Construction of the rational numbers

Let (m, n), (k, `), (i, j) N N such that (m, n) (k, `) and (k, `) (i, j).
Then m + ` = k + n and k + j = i + `. So

(m + `) + (k + j) = (k + n) + (i + `).

Hence, by (2.1), (2.2) and Theorem A.2,

m + j = i + n.

So (m, n) (i, j). q. e. d.

Definition 2.4. The set of integers Z is defined as the set of all equivalence classes
under this equivalence relation:

Z := (N N)/ = {[n m] | n, m N} .

The operators of addition and multiplication as well as a strict total order on Z

are defined as follows:

Definition 2.5. For m, n, k, ` N, we define

[m n] +Z [k `] := [(m + k) (n + `)], (2.12)

[m n] Z [k `] := [(m k + n `) (m ` + n k)], (2.13)
[m n] <Z [k `] : m + ` < k + n. (2.14)

Interpreting the symbol () with our traditional notion of subtraction, the ideas
behind the definitions should all become clear.

Proposition 2.6. The binary operations of addition (+Z ) and multiplication (Z )

as well as the order relation (<Z ) on Z are well-defined.

Proof. Let (m, n), (k, `), (m0 , n0 ), (k 0 , `0 ) N N such that (m, n) (m0 , n0 ) and
(k, `) (k 0 , `0 ). Then m + n0 = m0 + n and k + `0 = k 0 + `.
By applying (2.1) and (2.2),

(m + k) + (n0 + `0 ) = (m + n0 ) + (k + `0 )
= (m0 + n) + (k 0 + `)
= (m0 + k 0 ) + (n + `).

Hence, (m + k, n + `) (m0 + k 0 , n0 + `0 ).
The proof that multiplication is well-defined requires numerous simple algebraic
manipulations. It is described in [4], Lemma 5ZE.
Finally, by (2.1), (2.2), (2.8) and Theorem A.3,

m + ` < k + n m + ` + n0 + `0 < k + n + n0 + `0

5
 2 Construction of the rational numbers

m0 + n + ` + `0 < k 0 + ` + n + n0
m0 + `0 < k 0 + n0 .

Hence, [m n] < [k `] if and only if [m0 n0 ] < [k 0 `0 ].

q. e. d.

Notation 2.7. For any integer [m n], we denote [n m] by [m n] and call

it its negative. Integers of the form [n 0] are referred to as n or nZ . The natural
embedding of N in Z is given by

NZ := {[n 0] | n N} ,

with the inclusion map n 7 [n 0] = n.

A new binary operator on Z, subtraction (Z ), is defined as

x Z y := x + (y),

for any x, y Z.
An integer x is called positive if x >Z 0. It is called negative if x <Z 0. We
denote the set of positive integers by

Z+ := {x Z | x >Z 0} .

Note that Z+ = NZ \ {0}.

Proposition 2.8. Every integer is of the form m or m for some m N.

Proof. Let [k `] Z. By Theorem A.4, there are three distinct cases.

Case 1, k = `.
Then k + 0 = 0 + `. So [k `] = [0 0] = 0.
Case 2, m N : k = ` + m.
Then k + 0 = m + `. So [k `] = [m 0] = m.
Case 3, m N : ` = k + m.
Then k + m = 0 + `. So [k `] = [0 m] = [m 0] = m. q. e. d.

By the previous proposition, every integer x Z can be expressed as n for some

n N, where is the sign of x (either a place holder or a minus sign).

Proposition 2.9. For any m, n N and any signs 1 , 2 ,

1 m 2 n = 1 2 (mn)Z .

6
 2 Construction of the rational numbers

Proof. Four cases need to be checked:

Case 1.
m Z n = [m 0] Z [n 0] = [(mn) 0] = (mn)Z .

Case 2.

m Z n = [0 m] Z [n 0] = [0 (mn)] = [(mn) 0] = (mn)Z .

Case 3.

m Z (n) = [m 0] Z [0 n] = [0 (mn)] = [(mn) 0] = (mn)Z .

Case 4.

(m) Z (n) = [0 m] Z [0 n] = [(mn) 0] = (mn)Z = ((mn)Z ).

q. e. d.

Note that two consecutive minus signs leave the number unchanged. So 1 2 can
always be substituted by some sign 3 .
Using the tools for the natural numbers which were given in the previous section,
we can show that (Z, +Z , Z , <Z ) indeed satisfies the axioms for an ordered ring given
in Definition A.1:

Theorem 2.10. (Z, +Z , Z , <Z ) forms an ordered ring.

Proof. Let x = [x1 x2 ], y = [y1 y2 ], z = [z1 z2 ] Z.

(A1) (x +Z y) +Z z = [((x1 + y1 ) + z1 ) ((x2 + y2 ) + z2 )]

(2.1) = [(x1 + (y1 + z1 )) (x2 + (y2 + z2 ))]
= x +Z (y +Z z). (2.15)
(A2) x +Z y = [(x1 + y1 ) (x2 + y2 )]
(2.2) = [(y1 + x1 ) (y2 + x2 )]
= y +Z x. (2.16)
(A3) x +Z 0 = [(x1 + 0) (x2 + 0)]
(2.3) = [x1 x2 ]
= x. (2.17)
(A4) x +Z (x) = [(x1 + x2 ) (x2 + x1 )]
(2.2) = [(x1 + x2 ) (x1 + x2 )]
= [0 0]
= 0. (2.18)

7
 2 Construction of the rational numbers

By Proposition 2.8, there are m, n, k N and signs 1 , 2 , 3 such that x = 1 m, y =

2 n, z = 3 k. We can now apply Proposition 2.9.

(M1) (x Z y) Z z = (1 m Z 2 n) Z 3 k
= 1 2 3 ((m n) k)Z
(2.4) = 1 2 3 (m (n k))Z
= 1 m Z (2 n Z 3 k)
= x Z (y Z z). (2.19)
(M2) x Z y = 1 m Z 2 n
= 1 2 (m n)Z
(2.5) = 1 2 (n m)Z
= 1 n Z 2 m
= y Z x. (2.20)
(M3) x Z 1 = [((x1 1) + (x2 0)) ((x1 0) + (x2 1))]
(2.6) = [x1 x2 ]
= x. (2.21)
(D) x Z (y +Z z) = [x1 x2 ] Z ([y1 y2 ] +Z [z1 z2 ])
= [x1 x2 ] Z [(y1 + z1 ) (y2 + z2 )]
= [(x1 (y1 + z1 ) + x2 (y2 + z2 )) (x1 (y2 + z2 ) + x2 (y1 + z1 ))]
(2.7) = [(x1 y1 + x1 z1 + x2 y2 + x2 z2 ) (x1 y2 + x1 z2 + x2 y1 + x2 z1 )]
= ([x1 x2 ] Z [y1 y2 ]) +Z ([x1 x2 ] Z [z1 z2 ])
= (x Z y) +Z (x Z z). (2.22)

Note that for (M3) we also used the fact that ` 0 = 0 for every ` N.

(O1) x <Z y = x1 + y2 < x2 + y1

(2.8) = x1 + z1 + y2 + z2 < x2 + z2 + y1 + z1
= x +Z z <Z y +Z z. (2.23)

(O2) Suppose that z >Z 0. Then z = k = [k 0]. Suppose further that x <Z y.
Since then x1 + y2 < x2 + y1 , we obtain by (2.9) that k(x1 + y2 ) < k(x2 + y1 ). Hence,

z Z x = [(kx1 ) (kx2 )] <Z [(ky1 ) (ky2 )] = z Z y. (2.24)

(O3) Suppose that x <Z y and y <Z z. Since then x1 + y2 < x2 + y1 and
y1 + z2 < y2 + z1 , we obtain by (2.8) that x1 + y1 + y2 + z2 < x2 + y1 + y2 + z1 .
Hence, by Theorem A.3, x1 + z2 < x2 + z1 . Thus,

x <Z z. (2.25)

8
 2 Construction of the rational numbers

(O4) By (2.11), exactly one of the following three cases holds: x1 + y2 < x2 + y1 ,
or x1 + y2 = x2 + y1 , or x1 + y2 > x2 + y1 . These correspond to:

x <Z y, x = y, x >Z y, (2.26)

respectively. q. e. d.

