Condensed Book

An introduction to
condensed mathematics
Bernard Le Stum (April 3, 2024)

– What I care most about are definitions (Peter Scholze - quoted by Michael Harris).
Réalisé en LATEX to partir du modèle Legrand Orange Book

Copyright © 2024 Bernard Le Stum
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1 Categories and functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1.1 Category 11
1.1.1 Definition/Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.1.2 Isomorphism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.1.3 Subcategory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2 Functor 14
1.2.1 Definition/Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2.2 Natural transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.2.3 Representable functor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.3 Limit 20
1.3.1 Diagrams, cones and limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.3.2 Specific limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.3.3 Constructions of limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.4 Applications 25
1.4.1 Filtered colimit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.4.2 Preservation of limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.4.3 Projective/Injective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.4.4 Algebraic structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.4.5 Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.5 Adjoint 31
1.5.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.5.2 Unit and counit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.5.3 Adjoint and limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
1.5.4 Reflective subcategory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
1.5.5 Kan extension (optional) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2 Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.1 Compact Hausdorff space 39
2.1.1 Compact/Hausdorff space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.1.2 Stone-Čech compactification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.2 Projective objects 43
2.2.1 Free compact Hausdorff space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.2.2 Totally/Extremally disconnected space . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.2.3 Stone space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.2.4 Stonean space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.3 Compactly generated space 48
2.3.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.3.2 Compact-open topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.3.3 Weak Hausdorff space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3 Sites and topos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

3.1 Presheaf 53
3.1.1 Defintion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.1.2 Yoneda embedding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.1.3 Equivalence relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.1.4 Slice category . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.2 Site 57
3.2.1 Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.2.2 Sheaf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.2.3 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.3 Topos 64
3.3.1 Pretopos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.3.2 Precanonical topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.3.3 Generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.3.4 Topos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.3.5 Internal Hom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.3.6 Quasi-compact/separated . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.4 Morphism of topos (optional) 72
3.4.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.4.2 Presheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.4.3 Morphisms of sites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.4.4 Induced topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.4.5 Topological spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4 Condensed sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.1 Condensed set 79
4.1.1 Compact Hausdorff spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.1.2 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.1.3 Free compact Hausdorff spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.2 Topological space and condensed set 83
4.2.1 Associated condensed set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.2.2 Internal Hom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.2.3 Quasi-compact/separated condensed set . . . . . . . . . . . . . . . . . . . . . . . . 88
5 Commutative algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.1 Additive category 91
5.1.1 Pre-additive category . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.1.2 Additive category . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.1.3 Exact sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.2 Abelian category 94
5.2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.2.2 Grothendieck category . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.3 Abelian sheaf 97
5.3.1 Defintiion/Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.3.2 Abelian group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.3.3 Internal Hom and tensor product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6 Condensed abelian groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

6.1 Condensed abelian group 103
6.1.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
6.1.2 Grothendieck category . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.2 Topological abelian groups 105
6.2.1 Associated condensed group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.2.2 Locally compact abelian groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
7 Cohomology (optional) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

7.1 Complex 111
7.1.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.1.2 Homotopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
7.1.3 Mapping cone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
7.1.4 Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
7.1.5 Quasi-isomorphism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
7.2 Derived functor 120
7.2.1 Derived category . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
7.2.2 Derived functor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
7.2.3 Spectral sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6
7.3 Sheaf cohomology 127

7.3.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
7.3.2 Simplicial method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
7.3.3 Čech cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
7.4 Morphisms of topos (optional) 133
7.4.1 Morphism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
7.4.2 Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
7.4.3 Topological spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
8 Condensed cohomology (optional) . . . . . . . . . . . . . . . . . . . 137

8.1 Cohomology 137
8.1.1 On Stonean spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
8.1.2 On Stone spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
8.1.3 On locally compact spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
8.2 Banach abelian groups 139
8.2.1 K-exactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
8.2.2 On Stone spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
8.2.3 Banach spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
8.3 Extensions of abelian groups 143
8.3.1 First computations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
8.3.2 Breen-Deligne resolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
8.3.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
8.3.4 Locally compact abelian groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
Références . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
Introduction
Teaser
The right environment in order to perform linear algebra is that of an

