Gelfand spectra of C*-algebras in topos theory

Masters thesis, by Martijn Caspers
Date: August 2008
Author: Martijn Caspers, 0314080
Master in Mathematics, Radboud University Nijmegen
Supervisor: Prof. Dr. N.P. Landsman
Second reader: Prof. Dr. M. Gehrke
“Proof is the idol before whom the pure mathematician tortures himself.”
Sir Arthur Eddington (1882-1944)
Contents
Preface 6
1 Categories and topoi 8
1.1 Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.1.1 Categories and examples thereof . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.1.2 Special objects and morphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.1.3 Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.1.4 Natural transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2 Topoi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2.1 Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.2.2 Subobjects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.2.3 Powerobjects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.2.4 Topoi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.2.5 Rational numbers objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2 Topos theorems 28
2.1 The Yoneda lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2 Pullbacks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.3 Epi-monic factorization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4 Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.5 Special objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3 The algebra of subobjects 39
3.1 Subobjects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2 Lattices and subobjects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.3 The internal Heyting algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.4 ∃ and ∀ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4 The Mitchell-B´enabou language 58
4.1 The language . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.2 Some examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.3 Some further examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.4 Partial truth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.5 Internal categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4
5 The construction of the spectrum 70
5.1 The topos T (A) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.2 Free lattices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.2.1 Construction of free distributive lattices . . . . . . . . . . . . . . . . . . . . . . . . 73
5.2.2 Relations and the lattice L

¯
Asa
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.2.3 Intermezzo: representing the local lattice . . . . . . . . . . . . . . . . . . . . . . . 81
5.3 The spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.3.1 The map /
0
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.3.2 The morphism / . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.4 Final remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
6 Computations of spectra 90
6.1 The finite, commutative case: C
n
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
6.2 The finite case in general . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.3 Some remarks on C(X) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Further work 104
Index 107
Bibliography 107
Acknowledgements 108
5
Preface
The aim of this thesis is to compute Gelfand spectra of internal C*-algebras in a topos. The text
gives a brief introduction to topos theory, starting from the basic definitions and ending with the internal
Mitchell-B´enabou language of a topos. Once we developped enough topos theory we turn our attentention
to the generalized Gelfand spectrum in topos theory. We show how to compute these spectra and work
out some examples.
This thesis consists of two parts. The first four chapters contain a brief introduction into topos theory,
the underlying topos logic and the Mitchell-B´enabou language. This theory was developped in the second
half of the 20th century and a vast amount of literature is available, see for example the reference list. A
special note should be devoted to the book by Moerdijk and MacLane ‘Sheaves in Geometry and Logic:
A First Introduction to Topos Theory’, which is a very nice caleidoscope of subjects in topos theory. The
first chapter of this thesis contains precisely all definitions between a category and a topos. It introduces
the notation we use and gives the important examples for this thesis. A reader who is familiar with for
example [10] can skip this chapter, except for its last proof which becomes important. The second chapter
contains all proofs of theorems widely used in topos theory. This chapter has been added for completeness
and we will refer to these pages several times. In chapters three and four we introduce the underlying
logic of a topos and the Mitchell-B´enabou language respectively. These chapters show the power of
topos theory and the Mitchell-B´enabou language shows that a topos is a really nice generalization of set
theory. In chapter 4 some examples are given to emphasize this fact. The Mitchell-B´enabou language is
important for the definitions in the last chapters. However, this language has a wide amount of properties
which would take a whole book to describe (see [1] for example). Here we restrict ourselves to the central
example of monicity and prove an important fact which is used extensively in the last chapters: equality
in the Mitchell-B´enabou language is equivalent to equality of morphisms (proposition 4.3.6).
Whereas most theory in the first part of this thesis can be found in literature, the second part is largely
new. We turn our attention to an article written by Chris Heunen, Klaas Landsman and Bas Spitters
[7]. This article introduces a very specific topos in which the authors define the notion of the localic
spectrum of a C*-algebra, which is a generalization of the spectrum of a commutative C*-algebra as used
in the Gelfand-Naimark theorem. In terms of the Mitchell-B´enabou language they apply generalizations
of set theoretic constructions to their topos. In chapter 5 we introduce these constructions and make
the definitions as described in [7] more precise and exlicit. We show that all constructions are actually
well-defined and doing so, we prove some important theorems that allow us to compute the spectrum.
In chapter 6 it is time to use the tools developed in chapter 5 to compute spectra of C*-algebras. At the
start of writing this thesis, our main goal was to compute the localic spectrum of the 2 2-matrices over
the complex numbers. This result is presented in chapter 6. Furthermore, we show what the spectrum
of a finite-dimensional, commutative C*-algebra looks like and give the tools to compute the spectrum
of any finite-dimensional (non-commutative) C*-algebra as long as we can compute its commutative
6
subalgebras. The case of the 2 2 matrices then drops out as a special case. Finally, we dare to taste
a bit of the general unital, commutative case: the continuous complex valued functions on a compact
Hausdorff space. Here the spectrum becomes quite complicated and a nice representation has not yet
been found.
Concerning the prerequisites, we assume the reader has the level of an M. Sc. student. There is no need
to know anything about category theory, to which the first chapter gives an introduction. However, this
introduction is very brief and it would be more interesting to read [10] to see that category theory can
be applied to a wide variety of subjects in mathematics and really unifies these subjects, so that it can
be used in ‘algebraic geometry’, ‘algebraic topology’, et cetera. We do assume that the reader knows the
very basics of lattice theory and logic. Furthermore, we assume the reader is familiar with C*-algebras
and the Gelfand theorem.
7
Chapter 1
Categories and topoi
“Man muss immer generalisieren” - Carl Jacobi
This chapter introduces briefly the main concepts of category theory, starting from the definition of a
category and ending with the definition of a topos (plural: topoi or toposes), a special kind of category.
A reader who is familiar with category theory can skip the first paragraph, although one might find it
instructive to see what notation is used in the rest of the thesis. The main definitions will be clarified by
some examples, which are worked out to some extent. Most of this material can be found in [10]. For
more examples and applications of topoi, see for instance [5] and [11].
1.1 Categories
In this chapter we define categories, functors, natural transformations and some definitions which are
closely related to them. The most important examples for this thesis will be covered.
1.1.1 Categories and examples thereof
Definition 1.1.1. A category ( consists of the following data:
• A class of objects, denoted as Obj(().
• A class of morphisms, denoted as Mor(().
• A map called source : Mor(() → Obj(().
• A map called target : Mor(() → Obj(().
• A map called identity : Obj(() → Mor((). The image of an object A under the identity map is
usually denoted as Id
A
and is called the identity on A.
• A composition map called ◦ : P → Mor((), defined on
P = ¦(f, g) ∈ Mor(() Mor(()[target(g) = source(f)¦ .
P is called the set of composable pairs. We will write f ◦ g or simply fg instead of ◦(f, g).
These data satisfy the following rules:
8
1.1. CATEGORIES CHAPTER 1. CATEGORIES AND TOPOI
1. For A ∈ Obj((), source(Id
A
) = target(Id
A
) = A.
2. If f is a morphism with source(f) = A and target(f) = B, then f ◦ Id
A
= f = Id
B
◦ f.
3. For (f, g) and (g, h) composable pairs, (f ◦ g) ◦ h = f ◦ (g ◦ h).
Note that we assumed the objects and morphisms to form a class, which is not necessarily a set. This
is because the categories we wish to examine have a large number of objects, i.e. the objects do not
form a set anymore. If both the objects and morphisms form a set, the category is called small . All
other categories are called large necessarily. Going into great detail about the difference between small
and large categories is beyond the scope of this thesis, but one should at least notice this choice in the
definition.
The morphisms are sometimes depicted as arrows. In some literature they are even called ‘arrows’. The
source map and target map indicate from which object an arrow ‘departs’ and at which object the arrow
‘arrives’. In this thesis we use the notation f : A → B to say ‘f is a morphism such that source(f) = A
and target(f) = B’. We now state the notion of an isomorphism.
Definition 1.1.2. In a category ( a morphism f : A → B is called an isomorphism if there exists a
morphism g : B → A such that Id
A
= gf and Id
B
= fg. If f : A → B is an isomorphism, then A and
B are called isomorphic. Notation A · B.
The most instructive way of grasping the concept of a category is by seeing some examples:
1. The empty category (
0
, consisting of no objects and no morphisms.
2. The category (
1
, consisting of 1 object and 1 morphism, which is necessarily the identity on that
object.
3. The category (
2
,

>
,
consisting of two objects (the bullet and the star) and three morphisms (two identities and the
arrow in the above diagram). Although this is not a very complicated category, we will refer to this
example several times.
4. The category Set of all small sets. For this category one takes a universe U of sets and defines
the small sets as the sets in that universe. This construction is needed in order to make sure the
objects form a class. The morphisms are functions between sets. The other maps are evident.
5. The category Grp of (small) groups with the group homomorphisms as morphisms.
6. Let X be a topological space. Let P be the set of points of X and for p, q ∈ P, let Π(p, q) be the
set of homotopy classes of paths from p to q. Then this forms a category with the elements of P as
objects and the union ∪
(p,q)
Π(p, q) as the class of morphisms. Composition is the composition of
paths.
7. The category cCStar of commutative, unital C

-algebra’s with the (unital) C

-morphisms as mor-
phisms.
9
1.1. CATEGORIES CHAPTER 1. CATEGORIES AND TOPOI
8. A category T is called a preorder if for every two objects A, B ∈ Obj(T) there is at most one
morphism from A to B. Conversely, if P is some set with a partial order, P can be regarded as a
category if we take the elements of P as objects and define a morphism to be a pair (p, q) ∈ P P
such that p ≤ q. Define source((p, q)) = p, target((p, q)) = q and (q, r) ◦ (p, q) = (p, r). Note that
the transitivity of the partial order shows that this composition is a valid operation. With this
composition, (p, p) is the identity on p. Conversely, a preorder T regarded as a category gives rise
to a partial order by putting for any two objects p, q ∈ Obj(P), p ≤ q if and only if there is a
morphism from p to q. The elements of the partial order are the objects of T modulo isomorphism.
In that case T is a preorder and there exists a morphism between objects p and q, we denote it by

p,q
: p → q or simply ⊆: p → q.
9. If ( is a category, we can define the opposite category (
op
by:
• Obj((
op
) := Obj(().
• Mor((
op
) := Mor(().
• source
op
:= target. So the source map in the opposite category is the target map of the
original category.
• target
op
:= source.
• Id
op
A
:= Id
A
.
• If g and f are composable morphisms in (, i.e. target(f) = source(g), then the composition
in the opposite category is defined by:
f
op

op
g
op
:= (g ◦ f)
op
.
Remark that by switching the source and target maps in the opposite category this composition
is well-defined.
Examples 3, 4 and 8 are very important for this thesis. The other examples will be useful too.
An important concept will be the one of ‘Hom-sets’.
Definition 1.1.3. Let ( be a category. Let A, B ∈ Obj(C). We define the Hom-set of A and B as:
Hom
C
(A, B) = ¦f [ f ∈ Mor((), source(f) = A, target(f) = B¦ .
Do not be misled by this terminology. A Hom-set does not necessarily have to be a set (but it certainly
is a class).
1.1.2 Special objects and morphisms
We define two special objects and two special types of morphisms.
Definition 1.1.4. An initial object 0 in a category ( is an object such that for each object C in ( there
is a unique morphism from 0 to C. This morphism will be denoted by !.
Definition 1.1.5. A terminal object 1 in a category ( is an object such that for each object C in (
there is a unique morphism from C to 1. This morphism will be denoted by ! as well. This might lead to
some confusion, but we will always indicate which morphism is meant.
10
1.1. CATEGORIES CHAPTER 1. CATEGORIES AND TOPOI
For example, in Set the initial object is the empty set, since there is only one function from ∅ to any other
set. The terminal object is any one-point set, since any function to the one-point set must be constant.
Remark that if we have two initial objects 0 and 0

, then they must be isomorphic. By definition there
are unique morphisms ! : 0 → 0

and !

: 0

→ 0. The composite !

◦! is the unique morphism from 0 tot
itself, which must be the identity. For the same reason !◦!

= Id
0
. Hence 0 · 0

. In the same way one
finds that 1 is unique up to isomorphism.
Definition 1.1.6. A morphism m is called monic or ‘a monic’ if mf = mg implies f = g, i.e. it is
left-cancellable. In diagrams we will denote monics by →.
Definition 1.1.7. A morphism e is called epi or ‘an epi’ if fe = ge implies f = g, i.e. it is right-
cancellable. In diagrams we will denote epis by →→.
It may be checked that in Set monic is equivalent to injective and epi is equivalent to surjective. However,
this is not always true. For example, one can take a category which does not have a notion of injectivity,
like example 6. Another example is the embedding of Q in R as topological spaces with the euclidian
topology. The embedding is not surjective, but it is epi.
1.1.3 Functors
Next we define ‘morphisms between categories’, the so-called functors.
Definition 1.1.8. Let ( and T be two categories. A functor F from ( to T consists of:
• A function F
0
: Obj(() → Obj(T).
• A function F
1
: Mor(() → Mor(T).
These functions satisfy the following properties. For any f, g ∈ Mor(() such that f and g are composable,
the following relations holds:
• F
0
(source(f)) = source(F
1
(f)).
• F
0
(target(f)) = target(F
1
(f)).
• F
1
(f ◦ g) = F
1
(f) ◦ F
1
(g).
• F
1
(Id
A
) = Id
F
0
(A)
.
We usually write F : ( → T, saying F is a functor from ( to T. Remark that this notation is consistent
with the idea of viewing F as a morphism between categories. For convenience of notation we write
F(A) := F
0
(A) and F(f) := F
0
(f) if A is an object and f is a morphism of (.
There exists a notion of a contravariant functor. The definition is the same as definition 1.1.8 except for
the relations. They become:
• F
0
(source(f)) = target(F
1
(f)).
• F
0
(target(f)) = source(F
1
(f)).
• F
1
(f ◦ g) = F
1
(g) ◦ F
1
(f).
11
1.1. CATEGORIES CHAPTER 1. CATEGORIES AND TOPOI
• F
1
(Id
A
) = Id
F
0
(A)
. Remark that this relation is the same as for a functor.
The intuitive idea is that a contravariant functor reverses the direction of the morphisms, whence a
contravariant functor from a category ( to a category T can be regarded as a functor from the opposite
category (
op
to T. The functors of definition 1.1.8 are often called covariant functors.
Again we give some examples.
1. Let U : Grp → Set be the forgetful functor, forgetting the group structure of Grp. This is a
functor mapping the objects of Grp to Set.
2. If F : ( → T and G : T → c are functors, the composite G ◦ F is a functor. Here the composite
maps are defined as (G◦ F)
0
= G
0
◦ F
0
and (G◦ F)
1
= G
1
◦ F
1
.
3. Let ( be a category with small Hom-sets, i.e. the Hom-sets are sets. Let A be an object. Then
there is a functor Hom
C
(A, −) : ( → Set, defined on objects as:
Hom
C
(A, −) : B → Hom
C
(A, B),
and on the morphisms as:
Hom
C
(A, −) : f → f

,
where f

(g) = f ◦ g, i.e. composing with f on the left. This functor is called the covariant Yoneda
functor.
4. Let ( be a category with small Hom-sets. Let A be an object. Then there is a functor Hom
C
(−, A) :
(
op
→ Set, defined on objects as:
Hom
C
(−, A) : B → Hom(B, A),
and on the morphisms as:
Hom
C
(A, −) : f → f

,
where f

(g) = g ◦ f is composing with f on the right. This functor is called the contravariant
Yoneda functor.
5. Let ¸ and ( be categories. Let C be an object of (. Define the constant functor ∆
C
: ¸ → ( on
objects as ∆
0
C
: Obj(¸) → Obj(() : J → C and on morphisms as ∆
1
C
: Mor(¸) → Mor(() : f →
Id
C
.
The functors defined in 3, 4 and 5 will become important later on.
12
1.1. CATEGORIES CHAPTER 1. CATEGORIES AND TOPOI
1.1.4 Natural transformations
Next we define the morphisms between functors, the so-called natural transformations.
Definition 1.1.9. Let F and G be functors from a category ( to a category T. A natural transfor-
mation τ from F to G is a function
τ : Obj(() → Mor(T),
such that for any object A in (
τ(A) ∈ Hom
D
(F(A), G(A)),
and such that the following diagram commutes for every f ∈ Hom
C
(A, B) ⊆ Mor(():
F(A)
τ(A)
>
G(A)
F(B)
F(f)

τ(B)
>
G(B).
G(f)

Furthermore, τ is a natural isomorphism if there is a natural transformation σ from G to F such that
for any object A ∈ Obj((), one has τ(A) ◦σ(A) = Id
G(A)
and σ(A) ◦τ(A) = Id
F(A)
. We write F · G. If
we view the functors between ( and T as objects and the natural transformations as morphisms between
these objects, this notation coincides with definition 1.1.2.
We write τ : F → G when τ is a natural transformation from the functor F to the functor G. If σ : G → H
is another natural transformation, then there is a way of composing σ and τ. This is done point-wise, i.e.
for every object A, (σ ◦ τ)(A) = σ(A) ◦ τ(A). In some literature this is called the horizontal composition.
The next definition is the central object of this thesis. It is the notion of a functor category. The idea is
to regard functors as objects and natural transformation as the morphisms between functors.
Definition 1.1.10. Let ( and T be categories. We define the category T
C
to be the category whose
objects are the functors from ( to T and whose morphisms are the natural transformations between
the functors. For an object functor F, the identity Id
F
is the natural transformation assigning to every
object A ∈ Obj(() the identity Id
F(A)
. The composition of natural transformation is defined as point-wise
composition, i.e. for two (composable) natural transformations τ and σ, one has (τ ◦σ)(A) = τ(A)◦σ(A).
This category is referred to as a functor category.
If F is an object of T
C
and if f : A → B is a morphism in (, then we write
f
F
A,B
:= F(f).
When we write f
F
A,B
without specifying f,A and B, f is implicitly defined to be a morphism from A to B
in ( . We call the morphisms f
F
A,B
the intrinsic maps (of F).
One easily checks that this indeed defines a category. It is instructive to think about this definition a bit
more. As an example we study the category Set
C
2
. An object of this category is a functor F : (
2
→ Set.
This functor can be regarded as two sets, namely F
0
(•) and F
0
(), together with a morphism between
13
1.2. TOPOI CHAPTER 1. CATEGORIES AND TOPOI
those sets, namely F
1
(→). So the objects of Set
C
2
are the diagrams of the form ¦¦ → ¦¦, where ¦¦
are sets. In the literature, diagrams in a category ( are defined as functors from a category ¸ (which
‘has the form of the diagram’) to the category (. Intuitively the definition of a diagram is quite clear.
A morphism in Set
C
2
between objects F and G is a natural transformation τ : F → G, i.e. a pair of
morphisms such that the following diagram commutes:
F
0
(•)
F
1
(→)
>
F
0
()
G
0
(•)
τ(•)

G
1
(→)
>
G
0
().
τ()

So the morphisms in Set
C
2
are pairs of arrows forming commutative diagrams.
To complete this paragraph we give two more examples of categories. These categories turn out to be
topoi and are widely used in the literature.
• As a special case of definition 1.1.10 we take Set
P
, where T is a preorder. This category will
become important.
• Let ( be a category. Define the category of presheaves on ( to be Set
C
op
. This is the category
which has as objects the contravariant functors ( →Set. If F is an object of Set
C
op
and f : A → B
is a morphism in (, then the intrinsic morphism F(f) is denoted as f
F
B,A
.
• Let X be a topological space and let O(X) be its set of opens. Regard O(X) as a preorder and
hence as a category using the inclusion to define a partial order. Recall that if V, U ∈ O(X) and V
is a subset of U, we denote the corresponding morphism by ⊆: V → U. An object F in Set
O(X)
op
is called a sheaf is it satisfies the following properties:
– If U is an open set in X, if ¦V
i
¦ is an open covering of U and if p, q ∈ F(U) are elements such
that for each i, ⊆
F
U,V
i
(p) =⊆
F
U,V
i
(q), then p = q.
– If U is an open set in X, if ¦V
i
¦ is an open covering of U and if we have elements p
i
∈ F(V
i
)
such that for every i, j ⊆
F
V
i
,V
i
∩V
j
(p
i
) =⊆
F
V
j
,V
i
∩V
j
(p
j
), then there is an element p ∈ F(U) such
that for every i, ⊆
F
U,V
i
(p) = p
i
.
The morphisms between sheaves are just the morphisms of presheaves.
1.2 Topoi
A topos is a special kind of category. One might think of it as a generalization of the concept of a set.
The most important properties of a set can be translated into a categorical formulation. If one takes
these formulations as axioms, you could wonder what categories satisfy these axioms. These set-like
categories are called topoi. Formally topoi are categories in which objects have subobjects (analogous to
subsets), power objects (analogous to the power set of a set), and finite (co)limits (which allows us to do
set arithmetics).
14
1.2. TOPOI CHAPTER 1. CATEGORIES AND TOPOI
1.2.1 Limits
Here we introduce the notion of a categorical limit. The name ‘limit’ may let you think of an analytic
limit in terms of and δ. However, a categorical limit is totally different. Limits are objects that are
unique for some universal property. We state the precise definition now.
Definition 1.2.1. Let ¸ and ( be categories and let F : ¸ → ( be a functor. We define the limit of
F to be an object C in ( together with a natural transformation τ from ∆
C
(the constant functor) to F,
such that for every other object D in ( and natural transformation σ from ∆
D
to F there is a unique
morphism f : D → C such that for every object J in ¸, σ(J) = τ(J) ◦ f.
Remark that if both (C, τ) and (C

, τ

) are limits, there exist unique morphisms f : C → C

and
g : C

→ C such that τ(J) = τ

(J) ◦ f and τ

(J) = τ(J) ◦ g. Now τ(J) = τ(J) ◦ g ◦ f, so g ◦ f is the
unique morphism h such that τ(J) = τ(J) ◦ h. Obviously this h is the identity: g ◦ f = Id
C
. In the same
way f ◦ g = Id
C
. Hence limits are unique up to isomorphism, provided that they exist.
Sometimes we will omit the natural transformation τ, just saying C is the limit of F. We will denote
this as limF = C.
The definition of a limit is a very abstract one. However, often limits can be depicted as a diagram.
Remark that the functor F : ¸ → ( can be represented as a diagram ‘of the form ¸’ with on each
vertex an object of ( and between them morphisms of (. Then the limit is an object of ( together
with ‘commutative morphisms from the object to the diagram’ (the limiting cone). One should think of
limits of functors as ‘limits of diagrams’. This is best depicted in some important examples below. These
examples will be used extensively in this thesis.
1. Let ¸ be the category consisting of 2 objects and only the two identity morphisms:
J
1
• •
J
2
.
The limit of a functor F : ¸ → ( is called the product of F(J
1
) and F(J
2
), denoted as F(J
1
)F(J
2
).
For example, take ( =Set. Let A
1
and A
2
be two sets. Let F be the functor F : J
i
→ A
i
. Then the
following diagram shows the limiting product F(J
1
) F(J
2
) = A
1
A
2
is just the usual cartesian
product of sets along with its projections as the natural transformation.
A
1
B
............
∃!
>
>
A
1
A
2

τ(J
1
) := p
1
A
2
.

τ(J
2
) := p
2
>
In the future the ‘projections’ corresponding to the natural transformation of a product will always
be denoted by p or p
i
, where i is an index.
2. Let ¸ be the category:
J
1

>
>

J
2
.
15
1.2. TOPOI CHAPTER 1. CATEGORIES AND TOPOI
The limit of a functor F : ¸ → ( is called an equalizer. Again take for example ¸ = Set. Let A
1
and A
2
be sets and f, g both maps from A
1
to A
2
as indicated in this diagram:
A
1
f
>
g
>
A
2
.
Remark that this diagram represents a functor F : ¸ → (. Its limit is the set ¦a ∈ A
1
[ f(a) = g(a)¦,
as the following diagram obviously shows.
¦a ∈ A
1
[ f(a) = g(a)¦
inclusion
>
A
1
f
>
g
>
A
2
C
∃!

.
.
.
.
.
.
.
.
.
>
.
3. Let ¸ be the category:

J
1
J
3

>

J
2
.

The limit of a functor F : ¸ → ( is called a pullback. In Set the limit of the diagram
A
1
A
2
g
>
A
3
.
f

is the set
¦(a, b) ∈ A
1
A
3
[ f(a) = g(b)¦
with the projections on A
1
and A
3
as morphisms.
The next definition is the dual of the previous one. It is exactly the same except that the limiting cone
must be a class of morphisms from the diagram to the object (instead of in the other direction).
Definition 1.2.2. Let ¸ and ( be categories and let F : ¸ → ( be a functor. We define the colimit of
F to be an object C in ( and a natural transformation τ from F to ∆
C
such that for every other object
D of ( and natural transformation σ from F to ∆
D
there is a unique morphism f : C → D such that for
every object J in ¸, σ(J) = f ◦ τ(J).
In the same spirit as limits, colimits are unique up to isomorphism. We list some important examples of
colimits.
16
1.2. TOPOI CHAPTER 1. CATEGORIES AND TOPOI
1. Let ¸ be the category:
J
1
• •
J
2
.
The colimit of a functor F : ¸ → ( is called the coproduct of F(J
1
) and F(J
2
), denoted as
F(J
1
) + F(J
2
). For example, take ( =Set. Let A
1
and A
2
be two sets. Let F be the functor
F : J
i
→ A
i
. Then the following diagram shows the limiting product F(J
1
) + F(J
2
) = A
1

A
2
is
just the disjoint union of sets along with its inclusions as the natural transformation.
A
1
A
1

A
2
τ(J
1
):=i
A
1

.........
∃!
>
B
>
A
2
τ(J
2
):=i
A
2

>
.
2. Let ¸ be the category:
J
1

>
>

J
2
.
The colimit of a functor F : ¸ → ( is called a coequalizer. Again take for example ( = Set. Let
A
1
and A
2
be sets and f, g both maps from A
1
to A
2
as indicated in this diagram:
A
1
f
>
g
>
A
2
.
Remark that this diagram represents a functor F. Its colimit is the set
¦b [ b ∈ A
2
¦ / ∼,
where ∼ is the equivalence relation generated by the relation bRb

if and only if there exists an
a ∈ A
1
such that b = f(a) and b

= g(a). The following diagram then obviously shows that this set
is the colimit of the diagram.
A
1
f
>
g
>
A
2
>
¦[b] [ b ∈ B¦ / ∼
C.

.
.
.
.
.
.
.
.
.
>
A limit or colimit is said to be finite if the categorie ¸ has a finite number of objects and morphisms.
We say that for a given category ( all (co)limits exist if for every category ¸ and for every functor
F : ¸ → ( the (co)limit exists. In chapter 2 we will prove a theorem that says all (finite) limits exist
if all (finite) products and all equalizers exist. Dually, all (finite) colimits exist if all (finite) coproducts
and all coequalizers exist.
17
1.2. TOPOI CHAPTER 1. CATEGORIES AND TOPOI
1.2.2 Subobjects
Now we generalize the notion of a subset, which will be called a subobject. In Set every subset A of a
given set B can be described by its characteristic function (from B to ¦0, 1¦), i.e. the function that is
defined to be 1 at elements of A and 0 elsewhere. Conversely, every characteristic function describes a
subset of B. This correspondence can be generalized by the following definition.
Definition 1.2.3. A category ( with a terminal object is said to have a subobject classifier if there
exists a diagram
1

t
>
Ω.
with the following property. For any diagram
A

m
>
B.
there exist a unique morphism χ such that the following square is a pullback, i.e. A is the limit of the
bottom right part of the diagram.
A
!
>
1
B
m


...........
∃!χ
>
Ω.
t


Remark that in this definition m and t are monic.
Sometimes we briefly call Ω the subobject classifier, assuming that it is clear which morphism t is meant.
For the moment we call a diagram A → B a subobject of B which is usually abbreviated by saying
A is a subobject of B. The exact definition of a subobject will be postponed to chapter 3. For now
this definition will suffice. The map χ is called the characteristic morphism of the subobject. It turns
out there is a ‘bijection up to isomorphism’ between subobjects of B and maps from B to Ω (i.e. the
characteristic morphisms). For convenience of notation we write true
B
≡ t◦!, where ! is the unique map
from B to 1.
The details of subobjects and their classifiers will be postponed to chapter 3. Here we will only prove
the uniqueness of a subobject classifier and give some examples.
Proposition 1.2.4. If a category ( has a subobject classifier t : 1 → Ω, then it is unique up to isomor-
phism. That is, if t

: 1 → Ω

is another subobject classifier, then there is an isomorphism f such that
the following diagram commutes.
1
Id
1
>
1

t

f
>

.
t


18
1.2. TOPOI CHAPTER 1. CATEGORIES AND TOPOI
Proof:
1
<
Id
1
1
<
Id
1
1

t

<............
g

t


<...........
f
Ω.
t

The definition of the subobject classifier gives unique morphisms f and g that turn the above squares
into pullbacks. The pullback lemma (see chapter 2) proves that the outer rectangle is a pullback. Now
g ◦ f has the unique property of making the outer diagram a pullback. It is clear Id

has this property.
We conclude g ◦ f = Id

. In the same way f ◦ g = Id

, which proves the proposition.
In Set the terminal object is any one-point set. We claim the subobject classifier is defined by the
diagram:
¦1¦
¦0, 1¦ .
i


Here i is the inclusion. Remark that in a subobject diagram m : A → B the set A can be regarded
as a subset of B (since m is monic A is isomorphic to its image under m). So let A be a subset of B,
then we claim the characteristic morphism χ
A
is the unique morphism turning the following square into
a pullback.
A
!
>
¦1¦
B
m


.........
χ
A
>
¦0, 1¦ .
i


Indeed the diagram commutes. Furthermore, suppose we have a diagram:
C
A
!
>
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. >
¦1¦
h
>
B
m


χ
A
>
k
>
¦0, 1¦ ,
i


then we can show that the dotted morphism is just k, which maps to a smaller target set. To see this,
remark that for any c ∈ C, one has χ
A
k(c) = ih(c) = 1. So k(c) ∈ A, so we can restrict the target of k
to A.
19
1.2. TOPOI CHAPTER 1. CATEGORIES AND TOPOI
The map χ
A
is unique. Suppose b ∈ B, b ∈ A, then χ(b) = 1, since the diagram must commute. Suppose
b ∈ B, b ,∈ A, then χ(b) = 0, since A must be the pullback of the bottom right limiting diagram.
Another example is given by the category Set
C
2
of functors from (
2
to Set. The objects are just diagrams
in Set of the form A → B and the morphisms are the natural transformations. It is easy to see that a
natural transformation τ : F → G is monic if and only if both ‘components’ are injective, i.e.
F(•)

τ(•)
>
G(•)
F()


τ()
>
G().

both τ(•) and τ() are injective. We claim the right-hand side of the following cubic diagram is the
subobject classifier.
T
!
>
¦1¦
S
!
>
>
¦1¦
!
>
B
τ
2


....................................
χ
2
>
¦0, 1¦
i
1


A
τ
1


.............................
χ
1
>
h
>
¦0, 1, 2¦
i
2


f
>
.
The morphisms i
1
and i
2
are inclusions. The map f is defined as:
f : 0 → 0
1 → 1
2 → 1
.
In order to prove the claim, we may take S and T to be subsets of A and B respectively, like in the
previous example. Let χ
2
be the characteristic function of T and let χ
1
be defined as follows:
χ
1
: a → 0 if a ,∈ S and h(a) ,∈ T.
a → 1 if a ∈ S.
a → 2 if a ,∈ S and h(a) ∈ T.
This obviously makes the cube commutative. To check if this cube is a pullback we look at the following
diagram.
20
1.2. TOPOI CHAPTER 1. CATEGORIES AND TOPOI
Q
2
Q
1
>
T
!
>
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. >
¦1¦
h
2
>
S
!
>
>
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. >
¦1¦
!
>
h
1
>
B
τ
2


....................................
χ
2
>
k
2
>
¦0, 1¦
i
1


A
τ
1


.............................
χ
1
>
k
1
> h
>
¦0, 1, 2¦
i
2


f
>
.
For n ∈ ¦1, 2¦, for all q
n
∈ Q
n
notice that χ
n
k
n
(q
n
) = i
n
h
n
(q
n
) = 1. Hence k
n
(q
n
) is in S or T, depending
on wether n is 1 or 2, respectively. We conclude that k
n
factors (obviously uniquely) through τ
n
in such
a way that the diagram commutes. The map χ
n
is unique in making the cube a pullback. For all a ∈ A,
χ
1
(a) = 1 if and only if a ∈ S by demanding the square becomes a pullback. Also for all b ∈ B, χ
1
(b) = 1
if and only if b ∈ T. Hence the map χ
2
is determined. The commutativity of the bottom square now
determines the map χ
1
and we conclude that these χ’s are unique.
The elements of A that are mapped to 2 by χ
1
could be interpreted as “those that eventually will be part
of the subobject”. The condition that the bottom square has to commute forces the front bottom right
object (Ω(•)) to be a larger set than ¦0, 1¦. The question ‘is a certain element part of the subobject?’
cannot simply be answered by yes or no. The answer has become more subtle, varying between no, yes
and not yet: after applying h it will be.
The last example we examine, is a generalization of the previous ones and will be important for the most
important topos in this thesis. Let ( be an arbitrary category. We can prove that Set
C
has a subobject
classifier. To see this, we introduce the notion of a cosieve.
Definition 1.2.5. Let ( be a category and let C be an object in (. A cosieve on C is a subset S of
¦f ∈ Mor(() [ source(f) = C¦ ,
which has the following property: if f ∈ S, then hf ∈ S for all h which have the same source as the target
of f.
Now let
Ω(C) = ¦S[S is a cosieve on C¦ .
For a morphism f : C → D in (, recall that we defined f

C,D
:= Ω(f) and define
f

C,D
: Ω(C) → Ω(D) : S → ¦h ∈ Mor(() [ source(f) = D, hf ∈ S¦
and note that if S is a cosieve on C, then f(S) is indeed a cosieve on D. The cosieve on an object C
containing all morphisms with source C is called the full cosieve (on C). Remark that a cosieve on C is
full if and only if it contains the identity on C.
21
1.2. TOPOI CHAPTER 1. CATEGORIES AND TOPOI
The terminal object in Set
C
is obviously the functor ∆
{1}
sending each object C in ( to the one-point
set ¦1¦. Define a morphism (natural transformation) t : 1 → Ω by
t(C) : 1(C) → Ω(C) : 1 → the full cosieve on C.
We can prove that the diagram
1
t
>
Ω.
is the subobject classifier. Doing this is just a generalization of the proof that Set
C
2
has a subobject
classifier. Again, a morphism in Set
C
(i.e. a natural transformation) is monic if and only if all components
are injective. Again if F is an object in Set
C
and G is a subobject of F, we may take G(C) to be a subset
of F(C) for all objects C. The morphisms χ(C) that form the characteristic natural transformation
between F and Ω are defined as follows:
χ(C) : F(C) → Ω(C) : a
c

_
f : C → D[D ∈ Obj((), f
F
C,D
(a
c
) ∈ G(D)
_
.
As in the example of Set
C
2
, this combines to a commutative diagram of natural transformations. The
diagram forms a pullback and χ is proved to be unique in having this property. The conclusion is that
Set
C
has a subobject classifier. As a special case Set
C
2
has a subobject classifier and we compute:
1. Ω(•) = ¦S [ S is a cosieve on •¦ consists of the cosieves
0 := ∅;
1 := ¦• → ¦ ;
2 := ¦Id

, • → ¦ .
2. Ω() = ¦S [ S is a cosieve on ¦ consists of the cosieves
0 := ∅;
1 := ¦Id

¦ .
3. The map f

•,
indeed is defined as:
f

•,
: 0 → 0;
1 → 1;
2 → 1.
As another example we prove that Grp does not have a subobject classifier. Consider the pullback:
A
!
>
1 = ¦0¦
B


χ
>
Ω.
t


Suppose it does have a subobject classifier. Then in order to make the above diagram a pullback A must
be isomorphic to the kernel of χ. However, if the image of A under the left monic is not a normal subgroup
22
1.2. TOPOI CHAPTER 1. CATEGORIES AND TOPOI
of B, this certainly is not the case. Even if we consider the category of abelian groups Ab, in which
every subgroup is normal, there is no subobject classifier. Because then the image of χ is isomorphic to
the group B modulo A. Since every group is the quotient of itself by the trivial group, every group must
be a subgroup of Ω, which is impossible.
1.2.3 Powerobjects
Now we like to generalize the notion of the powerset of a set. This is done using the definition of adjoint
functors and the notion of cartesian closed categories.
Definition 1.2.6. Let ( and T be categories and let F : ( → T and G : T → ( be functors. We say that
F and G are adjoint functors if there exists a natural isomorphism
τ : Hom
D
(F(−), −) → Hom
C
(−, G(−)).
Remark that Hom
D
(F(−), −) and Hom
C
(−, G(−)) can be regarded as a functor from (
op
T to Set.
In this case F is called a left adjoint, and G is called a right adjoint.
If a functor F has a right adjoint G, then it is unique up to isomorphism. That is, if G

is a right adjoint
for F too, then there is a natural isomorphism from G to G

. Conversely, if G has a left adjoint, it is
unique up to isomorphism, too. A proof can be found in [10].
Adjoint functors are very important in category theory. The most useful result is theorem 2.4.1, showing
that if a functor has a left adjoint, then the functor preserves all limits. Similarly, if a functor has a right
adjoint, then it preserves all colimits. The exact meaning of this formulation is stated in theorem 2.4.1.
Definition 1.2.7. A category ( is said to be cartesian closed if it has finite products and if for any
given object B the functor A → A B (with the obvious map on morphisms) has a right adjoint, called

B
. That is, there is a natural isomorphism
Hom
C
(AB, C) · Hom
C
(A, C
B
). (1.1)
If h : C → C

, we denote the morphism after applying −
B
by h
B
: C
B
→ C
B
.
Note that we require naturality in A and C, not necessarily in B.
The category Set is cartesian closed and the above notation reduces to the familiar meaning on sets. I.e.
for a map f : AB → C:
τ : Hom
Set
(AB, C) → Hom
Set
(A, C
B
)
((a, b) → f(a, b)) → (a → (b → f(a, b)))
is a natural isomorphism.
Next, we prove some properties using of the Yoneda lemma, which is stated in chapter 2 (lemma 2.1.1).
Proposition 1.2.8. The following properties hold in any cartesian closed category (:
1. If both C
B
and (C
B
)

have the property of (1.1), then C
B
· (C
B
)

;
2. C
1
· C;
23
1.2. TOPOI CHAPTER 1. CATEGORIES AND TOPOI
3. C
B
D
B
· (C D)
B
.
Proof: For (1) we have
Hom
C
(A, C
B
) · Hom
C
(AB, C)
· Hom
C
(A, (C
B
)

).
Now, property (1) follows from the Yoneda lemma 2.1.1. The other properties follow from this lemma
by:
Hom
C
(A, C
1
) · Hom
C
(A1, C)
· Hom
C
(A, C)
;
Hom
C
(A, C
B
D
B
) · Hom
C
(A, C
B
) Hom
C
(A, D
B
)
· Hom
C
(AB, C) Hom
C
(AB, D)
· Hom
C
(AB, C D)
· Hom
C
(A, (C D)
B
).
So where is the power object? If a category is both cartesian closed and has a subobject classifier there
is a correspondence between subobjects of an object A and morphisms χ : A → Ω. Since the category is
cartesian closed such a morphism corresponds to a morphism ˆ χ : 1 → Ω
A
. Now in Set these morphisms
correspond to points in Ω
A
since 1 = ¦1¦. Remark that Ω
A
= ¦0, 1¦
A
is simply the powerset of A. Hence
we have found our generalization of the powerset of A: it is Ω
A
.
1.2.4 Topoi
We conclude this paragraph by stating the definition of a topos. There are many ways to define a topos,
that of course are all equivalent. Here we state a very categorical definition, which is based on the
previous paragraphs and is most suitable for this thesis.
Definition 1.2.9. A category T is a topos if:
1. it has all finite limits and colimits;
2. it is cartesian closed;
3. it has a subobject classifier.
We have already seen that Set is a topos. Also, Set
C
turns out to be a topos. For Set
C
we have proved
that it has a subobject classifier. It has all finite limits, since limits can be taken point-wise (see theorem
2.4.5). Here we will prove it is cartesian closed too. The reason we do not postpone this proof to the
second chapter is that we need the details of the proof for further use. All proofs in the second chapter
will not be of any further interest, although their results will be. The proof below is very technical in
nature and seems not very elegant at all. It shows how abstract category theory can be. However, after
chapter 2 we skip the technical details more and more to avoid long pages of computations.
Proposition 1.2.10. The category Set
C
is cartesian closed.
Proof: First remark that theorem 2.4.5 below tells us that this category has finite products. Hence it
remains to show that if F, G and H are functors from ( to Set we must define a functor H
G
such that
there is a natural isomorphism
24
1.2. TOPOI CHAPTER 1. CATEGORIES AND TOPOI
τ : Hom
Set
C (F G, H) · Hom
Set
C (F, H
G
).
In what follows, A and B are objects of (. One might expect H
G
= Hom
Set
C (G(−), H(−)). However,
there is no way we can regard Hom
Set
C (G(−), H(−)) as a functor from ( to Set (although it can be
regarded as a functor from (
op
( to Set). We show that the correct definition is
H
G
(A) = Hom
Set
C (Hom
C
(A, −) G(−), H(−)),
the set of natural transformations between the functors Hom(A, −) G and H. Note that the outer
Hom is taken in Set
C
while the inner Hom is taken in (. Since H
G
must be a functor, we must specify
the function on the morphisms f : A → B of (.
H
G
(f) : Hom
Set
C (Hom
C
(A, −) G(−), H(−)) → Hom
Set
C (Hom
C
(B, −) G(−), H(−));
σ → (f

Id
G
)

σ.
Remark that for composable morphisms f and g in (,
H
G
(g) ◦ H
G
(f) = (g

Id
G
)

◦ (f

Id
G
)

= ((f

Id) ◦ (g

Id
G
))

= ((f

◦ g

) Id
G
)

= (((g ◦ f)

) Id
G
)

= H
G
(g ◦ f);
H
G
(Id
A
) = (Id

A
Id
G
)

= Id
H
G
(A)
.
Hence H
G
is a functor, since other properties are trivial. We conclude H
G
is an object in Set
C
.
Next we define the natural isomorphism τ.
τ : Hom
Set
C (F G, H) → Hom
Set
C (F, H
G
);
σ → ˆ σ;
ˆ σ(A) : F(A) → H
G
(A) = Hom
Set
C (Hom
C
(A, −) G, H);
a → ˆ σ(A)(a);
((ˆ σ(A))(a))(B) : Hom
C
(A, B) G(B) → H(B);
(g, x) → σ(B)(g
F
A,B
(a), x).
First we check that ˆ σ is indeed a natural transformation. This is done by checking wether the following
diagram commutes for every f : A → B in (.
F(A)
ˆ σ(A)
>
Hom
Set
C (Hom
C
(A, −) G, H)
F(B)
f
F
A,B

ˆ
σ(B)
>
Hom
Set
C (Hom
C
(B, −) G, H)
(f

×Id
G
)


. (1.2)
25
1.2. TOPOI CHAPTER 1. CATEGORIES AND TOPOI
This is done by the following computation.
(((f

Id
G
)

((ˆ σ(A))(a)))(C))(g, x) = ((((ˆ σ(A))(a))(f

Id
G
))(C))(g, x)
= (((ˆ σ(A))(a))(C)(f

Id
G
)(C))(g, x)
= ((ˆ σ(A))(a))(C)(gf, x)
= σ(C)((gf)
F
A,C
, x)
= σ(C)(g
F
B,C
f
F
A,B
a, x)
= (((ˆ σ(B))(f
F
A,B
(a)))(C))(g, x).
We must check if the morphism τ is natural in F and H. So let η : F

→ F and µ : H → H

be morphisms
in Set
C
. Then we must check if the following diagram commutes:
Hom
Set
C (F G, H)
τ
>
Hom
Set
C (F, H
G
)
Hom
Set
C (F

G, H

)
(η×Id
G
)

µ∗

τ
>
Hom
Set
C (F

, H
G
).
ν

µ∗∗

This is indeed the case, as the following computation shows:
(((τ(µσ(ν Id
G
))(A))(a))(B))(g, x) = ((µσ(ν Id
G
))(B))(g
F

A,B
a, x)
= µ(B)(σ(B)(ν(B)g
F

A,B
a, x))
= µ(B)(σ(B)(g
F
A,B
ν(A)a, x))
= µ(B)(((ˆ σ(A))(ν(A)(a))(B))(g, x))
= ((((µ

ˆ σν)(A))(a))(B))(g, x).
The next step is to define the inverse of τ. This is done by
τ
−1
: Hom
Set
C (F, H
G
) → Hom
Set
C (F G, H);
σ → ˇ σ;
ˇ σ(A) : F(A) G(A) → H(A);
(a, b) → σ(A)(a)(A)(Id
A
, b).
It follows from a similar computation that τ
−1
indeed defines a natural transformation and that ˇ σ is a
natural transformation. The last thing to check is whether τ
−1
is indeed the inverse of τ:
(
ˇ
ˆ σ(A))(a, b) = (((ˆ σ(A))(a))(A))(Id
A
, b);
= (σ(A))(a, b),
(((
ˆ
ˇ σ(A))(a))(B))(g, b) = (ˇ σ(B))g
F
A,B
(a), x) =
= (((σ(B))(g
F
A,B
(a)))(B))(Id
B
, b) =
= (((σ(A))(a))(B))(g, b),
where the last equality follows from diagram 1.2.
In particular it now follows that Set
C
is a topos for any category (.
26
1.2. TOPOI CHAPTER 1. CATEGORIES AND TOPOI
1.2.5 Rational numbers objects
In this paragraph we will try find a generalization of the rational numbers Q in Set to a topos T . This
can not be done for every topos, but in the case of Set, Set
C
and Set
C
op
, with ( a category, it can.
To do so one requires the topos T to have a natural numbers object, i.e. an object N together with
morphisms 0 and s such that for every diagram
1
0
>
N
s
>
N
[[
1
0

>
X
i

.
.
.
.
.
.
.
.
.
.
s

>
Y,
i

.
.
.
.
.
.
.
.
.
.
there exists a unique morphism i making the diagram commutative. This object plays a similar role
as the natural numbers in Set. Not every topos has a natural numbers object, but Set
C
, Set
C
op
and
(obviously) Set do have one. In Set
C
and Set
C
op
the natural numbers object is the constant functor

N
. Notice that there is a morphism + : N N → N, which in the case of Set
C
is defined by letting
+(C) : N(C) N(C) → N(C) be addition in Set. We may use this to define the integral numbers object
Z as the coequalizer of the diagram:
E
(p
1
f,p
2
g)
>
(p
2
g,p
1
f)
>
NN
>
Z.
Here E is the upper left object in the following pullback:
E
f
>
NN
NN
g

+
>
N.
+

E is the object such that E(C) consists of 4-tuples (m, n, m

, n

) such that m+n = m

+n

. In the case
of Set
C
and Set
C
op
a short calculation gives Z = ∆
Z
. Note that this object comes with a morphism
m : Z N → Z that is locally the multiplication operator, i.e. m(C) : Z(C) N(C) → Z(C) is
multiplication in Set. In a similar way one defines the rational numbers object Q, which in case of Set
C
and Set
C
op
becomes
Q = ∆
Q
.
In section 4.5 we will argue why we would call this a generalization of the set-theoretic rational numbers
Q. In chapter 5 we will need the rational numbers object in order to define a generalized norm in topos
theory.
27
Chapter 2
Topos theorems
“I mean the word proof not in the sense of the lawyers, who set two half proofs equal to a whole one, but
in the sense of a mathematician, where half proof = 0, and it is demanded for proof that every doubt
becomes impossible.” - Karl Friedrich Gauss
Throughout this chapter T will be a topos and any morphism or object will be taken within this topos
unless specified otherwise. The aim of this chapter is to prove some often used theorems in topos theory.
Some of these theorems may hold in a category which has only some of the properties of a topos, but for
convenience we prove the statements within a topos T . Most of this material can be found in [10] and
[11], however sometimes without proof.
2.1 The Yoneda lemma
If A and B are objects of a category ( one might wonder what the natural transformations between
the Yoneda functors Hom
C
(A, −) and Hom
C
(B, −) look like. The Yoneda lemma and its immediate
corollary give the answer to this question.
Lemma 2.1.1. (Yoneda lemma) If F : ( →Set is a functor and A is an object of the category (, then
there is a bijection of sets
k : Hom
Set
C (Hom
C
(A, −), F) → F(A),
which sends each natural transformation τ : Hom
C
(A, −) → F to (τ(A))(Id
A
).
Proof: Let f : A → B be a morphism in (. Then the following diagram shows that k(τ) determines the
natural transformation τ, i.e. if you know what (τ(A))(Id
A
) is, τ is completely determined. Hence k is
injective. It is surjective too, since for every x ∈ F(A) one can define τ by putting (τ(A))(Id
A
) = x.
Hom
C
(A, A)
τ(A)
>
F(A)
Hom
C
(A, B)
f∗

τ(B)
>
F(B).

f
F
A,B
28
2.2. PULLBACKS CHAPTER 2. TOPOS THEOREMS
Corollary 2.1.2. Let A, B be objects of a category (, then every natural transformation from Hom
C
(A, −)
to Hom
C
(B, −) is of the form f

for a map f ∈ Hom
C
(B, A).
Proof: This is the special case of lemma 2.1.1 with F = Hom
C
(B, −).
2.2 Pullbacks
Here we will examine two properties of pullbacks. Together with products and equalizers the pullback is
the most important type of limit in topos theory.
Theorem 2.2.1. Let the diagram below be a pullback square:
A
n
>
B
C
f

m
>
D.
g

If m is monic, then n is monic.
Proof: Suppose we have morphisms h, k : E → A, such that nh = nk. Then m(fk) = g(nk) = g(nh) =
m(fh), so fk = fh. By the universal property of a pullback there is a unique morphism from l : E → A
such that nh = nl = nk and fh = fl = fk. The uniqueness of the pullback proves h = k.
The next lemma will be used extensively in the next chapters. In the literature one refers to this lemma
as the ‘pullback lemma’ or simply ‘PBL’.
Lemma 2.2.2. (Pullback lemma, PBL)
A
>
B
>
C
D

>
E

>
F.

1. If in the above diagram both squares are pullbacks, then the outer rectangle is a pullback.
2. If the outer rectangle is a pullback and the right square is a pullback, then the left square is a
pullback.
Proof:
1.
G
A
>
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. >
B
>
.................................... >
C
>
D

>
>
E

>
F.

Suppose we have morphisms from G to C and D as in the diagram. The pullback property of the
right and left square gives the unique dotted morphisms to B and A respectively.
29
2.3. EPI-MONIC FACTORIZATION CHAPTER 2. TOPOS THEOREMS
2.
G
A
>
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. >
B
>
>
C
>
D

>
>
E

>
F.