Since we have shown that Z is an ordered ring and therefore satisfies all the
properties of an ordered semiring, the structure of NZ is preserved. We will from
now on consider N in the form of its embedding NZ as a subset of Z. In particular,
we can now use the same symbols for constants, operations and relations whenever
we refer to naturals and integers.
The tools which we have established are enough for us to proceed to the rational
numbers in a similar fashion.

2.3 From the integers to the rationals

In this section, the set of the rational numbers, denoted by Q, will be constructed
from Z. The ideas and methods used will exhibit many similarities to the ones for
the construction of Z from N. We will therefore only give the relevant definitions and
prove the most important properties of the newly established set Q. Further justi-
fication can be found in [7] Section 2.4, [9] Chapter 2, and [4] Chapter 5, pp. 101 111.

Axiomatically, the set of rational numbers Q is introduced as the smallest ordered

field. That is, every ordered field K contains a subfield isomorphic to Q. This
subfield is the one created by 0K and 1K .
The additional axiom which a field satisfies but a ring does not is the existence
of a multiplicative inverse for every non-zero element.
Again, this axiomatic approach does not give us a set theoretical definition but in-
stead postulates the existence of multiplicative inverses and integer multiples thereof.
To define those new elements rigorously in a set theoretical context we need, in some
sense, a notion of division.
The trick, once again, is to consider equivalence classes of ordered pairs (m, n)
Z Z+ . Informally, the pair (m, n) corresponds to the rational number m n . As in
the previous section, the pair corresponding to a particular rational number is not
unique. We therefore define an equivalence relation on Z Z+ , based on the idea
that two fractions m k
n and ` are equal if and only if m ` = k n.

Definition 2.11. We define a relation () on Z Z+ by

(m, n) (k, `) : m ` = k n.

9
 2 Construction of the rational numbers

Proposition 2.12. () defines an equivalence relation.

Notation 2.13. We denote the equivalence class [(m, n)] of the pair of integers
(m, n) by
m
.
n
A formal definition of division will be given later on.

Definition 2.14. The set of rational numbers Q is defined as the set of all equiva-
lence classes under ():

m 
 
+
Q := (Z Z )/ = (m, n) Z Z+ .
n 
As is necessary for an ordered field, we need to define addition, multiplication
and an order relation.

Definition 2.15. For m, k Z, n, ` Z+ , we define

m k (m `) + (k n)
+Q := , (2.27)
n ` n`
m k mk
Q := , (2.28)
n ` n`
m k
<Q : m ` < k n. (2.29)
n `
Proposition 2.16. The binary operations of addition (+Q ) and multiplication (Q )
as well as the order relation (<Q ) on Q are well-defined.
n
Notation 2.17. For any n Z, we denote 1 by n or nQ . The natural embedding
of Z in Q is given by the map
n
n 7 = n = nQ .
1
This induces a copy of Z in Q:

n 
 
ZQ := nZ .
1 
m
For any rational number x = n, its negative is given by

m m
x = := .
n n
If m 6= 0, we define

m
1 n if m > 0,
m
:=
n n if m < 0.
m

The binary operator subtraction (Q ) is, as before, defined as

x Q y = x + (y),

10
 2 Construction of the rational numbers

for any x, y Q.
A new binary operator (/Q ), called division, on Q Q \ {0} is defined by

x
= x/Q y := x y1 .
y
We can now proceed to showing that (Q, +Q , Q , <Q ) with additive identity 0
and multiplicative identity 1 is an ordered field using the properties (2.15) (2.26)
of the ordered ring Z. We will, however, only prove the new property of Q the
existence of multiplicative inverses for non-zero elements. Detailed proofs of the
other properties can be found, for example, in [4], p. 104109.

Proposition 2.18. (Q, +Q , Q , <Q ) forms an ordered field.

m
Proof. (M4) Let x = n Q \ {0}. Suppose that m > 0. Then

m n
   
1
x Q x = Q
n m
mn
=
nm
(2.20) =
mn
.
mn
Since (m n) 1 = m n = 1 (m n), we obtain
mn 1
x Q x1 = = = 1. (2.30)
mn 1
n
If n < 0, we have x1 = m . In that case we can argue similarly to obtain

(m n)
x Q x1 = = 1.
(m n)

q. e. d.

Since Q satisfies the axioms for an ordered field, which entirely includes the ax-
ioms for an ordered ring, we can identify Z with the subset ZQ of Q and use the
same symbols for operators and relations on those sets.

Establishing the ordered field of rational numbers is the prerequisite for the
actual set we are aiming for: the complete ordered field of real numbers. We still
need to give a formal description of completeness as well as methods of filling the
gaps in Q, which we will explore in the next chapter.

11
 3 Classical approaches to completeness

3 Classical approaches to completeness

Already in the ancient world it was known that there exist numbers representing a
length in geometry which is not representable as a rational number. These irra-
tional numbers are the gaps on the continuous number line which are not filled
by rationals. This already gives us an intuitive concept of completeness: The real
number line is complete means that there are no such gaps which are not covered
by a real number.
As it turns out, the complete ordered field of real numbers R is unique up to iso-
morphism (see section 5.2); in fact, there is a unique isomorphism from one complete
ordered field to another. There are various equivalent ways of formally describing
the completeness property. For instance, one common notion of completeness, which
was described in the introduction, is the supremum property. Depending on how
the problem of filling the gaps is approached, different but equivalent notions of
completeness naturally arise.
In the following two sections, two classical approaches will be demonstrated
in detail: Dedekinds construction through cuts on Q and Cantors construction
through rational Cauchy sequences.
Dedekind cuts and resulting properties are fully described in [9] Chapter 3 & 4,
and [7] Section 2.2. Section 2.3 in [7] also presents Cantors approach through Cauchy
sequences. The following two sections are largely inspired by those parts of the
books.

3.1 Dedekinds construction through cuts

First we try to approach completeness by cuts on the set of rational numbers. Landau
defines in [9], p. 43, a cut as a set of rational numbers satisfying the following three
conditions:

1. It contains a rational number, but it does not contain all rational numbers.

2. Every rational number of the set is smaller than every rational number not
contained in the set.

3. It does not contain a greatest rational number.

We formalise this as follows:

Definition 3.1 (Dedekind cuts). A set r Q is called a (Dedekind) cut (on Q) if

it satisfies the following three conditions:

=
6 r 6= Q (3.1)
p r q Q \ r : p < q (3.2)

12
 3 Classical approaches to completeness

p r q r : p < q. (3.3)

We now define the set of real numbers R as the set of all Dedekind cuts on Q.
Once we have introduced other constructions of the real numbers, we will refer to
these as Dedekind real numbers and call them RD .
Using the usual notion of suprema, the set r does indeed represent the corre-
sponding number sup r in R. The idea behind the corresponding version of com-
pleteness is that every Dedekind cut has a least upper bound in R. This is in fact
already one form of the completeness axiom, which will be further discussed later in
this section.

Notation 3.2. For q Q, we denote {x Q | x < q} by q or qR .

The natural copy of Q in R is given by

QR := {q | q Q} ,

with the natural embedding map q 7 q.