abelian category. For example, vector spaces over R (or any field) form an
abelian category. More interesting, we can consider the category of vector
spaces endowed with an operator. This is again an abelian category. Also
abelian sheaves on a topological space (or more generally on a site) form
an abelian category. This is not true however for Banach (or Hilbert)
spaces for example. The presence of a topology makes Noether first
isomorphism (or equivalently Grothendieck AB2 axiom) fail.
The queen of abelian categories is the category of abelian groups.
Actually, this is even a “Grothendieck category satisfying AB6 and AB4*”.
Introducing a topology leads to consider the category of topological
abelian groups. This is not an abelian category unless we focus on
compact Hausdorff abelian groups in which case we miss usual (infinite)
abelian groups as well as (non trivial) Banach spaces. Dustin Clausen
and Peter Scholze define a condensed abelian group as an abelian sheaf on
the site (for the precanonical topology) of all compact Hausdorff spaces1 .
They form a Grothendieck abelian category satisfying AB6 and AB4*
that contains all compactly generated abelian groups.
Vector spaces often come with a topology, and linear maps between them are
then required to be continuous. In particular, an isomorphism is not only supposed
to be linear and bijective, but it also needs to be a homeomorphism. Unfortunately,
it is not true anymore, in this situation, that a linear map whose kernel and cokernel
1
Actually, they consider profinite sets but this is equivalent.
8
vanish is an isomorphism. The baby example is given by the identity on the real line
R when R is given the discrete topology on the one side and its usual topology on
the other. In particular, all the standard tools from commutative algebra are not
anymore available when some topology is involved. Mathematicians have been trying
to resolve this issue for some time now, starting maybe with the work of Choquet in
the late 40’s and first formalized by Johnstone in the late 70’s ([Joh79]).
His original idea was to not only consider the points of a given topological space
X but also the set of all convergent sequences in it (in order to keep track of the
topology). A point of X may be seen – in a very fancy way – as a (continuous) map
defined on a one point space • with values in X. Similarly, a convergent sequence
(together with the choice of a limit) may be seen as a continuous map defined on the
one-point compactification N of the set of all integers N with values in X. Let us
denote by X(•) and X(N) the corresponding sets. They are not unrelated because
there exists various maps between them obtained for example by composing with
the various points of N (meaning maps from • to N). One may then consider more
generally any couple of sets T (•) and T (N), endowed with some family of maps
between them and subject to some conditions. In modern language, this is a sheaf
on a site. The collection of all these sheaves may be seen as an enlargement of the
collection of all topological spaces. Abelian sheaves always form a Grothendieck
abelian category satisfying AB5 and AB3*.
Although promising, this is not completely satisfying. For example, there exists
the general notion of a (quasi-) compact sheaf but the unit interval [0, 1] is not
(quasi-) compact in this setting. This may be fixed by adding a new test-space, for
example the Cantor space 2N which surjects continuously onto [0, 1]. The idea of
Clausen and Scholze is to actually use the site of all (light) profinite spaces S and not
merely • and N or even 2N . A condensed set is then simply a sheaf (of sets) on this
site. And a condensed abelian group is an abelian sheaf on this site. It happens that
condensed abelian groups have all the properties expected from commutative algebra:
much like abelian groups, they form a Grothendieck abelian category satisfying AB6
and AB4*. Moreover, their cohomology provides the expected invariants for locally
compact abelian groups.
There also exists the notion of a completion in this theory called solidification
that matches the properties of usual non-Archimedean completion. Clausen and
Scholze are then able to provide very elegant proofs of several theorems from algebraic
geometry and even obtain new results. The idea is that many results in geometry
require strong global properties such as properness because of finiteness constraints.
Condensed geometry provides a way to remove the global condition and work locally,
which is much more natrual. Note that there also exists the notion of a liquid
vector space that matches the properties of Banach spaces in the archimedean world.
Their introduction is necessary in order to recover classical analytic geometry (and
functional analysis).
This course is intended for regular students. We shall present some material
which is necessary to understand the basics of the theory of condensed sets. We may
only reach the definition of a condensed abelian group in the end – and hopefully be
able to show that they satisfy the expected properties. We hope however that we
can cover some cohomological results. We apologize before the the reader that the
9
limited time for the course does not allow us to do more and send him to further
readings.
The first chapter is presentation of general category theory. The second one is a
review of topology with special focus on compact Hausdorff spaces. The third chapter
is an introduction to topos theory (it is completely independent from the second one).
Chapter 4 is devoted to condensed sets. Chapter 5 presents the theory of abelian
categories. Chapter 6 is devoted to condensed abelian groups. Chapter 7 deals
with cohomology theory. Chapter 8 is devoted to the computation of cohomology of
condensed abelian groups.
Here is non exhaustive list of links to yet unpublished documents related to the
theory of condensed sets (see also the bibliography at the end of the course):
• Lectures on condensed mathematics – Clausen, Scholze.
• Lectures on analytic geometry – Clausen, Scholze.
• Condensed mathematics and complex geometry – Clausen, Scholze.
• The foundation of condensed mathematics – Àsgeirsson.
• Mathématiques condensées – Le Bras.
• Condensed mathematics – Mathew.
• Crash Course Condensed Mathematics – Barton, Commelin.
• Condensed and locally compact abelian groups – Deglise.
• Condensed Mathematics Seminar – Morgan, Rodrigez-Camargo
Many thanks to the student who took my course - and those who freely attended
it - for their feedback that helped me improve on these notes.
1. Categories and functors
For those who might worry about set-theoretic issues (see [Shu08] for example),
we shall stay in a fixed universe (some large set, see definition 1.1.1 of [KS06] for
example). We shall then call a set only those sets that are contained (⊂) in our
universe. Such a set is said to be small if it actually belongs (∈) to our universe.
1.1 Category
1.1.1 Definition/Examples
Definition 1.1.1 A (locally small) category consists in the following data:
1. a set of objects C,
2. for all X, Y ∈ C, a small set of morphisms Hom(X, Y ),
3. for all X ∈ C, an identity morphism IdX ∈ End(X) := Hom(X, X),
4. for all X, Y, Z ∈ C, a composition rule
Hom(X, Y ) × Hom(Y, Z) → Hom(X, Z), (f, g) 7→ g ◦ f
such that
(a) f ◦ IdX = f and IdY ◦ f = f ,
(b) if h ∈ Hom(Z, T ), then h ◦ (g ◦ f ) = (h ◦ g) ◦ f .
The category is said to be small if its objects (or equivalently its morphisms)
form a small set and finite if there is a finite number of (objects and) morphisms.
We will usually write f : X → Y instead of f ∈ Hom(X, Y ) and call X (resp. Y )
the domain (resp. le codomain) of f . Note that IdX is uniquely determined by the
conditions (4a). In practice, we shall simply say that the set C is a category1 but we
must not forget that it involves some extra structure: morphisms and composition.
1
As one may denote a group by G without explicitly mentioning the multiplication rule.
12 Chapter 1. Categories and functors
Examples 1. If G is a monoid, one can consider the category G with a unique