Suppose we have morphisms from G to B and D as indicated in the diagram. This gives a morphism
from G to C such that the diagram commutes. Since the outer rectangle is a pullback, there is a
unique morphism from G to A such that the diagram commutes. The morphism G → D equals
the factorization G → A → D. Remark that the morphism G → A → B is a morphism such that
G → A → B → C equals G → C and G → A → B → E equals G → D → E. Since the right
square is a pullback, such a morphism is unique and hence it equals the morphism G → B.
2.3 Epi-monic factorization
In Set every morphism can be written as a composition of an epi with a monic. The epi is the map to
the direct image of the morphism and the monic is the inclusion in the target set of the morphism. This
can be generalized to a topos. This is the first result that shows topoi look like sets.
Definition 2.3.1. In a topos T , let f : A → B be a morphism. A monic m is called the image of f if
f factors through m and whenever f factors through a monic m

, m factors through m

. We denote the
source of m by Im(f), thus f : A → Im(f) →
m
B.
This definition leads to an important theorem about (unique) epi-monic factorization.
Theorem 2.3.2. In a topos T , let f : A → B be a morphism. Then f = me where m is an image of f
(thus monic) and e is epi.
Proof: the proof makes use of two facts:
1. For every morphism f : A → B there exists a so-called cokernel pair of f. This is a pair of
morphisms s, t from B to an object C which has the universal property sf = tf, i.e. if there is
another pair u, v : B → C

such that uf = vf, then there is a unique morphism w such that the
following diagram commutes:
A
f
>
B
s
>
t
>
C
C

.
w

.
.
.
.
.
.
.
.
.
v
>
u
>
The proof of this fact is by taking s, t to be the morphisms obtained from the colimit of
30
2.3. EPI-MONIC FACTORIZATION CHAPTER 2. TOPOS THEOREMS
B
f
A
>
f
B.
>
This colimit is called a pushout.
2. Every monic m : C → D is the equalizer of two morphisms. It follows from the definition of a
subobject classifier that m equalizes the characteristic morphism of m and the map true
D
.
A
f
>
A
[[ [[
A
...........
e
>
M
...........
m
>
B
s
>
t
>
C
[[ [[
A
g
>
N
h
>
B
u
>
v
>
D.
w

Take s, t : B → C the cokernel pair of f and let m be the equalizer of s, t. Since f equalizes s, t, it factors
through m, say f = me. Now take another factorization f = hg. By (2) h is an equalizer, say of u, v.
Since h equalizes u, v, so does f. By the uniqueness property of a cokernel pair there is a morphism w
making the diagram above commutative. Now note that um = wsm = wtm = vm, hence m equalizes u
and v, hence m factors through h. This proves that m is the image of f.
Recall that an equalizer is always monic. It remains to prove that e is epi. Suppose m is an isomorphism.
Then s = t, since m is the equalizer of s and t. Now since s, s is the cokernel pair of f, e must be epi.
Now suppose that we have f = me without m being an isomorphism. Take an epi-monic factorization of
e, e = m

e

.
A
e

>
M

m

>
M
m
>
B.
We find f factors through mm

and so does the image m, say m = mm

q for some q. By m being monic
this implies 1 = m

q. This gives m

qm

= m

and since m

is monic 1 = qm

. So m

is an isomorphism
and e

is epi. Hence e = m

e

is epi.
In the previous theorem we proved the existence of an epi-monic factorization. The factorization turns
out to be unique up to isomorphism as the following theorem and its corollary shows.
Theorem 2.3.3. If f = me and f

= m

e

with m, m

monic and e, e

epi, then for morphisms r, t in
the diagram below, there is a unique map s such that the diagram commutes.
31
2.4. LIMITS CHAPTER 2. TOPOS THEOREMS
A
e
>
C
m
>
B
P
>
A

r

e

>
C

s

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
m

>
<
B

.
t

Proof: First suppose m is the image of f. Let P be the pullback of t along m

, like in the diagram.
Because m

is monic P → B is monic (theorem 2.2.1). Since tf = m

e

r, f factors through P → B by the
pullback property. Since m was assumed to be the image of f and P → B is monic, m factors through
P → B. Hence there is a map s such that the right square commutes. Since m

is monic, this s is unique.
Since m

se = tme = m

e

r it follows from the fact that m

is monic that se = e

r and so the left square
commutes.
Now suppose f = ˜ m˜ e is an arbitrary epi-monic factorization of f. Then look at:
A
e
>
C
m
>
B
[[ [[
A

˜ e
>
C

s

˜ m
>
B

.
s is monic, since ˜ ms = m. s is epi since se = ˜ e. Now as in fact (1) of theorem 2.3.2, s is the equalizer of
some cokernel pair. Because it is epi, the cokernel pair must be a pair of equal morphisms. But if s is
the equalizer of two identical morphisms, s must be an isomorphism. Hence ˜ m is the image of f.
Corollary 2.3.4. An epi-monic factorization of a morphism f is unique up to isomorphism.
Proof: This follows straight from theorem 2.3.3 taking f

= f, r = Id
A
and t = Id
B
.
2.4 Limits
In chapter 1 we introduced limits and stated some of their properties. We promised to give the proofs in
this chapter. First we take care of the preservation of limits under adjoint functors.
Theorem 2.4.1. Let F : ¸ → ( and G : ( → T be functors. Let H : T → ( be a left adjoint for G.
Suppose limF = C. Then limGF = GC.
Proof: Call the natural isomorphism obtained from the adjunction
ϕ : Hom
D
(D, GC) ˜ →Hom
C
(HD, C).
Let D be an object in T and τ a natural transformation from ∆
D
to GF. Compare the following diagrams
in the category T and ( respectively:
32
2.4. LIMITS CHAPTER 2. TOPOS THEOREMS
GFj
α
D
......... >
τα
>
GC
Gσα
>
GFj
β
.
GFf
αβ


β
>
τ
β
>
Fj
α
HD
.........
∃!k
>
ϕτα
>
C
σα
>
Fj
β
.
Ff
αβ

σ
β
> ϕτα
>
.
By adjunction the first diagram commutes if and only if the second diagram commutes. In the second
diagram there exist a unique morphism k such that the diagram commutes. By adjunction this gives the
unique (!) (dotted) morphism ϕ
−1
(k) making the first diagram commutative.
Of course, there is a dual version of the previous theorem in terms of colimits and right adjoints.
Theorem 2.4.2. Let F : ¸ → ( and G : ( → T be functors. Let H : T → ( be a right adjoint for G.
Suppose colimF = C. Then colimGF = GC.
Proof: The proof is obtained in exactly the same way as theorem 2.4.1 by comparing the following
diagrams.
GFj
α
GC
......... >
Gσα >
D
τα
>

β
GFj
β
GFf
αβ

τ
β
>
>
.
Fj
α
C
.........
∃!k
>
σα
>
HD
ϕτα
>
σ
β
Fj
β
Ff
αβ

ϕτ
β
>
>
.
One of the properties of a category in order to be a topos is having all finite limits and colomits. In
general it is very hard to check if a given category has all finite (co)limits. The following theorem will
make life a lot easier; one only has to check wether a category has (co)equalizers and finite (co)products.
Theorem 2.4.3. Let ¸ and ( be categories. If ( has equalizers and if ( has all products of length
card(Obj(¸)) and card(Mor(¸)) then the limit of every functor F : ¸ → ( exists.
Proof: In the following diagram A, B and C are objects in ¸ and the product

A
is taken over all
objects of ¸. f and g are morphisms of ¸.

f
is taken over all morphisms of ¸. p denotes the projection
on the object in its index.
F(target(g)) = F(target(g)) F(B)

f
F(target(f))
p
target(g)

<
h
<
k

A
F(A)
p
target(g)

<
e
p
B
>
C

F(target(g))
p
target(g)

<
F(g)
F(source(g))
p
source(g)

.
33
2.4. LIMITS CHAPTER 2. TOPOS THEOREMS
By definition of the left product in the diagram, there is a morphism h such that the upper square
commutes for every g and there is a morphism k such that the lower square commutes for every g. Let
e be the equalizer of h and k. Write τ(B) := p
B
e. Then for a morphism g : B → B

of ¸:
F(g) ◦ τ(B) = F(g)p
B
e
= p
B
ke
= p
B
he
= p
B
e
= τ(B

).
So the morphisms τ(B) combine into a natural transformation from ∆
C
to F.
Now suppose we have an object D together with a natural transformation σ from ∆
D
to F. Then
the morphisms σ(A) combine to a morphsim m : D →

A
F(A). The property that σ is a natural
transformation from ∆
D
to F is equivalent to the equation hm = km. Hence m = en for a morphism
n : D → C by the unique property of an equalizer. Then σ factors through τ, i.e. σ(B) = τ(B) ◦ n.
So we see that if T has all finite products and equalizers, then every finite limit exist since for these
limits the cardinality of Obj(¸) and Mor(¸) is finite. The same argument can be used for finite colimits
as the next theorem shows.
Theorem 2.4.4. Let ¸ and ( be categories. If ( has equalizers and if ( has all products of length
card(Obj(¸)) and card(Mor(¸)), then the limit of every functor F : ¸ → ( exist.
Proof: The theorem is proved by the following diagram:
F(source(g)) = F(source(g)) F(B)

f
F(source(f))
i
source(g)

h
>
k
>

A
F(A)
i
source(g)

e
>
i
B
<
C

F(source(g))
i
source(g)∧
F(g)
>
F(target(g))
i
target(g) ∧
.
The next theorem is might look complicated at first sight. It is not! The theorem is a simple result of a
diagram chase, which is worked out for a simple but non-trivial case.
Theorem 2.4.5. Let (, T and ¸ be categories (not necessarily topoi). Let F : ¸ → (
D
and let
G
D
: (
D
→ (;
F → F(D);
τ → τ(D).
Note that this defines a functor. Then if (lim G
D
F) exists in ( for all objects D of T, then lim F exists
and one has
(lim F)(D) = (lim G
D
F). (2.1)
34
2.4. LIMITS CHAPTER 2. TOPOS THEOREMS
Let σ : ∆
limF
→ F be the natural transformation implied by the limit of F and let τ
D
: ∆
limG
D
F
→ G
D
F
be the natural transformation implied by the limit of G
D
F. Then we have
σ(D) = τ
D
.
Furthermore, the intrinsic morphisms of (lim F) are determined by this equality.
Proof: The proof is a simple diagram chase. To keep things clear, we prove the theorem for ¸ = (
2
. In
this case the limit of the functor F can be represented as the following diagram:
K
>
H
G.

f
>
Where H := F(•), G := F(), f := F(→) and K = lim F. Let g : D → D

be a morphism of T. Then
the uniqueness of the point-wise limit of G
D
F = K(D

) shows that there is a unique dotted morphism
making the diagram below commutative:
K(D

)
τ
D
(•)
>
H(D

)
g
K
D,D

K(D)
τ
D
(•)
>
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
>
H(D)
g
H
D,D

>
G(D

)
f(D

)

τ
D
()
>
G(D)
f(D)

τ
D
()
> g
G
D,D

>
.
By a simple diagram chase, we have (Id
D
)
K
D,D
= Id
K(D)
and if h : D

→ D

is a morphism in T,
then h
K
D

,D

g
K
D,D

= (hg)
K
D,D

. This defines the intrinsic maps of K and determines K as a functor,
hence an object of (
D
. The above diagram shows that the morphisms τ
D
indeed combine into a natural
transformation σ : ∆
limF
→ F.
Suppose L is another object that has a natural transformation ρ : ∆
L
→ F. We have proved the theorem
if ρ factors through σ.
35
2.5. SPECIAL OBJECTS CHAPTER 2. TOPOS THEOREMS
L(D

)
β(D

)
>
K(D

)
τ
D
(•)
>
H(D

)
g
L
D,D

g
K
D,D

L(D)
β(D)
>
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
>
K(D)
τ
D
(•)
>
>
H(D)
g
H
D,D

>
G(D

)
f(D

)

τ
D
()
>
G(D)
f(D)

τ
D
()
> g
G
D,D

>
.
Since we took the limit point-wise, there are unique (!) morphisms β(D) : L(D) → K(D) and β(D

) :
L(D

) → K(D

) such that the above diagram commutes except maybe for the part where the dotted
morphism is included. Note that
α(•) := g
H
D,D
τ
D
(•)β(D)
and
α() := g
G
D,D
τ
D
()β(D)
combine into a natural transformation α : G
D

L
→ G
D
F. Hence α must factor uniquely through τ
D
.
From the diagram above it follows that α factors both through g
K
D,D

β(D) and β(D

)g
L
D,D

. Hence these
morphisms must be equal and the part where the dotted morphism is included commutes. Hence ρ factors
uniquely through σ.
2.5 Special objects
This section shows some properties of the initial and terminal object of a topos. We first show they
always exist.
Proposition 2.5.1. Every topos T has an initial object 0 and a terminal object 1.
Proof: Let F : (
0
→ T be a functor. Notice that there is only one such functor. The inital object is the
colimit of F and the terminal object is the limit of F, because in this case the definition of the (co)limit
is equivalent to the definition of an initial object or a final object.
For the next proof it is instructive to introduce the slice category. In the next chapters we will study
diagrams of the form B → A for a fixed object A. These diagrams can be seen as objects of what is
called the slice category.
Definition 2.5.2. Let T be a topos and let A be an object of T . The slice category T /A is defined as
the following category:
• The objects are diagrams in T of the form f : B → A.
36
2.5. SPECIAL OBJECTS CHAPTER 2. TOPOS THEOREMS
• A morphism from an object f : B → A to an object g : C → A is a morphism h : B → C such that
the following diagram commutes:
B
h
>
C
f
A
g
<
>
.
The next theorem is stated without proof. The reason is that the proof is very technical, not very
instructive to read and it would take about three pages to work out. The result is simple, but very
important.
Theorem 2.5.3. Let T be a topos and let A be an object of T . Then the slice category T /A is a topos.
Proof: See for instance [11].
A morphism f : A → B induces a functor f
−1
: T /B → T /A by pulling back, as the following definition
shows.
Definition 2.5.4. Let f : A → B be a morphism between objects A and B in a topos T . This gives a
functor f
−1
: T /B → T /A, by pulling back a morphism S → B along f.
A
B
S → S
A

f
>
B.

f
−1
: T /B → T /A;
(S → B) → (A
B
S → B) .
For a morphism k : T → S in T /A one defines f
−1
(k) as the unique dotted morphism making the
following diagram commutative:
A
B
T
>
T
f
−1
(k) k
A
B
S
>
<
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
S
<
A

f
>
<
B

<
.
We will refer to f
−1
as the pullback functor.
By diagram chasing this definition is well defined. A very important result is that the functor f
−1
has
both a left and right adjoint. The proof of this is postponed to the next chapter, though we will use the
result here to prove the following proposition.
37
2.5. SPECIAL OBJECTS CHAPTER 2. TOPOS THEOREMS
Proposition 2.5.5. If f : A → 0 is a morphism to the initial object in a topos T , then f is an
isomorphism. Hence A is isomorphic to the initial object. We say the initial object of a topos is strict.
Proof: In the slice category T /0 the object Id
0
: 0 → 0 is both initial and terminal. Since k
−1
: T /
0 → T /A has a left adjoint, k
−1
(Id
0
: 0 → 0) must be the the terminal object in T /A, which is
Id
A
: A → A. Since k
−1
has a right adjoint k
−1
(Id
0
: 0 → 0) must be the initial object in T /A, which
is ! : 0 → A. So we see that both the diagrams
A
f
>
0
A
Id
A

f
>
0.
!=Id
0

0
f
>
0
A
!

f
>
0.
!=Id
0

are pullbacks. Hence A must be isomorphic to 0 (in T ). Since A is initial, this isomorphism must be
f.
38
Chapter 3
The algebra of subobjects
“Logic is invincible because in order to combat logic it is necessary to use logic.” - Pierre Boutroux
In this chapter T is a topos and all objects and morphisms are taken within this topos unless specified
otherwise. We will explore the structure of the subobjects of a given object in a topos. We will begin
with defining a subobject. Remark that the definition we gave in chapter 1 only sufficed for the previous
chapters. In this chapter we make it more precise and work towards a lattice structure on the subobjects
of a given object. Most of this work is based on [5], although some of the proofs are neater and a few
corollaries have been added.
3.1 Subobjects
Definition 3.1.1. A presubobject f : B → A in a topos T is an object B in T together with a monic
f : B → A.
Mostly we omit the morphism f if it is clear which morphism is meant. We will just write B → A saying
B is a presubobject of A. When it is clear of which object B is a presubobject, we simply speak about
the presubobject B.
The following definition of a subobject should not surprise the reader at all. In the previous chapters
we had a notion of uniqueness up to isomorphism for many of the definitions. For example 0, 1, limits,
colimits, Ω, . . . are all unique up to isomorphism. The structure we would like to study in this chapter
is the one of the presubobjects modulo isomorphism.
Definition 3.1.2. We define an equivalence relation on the presubobjects of an object A by (B → A) ·
(C → A) if and only if there exists an isomorphism h such that the following diagram commutes:
B
h
>
C
A
<


>
.
It is easily verified that this relation is symmetric and satisfies reflexivity and transitivity.
Definition 3.1.3. Define Sub(A) to be the class of presubobjects of A modulo ·-equivalence. The ele-
ments of Sub(A) are called subobjects.
39
3.1. SUBOBJECTS CHAPTER 3. THE ALGEBRA OF SUBOBJECTS
The first proof of this chapter is not hard at all. We have already mentioned it: there is a correspondence
between subobjects and characteristic morphisms. Although the proof is easy, its consequences are
important as we will see later. From this point we alter our notation of a characteristic morphism. If
B → A is a subobject we write χ
B
for the characteristic morphism of B.
Theorem 3.1.4. In a topos T , there is a bijective correspondence between Sub(A) and characteristic
morphisms A → Ω.
Proof: Since T is a topos every subobject B ∈ Sub(A) admits a unique χ
B
: A → Ω such that:
B
!
>
1
A
f


χ
B
>
Ω,


t
is a pullback square. This assignment is:
• Well-defined. If B, C ∈ Sub(A) and B · C, then the following diagram commutes:
C

>
B
>
1
A


Id
A
>
A
f


χ
B
>
Ω.


t
Both the left and right square are pullbacks. By the PBL the whole rectangle is a pullback. So χ
B
must be the unique morphism turning
C
>
1
A


χ
B
>
Ω,


t
into a pullback.
• Injective. Suppose both B, C ∈ Sub(A) have characteristic morphism χ. Then both B and C are
the pullbacks of
1
A
χ
>
Ω.


t
Limits are unique up to isomorphism, so we may conclude B · C.
40
3.2. LATTICES AND SUBOBJECTS CHAPTER 3. THE ALGEBRA OF SUBOBJECTS
• Surjective. For a given characteristic morphism χ : A → Ω, its pullback along t : 1 → Ω will yield
a subobject B with χ as its characteristic morphism.
From now on we will always consider subobjects instead of presubobjects. We will denote the subobjects
like presubobjects. This can be done without confusion of notation, since every claim made below will
be true for presubobjects as well as for subobjects.
3.2 Lattices and subobjects
We will prove that Sub(A) has more structure than just the structure of a set. Sub(A) will appear to be
a Heyting algebra. This algebra is defined as a lattice together with some extra structure.
Definition 3.2.1. A lattice is a poset L together with binary operations ∪ and ∩ and nullary operations
⊥ and ·, such that for every x, y, z ∈ L:
x ∩ y ≥ z iff x ≥ z and y ≥ z;
x ∪ y ≤ z iff x ≤ z and y ≤ z;
x ∩ ⊥ = ⊥ ;
x ∪ · = ·.
A morphism of lattices is a map that preserves ∩, ∪, ⊥ and ·. Such a map is called a lattice homomor-
phism.
The symbols ∩ and ∪ are sometimes called the inf and sup, respectively. The symbols ⊥ and · are
called the bottom and top element, respectively. Some authors refer to the defintion above as a lattice
with top and bottom element and define a lattice without having these elements. Since all the lattices
in this thesis do have a top and bottom element we prefer this definition.
Definition 3.2.2. A Heyting algebra is a lattice L with a binary operation ⇒, such that for every
x, y, z ∈ L:
x ≤ (y ⇒ z) iff x ∩ y ≤ z.
A morphism of Heyting algebras is a map that preserves ∩, ∪, ⊥ and · as well as ⇒.
Heyting algebras where introduced by Arend Heyting and play a role in intuitionistic logic. In the next
chapter we will introduce (an interpretation of) a logical language which will turn out to be intuitionistic,
i.e. the law of the excluded middle is not necessarily true. For chapters 5 and 6 we will need the definition
of a locale, which we state now.
Definition 3.2.3. A locale L is a Heyting algebra that is complete, i.e. for every subset U of L (possibly
infinite) there exists an element z ∈ L such that x is the least upper bound of U:
∀x(x ∈ U ⇒ x ≤ z) and (∀x(x ∈ U ⇒ x ≤ y)) ⇒ (z ≤ y).
Notation: z =
_
U (z is the supremem of U). A morphism f from a locale L to a locale M is a morphism
of Heyting algebras
¯
f from M to L that preserves the (infinite) supremum. Note that a morphism from
L to M is a function in the reversed direction preserving the structure of a locale.
41
3.2. LATTICES AND SUBOBJECTS CHAPTER 3. THE ALGEBRA OF SUBOBJECTS
A locale L is called compact if for any subset U of L such that · =
_
U, there is a finite subset V ⊂ U
such that · =
_
V .
For a locale L and x, y ∈ L, we say x is rather below y if there exists an element z ∈ L such that:
x ∧ z = 0 and z ∨ y = 1.
Moreover, we say x is completely below y if there is a family z
q
∈ L indexed by rationals 0 ≤ q ≤ 1,
for which
z
0
= x, z
p
is rather below z
q
if p < q, and z
1
= y.
Now, a locale L is called completely regular if for every x ∈ L
x =

¦y ∈ L [ y is completely below x¦ .
If X is a topological space, the set of opens O(X) is a locale. If f : X → Y is a map of topological
spaces, it leads to a map of locales g : O(X) → O(Y ) by the inverse image function: ¯ g = f
−1
: O(Y ) →
O(X). Furthermore, if X is compact Hausdorff, the locale is a compact completely regular locale. The
correspondence between topological spaces and their underlying topology will be usefull for chapter 5. In
this chapter our attention is focused on the algebra structure of Sub(A). It turns out to be an Heyting
algebra as the following series of definitions an propositions shows.
Proposition 3.2.4. Sub(A) has the structure of a partially ordered set by writing B ⊆ C if and only if
there is a morphism h such that the following diagram commutes:
B
h
>
C
A
<


>
.
Proof:
• Reflexivity is obvious, by taking h = Id.
• Transitivity follows by:
B
h
>
C
k
>
D
A


<


>
.
Since the outer triangles commute, the large triangle commutes, which amounts to B ⊆ D.
• For anti-symmetry. Suppose B ⊆ C and C ⊆ B, thus:
B
h
>
C
k
>
B
A


f
<

f

>
.
We have f ◦Id
B
= f ◦k◦h. Since f is monic, we conclude Id
B
= k◦h. In the same way Id
C
= h◦k.
Hence k = h
−1
, so B · C.
42
3.2. LATTICES AND SUBOBJECTS CHAPTER 3. THE ALGEBRA OF SUBOBJECTS
Proposition 3.2.5. Sub(A) has a bottom element ⊥, namely the initial object and its unique morphism
! : 0 → A. Sub(A) has a top element ·, namely the subobject Id
A
: A → A.
Proof: Since 0 is the initial object, it admits a unique morphism 0 → A. This morphism is automatically
monic: since in a topos 0 is strict (see proposition 2.5.5), any map B → 0, must be the unique (iden-
tity)map 0 → 0. Now since 0 is the initial object, the following diagram commutes for any subobject
f : B → A:
0
!
>
B
A
f
<

!

>
.
It follows from the commutativity of
B
f
>
A
A
Id
A
<

f

>
,
that for every subobject f : B → A, B ⊆ A. Hence Id
A
: A → A is the top element of the poset
Sub(A).
In order to turn Sub(A) into a Heyting algebra, we wish to define ∩, ∪ and ⇒. This can be done as
follows.
Definition 3.2.6. Let f : B → A and g : C → A subobjects of A. Define the subobject B ∩ C → A of
A by pulling back f allong g. That is:
B ∩ C

k
>
B
C


h

g
>
A.


f
Proposition 3.2.7. Definition 3.2.6 really defines the ∩ in the poset Sub(A). I.e. if D ⊆ B and D ⊆ C,
then D ⊆ B ∩ C.
Proof: Since f and g are monic, so are h and k (proposition 2.2.1), hence f ◦ k is. We conclude
f ◦ k : B ∩ C → A is a subobject of A.
Now suppose we have a subobject m : D → A such that D ⊆ B and D ⊆ C. Then:
43
3.2. LATTICES AND SUBOBJECTS CHAPTER 3. THE ALGEBRA OF SUBOBJECTS
D
B ∩ C

k
>
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
>
B
i
>
C
h



g
>
j
>
A.
f


By assumption f ◦ i = m = g ◦ j, so by definition of a pullback the dotted morphism exists, which means
D ⊆ B ∩ C.
The definition of ∩ already gives some interesting results. For example,
(B ∩ C) ∩ D = B ∩ (C ∩ D)
is true in an arbitrary partial order with a meet [15]. In our topos the ∩ is interpreted as a pullback
and hence this equation is an equality of objects which tells us pulling back B → A along C → A and
then pulling back the result along D → A is the same (up to isomorphism) as pulling back C → A along
D → A and then pulling back the result along B → A.
Next we define ∪.
Definition 3.2.8. Let f : B → A and g : C → A be subobjects of A. Define the subobject m : B∪C → A
of A by taking the image of the coproduct of f : B → A and g : C → A. That is:
B +C
<
k
B
B ∪ C
e
>
>
C
h


g
>
A.
f


m

>
Proposition 3.2.9. Definition 3.2.8 really defines the ∪ in the poset Sub(A). I.e. if B ⊆ D and C ⊆ D,
then B ∪ C ⊆ D.
Proof: Suppose we have a subobject j : D → A such that B ⊆ D and C ⊆ D. The universal property
of the coproduct amounts to the following commutative diagram:
B +C
<
k
B
D
i
>
C
h


g
>
A.
f


j

>
44
3.2. LATTICES AND SUBOBJECTS CHAPTER 3. THE ALGEBRA OF SUBOBJECTS
Note that both m◦e in definition 3.2.8 and j ◦i in the above diagram are the unique morphism such that
both f and g factor through B + C. Hence they are equal: m◦ e = j ◦ i. Now split i into its epi-monic
factorization:
N

>
D
B +C
>
>
A
j

>
B ∪ C

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
m

>
e >
>
.
According to theorem 2.3.3 there is an isomorphism (the dotted arrow) M → N such that the above
diagram commutes. It now follows that B ∪ C ⊆ D.
We will denote the morphism m in the above definition as f ∪ g. Finally, we define the implication ⇒.
Definition 3.2.10. Let f : B → A and g : C → A be subobjects of A. Define the subobject e :
(B ⇒ C) → A as the equalizer of the characteristic morphisms χ
B
and χ
B∩C
.
(B ⇒ C)

e
>
A
χ
B
>
χ
B∩C
>
Ω.
Proposition 3.2.11. Definition 3.2.10 really defines the ⇒ in the poset Sub(A). I.e. for subobjects
B, C, D ∈ Sub(A): D ⊆ (B ⇒ C) if and only if D ∩ B ⊆ C.
Proof: First note that e in definition 3.2.10 really is monic since it is an equalizer.
Write m : D → A for the subobject D. By definition of the partial order and the equalizer one has
D ⊆ (B ⇒ C) if and only if m factors through e if and only if m equalizes χ
B
and χ
B∩C
. We find:
D ⊆ (B ⇒ C) iff χ
B
◦ m = χ
B∩C
◦ m;
iff D ∩ B = D ∩ B ∩ C;
iff D ∩ B ⊆ C.
In the second iff-statement we used proposition 3.1.4 as well as the fact that in
D ∩ B
>
B
>
1
D


m
>
A


χ
B
>
Ω,


true
both the left and right square are pullbacks. By the PBL the outer rectangle is a pullback, hence χ
B
◦ m
is the characteristic morphism of D ∩ B. Replacing B by B ∩ C, one readily finds that χ
B∩C
◦ m is the
characteristic morphism of D ∩ B ∩ C.
Finally we summarize the result in one theorem.
Theorem 3.2.12. With ∩, ∪ and ⇒ defined as above, Sub(A) is a Heyting algebra.
45
3.2. LATTICES AND SUBOBJECTS CHAPTER 3. THE ALGEBRA OF SUBOBJECTS
You could wonder if Sub(A) has the structure of a boolean algebra, i.e. if there is some operation on
subobjects such that B ∩ B = ⊥ and B ∪ B = ·. This is not necessarily the case. For example, one
might look at the topos Set
C
2
. Let F be the functor, depicted as a diagram:
¦0, 1¦
!
>
¦1¦ .
Let G be the subfunctor
¦0¦
!
>
¦1¦ .
By applying theorem 2.4.5 to the definition of ∩ and ∪ and noticing that the epi-monic factorization can
be taken pointwise, the ∪ and ∩ are the naive point-wise meet and intersection. Then G must be equal
to
¦1¦
?
>
∅,
which is not a proper function anymore.
Since we are doing category theory, the obvious next step is to define morphisms between algebras Sub(A)
and Sub(B). We recall definition 2.5.4.
Definition 3.2.13. Let f : A → B be a morphism between objects A and B in a topos T . This gives a
functor f
−1
: T /B → T /A, by pulling back a morphism S → B along f.
A
B
S → S
A

f
>
B.

f
−1
: T /B → T /A
(S → B) → (A
B
S → B)
For a morphism k : T → S in T /A one defines f
−1
(k) by the unique dotted morphism making the
following diagram commutative.
A
B
T
>
T
f
−1
(k) k
A
B
S
>
<
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
S
<
A

f
>
<
B

<
.
46
3.3. THE INTERNAL HEYTING ALGEBRA CHAPTER 3. THE ALGEBRA OF SUBOBJECTS
Definition 3.2.14. Since the pullback of a monic is monic (proposition 2.2.1), f
−1
in definition 2.5.4
descents to a morphism from Sub(B) to Sub(C). Since limits are unique up to isomorphism the pullback
functor is well-defined on the equivalence classes defined in definition 3.1.2.
The most important property of the pullback functor is that it preserves the structure of the Heyting
algebra.
Proposition 3.2.15. The map f
−1
of definition 3.2.14 is a morphism of Heyting algebras.
Proof: One can give a proof by diagram chasing, using the universal property of pullbacks, the fact that
pulling back an epi (monic) gives again an epi (monic), theorem 2.3.3 about epi-monic factorization and
the PBL.
The proof of the last theorem is not very elegant. In the next chapter, we will give a short proof of
proposition 3.2.15 using theorem 3.1.4. Actually, proposition 3.2.15 will turn out to be a special case
of a more general situation. Namely, we will observe that f
−1
has a left and right adjoint. Then f
−1
preserves limits as well as colimits, and hence it perserves ∩ and ∪ since they are the product, respectively
coproduct of the lattice Sub(B) as a preorder.
3.3 The internal Heyting algebra
Theorem 3.1.4 gives a correspondence between the Heyting algebra Sub(A) of an object A in a topos T
and the characteristic morphisms A → Ω. We may use this correspondence to turn the set Hom
T
(A, Ω)
into a Heyting algebra. For B, C ∈ Sub(A), put:
χ
B
∩ χ
C
= χ
B∩C
;
χ
B
∪ χ
C
= χ
B∪C
;
χ
B
⇒ χ
C
= χ
B⇒C
;
⊥ = = χ

;
· = true
A
= χ

.
The last two items define the initial and terminal object in Hom
T
(A, Ω). It follows directly from the
definitions that these are indeed the initial and terminal objects.
The poset structure on Hom
T
(A, Ω) is given by the obvious relation:
χ
B
⊆ χ
C
iff B ⊆ C;
iff χ
B
∩ χ
C
= χ
B
;
iff B ∩ C = B.
The aim of this paragraph is to give a more direct description of the operators ∩, ∪ and ⇒on Hom
T
(A, Ω).
Proposition 3.3.1. Let ∧ : Ω Ω → Ω be the characteristic morphism of (t, t) : 1 → Ω Ω. Then for
B, C ∈ Sub(A), ∧ ◦ (χ
B
, χ
C
) = χ
B
∩ χ
C
.
Proof: We are finished if the outer rectangle in the following diagram is a pullback.
B ∩ C
!
>
1

>
1
A



B

C
)
>
Ω Ω
(t,t)



>

t


47
3.3. THE INTERNAL HEYTING ALGEBRA CHAPTER 3. THE ALGEBRA OF SUBOBJECTS
The right square is a pullback by definition of ∧. If we can prove the left square is a pullback, the PBL
shows the outer rectangle is a pullback. Suppose we have a commutative diagram:
D
B ∩ C
>
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
>
1
>
A


B

C
)
>
>
Ω Ω.
(t,t)

(3.1)
We need to prove that there is a unique morphism from D to B ∩ C, indicated with the dotted arrow in
the diagram above. From this diagram, we obtain:
D
B ∩ C
>
B
>
>
1
>
C

>
>
A

χB
>
>


t
1

t
>
>

χC

.
(3.2)
The bottom and the right squares are pullbacks. Hence there are unique morphisms from D to B and
from D to C making the diagram commute. The upper left square is a pullback, hence there is a unique
morphism from D to B ∩ C making the diagram commute. Since diagram 3.1 is equivalent to diagram
3.2, this is the unique dotted morphism making diagram 3.1 commute.
If you do not accept this last argument, do a diagram chase on diagram 3.1.
Proposition 3.3.2. Let ∨ : Ω Ω → Ω be the characteristic morphism of (true

, 1

) ∪ (1

, true

) :
Ω ∪ Ω → Ω Ω. Then for B, C ∈ Sub(A), ∨ ◦ (χ
B
, χ
C
) = χ
B
∪ χ
C
.
We split the proof in three steps:
1. We first prove the following square to be a pullback:
B

f
>
A

χ
C
◦f

(true

,1

)
>
Ω Ω.

B

C
)

48
3.3. THE INTERNAL HEYTING ALGEBRA CHAPTER 3. THE ALGEBRA OF SUBOBJECTS
Suppose we have an object D such that the following diagram commutes:
D
B

f
>
l
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. >
A
k
>

χ
C
◦f

1

>
m
>

χ
C
>
1
!

t
>

χ
B

true

>
.
The rectangle ‘B −A−Ω −1’ is a pullback, hence there exists an l as indicated, such that
f ◦ l = k;
! ◦ χ
C
◦ f ◦ l = ! ◦ m.
From the diagram we obtain m = 1

◦ m = χ
C
◦ k. Hence
χ
C
◦ f ◦ l = χ
C
◦ k = m.
So the whole diagram commutes. Furthermore, since f is monic it follows that l is a unique such
arrow, for if l

◦ f = k = l ◦ f, then l

= l. Looking closely, one sees that both the above diagrams
represent the same situation and the morphism l is the unique morphism turning the first diagram
into a pullback.
2. In the same spirit, the following square is a pullback:
C

g
>
A

χ
B
◦g

(1

,true

)
>
Ω Ω.

B

C
)

3. Now take a look at the diagram:
B +C
i
1
◦χ
B
◦f+i
2
◦χ
C
◦g
>
Ω + Ω
B ∪ C


............................................... >
Ω ∪ Ω


!
>
1
A
f
∪ g



B

C
)
>
Ω Ω
(true

,1

)∪(1

,true

)



>
Ω.
t

49
3.3. THE INTERNAL HEYTING ALGEBRA CHAPTER 3. THE ALGEBRA OF SUBOBJECTS
In the left outer rectangle, we took the epi-monic factorization of f +g and (true

, 1

)+(1

, true

).
Since the pullback functor (χ
B
, χ
C
)
−1
has a left adjoint (as we will prove in theorem 3.4.1), it
preserves colimits. Combining this with step 1 and step 2, the left (outer) rectangle is a pullback.
Theorem 2.3.3 gives a unique map from B ∪ C to Ω ∪ Ω. This turns the bottom left square into a
pullback. To see this, pull back (χ
B
, χ
C
) along (true

, 1

) ∪ (1

, true

) and pull the result back
along Ω+Ω → Ω∪Ω. The PBL tells us this double pullback is a pullback itself, so that the outer left
square is a pullback too. Since the pullback of a monic is monic (theorem 2.2.1), or alternatively,
since the pullback functor has a left adjoint (as we will prove later) and since the pullback of an
epi is epi since the pullback functor has a right adjoint (as we will prove later), it follows that the
two morphisms obtained from the double pullback must form the epi-monic factorization of f +g.
This factorization is unique, hence the bottom square must be the first pullback.
The (bottom) right square is a pullback by definition. The PBL gives that the bottom (outer)
rectangle is a pullback. Hence ∨ ◦ (χ
B
, χ
C
) is the characteristic morphism of B ∪ C, which is
χ
B
∪ χ
C
by definition.
Proposition 3.3.3. Let →: Ω Ω → Ω be the characteristic morphism of the equalizer e of

e
>
Ω Ω

>
p
1
>
Ω,
where p
1
is the projection on the first coordinate. Then for B, C ∈ Sub(A), → ◦(χ
B
, χ
C
) = (χ
B
⇒ χ
C
).
Proof: Our strategy will be the same as in propopositions 3.3.1 and 3.3.2. Take the diagram
D
B ⇒ C
....................
h
>
l
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
>

!
>
k
>
1
A
f


B

C
)
>
m
>
Ω Ω
e



>

t

,
Ω.
p
1



where f denotes the equalizer of both compositions. Note that p
1
◦ (χ
B
, χ
C
) = χ
B
and ∧ ◦ (χ
B
, χ
C
) =
χ
B
∩ χ
C
.
The right square is a pullback (by definition). If we can prove that the left square is a pullback, the PBL
tells us that the whole rectangle is a pullback, which proves the proposition. First note that since
p
1
◦ (χ
B
, χ
C
) ◦ f = ∧ ◦ (χ
B
, χ
C
) ◦ f,

B
, χ
C
) ◦ f equalizes p
1
and ∧. Hence there is a unique morphism h : (B ⇒ C) →≤, that makes the
left square commute. This square actually turns out to be a pullback. Suppose we have an object D and
morphisms k, m as indicated in the diagram. Now,
p
1
◦ (χ
B
, χ
C
) ◦ m = p
1
◦ e ◦ k = ∧ ◦ e ◦ k = ∧ ◦ (χ
B
, χ
C
) ◦ m.
50
3.3. THE INTERNAL HEYTING ALGEBRA CHAPTER 3. THE ALGEBRA OF SUBOBJECTS
So m equalizes χ
B
and χ
B
∩ χ
C
, hence it factors uniquely through f by l. To see that for this unique l
one has h ◦ l = k, compute
e ◦ k = (χ
B
, χ
C
) ◦ m = (χ
B
, χ
C
) ◦ f ◦ l = e ◦ h ◦ l.
Since e is monic, we conclude h ◦ l = k.
We are now in a position to prove proposition 3.2.15. A morphism f : A → B gives a morphism
f
−1
: Sub(B) → Sub(A) as defined in the previous chapter. The corresponding map on the characteristic
morphisms is
f

: Hom
T
(A, Ω) → Hom
T
(B, Ω) : χ
T
→ χ
T
◦ f.
Since the left and right squares in the following diagram are pullbacks, the PBL turns the outer rectangle
into a pullback. Hence f

χ
T
is indeed the characteristic morphism of f
−1
(T).
f
−1
(T)
>
T
!
>
1
A


f
>
B


χ
T
>
Ω.


Proof (of proposition 3.2.15): For S, T ∈ Sub(B) one has
f


T∪S
) = ∧ ◦ (χ
T
, χ
S
) ◦ f
= ∧ ◦ (χ
T
◦ f, χ
S
◦ f)
= χ
f
−1
(T)
∩ χ
f
−1
(S)
.
With theorem 3.1.4 we conclude
f
−1
(T ∩ S) = f
−1
(T) ∩ f
−1
(S).
Replacing ∧ and ∩ by ∨ and ∪ or → and ⇒ in the above equations, one finds
f
−1
(T ∪ S) = f
−1
(T) ∪ f
−1
(S)
f
−1
(T ⇒ S) = f
−1
(T) ⇒ f
−1
(S).
Propositions 3.3.1, 3.3.2 and 3.3.3 and the definition of f

gave a direct description of the Heyting algebra
structure of Hom
T
(A, Ω). Now if we apply the functor −
A
obtained from the cartesian closedness of our
topos, we find morphisms:

A
: Ω
A

A
→ Ω
A
;

A
: Ω
A

A
→ Ω
A
;

A
: Ω
A

A
→ Ω
A
;
⊥ : 1 → Ω
A
;
· : 1 → Ω
A
.
51
3.4. ∃ AND ∀ CHAPTER 3. THE ALGEBRA OF SUBOBJECTS
Note that we used the isomorphism (Ω Ω)
A
· Ω
A

A
here. These morphisms have the property
that resemble of a Heyting algebra in Set. For example, ∧
A
is associative, since the following diagram
commutes:

A

A

A
Id

A
×∧
A
>

A

A

A

A

A
×Id

A


A
>

A
.

A

In the same way one finds that ∧
A
is commutative, and that it satisfies the idempotence law (φ∧
A
φ = φ)
and the unit law (1 ∧
A
φ = φ). The dual relations for ∨
A
are true too, i.e. ∨
A
satisfies associativity,
commutativity, idempotence and the unit law (0 ∨
A
φ = φ). ∧
A
and ∨
A
satisfy the absorbtion laws.
Furthermore, if E is the equalizer of
E
e
>

A

A

>
π
1
>

A
,
then both the following squares are pullbacks:
Q
>
E

A

A

A


A
×Id

A
>

A

A
.
e

Q
>
E

A

A

A

Id

A
×⇒
A
>

A

A
.
e

Note that both diagrams contain the same object Q. An object L with morphisms ∧, ∨, ⇒, · and
⊥ that satisfies all the above properties is called an internal Heyting algebra in the topos T . This is a
generalization of the concept of a Heyting algebra in Set. Thus we have proved Ω
A
is an internal Heyting
algebra.
3.4 ∃ and ∀
In this section we will prove that the pullback functor has both a left and a right adjoint. These are called
∃ and ∀, respectively and are generalizations of the symbols ∃ and ∀ in Set. In the next chapter we will
see that most of the set theoretic properties of ∃ and ∀ generalize to topoi. Defining the topos-theoretic
operations ∃ and ∀ is not as simple as in the case of Set, however.
Definition 3.4.1. Let f : A → B be a morphism in a topos. The induced map f
−1
: Sub(B) → Sub(A)
has a left adjoint as a morphism of preorders. The left adjoint is given by:
S
e
>>

f
S
A
i



f
B,
m


where S →→
e

f
(S) →
m
B is the epi-monic factorization of the composite S →
i
A →
f
B.
52
3.4. ∃ AND ∀ CHAPTER 3. THE ALGEBRA OF SUBOBJECTS
We prove that this definition is correct.
Proposition 3.4.2. The function ∃
f
in definition 3.4.1 is the left adjoint of f
−1
as morphism of pre-
orders.
Proof: First observe that the isomorphism
Hom
Sub(B)
(∃
f
S, T) · Hom
Sub(A)
(S, f
−1
T)
is equivalent to the statement: a morphism u exists such that the diagram below commutes if and only
if a morphism v exists such that the diagram below commutes.
f
−1
T
k
>
T
S
e
>>
u
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
>

f
S
v
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
>
A
i


f
>
r
<

B
m


s
<

.
Now if u exists, then mr = fi = fru = sku. Now according to theorem 2.3.3, sku has a unique epi-monic
factorization. Hence there is a unique isomorphism between ∃
f
S and Im(sku) · Im(ku) (s was already
monic), which gives the unique (!) morphism v.
If v exists, s ◦ v ◦ e = m◦ e = f ◦ i. By the universal property of a pullback there is a unique morphism
u such that the diagram commutes.
Some properties one might expect to be true for the left adjoint ∃
f
are infect false. For instance, one
might ask if ∃
f
preserves ∩, ∪ and ⇒. This is in general only true for some of these logical connectives.
To see that some examples of properties do not hold, take the topos Set. Let X = ¦0, 1¦ and Y = ¦0, 1¦.
Let p : X Y → X be the projection on the first co¨ordinate. Then:
• The following diagram is in general not a pullback:
A
e
>>

p
A
X Y
i


p
>
X.
m


To see this, choose A = ¦(0, 0)¦. Then ∃
p
A = Im(p ◦ i) = e (¦(0, 0)¦) = ¦0¦. But then the pullback
of m along p is
¦(0, 0), (0, 1)¦
e
>>
¦0¦
¦0, 1¦ ¦0, 1¦
i


p
>
¦0, 1¦.
m


53
3.4. ∃ AND ∀ CHAPTER 3. THE ALGEBRA OF SUBOBJECTS
Hence the first diagram can not be pullback diagram.
• ∃
p
does not preserve ∩. Take A = ¦(0, 1)¦ ⊆ X Y and B = ¦(0, 0)¦ ⊆ X Y . Then ∃
p
(A∩B) =

p
(∅) = ∅. But ∃
p
(A) ∩ ∃
p
(B) = ¦0¦ ∩ ¦0¦ = ¦0¦.
• ∃
p
does not preserve ⇒. In Set one can easily show that for subobjects A and B of X Y that
A ⇒ B = B
c
∪ A. Take A = ∅ and B = ¦(1, 0)¦. Then ∃
p
(A ⇒ B) = ¦0, 1¦ ,= ¦0¦ = ∃
p
(A) ⇒

p
(B).
• ∃
p
does preserve ∪. Let A be an object in a topos. Then we can regard Sub(A) as a preorder,
hence a category. ∪ is the coproduct of this category. Since ∃
p
has a right adjoint (namely p
−1
) it
must preserve colimits, hence in particular the coproduct (theorem 2.4.2).
We will prove that f
−1
has a right adjoint too. This is not as easy as proving the existence of its left
adjoint. We will split the proof in two steps, making use of the slice category (see chapter 2). The proof
can be found in [11]. This proof uses the theorem stated in chapter 2, proving that the slice category
T /A is a topos. The most important proposition is the following one.
Proposition 3.4.3. Let T be a topos and let A and B be objects of T . Let f be a morphism from A to
B. The functor f
−1
: T /B → T /A has a right adjoint, called π
f
.
Proof: The proof is obtained in two steps:
1. First assume B is the terminal object 1. We find that
S A
p
1
>
S
A
p
2

f
>
1,

is a pull back square, since it is easily verified that every pullback over 1 is just the product. We
conclude that f
−1
(S) = S A.
Now let h : T → A be an object over A. A morphism from S A to T over A is a morhpism
t : S A → T such that
h ◦ t = p
2
.
We use the adjunction of cartesian closedness to apply the functor −
A
on both sides of the equal-
ity. The morphisms t correspond to morphisms t

: S → T
A
such that h
A
◦ t

(the LHS of the
equality) equals the composite S →
!
1 →
j
A
A
(the RHS of the equality). Here j is the morphism
corresponding by cartesian closedness to the identity on A. This can be seen from the following
diagrams:
Hom
T
(S A, T)

ϕ
>
Hom
T
(S, T
A
)
Hom
T
(S A, A)
h∗

ϕ
>
Hom
T
(S, A
A
),
(h
A
)∗

54
3.4. ∃ AND ∀ CHAPTER 3. THE ALGEBRA OF SUBOBJECTS
ϕ(h ◦ t) = ϕ(h

◦ t) = (h
A
)

◦ ϕ(t) = h
A
◦ t

;
Hom
T
(A, A)

φ
>
Hom
T
(1, A
A
)
Hom
T
(S A, A)
p

2

φ
>
Hom
T
(S, A
A
),
!


φ(p
2
) = φ(p

2
◦ Id
A
) =!

◦ φ(Id
A
) = j◦!.
Now consider the pullback square:
S
1
A
A T
A
>
t

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. >
T
A
t

>
1

j
>
!
>
A
A
.
h
A

By the universal property of pullbacks, every t

corresponds to a morphism t

: S → 1
A
A T
A
. We
conclude that every morphism t : S A → T over A, corresponds bijectively to a unique morphism
t

: S → 1
A
A T
A
(over 1 if you like). Hence this defines the desired right adjoint.
2. Now suppose B is not necessarily equal to 1. The morphism f : A → B can be regarded as an
object of the slice category T /B. An object over f : A → B in the double slice category (T /B) /
(f : A → B) is a commutative square
C
g
>
B
A
h

f
>
B.
Id
B

Now for each h such that this diagram commutes, the arrow g is determined. Conversely, every
such g defines an h. This correspondence defines a functor F : (T /B) /(f : A → B) → T /A as well
as a functor G : T /A → (T /B) /(f : A → B), which are inverses of each other (i.e. the maps on
the objects are mutual inverses and the maps on the morphisms are mutual inverses). Where we
take the obvious map on morphisms.
Now we can regard f
−1
as a functor
f
−1
: (T /B)/1 · T /B → T /A · (T /B) /(f : A → B).
Hence, since T /B and T /A are toposes, we are in the situation of step 1 and so f
−1
has a right
adjoint. This is the required π
f
.
55
3.4. ∃ AND ∀ CHAPTER 3. THE ALGEBRA OF SUBOBJECTS
Corollary 3.4.4. Let f : A → B be a map between objects in a topos. The induced map f
−1
: Sub(B) →
Sub(A) has a right adjoint as map of partially ordered sets, called ∀
f
.
Proof: In order to prove this theorem, it is enough to show that if m : C → A is monic, then π
f
m is
monic. To see this, notice that since π
f
is a right adjoint, it must preserve the terminal object of T /A,
which is Id
A
: A → A.
Suppose m is a monic in T /A. Then

f
m)h = (π
f
m)k iff
m(f
−1
(h)) = m(f
−1
(k)) iff
h = k.
Hence π
f
preserves monics. So if m is a monic, it is preserved under the following operations:
C
A
m


C
m
>
A
m
A
Id
A

>

π
f
C
π
f
m
>
B
π
f
m
B
Id
B
∨ >

π
f
C
A
π
f
m

.
Hence π
f
restrict to the ‘monic objects’. This restriction is by definition ∀
f
.
We examine a few properties of this adjoint functor. Consider once more the topos Set. Let X = ¦0, 1¦
and Y = ¦0, 1¦. Let p : X Y → X be the projection on the first co¨ordinate. Then:
• There is not necessarily a pullback diagram of the form
A
............ >

p
A
X Y


p
>
X.


Indeed the construction of ∀
p
does not give the dotted morphism. And actually such a mor-
phism does not have to exist. For an example, take A = ¦(0, 0)¦. Then ∀
p
(A) = ∅ (this
follows from direct calculation, or in a more elegant way, by making use of the correspondence
Hom
Sub(X×Y )
(p
−1
−, −) · Hom
Sub(X)
(−, ∀
f
−)). Then there is no morphism making the following
diagram a pullback:
¦(0, 0)¦
......... >

X Y


p
>
X.


• Unlike ∃
f
, ∀
f
does not preserve ∪. Take A = ¦(0, 1)¦ ⊆ X Y and B = ¦(0, 0)¦ ⊆ X Y .
Then ∀
p
(A ∪ B) = ∀
p
(¦(0, 0), (0, 1)¦) = ¦0¦ (again this follows from Hom
Sub(X×Y )
(p
−1
−, −) ·
Hom
Sub(X)
(−, ∀
f
−)). But ∀
p
(A) ∪ ∀
p
(B) = ∅ ∪ ∅ = ∅.
56
3.4. ∃ AND ∀ CHAPTER 3. THE ALGEBRA OF SUBOBJECTS
• ∀
f
does not preserve ⇒. In Set, A ⇒ B = B
c
∪A. Take A = ∅ and B¦(1, 0)¦. Then ∀
p
(A ⇒ B) =
¦0¦ , = ¦0, 1¦ = ∀
p
(A) ⇒ ∀
p
(B).
• Unlike ∃
f
, ∀
f
does preserve ∩. Let A be an object in a topos. Then we can regard Sub(A) as a
preorder, hence a category, whose product is ∩. Since ∀
f
has a left adjoint (namely p
−1
), it must
preserve limits, hence in particular, the product.
For objects A and B in a topos and a morphism f : A → B, we would like to know how the characteristic
morphisms of S ∈ Sub(A) and ∃
f
S ∈ Sub(B) as well as ∀
f
S ∈ Sub(B) depend on each other.
Definition 3.4.5. Let A and B be objects in a topos and let f : A → B be a morphism from A to B.
For S ∈ Sub(A), define:

f
χ
S
:= χ

f
S

f
χ
S
:= χ

f
S
Now we have shown the algebra of subobjects as a logical structure, namely the structure of a Heyting
algebra. Furthermore we have a pullback functor f
−1
and a notion of ∃
f
and ∀
f
. We will use these
definitions to introduce an internal logical language in a topos.
57
Chapter 4
The Mitchell-B´enabou language
“Such is the advantage of a well constructed language that its simplified notation often becomes the
source of profound theories.” - Pierre-Simon de Laplace
We have defined a logical structure using ∩, ∪, ⇒ on objects of the form Sub(A) (externally) or Ω
A
(internally), where A is an object in some topos T , as well as quantifiers ∀
f
and ∃
f
relating Sub(A) and
Sub(B), where both A and B are objects of a topos T and f : A → B is a morphism. Now we would
like to build logical formulae from it. This can be done by using terms: expressions of variables, logical
connectives and quantifiers that are defined inductively. In any topos T , these formal expressions have
an interpretation, which is a morphism in T , see definition 4.1.1. The target of the interpretation of a
term is called the type of the term. If the interpretation of a term is a morphism which target is Ω, the
term is called a formula. In the previous chapters we used the characters A and B to specify objects. In
this chapter we use X, Y , X
1
, X
2
, et cetera to denote objects. This notation will remind one more of
variables. The basic definitions and theorems in this chapter can be found in [11] and [1].
4.1 The language
In this chapter we introduce the language formulated by W. Mitchell and J. B´enabou. Recall that our
perspective so far has been that a topos is a generalization of the category of sets. In this chapter we
see how well the three defining properties (i.e. limits, subobjects and cartesian closedness) where chosen.
Indeed, it turns out that a whole logical language can be built within a topos.
Definition 4.1.1. We define a term in a topos T to be an expression that can be obtained by the
following inductive steps:
1. For an object X, the character x is a term of type X. Its interpretation is the identity Id
X
.
2. Let f : X → Y be a morphism and let τ be a term of type X. Then the formal expression fτ is a
term of type Y . Its interpretation is the morphism f ◦ τ.
3. Let σ be a term of type X with interpretation σ : U → X and let τ be a term of type Y with
interpretation τ : V → Y . Then (σ, τ) is a term of type X Y . Its interpretation is the morphism
(σ ◦ p
1
, τ ◦ p
2
) : U V → X Y .
4. Let σ and τ be terms of type X, interpreted by σ : U → X and τ : V → X respectively. Then σ = τ
is a term of type Ω, i.e. a formula. Its interpretation is the morphism
58
4.1. THE LANGUAGE CHAPTER 4. THE MITCHELL-B
´
ENABOU LANGUAGE
σ = τ :U V
(σ◦p
1
,τ◦p
2
)
>
X X
δ
X
>
Ω.
Here δ
X
is the characteristic morphism of the diagonal morphism ∆
X
: X → XX. Note that the
equality sign ‘=’ is not an equality of morphisms. For example, σ and τ may have different sources.
5. Let σ be a term of type X with interpretation σ : U → X and let τ be a term of type Ω
X
with
interpretation τ : V → Ω
X
. This yields a term σ ∈ τ of type Ω, interpreted by
σ ∈ τ : U V
>
X Ω
X
e
>
.Ω.
Here e is the morphism in Hom
T
(X Ω
X
, Ω) corresponding to Id

X ∈ Hom
T
(Ω
X
, Ω
X
) by carte-
sian closedness.
6. Let θ be a term of type Y
X
, interpreted by θ : V → Y
X
and let σ be a term of type X, interpreted
by σ : U → X. This gives a term θ(σ) of type Y , interpreted by
θ(σ) : V U
>
Y
X
X
e
>
Y.
7. Let x be a variable of type X and σ be a term of type Z with interpretation σ : X U → Z. This
yields a term λxσ of type Z
X
whose interpretation is the transpose of σ.
λxσ : U
>
Z
X
.
8. Let φ and ψ be terms of type Ω with interpretation φ : U → Ω and ψ : V → Ω respectively. Then
the following expressions and their interpretations define terms of type Ω.
φ ∧ ψ :U V
(φp
1
,ψp
2
)
>
Ω Ω

>
Ω;
φ ∨ ψ :U V
(φp
1
,ψp
2
)
>
Ω Ω

>
Ω;
φ ⇒ ψ :U V
(φp
1
,ψp
2
)
>
Ω Ω

>
Ω.
9. Let φ be a term of type Ω with interpretation φ : U X → Ω. Then the following expressions and
their interpretation define terms of type Ω:
∃xφ : U
∃pφ
>
Ω;
∀xφ : U
∀pφ
>
Ω.
Here p : U X → U is the projection on U.
59
4.2. SOME EXAMPLES CHAPTER 4. THE MITCHELL-B
´
ENABOU LANGUAGE
It is important to remark is that the definition of a term is a formal expression defined by the inductive
steps in above definition. These formal expressions are just strings of logical signs, variables and functions
having a priori nothing to do with morphisms within a topos. Regarding them as morphisms in a topos
can be done though (by above definition), and is called the interpretation of the logical language.
In the next part of the thesis expressions that are obtained using inductive steps (5), (6) and (7) will not
occur. These steps are quoted for the completeness of the definition. The Mithcell-B´enabou language has
a wide amount of properties, many of which can be found in [1] and [11]. Some proofs in the literature
make use of all of these steps, but the details of this language distract too much from the scope of
this thesis. Nevertheless, we will give a few examples which demonstrate the power of the Mitchell-
B´enabou language. Our main example will be to prove that for a morphism f in a topos the formula
∀x∀x

fx = fx

⇒ x = x

‘is true’ if and only if f is monic. After briefly playing with this language I will
refer to literature.
If φ : X
1
. . . X
n
→ Ω is a formula, then we write φ(x
1
, . . . , x
n
) to explicitly indicate its source.
Definition 4.1.2. A formula is a term of type Ω.
We now state when a formula is true.
Definition 4.1.3. A formula φ(x) : X → Ω is called true, written [= φ(x), if it factors through t : 1 → Ω.
1
!
X
φ(x)
>
>
Ω.
t

This definition can be extended in an obvious way to formulae of multiple variables.
For our convenience we introduce a different, but equivalent notion of ‘true’.
Corollary 4.1.4. We say a formula φ(x) : X → Ω is true if and only if the following square is a pullback:
X
!
>
1
X
Id
X

φ(x)
>
Ω.
t

Proof: The proof is straightforward.
We will not distinguish between these definitions.
If φ(x) : X → Ω is a formula we will write ¦x [ φ(x)¦ for the corresponding subobject of X.
4.2 Some examples
At this point the language built, in the previous chapters is very abstract. Before building any further
machinery, it is instructive to see some examples of the use of this language. This chapter uses a lot of
60
4.2. SOME EXAMPLES CHAPTER 4. THE MITCHELL-B
´
ENABOU LANGUAGE
diagram chasing to tackle its problems. This will be useful to understand the definitions of the language.
Furthermore, it will give a feeling for the theory built up in the next chapter.
Actually, the logical language will produce quite useful results. Many of the logical formulae which are
true intuitively are indeed true in the topos logic. This shows the power of the language.
Example 4.2.1. Let φ(x) : X → Ω and ψ(x) : X → Ω be formulae. Then we can show that ¦x[φ(x)¦ ⊆
¦x[ψ(x)¦ if and only if φ(x) → ψ(x) is true.
Remark that:
¦x[φ(x)¦ ⊆ ¦x[ψ(x)¦ iff
¦x[φ(x)¦ ∩ ¦x[ψ(x)¦ = ¦x[φ(x)¦ iff
φ(x) ∩ ψ(x) = φ(x) (equality as morphisms).
Hence we must show that φ(x) ∩ ψ(x) = φ(x) if and only if [= (φ(x) → ψ(x)). Now look at the diagram:
¦x[φ(x) → ψ(x)¦
h
>

!
>
1
X
f

(φ(x),ψ(x))
>
Ω Ω
e



>

t

Ω.
p
1



and remark that f was defined as the equalizer of φ(x) and φ(x) ∩ ψ(x). Now if φ(x) → ψ(x) is true,
we find f is the identity. Hence φ(x) ∩ ψ(x) = φ(x). If φ(x) ∩ ψ(x) = φ(x), the equalizer of φ(x) and
φ(x) ∩ ψ(x) apparently is Id
X
: X → X. Hence we conclude that φ(x) → ψ(x) is true.
Example 4.2.2. In this example both x and x

will be variables of type X. Let f : X → Y be a morphism.
The following is statement holds: ∀x∀x

fx = fx

⇒ x = x

is true if and only if f is monic.
Remark that ‘monic’ has the intuitive meaning of ‘injective’. But since in a topos we are dealing with mor-
phisms rather than functions, there is no proper meaning of ‘injective’. In the topos Set the morphisms are
functions and one can easily prove monic is equivalent to injective. The formula ∀x∀x

fx = fx

⇒ x = x

is intuitively the definition of injectivity. However, in the Mitchell-B´enabou language this expression has
a far more complicated meaning. Therefore it may be surprising that this formula is true if and only if
f is monic. We will prove this using example 4.2.1.
First suppose the formula is true. Consider:
_
(x, x

)[fx = fx

⇒ x = x

_
>
X X
_
(x, x

)[∀x

fx = fx

⇒ x = x

_
>
X
p
1

_
(x, x

)[∀x∀x

fx = fx

⇒ x = x

_
>
1
!