Note that for any p Q, we have p q if and only if p < q.

It is easy to see that q defines a Dedekind cut whose least upper bound is q.
Once we define the necessary operators and order relation, one can show that QR is
indeed an ordered field. For a detailed proof, see [4] Theorem 5RJ.
We proceed by defining addition (+R ), multiplication (R ) and an order relation
(<R ) on R.

Definition 3.3. Let r, s R.

r +R s := {p + q | p r, q s} , (3.4)
r <R s : r ( s. (3.5)

Undoubtedly the order relation <R is well-defined. It will be useful to show that
trichotomy holds.

Proposition 3.4. For any r, s R, exactly one of the following holds:

r <R s, r = s, r >R s.

Proof. Let r, s R such that r 6= s. Then either r \ s 6= or s \ r 6= . Suppose that

r \ s 6= and take q r \ s. Since q r, all rationals less than q are also contained
in r. As q / s, all rationals in s are smaller than q. Hence, all elements of s are
contained in r. So s ( r, whence s <R r.
We can argue similarly to show that if s \ r 6= , then r <R s. q. e. d.

For the definition of multiplication on the real numbers we have to be more

careful. We will first define the product of non-negative real numbers.

13
 3 Classical approaches to completeness

Definition 3.5. Let r, s R such that r, s R 0. Then

r R s := {p q | p r \ 0, q s \ 0} 0. (3.6)

We must exclude negative numbers from the sets r and s, so that the products
of the rationals they contain do not become arbitrarily large. As similar problems
are encountered for the definition of products of other real numbers, it will be useful
to define the negative and the modulus of a real number first.

Definition 3.6. Let r R. The negative of r is defined as

r := {q Q | p > q : p Q \ r} .

Proposition 3.7. For any r R, its negative r is a Dedekind cut.

Proof. Let q r and s Q such that s > q. Then s < q, so s r. Hence,

q / r. Now let u Q \ r. Since u > u 1 and (u) = u Q \ r, the
rational u 1 lies in r.
Suppose that q 0 is a rational less than q. Since q r, there exists p > q > q 0
such that p Q \ r. Thus, q 0 r.
Next we need to show that r contains no greatest element. Let t := q+p 2 . Then
q < t < p. As t is rational and p Q \ r, it is also contained in r. So for every
element in r, we can find a greater element also lying in r. q. e. d.

Definition 3.8. The modulus function from R to the set of non-negative real num-
bers R0 is defined as

r if r R 0,
| . | : R R0 , r 7
r if r <R 0.
|r| is called the absolute value of r.

An equivalent, rather set theoretical definition of the modulus would be |r| =

r r.
Now we can define multiplication of two real numbers in the remaining cases.

Definition 3.9. Let r, s R.

(r R |s|) if r R 0, s <R 0,

r R s := (|r| R s) if r <R 0, s R 0,



|r| R |s| if r, s <R 0.

Unlike in the previous section, it is not obvious that the real numbers are closed
under addition and multiplication. We need to check whether the sum and product
of two reals are contained in R.

14
 3 Classical approaches to completeness

Proposition 3.10. R is closed under addition (+R ) and multiplication (R ).

Proof. Let r, s R.
(1) Let q r, a Q \ r, p s and b Q \ s. Then q + p r + s. For any
q r and p0 s, we have q 0 < a and p0 < b. So q 0 + p0 < a + b. Thus, a + b
0 / r + s.
Now let c Q \ (r + s). Since c 6= q + p, either c < q + p or c > q + p. But if
c < q + p, then c q < p. So c q s, and thus c r + s, a contradiction. Hence,
c > q + p.
Finally, we need to show that q + s contains no greatest element. Let u r and
v s such that q < u and p < v. Then p + q < u + v r +R s.

(2) Suppose that r, s >R 0. Let q r \ 0 and p s \ 0. The product r R s is

nonempty, as 0 r R s. Let a Q \ r and b Q \ s. Then 0 < q < a and 0 < p < b.
So q p < a b, whence a b
/ r R s.
Let c Q such that c < q p. If c 0, then c r R s already. If c
/ 0, then
1 1
c 0. In particular then q > 0, and 0 c q < p. So c q s \ 0. Thus,
c r R s.
Finally, for u r and v s such that q < u and p < v, since 0 < u and 0 < v,
we obtain q p < u v r R s.
If r = 0 or s = 0, their product in R is 0.
In the remaining cases the product is a Dedekind cut by definition. q. e. d.

To show that R is a field, we need to introduce multiplicative inverses.

Definition 3.11. For any s R \ {0}, we define its multiplicative inverse as follows:
If s >R 0, then
n o
s1 := q Q \ {0} | p Q \ s : p < q 1 0 {0} .

If s <R 0, then
s1 := |s|1 .

For a proof that s1 defines a cut, see [9] Theorem 152.

Theorem 3.12. (R, +R , R , <R ) is an ordered field.

Proof. Let x, y, z R.
Properties (A1)(A2) follow from accociativity and commutativity of Q under
addition.
(A3) To show: x + 0 = 0.
Let q x and p 0. Since p < 0, we obtain q + p < q. So q + p x. Hence,
x +R 0 x.

15
 3 Classical approaches to completeness

Now suppose that t x. Let t0 x such that t < t0 . Then t t0 0. Hence,

t = t0 + (t t0 ) x + 0. Hence, x x +R 0.

(A4) See [4] Theorem 5RF.

(M1)(M3) These can first be demonstrated for positive real numbers. We can
then use the identity (x) = x, which follows from (A2)(A4), for the other cases.
(M4) First consider x >R 0. Let u x \ 0 and v x1 \ 0. Then v 1 Q \ x
and u1 Q \ x1 . If x <R x1 , then v < u1 . If x1 <R x, then u < v 1 . Hence,
uv < 1. So x R x1 1.
The other direction, i. e. showing that 1 x R x1 , is fully demonstrated in [9],
Theorem 152.
For x <R 0, the proof follows from the definition of multiplicative inverses of
negative numbers.
(D) This follows immediately from distributivity in Q.
Conditions (O1)(O3) all follow immediately from the properties of the proper
subset relation ((). Trichotomy (O4) was already proved. q. e. d.

As R is an ordered field, and the structure of Q in R considered as its natural

copy QR is preserved, we can from now on use the same notation.
The important additional property of completeness, which distinguishes R from
Q, remains to be introduced and explained.

In [2], p. 322, Dedekind himself gives a notion of completeness of the straight

line (meaning the real number line) which is closely related to Dedekind cuts:

If all points on the straight line fall into two classes, such that every point of
the first class lies to the left of every point of the second class, then there exists one
and only one point which produces this division of all points into two classes, this
dissection of the line into two pieces.

Notably, this notion of completeness uses Dedekind cuts on the real numbers and
expresses a one-to-one correspondence between those cuts and the real numbers. We
will formalise it and express this version of completeness for general ordered fields
in the following.

Definition 3.13 (Dedekind completeness). An ordered field F is Dedekind complete

if every Dedekind cut on F has a least upper bound in F .

Remark 3.14. Dedekind cuts on R or any other ordered field are defined similarly
to the ones on Q.

16
 3 Classical approaches to completeness

Theorem 3.15. The Dedekind real number system is Dedekind complete.