object •, End(•) := G and composition given by multiplication on G. This
way, we get essentially all categories with a unique object.
2. If ≤ is a preorder on a set X, we will then denote by X the category whose
objects are the elements x ∈ X and morphisms are couples (x, y) for x ≤ y.
This way, we get essentially all the small categories whose Hom have at most
one element.
3. As a particular case, one may always endow a set with equality and consider
any set as a category (with only identities). This provides all small category
with only identities as morphisms.
4. The naturel number n := {0, . . . , n − 1} is an ordered set and we will denote by
n the corresponding category. For example, 0 is the empty category that has
no objects (and no morphisms), 1 is the category that has exactly one object
(and one morphism), 2 is a category with two distinct objects and a unique
morphism between them.
5. If X is a topological space, then Open(X) is ordered by inclusion and we shall
denote by Open(X) the corresponding category.
6. We shall denote by Set the (large) category whose objects are sets and mor-
phisms are maps between them. We shall write F(X, Y ) for the set of maps
between two sets.
7. In the same way, we shall consider the (large) category Top whose objects
are topological spaces and morphisms are continuous maps. We shall write
C(X, Y ) for the set of continuous maps between two topological spaces.
8. Finally, we will denote by Ab the (large) category whose objects are abelian
groups and morphisms are homomorphisms. We shall write HomZ (M, N ) for
the set of homomorphisms of abelian groups.
Exercise 1.1 Define the categories Mon, Gr, Rng, G-Set, A-Mod and k-Alg of
monoids, groups, rings, G-sets, A-modules and k-algebras.
Definition 1.1.2 The simplex category ∆ has positive integers [n] := n + 1 for
n ∈ N as objects and order preserving maps as morphisms. The injective (resp.
surjective) maps
δni : [n − 1] → [n] (resp. σni : [n + 1] → [n])
that forget (resp. repeats) the i-th term are called the face (resp. degeneracy)
maps.
Exercise 1.2 Show that any morphism in ∆ is a composition of face and degeneracy
maps.
The product C × C ′ of two categories C and C ′ is itself a category (everything is

done termwise). The opposite category to a category C is the category C op with the
same objects as C but HomC op (X, Y ) = HomC (Y, X) (and composition going in the
reverse direction). We have (C op )op = C.
1.1 Category 13
Exercise 1.3 Show that if C is any category, then there exists a category Mor(C)
defined as follows: an object is a morphism of C and a morphism between
f : X → Y and g : X ′ → Y ′ is a pair of morphisms in C, φ : X → X ′ and
ψ : Y → Y ′ , such that g ◦ φ = ψ ◦ f .
Exercise 1.4 Show that, if G is a monoid, then Mor(G) = G and that, if X is a

preordered set, then Mor(X) is the graph of the relation.
Exercise 1.5 Show that if C is a category and X ∈ C, then there exists a category
X\ C (of X-objects of C) defined as follows: an object of X\ C is a morphism
f : X → Y and a morphism from f : X → Y to g : X → Z is a morphism
h : Y → Z such that h ◦ f = g. Make explicit the category C/X := (X\ C op )op (of
objects of C over X). For example, in Top, an object Y over X is called a bundle
or a fibration.
1.1.2 Isomorphism
Definition 1.1.3 In a category C,
1. a section (resp. a retraction) of a morphism f : X → Y is a morphism
g : Y → X such that f ◦ g = IdY (resp. g ◦ f = IdX ):
!
g g
u /Y u /
X f g resp. 7X f
Y .
2. an isomorphism is a morphism that has at the same time a section and a

retraction. When there exists an isomorphism X → Y , one says that X and
Y are isomorphic and we write X ≃ Y . An isomorphism between X and
itself is called an automorphism.
A section for g is a retraction for f and conversely. A retraction in C is the same
thing as a section in C op and conversely. More generally, for any notion, there exists
a dual notion obtained by applying the same definition to the opposite category.
Examples 1. A retract A of a topological space X is a subspace such that the

inclusion map A ,→ X has a retraction.
2. A direct factor M ′ of a module M is a submodule such that the inclusion map
M ′ ,→ M has a retraction.
Proposition 1.1.4 If f is an isomorphism, then it has a unique section and a unique

retraction and they are the same.
Proof. If f ◦ g = idY and h ◦ f = idX , then
h = h ◦ idY = h ◦ f ◦ g = idX ◦ g = g. ■
The unique section/retraction of an isomorphism f is called its inverse and

denoted by f −1 .
Exercise 1.6 What is an isomorphism in Set, in Top, in Ab, etc. ? In X if X is a
preordered set ? In G if G is a monoid ?
1.1.3 Subcategory
Definition 1.1.5 A subcategory of a category C is the data of
1. a subset C ′ ⊂ C,
2. for all X, Y ∈ C ′ , a subset HomC ′ (X, Y ) ⊂ HomC (X, Y ), such that
(a) if X ∈ C ′ , then IdX ∈ EndC ′ (X) := HomC ′ (X, X),
(b) if X, Y, Z ∈ C ′ , f ∈ HomC ′ (X, Y ) and g ∈ HomC ′ (Y, Z), then g ◦ f ∈
HomC ′ (X, Z).
It is a full subcategory if actually
∀X, Y ∈ C ′ , HomC ′ (X, Y ) = HomC (X, Y ).
It is a wide subcategory if any object of C is in C ′ .
A subcategory becomes a category with the induced composition. A full subcate-

gory is uniquely determined by its objects.
Examples 1. Ab is a full subcategory of Gr which itself is a full subcategory of