1

t
>
Ω.
true
1
=(∀x∀x

fx=fx

⇒x=x

)

61
4.2. SOME EXAMPLES CHAPTER 4. THE MITCHELL-B
´
ENABOU LANGUAGE
The bottom square is a pullback. Hence we conclude ¦(x, x

)[∀x∀x

fx = fx

⇒ x = x

¦ = 1. By adjunc-
tion we know:
Hom
Sub(1)
(1, 1) =
Hom
Sub(1)
(∀x¦(x, x

)[∀x

fx = fx

⇒ x = x

¦ , ∀x¦(x, x

)[∀x

fx = fx

⇒ x = x

¦) ·
Hom
Sub(X)
(!
−1
∀x¦(x, x

)[∀x

fx = fx

⇒ x = x

¦ , ¦(x, x

)[∀x

fx = fx

⇒ x = x

¦) =
Hom
Sub(X)
(p
−1
1
1, ¦(x, x

)[∀x

fx = fx

⇒ x = x

¦) =
Hom
Sub(X)
(X, ¦(x, x

)[∀x

fx = fx

⇒ x = x

¦)
Since Hom
Sub(1)
(1, 1) contains exactly one morphism, so does Hom
Sub(X)
(X, ¦(x, x

)[∀x

fx = fx

⇒ x = x

¦).
Since X is the terminal object in Sub(X), we must have
_
(x, x

)[∀x

fx = fx

⇒ x = x

_
= X
.
In the same way we find:
_
(x, x

)[fx = fx

⇒ x = x

_
= X X.
Hence we conclude that the formula fx = fx

⇒ x = x

is true. This means (by example 4.2.2) that
¦(x, x

)[fx = fx

¦ ⊆ ¦(x, x

)[x = x

¦. Now consider:
A X
!
>
1
_
(x, x

)[fx = fx

_

.
.
.
.
.
.
.
.
.
>
>
X X

X

δ
X
>
(g,h)
>


Y


Y
>
Y Y
(fp
1
,fp
2
)

1
!

>

δ
Y

.
The two lower squares are pullbacks. We would like to prove f is monic, so suppose fg = fh. Then
(fp
1
, fp
2
) ◦ (g, h) factors through ∆
Y
. Since the upper square is a pullback, (g, h) factors through
¦(x, x

)[fx = fx

¦. With the inclusion ¦(x, x

)[fx = fx

¦ ⊆ ¦(x, x

)[x = x

¦ we find (g, h) factors through

X
. Hence g = h and f is monic.
The converse is also true. Suppose f is monic. Since ∀x and ∀x

are right adjoints, they preserve the
terminal object. Hence it is enough to check if fx = fx

⇒ x = x

is true. Consider:
X
f
>
Y
!
>
1
X X

X

(f,f)
>
Y Y

Y

δ
Y
>
Ω.
t

62
4.2. SOME EXAMPLES CHAPTER 4. THE MITCHELL-B
´
ENABOU LANGUAGE
The right square is a pullback by definition. By diagram chasing the left square is a pullback. The
PBL makes the whole rectangle is a pullback, so by the universal property of the subobject classifier
one has δ
Y
◦ (f, f) = δ
X
. Switching from characteristic morphisms to subobjects this equation becomes
¦(x, x

)[fx = fx

¦ = ¦(x, x

)[x = x

¦. So example 4.2.1 makes fx = fx

⇒ x = x

true. Since ∀x and ∀x

are right adjoints, they preserve the terminal object and hence ∀x∀x

fx = fx

⇒ x = x

is true.
Of course, the equivalent statement is true for epis. Clearly we could view ‘epi’ as a generalization of
‘surjective’ in the topos Set.
Example 4.2.3. Let f be a morphism between objects X and Y . Let x be a variable of type X and y a
variable of type Y . Then ∀y∃xfx = y is true if and only if f is epi.
Like in example 4.2.2, it suffices to prove that ∃xfx = y is true if and only if f is epi. This can be
expressed by the following diagram:
X
>
Y
!
>
1
Y
<
p
2
<
X Y
(Id
X
,f)

(fp
1
,Id
Y
p
2
)
>
Y Y

Y

δ
Y
>
Ω.
t

Now clearly, f is epi if and only if the left diagonal morphism is epi if and only if ∃xfx = y is true.
The next example is concerned with unique factorization. Suppose we have functions m : X → Z and
f : Y → Z in Set such that m is injective. Suppose furthermore ∀y∃x mx = fy. Then it is clear that f
factors uniquely through m. This can be generalized to an arbitrary topos.
Example 4.2.4. Let m : X → Z be a monic morphism and f : Y → Z a morphism. Suppose
[= ∀y∃x mx = fy.
Then there is a unique morphism g such that mg = f.
First, we make the meaning of the formula [= ∀y∃x mx = fy explicit. Like in example 4.2.2, it follows
that [= ∃x mx = fy.
¦y[∃x mx = fy¦
<<
¦(x, y)[mx = fy¦
me
>
Z
>
1
Y
Id


<
p
2
<
<
X Y
e

(mp
1
,fp
2
)
>
Z Z


δ
>
Ω.
t

The middle square is a pullback (by definition) and by a simple diagram chase it turns out that
¦(x, y)[mx = fy¦ is such that the following diagram is an equalizer:
¦(x, y)[mx = fy¦
e
>
X Y
mp
1
>
fp
2
>
Z.
Now look at
63
4.2. SOME EXAMPLES CHAPTER 4. THE MITCHELL-B
´
ENABOU LANGUAGE
E
Y
<
p
2
<
<
X Y
e

p
1
>
X
!
X
>
1
>
Z
m

χm
>
f
>
Ω.
t

Remark that χ
m
fp
2
e = χ
m
mp
1
e = t!
X
p
1
e = t!
Y
p
2
e, where !
Y
is the unique morphism from Y to 1.
Since p
2
e is epi, χ
m
f = t!
Y
. Hence there is a unique morphism g : Y → X such that the diagram above
commutes (g is not indicated in the diagram). This morphism makes the following diagram commute:
X

m
>
Z
Y
g

f
>
.
Since m is monic, this g is unique.
Finally, we show that limits in a topos are generalizations of the limits in Set. That is, in Set one can
write the limit of a diagram as a subobject of the product of the objects in that diagram. The subobject
is implied by the morphisms in the diagram and can be expressed by a characteristic morphism. Since
subobjects and characteristic morphisms are generalized from a set to a topos it turns out that limits
can be defined in exact the same way. We give two examples, pullbacks and equalizers.
Example 4.2.5. The following diagram is a pullback:
¦(x, y)[fx = fy¦
>
X
Y

g
>
Z.
f

Note that this is exactly the pullback of Set. That is, pulling back f along g yields the set of pairs
(x, y) ∈ X Y such that fx = gy. In previous chapter we gave the expression fx = gy an interpretation
in an arbitrary topos, which is:
¦(x, y)[fx = fy¦
>
Z
>
1
U
.....................
(h,k)
>
X Y

(fp
1
,gp
2
)
>
Z Z

Z

δ
Z
>
Ω.

Suppose there is an object U and a morhpism (h, k) such that fh = gk. Then (fp, gq) ◦ (h, k) factors
through ∆
Z
. The left square is a pullback, hence there is a unique morphism from U to ¦(x, y)[fx = fy¦
such that the diagram commutes. This proves that ¦(x, y)[fx = fy¦ is the pullback of f along g.
64
4.3. SOME FURTHER EXAMPLES CHAPTER 4. THE MITCHELL-B
´
ENABOU LANGUAGE
Example 4.2.6. ¦x[fx = gx¦ is the equalizer of the diagram:
X
f
>
g
>
Y.
The proof is similar as in example 4.2.5.
4.3 Some further examples
The aim of this section is to establish some techniques to handle the formulae that are introduced in the
beginning of this chapter. In the previous section many examples were solved by diagram chasing. In
this section we introduce theorems which enable you to solve problems using logical rules. This will lead
to much shorter and more elegant solutions to many problems. Also, it will give a good logical tool to
prove deeper results of the Mitchell-B´enabou language. The aim of this chapter is not to give a general
overview of true formulae. We restrict ourselves to some elegant examples.
We start with a proposition which has already been proved in example 4.2.1. For convenience we quote
it again.
Proposition 4.3.1. Let φ(x) : X → Ω and ψ(x) : X → Ω be formulae. Then ¦x[φ(x)¦ ⊆ ¦x[ψ(x)¦ if
and only if [= φ(x) ⇒ ψ(x).
Proof: see example 4.2.1.
Let’s start with some simple formulae.
Theorem 4.3.2. Let φ(x) and ψ(x) be formulae. Then the following formulae are true:
1. [= φ(x) ∧ ψ(x) ⇒ φ(x).
2. [= φ(x) ∧ ψ(x) ⇒ ψ(x).
3. [= φ(x) ⇒ φ(x) ∨ ψ(x).
4. [= ψ(x) ⇒ φ(x) ∨ ψ(x).
Proof: According to proposition 4.3.1, proving that one of the formulae above is true amounts to proving
an inclusion:
1.
¦x[φ(x) ∧ ψ(x)¦ ⊆ ¦x[φ(x)¦ iff
¦x[φ(x)¦ ∩ ¦x[ψ(x)¦ ⊆ ¦x[φ(x)¦ .
2.
¦x[φ(x) ∧ ψ(x)¦ ⊆ ¦x[ψ(x)¦ iff
¦x[φ(x)¦ ∩ ¦x[ψ(x)¦ ⊆ ¦x[ψ(x)¦ .
3.
¦x[φ(x)¦ ⊆ ¦x[φ(x) ∨ ψ(x)¦ iff
¦x[φ(x)¦ ⊆ ¦x[φ(x)¦ ∪ ¦x[ψ(x)¦ .
65
4.3. SOME FURTHER EXAMPLES CHAPTER 4. THE MITCHELL-B
´
ENABOU LANGUAGE
4.
¦x[ψ(x)¦ ⊆ ¦x[φ(x) ∨ ψ(x)¦ iff
¦x[ψ(x)¦ ⊆ ¦x[φ(x)¦ ∪ ¦x[ψ(x)¦ .
Each time, the last inclusion is obviously true.
If φ(x
1
, . . . , x
n
) is a formula with variables x
1
, . . . , x
n
, there is a way of ‘extending its source’, i.e. turning
it into a formula of more variables: φ(x
1
, . . . , x
n
, x
n+1
, . . . , x
m
), m > n. This is done by the following
diagram:
φ(x
1
, . . . , x
m
) :=X
1
. . . X
m
p
>
X
1
X
n
φ(x
1
,...,xn)
>
Ω,
where p is the obvious projection.
Lemma 4.3.3. Let φ(x
1
, . . . , x
n
) be a formula with variables x
1
, . . . , x
n
. Let φ(x
1
, . . . , x
m
), m > n be
its extension. Then:
1. If [= φ(x
1
, . . . , x
n
), then [= φ(x
1
, . . . , x
m
).
2. If the projection X
1
. . . X
n
→ X
1
. . . X
m
is epi and [= φ(x
1
, . . . , x
m
), then [= φ(x
1
, . . . , x
n
).
Proof:
1. The statement is obvious from the following diagram:
1
X
1
. . . X
m
p
>
X
1
. . . X
n
φ(x
1
,...,xn)
>
>
Ω.

If φ(x
1
, . . . , x
n
) factors through 1, so does φ(x
1
, . . . , x
n
) ◦ p.
2. We have:
1
X
1
. . . X
m
p
>
!
X
1
×...×Xm
>
X
1
. . . X
n
φ(x
1
,...,xn)
>
!
X
1
×...×Xn
>
Ω.
t

Now,
φ(x
1
, . . . , x
n
) ◦ p = t◦!
X
1
×...×Xm
= t◦!
X
1
×...×Xn
◦ p.
If p is epi, we find:
φ(x
1
, . . . , x
n
) = t◦!
X
1
×...×Xn
.
66
4.3. SOME FURTHER EXAMPLES CHAPTER 4. THE MITCHELL-B
´
ENABOU LANGUAGE
The following theorem is a very powerful one. It proves that the Modus Ponens is often true in topos
logic. For many topoi the conditions for the theorem are easily checked. For example the topoi Set and
Set
C
satisfy these conditions. We refer to topoi in which the Modus Ponens is true as topoi with Modus
Ponens.
Theorem 4.3.4. (Modus Ponens). Suppose φ and ψ are formulae. Let X
1
, . . . , X
n
be the variables of
ψ and let X
1
, . . . , X
m
be the variables of φ and ψ. Suppose the projection X
1
. . . X
m
→ X
1
. . . X
n
is epi. If [= φ and [= φ ⇒ ψ then [= ψ.
Proof: First, extend both φ and ψ to the variables X
1
, . . . , X
m
. Call the extensions
¯
φ and
¯
ψ. Remark
that lemma 4.3.3 gives [=
¯
φ and [=
¯
φ ⇒
¯
ψ, which is equivalent to
_
(x
1
, . . . , x
m
)[
¯
φ(x
1
, . . . , x
m
)
_
= 1 and
_
(x
1
, . . . , x
m
)[
¯
φ(x
1
, . . . , x
m
)
_

_
(x
1
, . . . , x
m
)[
¯
ψ(x
1
, . . . , x
m
)
_
. From this, we conclude
_
(x
1
, . . . , x
m
)[
¯
ψ(x
1
, . . . , x
m
)
_
= 1.
Hence [=
¯
ψ. Now, lemma 4.3.3 gives [= ψ.
Suppose we want to prove a statement of the form
‘If [= φ, then [= ψ’.
In a topos with Modus Ponens, it then is enough to check whether the formula φ ⇒ ψ is true. In general
this statement is easier to prove, since it amounts to proving an inclusion of subobjects. Thanks to the
Modus Ponens the following theorem becomes useful.
Theorem 4.3.5. Let φ(x
1
, . . . , x
n
) be a formula with variables X
1
, . . . , X
n
and let φ(x, x
1
, . . . , x
n
) be its
extension to variables X, X
1
, . . . , X
n
. Let p : X X
1
. . . X
n
→ X
1
X
n
be the obvious projection.
Let f : Y → X be a morphism. Then:
1. [= φ(x
1
, . . . , x
n
) ⇒ (∀xφ(x, x
1
, . . . , x
n
)).
2. [= (∀xφ(x, x
1
, . . . , x
n
))(x, x
1
, . . . , x
n
) ⇒ φ(x, x
1
, . . . , x
n
),
where the part to the left of ⇒ should be regarded as the extension of the formula ∀xφ(x, x
1
, . . . , x
n
)
to the variables x, x
1
, . . . , x
n
.
3. If [= φ(x, x
1
, . . . , x
n
), then [= φ(fy, x
1
, . . . , x
n
).
Proof:
1.
[= φ(x
1
, . . . , x
n
) ⇒ (∀xφ(x, x
1
, . . . , x
n
)) iff
¦x
1
. . . x
n
[ φ(x
1
, . . . , x
n
)¦ ⊆ ¦x
1
. . . x
n
[ ∀xφ(x, x
1
, . . . , x
n
)¦ iff
p
−1
¦x
1
. . . x
n
[ φ(x
1
, . . . , x
n
)¦ ⊆ ¦x
1
. . . x
n
[ ∀xφ(x, x
1
, . . . , x
n

The last inclusion holds, since it actually is an equality.
2.
[= (∀xφ(x, x
1
, . . . , x
n
))(x, x
1
, . . . , x
n
) ⇒ φ(x, x
1
, . . . , x
n
) iff
¦x x
1
. . . x
n
[ (∀xφ(x, x
1
, . . . , x
n
))(x, x
1
, . . . , x
n
)¦ ⊆ ¦x x
1
. . . x
n
[ φ(x, x
1
, . . . , x
n
)¦ iff
¦x x
1
. . . x
n
[ (∀xφ(x, x
1
, . . . , x
n
))(x, x
1
, . . . , x
n
)¦ ⊆ ¦x x
1
. . . x
n
[ φ(x, x
1
, . . . , x
n

67
4.4. PARTIAL TRUTH CHAPTER 4. THE MITCHELL-B
´
ENABOU LANGUAGE
3. The proof is exactly the same as in lemma 4.3.3 (2), with p replaced by f Id
X
1
×...×Xn
.
The following proposition will be of great use too:
Proposition 4.3.6. Let f : X → Y and g : X → Y be morphisms. Then, in a topos with Modus Ponens
f = g (equality as morphisms!) if and only if:
[= ∀xfx = gx
Proof: This is almost immediate from the universal property of a pullback:
Y
O
Y
>
1
X
(f,g)
>
.........................................
>
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
>
Y Y

Y

δ
Y
>
Ω.
t

The left dotted morphism exists such that the diagram commutes if and only if the right dotted morphism
exists such that the diagram commutes. The existence of the left dotted morphism is equivalent to f = g
(equality as morphisms). The existence of the right dotted morphism is equivalent to [= fx = gx (as a
formula). Theorem 4.3.5 now gives f = g if and only if [= ∀xfx = gx.
Now let us examine example 4.2.2 again.
Example 4.3.7. In a category with Modus Ponens, [= ∀x∀x

fx = fx

⇒ x = x

if and only if f is
monic.
First the only if part. Suppose fh = fg, then [= fhx = fgx according to theorem 4.3.6. From
[= ∀x∀x

fx = fx

⇒ x = x

we obtain [= fx = fx

⇒ x = x

according to theorem 4.3.5 (2). So
[= fhy = fgy ⇒ hy = gy by applying theorem 4.3.5 (3) twice. Modus Ponens gives [= hx = gx and
hence h = g.
The converse is probably best done in previous section.
4.4 Partial truth
In this section we briefly describe the notion of partial truth, since in the next chapter we will refer to
some articles which make use of this construction. Since we will not use the details of these articles, we
just give the definition. Some important theorems about partial truth can be found in [11]. In particular
the Kripke-Joyal semantics form an important tool for problems involving partial truth.
Definition 4.4.1. Let φ(x
1
, . . . , x
n
) : X
1
, . . . , X
n
→ Ω be a formula in a topos. Let U be an object of the
topos and α : U → X
1
. . . X
n
a morphism. We write:
U [= φ(α)
if there exists a morphism ˜ α such that the following diagram commutes.
¦(x
1
, . . . , x
n
)[φ(x
1
, . . . , x
n

>
1
˜ α
U
α
>
....................................
>
X


φ(x)
>
Ω.


68
4.5. INTERNAL CATEGORIES CHAPTER 4. THE MITCHELL-B
´
ENABOU LANGUAGE
Sometimes we abbreviate this as U [= φ.
The following theorem shows the link to the previous section.
Theorem 4.4.2. Let φ(x
1
, . . . , x
n
) : X
1
, . . . , X
n
→ Ω be a formula in a topos. Then [= φ if and only if
U [= φ(α) for all objects U and morphisms α.
Proof: Suppose that [= φ. Then the pullback property of the (right) square in the diagram below tells
us there is a (dotted) morphism ˜ α for every α.
¦(x
1
, . . . , x
n
)[φ(x
1
, . . . , x
n

>
1
˜ α
U
α
>
.....................................
>
X


φ(x)
>
>
Ω.


Conversely, suppose U [= φ(α) for every U and α then certainly X
1
. . . X
n
[= φ(Id
X
1
×...×Xn
). Hence
φ factors through 1.
A formula φ might not be true for all U, but if one puts a U with a morphism in front of it, it becomes
‘true on U’, i.e. U [= φ. For deeper results about partial truth, see for example [5] or [11].
4.5 Internal categories
The Mitchell-B´enabou language forms an important tool for generalizing objects that are defined as a
set with a certain structure. For example, a group is a set togheter with an addition + and a special
element 0, satisfying the properties:
[= ∀a∀b∀c (a +b) +c = a + (b +c) ;
[= ∀a (a + 0) = 0 ;
[= ∀a∃b (a +b) = 0 .
Since these properties, expressed in the Mitchell-B´enabou-language, have a (generalized) interpretation
in any topos, there is no need to require that a group needs to be a set with some additional structure.
It is enough to require that a group is an object G within a topos T , together with morphisms:
0 : 1 → G;
+ : B B → B,
such that these morphisms satisfy the above properties written in the Mitchell-B´enabou language. Such
an object G is called an internal group. In an analogous way, one can define internal fiels, internal
lattices, et cetera. In chapter 1, we introduced the rational numbers object Q and like one would expect,
one can proof that this is an internal field. More general, there is a notion of an internal category whose
the definition can be found in [8]. This definition is quite abstract and we rather give specific definitions
of internal lattices, internal locales, et cetera.
69
Chapter 5
The construction of the spectrum
“The mathematician is fascinated with the marvelous beauty of the forms he constructs, and in their
beauty he finds everlasting truth.” - G.B. Shaw
In this chapter we turn our attention to a generalization of the Gelfand theorem to topos theory. The
original theorem proved by I.M. Gelfand and M. Naimark shows that there is a duality between unital
commutative C*-algebras and compact Hausdorff spaces (in Set). If we consider the locale formed by
the underlying topology of a compact Hausdorff space instead of the space itself, this duality becomes
one between unital commutative C*-algebras and compact completely regular locales.
cCStar ↔ KRegLoc
By the work of Mulvey and Banaschewski this duality may be generalized to any topos T , see [12], [13]
and [14]. This is done by introducing a generalization of a C*-algebra to the topos T , i.e. an object
of T which has all the properties of a C*-algebra (we state the precise definition later in this chapter),
called an internal C*-algebra of T . Also the definition of a locale may be generalized to an object called
an internal locale of T . Mulvey and Banaschewski showed that there is a duality between the internal
C*-algebras and the internal compact completely regular locales of any topos T .
In this thesis we study an article by Heunen, Landsman and Spitters [7]. The authors start with a
unital C*-algebra A in Set and construct a particular topos T (A) from this C*-algebra. Furthermore,
they define an internal C*-algebra of T (A), called
¯
A. In the next two chapters we try to compute the
compact completely regular locale corresponding to
¯
A by the morphism cCStar → KRegLoc. This
locale is called the spectrum of A and is denoted as Σ(A), or briefly Σ, since A is fixed from the start
of the construction. In this chapter we write out the details of the construction of the spectrum of the
C*-algebra A and show that every step is well-defined. The construction consists of the following steps:
1. In section 5.1, we define the topos T (A) and the internal C*-algebra
¯
A.
2. In section 5.2 we introduce some theory about lattices in topoi and construct an internal lattice in
a topos, called L

¯
Asa
. We first consider a generalization of a free distributive lattice to topos theory
(paragraph 5.2.1) and then turn our attention to the more general case of generalized distributive
lattices freely generated by an object subject to relations in a topos. Paragraph 5.2.3 should be
considered an intermezzo: this paragraph only considers lattices in Set and is heavily based on [2]
and [3]. This paragraph is included to show that the constructions that follow are well defined.
Furthermore, it gives a nice representation of the lattices that we are interested in, in Set.
70
5.1. THE TOPOS T (A) CHAPTER 5. THE CONSTRUCTION OF THE SPECTRUM
3. Finally, in section 5.3 we introduce a morphism / : Ω
L

¯
Asa
→ Ω
L

¯
Asa
and define the spectrum Σ(A)
as a certain subobject of Ω
L

¯
Asa
.
During the construction we will introduce some theorems that allow us to compute the spectrum. The
Mitchell-B´enabou language will be of great use. However, we sometimes avoid some technicalities of this
language by making use of elementary category theory or lattice theory.
The actual computations are done in chapter 6, where our main examples will be the spectra of the
C*-algebras C
2
and M(2, C), the 2 2-matrices. We will spend some words on C, C
n
and C(X), with
X a compact Hausdorff space, too.
5.1 The topos T (A)
The construction of the spectrum starts with a unital C*-algebra A, which is defined as a set with the
usual structure of a Banach Space over C with unit and involution called

satisfying |a

a| = |a|
2
.
The reason we emphasize that this construction has a set as input is that we are going to construct (i.e.
define) a C*-algebra that is not a set anymore, but a generalization of a set: an object in a certain topos.
In what follows A will always be our unital C*-algebra as a set, unless specified otherwise.
Now we introduce ((A) as the following preorder:
((A) = ¦B [ B is a unital commutative C*-subalgebra of A¦ . (5.1)
Here the unit in B is supposed to be inherited from A. The partial order of ((A) is given by the inclusion.
This partial order has a meet, given by the intersection of unital, commutative C*-subalgebras. It does
not have a join, since the join of two commutative C*-subalgebras does not have to be commutative.
An important remark is that ((A) has a bottom element, which is the unique commutative, unital
C*-subalgebra of dimension 1: C 1. The topos we are interested in will be the following one.
Definition 5.1.1. Let T (A) denote the topos Set
C(A)
, where ((A) is the categorical preorder (5.1).
In the previous chapters we proved that T (A) is a topos. The partial order and the bottom element of
((A) will turn T (A) into a topos with nice properties, as we will see. The next step is to introduce a
special object in this topos, which is called the tautological functor.
Definition 5.1.2. Let
¯
A : ((A) → Set denote the functor that sends a unital, commutative C*-subalgebra
B ⊆ A to itself. I.e. it is defined on objects as:
¯
A : B → B.
¯
A is defined on morphisms (in the preorder ((A)) as the inclusion of C*-subalgebras.
What is so interesting about this tautological object? First, it turns out to be an internal C*-algebra in
the topos T (A). We introduce the internal C*-algebra in an analogous way to the internal categories of
section 4.5. The only problem is the analytic part of the definition of a C*-algebra: the norm and its
completeness. We will not cover this in full detail, but only elaborate a bit.
A norm of a C*-algebra A in Set can be regarded as a subset N of AQ
+
, where one requires (a, q) ∈ N
if and only if |a| < q. Since this concept is set theoretic in nature, it may be generalized to a topos that
71
5.1. THE TOPOS T (A) CHAPTER 5. THE CONSTRUCTION OF THE SPECTRUM
has a rational numbers object Q. Recall that we defined the rational numbers object in chapter 1. Then
define
Q
+
=
_
q [ ∃r (q = r
2
)
_
⊆ Q.
Now a norm on A is a subobject of A Q
+
satisfying certain axioms. These axioms can be found in
[14] as well as the axioms for an internal C*-algebra to be complete. Now, we state the definition of an
internal C*-algebra.
Definition 5.1.3. An internal C*-algebra in a topos T is an object B of the topos, together with
morphisms
0 : 1 → B ;
λ : C B → B (scalar multiplication);
: B B → B (multiplication);
+ : B B → B ;
∗ : B → B ,
such that:
[= ∀a∀b∀c (a +b) +c = a + (b +c) ;
[= ∀a∀b a +b = b +a ;
[= ∀a 0 +a = a ;
[= ∀a∃b a +b = 0 ;
[= ∀a∀µ∀ν λ(µ, λ(ν, a)) = λ(µν, a) ;
[= ∀a∀b∀µ λ(µ, a +b) = λ(µ, a) +λ(µ, b) ;
[= ∀a∀b∀c (a b) c = a (b c) ;
[= ∀a∀b∀c a (b +c) = (a b) + (a c) ;
[= ∀a a
∗∗
= a ;
[= ∀a∀b (a +b)

= a

+b

;
[= ∀a∀b (ab)

= b

a

;
[= ∀a∀µ λ(µ, a)

= λ(¯ µ, a

) .
Moreover, B has a norm N ⊆ B Q
+
, in which it is complete.
Furthermore, we call the C*-algebra B commutative if
[= ∀a∀b a b = b a.
We say the C*-algebra B is unital if there is a morphism 1 : 1 → B such that
[= ∀a ((1 a = a) ∧ (a 1 = a)).
It is very easy to verify that
¯
A as in definition 5.1.2 satisfies the axioms of definition 5.1.3 for the (point-
wise) addition, multiplication and scalar multiplication. For the completeness part we refer to [7], where
it is proved that
¯
A is complete using the Kripke-Joyal semantics. It is easy to see that
¯
A is commutative
and has a unit (since we have chosen all algebras of ((A) to be commutative and unital). We state the
result as a theorem.
Theorem 5.1.4.
¯
A is an internal commutative unital C*-algebra of T (A).
Proof: See [7].
72
5.2. FREE LATTICES CHAPTER 5. THE CONSTRUCTION OF THE SPECTRUM
5.2 Free lattices
The next step of constructing the spectrum is to calculate the free distributive lattice (with top and
bottom) generated by
¯
A subject to certain relations which we will introduce later. First, we study the
free distributive lattice generated by
¯
A without subjecting it to any relation since this object turns out
to be quite interesting. Recall that a lattice is always assumed to have a top and a bottom element.
5.2.1 Construction of free distributive lattices
For a free distributive lattice in a topos we need the notion of an internal distrivutive lattice. The
definition is analogous to the one of an internal Heyting algebra and the one of an internal C*-algebra.
An internal distrivutive lattice in a topos T is an object L of the topos together with morphisms
∨ : L L → L;
∧ : L L → L;
· : 1 → L;
⊥ : 1 → L,
satisfying the obvious expressions in the Mitchell-B´enabou language for associativity, commutativity, the
unit law, idempotence and distributivity for both ∨ and ∧. Furthermore, it satisfies the absorption laws.
In Set these relations reduce to the familiar definition of a distributive lattice. Now, we formally state
the definition of a free distributive lattice in a topos.
Definition 5.2.1. Let T be a topos, and let S be an object in that topos. An internal distributive lattice
L
S
is called freely generated by S if there is a morphism i : S → L
S
such that for every internal distributive
lattice M and morphism f : S → M there is a unique morphism g : L
S
→ M such that the following
diagram commutes:
S
f
>
M
L
S
i

g
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
>
and such that g commutes with the internal lattice structure of L
S
and M, i.e. the following diagrams
commute:
L
S
L
S

>
L
S
M M
g×g


>
M
g

L
S
L
S

>
L
S
M M
g×g


>
M
g

L
S
g
>
M
1



>
L
S
g
>
M
1

>
.
73
5.2. FREE LATTICES CHAPTER 5. THE CONSTRUCTION OF THE SPECTRUM
The choice of the notation i for the map between the object and the free distributive lattice it generates
may remind the reader of an inclusion. In many cases it will be an ‘inclusion’, i.e. the morphism is monic.
However, after we subject the lattice to relations, i will not be monic anymore. Nevertheless, we will still
speak about the inclusion in the free distributive lattice, since this is a convenient way of indicating that
a variable has to be regarded as an element of the free distributive lattice.
As an example we calculate the free distributive lattice generated by a set S in the topos Set. We claim
the lattice obtained by the following procedure is the distributive lattice freely generated by S. Construct
a set L inductively:
1. S ⊆ L. ⊥, · ∈ L (as formal symbols).
2. If φ and ψ are elements of L, so are the formal expressions φ ∨ ψ and φ ∧ ψ.
Now divide out the following equational laws in L.
commutativity : φ ∧ ψ = ψ ∧ φ
associativity : φ ∧ (ψ ∧ ρ) = (φ ∧ ψ) ∧ ρ
unit law : φ ∧ · = φ
idempotence : φ ∧ φ = φ
commutativity : φ ∨ ψ = ψ ∨ φ
associativity : φ ∨ (ψ ∨ ρ) = (φ ∨ ψ) ∨ ρ
unit law : φ ∨ ⊥ = φ
idempotence : φ ∨ φ = φ
absorbtion : φ ∨ (φ ∧ ψ) = φ
φ ∧ (φ ∨ ψ) = φ
distributivity : φ ∨ (ψ ∧ ρ) = (φ ∨ ψ) ∧ (φ ∨ ρ)
Define L
S
= L/ ∼
l
, where the ∼
l
stands for the equivalence relation generated by above relations. For
convenience of notation we do not add extra notation to elements of L
S
to indicate that its elements are
actually equivalence classes. The internal lattice structure is given by:
bottomelement ⊥
top element ·
meet ∧ : L
S
L
S
→ L
S
: (φ, ψ) → φ ∧ ψ where the last wedge is defined as above
join ∨ : L
S
L
S
→ L
S
: (φ, ψ) → φ ∨ ψ where the last vee is defined as above.
It is important to remark that one has to prove that this defines a distributive lattice. This is done in
[15]. Now if we have a diagram in Set like:
S
f
>
M
L
S
i

g
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
>
,
where i is the obvious inclusion, then we can define g inductively by:
74
5.2. FREE LATTICES CHAPTER 5. THE CONSTRUCTION OF THE SPECTRUM
1. Define
g : s → f(s) s ∈ S;
g : · → · ;
g : ⊥ → ⊥.
2. If φ(s
1
, . . . , s
n
, ⊥, ·) ∈ L
S
is an expression in which the elements s
1
, . . . , s
n
∈ S and possibly ⊥ or
· appear, define
g : φ(s
1
, . . . , s
n
) → φ(g(s
1
), . . . , g(s
n
)).
The hard part is proving that the function g is well defined. This relates to the problem of determining
which of the formal expressions that are defined above, are equivalent. This problem is known as the
‘word problem’ (for free lattices) and has been solved. Indeed one can prove that g is well defined. It is
obvious that f = gi. By definition g is indeed a morphism of lattices (i.e. it preserves meets, joins, ·
and ⊥). Furthermore, g is unique in having this property, since there is no free choice for the initial step
1 and the inductive step 2 of the definition of g in order to let g respect the lattice structure of L
S
and
M. We restate the result:
Theorem 5.2.2. In Set the distributive lattice freely generated by a set S is defined as the object L
S
above.
The theorem may not be surprising, but it is of great use to compute free lattices generated by an object
in Set
C
. The following theorem shows that a free lattice in this category can easily be computed, namely
point-wise (sometimes called locally).
Theorem 5.2.3. Let ( be a category. In the topos Set
C
the free distributive lattice generated by an object
S can be computed point-wise. That is, for any object C ∈ (, L
S
(C) = L
S(C)
and the internal meets and
joins are defined by the local meets and joins.
Proof:
1. First we must define the intrinsic maps f
L
S
C,D
. Recall that this notation is defined as f
L
S
C,D
= L
S
(f),
where f is a morphism in ( from C to D. Since L
S
(D) = L
S(D)
is a distributive lattice in Set by
definition of L
S(D)
and L
S
(C) = L
S(C)
is the free distributive lattice generated by S(C), there is a
unique map f
L
S
C,D
as in the diagram below such that f
L
S
C,D
preserves the meet and join in Set:
S(C)
f
S
C,D
>
S(D)
L
S(C)
i
S(C)

......... >
L
S(D)
.
i
S(D)

Remark that by uniqueness of this map and some diagram chasing the following equations hold.
Id
L
S
C,C
= Id
L
S
(C)
;
g
L
S
D,E
◦ f
L
S
C,D
= (g ◦ f)
L
S
C,E
.
Hence the assignment (f : C → D) → f
L
S
C,D
is functorial.
75
5.2. FREE LATTICES CHAPTER 5. THE CONSTRUCTION OF THE SPECTRUM
2. The meet and join in L
S
commute with the intrinsic maps f
L
S
C,D
. Since this meet and join are
defined locally in L
S
this claim becomes a tautology; we actually defined the maps f
L
S
C,D
in such a
way they commute with the meet and join.
3. Now suppose M is an internal lattice in Set
C
.
S
f
>
M
L
S
i

g
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
>
.
We have to prove there exists a unique g such that this diagram commutes. The uniqueness is the
easy part. For if there exists such a morphism g, the diagram must at least commute locally, i.e.
S(C)
f(C)
>
M(C)
L
S
(C)
i
∨ g(C)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
>
,
commutes. Note that M(C) is a distributive lattice in Set, since the local meets and joins inherited
from M equip M(C) with a distributive lattice structure. For every C there exists at most one
such g(C) since M(C) is a distributive lattice in Set and L
S
(C) is the distributive lattice freely
generated by S(C) in Set.
For the existence of g, note that we can define a global map by combining the unique local maps
which arise from the local diagrams in Set. If this is a well-defined map it automatically commutes
with the meet and join in L
A
and M, since it does commute locally. In order to let this map g be
well-defined, we need to check if it commutes with the intrinsic maps of L
S
and M, i.e. for any
morphism k : C → D in ( the following equation should hold.
g(D)k
L
S
C,D
= k
M
C,D
g(C) (5.2)
Both the left hand side and right hand side are maps from L
S
(C) to M(D). The left hand side is
the unique map arising from
L
S
(C)
S(C)

>
L
S
(C)
k
L
S
C,D
>
L
S
(D)
g(D)
>
M(D),
.............................................................. >
(5.3)
and the right hand side is the unique map arising from
L
S
(C)
S(C)

>
L
S
(C)
g(C)
>
M(C)
k
M
C,D
>
M(D).
............................................................. >
(5.4)
76
5.2. FREE LATTICES CHAPTER 5. THE CONSTRUCTION OF THE SPECTRUM
For x ∈ S(C), one has
g(D)k
L
S
C,D
i(C)(x) = g(D)i(D)k
S
C,D
(x)
= f(D)k
S
C,D
(x)
= k
M
C,D
f(C)(x)
= k
M
C,D
g(C)i(C)(x).
So the composition of the horizontal arrows in (5.3) and (5.4) are equal. The uniqueness of the
dotted arrow then results in (5.2).
5.2.2 Relations and the lattice L

¯
Asa
For the construction of the spectrum we first have to take the self-adjoint part of our object
¯
A, which is
defined as the following subobject of
¯
A:
¯
A
sa
= ¦a[a

= a¦ .
It follows immediately from the fact that the adjoint

is defined locally, that
¯
A is the local self-adjoint
part of every C*-subalgebra B of A. That is,
¯
A
sa
is given on objects as:
B → B
sa
.
In this paragraph we are going to define the distributive lattice freely generated by the object
¯
A
sa
subject to relations. This turns out to be an object, called L

¯
Asa
, together with a morphism, called
i

:
¯
A
sa
→ L

¯
Asa
. Suppose a is a variable of type
¯
A (see definition 4.1.1 for its interpretation), then we
write D
a
for the term i

a. Now the relations we need in order to define the spectrum are the following
terms of the Mitchell-B´enabou-language.
[= D
1
= · (5)
[= ∀a D
a
∧ D
−a
= ⊥ (6)
[= ∀a D
−a
2 = ⊥ (7)
[= ∀a∀b D
a+b
≤ D
a
∨ D
b
(8)
[= ∀a∀b D
ab
≤ (D
a
∧ D
b
) ∨ (D
−a
∧ D
−b
). (9)
By definition of ≤, the last two relations are equivalent to:
[= ∀a∀b D
a+b
∨ (D
a
∨ D
b
) = D
a
∨ D
b
(8)
[= ∀a∀b D
ab
∨ ((D
a
∧ D
b
) ∨ (D
−a
∧ D
−b
)) = (D
a
∧ D
b
) ∨ (D
−a
∧ D
−b
) (9).
How is this lattice defined? The reader may, meanwhile have acquired some intuition about this.
Definition 5.2.4. The free distribuitive lattice generated by
¯
A
sa
subject to relations (5) - (9) is an
internal distributive lattice L

¯
Asa
that satisfies (5) - (9)together with a morphism:
¯
A
sa
i

>
L

¯
Asa
.
Furthermore, if there is another internal distributive lattice M satisfying (5) - (9) together with a map
f :
¯
A
sa
→ M, then f factors uniquely through i

by a morphism of lattices.
77
5.2. FREE LATTICES CHAPTER 5. THE CONSTRUCTION OF THE SPECTRUM
Remarks and notation: Of course this definition can be generalized to distributive lattices generated
by an arbitrary object subject to an arbitrary finite number of relations in an arbitrary topos. For
any such object we adopt the notation D
a
to indicate the term i

a, specifying which relations, which
generating object and which topos define the morphism i

. Especially, we are going to use both the
interpretation in Set and its generalization to T (A) a lot.
Here we study the relations (5) - (9). We first prove that in Set the distributive lattice freely generated
by a set S subject to (5) - (9) always exists.
Theorem 5.2.5. In Set the lattice freely generated by a set S subject to relations (5) - (9) exists.
Proof: First, construct L
S
, the free lattice generated by S. Then find the smallest equivalence relation
∼ on L
S
such that L
S
/ ∼ is a lattice when it inherits the lattice structure of L
S
and such that L
S
/ ∼
satisfies the relations (5) - (9). Suppose there is another lattice M subject to the given relations. Then,
by definition of the free lattice, there exists a morphism L
S
→ M which respects the lattice structure
of L
S
and M. By definition of the equivalence relation, if two elements of L
S
are equivalent by ∼, then
their images under L
S
→ M are equal. Hence there is a unique lattice homomorphism (L
S
/ ∼) → M
making the diagram below commutative:
S
>
M
L
S

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
>
L

S

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
>
.
Now note that by theorem 4.3.6 relations (5) - (9) can be expressed as equalities of morphisms in any
topos. For example, relation (7) is equivalent to the following diagram being commutative:
¯
A
sa
−.
2
>
¯
A
sa
1
!


>
L ¯
Asa
.
i

All the other relations can be expressed in this way, too. The next theorem shows that the free distributive
lattice generated by
¯
A
sa
subject to relations (5) - (9) can be computed point-wise. The proof is similar
to theorem 5.2.3, but involves some extra diagram chasing. For our convenience we will only subject the
free distributive lattice in the proof below to relation (7). There is no special reason why this relations
is chosen. The proof for subjecting the lattice to all the relations is analogous to the one below. The
reader can easily work out the details on a huge amount of scratch paper.
Theorem 5.2.6. Let S be an object of Set
C
and let α : S → S be a morphism. Then L

S
, the distributive
lattice generated freely by S subject to relation
78
5.2. FREE LATTICES CHAPTER 5. THE CONSTRUCTION OF THE SPECTRUM
D
α(s)
= ⊥, (5.10)
can be computed locally. That is, for any object C ∈ (, L

S
(C) = L

S(C)
and the internal meets and joins
are defined by the local meets and joins.
Proof: The proof is similar to theorem 5.2.3. Notice that α will play the role of the function s → −s
2
.
Furthermore, theorem 5.2.5 proves that the local distributive lattices generated by the local sets subject
to relation (5.10) exist. We need three steps:
1. First we must define the internal maps f
L

S
C,D
S(C)
f
S
C,D
>
S(D)
L

S
(C)
i(C)

..................................................... >
L

S
(D)
i(D)

i

(C) i

(D)
1(C)
!
>
⊥(C)
<
1(D)
⊥(D)
<
S(C)
f
S
C,D
>
>
S(D)
>
(α)(C) (α)(D)
S(C)
f
S
C,D
>
!
>
<
S(D)
!
>
<
.
We know that the non-dotted arrows form a commutative diagram. Note that this commutativity is
exactly the condition stating that there exists a unique morphism f
L

S
C,D
as indicated in the diagram.
By uniqueness and an extensive diagram chase we find:
Id
L

S
C,C
= Id
L

S
(C)
;
g
L

S
D,E
◦ f
L

S
C,D
= (g ◦ f)
L

S
C,E
.
Hence the assignment (f : C → D) → f
L

S
C,D
is functorial.
2. f
L

S
C,D
is a lattice homomorphism by construction.
3. Now suppose M is an internal lattice in Set
C
and f : A → M is a morphism such that the right
diagram below commutes.
S
f
>
M
L

S
i


g
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
>
S
α
>
S
1
!


>
M
f

.
79
5.2. FREE LATTICES CHAPTER 5. THE CONSTRUCTION OF THE SPECTRUM
We have to prove that there exists a unique g such that the left diagram above commutes. First we
prove uniqueness. If there exists such a morphism g, the diagrams must at least commute locally,
i.e.
S(C)
f(C)
>
M(C)
L
S
(C)
i

(C)
∨ g(C)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
>
S(C)
α(C)
>
S(C)
1(C)
!