S
Proof. Let A be a Dedekind cut on R. We will first show that a := A is a
Dedekind cut on Q and thus a real number, and then that a is the least upper
bound of A.
Since A is non-empty and every element of A is nonempty, a is also non-empty.
Let b R \ A. Then for every c A, we have c <R b. Thus, c ( b. In particular,
a = A b. Hence, a ( b +R 1. So Q \ a 6= .
S

Let q a and p Q \ a = cA (Q \ c). Consider p = pR R. Since p

T
/ c for
all c A, it follows that c ( p for all c A. Thus, a p. So q p, and hence
q < p.
Suppose that s Q is an upper bound for a. Since c a for all c A, that s is
also an upper bound for each c in A. So s is not contained in any c and hence not
in a.
Hence, a is a Dedekind cut on Q.
By construction, c a, so c R a for all c A. Assume for a contradiction
that there were an upper bound a0 R for A such that a0 < a. Then a0 ( a. For
t a \ a0 , we obtain a0 ( t ( a and t A. But then a0 <R t, contradicting that a0
is an upper bound.
Hence, a is the least upper bound of A.
q. e. d.

Dedekinds notion of completeness is closely related to the version of the com-

pleteness axiom in the form of the supremum property. In fact, those two versions
(as all other versions) are equivalent (see Section 5.1).
We will show that the Dedekind real number system satisfies this axiom.

Theorem 3.16. The Dedekind real number system has the supremum property.

Proof. Let A be a non-empty set of real numbers which is bounded above. Define
the set B as follows:
B := {x R | y A : x < y} .

First we will show that B is a Dedekind cut on R, and then that the least upper
bound of B is the least upper bound of A.
As A is non-empty, there exists a A. Then a 1 B. As A is bounded above,
there exists m R such that z m for all z A. So m / B.
Let b B and c R \ B. Note that then c is an upper bound for B. Moreover,
there exists y A such that b < y. So b < y c.
Let b0 = b+y 0
2 . Then b < b < y. Hence, B has no maximum. Thus, B is a
Dedekind cut on R.

17
 3 Classical approaches to completeness

By the previous theorem, B has a least upper bound s, say. If it were not an
upper bound for A, there would be y A such that s < y. But then, for b0 = s+y 2 ,
0 0
we would have s < b < y. This is a contradiction, since b B. Hence, s is an
upper bound for A.
Assume for a contradiction that A has an upper bound s0 such that s0 < s. Since
s is the least upper bound of B, which is not contained in B, there exists b B such
that s0 < b < s. But then there exists a0 A such that s0 < b < a0 , a contradiction.
Hence, s is the least upper bound of A. q. e. d.

This finishes one possible construction of the complete ordered field of real num-
bers. Another classical approach leading to a different version of the completeness
axiom will be demonstrated in the following section.

3.2 Cantors construction through Cauchy sequences

When we define the real numbers axiomatically and consider the set of rational
numbers as a subset, we know from the study of sequences that every real Cauchy
sequence converges to a real number. We also know that for every real number p,
there exists a convergent sequence and hence Cauchy sequence of rational numbers
converging to p. This gives us a correspondence between rational Cauchy sequences
and real numbers, namely that a rational Cauchy sequence (an ) corresponds to its
limit limn an . We will formalise this concept in this section. Throughout the
section we use the standard notation for sequences and only prove results which are
not usually covered in a first year Analysis course (e.g. Earl [3]).

Let C be the set of all Cauchy sequences in Q. As we want to define the real
numbers as limits of rational Cauchy sequences, we need to make sure that two
distinct Cauchy sequences with the same limit do not correspond to different real
numbers. As in the previous section, the solution is to define an equivalence relation.

Definition 3.17. We define a relation () on C by

(an ) (bn ) : lim (an bn ) = 0.

n

This uses the idea that two convergent sequences tend to the same limit if and
only if their difference tends to zero.
We could proceed by proving that this relation is indeed an equivalence relation
and then define the necessary operators and order relation, which is fully demon-
strated in [7] Section 2.3. However, knowing that C with standard addition and
multiplication of sequences forms a ring, we will rephrase the equivalence relation
in the context of rings and ideals and exploit results from Algebra.

18
 3 Classical approaches to completeness

Definition 3.18. We define the subset I of C as the set of zero-sequences in C:

n  o
I := (an ) C  lim an = 0 .

n

Proposition 3.19. I is an ideal of C.

Proof. We need to show that I is closed under addition, and under multiplication
by elements in C.
Let (an ), (bn ) I and (cn ) C. Since (cn ) is bounded (see Theorem A.5),
limn cn an = 0. Hence, (an ) C (cn ) I. Moreover, since (an ) and (bn ) both
converge to 0, so does their sum. Hence, (an ) +C (bn ) I. q. e. d.

As I / C, the quotient C/I forms a ring with addition and multiplication induced
by C. In order to show that C/I forms a field, it suffices to prove the existence of
multiplicative inverses of elements not contained in I. This is equivalent to showing
that I is a maximal ideal.

Proposition 3.20. The quotient ring C/I is a field.

Proof. Let (dn ) C \ I.

Define a sequence (d0n ) by

d1 if dn 6= 0,
n
d0n :=
0 if dn = 0.

Since (dn ) is not a zero-sequence, there exists N N such that for all n N ,
we have dn 6= 0, and the N -th tail (d0N +n ) of (d0n ) is equal to (d1
N +n ). Thus, by
1 0
Theorem A.6, (dN +n ) is a Cauchy sequence. Hence, also (dn ) is a Cauchy sequence.
Moreover, we obtain for all n N that d0n dn = 1. So (dn d0n 1) is a zero-sequence.
Hence, (dn ) C (d0n ) + I = 1C + I. q. e. d.

Definition 3.21. We define the field of real numbers R as the quotient of the ring
of rational Cauchy sequences C and its ideal of zero-sequences I:

R := C/I.

Addition (+R ) and multiplication (R ) are induced by C.

We also call R the Cantor real number system and denote it by RC if ambiguity
might arise.

Remark 3.22. For any (an ), (bn ) C,

(an ) (bn ) if and only if (an ) C (bn ) I.

19
 3 Classical approaches to completeness

This follows directly from the definition of I and (). As a result,

(an ) (bn ) if and only if (an ) + I = (bn ) + I.

Hence,
[(an )] = (an ) + I and (C/ ) = C/I.

It therefore makes no difference whether we define the real numbers as the set of
equivalence classes of rational Cauchy sequences or as the quotient ring of the ring
of rational Cauchy sequences with its maximal ideal of zero-sequences.

Notation 3.23. For any rational number q Q, the natural copy of q in R is given
by (qn ) + I, where (qn ) is the constant sequence qn = q for all n N.
The notation for additive and multiplicative inverses as well as subtraction and
division are defined in the same way as on Q.

Also note that 0 = I.

So far we have not mentioned how the order on Q can be extended to an order
on R.

Definition 3.24. We define a relation (<R ) on C/ as follows:

[(an )] <R [(bn )] : ( Q>0 N N n N : an + < bn ).

Proposition 3.25. The relation (<R ) is well defined.

Proof. Let (an ), (bn ), (a0n ), (b0n ) C. Suppose that there exist N1 N and Q
such that > 0 and for all n N1 we have an > bn + . Suppose further that
(an ) (a0n ), (bn ) (b0n ). Then there exists N2 N such that a0n > an 31 for all
n N2 , and there exists N3 N such that bn + 13 > b0n for all n N3 .
Let N := max {N1 , N2 , N3 }. Then for all n N :
1 2 1
a0n > an > bn + > b0n + .
3 3 3
This yields that (<R ) is well-defined.
q. e. d.

Theorem 3.26. (R, +R , R , <R ) forms an ordered field.