Mon (which itself is a non-full subcategory of the category of semigroups or
magmas for example).
2. If X is a topological space, then Open(X) is a subcategory of Set which is not
full.
3. Top is not a subcategory of Set and neither is Ab (but see the notion of a
forgetful functor below).
4. If X is any topological space, then an espace étalé over X is a local homeomor-
phism X ′ → X. They form a full subcategory Et(X) ⊂ Top/X .
5. We may consider the subcategory ∆inj of ∆ with the same objects but order
preserving injective maps as morphism.
6. Conversely, we may consider ∆ as a subcategory of ∆+ which is defined by
adding [−1] = ∅.
1.2 Functor
1.2.1 Definition/Examples
Definition 1.2.1 1. A (covariant) functor F : C → C ′ between two categories
is the data for all X ∈ C of F (X) ∈ C ′ and for all f : X → Y of F (f ) :
F (X) → F (Y ), in such a way that we always have F (IdX ) = IdF (X) and
F (g ◦ f ) = F (g) ◦ F (f ).
2. If G : C ′ → C ′′ is another functor, then their composite is the functor G ◦ F
given by (G ◦ F )(X) = G(F (X)) and (G ◦ F )(f ) = G(F (f )).
We will denote by Hom(C, C ′ ) the set of all functors C → C ′ . We will often

describe the functors by their action on the objects and let the reader guess what
happens on the morphisms.
1.2 Functor 15
There always exists an identity functor IdC : C → C that doesn’t change anything.
A functor F : C → C ′ is called an isomorphism it there exists a functor G such that
G ◦ F = IdC and F ◦ G = IdC ′ (but this is not an interesting notion).
A functor F : C op → C ′ is also called a contravariant functor from C to C ′ . Any
functor F : C → C ′ provides a fucntor F op : C op → C ′op and this construction is
“functorial” : IdC op = Idop
C and (G ◦ F )
op
= Gop ◦ F op .
Examples 1. There exists a functor that forgets topology (and continuity)

Top → Set (underlying set). In the other direction, there exits two func-
tors X 7→ X disc and X 7→ X coarse that endow a set X with the discrete
topology (maximal) or the coarse topology (minimal).
2. There exists a functor that forgets the algebraic structure Ab → Set (underlying
set) and, in the other direction, a functor X 7→ Z · X (or Z(X) ) which sends a
set to the free free abelian group 2 generated by X:
( )
X
Z · X := nx x : nx ∈ Z, x ∈ X .
finite
3. There exists an inclusion functor Gr ,→ Mon and two functors G 7→ G× and
G 7→ Ggr := ⟨{xg }g∈G / {x−1

gh xg xh }g,h∈G ⟩
in the other direction.

4. Small categories and functors between them form a category that we shall
denote Cat and there exists a (contravariant) functor
Topop → Cat, X 7→ Ouv(X), u 7→ u−1 .
Exercise 1.7 What is a functor G → H between categories associated to monoïds

? What is a functor X → Y between categories associated to preordered sets ?
Exercise 1.8 What are the analog of the “free abelian group” functor X 7→ Z · X
for the categories Mon, Gr, Rng, G-Set, A-Mod and k-Alg.
Solution. For example, the free monoid generated by a set X is G := {g : n →

X, n ∈ N} endowed with

g(i) if i < n
gh : n + m → X, i 7→
h(i − n) otherwise.
In other words, X is the alphabet, G is the set of words in this alphabet and the
operation is concatenation. ■
Exercise 1.9 Show that, besides the inclusion functor Ab ,→ Gr, there exists an
abelianization functor G 7→ Gab = G/[G, G] in the other direction. Show however
hat the center of a group is not functorial: a group homomorphism φ : G → H
does not necessarily induce a morphism of abelian groups Z(G) → Z(H).
2
Some people write Z[X] instead of Z · X but this may be confused with a polynomial ring.
Exercise 1.10 Show that the categories Z-Mod and Ab are isomorphic. Same
thing with the categories Z-Alg and Rng, and, more generally, with k-Alg and a
full subcategory of k\ Rng (the image of k must be in the center).
If we are given two categories C and C ′ , then the projections C × C ′ on C and

C are functorial. The same holds for the obvious partial functors C ′ ,→ C × C ′ or
′
C ,→ C × C ′ associated to a fixed object X ∈ C or X ′ ∈ C ′ .