⊥(C)
>
M(C)
f(C)

,
commute. For every C there exists at most one such g(C), since M(C) is a lattice subject to
relation (5.10) in Set and L

S
(C) is the lattice freely generated by S(C) subjected to relation (5.10)
in Set.
For the existence of g, note that we can define a global map by combining the local maps obtained
by the local diagrams in Set. We only need to prove that this map commutes with the intrinsic
maps f
L

S
C,D
, i.e. for any morphism k : C → D in ( the following equation should hold:
g(D)k
L

S
C,D
= k
M
C,D
g(C). (5.11)
Both the left-hand side and the right-hand side are maps from L

S
(C) to M(D). The left-hand side
is the unique map arising from
L

S
(C)
S(C)
i


i

>
L

S
(C)
k
L

S
C,D
>
L

S
(D)
g(D)
>
M(D)
............................................................... >
S(C)
α(C)
>
S(C)
1(C)
!(C)


>
M(D)
g(D)k
L

S
C,D
i

, (5.12)
and the right-hand side is the unique map arising from
L

S
(C)
A(C)
i


i

>
L

S
(C)
g(C)
>
M(C)
k
M
C,D
>
M(D)
............................................................. >
A(C)
α(C)
>
A(C)
1(C)
!(C)


>
M(D)
k
M
C,D
g(C)i

, (5.13)
For x ∈ S(C), one has
g(D)k
L

S
C,D
i

(C)(x) = g(D)i

(D)k
S
C,D
(x)
= f(D)k
S
C,D
(x)
= k
M
C,D
f(C)(x)
= k
M
C,D
g(C)i

(C)(x).
Hence the compositions of the horizontal arrows in the left diagrams of (5.12) and (5.13) are equal.
The uniqueness of the dotted arrow then results in (5.11).
80
5.2. FREE LATTICES CHAPTER 5. THE CONSTRUCTION OF THE SPECTRUM
We derive a short corollary from the theorem above.
Corollary 5.2.7. For any C*-algebra A, the free distributive lattice generated by
¯
A subject to relations
(5) - (9) exists. This lattice is denoted as L

¯
Asa
.
5.2.3 Intermezzo: representing the local lattice
The two theorems in this paragraph will be quite important for for what follows. We first introduce
a few lemmas that give us tools to compute the distributive lattice freely generated by a commutative
unital C*-algebra subject to (5) - (9) in Set. We will use the Gelfand theorem to infer that if A is a
commutative unital C*-algebra, then it is of the form C(X), where X is a compact Hausdorff space. In
this situation, A
sa
is the space of real valued continuous functions on X. The theorems below are based
on [2] and [3].
In this paragraph all statements are taken within Set. Hence, in this paragraph, we write D
a
for
i

a : A
sa
→ L

Asa
, the inclusion of the variable a of type A
sa
into the distributive lattice freely generated
by A
sa
, subject to relations (5) - (9).
Lemma 5.2.8. Let A = C(X) be a commutative unital C*-algebra. Suppose a, b ∈ A
sa
and a ≤ b (i.e.
∀x ∈ X a(x) ≤ b(x)). Then for D
a
, D
b
∈ L

Asa
, D
a
≤ D
b
.
Proof:
D
a
= D
(a−b)+b
≤ D
a−b
∨ D
b
(by relation (8))
= ⊥∨ D
b
(by relation (7))
= D
b
For a, b ∈ C(X)
sa
define (a ∨ b) ∈ C(X)
sa
and (a ∧ b) ∈ C(X)
sa
to be the point-wise maximum and
minimum of a and b, respectively. Let a
+
= a ∨ 0 and a

= a ∧ 0. Note that a = a
+
+a

.
Lemma 5.2.9. Let A = C(X) be a commutative unital C*-algebra. For a ∈ A
sa
, D
a
= D
a
+ and
D
−a
= D
−a
−.
Proof: Lemma 5.2.8 gives D
a
≤ D
a
+. Conversely:
D
a
+ = D
a∨0
≤ D
a
∨ D
0
(by relation (8))
= D
a
(by relation (7)).
We obtain D
a
= D
a
+, from which D
−a
− = D
(−a)
+ = D
−a
.
Lemma 5.2.10. Let A = C(X) be a commutative unital C*-algebra. For a ∈ A
sa
and n ∈ N

, one has
D
a
= D
na
.
Proof:
· = D
1
(by relation (5))
= D
n
1
n
≤ D1
n
(by applying relation (8) n times)
Now relation (9) gives D
na
≤ D
a
and D
a
= D1
n
na
≤ D
na
, from which the lemma follows.
81
5.2. FREE LATTICES CHAPTER 5. THE CONSTRUCTION OF THE SPECTRUM
Theorem 5.2.11. Let A = C(X) be a commutative unital C*-algebra. Then the map of sets i : A
sa

L

Asa
is surjective, i.e. every element of the lattice L

Asa
has a representative of the form D
a
for some
a ∈ A
sa
.
First notice that the proof of theorem 5.2.7 tells us L

Asa
is equal to the free lattice generated by A
sa
modulo some equivalence relation. We have proved the theorem if we can show that:
1. ⊥ is equal to D
a
for some a ∈ A
sa
. This is true, since ⊥ = D
−1
by relation (7).
2. · is equal to D
a
for some a ∈ A
sa
. This is true, since · = D
1
by relation (5).
3. D
a
∨ D
b
is equal to D
c
for some c ∈ A
sa
. We prove D
c
= D
a∨b
. The property D
a
∨ D
b
≤ D
a∨b
follows straight from lemma 5.2.8. For the other inequality, we note
D
a∨b
= D
(a−b)
+
+b
≤ D
(a−b)
+ ∨ D
b
(by relation (8))
≤ D
(a−b)
+ ∨ (D
b
∨ D
a
);
D
a∨b
≤ D
(b−a)
+ ∨ (D
b
∨ D
a
).
Now, the following inequality follows:
D
a∨b
≤ (D
(b−a)
+ ∨ (D
b
∨ D
a
)) ∧ (D
(a−b)
+ ∨ (D
b
∨ D
a
))
= (D
(b−a)
+ ∧ D
(a−b)
+) ∨ (D
b
∨ D
a
) ∨ ((D
b
∨ D
a
) ∧ D
(a−b)
+) ∨ ((D
b
∨ D
a
) ∧ D
(b−a)
+)
= (D
b−a
∧ D
a−b
) ∨ (D
b
∨ D
a
)
= D
b
∨ D
a
.
The first equality follows from distributivity, the second one follows from lemma 5.2.9 and the last
one follows from relation (6).
4. D
a
∧ D
b
is equivalent to D
c
for some c ∈ A
sa
. We prove that D
c
= D
a∧b
. D
a
∧ D
b
≥ D
a∧b
follows
straight from lemma 5.2.8. For the other inequality:
D
a
= D
(a∧b)+(a−b)
+
≤ D
a∧b
∨ D
(a−b)
+ , by relation (8)
D
b
≤ D
a∧b
∨ D
−(a−b)
−.
So:
D
a
∧ D
b
≤ (D
a∧b
∨ D
(a−b)
+) ∧ (D
a∧b
∨ D
−(a−b)
−)
= D
a∧b
∨ (D
(a−b)
+ ∧ D
−(a−b)
−) ∨ (D
a∧b
∧ D
(a−b)
+) ∨ (D
a∧b
∧ D
−(a−b)
−)
= D
a∧b
∨ (D
(a−b)
∧ D
−(a−b)
)
= D
a∧b
∨ ⊥
= D
a∧b
.
82
5.2. FREE LATTICES CHAPTER 5. THE CONSTRUCTION OF THE SPECTRUM
So all elements in L
Asa
that are obtained by the inductive steps in the proof of theorem 5.2.2 are equivalent
to D
a
for some a ∈ A
sa
. This proves the theorem.
The theorem above tells us that every element of L

Asa
is of the form D
a
for some a ∈ A
sa
. So if we
want to compute L

Asa
, we can start with the set A
sa
and divide out relations (5) - (9) in A
sa
instead
of dividing out these relations in L
Asa
. This makes the lattice far easier to compute. Futhermore, this
justifies to denote all the elements of the lattice L

Asa
as D
a
for a fixed element a ∈ A
sa
.
Notation: recapulating the discussion above, the notation D
a
can have three meanings:
1. We write D
a
for the term i

a if i

:
¯
A
sa
→ L

¯
Asa
is defined by the morphism of definition 5.2.6 in
the topos T (A). When this is meant, we will emphasize that this term is interpreted in T (A).
2. We write D
a
for the term i

a if i

: A
sa
→ L

Asa
is defined by the morphism of definition 5.2.6 in
the topos Set. We will emphasize that this term has to be interpreted in Set.
3. We write D
a
for the equivalence class of L

Asa
that contains a. In this case, we will always specify
the value of a and replace ‘a’ in the notation by its value. So, for example, we will always write
D
1
, D
−1
, et cetera.
This notation is not as ambiguous as it seems. The interpretations of the terms of the Mitchell-B´enabou
language in Set are exactly the conventional set-theoretic meanings of the expressions in terms of vari-
ables, logical connectives and quantifiers. Since in Set it is our practise to denote D
a
for both the
meanings as stated in 2 and 3, this notation will not lead to confusion. Also, note that meaning number
1 is a generalization of meaning number 2. Adopting the notation from meaning number 2 for meaning
number 1 will not lead to confusion as long as we specify in which topos we are working.
Returning to the computations in Set we prove one last theorem, which tells exactly which equivalence
relation we must divide out in order to compute the distributive lattice generated by A
sa
subject to (5)
- (9) in Set.
Theorem 5.2.12. Let A = C(X) be a commutative unital C*-algebra. Let L be the set A
sa
modulo the
following equivalence relation:
a ∼
+
b if and only if ∃n ∈ N

: a
+
≤ nb
+
and ∃m ∈ N

: b
+
≤ ma
+
.
Suppose L is equipped with the structure of the partial order _
+
defined by:
a _
+
b if and only if ∃n ∈ N

: a
+
≤ nb
+
.
Then L is a lattice isomorphic to L

Asa
.
Proof: Theorem 5.2.11 and the discussion after its proof show that the free lattice can be computed by
dividing out the equivalence relation generated by (5) - (9) in A
sa
with its partial order on functions.
Now clearly at least the equivalence relation above must at least be divided out, since if
∃n ∈ N

: a
+
≤ nb
+
and ∃m ∈ N

: b
+
≤ ma
+
then
D
a
= D
b
,
83
5.3. THE SPECTRUM CHAPTER 5. THE CONSTRUCTION OF THE SPECTRUM
by lemma 5.2.9 and lemma 5.2.10. The lattice L = A
sa
/ ∼
+
satisfies (5) - (9). Hence ∼
+
is the smallest
equivalence relation ∼ such that A
sa
/ ∼ is a lattice satisfying (5) - (9), and hence L must be the free
lattice. One easily verifies that _
+
is a well-defined partial order and if a ≤ b, then a _
+
b. Hence _
+
is the partial order on L that is inherited from the partial order on functions on A
sa
.
The theorem above gives a very nice representation of the free lattice generated by A
sa
. Now that we
now what this lattice looks like, we proceed to the next step, the construction of the spectrum.
5.3 The spectrum
Thusfar, we started with a commutative unital C*-algebra A in Set. We defined the topos T (A) and
the internal C*-algebra
¯
A. By means of section 5.2 we defined the distributive lattice freely generated by
¯
A
sa
subject to relations (5) - (9) in T (A). In this section we introduce a morphism / : Ω
L

¯
Asa
→ Ω
L

¯
Asa
of T (A) and define the spectrum Σ as a subobject of Ω
L

¯
Asa
by means of this morphism /.
In this section V , V
0
or any object that includes the character ‘V ’ in its notation is supposed to be a
set. We use the character U to indicate a subobject of L

¯
Asa
or more generally something that ‘looks
like’ or ‘corresponds to’ a subfunctor of L

¯
Asa
. The symbols C and D will always be objects in ((A), i.e.
commutative unital C*-subalgebras of A. Recall that L

¯
Asa
was an object of T (A). Hence L

¯
Asa
(C) for
an object C ∈ ((A) is a set.
5.3.1 The map /
0
In the following construction we define a map /
0
: Sub(L

¯
Asa
) → Sub(L

¯
Asa
) of subobjects of L

¯
Asa
in the
topos T (A). This is done by defining a map /
0
(C) : T(L

¯
Asa
(C)) → T(L

¯
Asa
(C)) of subsets of L

¯
Asa
(C)
for all objects C ∈ ((A) and combining them together into a global map (here T denotes the power set).
The maps /
0
(C) are called the (local) completions.
To define the completions we first need to know something about joins of finite subsets of a lattice.
Suppose L is an internal lattice in Set and suppose V
0
is a finite subset of L. We define
_
V
0
to be the
supremum of the set V
0
. Thus
_
V
0
is the element z ∈ L such that
[= (∀x x ∈ V
0
⇒ x ≤ z) ∧ (∀x(∀y y ∈ V
0
⇒ y ≤ x) ⇒ z ≤ x).
Here the variables x and y are of type L. The expression above has an interpretation in the Mitchell-
B´enabou language of the topos Set, which coincides with its familiar meaning in set theory. Note that
_
V
0
always exists: it is the join of all the elements of V
0
, which is finite by definition. Note that by this
definition
_
∅ is defined to be the bottom element ⊥ of L.
We start with a short lemma in Set.
Lemma 5.3.1. Suppose C is a unital commutative C*-algebra in Set. Let D
a
, D
b
∈ L

Csa
such that
D
a
= D
b
. Then for every q ∈ Q there exists an r ∈ Q such that
D
a−q
≤ D
b−r
.
84
5.3. THE SPECTRUM CHAPTER 5. THE CONSTRUCTION OF THE SPECTRUM
Proof: Choose an n ∈ N

such that a ≤ nb, which can be done by theorem 5.2.12. Then
D
a−q
≤ D
nb−q
= D
n(b−
q
n
)
= D
b−
q
n
.
Hence r =
q
n
works.
Now if C is an object of ((A) and U is a subfunctor of L

¯
Asa
, we define locally what is called an ideal
generated by U(C) with respect to the covering by:
/
0
(C)U(C) =
_
D
a
∈ L

¯
Asa
(C) [ D
a
U(C)
_
,
where D
a
U(C) is defined as:
For every q ∈ Q
+
there exists a finite subset V
0
of U(C) such that D
a−q

V
0
. (5.14)
We say U(C) covers D
a
. It follows from lemma 5.3.1 that if D
a
= D
b
and D
a
U(C), then D
b
U(C).
Hence /
0
(C) is well-defined. Furthermore, the completions /
0
(C)U(C) glue together to a functor called
/
0
U, which is an object of T (A). To see this, we must check if the intrinsic maps ⊆
A
0
U
C,D
are well-
defined. Since /
0
U is a subfunctor of L

¯
Asa
, we have ⊆
A
0
U
C,D
=⊆
L

¯
Asa
C,D
[
A
0
U(C)
and we must check whether

L

¯
Asa
C,D
[
A
0
U(C)
(D
a
) is in (/
0
U)(D) = /
0
(D)U(D). This amounts to checking whether ⊆
L

¯
Asa
C,D
[
A
0
U(C)
(D
a
)
satisfies property 5.14. Now since ⊆
L

¯
Asa
C,D
is a lattice homomorphism (cf. proof of theorem 5.2.3), and since
applying
_
to a finite set gives the join of all the elements of that set, we find that ⊆
L

¯
Asa
C,D
[
A
0
U(C)
(D
a
) is
covered by ⊆
L

¯
Asa
C,D
[
A
0
U(C)
(V
0
), where V
0
is a finite subset of U(C) covering D
a
. So we have successfully
defined our map /
0
: Sub(L

¯
Asa
) → Sub(L

¯
Asa
), by
(/
0
U)(C) := /
0
(C)U(C).
The following lemma will be useful for later computations.
Lemma 5.3.2. /
0
is idempotent.
Proof: /
0
is idempotent if and only if /
0
(C) is idempotent for every C ∈ ((A). So we must show that
if U is a subfunctor of L

¯
Asa
, then
/
0
(C)/
0
(C)U(C) = /
0
(C)U(C).
The inclusion ⊇ is obvious, since every element of /
0
(C)U(C) is covered by itself. Conversely, suppose
D
a
∈ /
0
(C)/
0
(C)U(C). Then we must show that for every q ∈ Q
+
there exists a finite subset V
0
of
U(C) such that D
a−q

_
V
0
. By the assumption D
a
∈ /
0
(C)/
0
(C)U(C) there exists a finite subset V

0
of /
0
(C)U(C) such that D
a−
q
2

_
V

0
. Now if D
b
∈ V

0
then there exists a finite subset V
0,b
of U(C)
such that D
b−
q
2

_
V
0,b
. Now we have
D
a−q

_
D
b−
q
2
[ D
b
∈ V

0
_

b

V
0,b
,
where the double supremum on the right hand side of the equality is the supremum of a finite set.
85
5.3. THE SPECTRUM CHAPTER 5. THE CONSTRUCTION OF THE SPECTRUM
5.3.2 The morphism /
Now we will define a morphism / : Ω
L

¯
Asa
→ Ω
L

¯
Asa
, which is the internal version of /
0
. Before we
proceed we introduce an important lemma.
Lemma 5.3.3. Let T be a categorical preorder and let A be an object of Set
P
. Then all intrinsic
morphisms of Ω
A
are surjective maps of sets.
Proof: In this proof, let C, D, E and F be objects of the preorder T. Recall that Ω
A
was defined as

A
(C) = Hom
Set
P(y(C) A, Ω),
where we have written y(C) for the covariant Yoneda-functor Hom
P
(C, −). Notice that
y(C)(E) =
_
¦1¦ if C ⊆ E;
∅ if C ,⊆ E;
and remark that
y(C)(E) A(E) =
_
A(E) if C ⊆ E;
∅ if C ,⊆ E.
Now suppose C ⊆ D (denote this as a morphism by ⊆
C,D
) and τ ∈ Hom
Set
P(y(D) A, Ω). We define
a natural transformation σ ∈ Hom
Set
P(y(C) A, Ω) by
σ(E) = τ(E) if D ⊆ E or C ,⊆ E
σ(E) : A(E) → Ω(E)
a →
_
F[D ⊆ F, (τ(F))(¦1¦ (⊆
A
E,F
)(a)) = m.c. on F
_
if C ⊆ E and D ,⊆ E,
where m.c. is short for ‘maximal cosieve’. It is not too hard to check that σ : y(C) A → Ω is indeed a
natural transformation. We prove that (⊆

C,D
Id
A
)

σ = τ. We separate three cases:
1. First, suppose D ⊆ E. Then
((⊆

C,D
Id
A
)

σ)(E) = (σ ◦ (⊆

C,D
Id
A
))(E)
= σ(E) ◦ (⊆

C,D
Id
A
)(E)
= τ(E).
The last equation follows from the fact that ⊆

C,D
(E) : y(D)(E) → y(C)(E) is the identity Id
{1}
.
2. Now, suppose C ,⊆ E. Then again
((⊆

C,D
Id
A
)

σ)(E) = (σ ◦ (⊆

C,D
Id
A
))(E)
= σ(E) ◦ (⊆

C,D
Id
A
)(E)
= τ(E).
The last equation follows from the fact that ⊆

C,D
(E) : y(D)(E) → y(C)(E) is the identity Id

.
86
5.3. THE SPECTRUM CHAPTER 5. THE CONSTRUCTION OF THE SPECTRUM
3. Suppose C ⊆ E and D ,⊆ E. Note that ((⊆

C,D
Id
A
)

σ)(E) : ∅ = y(D)(E) A(E) → Ω(E) is the
unique morphism from the initial object to Ω(E). Hence it must equal τ(E).
We have shown that (⊆

C,D
Id
A
)

σ = τ, hence ⊆

A
C,D
= (⊆

C,D
Id
A
)

is surjective.
Corollary 5.3.4. Suppose in lemma 5.3.3 that T has a bottom element ⊥. Then, for any object B of
Set
P
, any morphism / : Ω
A
→ B is completely determined by the morphism /(⊥) : Ω
A
(⊥) → B(⊥).
Proof: Suppose C ∈ T; we must define /(C) : Ω
A
(C) → B(C). If x ∈ Ω
A
(C), then there exists a
y ∈ Ω
A
(⊥) such that ⊆

A
⊥,C
(y) = x. Hence
(/(C))(x) = (/(C))(⊆

A
⊥,C
(y)) = (⊆
B
⊥,C
)(/(⊥))(y). (5.15)
Now observe that a subfunctor U of L

¯
Asa
corresponds to a morphism χ
U
: L

¯
Asa
→ Ω, which in turn
corresponds to a morphism ˆ χ
U
: 1 → Ω
L

¯
Asa
by cartesian closedness. So the completion /
0
corresponds
to a morphism
( ˆ χ
U
: 1 → Ω
L

¯
Asa
) → ( ˆ χ
A
0
U
: 1 → Ω
L

¯
Asa
).
Since each morphism 1 → Ω
L

¯
Asa
is determined by 1(⊥) → Ω
L

¯
Asa
(⊥) (noting that the intrinsic maps of
1 are surjective), this corresponds to a morphism

L

¯
Asa
(⊥) → Ω
L

¯
Asa
(⊥),
which in turn induces a unique morphism / by equation (5.15) in corollary 5.3.4:
/ : Ω
L

¯
Asa
→ Ω
L

¯
Asa
.
Finally, we must check if / commutes with the intrinsic maps of Ω
L

¯
Asa
. Suppose x =⊆

L

¯
Asa
⊥,C
y, then:
(⊆

L

¯
Asa
C,D
)(/(C))(x) = (⊆

L

¯
Asa
C,D
)(⊆

L

¯
Asa
⊥,C
)(/(⊥))(y)
= (⊆

L

¯
Asa
⊥,D
)(/(⊥))(y)
= (/(D))(⊆

L

¯
Asa
⊥,D
)(y)
= (/(D))(⊆

L

¯
Asa
C,D
)(⊆

L

¯
Asa
⊥,C
)(y)
= (/(D))(⊆

L

¯
Asa
C,D
)(x).
Hence / is well-defined. We next prove the counterpart of lemma 5.3.2 for /.
Lemma 5.3.5. / is idempotent.
Proof: / is idempotent if and only if /(C) is idempotent for all C ∈ ((A). The lemma follows from the
following computation. Suppose x =⊆

L

¯
Asa
⊥,C
y, then
(/(C))(x) = (⊆

L

¯
Asa
⊥,C
)(/(⊥))y
= (⊆

L

¯
Asa
⊥,C
)(/(⊥))(/(⊥))y
= (/(C))(/(C))(x).
87
5.3. THE SPECTRUM CHAPTER 5. THE CONSTRUCTION OF THE SPECTRUM
The first and last equations hold by definition. The second equation follows from lemma 5.3.2 and the
correspondence between /
0
and /(⊥).
We conclude this section by giving the definition of the spectrum as well as a proposition which is a
consequence of lemma 5.3.2 and lemma 5.3.5.
Definition 5.3.6. The spectrum Σ of an internal unital commutative C*-algebra
¯
A in the topos T (A)
is defined as
Σ = ¦U [/U = U¦ ⊆ Ω
L

¯
Asa
.
As we said in the beginning of this chapter Mulvey and Banaschewski proved the duality
cCStar ↔ KRegLoc,
in any topos. The construction above starting from the internal C*-algebra
¯
A and ending with the
spectrum Σ is precisely the functor cCStar → KRegLoc in the topos T (A), as proved in [7]. The
morphism cCStar ← KRegLoc then is Σ → Loc(Σ, C).
The following theorem allows us to compute the spectrum once we have computed Σ(⊥).
Theorem 5.3.7. The maps ⊆
Σ
⊥,C
: Σ(⊥) → Σ(C) are surjective. Hence Σ(C) is the image set of the the
intrinsic map ⊆
Σ
⊥,C
.
Proof: Suppose U
C
∈ Σ(C) and U
C
= (⊆

L

¯
Asa
⊥,C
)(U

). Then
U
C
= (/(C))U
C
= (⊆

L

¯
Asa
⊥,C
)(/(⊥))U

= (⊆
Σ
⊥,C
)(/(⊥))U

.
For the last equality, notice that lemma 5.3.5 gives (/(⊥))(/(⊥))U

= (/(⊥))U

, hence (/(⊥))U


Σ(⊥).
The correspondences between objects in this section may seem very abstract. However, the correspon-
dences are actually quite straight forward. Suppose that we have computed the lattice L

¯
Asa
and suppose
we know which subfunctors U of L

¯
Asa
have the property that /
0
U = U. Then in order to compute the
spectrum Σ, we have the following correspondences:
1. Any subfunctor U of L

¯
Asa
corresponds to a characteristic morphism χ
U
: L

¯
Asa
→ Ω by theorem
3.1.4.
2. Any characteristic morphism χ
U
in turn corresponds to a morphism ˆ χ
U
: 1 → Ω
L

¯
Asa
by cartesian
closedness of the topos T (A).
3. Since ((A) has a bottom element ⊥ the morphism ˆ χ
U
: 1 → Ω
L

¯
Asa
is completely determined by
the element ( ˆ χ
U
(⊥)(∗

)) ∈ Σ(⊥) and the computation:
ˆ χ
U
(C)(∗) = ˆ χ
U
(C)(⊆
1
⊥,C
)(∗

) =⊆

L

¯
Asa
⊥,C
( ˆ χ
U
(⊥)(∗

)),
where ∗ and ∗

denote the unique elements of 1(C) and 1(⊥) respectively. Hence morphisms ˆ χ
U
:
1 → Ω
L

¯
Asa
correspond one-on-one to elements of Ω
L

¯
Asa
(⊥). By the definition of the correspondence
88
5.4. FINAL REMARKS CHAPTER 5. THE CONSTRUCTION OF THE SPECTRUM
between / and /
0
, the subobjects U of L

¯
Asa
such that /
0
U = U correspond exactly to the elements
x of Ω
L

¯
Asa
(⊥) such that /(⊥)x = x.
4. By the above correspondences we can compute Σ(⊥) from the set of subfunctors U of L

¯
Asa
that
satisfy /
0
U = U. Then theorem 5.3.7 shows that we can compute each Σ(C) and hence the functor
Σ from Σ(⊥), resulting in the spectrum of our initial C*-algebra A.
5.4 Final remarks
We conclude this chapter by summarizing the construction of the spectrum of a C*-algebra A and the
most important theorems for computations.
1. First, one tries to find all unital C*-subalgebras of A, resulting in the preorder ((A), the topos
T (A) and the tautological functor
¯
A. In the next chapter we will prove some theorems to compute
((A).
2. Next, one computes the distributive lattice freely generated by
¯
A
sa
subject to relations (5) - (9).
Theorem 5.2.6 shows that this can be done locally, i.e. one computes the distributive lattices freely
generated by B
sa
subject to relations (5) - (9) in Set, where B is a subalgebra of A. These local
freely generated distributive lattices can be combined into the functor L

¯
Asa
.
3. Finally, one computes the subfunctors U of L

¯
Asa
such that /
0
U = U. In the next chapter we will
see that if A is finite, this can be done by finding the subfunctors U such that U(C) is a down-set for
every object C ∈ ((A). By the correspondences stated at the end of paragraph 5.3 one then directly
finds Σ(⊥). Finally, theorem 5.3.7 can be applied to find the spectrum Σ. In the next chapter we
will introduce a theorem (see 6.1.3) that allows us to compute Σ easily from the subfunctors U of
L

¯
Asa
such that /
0
U = U.
The duality by Mulvey and Banaschewski
cCStar ↔ KRegLoc,
is proved by a general construction for the morphism cCStar → KRegLoc for any topos T . This
general construction introduced by Mulvey and Banaschewski (cf. [12], [13] and [14]) and reformulated
by Coquand and Spitters [4] is quite similar. Again, one first computes the distributive lattice freely
generated by the self-adjoint part of an (arbitrary) internal C*-algebra
¯
A subject to (5) - (9). Then, one
turns this into a frame that satisfies the relation
D
a

q∈Q
+
D
a−q
.
The construction of this frame can be done by introducing a general morphism/, which involves technical
definitions of finite objects of a topos and infinite joins. For brevity, we have avoided the direct definition
of / in T (A) and chosen to introduce the (less complicated) map /
0
first. In the next chapter we will
do some computations and fortunately, with a good understanding of the definition of /
0
the spectrum
can be computed.
89
Chapter 6
Computations of spectra
“Can you do addition?” the White Queen asked. “What’s one and one and one and one and one and
one and one and one and one and one?” “I don’t know,” said Alice. “I lost count.” - Lewis Carroll
In this chapter we compute some examples of Gelfand spectra. Our main examples will be C
2
and
M(2, C). We have chosen these examples since C
2
is the first C*-algebra whose spectrum is nontrivial,
whereas M(2, C) is the simplest noncommutative algebra. Furthermore, it turns out that C has a trivial
spectrum, and that C
n
is computable, but quite complicated for n ≥ 3. Further examples that would
be interesting are C(X), with X a compact Hausdorff space, and B(H), the bounded operators on a
Hilbert-space. However, whenever a C*-algebra has a commutative C*-subalgebra of dimension 3 or
higher, the spectrum turns out to be hard to compute by hand. Nevertheless we will spend a few words
on C(X).
In the following examples A will be the C*-algebra that we study. We recapitulate that the spectrum
can be computed in three steps:
1. The computation of ((A) and hence the corresponding topos T (A).
2. The computation of the free distributive lattice generated by
¯
A
sa
subject to the relations:
[= D
1
= ·; (5)
[= ∀a D
a
∧ D
−a
= ⊥; (6)
[= ∀a D
−a
2 = ⊥; (7)
[= ∀a∀b D
a+b
≤ D
a
∨ D
b
; (8)
[= ∀a∀b D
ab
≤ (D
a
∧ D
b
) ∨ (D
−a
∧ D
−b
). (9)
3. The computation of the spectrum Σ, via Σ(⊥) as described at the end of section 5.3.
6.1 The finite, commutative case: C
n
First we study the commutative C*-algebra A = C
n
. It turns out that we can compute the spectrum Σ
of C
n
for every n. However, for n = 3 things become quite complicated and for n ≥ 4 one would rather
use a computer. Here we will make the structure precise for the special cases n = 1 and n = 2, avoiding
the extensive amount of computations for n ≥ 3.
90
6.1. THE FINITE, COMMUTATIVE CASE: C
N
CHAPTER 6. COMPUTATIONS OF SPECTRA
C*-subalgebra and topos
The first question we should answer is what the unital C*-subalgebras of C
n
look like. Recall that we
can regard C
n
as the C*-algebra of functions from an n-point space with discrete topology to C. Suppose
Π is a partition of these points. We prove that the commutative unital C*-subalgebras are the ones that
consist of functions that are constant on the equivalence classes of some partition using elementary linear
algebra. In section 6.3 we will give another proof using the Gelfand isomorphism, which shortens the
case of C
n
, but requires more technical tools.
Theorem 6.1.1. Let A be the C*-algebra C
n
and regard it as the space of functions on the topological
space X with n points and discrete topology. Let Π be a partition of the points of X and let A
Π
be the
set of functions that are constant on every equivalence class of the partition. Then:
1. A
Π
is a C*-subalgebra.
2. Every C*-subalgebra of A is equal to A
Π
for a unique partition of n points called Π.
Proof:
1. Trivial.
2. Given any C*-subalgebra B of A, we must find the pertinent partition Π. Let e
p
i
: B → C be the
evaluation map at the i
th
point of X. Now define a relation on the points of X by p
i

X
p
j
if and
only if Ker(e
p
i
−e
p
j
) = B. It is straightforward that this relation is an equivalence relation on X.
Let Π be the corresponding partition.
If f ∈ B is a function on X, then it is constant on equivalence classes of Π by definition of ∼
X
.
Hence f ∈ A
Π
, so that B ⊆ A
Π
.
Conversely, let f ∈ A
Π
. Let R be a set of representatives of Π. If we can find a g ∈ B that equals f
on R we may conclude A
Π
⊆ B. By induction (to i) we show that for any set of i different points of
R and for any f ∈ A
Π
we can construct a function g ∈ B that equals f at these i different points.
• For i = 1 this is trivial, since B contains the identity. If the partition has only one equivalence
class, we are finished.
• For i = 2 note that B contains at least one function h that has different values at two
given representatives r
k
1
, r
k
2
∈ R. Since the vectors (h(r
k
1
), h(r
k
2
)) and (Id(r
k
1
), Id(r
k
2
)) are
independent vectors, there is a linear combination of h and the identity whose values at r
k
1
and r
k
2
equal f.
• For i + 1, take elements r
1
, r
2
, . . . , r
i+1
∈ R. Let f : R → C be a function and suppose we
chose a point r
k
∈ ¦r
1
, r
2
, . . . , r
i+1
¦. Then there is, by induction, a function h ∈ B such that
h equals f at all points r
1
, r
2
, . . . , r
i+1
except for maybe r
k
.
– Suppose i + 1 is odd. Then construct functions h
1
, . . . , h
i+1
that satisfy
_
_
_
_
_
_
_
h
1
(r
1
) h
1
(r
2
) . . . h
1
(r
i
) h
1
(r
i+1
)
h
2
(r
1
) h
2
(r
2
) . . . h
2
(r
i
) h
2
(r
i+1
)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
h
i
(r
1
) h
i
(r
2
) . . . h
i
(r
i
) h
i
(r
i+1
)
h
i+1
(r
1
) h
i+1
(r
2
) . . . h
i+1
(r
i
) h
i+1
(r
i+1
)
_
_
_
_
_
_
_
=
_
_
_
_
_
_
_
_
_
1 a
1
0 . . . 0 0
0 1 a
2
. . . 0 0
0 0 1 . . . 0 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 0 0 . . . 1 a
i
a
i+1
0 0 . . . 0 1
_
_
_
_
_
_
_
_
_
,
91
6.1. THE FINITE, COMMUTATIVE CASE: C
N
CHAPTER 6. COMPUTATIONS OF SPECTRA
so every indexed function h is specified at precisely i points and its value is unknown at
only one point. These unknown values are denoted by a
1
, . . . , a
i+1
in the matrix above.
Since i +1 is odd, the determinant of this matrix is 1+a
1
. . . a
i+1
. If this is non-zero, the
functions are independent. Hence for any g ∈ A
Π
there is a linear combination of functions
in B such that this linear combination equals g at the i + 1 points. If the determinant
is zero, note that the determinant of the matrix obtained by evaluating the squares of
h
1
, . . . , h
i+1
at r
1
, . . . , r
i+1
is 1 + (a
1
. . . a
i+1
)
2
= 2 which leads to the same conclusion.
– Suppose i + 1 is even. Then consider
_
_
_
_
_
_
_
h
1
(r
1
) h
1
(r
2
) . . . h
1
(r
i
) h
1
(r
i+1
)
h
2
(r
1
) h
2
(r
2
) . . . h
2
(r
i
) h
2
(r
i+1
)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
h
i
(r
1
) h
i
(r
2
) . . . h
i
(r
i
) h
i
(r
i+1
)
h
i+1
(r
1
) h
i+1
(r
2
) . . . h
i+1
(r
i
) h
i+1
(r
i+1
)
_
_
_
_
_
_
_
=
_
_
_
_
_
_
_
_
_
1 a
1
0 . . . 0 0
0 1 a
2
. . . 0 0
0 0 1 . . . 0 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
a
i
0 0 . . . 1 0
a
i+1
0 0 . . . 0 1
_
_
_
_
_
_
_
_
_
.
A simple calculation gives that the determinant of this matrix is 1 + a
1
. . . a
i
and the
induction hypothesis is proved by the same reasoning as in the case where i + 1 is odd.
The theorem above tells us that ((C
n
) is the partial order of partitions of n points, where the order is
given by refinements of partitions. As special cases we have the Hasse diagrams:
((C
1
) = C; ((C
2
) =
_
_
_
_
_
C
2
C

_
_
_
_
_
; ((C
3
) =
_
_
_
_
_
_
_
_
_
_
_
C
3
A
{{1},{2,3}}
>
A
{{2},{1,3}}

A
{{3},{1,2}}
<
C

>
<
_
_
_
_
_
_
_
_
_
_
_
.
These diagrams also represent the tautological functor
¯
A as an object of T (C
n
) =Set
C(C
n
)
. Note that if
n = 1, T (C) is the category Set. If n = 2 we have T (C
2
) =Set
C
2
.
The lattice
In theorem 5.2.6 we showed that we may compute the lattice generated by
¯
A
sa
subject to relations (5)
- (9) locally. The local lattices are generated by R
m
, m ≤ n subject to the given relations. Theorem
5.2.12 provides the means to compute the lattice generated by R
m
subject to the relations. We restate
the theorem for the special case C
m
.
Theorem 6.1.2. Let B = C
m
. The lattice generated by B
sa
subject to relations (5) - (9) consists of the
elements
_
D
(±1,±1,...,±1)
_
, where D
a
≤ D
b
if and only if a ≤ b, i.e. a
i
≤ b
i
for all coordinates i.
Proof: Define a function sign:
92
6.1. THE FINITE, COMMUTATIVE CASE: C
N
CHAPTER 6. COMPUTATIONS OF SPECTRA
sign : R → R ;
x → 1 if x > 0;
x → −1 if x ≤ 0.
For a ∈ B
sa
, theorem 5.2.12 gives D
a
= D
a
+ = D
(sign(a
1
),...,sign(am))
. Hence every element of the lattice
generated by B
sa
subject to (5) - (9) is of the form D
(±1,...,±1)
. By definition of ∼
+
, none of the elements
of
_
D
(±1,±1,...,±1)
_
is equivalent to any other. The statement about the partial order follows from theorem
5.2.12.
We could express theorem 6.1.2 geometrically by saying that the lattice C
m
produces is the m-dimensional
hypercube. We will try to visualize L

¯
Asa
for the cases n = 1, n = 2 and n = 3. The case n = 1 looks like:
_
_
_
_
D
1
D
−1
_
_
_
_
.
The case n = 2 has the form:
_
_
_
_
_
_
_
_
_
_
_
D
(1,1)
D
(−1,1)
D
(1,−1)
D
(−1,−1)
_
_
_
_
_
_
_
_
_
_
_

_
_
_
_
_
D
(1,1)
D
(−1,−1)
_
_
_
_
_
.
The case n = 3 takes the form:
93
6.1. THE FINITE, COMMUTATIVE CASE: C
N
CHAPTER 6. COMPUTATIONS OF SPECTRA
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
D
(1,1,−1)
D
(1,1,1)
D
(1,−1,−1)
D
(1,−1,1)
D
(−1,1,−1)
D
(−1,1,1)
D
(−1,−1,−1)
D
(−1,−1,1)
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
D
(1,−1,−1)
D
(1,1,1)
D
(−1,−1,−1)
D
(−1,1,1)
_
_
_
_
_
>
_
_
_
_
_
D
(−1,1,−1)
D
(1,1,1)
D
(−1,−1,−1)
D
(1,−1,1)
_
_
_
_
_

_
_
_
_
_
D
(1,1,−1)
D
(1,1,1)
D
(−1,−1,−1)
D
(−1,−1,1)
_
_
_
_
_
<
_
_
_
_
_
D
(1,1,1)
D
(−1,−1,−1)
_
_
_
_
_
.

>
<
The spectrum
Now we compute Σ. We will do this for n = 1 and n = 2. For n = 3 the situation starts to become too
complicated to calculate by hand, as we will see.
First we recall the definition of a down-set. If L is a lattice and x ∈ L, then the down-set x ↓ of x is
defined by
x↓= ¦y [ y ≤ x¦ .
Now, let us recall the definition of /
0
. Let U be a subfunctor of L

¯
Asa
and C ∈ ((A). Then
/
0
(C)U(C) =
_
D
a
∈ L

¯
Asa
(C) [ D
a
U(C)
_
=
_
D
a
∈ L

¯
Asa
(C) [ For every q ∈ Q
+
: D
a−q

_
U(C)
_
=
_
D
a
∈ L

¯
Asa
(C) [ D
a

_
U(C)
_
The first equation is the definition. The second follows from the fact that U(C) is already finite in the
case of C
n
. The last equation follows from the fact that D
a−q
= D
a
for small q, which follows from
theorem 6.1.2. The last equation says that /
0
(C)U(C) is the down-set of the meet of U(C). Conversely,
every down-set is the completion under /
0
of itself. Recall from section 5.3 that Σ(⊥) as a lattice in Set
may be represented by the subfunctors U of L

¯
Asa
such that /
0
U = U. We find that these subfunctors
are the ones that are locally down-sets.
94
6.1. THE FINITE, COMMUTATIVE CASE: C
N
CHAPTER 6. COMPUTATIONS OF SPECTRA
n = 1: We find that Σ(⊥) is equal to the following lattice:
D
1

D
−1
↓ .
Since ((A) consists only of one element, this determines Σ globally.
n = 2: The following subfunctors of L

¯
Asa
are the ones that are locally down-sets and hence these subfunc-
tors will form the lattice Σ(⊥):
A
1,1
=
D
(1,1)
D
(−1,−1)
>
D
(1,1)
D
(−1,1)
D
(1,−1)
D
(−1,−1)
A
2,1
=
0
D
(−1,−1)
>
D
(1,1)
D
(−1,1)
D
(1,−1)
D
(−1,−1)
A
2,2
=
0
D
(−1,−1)
>
0
0 D
(1,−1)
D
(−1,−1)
A
2,3
=
0
D
(−1,−1)
>
0
D
(−1,1)
0
D
(−1,−1)
95
6.1. THE FINITE, COMMUTATIVE CASE: C
N
CHAPTER 6. COMPUTATIONS OF SPECTRA
A
2,4
=
0
D
(−1,−1)
>
0
0 0
D
(−1,−1)
Here the 0 indicates that the element is not included in the subfunctor. The lattice structure of
Σ(⊥) is then given by
A
1,1
A
2,1
A
2,2
A
2,3
A
2,4
.
We now compute Σ(C
2
) = Σ(·), which by theorem 5.3.7 and the subsequent discussion equals
Im(⊆
Σ
⊥,
). Recall that images can be taken point-wise. We compute:
(⊆

⊥,
Id
L

¯
Asa
)(·) = Id
{1×L

¯
Asa
}
;
(⊆

⊥,
Id
L

¯
Asa
)(⊥) = !,
where ! : 0 → L

¯
Asa
is the unique arrow from the initial object of Set. It follows from these equations
that ⊆
Σ
⊥,
χ
U
= χ
U
(⊆

⊥,
Id
L

¯
Asa
) = χ
V
(⊆

⊥,
Id
L

¯
Asa
) =⊆
Σ
⊥,
χ
V
if and only if U(·) = V (·).
Hence Σ(·) has the form
A
2,1
· A
1,1
A
2,2
A
2,3
A
2,4
.
We will restate the argument above for the computation of Σ(·) in the more general theorem 6.1.3
below.
96
6.1. THE FINITE, COMMUTATIVE CASE: C
N
CHAPTER 6. COMPUTATIONS OF SPECTRA
n ≥ 3: For this case one proceeds in exactly the same way. One has to compute all subfunctors of L

¯
Asa
such that locally the subfunctor is a down-set. This can easily be done by a computer. In the case
n = 3 one finds 96 such functors, which form Σ(⊥). The following theorem shows what the images
of the intrinsic maps ⊆
Σ
⊥,C
look like.
Theorem 6.1.3. Let U and V be subfunctors of L

¯
Asa
. Then:
(a) The corresponding elements of U in V in Ω
L

¯
Asa
(⊥) = Hom
T (A)
(y(⊥)A, Ω) = Hom
T (A)
(A, Ω)
are the natural transformations χ
U
and χ
V
.
(b) For C ∈ ((A), one has ⊆
Σ
⊥,C
χ
U
=⊆
Σ
⊥,C
χ
V
if and only if U(D) = V (D) for all objects C ⊆ D.
Proof:
(a) This is obvious, being exactly the definition of exponentiation (see theorem 1.2.10).
(b) Recall that ⊆
Σ
⊥,C
= (⊆

⊥,C
Id
L

¯
Asa
)

. The second assertion is a consequence of the following
computation of the map (⊆

⊥,C
Id
L

¯
Asa
) : y(C) L

¯
Asa
→ y(⊥) L

¯
Asa
, which is given by
(⊆

⊥,C
Id
L

¯
Asa
)(D) = Id
{1}×L

¯
Asa
(D)
if C ⊆ D;
(⊆

⊥,C
Id
L

¯
Asa
)(D) = ! if C ,⊆ D,
where ! : 0 = (y(C))(D) L

¯
Asa
(D) → (y(⊥))(D) L

¯
Asa
(D) = L

¯
Asa
(D) is the unique arrow
from the initial object, i.e. the empty set. Now we have the following situation:
i. If D ⊇ C, then
((⊆

⊥,C
Id
L

¯
Asa
)

χ
U
)(D) = χ
U
(D)(⊆

⊥,C
Id
L

¯
Asa
)(D);
= χ
U
(D).
ii. If D ,⊇ C, then
((⊆

⊥,C
Id
L

¯
Asa
)

χ
U
)(D) = χ
U
(D)(⊆

⊥,C
Id
L

¯
Asa
)(D);
= ! : 0 → Ω(D).
Now U(D) = V (D) for all D ⊇ C if and only if χ
U
(D) = χ
V
(D) for all D ⊇ C if and only if
(⊆
Σ
⊥,C
χ
U
)(D) = (⊆
Σ
⊥,C
χ
V
)(D) for all D ⊇ C.
From this theorem we derive a corollary that gives the localic spectrum of C
n
by the original Gelfand
theorem in Set.
Corollary 6.1.4. In the case of A = C
n
, ((A) has a top element and the top level of the spectrum
Σ(·) is given by the lattice T(¦1, . . . , n¦).
Proof: In this case A itself is the top element of ((A). By theorem 6.1.2 L

¯
Asa
(·) is of the form
T(¦1, . . . , n¦). Then:
(a) Clearly, if V ⊆ L

¯
Asa
(·) then there is a subfunctor U
V
⊆ L

¯
Asa
such that U
V
(·) = V . It follows
from theorem 6.1.3 that if ⊆
Σ
⊥,
χ
U
V
=⊆
Σ
⊥,
χ
U
V

, then V = V

. Hence the images ⊆
Σ
⊥,
χ
U
V
represent different elements of Σ(·) for every V .
(b) From theorem 6.1.3 it immediately follows that for every subfunctor U ⊆ L

¯
Asa
one has ⊆
Σ
⊥,
χ
U
=⊆
Σ
⊥,
χ
U
U()
. Hence every element of Σ(·) is of the form ⊆
Σ
⊥,
χ
U
V
.
From this we conclude that Σ(·) = T(¦1, . . . , n¦).
97
6.2. THE FINITE CASE IN GENERAL CHAPTER 6. COMPUTATIONS OF SPECTRA
6.2 The finite case in general
Now that we know what the spectrum of C
n
looks like, computing the spectrum of any finite-dimensional
(non-commutative) C*-algebra becomes much easier! Once we know the structure of ((A), the lattice
L

¯
Asa
can be computed locally at C and every local lattice has the form of an n-dimensional hypercube.
Then Σ(⊥) has the form of all subfunctors such that locally the subfunctor is a down-set. Theorems
5.3.7 and 6.1.3 then give a complete description of Σ globally.
The spectrum of M(2, C)
As an example, we study M(2, C). We first prove a proposition that gives all commutative unital C*-
subalgebras of A = M(2, C).
Proposition 6.2.1. All nontrivial unital C*-subalgebra’s of M(2, C) are of the form
B
u
=
_
u
_
z
1
0
0 z
2
_
u
−1
[ z
1
, z
2
∈ C
_
,
where u is a unitary matrix.
Proof: We denote matrices by small characters. Let B be a unital commutative C*-subalgebra of
M(2, C). Let b ∈ B, then b = udu
−1
. Where d is either diagonal or of the form:
_
d
1,1
1
0 d
2,2
_
.
If d has a 1 in the upper right entry one can show that B = M(2, C). So suppose b = udu
−1
, where d
is diagonal. Then b

= (u

)
−1
d

u

commutes with b, which means that b and b

have a common set of
eigenvalues. Hence u

= u
−1
, so u is unitary. Now suppose b

∈ B, then b

= u

d

u
−1
. Since b

commutes
with b, b and b

have a common set of eigenvalues, hence u

= u. This proves the theorem.
We thus find:
((M(2, C)) =
_
_
_
_
_
_
_
_
_
_
. . . B
u
. . .
C
˙
1 =
__
z 0
0 z
_
[ z ∈ C
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
,
where the dots stand for uncountably many C*-subalgebras running through all unitary u ∈ M(2, C),
between which there is no (nontrivial) inclusion relation. Then L

¯
Asa
is given by the following functor, as
follows from the example of C
n
and the fact that we may compute this lattice point-wise.
98
6.3. SOME REMARKS ON C(X) CHAPTER 6. COMPUTATIONS OF SPECTRA
. . .
_
_
_
_
_
_
_
_
_
_
_
D
Id
2
D
ua
−+
u
−1 D
ua
+−
u
−1
D
−Id
2
_
_
_
_
_
_
_
_
_
_
_
. . .
_
_
_
_
D
Id
2
D
−Id
2
_
_
_
_
_
_
.
Here
a
−+
= −a
+−
=
_
−1 0
0 1
_
.
Now what are the subfunctors that are locally down-sets? First, there is L

¯
Asa
itself. For any other
subfunctor S ⊆ L

¯
Asa
we have D
Id
2
,∈ S(⊥). Then for every unitary matrix u we have four choices for
down-sets of L

¯
Asa
(B
u
), namely D
Id
2
↓, D
ua
−+
u
−1 ↓, D
ua
+−
u
−1 ↓ and D
−Id
2
↓. For each B
u
we are free to
chose any of these four down-sets, as each defines a subfunctor of L

¯
Asa
. Hence we conclude that Σ(⊥) is
the lattice
L

¯
Asa
T(¦0, 1¦)
R
,
where R indicates the cardinal number of ¦u [ u is a unitary matrix¦. It follows straight from theorem
6.1.3 that
Σ(B
u
) = L

¯
Asa
(B
u
) =
_
_
_
_
_
_
_
_
_
_
_
D
Id
2

D
ua
−+
u
−1 ↓ D
ua
+−
u
−1 ↓
D
−Id
2

_
_
_
_
_
_
_
_
_
_
_
,
where the map ⊆
Σ
⊥,Bu
maps a subfunctor U ⊆ L

¯
Asa
, regarded as an element of Σ(⊥), to U(B
u
), regarded
as an element of Σ(B
u
).
6.3 Some remarks on C(X)
In this paragraph we study ((A) for A = C(X), the continuous functions on a compact Hausdorff space
X. For convenience we first assume X = [0, 1] ⊆ R, commenting the general case afterwards. Let Π be
99
6.3. SOME REMARKS ON C(X) CHAPTER 6. COMPUTATIONS OF SPECTRA
a closed partition of [0, 1], that is, a partition of [0, 1] such that every equivalence class is closed in [0, 1]
(with the euclidian topology). One can prove that the cardinality of the number of equivalence classes of
Π is not countable. If the number of equivalence classes is finite, then the partition is trivial. Now, as in
the situation of C
n
, we introduce the algebra C
Π
of all continuous functions on [0, 1] that are constant
on equivalence classes. There are three natural questions to ask:
1. Is C
Π
a C*-algebra?
2. For two closed partitions Π
1
and Π
2
, does the following implication hold: C
Π
1
= C
Π
2
⇒ Π
1
= Π
2
?
3. Is every unital C*-subalgebra of C([0, 1]) of the form C
Π
for some closed partition Π?
The first question
The answer to the first question is obviously yes. From this we make a short remark: we could, on the
one hand consider the category of partitions of [0, 1] with the refinements of partitions as morphisms, and
on the other hand the category of unital C*-subalgebras of C([0, 1]) with the inclusions. Then Π → C
Π
is a functor between these categories.
The second question
The answer to the second question is no. This can be illustrated by the following example. Let φ be a
function from the open inteval (0, 1) to itself having the following properties:
1. φ is a bijection.
2. For every open subinterval of (0, 1) the image of this subinterval under φ is dense in (0, 1).
Such a function exists as we shall prove later. Let
ϕ(x) =
1
2
+
1
2
φ(2x) : (0,
1
2
) → (
1
2
, 1),
and define ∼
ϕ
to be the smallest equivalence relation on [0, 1] such that if y = ϕ(x) then y ∼
ϕ
x.
Note that each of 0,
1
2
and 1 is only equivalent to itself and that every number in (0,
1
2
) is equivalent
to exactly two numbers, namely itself and the image of ϕ (which lies in (
1
2
, 1)). Call the corresponding
partition Π
∼ϕ
and call the trivial partition consisting of only one equivalence class Π
0
. We claim that
that C
Π∼ϕ
= C
Π
0
. Note that C
Π
0
only contains the constant functions. Hence we would have proved the
claim if every function in C
Π∼ϕ
is constant. Let f ∈ C
Π∼ϕ
. By definition of the equivalence relation and
the continuity of f we are finished if we can prove f to be constant on the interval (0,
1
2
). Suppose it is
not. Then there exist p, q ∈ (0,
1
2
) such that f(p) ,= f(q). Since f is continuous we can find a δ such that
if x ∈ (p −δ, p +δ) then
[f(x) −f(p)[ < [f(p) −f(q)[/2.
Now remark that f(ϕ(x)) = f(x) and ϕ((p −δ, p +δ)) is dense in the interval (
1
2
, 1). From this and the
continuity of f we may conclude f(x) takes values between f(p) ± [f(p) − f(q)[/2 for every x ∈ (
1
2
, 1).
Then
f(q) = f(ϕ(q)) ∈ (f(p) −[f(p) −f(q)[/2, f(p) +[f(p) −f(q)[/2),
which is a contradiction. Hence f is constant.
100
6.3. SOME REMARKS ON C(X) CHAPTER 6. COMPUTATIONS OF SPECTRA
It remains to prove the existence of the function φ. Let a
k,n
=
k
2
n
for natural numbers 2 ≤ n, 1 ≤ k < 2
n
and 2 ,[k. Let b
k,n,i
be a sequence in Q converging to a
k,n
such that b
k,n,i
= b
k

,n

,i
⇒ k = k

, n = n

, i = i

.
Such sequences b
k,n,i
exist. Define:
B
k,n
= ¦b
n,k,i
[i ∈ N¦ .
For p prime, define:
D
p
=
_
k
p
n
[k, n ∈ N, 2 ≤ n, 1 ≤ k < p
n
, p ,[k
_
.
Let f be a bijection between the countable sets
¦(k, n)[k, n ∈ N, 2 ≤ n, 1 ≤ k < 2
n
, 2 ,[k¦ ,
and the set of all primes. Note that both the sets B
k,n
and D
p
are countable. Hence there exist bijections
g
k,n
: B
k,n
→ D
f(k,n)
.
Let g be a bijection between the uncountable sets (0, 1)¸

k,n
B
k,n
and (0, 1)¸

p
D
p
. Finally, define
φ : x → g
k,n
(x) x ∈ B
k,n
;
x → g(x) x ∈ (0, 1)¸

k,n
B
k,n
.
We claim this function to be the desired φ . It is obvious that this function is a bijection. Now suppose
I ⊆ (0,
1
2
) is an open subinterval. It at least contains one point a
n,k
. Fix n and k for now. Then there
exists a N ∈ N such that for all i > N, b
k,n,i
∈ I. Then φ(I) contains D
f(k,n)
except for a finite subset
of D
f(k,n)
. From this it follows φ(I) is dense in (0,
1
2
).
The third question
The answer to the third question is yes. A proof can be established from the Gelfand - Naimark theorem,
some category theory and topology. Suppose B to be a unital C*-subalgebra of C([0, 1]). We would
like to prove B is equal to the continous functions on [0, 1] to C that are constant on equivalence classes
of some partition Π. We first introduce some notation. Let A be the C*-algebra C[(0, 1)]. If C is any
C*-algebra, write Ω(C) for the spectrum of C, meant by the Gelfand-Naimark theorem (rather than the
spectrum as defined in previous chapters). Let i : B → A denote the inclusion of B in A. Then we have:
B →
i
A
Ω(B) ←
i

Ω(A)
First remark that i is monic since it is injective. The Gelfand-Naimark theorem states that there is a
duality between compact Hausdorff spaces and unital commutative C*-algebras. From this one obtains
i

is epi. Now we claim i

is surjective. Suppose i

is not surjective. Then there is a point p ∈ Ω(B) such
that p is not in the image of i

. Remark that the one-point set ¦p¦ is closed, hence compact. The image
of i

is compact since i

is continuous and Ω(A) is compact. Ω(B) is compact Hausdorff, hence normal.
Hence we can apply Urysohn’s lemma giving a continuous function f : Ω(B) → [0, 1] such that f(p) = 1
and f(x) = 0 for all x ∈ i

(Ω(A)). Now let the function h : Ω(B) → [0, 1] be constant equal to 0. Then
f ◦ i

= h ◦ i

but f ,= h which contradicts with the fact that i

is epi. We conclude that i

is surjective.
Now define an equivalence relation on Ω(A) by:
e
p


e
q
if and only if i

(e
p
) = i

(e
q
)
101
6.3. SOME REMARKS ON C(X) CHAPTER 6. COMPUTATIONS OF SPECTRA
Here e
p
, e
q
∈ Ω(A) are the evaluations in p and q.
Then i

factors through (Ω(A)/ ∼

) by i

: Ω(A) →
π
(Ω(A)/ ∼

) →
¯
i

Ω(B), where both π and
¯
i

are
continuous. From the facts that:
-
¯
i

is bijective and continuous,
- (Ω(A)/ ∼

) is compact (since the quotient of a compact space is compact),
- Ω(B) is Hausdorff,
it follows that
¯
i

is a homeomorphism. So we may view i

as a map from Ω(A) to (Ω(A)/ ∼) (this is
actually the map π). The Gelfand-Naimark theorem tells us that
B · C(Ω(B)) · C(Ω(A)/ ∼

) · C([0, 1]/ ∼
0
).
For the last isomorphism, define for p, q ∈ [0, 1]:
p ∼
0
q if and only if f(p) = f(q), ∀f ∈ B,
and notice that f(p) = f(q), ∀f ∈ B if and only if i

(e
p
) = i

(e
q
), where e
p
, e
q
∈ Ω(A) are evaluation at
p and q, respectively.
Finally, we claim that C([0, 1]/ ∼
0
) is nothing else but the continuous functions from [0, 1] to C that
are constant on the equivalence classes defined by ∼
0
. It is easy to see that each continuous function on
[0, 1] → C that is constant on equivalence classes defines a continuous function on [0, 1]/ ∼
0
and that
this assignment is injective. On the other hand, if f is a function in C([0, 1]/ ∼
0
), then f ◦ π
0
, where
π
0
: [0, 1] → [0, 1]/ ∼
0
is the quotient map, is a continuous function that is constant on equivalence
classes, and this is the inverse assingment. Hence C([0, 1]/ ∼
0
) consist of all continuous functions that
are constant on (closed) equivalence classes. This proves claim number 3.
Further remarks
In the above discussion we only used the topological properties of [0, 1] being compact and Hausdorff.
The answers to the first and third question stay the same in the setting of any compact Hausdorff space
X instead of [0, 1]. The answer to the second question does depend on X, as we saw in the example of
a discrete n-point space and in the example of [0, 1]. The three questions amount to a representation of
all unital C*-subalgebras of C(X) where X is a compact Hausdorff space. Namely:
((C(X)) = ¦B ⊆ C(X)[B a unital C*-subalgebra of C(X)¦
· ¦Π[ Π is a partition of X¦ / ∼
Π
,
where Π
1

Π
Π
2
iff C
Π
1
(X) = C
Π
2
(X).
Of course the question wether two partitions are equialent with respect to ∼
Π
may become difficult to
anwer, but at least this representation gives an idea what the C*-subalgebra’s of C(X) look like and,
if the equivalence relation is neat enough, it even tells you when a C*-subalgebra is contained in one
another, i.e. it gives a hint what the lattice structure looks like. For the computation of the spectrum,
theorem 5.2.12 seems to be useful. However, the structure of ((A) is still too rough to compute the
spectrum properly.
102
6.3. SOME REMARKS ON C(X) CHAPTER 6. COMPUTATIONS OF SPECTRA
Remark: after finishing this thesis we found out that the C*-subalgebras of C(X) correspond to the
partitions of X that are obtained by a closed equivalence relation R. That is R ⊆ XX should be closed
in the euclidian topology of X X. Then the quotient space X/R is compact Hausdorff, as this follows
from elementary topology. The space of complex functions on this quotient space is a C*-subalgebra of
C(X) and it can be proved that every such algebra is of this form for a unique equivalence relation R.
103
Further work
In this chapter we make some final remarks. We have shown that if a C*-algebra is finite dimensional,
we are able to compute its spectrum once we have found its commutative subalgebras. As a special
case, we computed the spectrum of M(2, C), which was the main goal at the start of this research. For
the case of A = C(X), there is some research to do to find a representation of the structure of ((A).
Also, it would be interesting if the questions stated in paragraph 6.3 can be answered in the same way
as C
0
(X), where X only locally compact, instead of compact; i.e. in the general case of a commutative
C*-algebra. Of course, B(H) would be interesting too, especially since the GNS-construction shows that
every C*-algebra admits an injective representation in B(H).
Another generalization would be to change our topos T (A) to the setting of sheaves. The most natural
way to do this is by remarking that Set
C(A)
itself can be regarded as a sheaf if one equips ((A) with the
so-called Alexandrov topology. In this topology the open sets are defined as the up-sets of the partial
order ((A). If one puts some other topology on this space, one can take the sheaves on this topological
space as the definition of T (A). One can still define the free lattice subjected to relations and one can
still define a completion map (maybe you will need the to take the sheafification of a presheaf once in a
while). If we let U ⊆ ((A) be an open set and V
i
an open covering of U, then the properties of a sheaf at
V
i
give information about the sheaf at U, which might lead to more interaction between the information
of the C*-subalgebras of A and hence spectra may become different.
104
Index
Adjoint, 23
Cartesian closed, 23
functor category, 24
Category, 8
Closed partition, 100
Coequalizer, 17
Cokernel pair, 30
Colimit, 16
Compact, 42
Completion, 84
Composable pair, 8
Composition, 8
Composition map, 8
Constant functor, 12
Contravariant functor, 11
Contravariant Yonedafunctor, 12
Coproduct, 17
Cosieve, 21
Covariant functor, 12
Covering, 85
Down-set, 94
Epi, 11
Epi-monic factorization, 30
Equalizer, 15
Exists, 52
For all, 56
Forgetful functor, 12
Formula, 58, 60
Freely generated lattice, 73
Freely generated lattice subject to relations, 77
Functor, 11
Functor category, 13
Heyting algebra, 41
Hom-set, 10
Horizontal composition, 13
Identity, 8
Initial object, 10
Integral numbers object, 27
Internal
C*-algebra, 72
Heyting algebra, 52
lattice, 73
Intrinsic map, 13
Isomorphism, 9
Large category, 9
Lattice, 41
Left adjoint, 23
Limit, 15
finite, 17
Limiting cone, 15
Local, 75
Locale, 41
Mitchell-B´ nabou language, 58
Modus Ponens, 67
Monic, 11
pullback of a monic, 29
Morphism, 8
Natural numbers object, 27
Natural transformation, 13
Norm, 71
Object, 8
Opposite category, 10
PBL, 29
Point-wise, 75
Power object, 24
Preorder, 10
Presheaf, 14
Presubobject, 39
Product, 15
Pullback, 16
105
INDEX INDEX
Pullback functor, 37, 46
Pullback lemma, 29
Pushout, 31
Rational numbers object, 27
Right adjoint, 23
Sheaf, 14
Slice category, 36
Small category, 9
Source, 8
Spectrum, 88
Strict, 38
Subobject, 18, 39
Heyting algebra, 41
Subobject classifier, 18
uniqueness, 18
Target, 8
Tautological functor, 71
Terminal object, 10
Topos, 24
Type, 58
Yoneda functor, 12
Yoneda lemma, 28
106
Bibliography
[1] Borceux, F. Handbook of categorical algebra volume I, II, III, Cambridge University Press, 1994.
[2] Coquand, T. About Stone’s notion of Spectrum, Journal of Pure and Applied Algebra, volume 197,
2005.
[3] Coquand, T.; Spitters, B. Formal topology and constructive mathematics: the Gelfand and Stone-
Yosida representation theorems, Journal of Universal Computer Science, 28 December 2005.
[4] Coquand, T.; Spitters, B. An elementary constructive proof of Gelfand duality for C*-algebras, sub-
mitted for publication, 2007.
[5] Goldblatt, R. Topoi, the categorical analysis of logic, Elsevier Science Publishers, 1984.
[6] Hartshorne, R. Algebraic Geometry (8th edition), Springer-Verlag, 1997
[7] Heunen, C.; Lansman, N.P.; Spitters, B. A topos for algebraic quantum theory,
http://arxiv.org/abs/0709.4364, 6 April 2008.
[8] Johnstone, P.T. Sketches of an elephant, a topos theory compendium volume I, Oxford Science Pub-
lications, 2002.
[9] Lambek, J.; Scott, P.J. Introduction to higher order categorical logic, Cambridge University Press,
1986.
[10] MacLane, S. Categories for the working mathematician, Springer-Verlag, 1971.
[11] Moerdijk, I; MacLane, S. Sheaves in geometry and logic, Springer-Verlag, 1992.
[12] Mulvey, C.J.; Banaschewski, B.; The spectral theory of commutative C*-algebras: the constructive
Gelfand-Mazur theorem, Quaestiones Mathematicae, 23(4):465-488, 2000.
[13] Mulvey, C.J.; Banaschewski, B.; The spectral theory of commutative C*-algebras: the constructive
spectrum, Quaestiones Mathematicae, 23(4):425-464, 2000.
[14] Mulvey, C.J.; Banaschewski, B.; A globalization of the gelfand duality theorem, Annals of Pure and
Applied Logic, volume 137, January 2006.
[15] Priestly, H.A.; Davey, B.A. Introduction to lattices and order, Cambridge University Press, 1990.
[16] Vickers, S. Topology via logic, Cambridge University Press, 1989.
107
Acknowledgements
After devoting five years of my life to mathematics, this Masters thesis is the final result in becoming
a Master of Science. This is the right place to thank some people who have helped me over the past
few years. First, I would like to thank the staff of the mathematics department for teaching me, and
especially Klaas Landsman who taught me on the subjects of C*-algebra’s, Hilbert spaces and Category
theory and supervised my work for the past half year. His enthusiasm and helpfulness motivated me
during the past two years and eventually led to this thesis. Also I would like to thank Chris Heunen
and Bas Spitters for answering my questions and Mai Gehrke for spending her time to act as the second
reader.
Some special words should be spent on the Nijmegen mathematics students, who are very cooperative
and helped me a lot throughout the past five years. This was especially the cas in the last months, when
I tried to explain what my Masters thesis was about, even though some of them had no idea what a
C*-algebra or a topos was and hence possibly had no idea what I was talking about. Also, I would like
to mention Desda, the students association.
Finally, I would like to thank my parents, since they have always showed interest in the things I have
done and who were always prepared to help, to spend time to me and advise me. Thanks.
108

“Proof is the idol before whom the pure mathematician tortures himself.”