Proof. Let a = [(an )], b = [(bn )], c = [(cn )] R. Conditions (O1) and (O2) follow
directly from the properties of Q as an ordered field.
Property (O3) follows from (O1) and the fact that a <R b if and only if 0 <R
b a.
(O4) Suppose that a 6= b. Then (an bn ) does not tend to 0. Hence, there
exists > 0 such that for all N N there exists n N such that |an bn | . Since

20
 3 Classical approaches to completeness

(an ) and (bn ) are Cauchy sequences, there exist N1 , N2 N such that |an am | < 13
for all n, m N1 and |bn bm | < 31 for all n, m N2 . Let N := max {N1 , N2 }.
Then there exists n N such that (an bn ) or (an bn ) . In the first
case, for all m N ,
1 2 1
am > an bn + > bm + .
3 3 3
Thus, a >R b.
In the second case, for all m N ,
1 2 1
am < an + bn < bm + .
3 3 3
Thus, a <R b. q. e. d.

Now that we have shown that R is an ordered field, we can consider Q in the
form of its natural embedding as a subset of R and adopt the notation on Q. The
only property left to describe is completeness.

Definition 3.27 (Cauchy completeness). An ordered field F is Cauchy complete if

every Cauchy sequence in F converges to a unique limit in F .

Remark 3.28. If F is a metric space, such as R with metric induced by the modulus
function (| . |), then the limit of a convergent sequence is automatically unique.

Theorem 3.29. The Cantor real number system R is Cauchy complete.

Proof. Let (an )n be a Cauchy sequence in R. For each n, m N, let

 
(n)
am
m
 
(n)
be a representative of the equivalence class an R. That is, am is a Cauchy
m
sequence in Q such that h  i
an = a(n) m .
m
 
(n)
Fix n N. Since am is a rational Cauchy sequence, there exists `n such that
m
for all m, m0 `n ,

 (n)

(n)  1
am am0  < . (3.7)
n
Without loss of generality, we can assume that the sequence (`n ) is strictly increasing,
as each `n can be chosen arbitrarily large. Now construct a new sequence in Q by
 
(n)
(bn )n := a`n .
n

We want to show that b := [(bn )n ] is a real number and the limit of (an ).

21
 3 Classical approaches to completeness

First, we need to show that (bn )n is a Cauchy sequence. Let Q>0 and R the
corresponding real number. Choose N N such that for all n, n0 N ,
R
|an an0 | <R .
2
This implies that there exists M N such that for all m M ,

 (n)

0)
am a(n
m < . (3.8)
2
n o
Choose a natural K max N, M, 2 . Then for all k > k 0 K,

(k0 ) 
 
 (k)
|bk bk0 | = a`k a`
k0

0
  0
(k0 ) 
 
 (k) (k )   (k )
a`k a`k  + a`k a` 0  .
k

Since `k k > K M and k, k 0 K N , it follows by (3.8) that


 (k) (k0 ) 

a`k a`k  < .
2
As (`n ) is strictly increasing, `k > `k0 . It follows by (3.7) that
 0
 (k ) (k0 ) 
 1 1
a`k a` 0  < 0 .
k k K 2
Hence,
|bk bk0 | < ,

for all k > k 0 K. So (bn )n is a Cauchy sequence.

Now consider the sequence (|an b|)n . Fix n N. Then
h  i
 (n) (n) 
|an b| = am a`n  .
m

Since for all m `n ,


 (n)

(n)  1 2 1
am a`n  < = ,
n n n
we obtain h
 (n)
 i
(n)  2
|an b| = am a`n  <R .
n m

Letting n , we obtain that (|an b|) is a zero-sequence in R. Hence,

lim an = b.
n

q. e. d.

22
 4 A non-standard approach

4 A non-standard approach
Generally a non-standard extension of a linearly ordered algebraic structure A is a
new algebraic structure A which satisfies the same axioms as A but also contains an
infinitely large element. If we call A the natural embedding of A in A, this means
that A contains an element being greater than any element in A. For semirings,
rings and fields, this existence of an infinite element does not contradict any of the
other axioms. Therefore a non-standard model exists. As the non-standard model
satisfies the same axioms as the standard one, a non-standard ring also contains
, which is smaller than any standard element, and a field contains 1 , which is
closer to zero than any non-zero standard element.
There are different ways of constructing non-standard sets. One way is based on
Internal Set Theory, which is introduced by Nelson in [11]. Internal Set Theory starts
with the axioms of ZermeloFraenkel set theory with the Axiom of Choice (ZFC) plus
a new (undefined) unary predicate standard and three additional axiom schemes
transfer principle, principle of idealisation, principle of standardization. The
transfer priciple states that every statement in first-order logic which holds for a
standard set also holds for its non-standard model.
However, Internal Set Theory is more abstract, applicable more generally and
thus more powerful than what we need for our purposes. We will concentrate on
the construction of the set of hyperrational and later hyperreal numbers using the
means which we have established so far, plus a new set theoretical construct called
ultrafilter. This construction is also called an ultrapower construction. It is mainly
inspired by Stroyan and Luxemburg [16]. Different approaches are contrasted in
further depth in Laugwitz [10].

4.1 Ultrafilters
Initially, we need to introduce the notion of a free ultrafilter on an infinite set. We
will firstly do this axiomatically and then demonstrate a non-constructive proof for
the existence of a free ultrafilter. We use the axioms and terminology for free filters
as used in both Zelenyuk [17] and [16].
Definition 4.1. Let J be an infinite set. A set U P(J) is called a (proper) filter
on J provided:
J U and
/U (properness), (4.1)
if A, B U, then A B U (finite intersection property), (4.2)
if A U and A B J, then B U (superset property). (4.3)
A filter is called an ultrafilter if it is a maximal filter, that is, it is not properly
contained in any other filter on J.

23
 4 A non-standard approach

It is called free (or non-principal) provided:

If A U, then A is infinite. (4.4)

In the following we will only consider proper filters on an infinite set J.

The first proof of the existence of a free ultrafilter by Tarski dates back to 1930
(see [13] p. 53) and strongly depends on the Axiom of Choice in the form of Zorns
Lemma. We will prove the existence of a free ultrafilter by firstly constructing a
free filter and then extending it to a free ultrafilter using the same method as in the
proof of the Ultrafilter Theorem ([17] Proposition 2.5).

Definition 4.2. The free filter defined by

Fr(J) := {A P(J) | (J \ A) is finite}

is called the Frchet filter on J.

Showing that Fr(J) is indeed a free filter is an easy exercise.

Lemma 4.3. Every filter containing the Frchet filter is free.

Proof. Let B be an ultrafilter such that Fr(J) B. If B contained a finite set D,

then since J \ D Fr(J) B, we would obtain B 3 (J \ D) D = , contradicting
properness. Hence, B is free. q. e. d.

Theorem 4.4 (Ultrafilter Theorem). Assuming the Axiom of Choice, Fr(J) can be
extended to a free ultrafilter.

Proof. Let
A := {F P(J) | Fr(J) F and F is a filter} .
Let B A be any chain, and consider B.
S

Clearly, Fr(J) B. So J B. Also / B, as the empty set is not contained

S S

in any set in B.
Since B is a chain, B trivially satisfies the superset property.
S

Let A, B B. As B is a chain, there exists a filter X B such that A, B X .

S

Hence, A B X B. So B has the finite intersection property.

S S

Hence, B is a filter containing Fr(J), whence B A.

S S

By Zorns Lemma (see Theorem A.7), A contains a maximal element M, which

is thus an ultrafilter.
Finally, as M contains Fr(J), it is a free ultrafilter by Lemma 4.3.
q. e. d.

An equivalent condition for maximality is given by the next theorem. We will,

however, only present one direction of the implication exploiting Proposition 2.3
from [17].

24
 4 A non-standard approach

Theorem 4.5. An ultrafilter U has the following property:

If A P(J), then either A U or J \ A U. (4.5)

Proof. Let A P(J).

Case 1: There exists B U such that B A = . Then B J \ A J. Hence,
by the superset property, J \ A U.
Case 2: For every B U, we have B A 6= . Let F P(J) be defined as the
family of non-empty subsets

F := {C P(J) | B U : B A C} .