If C is any category, then there always exists a (bi-) functor
Hom : C op × C → Set, (X, Y ) 7→ Hom(X, Y )
that sends (f, g) to the map h 7→ g ◦ h ◦ f . If we compose with the partial functors,
we get the (fundamental) functors
hX : C → Set, Y 7→ Hom(X, Y )
and
hY : C op → Set, X 7→ Hom(X, Y ).
Exercise 1.11 Define the domain and codomain functors Mor(C) → C as well as
the forgetful functors X\ C → C and C/X → C.
Exercise 1.12 Let us denote by Op(C) ⊂ Mor(C) the full subcategory made of
morphisms with same domain and codomain (objects with operator). Show that,
if k is commutative ring, then Op(k-Mod) ≃ k[T ]-Mod.
1.2.2 Natural transformation

Definition 1.2.2 1. If F, G : C → C ′ are two functors, then a natural transforma-
tion α : F ⇒ G is the data for all X ∈ C of a morphism αX : F (X) → G(X)
such that, for all f : X → Y , we have αY ◦ F (f ) = G(f ) ◦ αX :
F (f )
F (X) / F (Y )
αY αX
G(f )
G(X) / G(Y )
We shall say natural isomorphism if all αX are isomorphisms.

2. If β : G ⇒ H is another natural transformation, then there composite
β ◦ α : F ⇒ H is the natural transformation defined by (β ◦ α)X = βX ◦ αX
for X ∈ C.
We shall denote by Hom(F, G) the set of all natural transformations form F to
G. If C is a small category, then the functors F : C → C ′ make a category Hom(C, C ′ )
with natural transformations as morphisms. One can check that the isomorphisms
are exactly the natural isomorphisms defined above.
1.2 Functor 17
Examples 1. We obtain a natural transformation by considering detA : GLn (A) →

A (between functors from commutative rings to groups). This is a natural
×
isomorphism for n = 1.
2. If C is any category, then there exists isomorphisms of categories Hom(0, C) ≃ 1,
Hom(1, C) ≃ C and Hom(2, C) ≃ Mor(C).
Definition 1.2.3 A functor F : C → C ′ is

1. faithful (resp. full, resp. fully faithful ) if for all X, Y ∈ C, the map
Hom(X, Y ) → Hom(F (X), F (Y )), f 7→ F (f )
is injective (resp. surjective, resp. bijective).

2. essentially surjective if for all X ′ ∈ C ′ , there exists X ∈ C such that
X ′ ≃ F (X).
3. an equivalence of categories if there exists G : C ′ → C such that IdC ′ ≃ F ◦ G
and G ◦ F ≃ IdC (and G is then called a quasi-inverse).
One also defines the essential image of a functor F as the set of all X ′ ∈ C ′
such that there exists X ∈ C with X ′ ≃ F (X). The functor F is then essentially
surjective when the essential image is equal to C ′ .
The inclusion of a (full) subcategory is a (fully) faithful functor. An isomorphism
of categories is an equivalence (but not conversely). We shall use the notation C ≃ C ′
for the wider notion of equivalence (and not merely isomorphism).
Exercise 1.13 Show that the forgetful functors Top → Set and Ab → Set are
faithful but not fully faithful.
Exercise 1.14 Show that there exists a fully faithful functor X 7→ X from pre-
ordered sets to small categories. What is the essential image ? Same questions
with a functor G 7→ G from monoids to small categories.
Exercise 1.15 Show that if F ≃ F ′ , then F is faithful (resp. full, resp. fully faithful,
essentially surjective, an equivalence) if and only if F ′ is.
Exercise 1.16 Show that a fully faithful functor is essentially injective: if F (X) ≃
F (Y ), then X ≃ Y .
Theorem 1.2.4 A functor is an equivalence of categories if and only if it is fully

faithful and essentially surjective.
Proof. In order to show that the condition is necessary, we first remark that our
functor F : C → C ′ will be essentially surjective since we will always have X ′ ≃ F (X)
with X := G(X ′ ) if G is a quasi-inverse for F . Then, we consider the following
sequence of maps
F
Hom(X, Y ) → Hom(F (X), F (Y ))
G F
→ Hom(G(F (X)), G(F (Y ))) → Hom(F (G(F (X))), F (G(F (Y ))))
Since both G ◦ F and F ◦ G are fully faithful, all the maps are necessary bijective
and F is therefore fully faithful.
In order to show that the condition is sufficient, we choose for all X ′ ∈ C ′ an
object X ∈ C and an isomorphism αX ′ : X ′ ≃ F (X). We set G(X ′ ) := X. Since F
is fully faithful, there exists for each f ′ : X ′ → Y ′ a unique f : G(X ′ ) → G(Y ′ ) such
that F (f ) = αY ′ ◦f ′ ◦αX
−1
′ and we set G(f ) = f . One easily checks that G is a functor
′
and we obtain by construction a natrual isomorphism α : IdC ′ ≃ F ◦ G. In particular,

if X ∈ C, there exists a natural isomorphism αF−1(X) : F (G(F (X))) ≃ F (X) and, since
F is fully faithful, there exists a unique isomorphism βX : (G ◦ F )(X) ≃ X such that
F (βX ) = αF−1(X) . One easily checks that β is indeed a natural transformation. ■
Exercise 1.17 Show that if X is a preordered set and Y denotes its ordered
quotient, then the categories X and Y are equivalent.
Exercise 1.18 Show that, if k is a commutative ring, then Mat(k) := N, endowed

with Hom(m, n) = Mn×m (k) and multiplication of matrices, is a small category.
Show that if k is a field, then Mat(k) is equivalent, but not isomorphic, to the
category of finite dimensional k-vector spaces (which is large).
1.2.3 Representable functor