Sir Arthur Eddington (1882-1944)

Internal categories . . . . . . . . . . . . . . . . . . . .2 Pullbacks . . . . . . . . . . . . . . . . . . . . . . .1 The Yoneda lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Some examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . The internal Heyting algebra ∃ and ∀ . . . . . . . . .2 Special objects and morphisms . . . . . . . . . . . . . Mitchell-B´nabou language e The language . . . . . . . .2. . . . . . . .2 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Lattices and subobjects . . . . . . 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Some further examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 4 The 4. . .3 4. . . . . . .1 Categories and examples thereof 1. . . . . . . . . . . . . . . . . . . . . . . . . . 1. . . . . . . . . . . . . . .5 Rational numbers objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Epi-monic factorization 2. . . . . . . . . . . . . . . .3 Powerobjects . . . . 2 Topos theorems 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 4.4 4. . . . . . . . .4 Topoi . . . . . . . . .1. . . . . . . .1 Limits . . . . . . . . . . . . . . . 1. . . . . . . 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5 Special objects .4 Natural transformations . . . . . .1. . . . . . . . . . . . . . . . . . . . . .2 3. .2. . . . 1. . . . . . . . 1. . . . . . . . . . . . . . . . . . . .3 Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1. . . 2. . . . . . . . . . . . . . . . . . . . . . . . . . . 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Partial truth . . . . . .4 Limits . . . . . . .1 3. . . . . . . . . . 1. . . . . . . . . . . . . . . . . . .2 Subobjects . . . . . . . . . . . .2 Topoi . . . . . . . . . . . . . algebra of subobjects Subobjects . . . . . . . . . . . . .5 6 8 8 8 10 11 13 14 15 18 23 24 27 28 28 29 30 32 36 39 39 41 47 52 58 58 60 65 68 69 . . . . . . . . . . . . . . . . . . . . . . . . 3 The 3. . . . . . . . 2. . . . . . . . . . . . . . . . . . . . . . . . .3 3. . . . .1 Categories . . . . . . . . . . .Contents Preface 1 Categories and topoi 1. . . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . 1. . . . . . . . . . . . . . .2. . . . . . . . . . . .

. . . . . . . . . . . . . . . . Further work Index Bibliography Acknowledgements 5 . . . . . . . . .5 The construction of the spectrum 5. . . . . . . . . . . . . . . . . . . . . . 70 71 73 73 77 81 84 84 86 89 90 90 98 99 104 107 107 108 6 Computations of spectra 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3. . . .1 The map A0 . . . . . . . . . . . . . . . . 6. . . . . . . . . . . . . . . . . . . . . . . .1 The topos T (A) . . . . . . . . . . . . . . . . . . . . .3 Intermezzo: representing the local lattice 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . 5. . . . . . . . . . commutative case: Cn . . . . . . 5. . . . . . . . . . . . . . . . . . . . . . . . . . . .2.3 The spectrum . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . .2 Free lattices . . . . . . . . . . . . . . . . . . . . . . . . . .1 Construction of free distributive lattices . . . . . . . . . . . . . .2 Relations and the lattice L∼sa . . . . . . . .2 The finite case in general . . . . . . . . . .2 The morphism A . . . . . . . . . . . . . 5. . . . .1 The finite. . . . . . . . . . . . . . . . . . . . . . . 5. . . . . . . . . . . . . . . .2. . 5. . . . . . . . . . . . . . . . .3 Some remarks on C(X) . . . . . . . . . . ¯ A 5. . . .3. . . . . . . . . .4 Final remarks . . . . . . . . . . . . . . . . . . . . . 5. . . . . . . . . 6. . . . .

The first four chapters contain a brief introduction into topos theory. Klaas Landsman and Bas Spitters [7].6). This article introduces a very specific topos in which the authors define the notion of the localic spectrum of a C*-algebra. The text gives a brief introduction to topos theory. We turn our attention to an article written by Chris Heunen. A reader who is familiar with for example [10] can skip this chapter. which is a generalization of the spectrum of a commutative C*-algebra as used in the Gelfand-Naimark theorem. Once we developped enough topos theory we turn our attentention e to the generalized Gelfand spectrum in topos theory. The second chapter contains all proofs of theorems widely used in topos theory. commutative C*-algebra looks like and give the tools to compute the spectrum of any finite-dimensional (non-commutative) C*-algebra as long as we can compute its commutative 6 . This chapter has been added for completeness and we will refer to these pages several times. Here we restrict ourselves to the central example of monicity and prove an important fact which is used extensively in the last chapters: equality in the Mitchell-B´nabou language is equivalent to equality of morphisms (proposition 4. Furthermore. the second part is largely new. see for example the reference list. We show that all constructions are actually well-defined and doing so. These chapters show the power of e topos theory and the Mitchell-B´nabou language shows that a topos is a really nice generalization of set e theory. except for its last proof which becomes important. This theory was developped in the second e half of the 20th century and a vast amount of literature is available. our main goal was to compute the localic spectrum of the 2 × 2-matrices over the complex numbers. The Mitchell-B´nabou language is e important for the definitions in the last chapters. starting from the basic definitions and ending with the internal Mitchell-B´nabou language of a topos. At the start of writing this thesis. which is a very nice caleidoscope of subjects in topos theory. In terms of the Mitchell-B´nabou language they apply generalizations e of set theoretic constructions to their topos. we show what the spectrum of a finite-dimensional. In chapter 6 it is time to use the tools developed in chapter 5 to compute spectra of C*-algebras. In chapters three and four we introduce the underlying logic of a topos and the Mitchell-B´nabou language respectively. In chapter 4 some examples are given to emphasize this fact. We show how to compute these spectra and work out some examples. However. this language has a wide amount of properties which would take a whole book to describe (see [1] for example).Preface The aim of this thesis is to compute Gelfand spectra of internal C*-algebras in a topos. the underlying topos logic and the Mitchell-B´nabou language. we prove some important theorems that allow us to compute the spectrum. This thesis consists of two parts. A special note should be devoted to the book by Moerdijk and MacLane ‘Sheaves in Geometry and Logic: A First Introduction to Topos Theory’. It introduces the notation we use and gives the important examples for this thesis. In chapter 5 we introduce these constructions and make the definitions as described in [7] more precise and exlicit.3. e Whereas most theory in the first part of this thesis can be found in literature. This result is presented in chapter 6. The first chapter of this thesis contains precisely all definitions between a category and a topos.

There is no need to know anything about category theory. Here the spectrum becomes quite complicated and a nice representation has not yet been found. this introduction is very brief and it would be more interesting to read [10] to see that category theory can be applied to a wide variety of subjects in mathematics and really unifies these subjects.subalgebras. Concerning the prerequisites. The case of the 2 × 2 matrices then drops out as a special case. commutative case: the continuous complex valued functions on a compact Hausdorff space. student. ‘algebraic topology’. We do assume that the reader knows the very basics of lattice theory and logic. Sc. et cetera. 7 . we assume the reader is familiar with C*-algebras and the Gelfand theorem. However. to which the first chapter gives an introduction. we dare to taste a bit of the general unital. Finally. Furthermore. so that it can be used in ‘algebraic geometry’. we assume the reader has the level of an M.

defined on P = {(f. functors. These data satisfy the following rules: 8 . 1. The most important examples for this thesis will be covered.1. • A map called source : M or(C) → Obj(C).Chapter 1 Categories and topoi “Man muss immer generalisieren” . • A class of morphisms. which are worked out to some extent. a special kind of category.1 Categories In this chapter we define categories.1. Most of this material can be found in [10]. 1. We will write f ◦ g or simply f g instead of ◦(f. • A map called identity : Obj(C) → M or(C). although one might find it instructive to see what notation is used in the rest of the thesis. denoted as Obj(C). natural transformations and some definitions which are closely related to them.1.1 Categories and examples thereof Definition 1. g) ∈ M or(C) × M or(C)|target(g) = source(f )} . The main definitions will be clarified by some examples. • A composition map called ◦ : P → M or(C). A reader who is familiar with category theory can skip the first paragraph.Carl Jacobi This chapter introduces briefly the main concepts of category theory. For more examples and applications of topoi. The image of an object A under the identity map is usually denoted as IdA and is called the identity on A. g). denoted as M or(C). see for instance [5] and [11]. starting from the definition of a category and ending with the definition of a topos (plural: topoi or toposes). • A map called target : M or(C) → Obj(C). P is called the set of composable pairs. A category C consists of the following data: • A class of objects.

1. unital C ∗ -algebra’s with the (unital) C ∗ -morphisms as morphisms. The category Grp of (small) groups with the group homomorphisms as morphisms. CATEGORIES AND TOPOI 1. For this category one takes a universe U of sets and defines the small sets as the sets in that universe. 6. then f ◦ IdA = f = IdB ◦ f . Although this is not a very complicated category.1. consisting of 1 object and 1 morphism. which is not necessarily a set. the category is called small . If both the objects and morphisms form a set. Definition 1. consisting of two objects (the bullet and the star) and three morphisms (two identities and the arrow in the above diagram). Note that we assumed the objects and morphisms to form a class.e. 3. 3.q) Π(p. g) and (g. If f : A → B is an isomorphism. Composition is the composition of paths. 4.1. This is because the categories we wish to examine have a large number of objects. CATEGORIES CHAPTER 1. • > . 5. which is necessarily the identity on that object. q) be the set of homotopy classes of paths from p to q. The category Set of all small sets. For A ∈ Obj(C). In a category C a morphism f : A → B is called an isomorphism if there exists a morphism g : B → A such that IdA = gf and IdB = f g. Going into great detail about the difference between small and large categories is beyond the scope of this thesis. 7. 2. The morphisms are functions between sets. q ∈ P . source(IdA ) = target(IdA ) = A. (f ◦ g) ◦ h = f ◦ (g ◦ h). then A and B are called isomorphic. The most instructive way of grasping the concept of a category is by seeing some examples: 1. If f is a morphism with source(f ) = A and target(f ) = B. q) as the class of morphisms. The empty category C0 . Let P be the set of points of X and for p.2. we will refer to this example several times. All other categories are called large necessarily. h) composable pairs. In this thesis we use the notation f : A → B to say ‘f is a morphism such that source(f ) = A and target(f ) = B’. For (f. let Π(p. In some literature they are even called ‘arrows’. The other maps are evident. This construction is needed in order to make sure the objects form a class. consisting of no objects and no morphisms. Let X be a topological space. 9 . but one should at least notice this choice in the definition. Notation A B. The category cCStar of commutative. The source map and target map indicate from which object an arrow ‘departs’ and at which object the arrow ‘arrives’. We now state the notion of an isomorphism. The category C1 . Then this forms a category with the elements of P as objects and the union ∪(p. The morphisms are sometimes depicted as arrows. 2. i. the objects do not form a set anymore. The category C2 .

Definition 1. Let A. Examples 3. we can define the opposite category C op by: • Obj(C op ) := Obj(C). q ∈ Obj(P ). a preorder P regarded as a category gives rise to a partial order by putting for any two objects p. Definition 1. 10 . A category P is called a preorder if for every two objects A. P can be regarded as a category if we take the elements of P as objects and define a morphism to be a pair (p. This morphism will be denoted by !. 4 and 8 are very important for this thesis.3.q : p → q or simply ⊆: p → q.1. r) ◦ (p. With this composition. but we will always indicate which morphism is meant. 9.1. Do not be misled by this terminology. r). target((p. Conversely. i. target(f ) = B} . source(f ) = A. p ≤ q if and only if there is a morphism from p to q. A Hom-set does not necessarily have to be a set (but it certainly is a class). q) = (p. This might lead to some confusion. Let C be a category. q)) = q and (q. Define source((p.1. • M or(C op ) := M or(C). B ∈ Obj(C). if P is some set with a partial order.1. A • If g and f are composable morphisms in C. q) ∈ P × P such that p ≤ q. The elements of the partial order are the objects of P modulo isomorphism. 1.e. CATEGORIES AND TOPOI 8. Note that the transitivity of the partial order shows that this composition is a valid operation. Remark that by switching the source and target maps in the opposite category this composition is well-defined. target(f ) = source(g). Definition 1. B) = {f | f ∈ M or(C). • Idop := IdA . In that case P is a preorder and there exists a morphism between objects p and q. So the source map in the opposite category is the target map of the original category. If C is a category.2 Special objects and morphisms We define two special objects and two special types of morphisms. p) is the identity on p.4. This morphism will be denoted by ! as well. • targetop := source. An initial object 0 in a category C is an object such that for each object C in C there is a unique morphism from 0 to C. An important concept will be the one of ‘Hom-sets’. q)) = p.1. Conversely. then the composition in the opposite category is defined by: f op ◦op g op := (g ◦ f )op . We define the Hom-set of A and B as: HomC (A. B ∈ Obj(P) there is at most one morphism from A to B. (p. CATEGORIES CHAPTER 1. The other examples will be useful too.5.1. we denote it by ⊆p. A terminal object 1 in a category C is an object such that for each object C in C there is a unique morphism from C to 1. • sourceop := target.

The terminal object is any one-point set.e. in Set the initial object is the empty set.1. it is rightcancellable. CATEGORIES CHAPTER 1. the so-called functors. In diagrams we will denote epis by → →. 11 . • F 1 (f ◦ g) = F 1 (f ) ◦ F 1 (g). • A function F 1 : M or(C) → M or(D). By definition there are unique morphisms ! : 0 → 0 and ! : 0 → 0. • F 0 (target(f )) = target(F 1 (f )). They become: • F 0 (source(f )) = target(F 1 (f )).1. For example. i. For the same reason !◦! = Id0 .e. We usually write F : C → D. CATEGORIES AND TOPOI For example. • F 1 (f ◦ g) = F 1 (g) ◦ F 1 (f ). then they must be isomorphic. The embedding is not surjective. A morphism e is called epi or ‘an epi’ if f e = ge implies f = g. • F 0 (target(f )) = source(F 1 (f )). A functor F from C to D consists of: • A function F 0 : Obj(C) → Obj(D). Definition 1. saying F is a functor from C to D. For convenience of notation we write F (A) := F 0 (A) and F (f ) := F 0 (f ) if A is an object and f is a morphism of C.1.8 except for the relations. In diagrams we will denote monics by →. but it is epi.7. the following relations holds: • F 0 (source(f )) = source(F 1 (f )).8. Remark that this notation is consistent with the idea of viewing F as a morphism between categories. g ∈ M or(C) such that f and g are composable. In the same way one finds that 1 is unique up to isomorphism. since any function to the one-point set must be constant.1. one can take a category which does not have a notion of injectivity. These functions satisfy the following properties.6. Remark that if we have two initial objects 0 and 0 . Another example is the embedding of Q in R as topological spaces with the euclidian topology. 1. A morphism m is called monic or ‘a monic’ if mf = mg implies f = g. which must be the identity. The composite ! ◦! is the unique morphism from 0 tot itself. since there is only one function from ∅ to any other set. • F 1 (IdA ) = IdF 0 (A) . However. For any f.1. like example 6. Definition 1.1.1. Hence 0 0 . this is not always true.3 Functors Next we define ‘morphisms between categories’. i. Definition 1. Let C and D be two categories. it is left-cancellable. There exists a notion of a contravariant functor . It may be checked that in Set monic is equivalent to injective and epi is equivalent to surjective. The definition is the same as definition 1.

−) : f → f∗ . Define the constant functor ∆C : J → C on objects as ∆0 : Obj(J ) → Obj(C) : J → C and on morphisms as ∆1 : M or(J ) → M or(C) : f → C C IdC . where f∗ (g) = f ◦ g. A). The functors defined in 3. 12 . Then there is a functor HomC (A. Let A be an object. −) : C → Set. Let C be an object of C. Let A be an object. and on the morphisms as: HomC (A. Again we give some examples. Let J and C be categories. defined on objects as: HomC (−. A) : B → Hom(B. This functor is called the covariant Yoneda functor. Let U : Grp → Set be the forgetful functor . defined on objects as: HomC (A. A) : C op → Set. and on the morphisms as: HomC (A. B). the Hom-sets are sets. 4.e. −) : f → f ∗ . where f ∗ (g) = g ◦ f is composing with f on the right. 1. The intuitive idea is that a contravariant functor reverses the direction of the morphisms. Let C be a category with small Hom-sets. −) : B → HomC (A. the composite G ◦ F is a functor. This is a functor mapping the objects of Grp to Set. Here the composite maps are defined as (G ◦ F )0 = G0 ◦ F 0 and (G ◦ F )1 = G1 ◦ F 1 . i.1. forgetting the group structure of Grp. i. 5. 2.1. 3.8 are often called covariant functors. If F : C → D and G : D → E are functors.e. The functors of definition 1. Remark that this relation is the same as for a functor. composing with f on the left. whence a contravariant functor from a category C to a category D can be regarded as a functor from the opposite category C op to D.1. 4 and 5 will become important later on. Let C be a category with small Hom-sets. CATEGORIES AND TOPOI • F 1 (IdA ) = IdF 0 (A) . Then there is a functor HomC (−. CATEGORIES CHAPTER 1. This functor is called the contravariant Yoneda functor.

If we view the functors between C and D as objects and the natural transformations as morphisms between these objects.4 Natural transformations Next we define the morphisms between functors. one has τ (A) ◦ σ(A) = IdG(A) and σ(A) ◦ τ (A) = IdF (A) . i. then there is a way of composing σ and τ . One easily checks that this indeed defines a category.B without specifying f . and such that the following diagram commutes for every f ∈ HomC (A.B the intrinsic maps (of F ). Let C and D be categories. We define the category DC to be the category whose objects are the functors from C to D and whose morphisms are the natural transformations between the functors. In some literature this is called the horizontal composition. A natural transformation τ from F to G is a function τ : Obj(C) → M or(D). It is instructive to think about this definition a bit more. An object of this category is a functor F : C2 → Set. This is done point-wise.2. The composition of natural transformation is defined as point-wise composition. G(A)). i.10. f is implicitly defined to be a morphism from A to B F in C . B) ⊆ M or(C): τ (A) > F (A) F (f ) ∨ G(A) G(f ) ∨ F (B) > τ (B) G(B). If σ : G → H is another natural transformation.e.1. namely F 0 (•) and F 0 ( ). such that for any object A in C τ (A) ∈ HomD (F (A).1. F When we write fA. the identity IdF is the natural transformation assigning to every object A ∈ Obj(C) the identity IdF (A) . this notation coincides with definition 1. Let F and G be functors from a category C to a category D. As an example we study the category SetC2 . The next definition is the central object of this thesis. CATEGORIES CHAPTER 1. τ is a natural isomorphism if there is a natural transformation σ from G to F such that for any object A ∈ Obj(C). We write F G.1.B := F (f ). It is the notion of a functor category. If F is an object of DC and if f : A → B is a morphism in C.e.1. for two (composable) natural transformations τ and σ. We write τ : F → G when τ is a natural transformation from the functor F to the functor G. for every object A. This category is referred to as a functor category.1.1. We call the morphisms fA. then we write F fA. the so-called natural transformations. (σ ◦ τ )(A) = σ(A) ◦ τ (A). Definition 1. This functor can be regarded as two sets. The idea is to regard functors as objects and natural transformation as the morphisms between functors. together with a morphism between 13 . Furthermore. Definition 1. CATEGORIES AND TOPOI 1. one has (τ ◦σ)(A) = τ (A)◦σ(A).9.A and B. For an object functor F .

Recall that if V. Formally topoi are categories in which objects have subobjects (analogous to subsets). In the literature. One might think of it as a generalization of the concept of a set. U ∈ O(X) and V op is a subset of U . power objects (analogous to the power set of a set).Vi (p) = pi .V U. To complete this paragraph we give two more examples of categories. These categories turn out to be topoi and are widely used in the literature.10 we take SetP . Regard O(X) as a preorder and hence as a category using the inclusion to define a partial order. ⊆U.1. So the objects of SetC2 are the diagrams of the form {} → {}. Define the category of presheaves on C to be SetC . U. An object F in SetO(X) is called a sheaf is it satisfies the following properties: – If U is an open set in X.2. where {} are sets. i.A . namely F 1 (→). TOPOI CHAPTER 1. The morphisms between sheaves are just the morphisms of presheaves. These set-like categories are called topoi. then the intrinsic morphism F (f ) is denoted as fB. a pair of morphisms such that the following diagram commutes: F 0 (•) τ (•) ∨ 0 F 1 (→) > F 0( ) τ( ) ∨ 0 G (•) G1 (→) > G ( ). If one takes these formulations as axioms. diagrams in a category C are defined as functors from a category J (which ‘has the form of the diagram’) to the category C. then p = q.V – If U is an open set in X. • As a special case of definition 1. This is the category op which has as objects the contravariant functors C →Set. and finite (co)limits (which allows us to do set arithmetics). 14 . ⊆F i (p) =⊆F i (q). The most important properties of a set can be translated into a categorical formulation.1. This category will become important. we denote the corresponding morphism by ⊆: V → U . if {Vi } is an open covering of U and if we have elements pi ∈ F (Vi ) such that for every i. then there is an element p ∈ F (U ) such V V F that for every i. If F is an object of SetC and f : A → B F is a morphism in C. A morphism in SetC2 between objects F and G is a natural transformation τ : F → G. you could wonder what categories satisfy these axioms. if {Vi } is an open covering of U and if p. j ⊆Fi . q ∈ F (U ) are elements such that for each i. Intuitively the definition of a diagram is quite clear.e.Vi ∩Vj (pi ) =⊆Fj .Vi ∩Vj (pj ). • Let C be a category. CATEGORIES AND TOPOI those sets. where P is a preorder.2 Topoi A topos is a special kind of category. op 1. • Let X be a topological space and let O(X) be its set of opens. So the morphisms in SetC2 are pairs of arrows forming commutative diagrams.

Definition 1. We define the limit of F to be an object C in C together with a natural transformation τ from ∆C (the constant functor) to F .1 Limits Here we introduce the notion of a categorical limit. In the future the ‘projections’ corresponding to the natural transformation of a product will always be denoted by p or pi .. We state the precise definition now. so g ◦ f is the unique morphism h such that τ (J) = τ (J) ◦ h. 15 . One should think of limits of functors as ‘limits of diagrams’. Let J be the category consisting of 2 objects and only the two identity morphisms: J1 • •J2 . Remark that the functor F : J → C can be represented as a diagram ‘of the form J ’ with on each vertex an object of C and between them morphisms of C. However. provided that they exist. 2. 1. A1 > ∧ τ (J1 ) := p1 > B . We will denote this as lim F = C. A1 × A2 ∃! τ (J2 ) := p2 > ∨ A2 . Let J be the category: J1 • > > •J2 . such that for every other object D in C and natural transformation σ from ∆D to F there is a unique morphism f : D → C such that for every object J in J . For example.. The name ‘limit’ may let you think of an analytic limit in terms of and δ. Now τ (J) = τ (J) ◦ g ◦ f . take C =Set. where i is an index. Let A1 and A2 be two sets. TOPOI CHAPTER 1..1. σ(J) = τ (J) ◦ f .2.. just saying C is the limit of F . Obviously this h is the identity: g ◦ f = IdC . This is best depicted in some important examples below. The definition of a limit is a very abstract one. CATEGORIES AND TOPOI 1. Let J and C be categories and let F : J → C be a functor. Hence limits are unique up to isomorphism. These examples will be used extensively in this thesis. there exist unique morphisms f : C → C and g : C → C such that τ (J) = τ (J) ◦ f and τ (J) = τ (J) ◦ g. However..2. a categorical limit is totally different. denoted as F (J1 )×F (J2 )...2. Then the limit is an object of C together with ‘commutative morphisms from the object to the diagram’ (the limiting cone).. Limits are objects that are unique for some universal property. τ ) are limits. The limit of a functor F : J → C is called the product of F (J1 ) and F (J2 ).. Sometimes we will omit the natural transformation τ .1. In the same way f ◦ g = IdC . Remark that if both (C. Let F be the functor F : Ji → Ai . often limits can be depicted as a diagram.. τ ) and (C .. Then the following diagram shows the limiting product F (J1 ) × F (J2 ) = A1 × A2 is just the usual cartesian product of sets along with its projections as the natural transformation.

We list some important examples of colimits. The limit of a functor F : J → C is called a pullback . {(a. CATEGORIES AND TOPOI The limit of a functor F : J → C is called an equalizer. Let J and C be categories and let F : J → C be a functor. Definition 1.2. b) ∈ A1 × A3 | f (a) = g(b)} with the projections on A1 and A3 as morphisms. . •J 1 ∨ J3 • > •J2 . ∧ . It is exactly the same except that the limiting cone must be a class of morphisms from the diagram to the object (instead of in the other direction).2.1. Again take for example J = Set. . C 3. In Set the limit of the diagram A1 f ∨ A2 is the set g > A3 . Let A1 and A2 be sets and f.2. . σ(J) = f ◦ τ (J). ∃!. g both maps from A1 to A2 as indicated in this diagram: f A1 g > > A2 . We define the colimit of F to be an object C in C and a natural transformation τ from F to ∆C such that for every other object D of C and natural transformation σ from F to ∆D there is a unique morphism f : C → D such that for every object J in J . 16 . Let J be the category: inclusion > > f A1 g > > A2 . colimits are unique up to isomorphism. . . Remark that this diagram represents a functor F : J → C. Its limit is the set {a ∈ A1 | f (a) = g(a)}. TOPOI CHAPTER 1. The next definition is the dual of the previous one. {a ∈ A1 | f (a) = g(a)} . . as the following diagram obviously shows. In the same spirit as limits.

• > > •J 2 ... g both maps from A1 to A2 as indicated in this diagram: f A1 g > > A2 . B ∃! > A2 2.. Let A1 and A2 be sets and f. . CATEGORIES AND TOPOI 1. . f A1 g > > A2 > {[b] | b ∈ B} / ∼ .. For example. Let J be the category: J1 . denoted as F (J1 ) + F (J2 ). . Remark that this diagram represents a functor F . In chapter 2 we will prove a theorem that says all (finite) limits exist if all (finite) products and all equalizers exist. where ∼ is the equivalence relation generated by the relation bRb if and only if there exists an a ∈ A1 such that b = f (a) and b = g(a). Let F be the functor F : Ji → Ai ... The colimit of a functor F : J → C is called the coproduct of F (J1 ) and F (J2 ). Dually. Again take for example C = Set.1. 17 . TOPOI CHAPTER 1.. . The colimit of a functor F : J → C is called a coequalizer.. Let J be the category: J1 • •J2 . The following diagram then obviously shows that this set is the colimit of the diagram. take C =Set. Let A1 and A2 be two sets. A limit or colimit is said to be finite if the categorie J has a finite number of objects and morphisms. . . A1 τ (J1 ):=iA1 ∨ > A1 ∧ τ (J2 ):=iA2 > A2 . .2. We say that for a given category C all (co)limits exist if for every category J and for every functor F : J → C the (co)limit exists. Its colimit is the set {b | b ∈ A2 } / ∼. Then the following diagram shows the limiting product F (J1 ) + F (J2 ) = A1 A2 is just the disjoint union of sets along with its inclusions as the natural transformation. > ∨ C. . all (finite) colimits exist if all (finite) coproducts and all coequalizers exist.

2. The details of subobjects and their classifiers will be postponed to chapter 3. Id1 1 t ∨ > 1 t ∨ Ω f > Ω. A is the limit of the bottom right part of the diagram. The map χ is called the characteristic morphism of the subobject.2 Subobjects Now we generalize the notion of a subset. Proposition 1. the characteristic morphisms). if t : 1 → Ω is another subobject classifier.. Definition 1.e. B.e. there exist a unique morphism χ such that the following square is a pullback.1.2. This correspondence can be generalized by the following definition. ! A ∩ > t 1 ∩ m ∨ ∃!χ ∨ > B ..2. For now this definition will suffice. Here we will only prove the uniqueness of a subobject classifier and give some examples. Conversely. 1}). 18 .e. then it is unique up to isomorphism... assuming that it is clear which morphism t is meant. Sometimes we briefly call Ω the subobject classifier. For any diagram A⊂ m > t > Ω. i.4. Remark that in this definition m and t are monic. In Set every subset A of a given set B can be described by its characteristic function (from B to {0. where ! is the unique map from B to 1... If a category C has a subobject classifier t : 1 → Ω.. the function that is defined to be 1 at elements of A and 0 elsewhere. every characteristic function describes a subset of B.2.. i. Ω. TOPOI CHAPTER 1.3. A category C with a terminal object is said to have a subobject classifier if there exists a diagram 1⊂ with the following property. The exact definition of a subobject will be postponed to chapter 3. then there is an isomorphism f such that the following diagram commutes.. which will be called a subobject. That is.. It turns out there is a ‘bijection up to isomorphism’ between subobjects of B and maps from B to Ω (i. CATEGORIES AND TOPOI 1. For the moment we call a diagram A → B a subobject of B which is usually abbreviated by saying A is a subobject of B. For convenience of notation we write trueB ≡ t◦!.

then we can show that the dotted morphism is just k...... TOPOI CHAPTER 1.. CATEGORIES AND TOPOI Proof: Id1 Id1 1< t ∨ 1< t ∨ 1 t ∨ . 1} ... . {0. so we can restrict the target of k to A. Remark that in a subobject diagram m : A → B the set A can be regarded as a subset of B (since m is monic A is isomorphic to its image under m). Here i is the inclusion.. The pullback lemma (see chapter 2) proves that the outer rectangle is a pullback.2. Ω< g f The definition of the subobject classifier gives unique morphisms f and g that turn the above squares into pullbacks.... In Set the terminal object is any one-point set. 1} ....> . 19 .... which proves the proposition....... ! A ∩ > {1} ∩ m ∨ χA i ∨ > B . k ∩ h ! > > A m ∨ > {1} ∩ i ∨ > χA B {0. .. It is clear IdΩ has this property. one has χA k(c) = ih(c) = 1.. So k(c) ∈ A. Furthermore. .. To see this. So let A be a subset of B. We conclude g ◦ f = IdΩ . Now g ◦ f has the unique property of making the outer diagram a pullback..... remark that for any c ∈ C. Indeed the diagram commutes.... 1} . .. Ω < . Ω. We claim the subobject classifier is defined by the diagram: {1} ∩ i ∨ {0.. In the same way f ◦ g = IdΩ .. suppose we have a diagram: C ... which maps to a smaller target set. then we claim the characteristic morphism χA is the unique morphism turning the following square into a pullback...1..

. a → 2 if a ∈ S and h(a) ∈ T.. i... 1.. ! ! > ∨ > T ∩ τ2 > {1} > ∩ i1 S ∩ ! > {1} ∩ ∨ χ2 > τ1 h χ1 > B .. like in the previous example.... It is easy to see that a natural transformation τ : F → G is monic if and only if both ‘components’ are injective.2. To check if this cube is a pullback we look at the following diagram. We claim the right-hand side of the following cubic diagram is the subobject classifier..... then χ(b) = 0... 2 → 1 In order to prove the claim. Suppose b ∈ B. since A must be the pullback of the bottom right limiting diagram.. {0. 20 . we may take S and T to be subsets of A and B respectively.. Suppose b ∈ B.... TOPOI CHAPTER 1. then χ(b) = 1. both τ (•) and τ ( ) are injective. Another example is given by the category SetC2 of functors from C2 to Set...1..... 2} ... The map f is defined as: f: 0 → 0 1 → 1 ... b ∈ A.... {0.......... Let χ2 be the characteristic function of T and let χ1 be defined as follows: χ1 : a → 0 if a ∈ S and h(a) ∈ T... a → 1 if a ∈ S. since the diagram must commute... This obviously makes the cube commutative.. 1} i2 ∨ f ∨ > A ... The objects are just diagrams in Set of the form A → B and the morphisms are the natural transformations.. τ (•) > F (•) ⊂ ∨ G(•) ∨ F( ) ⊂ > τ( ) G( ).e. CATEGORIES AND TOPOI The map χA is unique... The morphisms i1 and i2 are inclusions....... b ∈ A......

depending on wether n is 1 or 2.. . is a generalization of the previous ones and will be important for the most important topos in this thesis. The condition that the bottom square has to commute forces the front bottom right object (Ω(•)) to be a larger set than {0.. we introduce the notion of a cosieve. {0.. respectively. The map χn is unique in making the cube a pullback.> ......... ..1..D := Ω(f ) and define Ω fC. Also for all b ∈ B.2.... then hf ∈ S for all h which have the same source as the target of f ..... Hence the map χ2 is determined. Definition 1.. 1. 2} ... The question ‘is a certain element part of the subobject?’ cannot simply be answered by yes or no... Remark that a cosieve on C is full if and only if it contains the identity on C...D : Ω(C) → Ω(D) : S → {h ∈ M or(C) | source(f ) = D........... Ω For a morphism f : C → D in C... then f (S) is indeed a cosieve on D.. 21 . .> .. . The commutativity of the bottom square now determines the map χ1 and we conclude that these χ’s are unique.. for all qn ∈ Qn notice that χn kn (qn ) = in hn (qn ) = 1......... The last example we examine.... A cosieve on C is a subset S of {f ∈ M or(C) | source(f ) = C} . k1 τ2 S {1} ∩ ∩ ∨ ∨ χ2 > > τ1 h χ1 > B .. 2}. Let C be a category and let C be an object in C. Hence kn (qn ) is in S or T ... yes and not yet: after applying h it will be.. . . The cosieve on an object C containing all morphisms with source C is called the full cosieve (on C). {0. The answer has become more subtle.. .. CATEGORIES AND TOPOI Q2 .... varying between no... .. > h2 ! ! > > > > {1} > ∩ i1 Q1 k2 h1 ! > T ∩ > For n ∈ {1. For all a ∈ A..... 1}.2. χ1 (b) = 1 if and only if b ∈ T .. We conclude that kn factors (obviously uniquely) through τn in such a way that the diagram commutes. Now let Ω(C) = {S|S is a cosieve on C} ........ hf ∈ S} and note that if S is a cosieve on C.. . recall that we defined fC... Let C be an arbitrary category.5. which has the following property: if f ∈ S. TOPOI CHAPTER 1. To see this. 1} i2 ∨ f ∨ > > A .... The elements of A that are mapped to 2 by χ1 could be interpreted as “those that eventually will be part of the subobject”.... χ1 (a) = 1 if and only if a ∈ S by demanding the square becomes a pullback. We can prove that SetC has a subobject classifier..

As a special case SetC2 has a subobject classifier and we compute: 1. Suppose it does have a subobject classifier. • → } . this combines to a commutative diagram of natural transformations.e. Ω(•) = {S | S is a cosieve on •} consists of the cosieves 0 := ∅. : 0 → 0. 2 → 1. As in the example of SetC2 . The conclusion is that SetC has a subobject classifier. CATEGORIES AND TOPOI The terminal object in SetC is obviously the functor ∆{1} sending each object C in C to the one-point set {1}. Consider the pullback: A ∩ ! > 1 = {0} ∩ t ∨ ∨ χ > B Ω.2. The morphisms χ(C) that form the characteristic natural transformation between F and Ω are defined as follows: F χ(C) : F (C) → Ω(C) : ac → f : C → D|D ∈ Obj(C).D (ac ) ∈ G(D) . is the subobject classifier. 1 := {• → } . Then in order to make the above diagram a pullback A must be isomorphic to the kernel of χ. However. Ω( ) = {S | S is a cosieve on } consists of the cosieves 0 := ∅. 1 → 1. we may take G(C) to be a subset of F (C) for all objects C. 1 := {Id } . a morphism in SetC (i. The diagram forms a pullback and χ is proved to be unique in having this property. a natural transformation) is monic if and only if all components are injective. fC. The map f•.1. if the image of A under the left monic is not a normal subgroup 22 . Again if F is an object in SetC and G is a subobject of F . We can prove that the diagram 1 t > Ω. indeed is defined as: Ω f•. As another example we prove that Grp does not have a subobject classifier. Ω 3. 2. TOPOI CHAPTER 1. Define a morphism (natural transformation) t : 1 → Ω by t(C) : 1(C) → Ω(C) : 1 → the full cosieve on C. 2 := {Id• . Again. Doing this is just a generalization of the proof that SetC2 has a subobject classifier.

If both C B and (C B ) have the property of (1. this certainly is not the case. if G is a right adjoint for F too.1. Similarly. for a map f : A × B → C: τ : HomSet (A × B. which is impossible. C 1 C.4. .1.7. in which every subgroup is normal. every group must be a subgroup of Ω. b))) is a natural isomorphism. Remark that HomD (F (−). too. there is a natural isomorphism HomC (A × B. Even if we consider the category of abelian groups Ab. The category Set is cartesian closed and the above notation reduces to the familiar meaning on sets. G(−)) can be regarded as a functor from C op × D to Set. That is. TOPOI CHAPTER 1. The following properties hold in any cartesian closed category C: 1.e. (1. Proposition 1. Adjoint functors are very important in category theory. C) → HomSet (A. Because then the image of χ is isomorphic to the group B modulo A.2. Note that we require naturality in A and C. then the functor preserves all limits. If a functor F has a right adjoint G. if a functor has a right adjoint. it is unique up to isomorphism. The exact meaning of this formulation is stated in theorem 2.1) If h : C → C .1). Next. The most useful result is theorem 2. 1. if G has a left adjoint. Since every group is the quotient of itself by the trivial group.1.2.2. then it is unique up to isomorphism. G(−)). then C B 2. A proof can be found in [10]. there is no subobject classifier.2. Conversely. I. C B ).1. A category C is said to be cartesian closed if it has finite products and if for any given object B the functor A → A × B (with the obvious map on morphisms) has a right adjoint. Definition 1. C B ) ((a. −) and HomC (−. b)) → (a → (b → f (a. then there is a natural isomorphism from G to G . then it preserves all colimits. −) → HomC (−. not necessarily in B. C) HomC (A. called −B . b) → f (a. That is. showing that if a functor has a left adjoint.8. 23 (C B ) . Definition 1. Let C and D be categories and let F : C → D and G : D → C be functors. we prove some properties using of the Yoneda lemma.4. This is done using the definition of adjoint functors and the notion of cartesian closed categories. we denote the morphism after applying −B by hB : C B → C B . We say that F and G are adjoint functors if there exists a natural isomorphism τ : HomD (F (−). In this case F is called a left adjoint. and G is called a right adjoint.1). CATEGORIES AND TOPOI of B. which is stated in chapter 2 (lemma 2.2.6.3 Powerobjects Now we like to generalize the notion of the powerset of a set.

after chapter 2 we skip the technical details more and more to avoid long pages of computations. The proof below is very technical in nature and seems not very elegant at all. There are many ways to define a topos. The category SetC is cartesian closed. DB ) HomC (A × B. CATEGORIES AND TOPOI 3.1. A category T is a topos if: 1. (C × D)B ). since limits can be taken point-wise (see theorem 2. So where is the power object? If a category is both cartesian closed and has a subobject classifier there is a correspondence between subobjects of an object A and morphisms χ : A → Ω.1. C) × HomC (A × B. C B × DB ) HomC (A. HomC (A.2. C) HomC (A. C B ) × HomC (A. It shows how abstract category theory can be. C B × DB (C × D)B . Proof: For (1) we have Now. Now in Set these morphisms ˆ correspond to points in ΩA since 1 = {1}. Hence we have found our generalization of the powerset of A: it is ΩA . (C B ) ). Remark that ΩA = {0. TOPOI CHAPTER 1. 1}A is simply the powerset of A. Here we will prove it is cartesian closed too.2.9. D) HomC (A × B. However. Also. G and H are functors from C to Set we must define a functor H G such that there is a natural isomorphism 24 .4 Topoi We conclude this paragraph by stating the definition of a topos. The other properties follow from this lemma by: HomC (A × 1. 1. it has all finite limits and colimits. although their results will be. it has a subobject classifier. SetC turns out to be a topos. C × D) HomC (A.1. Since the category is cartesian closed such a morphism corresponds to a morphism χ : 1 → ΩA . We have already seen that Set is a topos. which is based on the previous paragraphs and is most suitable for this thesis. C 1 ) . Hence it remains to show that if F.5).10.5 below tells us that this category has finite products. Proof: First remark that theorem 2. It has all finite limits. All proofs in the second chapter will not be of any further interest. Here we state a very categorical definition. The reason we do not postpone this proof to the second chapter is that we need the details of the proof for further use. C B ) HomC (A × B. C) HomC (A. Proposition 1. HomC (A. C) HomC (A. For SetC we have proved that it has a subobject classifier.2.4. 2. 3. Definition 1.2.4. it is cartesian closed. that of course are all equivalent. property (1) follows from the Yoneda lemma 2.

Hence H G is a functor. H(−)) as a functor from C to Set (although it can be regarded as a functor from C op × C to Set). However.2.B ∨ σ (A) ˆ > Hom SetC (HomC (A. F (A) F fA. One might expect H G = Hom SetC (G(−). σ → σ. TOPOI CHAPTER 1. In what follows. −) × G(−). First we check that σ is indeed a natural transformation. This is done by checking wether the following ˆ diagram commutes for every f : A → B in C. a → σ (A)(a). x). H G (f ) : Hom SetC (HomC (A. H(−)). ˆ σ (A) : ˆ F (A) → H G (A) = Hom SetC (HomC (A. H G ). (1. −) × G. σ → (f ∗ × IdG )∗ σ. We show that the correct definition is H G (A) = Hom SetC (HomC (A. Since H G must be a functor. A and B are objects of C. H). −) × G. H) → Hom SetC (F. there is no way we can regard Hom SetC (G(−). H G (g) ◦ H G (f ) = = = = = (g ∗ × IdG )∗ ◦ (f ∗ × IdG )∗ ((f ∗ × Id) ◦ (g ∗ × IdG ))∗ ((f ∗ ◦ g ∗ ) × IdG )∗ (((g ◦ f )∗ ) × IdG )∗ H G (g ◦ f ). σ F (g. ˆ ((ˆ (A))(a))(B) : HomC (A.B (a). x) → σ(B)(gA. H) Hom SetC (F. H) (f ∗ ×IdG )∗ ∨ . Remark that for composable morphisms f and g in C. we must specify the function on the morphisms f : A → B of C. −) × G and H. CATEGORIES AND TOPOI τ : Hom SetC (F × G. H G ). H) 25 . since other properties are trivial. τ : Hom SetC (F × G. −) × G. Next we define the natural isomorphism τ . B) × G(B) → H(B). −) × G(−). H(−)). We conclude H G is an object in SetC . the set of natural transformations between the functors Hom(A. −) × G(−). Note that the outer Hom is taken in SetC while the inner Hom is taken in C.2) F (B) > ˆ σ(B) Hom SetC (HomC (B. H(−)) → Hom SetC (HomC (B.1. H G (IdA ) = (Id∗ × IdG )∗ A = IdH G (A) . H(−)).

C F F σ(C)(gB.B σ We must check if the morphism τ is natural in F and H. It follows from a similar computation that τ −1 indeed defines a natural transformation and that σ is a ˇ −1 is indeed the inverse of τ : natural transformation. 26 . In particular it now follows that SetC is a topos for any category C. H G ) → Hom SetC (F × G. x) σ ((ˆ (A))(a))(C)(gf. x) σ σ(C)((gf )F .1.B (a)))(B))(IdB . H G ) ∨ ν ∗ µ∗∗ Hom SetC (F × G. b).B ν(A)a. b) = = (((σ(A))(a))(B))(g. H). ˆ The next step is to define the inverse of τ . x). (a. b) = (ˇ (B))gA. CATEGORIES AND TOPOI This is done by the following computation.B a. x)) F µ(B)(σ(B)(gA.B a.2. b) = (((ˆ (A))(a))(A))(IdA . σ (((f ∗ × IdG )∗ ((ˆ (A))(a)))(C))(g. as the following computation shows: (((τ (µσ(ν × IdG ))(A))(a))(B))(g.B (a). H G ). ˆ σ = (σ(A))(a.C fA. b). x).2. Then we must check if the following diagram commutes: τ Hom SetC (F × G. H ) > τ Hom SetC (F . The last thing to check is whether τ ˇ (σ (A))(a. x) = = = = = = ((((ˆ (A))(a))(f ∗ × IdG ))(C))(g. So let η : F → F and µ : H → H be morphisms in SetC . x) = = = = = F ((µσ(ν × IdG ))(B))(gA. x)) σ ((((µ∗ σ ν)(A))(a))(B))(g. x) = ˇ σ F = (((σ(B))(gA. x) F µ(B)(σ(B)(ν(B)gA. b). This is indeed the case. x) σ (((ˆ (A))(a))(C)(f ∗ × IdG )(C))(g. where the last equality follows from diagram 1. x) A. x) F (a)))(C))(g.B a. σ → σ. b) → σ(A)(a)(A)(IdA . TOPOI CHAPTER 1. ˇ σ (A) : ˇ F (A) × G(A) → H(A). b). This is done by τ −1 : Hom SetC (F. F ˆ (((σ (A))(a))(B))(g. (((ˆ (B))(fA. H) (η×IdG )∗ µ∗ ∨ > Hom SetC (F. x)) µ(B)(((ˆ (A))(ν(A)(a))(B))(g.

1.2. TOPOI

CHAPTER 1. CATEGORIES AND TOPOI

1.2.5

Rational numbers objects

In this paragraph we will try find a generalization of the rational numbers Q in Set to a topos T . This op can not be done for every topos, but in the case of Set, SetC and SetC , with C a category, it can. To do so one requires the topos T to have a natural numbers object, i.e. an object N together with morphisms 0 and s such that for every diagram
0 s

1 || 1

0

N . . . . . i. . . . . ∨ >X
>

s

N . . . . . i. . . . . ∨ > Y,
>

there exists a unique morphism i making the diagram commutative. This object plays a similar role op as the natural numbers in Set. Not every topos has a natural numbers object, but SetC , SetC and op (obviously) Set do have one. In SetC and SetC the natural numbers object is the constant functor ∆N . Notice that there is a morphism + : N × N → N, which in the case of SetC is defined by letting +(C) : N(C) × N(C) → N(C) be addition in Set. We may use this to define the integral numbers object Z as the coequalizer of the diagram:
(p1 f,p2 g) > > (p2 g,p1 f )

E

N×N

>

Z.

Here E is the upper left object in the following pullback:
f >

E
g ∨

N×N
+ ∨ >

N×N

+

N.

E is the object such that E(C) consists of 4-tuples (m, n, m , n ) such that m + n = m + n . In the case op of SetC and SetC a short calculation gives Z = ∆Z . Note that this object comes with a morphism m : Z × N → Z that is locally the multiplication operator, i.e. m(C) : Z(C) × N(C) → Z(C) is multiplication in Set. In a similar way one defines the rational numbers object Q, which in case of SetC op and SetC becomes Q = ∆Q . In section 4.5 we will argue why we would call this a generalization of the set-theoretic rational numbers Q. In chapter 5 we will need the rational numbers object in order to define a generalized norm in topos theory.

27

Chapter 2

Topos theorems
“I mean the word proof not in the sense of the lawyers, who set two half proofs equal to a whole one, but in the sense of a mathematician, where half proof = 0, and it is demanded for proof that every doubt becomes impossible.” - Karl Friedrich Gauss Throughout this chapter T will be a topos and any morphism or object will be taken within this topos unless specified otherwise. The aim of this chapter is to prove some often used theorems in topos theory. Some of these theorems may hold in a category which has only some of the properties of a topos, but for convenience we prove the statements within a topos T . Most of this material can be found in [10] and [11], however sometimes without proof.

2.1

The Yoneda lemma

If A and B are objects of a category C one might wonder what the natural transformations between the Yoneda functors HomC (A, −) and HomC (B, −) look like. The Yoneda lemma and its immediate corollary give the answer to this question. Lemma 2.1.1. (Yoneda lemma) If F : C →Set is a functor and A is an object of the category C, then there is a bijection of sets k : Hom SetC (HomC (A, −), F ) → F (A), which sends each natural transformation τ : HomC (A, −) → F to (τ (A))(IdA ). Proof: Let f : A → B be a morphism in C. Then the following diagram shows that k(τ ) determines the natural transformation τ , i.e. if you know what (τ (A))(IdA ) is, τ is completely determined. Hence k is injective. It is surjective too, since for every x ∈ F (A) one can define τ by putting (τ (A))(IdA ) = x.
τ (A) >

HomC (A, A)
f∗ ∨

F (A)
∨ F fA,B

HomC (A, B)

> τ (B)

F (B).

28

2.2. PULLBACKS

CHAPTER 2. TOPOS THEOREMS

Corollary 2.1.2. Let A, B be objects of a category C, then every natural transformation from HomC (A, −) to HomC (B, −) is of the form f ∗ for a map f ∈ HomC (B, A). Proof: This is the special case of lemma 2.1.1 with F = HomC (B, −).

2.2

Pullbacks

Here we will examine two properties of pullbacks. Together with products and equalizers the pullback is the most important type of limit in topos theory. Theorem 2.2.1. Let the diagram below be a pullback square: A
f ∨ ∨ m > n >

B
g

C If m is monic, then n is monic.

D.

Proof: Suppose we have morphisms h, k : E → A, such that nh = nk. Then m(f k) = g(nk) = g(nh) = m(f h), so f k = f h. By the universal property of a pullback there is a unique morphism from l : E → A such that nh = nl = nk and f h = f l = f k. The uniqueness of the pullback proves h = k. The next lemma will be used extensively in the next chapters. In the literature one refers to this lemma as the ‘pullback lemma’ or simply ‘PBL’. Lemma 2.2.2. (Pullback lemma, PBL) A
∨ >

B

>

C

>E > F. D 1. If in the above diagram both squares are pullbacks, then the outer rectangle is a pullback.