It clearly has the superset property.

For any C1 , C2 F, there exist B1 , B2 U such that C1 B1 A and C2 B2 A.
Thus,
C1 C2 (B1 B2 ) A.
| {z }
U

Hence, C1 C2 F. So F has the finite intersection property.

Consequently, F is a filter.
Clearly, U F and A F. Since U is not properly contained in any other filter, it
follows U = F, and thus A U.
q. e. d.

Later on, we will refer to property (4.5) as maximality of the ultrafilter.

Remark 4.6. At the point where we use Zorns Lemma, which is equivalent to
the Axiom of Choice, the proof of the existence of a free ultrafilter becomes non-
constructive. Hence, we do not obtain any information about any further properties
other than the ones which we can directly derive from the axioms it satisfies. As the
definition of hyperrationals only depends on the existence of a free ultrafilter, this
does not lead to any complications.

4.2 Hyperrationals
First we specify the infinite set J to be the set of natural numbers N. Let U be an
ultrafilter on N, and QN the set of rational sequences.

Definition 4.7. We define a relation on QN by

(an ) (bn ) : {n | an = bn } U.

Proposition 4.8. The relation defines an equivalence relation.

25
 4 A non-standard approach

Proof. Let (an ), (bn ), (cn ) QN .

(Reflexivity) {n | an = an } = N U, so (an ) (an ).
(Symmetry) Suppose that (an ) (bn ). Then {n | bn = an } = {n | an = bn } U.
So (bn ) (an ).
(Transitivity) Suppose that (an ) (bn ) and (bn ) (cn ). Then

{n | an = cn } {n | an = cn bn = cn }
= {n | an = bn bn = cn }
= {n | an = bn } {n | bn = cn }
| {z } | {z }
U U

Because of the finite intersection and superset property of U, we obtain {n | an = cn }

U. So (an ) (cn ). q. e. d.

If (an ) (bn ), we say that the sequences are U-equivalent.

Definition 4.9. The set of hyperrational numbers Q is defined as the set of all
equivalence classes of rational sequences under the equivalence relation ():

Q := QN / .

Definition 4.10. Addition, multiplication and an order relation on Q are defined

as follows:

[(an )] + [(bn )] := [(an + bn )] (4.6)

[(an )] [(bn )] := [(an bn )] (4.7)
[(an )] < [(bn )] : {n | an < bn } U (4.8)

Proposition 4.11. Addition and multiplication on Q are well-defined.

Proof. First consider addition. Let (an ), (a0n ), (bn ), (b0n ) QN such that (an ) (a0n )
and (bn ) (b0n ). We need to show that (an +bn ) (a0n +b0n ). Since {n | an = a0n } U
and {n | bn = b0n } U, also {n | an = a0n } {n | bn = b0n } U by the finite intersec-
tion property. Hence,

n | an + bn = a0n + b0n n | an = a0n n | bn = b0n U.

  

By the superset property, (an + bn ) (a0n + b0n ).

One can give a similar proof for multiplication. q. e. d.

Proposition 4.12. The order relation < is well-defined.

26
 4 A non-standard approach

Proof. Let (an ), (a0n ), (bn ), (b0n ) QN such that (an ) (a0n ) and (bn ) (b0n ). Suppose
that {n | an < bn } U. Then

n | a0n < b0n n | an < bn an = a0n bn = b0n

 

= {n | an < bn } n | an = a0n n | bn = b0n

 
| {z } | {z } | {z }
U U U

U,

and hence {n | a0n < b0n } U by the finite intersection and superset properties.
q. e. d.

Remark 4.13. Since we cannot make a statement about the uniqueness of a free
ultrafilter U, the set Q might not be uniquely defined. We will, however, refer to

Q as the set of hyperrationals, since all of the properties of ultrafilters used to prove
properties of Q are derived from the axioms for free ultrafilters and will therefore
not lead to imprecise statements.

Notation 4.14. We now denote hyperrationals of the form [(an )] where (an ) is a
constant sequence an = c for some c Q by c . The natural standard copy of Q in

Q by this identification is denoted by Q.

The one additional axiom extending the rational numbers to the hyperrational
numbers states the existence of an element Q which is strictly greater than
any standard rational number. We can, for example, define as

:= [(n )],

where n = n.

In order to show that Q satisfies all the axioms for a field, we use a trick similar
to the one used for our construction of the Cantor real number system, namely, to
express Q as a quotient ring with its maximal ideal which consists of all elements
equivalent to the zero-element of the ring.
First note that QN is a ring. Consider the following subset of QN :
n  o
(an ) QN  (an ) 0 .


This is infact the set 0 , the equivalence class of the constant zero-sequence.
That 0 is an ideal follows directly from the finite intersection and superset
property of U. So QN /0 is a ring. Again, we will demonstrate the existence of
multiplicative inverses.

Proposition 4.15. For every (pn ) QN such that (pn ) 6 0 there exists (qn ) QN
such that (pn qn ) 1 .

27
 4 A non-standard approach

Proof. Let (pn ) QN such that (pn ) 6 0 . Then {n | pn = 0} / U. By the maxi-

mality of U, {n | pn 6= 0} = N \ {n | pn = 0} U. Let (qn ) Q be defined by
N

1 if pn 6= 0,
pn
qn :=
0 if pn = 0.

Then {n | pn qn = 1} = {n | pn 6= 0} U. Hence, (pn qn ) 1 . q. e. d.

We denote the multiplicative inverse of Q \ {0 } by 1 .

Corollary 4.16. ( Q, + , ) is a field.

Proof. Two sequences in QN are U-equivalent if and only if they differ by a sequence
which is U-equivalent to the constant zero-sequence. Hence,

Q = (QN / ) = QN /0 .

By Proposition 4.15, Q is a field. q. e. d.

Proposition 4.17. ( Q, + , , < ) forms an ordered field.

Proof. Conditions (O1) and (O2) result directly from the order properties of Q.
(O3) Let (an ), (bn ), (cn ) QN such that [(an )] < [(bn )] and [(bn )] < [(cn )].
Then

{n | an < cn } {n | an < bn bn < cn }

= {n | an < bn } {n | bn < cn } U.

Hence, by the superset property, {n | an < cn } U.

(O4) Let (an ), (bn ) QN such that [(an )] 6= [(bn )]. Then {n | an = bn }
/ U.
So
{n | an 6= bn } = N \ {n | an = bn } U.

Hence,

{n | an 6= bn } = {n | an < bn an > bn }
= {n | an < bn }
{n | an > bn } U.

At least one of the two sets of that union must be infinite. Suppose that {n | (an < bn )}
is infinite. If {n | (an < bn )} U, then [(an )] < [(bn )]. Otherwise, by maximality
and finite intersection property of U,

(N \ {n | (an < bn )}) {n | an 6= bn } = {n | an > bn } U.

In that case, [(an )] > [(bn )].

A similar argument can be applied if {n | an > bn } is infinite. q. e. d.

28
 4 A non-standard approach

We have established the ordered field of hyperrationals Q with an infinite el-

ement . Because of the existence of infinitely large and small elements, this field
has members which we do not want in the real numbers. In the next section we will
show how we can reduce the number of elements using, again, equivalence classes.

4.3 Completeness through a quotient ring

Let O be the ring of finite hyperrationals, i. e. the following set:

O := { Q | q Q : || < q } .

We will continue working with O so that we do not have to deal with infinitely
large hyperrationals.
Next we need to consider hyperrationals which are infinitely close to each other,
as we do not want R to contain such elements. This can be done with an equivalence
relation. We will, however, start directly by defining an ideal consisting of the
infinitely small hyperrationals (infinitesimals) and 0 :

I := { Q | q Q \ {0 } : || < |q |} .