Definition 1.2.5 An object X ∈ C, together with an element s ∈ F (X), is said to
be universal for a functor F : C → Set if
∀Y ∈ C, ∀t ∈ F (Y ), ∃!f : X → Y, F (f )(s) = t.
We shall also say that F is represented by X (and s).
We may also say that the couple (X ∈ C, s ∈ F (X)) is universal among all
couples (Y ∈ C, t ∈ F (Y )), or that s ∈ F (X) is universal for t ∈ F (Y ).
Examples 1. If k is a commutative ring then k[t] represents the forgetful functor

k−Alg → Set. More precisely, (k[t], t) is universal among all (A, a) where
A is a k-algebra and a ∈ A: there exists a unique morphism of k-algebras
φ : k[t] → A such that φ(t) = a.
2. Z is universal among all rings (F = ∅).
3. The inclusion Y ,→ X of a subspace into a topological space is universal for
continuous maps f : Z → X such that f (Z) ⊂ Y (contravariant F ).
Exercise 1.19 Show that the other usual forgetful functors with value in Set are
also representable.
Exercise 1.20 Show that the tensor product (M, N ) 7→ M ⊗A N (resp. (M, N ) 7→
M ⊗k N ) is universal for Z-bilinear (resp. k-bilinear) maps.
Exercise 1.21 Let k be a commutative ring and f1 , . . . , fr ∈ k[t1 , . . . , tn ]. Show

1.2 Functor 19
that the functor that sends a commutative k-algebra A to the set
S(A) := {(a1 , . . . , an ) ∈ An / f1 (a1 , . . . , an ) = · · · = fr (a1 , . . . , an ) = 0}
of all solutions with values in A, is representable.
Exercise 1.22 Show that, if F is represented by both X and X ′ , then X ≃ X ′ .

More precisely, show that if both (X, s) and (X ′ , s′ ) are universal for F , then
there exists a unique isomorphism f : X ≃ X ′ such that F (f )(s) = s′ .
Lemma 1.2.6 — Yoneda. If F : C → Set is any functor and X ∈ C, then there

exists a natural bijection
Hom(hX , F ) ≃ F (X), α 7→ αX (IdX ).
Proof. Given s ∈ F (X), if Y ∈ C and f : X → Y , then we set αY (f ) := F (f )(s).

This defines a map αY : Hom(X, Y ) → F (Y ) and we shall show that this is natural,
meaning that
αZ ◦ hX (g) = F (g) ◦ αY
if g : Y → Z. Indeed, we do have
(αZ ◦ hX (g))(f ) = αZ (hX (g)(f )) = αZ (g ◦ f ) = F (g ◦ f )(s)

= (F (g) ◦ F (f ))(s) = (F (g)(F (f )(s)) = F (g)(αY (f )) = (F (g) ◦ αY )(f ).
It only remains to check that we did define an inverse (exercise). ■
Proposition 1.2.7 A functor F : C → Set is represented by X ∈ C if and only if

hX ≃ F .
Proof. We may simply apply Yoneda lemma: the condition means that there exists
a natural transformation α : hX → F which is an isomorphism. This α corresponds
to some s ∈ F (X) and the condition exactly means that αY is always bijective since,
necessarily, F (f )(s) = αY (f ). ■
In other words, F is represented by X if and only if there exists a natural bijection
Hom(X, Y ) ≃ F (Y ). Note that, for a contravariant functor F : C op → Set, the
condition reads Hom(Y, X) ≃ F (Y ) or, equivalently, hX ≃ F .
Exercise 1.23 Show that if C is a small category, then the functor
C op → Hom(C, Set), X 7→ hX
is fully faithful. Deduce that C op (resp. C) is equivalent to the full subcategory

made of representable functors on C (resp. C op ).
Solution. If X, Y ∈ C, then Yoneda’s lemma implies that the map
Hom(hY , hX ) ≃ hX (Y ) = Hom(X, Y ), α 7→ αY (IdY )

is bijective. It is therefore sufficient to notice that, for f : X → Y , we have

hfY (IdY ) = IdY ◦ f = f . If we denote by R the full subcategory of representable
functors, then the induced functor C op → R is fully faithful and essentially surjective.
It is therefore an equivalence. The resp. assertion is obtained by duality. ■
1.3 Limit
1.3.1 Diagrams, cones and limits
Definition 1.3.1 Let I be a small category and C any category. Then, a commutative
diagram on I in C is a functor D : I → C.
A commutative diagram on I in C is therefore the data of an object Xi for all
i ∈ I and a morphism fα : Xi → Xj for all α : i → j satisfying fβ◦α = fβ ◦ fα and
fIdi = IdXi :
fα
Xi / Xj
fβ◦α
~ fβ
Xk
We shall denote such a diagram as (fα : Xi → Xj ) or (Xi , fα ), and the category of

all commutative diagrams on I in C by C I := Hom(I, C).
Definition 1.3.2 A simplicial (resp. semi-simplicial ) object is a diagram on ∆op
(resp. ∆op
inj ). A (semi-) cosimplicial object is a (semi-) simplicial object in C . It
op
is said to be augmented if we actually use ∆+ instead of ∆
Exercise 1.24 Show that giving a semi-simplicial object is equivalent to giving a

sequence of morphisms
dn+1 dn
.. / .. / / d10
0 0
/ /
Xn+1 . / Xn . / Xn−1 X2 / X1 / X0
dn d11
dn+1
n+1 n
such that
∀n ∈ N, ∀0 ≤ i < j ≤ n + 1, dni ◦ dn+1

j = dnj−1 ◦ dn+1
i .
Analog for simplicial objects.
Example There exists a cosimplicial object in Top sending [n] to the standard
topological simplex
( n
)
X
∆n := (x0 , . . . , xn ) ∈ Rn+1
≥0 , xi = 1
i=0
and u : [n] → [m] to the unique linear map sending ei to eu(i) if (e0 , . . . , en ) denotes
the usual basis.
1.3 Limit 21
By composition, any functor λ : I → J between small categories will provide a

functor λ∗ : C J → C I between diagrams (and this is functorial). As a particular case,
the unique functor I → 1 induces the constant diagram functor
C ≃ C1 → CI , X 7→ X.
Definition 1.3.3 A cone for a diagram D in C is a morphism X → D with X ∈ C.
In more down to earth terms, a cone for (Xi , fα ) is a family of morphisms