2. If the outer rectangle is a pullback and the right square is a pullback, then the left square is a pullback. Proof: 1. G ..... ...... ...... ...... ...... ...... > . >B A
. ..> .... .... .... ....
>

> >

C

> F. >E D Suppose we have morphisms from G to C and D as in the diagram. The pullback property of the right and left square gives the unique dotted morphisms to B and A respectively.

29

Then f = me where m is an image of f (thus monic) and e is epi. EPI-MONIC FACTORIZATION CHAPTER 2. ∨ C.3. . Remark that the morphism G → A → B is a morphism such that G → A → B → C equals G → C and G → A → B → E equals G → D → E. .3. .. if there is another pair u. This is the first result that shows topoi look like sets. This can be generalized to a topos..3 Epi-monic factorization In Set every morphism can be written as a composition of an epi with a monic. . Since the outer rectangle is a pullback.> . let f : A → B be a morphism. We denote the m source of m by Im(f ). t to be the morphisms obtained from the colimit of > > 30 .. .w . For every morphism f : A → B there exists a so-called cokernel pair of f .. . v : B → C such that uf = vf ... . In a topos T . such a morphism is unique and hence it equals the morphism G → B. .. there is a unique morphism from G to A such that the diagram commutes.. TOPOS THEOREMS 2. A ∨ > > > B ∨ > > C ∨ D > E > F. This gives a morphism from G to C such that the diagram commutes.. The epi is the map to the direct image of the morphism and the monic is the inclusion in the target set of the morphism. This is a pair of morphisms s. Definition 2.2. m factors through m . t from B to an object C which has the universal property sf = tf . The morphism G → D equals the factorization G → A → D.e. In a topos T .3.. G . . Since the right square is a pullback. Theorem 2. i. Proof: the proof makes use of two facts: 1.. . 2. thus f : A → Im(f ) → B...2.1. . Suppose we have morphisms from G to B and D as indicated in the diagram. then there is a unique morphism w such that the following diagram commutes: A f s > B t u v > > C . A monic m is called the image of f if f factors through m and whenever f factors through a monic m . This definition leads to an important theorem about (unique) epi-monic factorization. The proof of this fact is by taking s. let f : A → B be a morphism.

M . If f = me and f = m e with m. e m m A > M > M > B.. TOPOS THEOREMS B f > A f > B.. Suppose m is an isomorphism. e epi. 2.. there is a unique map s such that the diagram commutes. Take s. e must be epi... Then s = t. say f = me. 31 . e = m e . hence m equalizes u and v. By the uniqueness property of a cokernel pair there is a morphism w making the diagram above commutative. A || e f > A || s t > > A .2.. This gives m qm = m and since m is monic 1 = qm . Every monic m : C → D is the equalizer of two morphisms. This colimit is called a pushout.. By (2) h is an equalizer. say m = mm q for some q. t : B → C the cokernel pair of f and let m be the equalizer of s. Hence e = m e is epi... Theorem 2. In the previous theorem we proved the existence of an epi-monic factorization. Now note that um = wsm = wtm = vm. hence m factors through h. We find f factors through mm and so does the image m. since m is the equalizer of s and t.3. v. Now suppose that we have f = me without m being an isomorphism. Take an epi-monic factorization of e. Recall that an equalizer is always monic. t in the diagram below.. t. EPI-MONIC FACTORIZATION CHAPTER 2. then for morphisms r. It follows from the definition of a subobject classifier that m equalizes the characteristic morphism of m and the map trueD . say of u. It remains to prove that e is epi. Now take another factorization f = hg. The factorization turns out to be unique up to isomorphism as the following theorem and its corollary shows.. By m being monic this implies 1 = m q. so does f . B m > > C w ∨ || A g > || u N h > B v > > D. v. Since h equalizes u. s is the cokernel pair of f . Since f equalizes s. it factors through m.3. m monic and e.......3. t. This proves that m is the image of f .. So m is an isomorphism and e is epi.. Now since s..

. C). Then look at: ˜e e m A || A > C s ∨ > B || e ˜ > C m ˜ > B. Now as in fact (1) of theorem 2. Since m se = tme = m e r it follows from the fact that m is monic that se = e r and so the left square commutes. Hence m is the image of f . Since tf = m e r. . the cokernel pair must be a pair of equal morphisms.2. Compare the following diagrams in the category D and C respectively: 32 . We promised to give the proofs in this chapter. Let H : D → C be a left adjoint for G. Because m is monic P → B is monic (theorem 2. .1). ˜ Corollary 2. Since m was assumed to be the image of f and P → B is monic. Proof: First suppose m is the image of f .2. Because it is epi. . . Now suppose f = m˜ is an arbitrary epi-monic factorization of f . s is the equalizer of ˜ ˜ some cokernel pair. Then lim GF = GC.3.s .4. . .1. . Proof: This follows straight from theorem 2. ∨ m > > B P ∨ > m t B. Suppose lim F = C. TOPOS THEOREMS A e > r A > e C < ∨ C . LIMITS CHAPTER 2. . . . First we take care of the preservation of limits under adjoint functors. Let F : J → C and G : C → D be functors.3. Theorem 2. An epi-monic factorization of a morphism f is unique up to isomorphism.3. s is epi since se = e. GC)→HomC (HD.4. . Since m is monic.4. . . . . . Hence there is a map s such that the right square commutes. s is monic.3 taking f = f . . . s must be an isomorphism. this s is unique. 2. Let P be the pullback of t along m . But if s is the equalizer of two identical morphisms. f factors through P → B by the pullback property. ˜ Let D be an object in D and τ a natural transformation from ∆D to GF . m factors through P → B. .2.4 Limits In chapter 1 we introduced limits and stated some of their properties. . Proof: Call the natural isomorphism obtained from the adjunction ϕ : HomD (D. like in the diagram. . r = IdA and t = IdB . since ms = m.

1 by comparing the following diagrams.. Theorem 2. .4. Of course. By adjunction the first diagram commutes if and only if the second diagram commutes...2. HD > > ∃!k .. Proof: In the following diagram A. F jβ > One of the properties of a category in order to be a topos is having all finite limits and colomits. The following theorem will make life a lot easier...4. LIMITS CHAPTER 2. B and C are objects in J and the product A is taken over all objects of J ... σβ ∨ ϕτα > > GF jα > F jα > ∨ > GF jβ .. F (target(g)) ∧ ptarget(g) h k A psource(g) ∨ F (g) F (target(f )) < < f ptarget(g) ∨ F (target(g)) < > F jβ .4..... F jα ϕτα > σα > F fαβ > C . Let F : J → C and G : C → D be functors. = F (target(g)) ∧ ptarget(g) pB e F (A) < F (B) > ∧ C F (source(g)) 33 . f is taken over all morphisms of J .. Then colim GF = GC. Let H : D → C be a right adjoint for G.3... f and g are morphisms of J . D > σβ ϕτβ GF jβ . In general it is very hard to check if a given category has all finite (co)limits.... C ∃!k Gσα GF fαβ Gσβ τβ > σα F fαβ . one only has to check wether a category has (co)equalizers and finite (co)products. Theorem 2..4... In the second diagram there exist a unique morphism k such that the diagram commutes... By adjunction this gives the unique (!) (dotted) morphism ϕ−1 (k) making the first diagram commutative. Let J and C be categories. Proof: The proof is obtained in exactly the same way as theorem 2.. TOPOS THEOREMS > τα > D .2... GF jα τα Gσα GF fαβ Gσβ ∨ τβ ∨ > > > GC ... GC ϕτα > HD ... p denotes the projection on the object in its index. Suppose colim F = C. If C has equalizers and if C has all products of length card(Obj(J )) and card(Mor(J )) then the limit of every functor F : J → C exists.. there is a dual version of the previous theorem in terms of colimits and right adjoints.

So we see that if T has all finite products and equalizers. It is not! The theorem is a simple result of a diagram chase. Let F : J → C D and let GD : C D → C. D and J be categories (not necessarily topoi).4.2. σ(B) = τ (B) ◦ n. Let C. Then the morphisms σ(A) combine to a morphsim m : D → A F (A). If C has equalizers and if C has all products of length card(Obj(J )) and card(Mor(J )). τ → τ (D). F → F (D). then the limit of every functor F : J → C exist.4. Note that this defines a functor. 34 (2.e. Hence m = en for a morphism n : D → C by the unique property of an equalizer. TOPOS THEOREMS By definition of the left product in the diagram. Then if (lim GD F ) exists in C for all objects D of D. then lim F exists and one has (lim F )(D) = (lim GD F ). Let J and C be categories. then every finite limit exist since for these limits the cardinality of Obj(J ) and Mor(J ) is finite. So the morphisms τ (B) combine into a natural transformation from ∆C to F . LIMITS CHAPTER 2. Then σ factors through τ . Theorem 2. The same argument can be used for finite colimits as the next theorem shows. Now suppose we have an object D together with a natural transformation σ from ∆D to F .4. i.4. Then for a morphism g : B → B of J : F (g) ◦ τ (B) = = = = = F (g)pB e pB ke pB he pB e τ (B ).1) . The next theorem is might look complicated at first sight.5. Proof: The theorem is proved by the following diagram: F (source(g)) isource(g) ∨ = F (source(g)) h k ∨ > > A ∧ itarget(g) > F (g) isource(g) < iB e F (B) ∨ > F (source(f )) f isource(g) ∧ F (A) C F (source(g)) F (target(g)) . The property that σ is a natural transformation from ∆D to F is equivalent to the equation hm = km. Theorem 2. Let e be the equalizer of h and k. Write τ (B) := pB e. which is worked out for a simple but non-trivial case. there is a morphism h such that the upper square commutes for every g and there is a morphism k such that the lower square commutes for every g.

τD (•) K(D) > H(D ) > f (D ) H gD. Then we have σ(D) = τD .D gD...D hence an object of C D ....> K . D D.D > H(D) f (D) τD ( ) > ∨ τD ( ) G(D ) > ∨ G gD. TOPOS THEOREMS Let σ : ∆limF → F be the natural transformation implied by the limit of F and let τD : ∆limGD F → GD F be the natural transformation implied by the limit of GD F . The above diagram shows that the morphisms τD indeed combine into a natural transformation σ : ∆limF → F . the intrinsic morphisms of (lim F ) are determined by this equality.. Proof: The proof is a simple diagram chase... By a simple diagram chase. To keep things clear. G := F ( ). Furthermore.D ..D K then hK . we prove the theorem for J = C2 ...D G(D) > .. Let g : D → D be a morphism of D.D = (hg)K . Suppose L is another object that has a natural transformation ρ : ∆L → F . gD. . we have (IdD )K = IdK(D) and if h : D → D is a morphism in D. In this case the limit of the functor F can be represented as the following diagram: K > H f ∨ Where H := F (•). f := F (→) and K = lim F ... . 35 . We have proved the theorem if ρ factors through σ.2. K(D ) .. D. . Then the uniqueness of the point-wise limit of GD F = K(D ) shows that there is a unique dotted morphism making the diagram below commutative: τD (•) > G. This defines the intrinsic maps of K and determines K as a functor.4. LIMITS CHAPTER 2.

These diagrams can be seen as objects of what is called the slice category.D τD ( )β(D) combine into a natural transformation α : GD ∆L → GD F .. Hence these morphisms must be equal and the part where the dotted morphism is included commutes... In the next chapters we will study diagrams of the form B → A for a fixed object A. ..> L gD. Hence α must factor uniquely through τD .5.D. SPECIAL OBJECTS CHAPTER 2. Proof: Let F : C0 → T be a functor. there are unique (!) morphisms β(D) : L(D) → K(D) and β(D ) : L(D ) → K(D ) such that the above diagram commutes except maybe for the part where the dotted morphism is included.D τD (•)β(D) and G α( ) := gD. 2.2.2. . Let T be a topos and let A be an object of T . K L From the diagram above it follows that α factors both through gD. Hence ρ factors uniquely through σ.D G(D) > . . because in this case the definition of the (co)limit is equivalent to the definition of an initial object or a final object.. . Every topos T has an initial object 0 and a terminal object 1. We first show they always exist. Note that H α(•) := gD. For the next proof it is instructive to introduce the slice category.. TOPOS THEOREMS L(D ) . Notice that there is only one such functor.D β(D) and β(D )gD. 36 . Proposition 2.5 Special objects This section shows some properties of the initial and terminal object of a topos.β(D) > K(D) L(D) β(D ) > K gD... The inital object is the colimit of F and the terminal object is the limit of F ..5. .5..1. The slice category T /A is defined as the following category: • The objects are diagrams in T of the form f : B → A. Definition 2.D τD (•) K(D ) > τD (•) > H(D ) > f (D ) H gD.D > H(D) f (D) τD ( ) > ∨ τD ( ) G(D ) > ∨ G gD.. Since we took the limit point-wise...D .

. f −1.. . A very important result is that the functor f −1 has both a left and right adjoint.. Definition 2. 37 .3. The reason is that the proof is very technical. Let f : A → B be a morphism between objects A and B in a topos T .. We will refer to f −1 as the pullback functor.. by pulling back a morphism S → B along f .5. The result is simple..5.4. Proof: See for instance [11]. not very instructive to read and it would take about three pages to work out... This gives a functor f −1 : T /B → T /A.. (S → B) → (A ×B S → B) . . A ×B S → ∨ > . Theorem 2.. The proof of this is postponed to the next chapter.2. though we will use the result here to prove the following proposition. A morphism f : A → B induces a functor f −1 : T /B → T /A by pulling back. SPECIAL OBJECTS CHAPTER 2..(k) . but very important.. Then the slice category T /A is a topos. For a morphism k : T → S in T /A one defines f −1 (k) as the unique dotted morphism making the following diagram commutative: A ×B T ..5. Let T be a topos and let A be an object of T . < >S A ×B S < > k T A f > B < ∨ ∨ < . . . By diagram chasing this definition is well defined. TOPOS THEOREMS • A morphism from an object f : B → A to an object g : C → A is a morphism h : B → C such that the following diagram commutes: B h > C f g A < The next theorem is stated without proof... f −1 : T /B → T /A. S ∨ A f > B.. as the following definition shows.

are pullbacks. Since k −1 has a right adjoint k −1 (Id0 : 0 → 0) must be the initial object in T /A. k −1 (Id0 : 0 → 0) must be the the terminal object in T /A. SPECIAL OBJECTS CHAPTER 2. 38 .2.5. then f is an isomorphism.5. We say the initial object of a topos is strict. So we see that both the diagrams A IdA ∨ f > 0 !=Id0 ∨ 0 ! ∨ f > 0 !=Id0 ∨ A f > 0. Proof: In the slice category T /0 the object Id0 : 0 → 0 is both initial and terminal. Since A is initial. A f > 0. this isomorphism must be f.5. Hence A must be isomorphic to 0 (in T ). If f : A → 0 is a morphism to the initial object in a topos T . Since k −1 : T / 0 → T /A has a left adjoint. which is IdA : A → A. which is ! : 0 → A. TOPOS THEOREMS Proposition 2. Hence A is isomorphic to the initial object.

We will explore the structure of the subobjects of a given object in a topos. The structure we would like to study in this chapter is the one of the presubobjects modulo isomorphism.1. A presubobject f : B → A in a topos T is an object B in T together with a monic f : B → A.Pierre Boutroux In this chapter T is a topos and all objects and morphisms are taken within this topos unless specified otherwise. Define Sub(A) to be the class of presubobjects of A modulo ments of Sub(A) are called subobjects.1. We will just write B → A saying B is a presubobject of A. . The ele- > . 1.Chapter 3 The algebra of subobjects “Logic is invincible because in order to combat logic it is necessary to use logic. For example 0. The following definition of a subobject should not surprise the reader at all. In this chapter we make it more precise and work towards a lattice structure on the subobjects of a given object. . Remark that the definition we gave in chapter 1 only sufficed for the previous chapters. . colimits. Most of this work is based on [5].” . We define an equivalence relation on the presubobjects of an object A by (B → A) (C → A) if and only if there exists an isomorphism h such that the following diagram commutes: B h ⊂ > ⊃ C A < It is easily verified that this relation is symmetric and satisfies reflexivity and transitivity. When it is clear of which object B is a presubobject.2.3. 39 -equivalence. Mostly we omit the morphism f if it is clear which morphism is meant. we simply speak about the presubobject B. are all unique up to isomorphism. Definition 3.1 Subobjects Definition 3. 3. Ω.1. Definition 3. We will begin with defining a subobject. although some of the proofs are neater and a few corollaries have been added. limits. In the previous chapters we had a notion of uniqueness up to isomorphism for many of the definitions. .1.

We have already mentioned it: there is a correspondence between subobjects and characteristic morphisms. Both the left and right square are pullbacks. If B → A is a subobject we write χB for the characteristic morphism of B. then the following diagram commutes: ∼ C ∩ > B ∩ > 1 ∩ f ∨ ∨ IdA > t ∨ χB > A A Ω. SUBOBJECTS CHAPTER 3. THE ALGEBRA OF SUBOBJECTS The first proof of this chapter is not hard at all. Then both B and C are the pullbacks of 1 ∩ t ∨ A χ > Ω. Limits are unique up to isomorphism. Ω. Proof: Since T is a topos every subobject B ∈ Sub(A) admits a unique χB : A → Ω such that: B ∩ ! > 1 ∩ f ∨ t ∨ χB > A is a pullback square. C ∈ Sub(A) have characteristic morphism χ. Although the proof is easy.4. Suppose both B. From this point we alter our notation of a characteristic morphism. its consequences are important as we will see later.1. C ∈ Sub(A) and B Ω.3. This assignment is: • Well-defined. In a topos T . • Injective. So χB must be the unique morphism turning C ∩ > 1 ∩ t ∨ ∨ χB > A into a pullback. Theorem 3.1. so we may conclude B 40 . By the PBL the whole rectangle is a pullback. C. If B. there is a bijective correspondence between Sub(A) and characteristic morphisms A → Ω. C.

⊥ and phism.e. The symbols ⊥ and are called the bottom and top element. such that for every x. 3. A morphism f from a locale L to a locale M is a morphism ¯ of Heyting algebras f from M to L that preserves the (infinite) supremum. A morphism of Heyting algebras is a map that preserves ∩. ∪. Definition 3. Definition 3. Sub(A) will appear to be a Heyting algebra.2. for every subset U of L (possibly infinite) there exists an element z ∈ L such that x is the least upper bound of U : ∀x(x ∈ U ⇒ x ≤ z) and (∀x(x ∈ U ⇒ x ≤ y)) ⇒ (z ≤ y). ⊥ and as well as ⇒. such that for every x. since every claim made below will be true for presubobjects as well as for subobjects. respectively.3. respectively. i. From now on we will always consider subobjects instead of presubobjects. For a given characteristic morphism χ : A → Ω. i. A locale L is a Heyting algebra that is complete. Heyting algebras where introduced by Arend Heyting and play a role in intuitionistic logic. Definition 3.2. its pullback along t : 1 → Ω will yield a subobject B with χ as its characteristic morphism. y. . A morphism of lattices is a map that preserves ∩. Note that a morphism from L to M is a function in the reversed direction preserving the structure of a locale. For chapters 5 and 6 we will need the definition of a locale. THE ALGEBRA OF SUBOBJECTS • Surjective. ∪. This algebra is defined as a lattice together with some extra structure. Notation: z = U (z is the supremem of U ). LATTICES AND SUBOBJECTS CHAPTER 3.2 Lattices and subobjects We will prove that Sub(A) has more structure than just the structure of a set. which we state now.2. A Heyting algebra is a lattice L with a binary operation ⇒.2. x∩⊥=⊥ . Since all the lattices in this thesis do have a top and bottom element we prefer this definition. x ∪ y ≤ z iff x ≤ z and y ≤ z. y. Such a map is called a lattice homomor- The symbols ∩ and ∪ are sometimes called the inf and sup.3.1. z ∈ L: x ∩ y ≥ z iff x ≥ z and y ≥ z. z ∈ L: x ≤ (y ⇒ z) iff x ∩ y ≤ z. Some authors refer to the defintion above as a lattice with top and bottom element and define a lattice without having these elements. This can be done without confusion of notation.2. We will denote the subobjects like presubobjects. A lattice is a poset L together with binary operations ∪ and ∩ and nullary operations ⊥ and . the law of the excluded middle is not necessarily true. x∪ = . 41 . In the next chapter we will introduce (an interpretation of) a logical language which will turn out to be intuitionistic.e.

LATTICES AND SUBOBJECTS CHAPTER 3.4. Now. The correspondence between topological spaces and their underlying topology will be usefull for chapter 5. the set of opens O(X) is a locale. If f : X → Y is a map of topological spaces. it leads to a map of locales g : O(X) → O(Y ) by the inverse image function: g = f −1 : O(Y ) → ¯ O(X). thus: ∩ ⊃ B h ⊂ < > > C ∨ k > B f > f < A . by taking h = Id. Moreover. • For anti-symmetry. THE ALGEBRA OF SUBOBJECTS A locale L is called compact if for any subset U of L such that such that = V . zp is rather below zq if p < q. for which z0 = x. we say x is completely below y if there is a family zq ∈ L indexed by rationals 0 ≤ q ≤ 1. there is a finite subset V ⊂ U For a locale L and x. We have f ◦IdB = f ◦k ◦h. 42 . a locale L is called completely regular if for every x ∈ L x= {y ∈ L | y is completely below x} . B h ⊂ > C ∨ k > D .3. = U .2. if X is compact Hausdorff. Sub(A) has the structure of a partially ordered set by writing B ⊆ C if and only if there is a morphism h such that the following diagram commutes: B h ⊂ > ⊃ C < > A Proof: • Reflexivity is obvious. the locale is a compact completely regular locale. Furthermore. we say x is rather below y if there exists an element z ∈ L such that: x ∧ z = 0 and z ∨ y = 1. which amounts to B ⊆ D. Hence k = h−1 . y ∈ L. and z1 = y. so B C. the large triangle commutes.2. A Since the outer triangles commute. If X is a topological space. we conclude IdB = k ◦h. In this chapter our attention is focused on the algebra structure of Sub(A). Since f is monic. It turns out to be an Heyting algebra as the following series of definitions an propositions shows. In the same way IdC = h◦k. Suppose B ⊆ C and C ⊆ B. • Transitivity follows by: ∩ ⊃ . Proposition 3.

7. Let f : B → A and g : C → A subobjects of A. That is: k B∩C ∩ > .e. THE ALGEBRA OF SUBOBJECTS Proposition 3. In order to turn Sub(A) into a Heyting algebra. so are h and k (proposition 2. any map B → 0.2.1). This morphism is automatically monic: since in a topos 0 is strict (see proposition 2.2. it admits a unique morphism 0 → A. I. Now suppose we have a subobject m : D → A such that D ⊆ B and D ⊆ C. Hence IdA : A → A is the top element of the poset Sub(A). We conclude f ◦ k : B ∩ C → A is a subobject of A. This can be done as follows. Proposition 3. hence f ◦ k is.3.5. ∪ and ⇒. if D ⊆ B and D ⊆ C.2. Definition 3.6 really defines the ∩ in the poset Sub(A). B ⊆ A.2.2. Now since 0 is the initial object. namely the initial object and its unique morphism ! : 0 → A. we wish to define ∩.2. Proof: Since f and g are monic. f A < IdA ⊃ B > A that for every subobject f : B → A. Sub(A) has a top element . Sub(A) has a bottom element ⊥.6. ⊂ > B ∩ h ∨ f ∨ ⊂ C g > A. must be the unique (identity)map 0 → 0. LATTICES AND SUBOBJECTS CHAPTER 3. the following diagram commutes for any subobject f : B → A: ! ⊂ ! f A It follows from the commutativity of f ⊂ < ⊃ 0 > B > . Define the subobject B ∩ C → A of A by pulling back f allong g.5). Definition 3. Then: 43 . namely the subobject IdA : A → A. Proof: Since 0 is the initial object. then D ⊆ B ∩ C.5.

The universal property of the coproduct amounts to the following commutative diagram: B+C ∧ i > < k B ∩ h D f j ∨ > ⊂ C ⊂ g > A.. . In our topos the ∩ is interpreted as a pullback and hence this equation is an equality of objects which tells us pulling back B → A along C → A and then pulling back the result along D → A is the same (up to isomorphism) as pulling back C → A along D → A and then pulling back the result along B → A. Let f : B → A and g : C → A be subobjects of A.. That is: B+C ∧ e >> < k B ∩ h B∪C ⊂ f m ∨ > C ⊂ g > A. . Proposition 3.. C⊂ By assumption f ◦ i = m = g ◦ j. Definition 3. Next we define ∪.8.. Definition 3.8 really defines the ∪ in the poset Sub(A). Proof: Suppose we have a subobject j : D → A such that B ⊆ D and C ⊆ D. For example.2. which means D ⊆ B ∩ C.. THE ALGEBRA OF SUBOBJECTS D .e. The definition of ∩ already gives some interesting results.2.3.> .. so by definition of a pullback the dotted morphism exists. if B ⊆ D and C ⊆ D. j ∩ i ⊂ B∩C h ∨ > k > > B ∩ f g ∨ > A..2. then B ∪ C ⊆ D. (B ∩ C) ∩ D = B ∩ (C ∩ D) is true in an arbitrary partial order with a meet [15]..2. Define the subobject m : B ∪C → A of A by taking the image of the coproduct of f : B → A and g : C → A. . LATTICES AND SUBOBJECTS CHAPTER 3..9. I... 44 ... .

Replacing B by B ∩ C. . for subobjects B. hence χB ◦ m is the characteristic morphism of D ∩ B. Now split i into its epi-monic factorization: N⊂ . . . Let f : B → A and g : C → A be subobjects of A.10 really defines the ⇒ in the poset Sub(A). By definition of the partial order and the equalizer one has D ⊆ (B ⇒ C) if and only if m factors through e if and only if m equalizes χB and χB∩C . . LATTICES AND SUBOBJECTS CHAPTER 3. we define the implication ⇒. We will denote the morphism m in the above definition as f ∪ g.2. . . . . iff D ∩ B = D ∩ B ∩ C. iff D ∩ B ⊆ C.11. .2.3 there is an isomorphism (the dotted arrow) M → N such that the above diagram commutes. ∧ . Define the subobject e : (B ⇒ C) → A as the equalizer of the characteristic morphisms χB and χB∩C . . . A χB∩C > > Ω.4 as well as the fact that in D∩B ∩ >> .e.8 and j ◦ i in the above diagram are the unique morphism such that both f and g factor through B + C. Definition 3. By the PBL the outer rectangle is a pullback. e χB > (B ⇒ C) ⊂ Proposition 3. one readily finds that χB∩C ◦ m is the characteristic morphism of D ∩ B ∩ C. C. THE ALGEBRA OF SUBOBJECTS Note that both m ◦ e in definition 3. . I.2. .2.1. Write m : D → A for the subobject D. . . . . . With ∩. Hence they are equal: m ◦ e = j ◦ i.3. Sub(A) is a Heyting algebra. . Definition 3. D ∈ Sub(A): D ⊆ (B ⇒ C) if and only if D ∩ B ⊆ C. Finally we summarize the result in one theorem. > B ∩ > 1 ∩ true ∨ ∨ m > ∨ χB > D A Ω. .10. . In the second iff-statement we used proposition 3. . both the left and right square are pullbacks. 45 . .12. Theorem 3. It now follows that B ∪ C ⊆ D. Finally. Proof: First note that e in definition 3.3.2.2. We find: D ⊆ (B ⇒ C) iff χB ◦ m = χB∩C ◦ m.10 really is monic since it is an equalizer. ∪ and ⇒ defined as above. ⊂ B∪C > D ⊂ >> j > > B+C e A m According to theorem 2.2.

. A ×B S → ∨ S ∨ A f > B.e. Let F be the functor. {0} > {1} . Then ¬G must be equal to ? {1} which is not a proper function anymore. By applying theorem 2. the obvious next step is to define morphisms between algebras Sub(A) and Sub(B). This is not necessarily the case. by pulling back a morphism S → B along f .3...2... For example. 1} Let G be the subfunctor ! > {1} . We recall definition 2.(k) .. the ∪ and ∩ are the naive point-wise meet and intersection. THE ALGEBRA OF SUBOBJECTS You could wonder if Sub(A) has the structure of a boolean algebra.5. < >S A ×B S < > k T A f > B < ∨ ∨ < .5 to the definition of ∩ and ∪ and noticing that the epi-monic factorization can be taken pointwise. if there is some operation ¬ on subobjects such that B ∩ ¬B = ⊥ and B ∪ ¬B = .2. i. LATTICES AND SUBOBJECTS CHAPTER 3.13. . . f −1 : T /B → T /A (S → B) → (A ×B S → B) For a morphism k : T → S in T /A one defines f −1 (k) by the unique dotted morphism making the following diagram commutative.. Let f : A → B be a morphism between objects A and B in a topos T .4. . . one might look at the topos SetC2 ... This gives a functor f −1 : T /B → T /A. Definition 3. Since we are doing category theory... A ×B T .. depicted as a diagram: ! {0... f −1. 46 .4. > ∅...

∧ ◦ (χB . Let ∧ : Ω × Ω → Ω be the characteristic morphism of (t. iff B ∩ C = B.4 gives a correspondence between the Heyting algebra Sub(A) of an object A in a topos T and the characteristic morphisms A → Ω. we will observe that f −1 has a left and right adjoint. t) : 1 → Ω × Ω.3 about epi-monic factorization and the PBL. and hence it perserves ∩ and ∪ since they are the product. χC ) = χB ∩ χC .2.3 The internal Heyting algebra Theorem 3. C ∈ Sub(A).2. f −1 in definition 2. put: χB ∩ χC χB ∪ χC χB ⇒ χC ⊥ = = = = = χB∩C χB∪C χB⇒C ¬ trueA .14 is a morphism of Heyting algebras. B∩C ∩ ! > 1 ∩ ∼ > 1 ∩ (t.5.1. Since limits are unique up to isomorphism the pullback functor is well-defined on the equivalence classes defined in definition 3.4 descents to a morphism from Sub(B) to Sub(C).1). Ω). The map f −1 of definition 3. = χ⊥ . Namely.1.1.χC ) ∨ ∧ > t A Ω×Ω Ω 47 . We may use this correspondence to turn the set HomT (A.3. .15. The aim of this paragraph is to give a more direct description of the operators ∩.2.3. Actually.2. theorem 2. C ∈ Sub(A).3.15 will turn out to be a special case of a more general situation. Ω) is given by the obvious relation: χB ⊆ χC iff B ⊆ C. Ω) into a Heyting algebra.14.4. Proposition 3. THE INTERNAL HEYTING ALGEBRA CHAPTER 3.3. respectively coproduct of the lattice Sub(B) as a preorder. 3. In the next chapter. The most important property of the pullback functor is that it preserves the structure of the Heyting algebra. the fact that pulling back an epi (monic) gives again an epi (monic). The last two items define the initial and terminal object in HomT (A. The proof of the last theorem is not very elegant. Proof: We are finished if the outer rectangle in the following diagram is a pullback. Then f −1 preserves limits as well as colimits. using the universal property of pullbacks. Ω). For B.1.2. Since the pullback of a monic is monic (proposition 2. Proof: One can give a proof by diagram chasing.t) ∨ ∨ > (χB . It follows directly from the definitions that these are indeed the initial and terminal objects. ∪ and ⇒ on HomT (A. . Proposition 3. iff χB ∩ χC = χB . Then for B.2. The poset structure on HomT (A.2. THE ALGEBRA OF SUBOBJECTS Definition 3. we will give a short proof of proposition 3. proposition 3. = χ .15 using theorem 3.

..1. C ∈ Sub(A). .. We first prove the following square to be a pullback: B ∨ ⊂ f > A (χB .1 is equivalent to diagram 3. Hence there are unique morphisms from D to B and from D to C making the diagram commute. 1Ω ) ∪ (1Ω . the PBL shows the outer rectangle is a pullback. χC ) = χB ∪ χC . do a diagram chase on diagram 3.1) A Ω × Ω.. If we can prove the left square is a pullback. .. If you do not accept this last argument.. THE ALGEBRA OF SUBOBJECTS The right square is a pullback by definition of ∧. .3. The upper left square is a pullback. ∨ ◦ (χB . We need to prove that there is a unique morphism from D to B ∩ C.χC ) ∨ χC ◦f Ω > (trueΩ ...1Ω ) Ω × Ω. . hence there is a unique morphism from D to B ∩ C making the diagram commute.t) ∨ (3. The bottom and the right squares are pullbacks. Then for B. From this diagram.3.χC ) > > > 1 (t. Since diagram 3.1 commute.. we obtain: D B∩C ∨ > > > B > > 1 t (3... this is the unique dotted morphism making diagram 3. indicated with the dotted arrow in the diagram above.> . THE INTERNAL HEYTING ALGEBRA CHAPTER 3.2) ∨ C ∨ > > ∨ >A χB > Ω χC ∨ t > 1 Ω . B∩C ∨ > (χB .2. Let ∨ : Ω × Ω → Ω be the characteristic morphism of (trueΩ . We split the proof in three steps: 1. 48 .3. Suppose we have a commutative diagram: D . Proposition 3... trueΩ ) : Ω ∪ Ω → Ω × Ω.2.

...1Ω )∪(1Ω ... Ω ∪ Ω f ! > 1 t ∨ ∪g ∨ ∨ (χB ....χC ) > (trueΩ ..... 2.. since f is monic it follows that l is a unique such arrow..... .. THE INTERNAL HEYTING ALGEBRA CHAPTER 3..trueΩ ) A Ω×Ω ∨ > Ω. l m k ⊂ B ∨ > f > > A χC > χC ◦f χB Ω ! ∨ 1Ω trueΩ t > ∨ >Ω > Ω 1 ... then l = l. one sees that both the above diagrams represent the same situation and the morphism l is the unique morphism turning the first diagram into a pullback.. Hence χC ◦ f ◦ l = χC ◦ k = m...... THE ALGEBRA OF SUBOBJECTS Suppose we have an object D such that the following diagram commutes: D .. hence there exists an l as indicated... Now take a look at the diagram: > (1Ω . From the diagram we obtain m = 1Ω ◦ m = χC ◦ k.. In the same spirit... such that f ◦ l = k. .trueΩ ) Ω × Ω. . ! ◦ χC ◦ f ◦ l = ! ◦ m.3.. for if l ◦ f = k = l ◦ f . The rectangle ‘B − A − Ω − 1’ is a pullback... B+C ∨ ∨ ∩ i1 ◦χB ◦f +i2 ◦χC ◦g > Ω+Ω ∨ ∨ ∩ > B ∪ C . Looking closely.......> . 49 .... the following square is a pullback: C ∨ ⊂ g > A (χB .......3. Furthermore..χC ) ∨ χB ◦g Ω 3. So the whole diagram commutes.... .....

> .3. . Take the diagram D . pull back (χB . or alternatively. First note that since p1 ◦ (χB . 50 ... that makes the left square commute.. since the pullback functor has a left adjoint (as we will prove later) and since the pullback of an epi is epi since the pullback functor has a right adjoint (as we will prove later). Hence ∨ ◦ (χB .. trueΩ ). where p1 is the projection on the first coordinate. χC ) = χB and ∧ ◦ (χB .χC ) > ∩ e ∨ A Ω×Ω ∧ p1 ∨ ∨ → > Ω . 1Ω ) ∪ (1Ω . χC ) ◦ m = p1 ◦ e ◦ k = ∧ ◦ e ◦ k = ∧ ◦ (χB .. Since the pullback functor (χB .2. Ω. l m k ! > > B ⇒ C . χC ) ◦ f = ∧ ◦ (χB . χC )−1 has a left adjoint (as we will prove in theorem 3. ≤ > 1 t ∨ h f ∨ > (χB .. Since the pullback of a monic is monic (theorem 2..3. Proposition 3.. trueΩ ) and pull the result back along Ω+Ω → Ω∪Ω. it preserves colimits. If we can prove that the left square is a pullback. Theorem 2. χC ) ◦ m. m as indicated in the diagram..3. Hence there is a unique morphism h : (B ⇒ C) →≤. χC ) along (trueΩ .. Let →: Ω × Ω → Ω be the characteristic morphism of the equalizer e of e > ∧ ≤ Ω×Ω p1 > > Ω..3 gives a unique map from B ∪ C to Ω ∪ Ω. which is χB ∪ χC by definition...3. This turns the bottom left square into a pullback. This square actually turns out to be a pullback. where f denotes the equalizer of both compositions.. To see this.1). hence the bottom square must be the first pullback. This factorization is unique.. χC ) = χB ∩ χC . (χB .. Now. The (bottom) right square is a pullback by definition.....1). χC ) ◦ f equalizes p1 and ∧. The PBL tells us this double pullback is a pullback itself. Proof: Our strategy will be the same as in propopositions 3. .4. The PBL gives that the bottom (outer) rectangle is a pullback.. . the left (outer) rectangle is a pullback. THE INTERNAL HEYTING ALGEBRA CHAPTER 3..3. so that the outer left square is a pullback too.. which proves the proposition. Note that p1 ◦ (χB . Then for B. → ◦(χB . 1Ω )+(1Ω .2. χC ) is the characteristic morphism of B ∪ C. χC ) ◦ f. Combining this with step 1 and step 2. p1 ◦ (χB . The right square is a pullback (by definition)... χC ) = (χB ⇒ χC ).3.....3. THE ALGEBRA OF SUBOBJECTS In the left outer rectangle. Suppose we have an object D and morphisms k. the PBL tells us that the whole rectangle is a pullback... C ∈ Sub(A). . we took the epi-monic factorization of f +g and (trueΩ .. it follows that the two morphisms obtained from the double pullback must form the epi-monic factorization of f + g.1 and 3.

Ω) → HomT (B. we find morphisms: ∧A : ΩA × ΩA → ∨A : ΩA × ΩA → →A : ΩA × ΩA → ⊥: 1 → : 1 → ΩA .3 and the definition of f ∗ gave a direct description of the Heyting algebra structure of HomT (A. f −1 (T ) ∩ > T ∩ ! > 1 ∩ ∨ ∨ f > ∨ χT > A B Ω. 3. ΩA . ΩA . Ω) : χT → χT ◦ f. Propositions 3.3. one finds f −1 (T ∪ S) = f −1 (T ) ∪ f −1 (S) f −1 (T ⇒ S) = f −1 (T ) ⇒ f −1 (S). χC ) ◦ f ◦ l = e ◦ h ◦ l. With theorem 3. A morphism f : A → B gives a morphism f −1 : Sub(B) → Sub(A) as defined in the previous chapter. To see that for this unique l one has h ◦ l = k. hence it factors uniquely through f by l. THE ALGEBRA OF SUBOBJECTS So m equalizes χB and χB ∩ χC . we conclude h ◦ l = k. χS ◦ f ) = χf −1 (T ) ∩ χf −1 (S) .4 we conclude f −1 (T ∩ S) = f −1 (T ) ∩ f −1 (S). T ∈ Sub(B) one has f ∗ (χT ∪S ) = ∧ ◦ (χT . χC ) ◦ m = (χB . ΩA .3. ΩA . Since the left and right squares in the following diagram are pullbacks.2.2.3. χS ) ◦ f = ∧ ◦ (χT ◦ f.1. Proof (of proposition 3. The corresponding map on the characteristic morphisms is f ∗ : HomT (A. 51 .3.3. Replacing ∧ and ∩ by ∨ and ∪ or → and ⇒ in the above equations.2 and 3. Hence f ∗ χT is indeed the characteristic morphism of f −1 (T ). Now if we apply the functor −A obtained from the cartesian closedness of our topos. Since e is monic. Ω). THE INTERNAL HEYTING ALGEBRA CHAPTER 3. the PBL turns the outer rectangle into a pullback.15.15): For S.1. compute e ◦ k = (χB . We are now in a position to prove proposition 3.

∧A and ∨A satisfy the absorbtion laws.4. This is a generalization of the concept of a Heyting algebra in Set.e. ΩA × Ω × ΩA > IdΩA ×⇒A ΩA × ΩA . then both the following squares are pullbacks: Q ∨ A > E e ∨ Q ∨ A > E e ∨ ΩA × Ω × ΩA > ∧A ×IdΩA ΩA × ΩA . idempotence and the unit law (0 ∨A φ = φ). ⇒.4 ∃ and ∀ In this section we will prove that the pullback functor has both a left and a right adjoint. Note that both diagrams contain the same object Q. These are called ∃ and ∀. For example. however. if E is the equalizer of e > ∧ E ΩA × ΩA π1 > > ΩA .1. Definition 3. In the next chapter we will see that most of the set theoretic properties of ∃ and ∀ generalize to topoi. ∨. ∃ AND ∀ CHAPTER 3. Defining the topos-theoretic operations ∃ and ∀ is not as simple as in the case of Set. Let f : A → B be a morphism in a topos. and ⊥ that satisfies all the above properties is called an internal Heyting algebra in the topos T . i. where S → e ∃f (S) →m B is the epi-monic factorization of the composite S →i A →f B. The left adjoint is given by: S ∩ e > > ∃f S ∩ i ∨ m ∨ A →f B. The induced map f −1 : Sub(B) → Sub(A) has a left adjoint as a morphism of preorders. ∨A satisfies associativity. → 52 . Furthermore. In the same way one finds that ∧A is commutative.4. since the following diagram commutes: ΩA × ΩA × ΩA ∧A ×IdΩA ∨ A IdΩA ×∧A > ΩA × ΩA ∧A ∨ A Ω × ΩA ∧A > Ω . and that it satisfies the idempotence law (φ ∧A φ = φ) and the unit law (1 ∧A φ = φ). The dual relations for ∨A are true too.3. 3. commutativity. Thus we have proved ΩA is an internal Heyting algebra. An object L with morphisms ∧. THE ALGEBRA OF SUBOBJECTS Note that we used the isomorphism (Ω × Ω)A ΩA × ΩA here. These morphisms have the property that resemble of a Heyting algebra in Set. respectively and are generalizations of the symbols ∃ and ∀ in Set. ∧A is associative.

0). Then ∃p A = Im(p ◦ i) = e ({(0. 1}.. Hence there is a unique isomorphism between ∃f S and Im(sku) Im(ku) (s was already monic). .. Let p : X × Y → X be the projection on the first co¨rdinate.1 is the left adjoint of f −1 as morphism of preorders. If v exists.. To see that some examples of properties do not hold.. 1} × {0. f −1 T ) is equivalent to the statement: a morphism u exists such that the diagram below commutes if and only if a morphism v exists such that the diagram below commutes... ... Proof: First observe that the isomorphism HomSub(B) (∃f S. ∃ AND ∀ CHAPTER 3. By the universal property of a pullback there is a unique morphism u such that the diagram commutes. take the topos Set.. .⊃ .3.. For instance. .. > s m . one might ask if ∃f preserves ∩.. Let X = {0. sku has a unique epi-monic factorization.2.. The function ∃f in definition 3. 1}. then mr = f i = f ru = sku..4.> u . This is in general only true for some of these logical connectives. T ) HomSub(A) (S. THE ALGEBRA OF SUBOBJECTS We prove that this definition is correct.4. 0)}) = {0}.. . s ◦ v ◦ e = m ◦ e = f ◦ i. 1)} ∩ > > {0} ∩ i ∨ m ∨ > p {0. 0)}.. ⊃ k v > T S ∩ r e > > ∃f S ∩ A < f > B < i ∨ ∨ . (0. 1} and Y = {0. choose A = {(0.... To see this.3. But then the pullback of m along p is e {(0. . ∪ and ⇒. Some properties one might expect to be true for the left adjoint ∃f are infect false. . Now if u exists.. Proposition 3. Then: o • The following diagram is in general not a pullback: e > > A ∩ ∃p A ∩ i ∨ m ∨ p > X ×Y X. ..4.. 1} 53 {0. .3. f −1 T . Now according to theorem 2.. which gives the unique (!) morphism v....

3. hence a category. We conclude that f −1 (S) = S × A. Proof: The proof is obtained in two steps: 1. Take A = {(0. hence in particular the coproduct (theorem 2. AA ). proving that the slice category T /A is a topos. since it is easily verified that every pullback over 1 is just the product. 54 . This is not as easy as proving the existence of its left adjoint. ∪ is the coproduct of this category. A) > ϕ HomT (S. The proof can be found in [11].4.4.3. Let f be a morphism from A to B. The functor f −1 : T /B → T /A has a right adjoint. is a pull back square. We will split the proof in two steps. called πf . We find that p1 S×A p2 ∨ > S ∨ A f > 1. Now let h : T → A be an object over A. T A ) ∨ (hA )∗ HomT (S × A. T ) h∗ ∨ HomT (S. 0)} ⊆ X × Y . We will prove that f −1 has a right adjoint too. • ∃p does not preserve ∩. Since ∃p has a right adjoint (namely p−1 ) it must preserve colimits. Let T be a topos and let A and B be objects of T . making use of the slice category (see chapter 2). But ∃p (A) ∩ ∃p (B) = {0} ∩ {0} = {0}. 0)}. The morphisms t correspond to morphisms t : S → T A such that hA ◦ t (the LHS of the equality) equals the composite S →! 1 →j AA (the RHS of the equality). Then ∃p (A ⇒ B) = {0. 1)} ⊆ X × Y and B = {(0. • ∃p does not preserve ⇒.2). First assume B is the terminal object 1. Then ∃p (A ∩ B) = ∃p (∅) = ∅. • ∃p does preserve ∪. We use the adjunction of cartesian closedness to apply the functor −A on both sides of the equality.4. Let A be an object in a topos. Then we can regard Sub(A) as a preorder. 1} = {0} = ∃p (A) ⇒ ∃p (B). Take A = ∅ and B = {(1. In Set one can easily show that for subobjects A and B of X × Y that A ⇒ B = B c ∪ A. This proof uses the theorem stated in chapter 2. THE ALGEBRA OF SUBOBJECTS Hence the first diagram can not be pullback diagram. Here j is the morphism corresponding by cartesian closedness to the identity on A. The most important proposition is the following one. Proposition 3. A morphism from S × A to T over A is a morhpism t : S × A → T such that h ◦ t = p2 . ∃ AND ∀ CHAPTER 3. This can be seen from the following diagrams: ∼ > ϕ HomT (S × A.

Hence this defines the desired right adjoint. A ! 1 ×AA T ∨ j > > > > TA hA ∨ A 1 A ... . . Where we take the obvious map on morphisms. φ(p2 ) = φ(p∗ ◦ IdA ) =!∗ ◦ φ(IdA ) = j◦!.e. 55 . By the universal property of pullbacks. This is the required πf ... we are in the situation of step 1 and so f −1 has a right adjoint. ∃ AND ∀ CHAPTER 3. t t . An object over f : A → B in the double slice category (T /B) / (f : A → B) is a commutative square g C h ∨ > B IdB ∨ A f > B.3. The morphism f : A → B can be regarded as an object of the slice category T /B. AA ) !∗ ∨ HomT (S × A.. the maps on the objects are mutual inverses and the maps on the morphisms are mutual inverses). AA ). A) > φ HomT (S. . 2 Now consider the pullback square: S .. This correspondence defines a functor F : (T /B) /(f : A → B) → T /A as well as a functor G : T /A → (T /B) /(f : A → B).4.. A) p∗ 2 ∨ > HomT (1. Now for each h such that this diagram commutes.. 2.> . corresponds bijectively to a unique morphism t : S → 1 ×AA T A (over 1 if you like).... Conversely. since T /B and T /A are toposes.. We conclude that every morphism t : S × A → T over A. the arrow g is determined. every such g defines an h. THE ALGEBRA OF SUBOBJECTS ϕ(h ◦ t) = ϕ(h∗ ◦ t) = (hA )∗ ◦ ϕ(t) = hA ◦ t . Hence. ∼ φ HomT (A. Now suppose B is not necessarily equal to 1. which are inverses of each other (i. every t corresponds to a morphism t : S → 1 ×AA T A . Now we can regard f −1 as a functor f −1 : (T /B)/1 T /B → T /A (T /B) /(f : A → B).

Then (πf m)h = (πf m)k iff −1 (h)) = m(f −1 (k)) iff m(f h = k. Then ∀p (A) = ∅ (this follows from direct calculation. ∅ ∩ ∩ ∨ ∨ p > X ×Y X. ∀f −)). We examine a few properties of this adjoint functor... 0)} . Let f : A → B be a map between objects in a topos. 1)}) = {0} (again this follows from HomSub(X×Y ) (p−1 −. or in a more elegant way.. Let p : X × Y → X be the projection on the first co¨rdinate. And actually such a morphism does not have to exist.. then πf m is monic.4. 1} and Y = {0.. Then ∀p (A ∪ B) = ∀p ({(0. −) HomSub(X) (−. (0. ∀f −)).... Let X = {0. 0)}. it must preserve the terminal object of T /A.3. 56 .4. Consider once more the topos Set. Hence πf preserves monics. Take A = {(0... > ∨ A A B A Hence πf restrict to the ‘monic objects’. ∀p A ∩ ∩ ∨ ∨ p > X ×Y X. THE ALGEBRA OF SUBOBJECTS Corollary 3. 0)} ⊆ X × Y .. called ∀f .. The induced map f −1 : Sub(B) → Sub(A) has a right adjoint as map of partially ordered sets. For an example. it is enough to show that if m : C → A is monic..4. To see this.. by making use of the correspondence HomSub(X×Y ) (p−1 −. Suppose m is a monic in T /A. notice that since πf is a right adjoint. ∀f does not preserve ∪. ∃ AND ∀ CHAPTER 3. • Unlike ∃f . Then: o • There is not necessarily a pullback diagram of the form > A .. 1}. −) HomSub(X) (−.. This restriction is by definition ∀f . take A = {(0. Proof: In order to prove this theorem. 1)} ⊆ X × Y and B = {(0. Then there is no morphism making the following diagram a pullback: > {(0.. So if m is a monic. which is IdA : A → A. it is preserved under the following operations: C m ∨ C ⇒ m > A IdA⇒ ∨ πf C πf m > B IdB⇒ πf C πf m ∨ m > πf m . But ∀p (A) ∪ ∀p (B) = ∅ ∪ ∅ = ∅. 0)... Indeed the construction of ∀p does not give the dotted morphism.

For objects A and B in a topos and a morphism f : A → B. ∃ AND ∀ CHAPTER 3. hence a category. For S ∈ Sub(A). namely the structure of a Heyting algebra. it must preserve limits. 57 . Let A be an object in a topos. ∀f does preserve ∩. Definition 3.4. whose product is ∩.4. We will use these definitions to introduce an internal logical language in a topos. Furthermore we have a pullback functor f −1 and a notion of ∃f and ∀f . Then we can regard Sub(A) as a preorder. Then ∀p (A ⇒ B) = {0} = {0.3. the product. hence in particular. Let A and B be objects in a topos and let f : A → B be a morphism from A to B. Take A = ∅ and B{(1. • Unlike ∃f . A ⇒ B = B c ∪ A. 0)}.5. define: ∃f χS := χ∃f S ∀f χS := χ∀f S Now we have shown the algebra of subobjects as a logical structure. we would like to know how the characteristic morphisms of S ∈ Sub(A) and ∃f S ∈ Sub(B) as well as ∀f S ∈ Sub(B) depend on each other. THE ALGEBRA OF SUBOBJECTS • ∀f does not preserve ⇒. Since ∀f has a left adjoint (namely p−1 ). In Set. 1} = ∀p (A) ⇒ ∀p (B).

Let f : X → Y be a morphism and let τ be a term of type X. Let σ and τ be terms of type X. In any topos T . The target of the interpretation of a term is called the type of the term.Chapter 4 The Mitchell-B´nabou language e “Such is the advantage of a well constructed language that its simplified notation often becomes the source of profound theories. τ ◦ p2 ) : U × V → X × Y . Y . We define a term in a topos T to be an expression that can be obtained by the following inductive steps: 1. where both A and B are objects of a topos T and f : A → B is a morphism. Definition 4. the term is called a formula.e.1 The language In this chapter we introduce the language formulated by W. The basic definitions and theorems in this chapter can be found in [11] and [1]. it turns out that a whole logical language can be built within a topos. 2. In this chapter we use X. these formal expressions have an interpretation. see definition 4. a formula. Then the formal expression f τ is a term of type Y . This notation will remind one more of variables. Its interpretation is the morphism f ◦ τ . limits. which is a morphism in T .e. This can be done by using terms: expressions of variables. ⇒ on objects of the form Sub(A) (externally) or ΩA (internally). 3.1. For an object X. Mitchell and J. et cetera to denote objects.1. 4. Then (σ. Indeed. where A is an object in some topos T .1. subobjects and cartesian closedness) where chosen. If the interpretation of a term is a morphism which target is Ω.” . the character x is a term of type X. i. B´nabou. Now we would like to build logical formulae from it. In this chapter we see how well the three defining properties (i. Its interpretation is the identity IdX . Then σ = τ is a term of type Ω. Its interpretation is the morphism 58 . In the previous chapters we used the characters A and B to specify objects. X1 . Recall that our e perspective so far has been that a topos is a generalization of the category of sets. Its interpretation is the morphism (σ ◦ p1 .Pierre-Simon de Laplace We have defined a logical structure using ∩.1. interpreted by σ : U → X and τ : V → X respectively. X2 . τ ) is a term of type X × Y . 4. ∪. Let σ be a term of type X with interpretation σ : U → X and let τ be a term of type Y with interpretation τ : V → Y . logical connectives and quantifiers that are defined inductively. as well as quantifiers ∀f and ∃f relating Sub(A) and Sub(B).