Eventually the real numbers will be defined as the quotient ring of O with I.
This quotient ring exists if O is a ring and I is an ideal.

Proposition 4.18. O is an ordered subring of Q.

Proof. To show that O is a subring of Q, we only need to check that O contains 1

and is closed under multiplication and subtraction. 1 is in Q and is thus finite.
Let , O. Then there exist q , p Q \ {0 } such that || < q and
|| < p . Hence,

| | = || || < q p Q \ {0 } ,

and
| | || + || < q + p Q \ {0 } .

Hence, and lie in O. q. e. d.

Proposition 4.19. I is an ideal of O.

Proof. Let , I and O, and let q Q \ {0 }. Then

|q |
|| < ,
2
and
|q |
|| < .
2

29
 4 A non-standard approach

So
| + | || + || < |q |.

Moreover, there exists p Q>0 such that || < p .

Since p > 0 , its inverse p1
exists and lies in Q>0 . So

|q |
Q \ {0 } .
p
Hence,
|q |
| | = || || < p = |q |.
p
Thus, + and lie in I. q. e. d.

Proposition 4.20. O/I forms a field.

Proof. It is enough to show that every element in O \ I has a multiplicative inverse

in O.
Let O \ I. Then 6= 0 . Suppose that > 0 . Then 1 exists in Q>0 .
Moreover, there exist p , q Q>0 such that

p < < q .

Hence,
p1
<
1
< q1 ,

which implies 1 O \ I.
A similar argument applies if < 0 . q. e. d.

Definition 4.21. Define the field of real numbers R as the quotient ring

R := O/I.

Addition and multiplication are induced by the ring O. An order relation on R

is given by
+ I <R + I : ( < + I 6= + I).

A proof that R is ordered can be found in Davis [1] Chapter 2, Theorem 1.8. The
embedding of Q into R is given by q 7 q + I. Again, this preserves the proper-
ties of Q as an ordered field, and we can therefore use the same notation on R as on Q.

Finally, we need to show that R is complete. We would like to use a notion of

completeness which naturally arises from the way we constructed R. This will be
discussed in the following section.

30
 4 A non-standard approach

4.4 A non-typical notion of completeness

To find a suitable statement of completeness corresponding to the ultrapower con-
struction, we first have to analyse how different versions of the completeness property
arise from particular ways of constructing real number systems.
The Dedekind construction of the real numbers first uses Dedekind cuts only on
the rational numbers and defines the real numbers as those cuts. Completeness is
then the property that every Dedekind cut on the real numbers is represented by
exactly one real number, that is, the least upper bound of that particular Dedekind
cut. Hence, there is a one-to-one correspondence between Dedekind cuts on the real
numbers and the real numbers themselves.
The Cauchy real number system consists of equivalence classes of rational Cauchy
sequences. In this context, completeness means that every real Cauchy sequence
converges to a unique real number. We obtain a one-to-one correspondence between
the equivalence classes of real Cauchy sequences (i. e. sets of Cauchy sequences that
tend to the same limit in R) and the real numbers.
In both cases, the idea of completeness is obtained by extending the sets we use
on the rationals to construct the reals. Completeness then means that there is a
structure-preserving one-to-one correspondence an order-isomorphism between
the reals and those sets (Dedekind cuts on R and equivalence classes of real Cauchy
sequences).
To apply the same idea to our ultrapower construction, we first need to ex-
tend the concept of a non-standard model of the rational numbers to general fields:
The corresponding hyperfield F of an ordered field F is defined to be the set of
equivalence classes of sequences in F under the equivalence relation induced by an
ultrafilter U. Let QF be the natural embedding of Q in F . In analogy to Q, we
define the ring of finite elements in F as

OF := { F | q QF : || < q} ,

and its ideal of infinitesimals as

IF := { F | q QF \ {0F } : || < |q|} .

Completeness now means that OF /IF is order-isomorphic to F .

Recalling how F is constructed by equivalence classes through ultrafilters, one
can phrase a sensible statement of completeness:

An ordered field F is complete if and only if every bounded sequence in F is U-

equivalent to a convergent sequence in F , and every convergent sequence in F is
U-equivalent to a bounded sequence in F . (4.9)

31
 5 Contrasting methods

For standard fields without infinitely large elements we use the common notion of
boundedness and convergence. In this case, the condition that every convergent
sequence is U-equivalent to a bounded sequence holds automatically. For a non-
standard field F , a sequence (an ) in F is bounded (by a finite element) if there
exists an element q in QF , the natural copy of Q in F , such that for all n N,

|an | <F q.

The sequence converges to a limit a F if for all QF such that >F 0F , there
exists N N such that for all n N ,

|an a| < .

(This also gives us a hint as to why non-standard fields are neither Dedekind nor
Cauchy complete: A Dedekind cut is not created by a unique element, a Cauchy
sequence does not have a unique limit. This is also discussed in Schmieden and
Laugwitz [14].)
A proof that for any ordered field F the quotient ring OF /IF is an ordered field
with the supremum property is given in [1] Chapter 2, Theorem 2.5. This verifies that
the newly established notion of completeness (4.9) implies supremum-completeness.
In fact, as the complete ordered field of real numbers is unique up to isomorphism,
this tells us that (4.9) is equivalent to any other version.
The equivalence of different notions of completeness as well as the uniqueness of
the real numbers will be discussed in the final chapter.

5 Contrasting methods
5.1 Equivalence of real number systems
In this section we want to show that the Dedekind real number system RD and
the Cauchy real number system RC are equivalent and outline a proof that the real
numbers are unique up to isomorphism.
We have already shown that RD has the supremum property, only using the
properties of Dedekind complete ordered fields (see Theorem 3.16). The converse
also holds.

Theorem 5.1. Every ordered field F which is supremum-complete is also Dedekind

complete.

Proof. Let A be a Dedekind cut on F . Then A is non-empty and bounded above by

any element in F \ A. By the supremum property, sup A exists in F . q. e. d.

32
 5 Contrasting methods

The proof that every ordered field which is supremum-complete is also Cauchy
complete, and vice versa, can be found in various Real Analysis textbooks. For
example, Forster [6] introduces the real numbers as the ordered field which is Cauchy
complete and then shows that it is supremum-complete (see [6] 9 Theorem 3). In
contrast, Hart [8] introduces the real numbers with the supremum property and
shows that they are Cauchy complete (see [8] Theorem 2.5.5.).
This yields that the three axioms of completeness Dedekind, Cauchy and
supremum-completeness are equivalent. As a result, every statement about RD
which is only derived from the complete ordered field properties can also be derived
in RC and vice versa. Yet, there is one minor detail which needs to be consid-
ered: How do we know that we cannot construct an element in RC which has no
corresponding element in RD ?
This is a crucial point when we use a constructive way of creating a new set
opposed to the axiomatic introduction: Even after showing that all the axioms are
satisfied, we need to make sure that the new set does not have properties which
cannot be derived from those axioms.

It is possible to prove that all ordered fields which satisfy one of the equivalent
versions of the completeness axiom are order-isomorphic. A full proof is given in
Spivak [15] Chapter 30. We will only outline the proof based on the explanation in
[7], Theorem 2.6.

Theorem 5.2. Any two (Dedekind) complete ordered fields are order-isomorphic.2

Proof. Let F and K be two complete ordered fields.

First note that the natural copies of Q in F and K, denoted by QF and QK
respectively, are order-isomorphic (with a unique isomorphism mapping 1F to 1K ).
Next, since F is complete, every number in F is the least upper bound of a Dedekind
cut on QF . The same statement holds for K. We can finally extend the isomorphism
to the whole sets by mapping sup AF onto sup AK for every Dedekind cut AF on
QF with corresponding Dedekind cut AK on QK . q. e. d.