(pi : X → Xi ) such that for all α : i → j, we have pj = fα ◦ pi . It is said to be finite
when I is finite. The dual notion is that of a cocone.
Definition 1.3.4 A limit X of a commutative diagram D on I in C is a universal
cone.
In other words, X is a limit for D if and only if X is a cone for D and, given any
cone Y for D, there exists a unique morphism f : Y → X making commutative the
diagram
f
Y / X
~
D.
In down to earth terms, a commutative diagram (Xi , fα ) has X as a limit if and only
if we are given for all i ∈ I a morphism pi : X → Xi such that for all α : i → j, we
have pj = fα ◦ pi with the following universal property: if we are given some Y ∈ C
endowed for all i ∈ I with a morphism gi : Y → Xi such that for all u : i → j, we
have gj = fα ◦ gi , then there exists a unique morphism g : Y → X such that for all
i ∈ I, we have gi = pi ◦ g :
gi
*
pi 5 Xi
g
Y / X fα
)
pj
4 Xj
gj
Alternatively, the definition says that the composite functor hD ◦ is representable

by X in C op . Equivalently, there exists a natural isomorphism for Y ∈ C:
HomC I (Y , D) ≃ HomC (Y, X).
We call a limit finite when I is finite.
A limit is unique up to a unique isomorphism and we may sometimes say the
limit and denote it by X = lim D. A limit X of a diagram D in C op is also called
←−
a colimit in C and denoted by X = lim D. A limit (resp. colimit) is also called an
−→
inverse (resp. a direct) limit, a projective (resp. an inductive) limit or a left (resp. a
right) limit. We have the following formulas:
Hom(Y , D) ≃ Hom(Y, lim D) et Hom(lim D, Y ) ≃ Hom(D, Y ).
←− −→
When we write a limit or a colimit, we implicitly assume that it exists.
Exercise 1.25 Make the notion of a colimit explicit.
1.3.2 Specific limits

Definition 1.3.5 A limit of the empty diagram 0 → C is called a final object of C
and denoted by 1C . Dually, we get the notion of an initial object 0C .
if X ∈ C, there exists a unique morphism X → 1C (resp. 0C → X).
Examples 1. In Set, the initial object is ∅ and 1 is a final object (defined up to

a unique bijection).
2. Same thing in Top.
3. In Ab, {0} is both a final and an initial object.
Definition 1.3.6 A limit of a family (Xi )i∈I of objects of C (with no morphism

apart from identites) is called a product and denoted
` by i∈I Xi . Dually, a colimit
Q
of (Xi )i∈I is called a coproduct and and denoted i∈I Xi . When all Xi are equal
to the same X, we shall say power (resp. copower ) and we write X I (resp. X (I)
or I · X).
Note that a final (resp. initial) object is nothing else but the empty product (resp.
coproduct).
Examples 1. In Set, the cartesian product is a product and the disjoint union
is a coproduct.
2. In Top, this is the same thing with the coarser (resp. finer) topolofy making
the projections (resp. injections) continuous.
3. In Ab, the cartesian product is a product and the direct sum is a coproduct3
(product equals coproduct when I is finite).
f1 f2
Definition 1.3.7 A limit X of a diagram (X1 → X0 ← X2 ) is called a fibered
product of X1 and X2 over X0 and denoted by X = X1 ×X0 X2 . We shall then
also say that the diagram
p1
X / X1
p2 f1
f2
X2 / X0
is cartesian or that p2 is the pullback of f1 along f2 (and p1 is the pullback of

f2 along f1 ). Dually, there exists the notions of a fibered coproduct denoted by
X1 ⊔X0 X2 , a cocartesian square and a pushout.
Note that a product (resp. coproduct) of two objects is nothing but a fibered
product (resp. fibered coproduct) over a final (resp. initial) object (if it exists).
3
Be careful that what is called free product is a coproduct in the category of (non abelian)
groups.
1.3 Limit 23
Examples 1. In Set, we have
X1 ×X0 X2 = {(x1 , x2 ) / f1 (x1 ) = f2 (x2 )} ⊂ X1 × X2
and
X1 ⊔X0 X2 = (X1 ⊔ X2 )/ ∼
where ∼ is the equivalence relation generated by f1 (x0 ) ∼ f2 (x0 ) when x0 ∈ X0 .