This yields a term σ ∈ τ of type Ω. 59 Ω. ZX . Here δX is the characteristic morphism of the diagonal morphism ∆X : X → X × X.ψp2 ) > Ω×Ω → > Ω. σ and τ may have different sources. λxσ : U > 8. interpreted by σ : U → X. Let θ be a term of type Y X .τ ◦p2 ) > X ×X δX > Ω.Ω. . 5. Note that the equality sign ‘=’ is not an equality of morphisms. interpreted by YX ×X e θ(σ) : V × U > > Y. Let φ be a term of type Ω with interpretation φ : U × X → Ω. Let x be a variable of type X and σ be a term of type Z with interpretation σ : X × U → Z. Let σ be a term of type X with interpretation σ : U → X and let τ be a term of type ΩX with interpretation τ : V → ΩX .ψp2 ) > ∧ φ ∧ ψ :U × V Ω×Ω > Ω. THE MITCHELL-BENABOU LANGUAGE σ = τ :U × V (σ◦p1 . 9. 7. Ω. 6. interpreted by θ : V → Y X and let σ be a term of type X.ψp2 ) > Ω×Ω ∨ > Ω. φ ⇒ ψ :U × V (φp1 . interpreted by X × ΩX e σ ∈τ :U ×V > > . For example. Then the following expressions and their interpretations define terms of type Ω. Then the following expressions and their interpretation define terms of type Ω: ∃p φ > ∀p φ > ∃xφ : U ∀xφ : U Here p : U × X → U is the projection on U . φ ∨ ψ :U × V (φp1 . This gives a term θ(σ) of type Y . (φp1 . This yields a term λxσ of type Z X whose interpretation is the transpose of σ.4. Ω) corresponding to IdΩX ∈ HomT (ΩX . ΩX ) by cartesian closedness. Here e is the morphism in HomT (X × ΩX . THE LANGUAGE ´ CHAPTER 4.1. Let φ and ψ be terms of type Ω with interpretation φ : U → Ω and ψ : V → Ω respectively.

if it factors through t : 1 → Ω. variables and functions having a priori nothing to do with morphisms within a topos. This definition can be extended in an obvious way to formulae of multiple variables. These formal expressions are just strings of logical signs. In the next part of the thesis expressions that are obtained using inductive steps (5). This chapter uses a lot of 60 . × Xn → Ω is a formula. we will give a few examples which demonstrate the power of the MitchellB´nabou language. If φ : X1 × . . SOME EXAMPLES ´ CHAPTER 4. xn ) to explicitly indicate its source. > φ(x) Ω. For our convenience we introduce a different. We will not distinguish between these definitions.2. many of which can be found in [1] and [11]. it is instructive to see some examples of the use of this language.2. Before building any further machinery.3. .2 Some examples At this point the language built. Definition 4.1. (6) and (7) will not occur. . 4. but the details of this language distract too much from the scope of this thesis. Our main example will be to prove that for a morphism f in a topos the formula e ∀x∀x f x = f x ⇒ x = x ‘is true’ if and only if f is monic.1. and is called the interpretation of the logical language. After briefly playing with this language I will refer to literature. We now state when a formula is true. Definition 4.1. The Mithcell-B´nabou language has e a wide amount of properties. THE MITCHELL-BENABOU LANGUAGE It is important to remark is that the definition of a term is a formal expression defined by the inductive steps in above definition. but equivalent notion of ‘true’. . Some proofs in the literature make use of all of these steps. . If φ(x) : X → Ω is a formula we will write {x | φ(x)} for the corresponding subobject of X. then we write φ(x1 . These steps are quoted for the completeness of the definition. written |= φ(x). Corollary 4. Nevertheless. Regarding them as morphisms in a topos can be done though (by above definition). A formula is a term of type Ω.4. We say a formula φ(x) : X → Ω is true if and only if the following square is a pullback: ! X IdX ∨ > 1 t ∨ X Proof: The proof is straightforward. A formula φ(x) : X → Ω is called true. . 1 ! ∨ > t X > φ(x) Ω.4. in the previous chapters is very abstract.

Consider: (x. it will give a feeling for the theory built up in the next chapter.2. and remark that f was defined as the equalizer of φ(x) and φ(x) ∩ ψ(x). Remark that ‘monic’ has the intuitive meaning of ‘injective’. Many of the logical formulae which are true intuitively are indeed true in the topos logic. we find f is the identity. Hence we conclude that φ(x) → ψ(x) is true. in the Mitchell-B´nabou language this expression has e a far more complicated meaning. x )|∀x∀x f x = f x ⇒ x = x ∨ > 1 true1 =(∀x∀x f x=f x ⇒x=x ) ∨ 1 t > Ω. Therefore it may be surprising that this formula is true if and only if f is monic. Let f : X → Y be a morphism. We will prove this using example 4. Actually. Furthermore. However.4. Example 4. Example 4. the logical language will produce quite useful results. The following is statement holds: ∀x∀x f x = f x ⇒ x = x is true if and only if f is monic. Then we can show that {x|φ(x)} ⊆ {x|ψ(x)} if and only if φ(x) → ψ(x) is true.1. The formula ∀x∀x f x = f x ⇒ x = x is intuitively the definition of injectivity. Now if φ(x) → ψ(x) is true. In this example both x and x will be variables of type X. SOME EXAMPLES ´ CHAPTER 4. x )|∀x f x = f x ⇒ x = x > X ! ∨ (x.2. THE MITCHELL-BENABOU LANGUAGE diagram chasing to tackle its problems. But since in a topos we are dealing with morphisms rather than functions. Let φ(x) : X → Ω and ψ(x) : X → Ω be formulae. Hence we must show that φ(x) ∩ ψ(x) = φ(x) if and only if |= (φ(x) → ψ(x)).2.ψ(x)) h > ! ≤ ∩ > 1 t ∨ e X Ω×Ω ∧ p1 ∨ ∨ → > Ω Ω. the equalizer of φ(x) and φ(x) ∩ ψ(x) apparently is IdX : X → X. 61 . This will be useful to understand the definitions of the language. Hence φ(x) ∩ ψ(x) = φ(x). x )|f x = f x ⇒ x = x > X ×X p1 ∨ (x.1. there is no proper meaning of ‘injective’.2. If φ(x) ∩ ψ(x) = φ(x). First suppose the formula is true.2. Now look at the diagram: {x|φ(x) → ψ(x)} f ∨ ∨ > (φ(x). Remark that: {x|φ(x)} ⊆ {x|ψ(x)} iff {x|φ(x)} ∩ {x|ψ(x)} = {x|φ(x)} iff φ(x) ∩ ψ(x) = φ(x) (equality as morphisms). This shows the power of the language. In the topos Set the morphisms are functions and one can easily prove monic is equivalent to injective.

Since ∀x and ∀x are right adjoints. . . x )|∀x f x = f x ⇒ x = x } . SOME EXAMPLES ´ CHAPTER 4. {(x. x )|x = x }. x )|∀x∀x f x = f x ⇒ x = x } = 1.2. Suppose f is monic. Hence we conclude {(x. Hence it is enough to check if f x = f x ⇒ x = x is true. In the same way we find: (x. Hence g = h and f is monic. ∀x {(x.4. x )|f x = f x ⇒ x = x = X × X. =X Hence we conclude that the formula f x = f x ⇒ x = x is true. THE MITCHELL-BENABOU LANGUAGE The bottom square is a pullback. ∨ (x.2) that {(x. . x )|∀x f x = f x ⇒ x = x }) = 1 HomSub(X) (X. {(x.2. This means (by example 4. . x )|∀x f x = f x ⇒ x = x }). We would like to prove f is monic. h) factors through ∆Y . By adjunction we know: HomSub(1) (1. . so does HomSub(X) (X. The two lower squares are pullbacks. With the inclusion {(x. Consider: f ! X ∆X ∨ > Y ∆Y ∨ > 1 t ∨ X ×X > (f. Then (f p1 . . they preserve the terminal object. we must have (x. . x )|f x = f x }. x )|f x = f x } ⊆ {(x. {(x. 1) contains exactly one morphism. . {(x. x )|∀x f x = f x ⇒ x = x . . Now consider: ! A . so suppose f g = f h. x )|∀x f x = f x ⇒ x = x }) = HomSub(X) (p−1 1. h) factors through {(x. f p2 ) ◦ (g. x )|∀x f x = f x ⇒ x = x }) HomSub(X) (!−1 ∀x {(x. The converse is also true. Since the upper square is a pullback.h) > X ∆X ∨ > 1 ∨ > > X ×X ∨ δX > Ω (f p1 . h) factors through ∆X . x )|∀x f x = f x ⇒ x = x } . 1) = HomSub(1) (∀x {(x. Since X is the terminal object in Sub(X). x )|∀x f x = f x ⇒ x = x }) Since HomSub(1) (1.f p2 ) Y ! ∨ ∆Y > Y ×Y δY ∨ > 1 Ω .f ) Y ×Y 62 δY > Ω. x )|f x = f x ∨ (g. x )|f x = f x } ⊆ {(x. (g. x )|x = x } we find (g.

The next example is concerned with unique factorization. y)|mx = f y} Now look at X ×Y Z. By diagram chasing the left square is a pullback. y)|mx = f y} ∩ > Z ∆ ∨ > 1 t ∨ Id ∨ e < < < ∨ p2 Y X ×Y > (mp1 . Example 4. Suppose |= ∀y∃x mx = f y.4. Like in example 4. Of course. This can be expressed by the following diagram: ! X (IdX . the equivalent statement is true for epis.2. Suppose furthermore ∀y∃x mx = f y. Then it is clear that f factors uniquely through m.IdY p2 ) Y ×Y δY > Ω. Clearly we could view ‘epi’ as a generalization of ‘surjective’ in the topos Set.2. The middle square is a pullback (by definition) and by a simple diagram chase it turns out that {(x. x )|f x = f x } = {(x. it suffices to prove that ∃xf x = y is true if and only if f is epi. 63 . Then ∀y∃xf x = y is true if and only if f is epi.f ) ∨ p2 > ∆Y Y ∨ > 1 t ∨ Y < < X ×Y > (f p1 . it follows that |= ∃x mx = f y.f p2 ) Z ×Z δ > Ω.4. me < {y|∃x mx = f y} < {(x. The PBL makes the whole rectangle is a pullback. y)|mx = f y} is such that the following diagram is an equalizer: e > mp1 > > f p2 {(x. This can be generalized to an arbitrary topos.2. First. Let m : X → Z be a monic morphism and f : Y → Z a morphism. so by the universal property of the subobject classifier one has δY ◦ (f. Let x be a variable of type X and y a variable of type Y . f is epi if and only if the left diagonal morphism is epi if and only if ∃xf x = y is true. SOME EXAMPLES ´ CHAPTER 4. Switching from characteristic morphisms to subobjects this equation becomes {(x. Since ∀x and ∀x are right adjoints.2. we make the meaning of the formula |= ∀y∃x mx = f y explicit.2.2. Example 4.1 makes f x = f x ⇒ x = x true.2. f ) = δX . Now clearly.2. Then there is a unique morphism g such that mg = f . Suppose we have functions m : X → Z and f : Y → Z in Set such that m is injective. Like in example 4. So example 4. THE MITCHELL-BENABOU LANGUAGE The right square is a pullback by definition. Let f be a morphism between objects X and Y .3. they preserve the terminal object and hence ∀x∀x f x = f x ⇒ x = x is true. x )|x = x }.

THE MITCHELL-BENABOU LANGUAGE E e < < < ∨ p2 Y X ×Y f p1 > X m ∨ !X > > 1 t ∨ > Z χm > Ω..gp2 ) > Z ∆Z ∨ > 1 ∨ Z ×Z δZ > Ω. χm f = t!Y . Since p2 e is epi. y)|f x = f y} ∨ > U .. y)|f x = f y} such that the diagram commutes. k) such that f h = gk.. Example 4.. Then (f p.k) > (f p1 . this g is unique.... gq) ◦ (h. y)|f x = f y} ∨ > X f ∨ Y g > Z.. X × Y (h... y) ∈ X × Y such that f x = gy. In previous chapter we gave the expression f x = gy an interpretation in an arbitrary topos. Note that this is exactly the pullback of Set. in Set one can write the limit of a diagram as a subobject of the product of the objects in that diagram.2. That is.2. We give two examples.. we show that limits in a topos are generalizations of the limits in Set. Remark that χm f p2 e = χm mp1 e = t!X p1 e = t!Y p2 e. SOME EXAMPLES ´ CHAPTER 4.... This proves that {(x... Hence there is a unique morphism g : Y → X such that the diagram above commutes (g is not indicated in the diagram). Since subobjects and characteristic morphisms are generalized from a set to a topos it turns out that limits can be defined in exact the same way. Finally.4. The subobject is implied by the morphisms in the diagram and can be expressed by a characteristic morphism. .. which is: {(x.5. That is. where !Y is the unique morphism from Y to 1. y)|f x = f y} is the pullback of f along g. This morphism makes the following diagram commute: X ∧ g f ⊂ m > > Z Y Since m is monic. 64 . hence there is a unique morphism from U to {(x. k) factors through ∆Z . pulling back f along g yields the set of pairs (x. The following diagram is a pullback: {(x.. pullbacks and equalizers. Suppose there is an object U and a morhpism (h.. The left square is a pullback..

3.3. The proof is similar as in example 4. |= ψ(x) ⇒ φ(x) ∨ ψ(x). {x|φ(x)} ⊆ {x|φ(x) ∨ ψ(x)} iff {x|φ(x)} ⊆ {x|φ(x)} ∪ {x|ψ(x)} . SOME FURTHER EXAMPLES ´ CHAPTER 4. proving that one of the formulae above is true amounts to proving an inclusion: 1. Let φ(x) and ψ(x) be formulae. Then {x|φ(x)} ⊆ {x|ψ(x)} if and only if |= φ(x) ⇒ ψ(x). In this section we introduce theorems which enable you to solve problems using logical rules. Let φ(x) : X → Ω and ψ(x) : X → Ω be formulae.3 Some further examples The aim of this section is to establish some techniques to handle the formulae that are introduced in the beginning of this chapter.3. THE MITCHELL-BENABOU LANGUAGE Example 4.3.1. Theorem 4. Proof: According to proposition 4. Also. Proof: see example 4. 4. In the previous section many examples were solved by diagram chasing.1. For convenience we quote it again. Proposition 4. |= φ(x) ∧ ψ(x) ⇒ φ(x). We start with a proposition which has already been proved in example 4. it will give a good logical tool to prove deeper results of the Mitchell-B´nabou language.6. 2. {x|f x = gx} is the equalizer of the diagram: f X g > > Y. We restrict ourselves to some elegant examples.2. 2. |= φ(x) ∧ ψ(x) ⇒ ψ(x). |= φ(x) ⇒ φ(x) ∨ ψ(x).1.3. This will lead to much shorter and more elegant solutions to many problems. {x|φ(x) ∧ ψ(x)} ⊆ {x|ψ(x)} iff {x|φ(x)} ∩ {x|ψ(x)} ⊆ {x|ψ(x)} . The aim of this chapter is not to give a general e overview of true formulae.2.2. {x|φ(x) ∧ ψ(x)} ⊆ {x|φ(x)} iff {x|φ(x)} ∩ {x|ψ(x)} ⊆ {x|φ(x)} .2.1. Then the following formulae are true: 1.4. 4.5. Let’s start with some simple formulae. 65 . 3.2.

Then: 1. . . . Xn φ(x1 .×Xn ∨ > Ω. xm ). . . . then |= φ(x1 .. . xn ). .. xn ) factors through 1. m > n. . .e. . . xn+1 . . .. . . i. . SOME FURTHER EXAMPLES ´ CHAPTER 4. . Lemma 4. . . . . × Xm where p is the obvious projection. . so does φ(x1 . If φ(x1 . × Xm p > X1 × . . there is a way of ‘extending its source’.. xn ) = t◦!X1 ×.. xn ) is a formula with variables x1 . Each time.. THE MITCHELL-BENABOU LANGUAGE 4. × Xm Now... . .. Proof: 1. . . {x|ψ(x)} ⊆ {x|φ(x) ∨ ψ(x)} iff {x|ψ(x)} ⊆ {x|φ(x)} ∪ {x|ψ(x)} . . . . xm ) :=X1 × .. Let φ(x1 . xn .3. . . 66 ... If |= φ(x1 . .xn ) > Ω. . .×Xn . .. If the projection X1 × . 2. × Xm is epi and |= φ(x1 .. . . .xn ) > φ(x1 . . This is done by the following diagram: p > φ(x1 . . .×Xm X1 × .. . . .. .×Xm = t◦!X1 ×. m > n be its extension. then |= φ(x1 . p > X1 × .×Xn ◦ p.xn ) φ(x1 .. . . . . xn .. . .. . xn . . xn ) ◦ p. .4.. . . . Let φ(x1 . xm ). xn ). . the last inclusion is obviously true. .. ..3. If p is epi. . . X1 × Xn Ω. . xn ) ◦ p = t◦!X1 ×. turning it into a formula of more variables: φ(x1 . xm ). . . . . The statement is obvious from the following diagram: > 1 ∨ X1 × . . . . × Xn φ(x1 . × t !X1 ×. .3. . We have: >1 > !X1 ×. 2. . . we find: φ(x1 . xm ). . If φ(x1 . . xn ) be a formula with variables x1 . × Xn → X1 × .

. × x | φ(x . . x1 . In a topos with Modus Ponens. . x1 . ¯ ¯ Proof: First. .3 gives |= ψ. Call the extensions φ and ψ. xn ) ⇒ φ(x. . xm )|φ(x1 . . Xn . . . xn )} ⊆ {x × x1 × . xn ) be its extension to variables X.3. x1 .4. . . . . Xn and let φ(x. xn )} ⊆ {x × x1 × . ¯ Hence |= ψ. . . 3. . (Modus Ponens). . Theorem 4. xn ))(x. . . . x1 . x1 . . . . . . . . . . . Now.4. . . . . . since it amounts to proving an inclusion of subobjects. . xm )|ψ(x1 . x1 . . . . × Xn is epi. then |= φ(f y. . since it actually is an equality.3 gives |= φ ¯ ¯ (x1 . . . . xm )|ψ(x1 . . . 2. xm ) = 1. . . . . . . . . . × xn | φ(x. . x1 . . x1 . . xn )) iff iff {x1 × . xn ) be a formula with variables X1 . X1 . x1 . . |= (∀xφ(x. . Suppose φ and ψ are formulae. . . lemma 4. xn ). . . . . . xn .3. x )} p 1 n 1 n 1 n 1 n The last inclusion holds. . xn )} ⊆ {x1 × . . 2. × x | ∀xφ(x. Remark ¯ and |= φ ⇒ ψ. . xn ) ⇒ φ(x. . . Let φ(x1 . Suppose the projection X1 × . xn ))(x. × Xn → X1 × Xn be the obvious projection. Let f : Y → X be a morphism. . . xm ) = 1 and ¯ ¯ ¯ that lemma 4. . x1 .3. . . SOME FURTHER EXAMPLES ´ CHAPTER 4. . × xn | (∀xφ(x. . If |= φ and |= φ ⇒ ψ then |= ψ. . . . . . If |= φ(x. . xn )} −1 {x × . x1 . . . Let X1 . . Xn be the variables of ψ and let X1 . . . Theorem 4. . xn ) ⇒ (∀xφ(x. which is equivalent to (x1 . xn ) ⇒ (∀xφ(x. . . extend both φ and ψ to the variables X1 . . . xn ))(x. . x1 . . Proof: 1. . . x1 . . then |= ψ’. . . x1 . . where the part to the left of ⇒ should be regarded as the extension of the formula ∀xφ(x. . xn ). x1 . . |= φ(x1 . . . . xm ) . . × xn | φ(x. x1 .3. Let p : X × X1 × . . . x . xn )} iff {x × x1 × . . . . . . x1 . . . . . . . . . |= φ(x1 . . . . . . Xm . . . . . × xn | (∀xφ(x. . . . . xn ) to the variables x. . Xm be the variables of φ and ψ. . . × Xm → X1 × . |= (∀xφ(x. . . xn )} 67 . THE MITCHELL-BENABOU LANGUAGE The following theorem is a very powerful one.5. x1 . . . It proves that the Modus Ponens is often true in topos logic. . . . xm ) ⊆ (x1 . . xn ))(x. it then is enough to check whether the formula φ ⇒ ψ is true. . . . . For example the topoi Set and SetC satisfy these conditions. We refer to topoi in which the Modus Ponens is true as topoi with Modus Ponens. xn ) iff {x × x1 × . . . xn )). . × xn | ∀xφ(x. . . . × xn | φ(x1 . . xm )|φ(x1 . . . . Thanks to the Modus Ponens the following theorem becomes useful. . x )} ⊆ {x × . . Suppose we want to prove a statement of the form ‘If |= φ. . . . x1 . we conclude ¯ (x1 . . . From this. . xn ). . x1 . In general this statement is easier to prove. . For many topoi the conditions for the theorem are easily checked. . . Then: 1. .3. . . . . .

... Example 4. >X U α > 1 ∩ ∨ > φ(x) Ω... Now let us examine example 4. So |= f hy = f gy ⇒ hy = gy by applying theorem 4... ... ...Y ...3. Then. .. Theorem 4.. . . .. Since we will not use the details of these articles..... . then |= f hx = f gx according to theorem 4.×Xn . From |= ∀x∀x f x = f x ⇒ x = x we obtain |= f x = f x ⇒ x = x according to theorem 4.. The following proposition will be of great use too: Proposition 4.4. . We write: U |= φ(α) if there exists a morphism α such that the following diagram commutes. in a topos with Modus Ponens f = g (equality as morphisms!) if and only if: |= ∀xf x = gx Proof: This is almost immediate from the universal property of a pullback: > > ... . Definition 4. PARTIAL TRUTH ´ CHAPTER 4. xn ) : X1 ..3 (2)..... In particular the Kripke-Joyal semantics form an important tool for problems involving partial truth. . . The converse is probably best done in previous section.4 Partial truth In this section we briefly describe the notion of partial truth..3...5 (2).... ˜ {(x1 . . In a category with Modus Ponens. ∨ .. since in the next chapter we will refer to some articles which make use of this construction. First the only if part. The proof is exactly the same as in lemma 4...4.. . The existence of the right dotted morphism is equivalent to |= f x = gx (as a formula).1..3... ...3. we just give the definition.. α ˜ .3.. with p replaced by f × IdX1 ×. . The existence of the left dotted morphism is equivalent to f = g (equality as morphisms). × Xn a morphism. ∩ . .g) δY OY The left dotted morphism exists such that the diagram commutes if and only if the right dotted morphism exists such that the diagram commutes.. Some important theorems about partial truth can be found in [11].. .. ..3..3.∆Y t .. . .5 (3) twice. ...2 again... Modus Ponens gives |= hx = gx and hence h = g. 4.. xn )} > . ... |= ∀x∀x f x = f x ⇒ x = x if and only if f is monic..6.. . 1 .. THE MITCHELL-BENABOU LANGUAGE 3. xn )|φ(x1 .. 68 ....... . . Let U be an object of the topos and α : U → X1 × . Suppose f h = f g..6.. . . ∨ ∨ .> .....5 now gives f = g if and only if |= ∀xf x = gx.. >Y ×Y > Ω..4... Let f : X → Y and g : X → Y be morphisms.7.... X (f. Let φ(x1 . Xn → Ω be a formula in a topos.2.

. together with morphisms: 0: 1 → G. Such e an object G is called an internal group. . . . U |= φ. INTERNAL CATEGORIES ´ CHAPTER 4.. + : B × B → B. one can define internal fiels. More general. . This definition is quite abstract and we rather give specific definitions of internal lattices. |= ∀a∃b (a + b) = 0 . THE MITCHELL-BENABOU LANGUAGE Sometimes we abbreviate this as U |= φ.5 Internal categories The Mitchell-B´nabou language forms an important tool for generalizing objects that are defined as a e set with a certain structure. but if one puts a U with a morphism in front of it.5. Then the pullback property of the (right) square in the diagram below tells us there is a (dotted) morphism α for every α... xn ) : X1 . .... |= ∀a (a + 0) = 0 . For deeper results about partial truth. ˜ {(x1 ... such that these morphisms satisfy the above properties written in the Mitchell-B´nabou language. satisfying the properties: |= ∀a∀b∀c (a + b) + c = a + (b + c) . ... α ˜ .. . . . Since these properties. a group is a set togheter with an addition + and a special element 0. we introduced the rational numbers object Q and like one would expect.. . >X U α >1 >∩ ∨ > φ(x) Ω... ∩ > . see for example [5] or [11]. . internal lattices. 69 .. Xn → Ω be a formula in a topos. . Then |= φ if and only if U |= φ(α) for all objects U and morphisms α. In chapter 1.. i. 4. have a (generalized) interpretation e in any topos.2. It is enough to require that a group is an object G within a topos T . Conversely. expressed in the Mitchell-B´nabou-language... ... et cetera.. .. et cetera. Let φ(x1 .... A formula φ might not be true for all U . there is a notion of an internal category whose the definition can be found in [8].. there is no need to require that a group needs to be a set with some additional structure. internal locales... suppose U |= φ(α) for every U and α then certainly X1 × . The following theorem shows the link to the previous section. Hence φ factors through 1. . In an analogous way. For example.. .. . × Xn |= φ(IdX1 ×. one can proof that this is an internal field. it becomes ‘true on U ’. xn )|φ(x1 . xn )} . ∨ ..e. .×Xn ).. .4.4. Proof: Suppose that |= φ. Theorem 4. .

Also the definition of a locale may be generalized to an object called an internal locale of T . Gelfand and M. This paragraph is included to show that the constructions that follow are well defined. This locale is called the spectrum of A and is denoted as Σ(A). since A is fixed from the start of the construction. i. this duality becomes one between unital commutative C*-algebras and compact completely regular locales. If we consider the locale formed by the underlying topology of a compact Hausdorff space instead of the space itself.3 should be considered an intermezzo: this paragraph only considers lattices in Set and is heavily based on [2] and [3].e.” .G. Landsman and Spitters [7]. The authors start with a unital C*-algebra A in Set and construct a particular topos T (A) from this C*-algebra. In this chapter we write out the details of the construction of the spectrum of the C*-algebra A and show that every step is well-defined. Shaw In this chapter we turn our attention to a generalization of the Gelfand theorem to topos theory. called an internal C*-algebra of T .1) and then turn our attention to the more general case of generalized distributive lattices freely generated by an object subject to relations in a topos. called A. This is done by introducing a generalization of a C*-algebra to the topos T .2. ¯ they define an internal C*-algebra of T (A).2. it gives a nice representation of the lattices that we are interested in. we define the topos T (A) and the internal C*-algebra A. 70 . In section 5. or briefly Σ. In section 5. 2. an object of T which has all the properties of a C*-algebra (we state the precise definition later in this chapter). The construction consists of the following steps: ¯ 1. Paragraph 5.1. in Set. Mulvey and Banaschewski showed that there is a duality between the internal C*-algebras and the internal compact completely regular locales of any topos T . The original theorem proved by I.Chapter 5 The construction of the spectrum “The mathematician is fascinated with the marvelous beauty of the forms he constructs. called L∼sa .2 we introduce some theory about lattices in topoi and construct an internal lattice in a topos. see [12]. We first consider a generalization of a free distributive lattice to topos theory ¯ A (paragraph 5. In the next two chapters we try to compute the ¯ compact completely regular locale corresponding to A by the morphism cCStar → KRegLoc. Furthermore. In this thesis we study an article by Heunen. [13] and [14]. Naimark shows that there is a duality between unital commutative C*-algebras and compact Hausdorff spaces (in Set). Furthermore. and in their beauty he finds everlasting truth.M. cCStar ↔ KRegLoc By the work of Mulvey and Banaschewski this duality may be generalized to any topos T .B.

Finally. too. C). ¯ A is defined on morphisms (in the preorder C(A)) as the inclusion of C*-subalgebras.1. THE CONSTRUCTION OF THE SPECTRUM L∼ ¯ A L∼ ¯ A 3. We will spend some words on C. where C(A) is the categorical preorder (5. it is defined on objects as: ¯ A : B → B. We will not cover this in full detail. sa →Ω sa and define the spectrum Σ(A) During the construction we will introduce some theorems that allow us to compute the spectrum. it may be generalized to a topos that 71 .e.5. q) ∈ N if and only if a < q. unital C*-subalgebra of dimension 1: C · 1. Cn and C(X). The next step is to introduce a special object in this topos. commutative C*-subalgebras. In what follows A will always be our unital C*-algebra as a set. but only elaborate a bit. where one requires (a. The topos we are interested in will be the following one. which is called the tautological functor . as we will see. given by the intersection of unital. The reason we emphasize that this construction has a set as input is that we are going to construct (i. which is defined as a set with the usual structure of a Banach Space over C with unit and involution called ∗ satisfying a∗ a = a 2 .1). (5. ¯ Definition 5. it turns out to be an internal C*-algebra in the topos T (A). It does not have a join. with X a compact Hausdorff space.2. we sometimes avoid some technicalities of this e language by making use of elementary category theory or lattice theory.1) Here the unit in B is supposed to be inherited from A. since the join of two commutative C*-subalgebras does not have to be commutative.1. THE TOPOS T (A) CHAPTER 5.5. which is the unique commutative. Since this concept is set theoretic in nature. but a generalization of a set: an object in a certain topos. Definition 5.1.e. in section 5. However. We introduce the internal C*-algebra in an analogous way to the internal categories of section 4.1 The topos T (A) The construction of the spectrum starts with a unital C*-algebra A.1. I. The partial order and the bottom element of C(A) will turn T (A) into a topos with nice properties. What is so interesting about this tautological object? First.3 we introduce a morphism A : Ω L∼ ¯ as a certain subobject of Ω Asa . 5. Now we introduce C(A) as the following preorder: C(A) = {B | B is a unital commutative C*-subalgebra of A} . In the previous chapters we proved that T (A) is a topos. The Mitchell-B´nabou language will be of great use. An important remark is that C(A) has a bottom element. The only problem is the analytic part of the definition of a C*-algebra: the norm and its completeness. define) a C*-algebra that is not a set anymore. The partial order of C(A) is given by the inclusion. Let A : C(A) → Set denote the functor that sends a unital. commutative C*-subalgebra B ⊆ A to itself. unless specified otherwise. The actual computations are done in chapter 6. Let T (A) denote the topos SetC(A) . where our main examples will be the spectra of the C*-algebras C2 and M (2. This partial order has a meet. A norm of a C*-algebra A in Set can be regarded as a subset N of A × Q+ . the 2 × 2-matrices.

Furthermore. An internal C*-algebra in a topos T is an object B of the topos.3 for the (pointwise) addition. Now a norm on A is a subobject of A × Q+ satisfying certain axioms.1. . 72 .1. We state the result as a theorem. . . 1 C×B B×B B×B B → → → → → B B B B B . . Then define Q+ = q | ∃r (q = r2 ) ⊆ Q. in which it is complete. . multiplication and scalar multiplication. . we state the definition of an internal C*-algebra. We say the C*-algebra B is unital if there is a morphism 1 : 1 → B such that |= ∀a ((1 · a = a) ∧ (a · 1 = a)). where ¯ ¯ it is proved that A is complete using the Kripke-Joyal semantics.1.4. . together with morphisms 0: λ: · : +: ∗: such that: |= ∀a∀b∀c (a + b) + c = a + (b + c) |= ∀a∀b a+b = b+a |= ∀a 0+a = a |= ∀a∃b a+b = 0 |= ∀a∀µ∀ν λ(µ. a) |= ∀a∀b∀µ λ(µ. THE TOPOS T (A) CHAPTER 5. (scalar multiplication).3. . Now.1. Definition 5.1. A is an internal commutative unital C*-algebra of T (A). λ(ν. (multiplication). For the completeness part we refer to [7]. It is easy to see that A is commutative and has a unit (since we have chosen all algebras of C(A) to be commutative and unital). a + b) = λ(µ. . THE CONSTRUCTION OF THE SPECTRUM has a rational numbers object Q. . . . a) + λ(µ. . Recall that we defined the rational numbers object in chapter 1. ¯ It is very easy to verify that A as in definition 5.5. a)) = λ(µν. ¯ Theorem 5. we call the C*-algebra B commutative if |= ∀a∀b a · b = b · a. a)∗ = λ(¯.2 satisfies the axioms of definition 5. . These axioms can be found in [14] as well as the axioms for an internal C*-algebra to be complete. Proof: See [7]. B has a norm N ⊆ B × Q+ . a∗ ) µ Moreover. b) |= ∀a∀b∀c (a · b) · c = a · (b · c) |= ∀a∀b∀c a · (b + c) = (a · b) + (a · c) |= ∀a a∗∗ = a |= ∀a∀b (a + b)∗ = a∗ + b∗ |= ∀a∀b (ab)∗ = b∗ a∗ |= ∀a∀µ λ(µ.

THE CONSTRUCTION OF THE SPECTRUM 5.> . Furthermore. First. commutativity. L.e..5. 5.. Definition 5. The definition is analogous to the one of an internal Heyting algebra and the one of an internal C*-algebra.. An internal distrivutive lattice in a topos T is an object L of the topos together with morphisms ∨: L×L → ∧: L×L → : 1 → ⊥: 1 → L.2.. the e unit law.. Now.2.. An internal distributive lattice LS is called freely generated by S if there is a morphism i : S → LS such that for every internal distributive lattice M and morphism f : S → M there is a unique morphism g : LS → M such that the following diagram commutes: f S i ∨ > M LS and such that g commutes with the internal lattice structure of LS and M . In Set these relations reduce to the familiar definition of a distributive lattice. we formally state the definition of a free distributive lattice in a topos. Recall that a lattice is always assumed to have a top and a bottom element.. the following diagrams commute: LS × LS g×g ∨ ∨ > .2. L. FREE LATTICES CHAPTER 5.2 Free lattices The next step of constructing the spectrum is to calculate the free distributive lattice (with top and ¯ bottom) generated by A subject to certain relations which we will introduce later. we study the ¯ free distributive lattice generated by A without subjecting it to any relation since this object turns out to be quite interesting. and let S be an object in that topos. M > > ⊥ 1 1 73 .. Let T be a topos. g LS g ∨ LS × LS g×g ∨ ∧ > LS g ∨ M ×M LS ∧ ⊥ g > > ∨ > M M ×M LS ∧ ∧ g > M M .... . it satisfies the absorption laws. . .1.. satisfying the obvious expressions in the Mitchell-B´nabou language for associativity.. idempotence and distributivity for both ∨ and ∧. i. L. .1 Construction of free distributive lattices For a free distributive lattice in a topos we need the notion of an internal distrivutive lattice.

. We claim the lattice obtained by the following procedure is the distributive lattice freely generated by S. commutativity : φ ∧ ψ = ψ ∧ φ associativity : φ ∧ (ψ ∧ ρ) = (φ ∧ ψ) ∧ ρ unit law : φ∧ =φ idempotence : φ∧φ=φ commutativity : φ ∨ ψ = ψ ∨ φ associativity : φ ∨ (ψ ∨ ρ) = (φ ∨ ψ) ∨ ρ unit law : φ∨⊥=φ idempotence : φ∨φ=φ absorbtion : φ ∨ (φ ∧ ψ) = φ φ ∧ (φ ∨ ψ) = φ φ ∨ (ψ ∧ ρ) = (φ ∨ ψ) ∧ (φ ∨ ρ) distributivity : Define LS = L/ ∼l . so are the formal expressions φ ∨ ψ and φ ∧ ψ. after we subject the lattice to relations.. THE CONSTRUCTION OF THE SPECTRUM The choice of the notation i for the map between the object and the free distributive lattice it generates may remind the reader of an inclusion. .. ψ) → φ ∨ ψ where the last vee is defined as above. ∈ L (as formal symbols). where i is the obvious inclusion. This is done in [15]. we will still speak about the inclusion in the free distributive lattice. Now divide out the following equational laws in L. It is important to remark that one has to prove that this defines a distributive lattice. As an example we calculate the free distributive lattice generated by a set S in the topos Set. . the morphism is monic..2.5... where the ∼l stands for the equivalence relation generated by above relations. However.. Construct a set L inductively: 1.. .. ψ) → φ ∧ ψ where the last wedge is defined as above join ∨ : LS × LS → LS : (φ.. Now if we have a diagram in Set like: S i ∨ f > M LS . The internal lattice structure is given by: bottom element ⊥ top element meet ∧ : LS × LS → LS : (φ..e.. For convenience of notation we do not add extra notation to elements of LS to indicate that its elements are actually equivalence classes. . since this is a convenient way of indicating that a variable has to be regarded as an element of the free distributive lattice. FREE LATTICES CHAPTER 5. In many cases it will be an ‘inclusion’. then we can define g inductively by: 74 . If φ and ψ are elements of L. 2. Nevertheless.. S ⊆ L. ⊥. g . i will not be monic anymore. i.> .

2. . Theorem 5. . . First we must define the intrinsic maps fC. It is obvious that f = gi.5. where f is a morphism in C from C to D. sn ) → φ(g(s1 ). 75 . joins. This problem is known as the ‘word problem’ (for free lattices) and has been solved. g(sn )). C. . Recall that this notation is defined as fC.. .2. By definition g is indeed a morphism of lattices (i. . LS(D) . In Set the distributive lattice freely generated by a set S is defined as the object LS above. .D is functorial. FREE LATTICES CHAPTER 5.. . . . LS (C) = LS(C) and the internal meets and joins are defined by the local meets and joins. define g : φ(s1 .. Define g : s → f (s) s ∈ S. If φ(s1 . sn ∈ S and possibly ⊥ or appear.2. THE CONSTRUCTION OF THE SPECTRUM 1. ) ∈ LS is an expression in which the elements s1 . namely point-wise (sometimes called locally). g is unique in having this property.D = (g ◦ f )LS . and ⊥). Proof: LS LS 1. . . The hard part is proving that the function g is well defined. but it is of great use to compute free lattices generated by an object in SetC . C. We restate the result: Theorem 5. .E LS gD.D preserves the meet and join in Set: S(C) iS(C) ∨ S fC.D = LS (f ).e.E LS Hence the assignment (f : C → D) → fC. Since LS (D) = LS(D) is a distributive lattice in Set by definition of LS(D) and LS (C) = LS(C) is the free distributive lattice generated by S(C). This relates to the problem of determining which of the formal expressions that are defined above.D > S(D) iS(D) ∨ > LS(C) . Indeed one can prove that g is well defined. .D . ⊥. for any object C ∈ C. Let C be a category. g: → . In the topos SetC the free distributive lattice generated by an object S can be computed point-wise..C LS ◦ fC. 2... there is a LS LS unique map fC.3.D as in the diagram below such that fC. Furthermore. are equivalent. The theorem may not be surprising. The following theorem shows that a free lattice in this category can easily be computed. Remark that by uniqueness of this map and some diagram chasing the following equations hold.2.. . That is. . g : ⊥ → ⊥. IdLS = IdLS (C) . since there is no free choice for the initial step 1 and the inductive step 2 of the definition of g in order to let g respect the lattice structure of LS and M . sn . it preserves meets..

.. For if there exists such a morphism g. LS M g(D)kC.2) Both the left hand side and right hand side are maps from LS (C) to M (D)..... ∧ ... For every C there exists at most one such g(C) since M (C) is a distributive lattice in Set and LS (C) is the distributive lattice freely generated by S(C) in Set.. g ..... we need to check if it commutes with the intrinsic maps of LS and M ..... ∧ . FREE LATTICES CHAPTER 5......4) 76 ......e. we actually defined the maps fC. Note that M (C) is a distributive lattice in Set...... > > M (D)..... i . In order to let this map g be well-defined.. since the local meets and joins inherited from M equip M (C) with a distributive lattice structure.. THE CONSTRUCTION OF THE SPECTRUM LS 2...3) g(D) and the right hand side is the unique map arising from LS (C) .... ..D = kC............ ..5.. . .. The uniqueness is the easy part......... For the existence of g...... for any morphism k : C → D in C the following equation should hold..... . .. S(C) > LS (C) L > LS (D) S kC....... the diagram must at least commute locally.e. . We have to prove there exists a unique g such that this diagram commutes....D (5.. > > M (C) > M (D)..... since it does commute locally. i...... ... M (C) ... If this is a well-defined map it automatically commutes with the meet and join in LA and M ..... S(C) > LS (C) M g(C) kC...D .. LS (C) . ..D g(C) f (C) > (5... The meet and join in LS commute with the intrinsic maps fC.. S(C) commutes..> .. i.. g(C) ∨ ... .... note that we can define a global map by combining the unique local maps which arise from the local diagrams in Set. f S i ∨ > M LS ....> . ... Since this meet and join are LS defined locally in LS this claim becomes a tautology...D (5.. .. The left hand side is the unique map arising from LS (C) ... ... ..2. Now suppose M is an internal lattice in SetC . ... 3....D in such a way they commute with the meet and join.

2. if there is another internal distributive lattice M satisfying (5) .2. The free distribuitive lattice generated by Asa subject to relations (5) . one has LS g(D)kC. ¯ Definition 5. Now the relations we need in order to define the spectrum are the following terms of the Mitchell-B´nabou-language.(9)together with a morphism: ¯ ¯ Asa i∼ > L∼sa . THE CONSTRUCTION OF THE SPECTRUM For x ∈ S(C). This turns out to be an object. The uniqueness of the dotted arrow then results in (5. ¯ In this paragraph we are going to define the distributive lattice freely generated by the object Asa subject to relations. meanwhile have acquired some intuition about this.2 Relations and the lattice L∼sa ¯ A ¯ For the construction of the spectrum we first have to take the self-adjoint part of our object A.D g(C)i(C)(x). ¯ A Furthermore.D (x) S (x) f (D)kC.3) and (5.4) are equal. That is. 5. A B → Bsa . 77 . which is ¯ defined as the following subobject of A: ¯ Asa = {a|a∗ = a} . So the composition of the horizontal arrows in (5.1 for its interpretation).2.4. together with a morphism. the last two relations are equivalent to: |= ∀a∀b Da+b ∨ (Da ∨ Db ) = Da ∨ Db (8) |= ∀a∀b Dab ∨ ((Da ∧ Db ) ∨ (D−a ∧ D−b )) = (Da ∧ Db ) ∨ (D−a ∧ D−b ) (9).(9) together with a map ¯ f : Asa → M . ¯ It follows immediately from the fact that the adjoint ∗ is defined locally.2). called L∼sa . Suppose a is a variable of type A (see definition 4.1.D f (C)(x) M kC. e |= D1 |= ∀a Da ∧ D−a |= ∀a D−a2 |= ∀a∀b Da+b |= ∀a∀b Dab = = = ≤ ≤ (5) ⊥ (6) ⊥ (7) Da ∨ Db (8) (Da ∧ Db ) ∨ (D−a ∧ D−b ).5.(9) is an ∼ internal distributive lattice LAsa that satisfies (5) .D i(C)(x) = = = = S g(D)i(D)kC. then f factors uniquely through i∼ by a morphism of lattices. that A is the local self-adjoint ¯sa is given on objects as: part of every C*-subalgebra B of A. FREE LATTICES CHAPTER 5.D M kC. then we ¯ A write Da for the term i∼ a. called ¯ A ¯ ¯ i∼ : Asa → L∼sa . How is this lattice defined? The reader may. (9) By definition of ≤.

. Then L∼ . which generating object and which topos define the morphism i∼ . .. .. For example. specifying which relations......2.. Then find the smallest equivalence relation ∼ on LS such that LS / ∼ is a lattice when it inherits the lattice structure of LS and such that LS / ∼ satisfies the relations (5) . .. For any such object we adopt the notation Da to indicate the term i∼ a..2..5. THE CONSTRUCTION OF THE SPECTRUM Remarks and notation: Of course this definition can be generalized to distributive lattices generated by an arbitrary object subject to an arbitrary finite number of relations in an arbitrary topos. . relation (7) is equivalent to the following diagram being commutative: ¯ Asa ! ∨ −. there exists a morphism LS → M which respects the lattice structure of LS and M . . Now note that by theorem 4. ¯ All the other relations can be expressed in this way. if two elements of LS are equivalent by ∼. The proof is similar to theorem 5. . . For our convenience we will only subject the free distributive lattice in the proof below to relation (7). too.. The next theorem shows that the free distributive ¯ lattice generated by Asa subject to relations (5) . we are going to use both the interpretation in Set and its generalization to T (A) a lot.(9) can be computed point-wise. The proof for subjecting the lattice to all the relations is analogous to the one below. The reader can easily work out the details on a huge amount of scratch paper.... Theorem 5.>. Theorem 5. Hence there is a unique lattice homomorphism (LS / ∼) → M making the diagram below commutative: S ∨ > M LS ∨ L∼ S ..2. Proof: First.3.. . We first prove that in Set the distributive lattice freely generated by a set S subject to (5) . FREE LATTICES CHAPTER 5. . construct LS . Let S be an object of SetC and let α : S → S be a morphism... Especially.. Then.5. but involves some extra diagram chasing.2 > ¯ Asa i ∨ 1 > ⊥ LAsa .2.> .3.....(9).. the free lattice generated by S.(9)..... By definition of the equivalence relation. .. the distributive S lattice generated freely by S subject to relation 78 ... Suppose there is another lattice M subject to the given relations.(9) can be expressed as equalities of morphisms in any topos.(9) exists. In Set the lattice freely generated by a set S subject to relations (5) ...6 relations (5) .6..(9) always exists.. then their images under LS → M are equal. There is no special reason why this relations is chosen. . by definition of the free lattice...... Here we study the relations (5) .

..2. L∼ (D) > S ⊥(C) ⊥(D) i∼ (D) < 1(C) S(C) > < ! > > > 1(D) > fC. L∼ (C) = L∼ S S(C) and the internal meets and joins are defined by the local meets and joins..5 proves that the local distributive lattices generated by the local sets subject to relation (5.D is a lattice homomorphism by construction. By uniqueness and an extensive diagram chase we find: S IdC.. THE CONSTRUCTION OF THE SPECTRUM Dα(s) = ⊥. L∼ L∼ 3.10) can be computed locally..2.D is functorial... Note that this commutativity is L∼ S exactly the condition stating that there exists a unique morphism fC......E ◦ fC.. ∨ . for any object C ∈ C... fC..... S 2. We know that the non-dotted arrows form a commutative diagram. L∼ L∼ S Hence the assignment (f : C → D) → fC..... S i∼ ∨ L∼ S f > M S ! ∨ α > S f..E . . We need three steps: S 1....2..... That is...D as indicated in the diagram.. theorem 5..D !S S(D) < ! (α)(C) (α)(D) S(C) S fC....D L∼ S(C) i(C) ∨ L∼ (C) S < i∼ (C) > S fC.. . Proof: The proof is similar to theorem 5...... ..10) exist......5. Furthermore.. ....... First we must define the internal maps fC.. (5.C L∼ = IdL∼ (C) .D > S(D) .D > S(D) i(D) ∨ .> . FREE LATTICES CHAPTER 5... Now suppose M is an internal lattice in SetC and f : A → M is a morphism such that the right diagram below commutes.D = (g ◦ f )C...3.. g 1 79 ⊥ > M .. Notice that α will play the role of the function s → −s2 . S L∼ S S S gD..

.... ∨ .D .. ∧ . .....11).... The left-hand side S is the unique map arising from α(C) L∼ (C) . ∧ .e.... L∼ (5. For every C there exists at most one such g(C).13) ⊥ > M (D) For x ∈ S(C)...D (x) S f (D)kC......5...... If there exists such a morphism g. i∼ ... i.. A(C) S . the diagrams must at least commute locally. ∨ ∨ > M (C) LS (C) 1(C) ⊥(C) commute..... ..D i∼ (C)(x) = = = = L∼ S g(D)i∼ (D)kC. since M (C) is a lattice subject to relation (5.12) g(D) ⊥ > M (D) and the right-hand side is the unique map arising from α(C) L∼ (C) .D i ..D > S(C) ∨ S g(D)kC...... f (C) α(C) > M (C) S(C) > S(C) S(C) ....10) in Set and L∼ (C) is the lattice freely generated by S(C) subjected to relation (5. The uniqueness of the dotted arrow then results in (5.. i∼ !(C) . i...D g(C)..e. .....> .... ∨ .......D > A(C) ∨ M kC..D = kC. !(C) . note that we can define a global map by combining the local maps obtained by the local diagrams in Set. ∼ > > M (C) > M (D) 1(C) A(C) ∼ LS (C) M i g(C) kC.............11) Both the left-hand side and the right-hand side are maps from L∼ (C) to M (D). ....D g(C)i∼ (C)(x)....10) S in Set.D f (C)(x) M kC.... First we prove uniqueness........> . 80 ...... L∼ (5...... S(C) S ... FREE LATTICES CHAPTER 5.. . for any morphism k : C → D in C the following equation should hold: M S g(D)kC..13) are equal.g(C) ∨ ...12) and (5.. For the existence of g... THE CONSTRUCTION OF THE SPECTRUM We have to prove that there exists a unique g such that the left diagram above commutes..D (x) M kC.. (5....D g(C)i . We only need to prove that this map commutes with the intrinsic L∼ S maps fC. f (C) . > > S > M (D) 1(C) S(C) ∼ L∼ (C) L∼> L∼ (D) S i S kC.... ! i∼ (C) .... Hence the compositions of the horizontal arrows in the left diagrams of (5... one has S g(D)kC..2.

Let A = C(X) be a commutative unital C*-algebra. Suppose a. FREE LATTICES CHAPTER 5. where X is a compact Hausdorff space. Hence. Da ≤ Db .2. THE CONSTRUCTION OF THE SPECTRUM We derive a short corollary from the theorem above. Let A = C(X) be a commutative unital C*-algebra. Conversely: Da+ = Da∨0 ≤ Da ∨ D0 (by relation (8)) = Da (by relation (7)). We will use the Gelfand theorem to infer that if A is a commutative unital C*-algebra. Proof: Lemma 5. For any C*-algebra A. Da = Da+ and D−a = D−a− . Let A = C(X) be a commutative unital C*-algebra. b ∈ Asa and a ≤ b (i.2. Lemma 5. Lemma 5. in this paragraph.2.8 gives Da ≤ Da+ . Proof: = D1 = Dn 1 n ≤ D1 n (by relation (5)) (by applying relation (8) n times) n Now relation (9) gives Dna ≤ Da and Da = D 1 na ≤ Dna .2. Db ∈ L∼sa . ¯ Corollary 5. b ∈ C(X)sa define (a ∨ b) ∈ C(X)sa and (a ∧ b) ∈ C(X)sa to be the point-wise maximum and minimum of a and b. In this situation. from which the lemma follows.(9) in Set. subject to relations (5) .3 Intermezzo: representing the local lattice The two theorems in this paragraph will be quite important for for what follows. we write Da for i∼ a : Asa → L∼sa .5. (5) . the inclusion of the variable a of type Asa into the distributive lattice freely generated A by Asa .e. 81 . In this paragraph all statements are taken within Set. Asa is the space of real valued continuous functions on X.9.2.2.2.10.7. Let a+ = a ∨ 0 and a− = a ∧ 0. Then for Da . respectively. For a ∈ Asa and n ∈ N∗ . The theorems below are based on [2] and [3]. the free distributive lattice generated by A subject to relations ∼ . We first introduce a few lemmas that give us tools to compute the distributive lattice freely generated by a commutative unital C*-algebra subject to (5) . For a ∈ Asa . then it is of the form C(X). Note that a = a+ + a− . from which D−a− = D(−a)+ = D−a .8. A Proof: Da = ≤ = = D(a−b)+b Da−b ∨ Db (by relation (8)) ⊥ ∨ Db (by relation (7)) Db For a. This lattice is denoted as LAsa ¯ 5.(9) exists. We obtain Da = Da+ . one has Da = Dna .(9). Lemma 5. ∀x ∈ X a(x) ≤ b(x)).

i.8. the following inequality follows: Da∨b ≤ = = = (D(b−a)+ ∨ (Db ∨ Da )) ∧ (D(a−b)+ ∨ (Db ∨ Da )) (D(b−a)+ ∧ D(a−b)+ ) ∨ (Db ∨ Da ) ∨ ((Db ∨ Da ) ∧ D(a−b)+ ) ∨ ((Db ∨ Da ) ∧ D(b−a)+ ) (Db−a ∧ Da−b ) ∨ (Db ∨ Da ) Db ∨ Da .e.2. This is true. The property Da ∨ Db ≤ Da∨b follows straight from lemma 5. THE CONSTRUCTION OF THE SPECTRUM Theorem 5. 3.2. Then the map of sets i : Asa → L∼sa is surjective. We have proved the theorem if we can show that: 1.5. For the other inequality. 4.2. Da ∨ Db is equal to Dc for some c ∈ Asa . the second one follows from lemma 5. ⊥ is equal to Da for some a ∈ Asa . by relation (8) 82 . This is true. Now.2.9 and the last one follows from relation (6).7 tells us L∼sa is equal to the free lattice generated by Asa A modulo some equivalence relation. So: Da ∧ Db ≤ = = = = (Da∧b ∨ D(a−b)+ ) ∧ (Da∧b ∨ D−(a−b)− ) Da∧b ∨ (D(a−b)+ ∧ D−(a−b)− ) ∨ (Da∧b ∧ D(a−b)+ ) ∨ (Da∧b ∧ D−(a−b)− ) Da∧b ∨ (D(a−b) ∧ D−(a−b) ) Da∧b ∨ ⊥ Da∧b . is equal to Da for some a ∈ Asa . We prove Dc = Da∨b . Da∨b ≤ D(b−a)+ ∨ (Db ∨ Da ). FREE LATTICES CHAPTER 5.8.11. we note Da∨b = D(a−b)+ +b ≤ D(a−b)+ ∨ Db (by relation (8)) ≤ D(a−b)+ ∨ (Db ∨ Da ).2.2. Da ∧ Db ≥ Da∧b follows straight from lemma 5. First notice that the proof of theorem 5. since ⊥ = D−1 by relation (7). every element of the lattice L∼sa has a representative of the form Da for some A A a ∈ Asa . Let A = C(X) be a commutative unital C*-algebra. . Da ∧ Db is equivalent to Dc for some c ∈ Asa . We prove that Dc = Da∧b . For the other inequality: Da = D(a∧b)+(a−b)+ ≤ Da∧b ∨ D(a−b)+ Db ≤ Da∧b ∨ D−(a−b)− . since = D1 by relation (5). 2. The first equality follows from distributivity.

we will always specify A the value of a and replace ‘a’ in the notation by its value. THE CONSTRUCTION OF THE SPECTRUM So all elements in LAsa that are obtained by the inductive steps in the proof of theorem 5.2.2 are equivalent to Da for some a ∈ Asa . et cetera. 3. The theorem above tells us that every element of L∼sa is of the form Da for some a ∈ Asa .2. Theorem 5. So. Now clearly at least the equivalence relation above must at least be divided out.12. this notation will not lead to confusion. Futhermore. FREE LATTICES CHAPTER 5. Suppose L is equipped with the structure of the partial order a + + defined by: b if and only if ∃n ∈ N∗ : a+ ≤ nb+ . logical connectives and quantifiers. So if we A want to compute L∼sa . Since in Set it is our practise to denote Da for both the meanings as stated in 2 and 3. we will always write D1 .2.6 in ¯ A the topos T (A).(9) in Asa instead A of dividing out these relations in LAsa .2. we will emphasize that this term is interpreted in T (A). since if ∃n ∈ N∗ : a+ ≤ nb+ and ∃m ∈ N∗ : b+ ≤ ma+ then Da = Db . Then L is a lattice isomorphic to L∼sa . for example. Let L be the set Asa modulo the following equivalence relation: a ∼+ b if and only if ∃n ∈ N∗ : a+ ≤ nb+ and ∃m ∈ N∗ : b+ ≤ ma+ .(9) in Asa with its partial order on functions. The interpretations of the terms of the Mitchell-B´nabou e language in Set are exactly the conventional set-theoretic meanings of the expressions in terms of variables. When this is meant.6 in A the topos Set.2. We will emphasize that this term has to be interpreted in Set. We write Da for the equivalence class of L∼sa that contains a. Adopting the notation from meaning number 2 for meaning number 1 will not lead to confusion as long as we specify in which topos we are working. We write Da for the term i∼ a if i∼ : Asa → L∼sa is defined by the morphism of definition 5. this justifies to denote all the elements of the lattice L∼sa as Da for a fixed element a ∈ Asa . A Notation: recapulating the discussion above. We write Da for the term i∼ a if i∼ : Asa → L∼sa is defined by the morphism of definition 5. which tells exactly which equivalence relation we must divide out in order to compute the distributive lattice generated by Asa subject to (5) . This notation is not as ambiguous as it seems. 83 .5. In this case. Let A = C(X) be a commutative unital C*-algebra. note that meaning number 1 is a generalization of meaning number 2. Returning to the computations in Set we prove one last theorem. D−1 .2. This proves the theorem.11 and the discussion after its proof show that the free lattice can be computed by dividing out the equivalence relation generated by (5) . we can start with the set Asa and divide out relations (5) . 2. the notation Da can have three meanings: ¯ 1. This makes the lattice far easier to compute. Also.(9) in Set. A Proof: Theorem 5.