This finally allows us to talk about the complete ordered field of real numbers.
As a final result, regardless of how we construct a complete ordered field F , every
property of F is also a property of every other complete ordered field.
2
The proof uses Dedekind completeness, but we could use any other equivalent notion of com-
pleteness and find a similar proof.

33
 5 Contrasting methods

5.2 Conclusion
We have an intuitive idea of what completeness means: There are no gaps in the
real number line. But even after seeing several different ways of filling the gaps in
Q, we can still ask the question whether the real numbers do not have any further
gaps which could be filled. When looking at non-standard models, in which infinitely
small quantities exist, one could say that those hyperreal numbers fill further gaps
in the real numbers. It is hard to find a general intuitive notion of completeness
which fully describes the abstract concept. However, there is a crucial feature of
completeness which we described in Section 4.4: If a concept is applied to a field
with gaps in order to complete it, the resulting complete field stays unchanged if the
same concept is applied again One cannot make a complete field more complete.
Formally, we can only state that an ordered field is complete if and only if it
satisfies a completeness axiom.

Another question we can now ask is why we have different concepts of com-
pleteness if they all turn out to be equivalent. So far we have only worked with
completeness on linearly ordered fields, but if we look at the tools which are neces-
sary to define the Dedekind and the Cauchy real number systems, we see that some
properties of linearly ordered fields are not fully exploited for the construction of a
complete set or the corresponding completeness axiom.
By using Cauchy sequences, the concept of order is reduced to the concept of
distance between two points. In the field of Topology, Cauchy sequences and their
convergence can be used in general metric spaces. Indeed a metric space is called
complete if and only if every Cauchy sequence converges to a point in that space.
As mentioned before, in a metric space limits of sequences are always unique. We
therefore do not need the uniqueness condition in the statement of Cauchy com-
pleteness.
In contrast, Dedekind cuts vastly rely on the order properties of a field. They
are still well-defined if the order is not total, i. e. if there are elements which are not
comparable under the order relation. Dedekind cuts can therefore be performed on
any partially ordered set.
Other constructions lead to other general applications. Different versions of the
completeness axiom can be applied to different mathematical constructs.
In Faltin et al. [5], a paper demonstrating yet another construction, the authors
conclude:
Few mathematical structures have undergone as many revisions or have been
presented in as many guises as the real numbers. Every generation re-examines the
reals in the light of its values and mathematical objectives. ([5] p. 278)

34
 A Appendix

A Appendix
The appendix only includes basic results and their immediate consequences of the
lecture courses Real Analysis [3] and Set Theory [12] as well as standard textbooks
for these courses.

Definition A.1. A set S equipped with two binary operations addition (+) and
multiplication () is called a ring if it satisfies the following axioms:

(A1) x, y, z S : (x + y) + z = x + (y + z) (associativity of addition)

(A2) x, y S : x + y = y + x (commutativity of addition)

There exists an element 0S S such that:

(A3) x S : x + 0 = x (existence of an additive identity)

(A4) x S y S : x + y = 0S (existence of additive inverses)

(M1) x, y, z S : (x y) z = x (y z) (associativity of multiplication)

(M2) x, y S : x y = y x (commutativity of multiplication)

There exists an element 1S S such that 1S 6= 0S and the following axiom

holds:

(M3) x S : x 1S = x (existence of a multiplicative identity)

(D) x, y, z S : x (y + z) = (x y) + (y z) (distributivity)

S is called a field if it additionally satisfies the axiom:

(M4) x S \ {0S } y S : x y = 1S (existence of multiplicative inverses)

S is called an ordered ring (or ordered field respectively) if there is a binary relation
(<) defined on S satisfying:

(O1) x, y, z S : [x < y = x + z < y + z]

(O2) x, y, z S : [(x < y 0 < z) = z x < z y]

(O3) x, y, z S : [(x < y y < z) = x < z] (transitivity of order)

(O4) x, y S exactly one of the following holds: x < y, x = y, or y < x (tri-

chotomy).

We call S an ordered semiring if it satisfies the axioms (A1) (A3), (M1) (M3),
(D) and (O1) (O4).

35
 A Appendix

Theorem A.2. (see [9] Theorem 8)

For any natural numbers m, n, k N, if m + k = n + k, then m = n.

Theorem A.3. (see [4] Theorem 4N)

For any natural numbers m, n, k N,

m + k < n + k if and only if m = n.

Theorem A.4. (see [9] Theorem 9)

For any two natural numbers m, n N, exactly one of the following three cases
holds:

1. m = n,

2. k N : m = n + k,

3. k N : n = m + k.

Theorem A.5. Every rational Cauchy sequence is bounded.

Theorem A.6. Suppose that (an ) is a rational Cauchy sequence such that an 6= 0
for all n N. Then (a1
n ) is a rational Cauchy sequence.

Theorem A.7 (Zorns Lemma, see [4] Theorem 6M). A set B is a chain if and only
if for any X, Y B, either X Y or Y X.
The following statement is equivalent to the Axiom of Choice:
Suppose that A is a set such that for every chain B A, we have
[
B A.

Then A contains a maximal element M , that is, M is not a subset of any other set
in A.

36
 References

References
[1] Martin Davis, Applied nonstandard analysis, Wiley, New York, 1977.

[2] Richard Dedekind, Gesammelte mathematische Werke, vol. 3, 1931, reprint,

Chelsea Publishing Co., New York, 1969.

[3] Richard Earl, Lecture notes, Mods Analysis MT 2011.

[4] Herbert B. Enderton, Elements of set theory, Academic Press, London, 1977.

[5] F. Faltin, N. Metropolis, B. Ross and G.-C. Rota, The Real Numbers as a
Wreath Product, Advances in Mathematics 16 (1975), no. 3, 278304.

[6] Otto Forster, Analysis 1 Differential- und Integralrechnung einer Vernder-

lichen, eighth edition, Vieweg, Wiesbaden, 2006.

[7] Derek Goldrei, Classic Set Theory For guided independent study, first edition,
Chapman & Hall, London, 1996.

[8] F. Mary Hart, Guide to Analysis, second edition, Palgrave Macmillan, Bas-
ingstoke, 2001.

[9] Edmund Landau, Grundlagen der Analysis: das Rechnen mit ganzen, ratio-
nalen, irrationalen, komplexen Zahlen, first edition, Akademische Verlagsge-
sellschaft, Leipzig, 1930.

[10] Detlef Laugwitz, Zahlen und Kontinuum Eine Einfhrung in die Infinitesi-
malmathematik, Wissenschaftliche Buchgesellschaft, Darmstadt, 1986.

[11] Edward Nelson, Internal set theory: a new approach to nonstandard analysis,
Bulletin of the American Mathematical Society 83 (1977), no. 6, 11651198.

[12] Jonathan Pila, Lecture notes, B1b Set Theory HT 2014.

[13] J. Sousa Pinto, Infinitesimal methods for mathematical analysis, Horwood,

Chichester, 2004.

[14] Curt Schmieden and Detlef Laugwitz, Eine Erweiterung der Infinitesimalrech-
nung, Mathematische Zeitschrift 69 (1958), no. 1, 139.

[15] Michael Spivak, Calculus, third edition, Publish or Perish, Houston, 1994.

[16] K. D. Stroyan and W. A. J. Luxemburg, Introduction to the Theory of Infinites-

imals, Academic Press, New York, 1976.

[17] Yevhen G. Zelenyuk, Ultrafilters and topologies on groups, De Gruyter, Berlin,

2011.

37