1 The map A0 In the following construction we define a map A0 : Sub(L∼sa ) → Sub(L∼sa ) of subobjects of L∼sa in the ¯ ¯ ¯ A A A topos T (A). To define the completions we first need to know something about joins of finite subsets of a lattice. We start with a short lemma in Set. 5.2. Hence + is the partial order on L that is inherited from the partial order on functions on Asa . In this section we introduce a morphism A : ΩLAsa → ΩLAsa A L∼ ¯ of T (A) and define the spectrum Σ as a subobject of Ω Asa by means of this morphism A. By means of section 5. We use the character U to indicate a subobject of L∼sa or more generally something that ‘looks ¯ A ∼ like’ or ‘corresponds to’ a subfunctor of LAsa .(9) in T (A). Note that e V0 always exists: it is the join of all the elements of V0 . We define V0 to be the supremum of the set V0 .(9). 84 . THE SPECTRUM CHAPTER 5. Here the variables x and y are of type L. Db ∈ L∼sa such that C Da = Db .3 The spectrum Thusfar. Hence L∼sa (C) for ¯ ¯ A A an object C ∈ C(A) is a set.2 we defined the distributive lattice freely generated∼ by ∼ ¯ ¯ ¯sa subject to relations (5) .(9). and hence L must be the free lattice. The expression above has an interpretation in the MitchellB´nabou language of the topos Set. ¯ commutative unital C*-subalgebras of A. i. Note that by this definition ∅ is defined to be the bottom element ⊥ of L. Lemma 5. Now that we now what this lattice looks like. the construction of the spectrum.1. THE CONSTRUCTION OF THE SPECTRUM by lemma 5. Suppose L is an internal lattice in Set and suppose V0 is a finite subset of L.3.5. Then for every q ∈ Q there exists an r ∈ Q such that Da−q ≤ Db−r . 5. we started with a commutative unital C*-algebra A in Set. This is done by defining a map A0 (C) : P(L∼sa (C)) → P(L∼sa (C)) of subsets of L∼sa (C) ¯ ¯ ¯ A A A for all objects C ∈ C(A) and combining them together into a global map (here P denotes the power set).2. which coincides with its familiar meaning in set theory.3. Thus V0 is the element z ∈ L such that |= (∀x x ∈ V0 ⇒ x ≤ z) ∧ (∀x(∀y y ∈ V0 ⇒ y ≤ x) ⇒ z ≤ x). then a + b. We defined the topos T (A) and ¯ the internal C*-algebra A. which is finite by definition.e. The lattice L = Asa / ∼+ satisfies (5) . The theorem above gives a very nice representation of the free lattice generated by Asa . Hence ∼+ is the smallest equivalence relation ∼ such that Asa / ∼ is a lattice satisfying (5) .3. The maps A0 (C) are called the (local) completions.10. In this section V . Recall that L∼sa was an object of T (A). we proceed to the next step. One easily verifies that + is a well-defined partial order and if a ≤ b. V0 or any object that includes the character ‘V ’ in its notation is supposed to be a set. Suppose C is a unital commutative C*-algebra in Set. The symbols C and D will always be objects in C(A). Let Da .9 and lemma 5.

Conversely. we have ⊆A0 U =⊆C. proof of theorem 5.D |A0 U (C) (V0 ). THE CONSTRUCTION OF THE SPECTRUM Proof: Choose an n ∈ N∗ such that a ≤ nb. and since L∼ ¯ L∼ ¯ L∼ ¯ L∼ ¯ applying Asa to a finite set gives the join of all the elements of that set. Now if Db ∈ V0 then there exists a finite subset V0. we define locally what is called an ideal ¯ A generated by U (C) with respect to the covering by: A0 (C)U (C) = Da ∈ L∼sa (C) | Da ¯ A where Da U (C) is defined as: V0 . Now since ⊆C. THE SPECTRUM CHAPTER 5. by ¯ ¯ A A L∼ ¯ (A0 U )(C) := A0 (C)U (C). Since A0 U is a subfunctor of L∼sa . This amounts to checking whether ⊆C.b . then Db U (C). n Hence r = q n works. since every element of A0 (C)U (C) is covered by itself.2. The inclusion ⊇ is obvious.3.14) U (C) .3. the completions A0 (C)U (C) glue together to a functor called A0 U .D Asa defined. For every q ∈ Q+ there exists a finite subset V0 of U (C) such that Da−q ≤ We say U (C) covers Da . Lemma 5. Hence A0 (C) is well-defined. Proof: A0 is idempotent if and only if A0 (C) is idempotent for every C ∈ C(A). which is an object of T (A). A0 is idempotent.D A Asa Asa ⊆C.12. So we must show that if U is a subfunctor of L∼sa . So we have successfully defined our map A0 : Sub(L∼sa ) → Sub(L∼sa ).D |A0 U (C) (Da ) is L∼ ¯ Asa covered by ⊆C.D is a lattice homomorphism (cf. we must check if the intrinsic maps ⊆A0 U are wellC. Then we must show that for every q ∈ Q+ there exists a finite subset V0 of U (C) such that Da−q ≤ V0 . By the assumption Da ∈ A0 (C)A0 (C)U (C) there exists a finite subset V0 of A0 (C)U (C) such that Da− q ≤ V0 .b of U (C) 2 such that Db− q ≤ V0.D |A0 U (C) and we must check whether ¯ C. Then Da−q ≤ Dnb−q = Dn(b− q ) n = Db− q . 85 . Furthermore. we find that ⊆C.5. where V0 is a finite subset of U (C) covering Da . where the double supremum on the right hand side of the equality is the supremum of a finite set. The following lemma will be useful for later computations.14. which can be done by theorem 5. Now we have 2 Da−q ≤ Db− q | Db ∈ V0 2 ≤ b V0.3). To see this. It follows from lemma 5. (5.2. Now if C is an object of C(A) and U is a subfunctor of L∼sa .D |A0 U (C) (Da ) is in (A0 U )(D) = A0 (D)U (D).1 that if Da = Db and Da U (C).2. then ¯ A A0 (C)A0 (C)U (C) = A0 (C)U (C).D |A0 U (C) (Da ) Asa satisfies property 5. suppose Da ∈ A0 (C)A0 (C)U (C).b .3.

on F E. is short for ‘maximal cosieve’.D C. We define a natural transformation σ ∈ HomSetP (y(C) × A. let C. Then all intrinsic morphisms of ΩA are surjective maps of sets. Now. Ω).D ) and τ ∈ HomSetP (y(D) × A.F if D ⊆ E or C ⊆ E if C ⊆ E and D ⊆ E. The last equation follows from the fact that ⊆∗ (E) : y(D)(E) → y(C)(E) is the identity Id{1} . We prove that (⊆∗ ×IdA )∗ σ = τ . (τ (F ))({1} × (⊆A )(a)) = m.D C. It is not too hard to check that σ : y(C) × A → Ω is indeed a natural transformation. Before we Lemma 5. where m. Then again ((⊆∗ ×IdA )∗ σ)(E) = (σ ◦ (⊆∗ ×IdA ))(E) C. THE CONSTRUCTION OF THE SPECTRUM 5. The last equation follows from the fact that ⊆∗ (E) : y(D)(E) → y(C)(E) is the identity Id∅ .D 86 .D = τ (E).D = τ (E). −). Notice that y(C)(E) = and remark that y(C)(E) × A(E) = A(E) if C ⊆ E.D = σ(E) ◦ (⊆∗ ×IdA )(E) C.D 2. Proof: In this proof. Now suppose C ⊆ D (denote this as a morphism by ⊆C.D = σ(E) ◦ (⊆∗ ×IdA )(E) C. Recall that ΩA was defined as ΩA (C) = HomSetP (y(C) × A.D 1. THE SPECTRUM CHAPTER 5.3.c. C. suppose C ⊆ E. First. E and F be objects of the preorder P. D. Ω). Ω) by σ(E) = τ (E) σ(E) : A(E) → Ω(E) a → F |D ⊆ F. ∅ if C ⊆ E.3.3.c.5.2 The morphism A L∼ L∼ ¯ A sa ¯ Now we will define a morphism A : Ω Asa → Ω proceed we introduce an important lemma. ∅ if C ⊆ E. suppose D ⊆ E. {1} if C ⊆ E. which is the internal version of A0 . Then ((⊆∗ ×IdA )∗ σ)(E) = (σ ◦ (⊆∗ ×IdA ))(E) C. C. We separate three cases: C. . Let P be a categorical preorder and let A be an object of SetP .3. where we have written y(C) for the covariant Yoneda-functor HomP (C.

A is idempotent.D = (A(D))(⊆Ω sa )(y) ⊥. Suppose in lemma 5.3 that P has a bottom element ⊥.3. C.C = (A(C))(A(C))(x).3. The lemma follows from the Ω following computation.D Hence A is well-defined.D ⊥.C ⊥. hence ⊆Ω = (⊆∗ ×IdA )∗ is surjective. If x ∈ ΩA (C).5.C (A(C))(x) = (A(C))(⊆Ω (y)) = (⊆B )(A(⊥))(y).C = (⊆Ω sa )(A(⊥))(A(⊥))y ⊥.4.D L∼ ¯ A L∼ ¯ A L∼ ¯ A = (A(D))(⊆Ω sa )(⊆Ω sa )(y) C. Suppose x =⊆Ω sa y.3. ⊥. 87 L∼ ¯ A L∼ ¯ A .D C. Hence it must equal τ (E). We have shown that (⊆∗ ×IdA )∗ σ = τ .C = (⊆Ω sa )(A(⊥))(y) ⊥. which in turn ¯ ¯ A A corresponds to a morphism χU : 1 → Ω ˆ to a morphism L∼ ¯ A sa by cartesian closedness. then L∼ ¯ A L∼ ¯ A L∼ ¯ A (A(C))(x) = (⊆Ω sa )(A(⊥))y ⊥. we must define A(C) : ΩA (C) → B(C). any morphism A : ΩA → B is completely determined by the morphism A(⊥) : ΩA (⊥) → B(⊥).15) Now observe that a subfunctor U of L∼sa corresponds to a morphism χU : L∼sa → Ω.D ⊥. THE SPECTRUM CHAPTER 5.D C. Then. We next prove the counterpart of lemma 5.3. Proof: Suppose C ∈ P. Proof: A is idempotent if and only if A(C) is idempotent for all C ∈ C(A).D unique morphism from the initial object to Ω(E). we must check if A commutes with the intrinsic maps of Ω L∼ ¯ A L∼ ¯ A L∼ ¯ A . Hence ⊥.15) in corollary 5. L∼ ¯ A sa Finally.5. Note that ((⊆∗ ×IdA )∗ σ)(E) : ∅ = y(D)(E) × A(E) → Ω(E) is the C. So the completion A0 corresponds L∼ ¯ A (χU : 1 → Ω ˆ L∼ ¯ A L∼ ¯ A sa ) → (χA0 U : 1 → Ω ˆ L∼ ¯ A sa sa ).2 for A. C. then there exists a A y ∈ ΩA (⊥) such that ⊆Ω (y) = x.C sa y. THE CONSTRUCTION OF THE SPECTRUM 3.D Corollary 5. Suppose C ⊆ E and D ⊆ E.C A A (5. for any object B of SetP .C L∼ ¯ A Ω (⊆C. Lemma 5. Suppose x =⊆⊥.4: A:Ω L∼ ¯ A sa →Ω L∼ ¯ A sa .D sa )(A(C))(x) = (⊆Ω sa )(⊆Ω sa )(A(⊥))(y) C.C = (A(D))(⊆Ω sa )(x).3. Since each morphism 1 → Ω sa is determined by 1(⊥) → Ω 1 are surjective). this corresponds to a morphism Ω L∼ ¯ A sa (⊥) (noting that the intrinsic maps of (⊥) → Ω L∼ ¯ A sa (⊥). then: ⊥.3. which in turn induces a unique morphism A by equation (5.

Theorem 5. Hence Σ(C) is the image set of the the ⊥. 2.3. Since C(A) has a bottom element ⊥ the morphism χU : 1 → Ω ˆ the element (χU (⊥)(∗ )) ∈ Σ(⊥) and the computation: ˆ L∼ ¯ A L∼ ¯ A L∼ ¯ A L∼ ¯ A sa by cartesian L∼ ¯ A sa is completely determined by χU (C)(∗) = χU (C)(⊆1 )(∗ ) =⊆Ω sa (χU (⊥)(∗ )).C where ∗ and ∗ denote the unique elements of 1(C) and 1(⊥) respectively. we have the following correspondences: 1.4.5 gives (A(⊥))(A(⊥))U⊥ = (A(⊥))U⊥ .C Proof: Suppose UC ∈ Σ(C) and UC = (⊆Ω sa )(U⊥ ). However. As we said in the beginning of this chapter Mulvey and Banaschewski proved the duality cCStar ↔ KRegLoc. The spectrum Σ of an internal unital commutative C*-algebra A in the topos T (A) is defined as Σ = {U |AU = U } ⊆ Ω L∼ ¯ A sa . ¯ Definition 5. THE CONSTRUCTION OF THE SPECTRUM The first and last equations hold by definition.C UC = (A(C))UC = (⊆Ω sa )(A(⊥))U⊥ ⊥.3. notice that lemma 5. ⊥. By the definition of the correspondence 1→Ω 88 .C = (⊆Σ )(A(⊥))U⊥ . The correspondences between objects in this section may seem very abstract.2 and lemma 5.C For the last equality. ⊥.2 and the correspondence between A0 and A(⊥).7.3. Then ⊥. THE SPECTRUM CHAPTER 5. We conclude this section by giving the definition of the spectrum as well as a proposition which is a consequence of lemma 5. hence (A(⊥))U⊥ ∈ Σ(⊥). the correspondences are actually quite straight forward. The morphism cCStar ← KRegLoc then is Σ → Loc(Σ. Hence morphisms χU : ˆ L∼ L∼ ¯ ¯ Asa correspond one-on-one to elements of Ω Asa (⊥).C intrinsic map ⊆Σ . The maps ⊆Σ : Σ(⊥) → Σ(C) are surjective.5. Then in order to compute the ¯ A spectrum Σ.6.1. as proved in [7]. ˆ ˆ ˆ ⊥.5. C).3.C ⊥. The following theorem allows us to compute the spectrum once we have computed Σ(⊥).3.3. The second equation follows from lemma 5. The construction above starting from the internal C*-algebra A and ending with the spectrum Σ is precisely the functor cCStar → KRegLoc in the topos T (A). ¯ in any topos. 3. Any characteristic morphism χU in turn corresponds to a morphism χU : 1 → Ω ˆ closedness of the topos T (A). Suppose that we have computed the lattice L∼sa and suppose ¯ A we know which subfunctors U of L∼sa have the property that A0 U = U .3. Any subfunctor U of L∼sa corresponds to a characteristic morphism χU : L∼sa → Ω by theorem ¯ ¯ A A 3.

one first computes the distributive lattice freely ¯ generated by the self-adjoint part of an (arbitrary) internal C*-algebra A subject to (5) .6 shows that this can be done locally. one computes the distributive lattices freely generated by Bsa subject to relations (5) .7 shows that we can compute each Σ(C) and hence the functor Σ from Σ(⊥).(9).(9) in Set.4. resulting in the spectrum of our initial C*-algebra A. First.5. [13] and [14]) and reformulated by Coquand and Spitters [4] is quite similar. ¯ 2. The construction of this frame can be done by introducing a general morphism A.(9). 5. i.e.3. 1. 4. By the correspondences stated at the end of paragraph 5. ¯ A 3. one tries to find all unital C*-subalgebras of A. one turns this into a frame that satisfies the relation Da ≤ q∈Q+ Da−q . 89 . FINAL REMARKS CHAPTER 5. one computes the distributive lattice freely generated by Asa subject to relations (5) . Then. with a good understanding of the definition of A0 the spectrum can be computed. the topos ¯ T (A) and the tautological functor A.7 can be applied to find the spectrum Σ. This general construction introduced by Mulvey and Banaschewski (cf. where B is a subalgebra of A.2.3 one then directly finds Σ(⊥). which involves technical definitions of finite objects of a topos and infinite joins. In the next chapter we will introduce a theorem (see 6. Theorem 5. These local freely generated distributive lattices can be combined into the functor L∼sa . we have avoided the direct definition of A in T (A) and chosen to introduce the (less complicated) map A0 first.3. Next. theorem 5. Again. [12].3) that allows us to compute Σ easily from the subfunctors U of L∼sa such that A0 U = U . is proved by a general construction for the morphism cCStar → KRegLoc for any topos T . In the next chapter we will do some computations and fortunately. By the above correspondences we can compute Σ(⊥) from the set of subfunctors U of L∼sa that ¯ A satisfy A0 U = U . one computes the subfunctors U of L∼sa such that A0 U = U .4 Final remarks We conclude this chapter by summarizing the construction of the spectrum of a C*-algebra A and the most important theorems for computations. For brevity. this can be done by finding the subfunctors U such that U (C) is a down-set for every object C ∈ C(A). the subobjects U of L∼sa such that A0 U = U correspond exactly to the elements ¯ A x of Ω L∼ ¯ A sa (⊥) such that A(⊥)x = x. THE CONSTRUCTION OF THE SPECTRUM between A and A0 . Finally. Finally. Then theorem 5. resulting in the preorder C(A).1. In the next chapter we will prove some theorems to compute C(A). ¯ A The duality by Mulvey and Banaschewski cCStar ↔ KRegLoc. In the next chapter we will ¯ A see that if A is finite.

C) is the simplest noncommutative algebra.” . Our main examples will be C2 and M (2. Nevertheless we will spend a few words on C(X). whereas M (2. commutative case: Cn First we study the commutative C*-algebra A = Cn . the spectrum turns out to be hard to compute by hand. ⊥. avoiding the extensive amount of computations for n ≥ 3. The computation of C(A) and hence the corresponding topos T (A). It turns out that we can compute the spectrum Σ of Cn for every n. ¯ 2. We have chosen these examples since C2 is the first C*-algebra whose spectrum is nontrivial. We recapitulate that the spectrum can be computed in three steps: 1. with X a compact Hausdorff space. whenever a C*-algebra has a commutative C*-subalgebra of dimension 3 or higher. via Σ(⊥) as described at the end of section 5. ⊥. “I lost count. “What’s one and one and one and one and one and one and one and one and one and one?” “I don’t know.Chapter 6 Computations of spectra “Can you do addition?” the White Queen asked. Here we will make the structure precise for the special cases n = 1 and n = 2. In the following examples A will be the C*-algebra that we study. However. The computation of the free distributive lattice generated by Asa subject to the relations: |= D1 |= ∀a Da ∧ D−a |= ∀a D−a2 |= ∀a∀b Da+b |= ∀a∀b Dab = = = ≤ ≤ .3. However. Further examples that would be interesting are C(X).” said Alice. it turns out that C has a trivial spectrum. and B(H). the bounded operators on a Hilbert-space. and that Cn is computable. but quite complicated for n ≥ 3.Lewis Carroll In this chapter we compute some examples of Gelfand spectra.1 The finite. for n = 3 things become quite complicated and for n ≥ 4 one would rather use a computer. 90 . 6. Furthermore. C). (5) (6) (7) (8) (9) 3. Da ∨ Db . (Da ∧ Db ) ∨ (D−a ∧ D−b ). The computation of the spectrum Σ.

Let f : R → C be a function and suppose we chose a point rk ∈ {r1 . .1. . rk2 ∈ R. If the partition has only one equivalence class. 0 a2 . r2 . we are finished. .6. 0 . 0 . hi+1 that satisfy  1 0 0 .. Since the vectors (h(rk1 ). Hence f ∈ AΠ . . a function h ∈ B such that h equals f at all points r1 .. Then: 1. there is a linear combination of h and the identity whose values at rk1 and rk2 equal f . .. hi (ri ) hi (ri+1 ) hi+1 (r1 ) hi+1 (r2 ) . • For i = 2 note that B contains at least one function h that has different values at two given representatives rk1 . . . . . – Suppose i + 1 is odd.. Let epi : B → C be the evaluation map at the ith point of X. . COMPUTATIONS OF SPECTRA C*-subalgebra and topos The first question we should answer is what the unital C*-subalgebras of Cn look like. By induction (to i) we show that for any set of i different points of R and for any f ∈ AΠ we can construct a function g ∈ B that equals f at these i different points. by induction. . • For i + 1. . . . . Let Π be a partition of the points of X and let AΠ be the set of functions that are constant on every equivalence class of the partition.. . . h1 (ri+1 ) h2 (ri+1 ) . h(rk2 )) and (Id(rk1 ).3 we will give another proof using the Gelfand isomorphism. hi+1 (ri ) hi+1 (ri+1 ) 91 h1 (r1 ) h2 (r1 ) . .. Given any C*-subalgebra B of A. h1 (ri ) h2 (ri ) . which shortens the case of Cn .. Let Π be the corresponding partition. . Trivial. .. .. we must find the pertinent partition Π. Proof: 1. . . If f ∈ B is a function on X. . ri+1 }. Let R be a set of representatives of Π. Then construct functions h1 . Every C*-subalgebra of A is equal to AΠ for a unique partition of n points called Π. . but requires more technical tools.. . since B contains the identity. . Theorem 6. let f ∈ AΠ . . COMMUTATIVE CASE: CN CHAPTER 6. . . ri+1 ∈ R... then it is constant on equivalence classes of Π by definition of ∼X . . Now define a relation on the points of X by pi ∼X pj if and only if Ker(epi − epj ) = B.         hi (r1 ) hi (r2 ) . a1 1 0 . . Suppose Π is a partition of these points. . . 0 1 . h1 (r2 ) h2 (r2 ) . .1. 2. Recall that we can regard Cn as the C*-algebra of functions from an n-point space with discrete topology to C. . . If we can find a g ∈ B that equals f on R we may conclude AΠ ⊆ B. .   ai  1 . Then there is. 0 0 0 0 . ri+1 except for maybe rk . r2 . 1 0 .         =      0 0 ai+1 0     . take elements r1 . so that B ⊆ AΠ . We prove that the commutative unital C*-subalgebras are the ones that consist of functions that are constant on the equivalence classes of some partition using elementary linear algebra. . . . .1. . AΠ is a C*-subalgebra. . 0 . Conversely. 2.. Id(rk2 )) are independent vectors. r2 .. Let A be the C*-algebra Cn and regard it as the space of functions on the topological space X with n points and discrete topology. THE FINITE. • For i = 1 this is trivial. In section 6. . It is straightforward that this relation is an equivalence relation on X. . .

. Note that if 2 ) =SetC2 . If the determinant is zero. h1 (r1 ) h2 (r1 ) .. . hi+1 (r1 ) hi+1 (r2 ) .12 provides the means to compute the lattice generated by Rm subject to the relations. n = 1.e..6.         . h1 (ri ) h2 (ri ) . . .6 we showed that we may compute the lattice generated by Asa subject to relations (5) m .. . 0 1 . The local lattices are generated by R 5.1. 0 a2 . . hi (r1 ) hi (r2 ) .        =   hi (ri ) hi (ri+1 )    ai 0 hi+1 (ri ) hi+1 (ri+1 ) ai+1 0 h1 (ri+1 ) h2 (ri+1 ) . Let B = Cm . ri+1 is 1 + (a1 · . The lattice generated by Bsa subject to relations (5) .. Since i + 1 is odd.2... .±1. · ai+1 )2 = 2 which leads to the same conclusion.   C(C2 ) =   C2 ∧ >∧ C3<       . . If this is non-zero. 1 0 . . .(9) locally.. We restate the theorem for the special case Cm . T (C) is the category Set. i. · ai and the induction hypothesis is proved by the same reasoning as in the case where i + 1 is odd. . . .. . . . .. .        .{1. ..3}} ∧ A{{3}. the determinant of this matrix is 1 + a1 · .. THE FINITE. . . If n = 2 we have T (C n The lattice ¯ In theorem 5. h1 (r2 ) h2 (r2 ) . . Hence for any g ∈ AΠ there is a linear combination of functions in B such that this linear combination equals g at the i + 1 points. ..  C      C(C3 ) = A{{1}. . note that the determinant of the matrix obtained by evaluating the squares of h1 . ...1. 0 .. . 0 0 0 0 . Then consider  1 0 0 . The theorem above tells us that C(Cn ) is the partial order of partitions of n points.. . . . These unknown values are denoted by a1 . .(9) consists of the elements D(±1. .3}}  <    A{{2}. .      . where Da ≤ Db if and only if a ≤ b. . 0 . COMMUTATIVE CASE: CN CHAPTER 6. Theorem 6. .{2. ai ≤ bi for all coordinates i. · ai+1 ..{1. hi+1 at r1 . . . – Suppose i + 1 is even.±1) . COMPUTATIONS OF SPECTRA so every indexed function h is specified at precisely i points and its value is unknown at only one point. the functions are independent. .. . a1 1 0 . m ≤ n subject to the given relations. ai+1 in the matrix above.   0  1 A simple calculation gives that the determinant of this matrix is 1 + a1 · . . .2}} > C ¯ These diagrams also represent the tautological functor A as an object of T (Cn ) =SetC(C ) . 0 .. . . Theorem . where the order is given by refinements of partitions. Proof: Define a function sign: 92 . .2. .2.. . As special cases we have the Hasse diagrams:   C(C1 ) = C. .

x → 1 if x > 0.. By definition of ∼+ .−1) 93 .12.1)    ...2 geometrically by saying that the lattice Cm produces is the m-dimensional hypercube.−1)      The case n = 3 takes the form: D(1. Hence every element of the lattice generated by Bsa subject to (5) .  D(−1. COMPUTATIONS OF SPECTRA sign : R → R .−1) ∧ D(1.±1.. none of the elements of D(±1.. For a ∈ Bsa .sign(am )) .(9) is of the form D(±1. We will try to visualize L∼sa for the cases n = 1. THE FINITE..±1) is equivalent to any other. The statement about the partial order follows from theorem 5.    D−1 The case n = 2 has the form:       D(−1. theorem 5.. n = 2 and n = 3.1)            D(1..±1) .1..2.1. We could express theorem 6.. x → −1 if x ≤ 0.12 gives Da = Da+ = D(sign(a1 )..2..1)     D(−1. The case n = 1 looks like: ¯ A   D1   . COMMUTATIVE CASE: CN CHAPTER 6.6.

1)           D(−1. The last equation follows from the fact that Da−q = Da for small q. COMMUTATIVE CASE: CN CHAPTER 6.1.1)           D(−1. then the down-set x ↓ of x is defined by x ↓= {y | y ≤ x} .1. every down-set is the completion under A0 of itself.1) D(1.−1)            D(−1. We will do this for n = 1 and n = 2.−1.1.1)      D(−1.−1) < D(1. For n = 3 the situation starts to become too complicated to calculate by hand.1.1.−1) D(1. The second follows from the fact that U (C) is already finite in the case of Cn .1)    . COMPUTATIONS OF SPECTRA        D(1. as we will see.−1) ∧ D(1. Let U be a subfunctor of L∼sa and C ∈ C(A).1.1) D(−1.−1. The last equation says that A0 (C)U (C) is the down-set of the meet of U (C).1. If L is a lattice and x ∈ L.1.1) D(−1. We find that these subfunctors ¯ A are the ones that are locally down-sets.−1) D(1.−1) D(1. 94 . Now.2.−1) The spectrum Now we compute Σ.−1) D(1.−1.−1.1)  D(1.−1.1)      D(1. First we recall the definition of a down-set.−1.1)      > < ∧ D(−1.  D(−1.−1.−1.6.−1.−1)      D(−1.−1) > D(1.−1. which follows from theorem 6. let us recall the definition of A0 . Recall from section 5.−1)            D(−1.1.3 that Σ(⊥) as a lattice in Set may be represented by the subfunctors U of L∼sa such that A0 U = U .1.1) D(−1.−1.1.1. Conversely.1. THE FINITE. Then ¯ A A0 (C)U (C) = = = Da ∈ L∼sa (C) | Da ¯ A Da ∈ L∼sa (C) | Da ≤ ¯ A U (C) U (C) Da ∈ L∼sa (C) | For every q ∈ Q+ : Da−q ≤ ¯ A U (C) The first equation is the definition.

−1) D(−1. this determines Σ globally. COMMUTATIVE CASE: CN CHAPTER 6.−1) 0 0 A2.−1) D(−1.−1) 95 .1 = D(−1. THE FINITE.−1) D(1.1 = > > D(−1.1) 0 D(−1.−1) D(−1.1.1) D(1.−1) 0 0 A2.1) 0 A2.3 = > D(−1. COMPUTATIONS OF SPECTRA n = 1: We find that Σ(⊥) is equal to the following lattice: D1↓ D−1↓ .−1) D(−1. Since C(A) consists only of one element.−1) D(−1.−1) D(−1.1) D(1.1) D(1. n = 2: The following subfunctors of L∼sa are the ones that are locally down-sets and hence these subfunc¯ A tors will form the lattice Σ(⊥): D(1.1) A1.2 = > 0 D(1.6.−1) D(−1.

4 = > 0 0 D(−1. It follows from these equations ¯ A Σ that ⊆⊥. The lattice structure of Σ(⊥) is then given by A1. We will restate the argument above for the computation of Σ( ) in the more general theorem 6.1 A2.1 A2. ⊥. ¯ A sa where ! : 0 → L∼sa is the unique arrow from the initial object of Set.3.−1) D(−1. ⊥. ⊥. (⊆∗ ×IdL∼ )( ) = Id{1×L∼ } . We now compute Σ(C2 ) = Σ( ). ⊥. ¯ ¯ Asa Asa (⊆∗ ×IdL∼ )(⊥) = !. COMPUTATIONS OF SPECTRA 0 0 A2. We compute: ⊥.2 A2. which by theorem 5.1 A1.1 A2. Recall that images can be taken point-wise.7 and the subsequent discussion equals Im(⊆Σ ).3 below.4 .3 A2. χU = χU (⊆∗ ×IdL∼ ) = χV (⊆∗ ×IdL∼ ) =⊆Σ χV if and only if U ( ) = V ( ).4 .2 A2.6. ⊥. COMMUTATIVE CASE: CN CHAPTER 6. THE FINITE.−1) Here the 0 indicates that the element is not included in the subfunctor.3 A2. ¯ ¯ Asa Asa Hence Σ( ) has the form A2. 96 .1.1.

. . ⊥.6.4.C ¯ Asa computation of the map (⊆∗ ×IdL∼ ) : y(C) × L∼sa → y(⊥) × L∼sa . ii. (b) Recall that ⊆Σ = (⊆∗ ×IdL∼ )∗ .3 that if ⊆⊥. Let U and V be subfunctors of L∼sa . 97 .C ⊥.1. ⊥. Ω) are the natural transformations χU and χV . Proof: In this case A itself is the top element of C(A). Ω) = HomT (A) (A. It follows ¯ ¯ A A Σ Σ from theorem 6. A Σ Σ χU =⊆⊥. . ¯ A sa where ! : 0 = (y(C))(D) × L∼sa (D) → (y(⊥))(D) × L∼sa (D) = L∼sa (D) is the unique arrow ¯ ¯ ¯ A A A from the initial object. . then V = V .e. Now we have the following situation: i. n}). One has to compute all subfunctors of L∼sa ¯ A such that locally the subfunctor is a down-set. . . (b) From theorem 6. ⊥. χUU ( ) . which form Σ(⊥).1. . From this we conclude that Σ( ) = P({1. being exactly the definition of exponentiation (see theorem 1. ⊥. ⊥. By theorem 6. Hence the images ⊆Σ χUV ⊥.C ⊥. Hence every element of Σ( ) is of the form ⊆⊥. the empty set. . if V ⊆ L∼sa ( ) then there is a subfunctor UV ⊆ L∼sa such that UV ( ) = V . The following theorem shows what the images of the intrinsic maps ⊆Σ look like.C ⊥.3. If D ⊇ C.C From this theorem we derive a corollary that gives the localic spectrum of Cn by the original Gelfand theorem in Set. .1. i. Then: ¯ A ¯ (a) The corresponding elements of U in V in Ω Asa (⊥) = HomT (A) (y(⊥)×A.1. then ((⊆∗ ×IdL∼ )∗ χU )(D) = χU (D)(⊆∗ ×IdL∼ )(D). COMPUTATIONS OF SPECTRA n ≥ 3: For this case one proceeds in exactly the same way. The second assertion is a consequence of the following ⊥. Now U (D) = V (D) for all D ⊇ C if and only if χU (D) = χV (D) for all D ⊇ C if and only if (⊆Σ χU )(D) = (⊆Σ χV )(D) for all D ⊇ C.C ⊥. THE FINITE. C(A) has a top element and the top level of the spectrum Σ( ) is given by the lattice P({1.C L∼ Proof: (a) This is obvious. If D ⊇ C.C ×IdL∼ )(D) = Id{1}×L∼ (D) if C ⊆ D.2 L∼sa ( ) is of the form ¯ A P({1.C ¯ A A A sa (⊆∗ ⊥. χUV =⊆⊥. . represent different elements of Σ( ) for every V . This can easily be done by a computer. n}). χUV . In the case of A = Cn .1. n}). Then: (a) Clearly.C ¯ ¯ Asa Asa = ! : 0 → Ω(D). In the case n = 3 one finds 96 such functors.C ¯ ¯ Asa Asa = χU (D). χUV .3 it immediately follows that for every subfunctor U ⊆ L∼sa one has ⊆Σ ¯ ⊥. which is given by ¯ ¯ ⊥. COMMUTATIVE CASE: CN CHAPTER 6.10). (b) For C ∈ C(A). Corollary 6. one has ⊆Σ χU =⊆Σ χV if and only if U (D) = V (D) for all objects C ⊆ D.1. . ¯ ¯ Asa Asa ×IdL∼ )(D) = ! if C ⊆ D. .2.C Theorem 6. then ((⊆∗ ×IdL∼ )∗ χU )(D) = χU (D)(⊆∗ ×IdL∼ )(D).C (⊆∗ ⊥.C ⊥.

u z1 0 0 z2 u−1 | z1 . C)) =      ˙ C1 = Bu  . z2 ∈ C . hence u = u. Now suppose b ∈ B. 98 . All nontrivial unital C*-subalgebra’s of M (2. C). C).     z 0 0 z |z ∈C where the dots stand for uncountably many C*-subalgebras running through all unitary u ∈ M (2. the lattice ∼ LAsa can be computed locally at C and every local lattice has the form of an n-dimensional hypercube. We first prove a proposition that gives all commutative unital C*subalgebras of A = M (2. Hence u∗ = u−1 . Proposition 6. So suppose b = udu−1 . . Then b∗ = (u∗ )−1 d∗ u∗ commutes with b.    .7 and 6. so u is unitary. Let b ∈ B. If d has a 1 in the upper right entry one can show that B = M (2. Theorems 5. Proof: We denote matrices by small characters. C) are of the form Bu = where u is a unitary matrix. The spectrum of M (2.2.2. Let B be a unital commutative C*-subalgebra of M (2. . . Since b commutes with b. C).1 1 0 d2.1. Then L∼sa is given by the following functor. where d is diagonal.    C(M (2. then b = udu−1 . .1. C). computing the spectrum of any finite-dimensional (non-commutative) C*-algebra becomes much easier! Once we know the structure of C(A).3. then b = u d u −1 .3 then give a complete description of Σ globally.6. THE FINITE CASE IN GENERAL CHAPTER 6. we study M (2. COMPUTATIONS OF SPECTRA 6. between which there is no (nontrivial) inclusion relation. which means that b and b∗ have a common set of eigenvalues. C) As an example. as ¯ A follows from the example of Cn and the fact that we may compute this lattice point-wise. C). b and b have a common set of eigenvalues. ¯ Then Σ(⊥) has the form of all subfunctors such that locally the subfunctor is a down-set.2 . This proves the theorem. Where d is either diagonal or of the form: d1. We thus find:  .2 The finite case in general Now that we know what the spectrum of Cn looks like.

3.     D−Id2     D−Id2 Here a−+ = −a+− = −1 0 0 1 . the continuous functions on a compact Hausdorff space X. For any other ¯ A subfunctor S ⊆ L∼sa we have DId2 ∈ S(⊥). as each defines a subfunctor of L∼sa . COMPUTATIONS OF SPECTRA       . SOME REMARKS ON C(X) CHAPTER 6.3 Some remarks on C(X) In this paragraph we study C(A) for A = C(X). Let Π be 99 . . .  ¯         D−Id2 ↓ where the map ⊆Σ u maps a subfunctor U ⊆ L∼sa . For convenience we first assume X = [0. commenting the general case afterwards.Dua−+ u−1     DId2       Dua+− u−1. regarded as an element of Σ(⊥). DId2     . 6.1. It follows straight from theorem 6. regarded ¯ ⊥. there is L∼sa itself.6. to U (Bu ).3 that   DId2 ↓           ∼ Σ(Bu ) = LAsa (Bu ) = Dua−+ u−1 ↓ Dua+− u−1 ↓ . where R indicates the cardinal number of {u | u is a unitary matrix}. 1})R . Now what are the subfunctors that are locally down-sets? First. namely DId2 ↓. 1] ⊆ R. Hence we conclude that Σ(⊥) is ¯ A the lattice L∼sa ¯ A P({0.B A as an element of Σ(Bu ). . For each Bu we are free to ¯ A chose any of these four down-sets. . Then for every unitary matrix u we have four choices for ¯ A down-sets of L∼sa (Bu ). Dua+− u−1 ↓ and D−Id2 ↓. Dua−+ u−1 ↓.

There are three natural questions to ask: 1. 2 2 2 2 and define ∼ϕ to be the smallest equivalence relation on [0. and on the other hand the category of unital C*-subalgebras of C([0. namely itself and the image of ϕ (which lies in ( 1 . Since f is continuous we can find a δ such that if x ∈ (p − δ. p + δ)) is dense in the interval ( 1 . 1]) of the form CΠ for some closed partition Π? The first question The answer to the first question is obviously yes. Then there exist p. From this we make a short remark: we could. we introduce the algebra CΠ of all continuous functions on [0. φ is a bijection. Hence we would have proved the claim if every function in CΠ∼ϕ is constant. q ∈ (0. Then Π → CΠ is a functor between these categories. For every open subinterval of (0. Let φ be a function from the open inteval (0. 1] that are constant on equivalence classes. ) → ( . One can prove that the cardinality of the number of equivalence classes of Π is not countable. Let f ∈ CΠ∼ϕ . 1). SOME REMARKS ON C(X) CHAPTER 6. Suppose it is 1 not. The second question The answer to the second question is no. 1). on the one hand consider the category of partitions of [0. 1] such that if y = ϕ(x) then y ∼ϕ x. 2. Such a function exists as we shall prove later. a partition of [0.6. 1). This can be illustrated by the following example. p + δ) then |f (x) − f (p)| < |f (p) − f (q)|/2. 2 ) such that f (p) = f (q). 1] such that every equivalence class is closed in [0. 1) the image of this subinterval under φ is dense in (0. Now. Now remark that f (ϕ(x)) = f (x) and ϕ((p − δ. For two closed partitions Π1 and Π2 . f (p) + |f (p) − f (q)|/2). 1 ) is equivalent 2 to exactly two numbers. Is every unital C*-subalgebra of C([0. 1]) with the inclusions. 100 . Call the corresponding 2 partition Π∼ϕ and call the trivial partition consisting of only one equivalence class Π0 . 1). We claim that that CΠ∼ϕ = CΠ0 . 2 ). which is a contradiction. 2 and 1 is only equivalent to itself and that every number in (0.3. 1 Note that each of 0. COMPUTATIONS OF SPECTRA a closed partition of [0. 1) to itself having the following properties: 1. 1] (with the euclidian topology). Note that CΠ0 only contains the constant functions. that is. 1)). then the partition is trivial. 2 Then f (q) = f (ϕ(q)) ∈ (f (p) − |f (p) − f (q)|/2. From this and the 2 continuity of f we may conclude f (x) takes values between f (p) ± |f (p) − f (q)|/2 for every x ∈ ( 1 . as in the situation of Cn . Hence f is constant. does the following implication hold: CΠ1 = CΠ2 ⇒ Π1 = Π2 ? 3. 1]. Is CΠ a C*-algebra? 2. 1] with the refinements of partitions as morphisms. By definition of the equivalence relation and 1 the continuity of f we are finished if we can prove f to be constant on the interval (0. Let ϕ(x) = 1 1 1 1 + φ(2x) : (0. If the number of equivalence classes is finite.

Now we claim i∗ is surjective. If C is any C*-algebra.i be a sequence in Q converging to ak. Let bk. n ∈ N. 1] be constant equal to 0.n. pn Let f be a bijection between the countable sets {(k. Such sequences bk. Finally. Hence there exist bijections gk.n Bk. bk. We first introduce some notation. COMPUTATIONS OF SPECTRA It remains to prove the existence of the function φ. Define: Bk. n)|k. 1] to C that are constant on equivalence classes of some partition Π.n) except for a finite subset 1 of Df (k. p |k . Suppose i∗ is not surjective. some category theory and topology. Then there is a point p ∈ Ω(B) such that p is not in the image of i∗ .i exist.n) .i |i ∈ N} . A proof can be established from the Gelfand . Let A be the C*-algebra C[(0.n Bk.n. Now let the function h : Ω(B) → [0.3.n : Bk.k. We would like to prove B is equal to the continous functions on [0.n (x) x ∈ Bk. 1)\ k.n and (0.k . Note that both the sets Bk.n. We conclude that i∗ is surjective. 1)\ k. 1]). Ω(B) is compact Hausdorff. For p prime. 1)\ p Dp . n ∈ N. 2 |k} . Then f ◦ i∗ = h ◦ i∗ but f = h which contradicts with the fact that i∗ is epi. From this one obtains i∗ is epi. The image of i∗ is compact since i∗ is continuous and Ω(A) is compact. Remark that the one-point set {p} is closed.n = {bn.n = 2k for natural numbers 2 ≤ n.n . Let ak. i = i . It is obvious that this function is a bijection.6. Suppose B to be a unital C*-subalgebra of C([0. and the set of all primes. 2 ≤ n. 1] such that f (p) = 1 and f (x) = 0 for all x ∈ i∗ (Ω(A)). Let i : B → A denote the inclusion of B in A. The third question The answer to the third question is yes. Hence we can apply Urysohn’s lemma giving a continuous function f : Ω(B) → [0. Let g be a bijection between the uncountable sets (0. meant by the Gelfand-Naimark theorem (rather than the spectrum as defined in previous chapters).n . It at least contains one point an.i ∈ I.n) . write Ω(C) for the spectrum of C. Now define an equivalence relation on Ω(A) by: ep ∼Ω eq if and only if i∗ (ep ) = i∗ (eq ) 101 . Now suppose 1 I ⊆ (0. Fix n and k for now. 2 ≤ n.n such that bk. x → g(x) x ∈ (0. 1)]. Then there exists a N ∈ N such that for all i > N . 1 ≤ k < 2n .n.n and Dp are countable. 1 ≤ k < pn .n .i ⇒ k = k . Then φ(I) contains Df (k. define φ : x → gk. n = n .n → Df (k. define: Dp = k |k.Naimark theorem. SOME REMARKS ON C(X) CHAPTER 6. 2 ) is an open subinterval.i = bk . We claim this function to be the desired φ . The Gelfand-Naimark theorem states that there is a duality between compact Hausdorff spaces and unital commutative C*-algebras. Then we have: B →i ∗ Ω(B) ←i A Ω(A) First remark that i is monic since it is injective. 2 ). hence compact. From this it follows φ(I) is dense in (0. hence normal. 1 ≤ k < 2n n and 2 |k.

102 . respectively. is a continuous function that is constant on equivalence classes. 1] being compact and Hausdorff. 1] → C that is constant on equivalence classes defines a continuous function on [0. but at least this representation gives an idea what the C*-subalgebra’s of C(X) look like and. and notice that f (p) = f (q).3. COMPUTATIONS OF SPECTRA Here ep . 1]/ ∼0 and that this assignment is injective. . ∀f ∈ B. if the equivalence relation is neat enough. Finally. where both π and i∗ are continuous.i∗ is bijective and continuous. For the computation of the spectrum. Hence C([0. 1]/ ∼0 is the quotient map. The three questions amount to a representation of all unital C*-subalgebras of C(X) where X is a compact Hausdorff space. It is easy to see that each continuous function on [0. as we saw in the example of a discrete n-point space and in the example of [0. From the facts that: ¯ .12 seems to be useful. it gives a hint what the lattice structure looks like. The Gelfand-Naimark theorem tells us that B C(Ω(B)) C(Ω(A)/ ∼Ω ) C([0.6. 1]/ ∼0 ) is nothing else but the continuous functions from [0. On the other hand. where ep . The answer to the second question does depend on X. 1]/ ∼0 ). define for p. Further remarks In the above discussion we only used the topological properties of [0. 1] to C that are constant on the equivalence classes defined by ∼0 . . then f ◦ π0 .e. q ∈ [0. ∀f ∈ B if and only if i∗ (ep ) = i∗ (eq ). i. 1] → [0.2. where π0 : [0.Ω(B) is Hausdorff. SOME REMARKS ON C(X) CHAPTER 6. eq ∈ Ω(A) are the evaluations in p and q. 1]/ ∼0 ). Of course the question wether two partitions are equialent with respect to ∼Π may become difficult to anwer. ∗ ¯ ¯ Then i∗ factors through (Ω(A)/ ∼Ω ) by i∗ : Ω(A) →π (Ω(A)/ ∼Ω ) →i Ω(B). eq ∈ Ω(A) are evaluation at p and q. For the last isomorphism. we claim that C([0. This proves claim number 3. the structure of C(A) is still too rough to compute the spectrum properly. 1]. and this is the inverse assingment. if f is a function in C([0. 1]/ ∼0 ) consist of all continuous functions that are constant on (closed) equivalence classes. it even tells you when a C*-subalgebra is contained in one another. Namely: C(C(X)) = {B ⊆ C(X)|B a unital C*-subalgebra of C(X)} {Π | Π is a partition of X} / ∼Π . where Π1 ∼Π Π2 iff CΠ1 (X) = CΠ2 (X). theorem 5. ¯ it follows that i∗ is a homeomorphism. 1]: p ∼0 q if and only if f (p) = f (q). The answers to the first and third question stay the same in the setting of any compact Hausdorff space X instead of [0.(Ω(A)/ ∼Ω ) is compact (since the quotient of a compact space is compact). 1]. However. So we may view i∗ as a map from Ω(A) to (Ω(A)/ ∼) (this is actually the map π).

6.3. SOME REMARKS ON C(X)

CHAPTER 6. COMPUTATIONS OF SPECTRA

Remark: after finishing this thesis we found out that the C*-subalgebras of C(X) correspond to the partitions of X that are obtained by a closed equivalence relation R. That is R ⊆ X × X should be closed in the euclidian topology of X × X. Then the quotient space X/R is compact Hausdorff, as this follows from elementary topology. The space of complex functions on this quotient space is a C*-subalgebra of C(X) and it can be proved that every such algebra is of this form for a unique equivalence relation R.

103

Further work
In this chapter we make some final remarks. We have shown that if a C*-algebra is finite dimensional, we are able to compute its spectrum once we have found its commutative subalgebras. As a special case, we computed the spectrum of M (2, C), which was the main goal at the start of this research. For the case of A = C(X), there is some research to do to find a representation of the structure of C(A). Also, it would be interesting if the questions stated in paragraph 6.3 can be answered in the same way as C0 (X), where X only locally compact, instead of compact; i.e. in the general case of a commutative C*-algebra. Of course, B(H) would be interesting too, especially since the GNS-construction shows that every C*-algebra admits an injective representation in B(H). Another generalization would be to change our topos T (A) to the setting of sheaves. The most natural way to do this is by remarking that SetC(A) itself can be regarded as a sheaf if one equips C(A) with the so-called Alexandrov topology. In this topology the open sets are defined as the up-sets of the partial order C(A). If one puts some other topology on this space, one can take the sheaves on this topological space as the definition of T (A). One can still define the free lattice subjected to relations and one can still define a completion map (maybe you will need the to take the sheafification of a presheaf once in a while). If we let U ⊆ C(A) be an open set and Vi an open covering of U , then the properties of a sheaf at Vi give information about the sheaf at U , which might lead to more interaction between the information of the C*-subalgebras of A and hence spectra may become different.

104

Index
Adjoint, 23 Cartesian closed, 23 functor category, 24 Category, 8 Closed partition, 100 Coequalizer, 17 Cokernel pair, 30 Colimit, 16 Compact, 42 Completion, 84 Composable pair, 8 Composition, 8 Composition map, 8 Constant functor, 12 Contravariant functor, 11 Contravariant Yonedafunctor, 12 Coproduct, 17 Cosieve, 21 Covariant functor, 12 Covering, 85 Down-set, 94 Epi, 11 Epi-monic factorization, 30 Equalizer, 15 Exists, 52 For all, 56 Forgetful functor, 12 Formula, 58, 60 Freely generated lattice, 73 Freely generated lattice subject to relations, 77 Functor, 11 Functor category, 13 Heyting algebra, 41 Hom-set, 10 Horizontal composition, 13 Identity, 8 Initial object, 10 Integral numbers object, 27 Internal C*-algebra, 72 Heyting algebra, 52 lattice, 73 Intrinsic map, 13 Isomorphism, 9 Large category, 9 Lattice, 41 Left adjoint, 23 Limit, 15 finite, 17 Limiting cone, 15 Local, 75 Locale, 41 Mitchell-B´abou language, 58 n Modus Ponens, 67 Monic, 11 pullback of a monic, 29 Morphism, 8 Natural numbers object, 27 Natural transformation, 13 Norm, 71 Object, 8 Opposite category, 10 PBL, 29 Point-wise, 75 Power object, 24 Preorder, 10 Presheaf, 14 Presubobject, 39 Product, 15 Pullback, 16 105

18. 18 Target. 31 Rational numbers object. 39 Heyting algebra. 8 Tautological functor. 18 uniqueness. 37. 38 Subobject. 27 Right adjoint. 29 Pushout. 9 Source. 36 Small category. 88 Strict. 71 Terminal object. 46 Pullback lemma. 10 Topos. 24 Type. 23 Sheaf. 8 Spectrum.INDEX INDEX Pullback functor. 28 106 . 14 Slice category. 58 Yoneda functor. 41 Subobject classifier. 12 Yoneda lemma.

Davey. Topology via logic.org/abs/0709. Scott.. Journal of Universal Computer Science. 1990. B. II. [14] Mulvey. Cambridge University Press. Cambridge University Press. [2] Coquand. [16] Vickers. B.. 1994. [4] Coquand. Introduction to higher order categorical logic. Elsevier Science Publishers. Journal of Pure and Applied Algebra.. [11] Moerdijk.. [5] Goldblatt.. Lansman.A. P. Cambridge University Press. http://arxiv. B. Sheaves in geometry and logic. T. Handbook of categorical algebra volume I. S. A globalization of the gelfand duality theorem. MacLane.4364. 23(4):425-464. 1997 [7] Heunen. Annals of Pure and Applied Logic.. Springer-Verlag. [13] Mulvey.. R. Springer-Verlag. [15] Priestly. C. Banaschewski. 1984. Algebraic Geometry (8th edition). Sketches of an elephant.. volume 137. 2000. T. C. The spectral theory of commutative C*-algebras: the constructive spectrum. [8] Johnstone..A.. B. B. H.J. 1971.J. Spitters. [9] Lambek. 1992. S. [12] Mulvey. [10] MacLane. the categorical analysis of logic. P. 2007. submitted for publication. Quaestiones Mathematicae. 107 . III. Springer-Verlag. Introduction to lattices and order. 23(4):465-488. B. C. Cambridge University Press. Oxford Science Publications. [6] Hartshorne. About Stone’s notion of Spectrum. 2002. 1989. a topos theory compendium volume I.P. volume 197. Formal topology and constructive mathematics: the Gelfand and StoneYosida representation theorems. Quaestiones Mathematicae.Bibliography [1] Borceux. Categories for the working mathematician. R. N. Spitters. 2005. Topoi. C.T. I. J.. S. F. January 2006. Banaschewski. 2000. An elementary constructive proof of Gelfand duality for C*-algebras. [3] Coquand.J. 28 December 2005. Spitters. B. 1986. The spectral theory of commutative C*-algebras: the constructive Gelfand-Mazur theorem. T. A topos for algebraic quantum theory.J. Banaschewski. 6 April 2008..

Also. I would like to mention Desda. to spend time to me and advise me. even though some of them had no idea what a C*-algebra or a topos was and hence possibly had no idea what I was talking about. the students association. this Masters thesis is the final result in becoming a Master of Science. 108 . This was especially the cas in the last months. Thanks. and especially Klaas Landsman who taught me on the subjects of C*-algebra’s. I would like to thank the staff of the mathematics department for teaching me. Some special words should be spent on the Nijmegen mathematics students. I would like to thank my parents. since they have always showed interest in the things I have done and who were always prepared to help. Hilbert spaces and Category theory and supervised my work for the past half year.Acknowledgements After devoting five years of my life to mathematics. This is the right place to thank some people who have helped me over the past few years. First. Finally. when I tried to explain what my Masters thesis was about. His enthusiasm and helpfulness motivated me during the past two years and eventually led to this thesis. Also I would like to thank Chris Heunen and Bas Spitters for answering my questions and Mai Gehrke for spending her time to act as the second reader. who are very cooperative and helped me a lot throughout the past five years.

Sign up to vote on this title
UsefulNot useful