2. In Top, this is the same thing with the induced (resp. quotient) topology.
3. In Ab this is the kernel (resp. cokernel) of the canonical map from (resp. to)
the direct sum.
Definition 1.3.8 A limit X of a pair Y ⇒ Z is called a kernel (or equalizer ) and

denoted by X = ker(f, g). We shall also say that the sequence
f
/Y /
X / Z
g
is left exact. Dually, there exists the notion of a cokernel (or coequalizer ) coker (f, g)
and right exact sequencea .
a
We should say exact sequence and coexact sequence.
Examples 1. In Set, we have ker(f, g) = {y ∈ Y / f (y) = g(y)} and coker (f, g) =

Z/ ∼ where ∼ is the equivalence relation generated by f (y) ∼ g(y) for y ∈ Y .
2. Same thing in Top.
3. In Ab, ker(f ) := ker(f, 0) is the usual kernel and we have ker(f, g) = ker(f −g),
(and dual) .
Definition 1.3.9 When a commutative diagram of the form
Y Y
i

Y
i / X
is cartesian, then Y is called a monomorphism and we shall write i : Y ,→ X. It

is called a regular monomorphism if it is a kernel. The dual notion is that of a
(regular) epimorphism and we shall then write X ↠ Y .
Exercise 1.26 1. Write down the definition of an epimorphism.

2. Show that X → Y is a monomorphism (resp. an epimorphism) if and only
if for all Z ∈ C, the map Hom(Z, X) → Hom(Z, X) (resp. Hom(Y, Z) →
Hom(X, Z)) induced by composition is injective.
3. Show that a kernel (resp. cokernel) is always a monomorphism (resp. an
epimorphism).
Examples 1. A morphism in Set (resp. Ab, resp. Top) is a monomorphism/epimorphism

if and only if it is injective/surjective.
2. A regular monomorphism/epimorphism in Top is a homeomorphism with a

subspace/a quotient map.
3. The inclusion map Q ,→ R is at the same time a monomorphism and an
epimorphism (but not an isomorphism) in the category of Hausdorff topological
spaces.
4. Same thing for the inclusion map Z ,→ Q in the category of rings.
Definition 1.3.10 If Y ,→ X is a monomorphism, we may write Y ⊂ X and call Y

a subobject a of X. When Y, Z ⊂ X, their intersection is Y ∩ Z := Y ×X Z ⊂ X
(if it exists). If f : X ′ → X is any morphism and Y ⊂ X, then its inverse image
is f −1 (Y ) := Y ×X X ′ ⊂ X ′ (if it exists).
a
More precisely, a subobject is an equivalence class of such.
Exercise 1.27 Make explicit some classical limits and colimits in Mon, Gr, Rng,
G-Set, A-Mod and k-Alg.
Exercise 1.28 Show that the fibered coproduct in the category of commutative
rings is the tensor product.
Exercise 1.29 Show that in a poset, the limit (resp. colimit) is the greatest lower
bound or meet (resp. least upper bound or join). What about cone and cocone ?
Make explicit the case of a finite ordinal as well as the set of open subsets of a
topological space.
Exercise 1.30 Does the category Cat have a final object? an initial object ? finite
products ? Make them explicit. Show that if C is a small category, then Op(C) is
the kernel of the domain and codomain functors Mor(C) → C in Cat.
Exercise 1.31 Show that a morphism f : Y → X is a monomorphism (resp. an

epimorphism) if and only if the induced functor C/Y → C/X (resp. X\ C → Y \ C) is
faithful.
Definition 1.3.11 The image im(f ) of a morphism f : X → Y is a subobject I of Y
which is universal for factorizations f : X → I ,→ Y (if it exists). The dual notion
is that of a coimage coim(f ). A morphism is said to be strict if im(f ) = coim(f ).
Exercise 1.32 Show that any morphism is strict in Set or Ab but not in Top.
1.3.3 Constructions of limits

Exercise 1.33 We give ourselves a commutative diagram (Xi , fα ). Assume that
X ′ := i Xi and X ′′ := α:i→j Xj both exist. Denote by p (resp. f ) the morphisms
Q Q
X ′ → X ′′ induced by the projections onto the codomain (resp. the composition of
the projection onto the domain and fα ). Show that X = lim(Xi , fα ) if and only if
←−
there exists a left exact sequence
p
/ X′ /
X / X ′′ .
f
1.4 Applications 25
Exercise 1.34 Show that if (fα : Xi → Xj ) is a commutative diagram of sets, then
lim Xi = Xi / ∼ where ∼ denotes the equivalence relation generated by xi ∼ xj
`
−→
when fα (xi ) = xj .
Exercise 1.35 Show that, if we are given two morphisms f, g : X → Y in a

category C, then the commutative diagram
ker(f, g) /X

Y / Y × Y.
is cartesian. More precisely, show that if Y × Y exists, then there exists such a
fibered product if and only if ker(f, g) exists in which case they coincide.
Exercise 1.36 Let C be a category.

1. Show that, if all (finite) products and all kernels exist, then all (finite) limits
exist.
2. Show that if all fibered products exist and there is a final object, then all
finite limits exist.
Analogues for colimits.
When all (finite) limits exist, then C is said to be (finitely) complete. The dual
terminology is cocomplete.
Exercise 1.37 Show that all limits and colimits do exist in Set, Top, Ab, etc.
1.4 Applications
1.4.1 Filtered colimit
Definition 1.4.1 A small category I is said to be filtered if it has all finite cocones.
A filtered diagram is a diagram I → C with I filtered. A filtered colimit is a colimit
of a filtered diagram.
Exercise 1.38 Show that a poset is directed if and only if it is filtered (as a
category).
There exists a partial converse (we may always replace a filtered category with a
directed set):
Proposition 1.4.2 If I is a filtered category, then there exists a directed set J and a
functor u : J → I such that, for all diagram D : I → C, if lim(D ◦ u) exists, then
−→
lim D exists and lim(D ◦ u) ≃ lim D.
−→ −→ −→
Proof. To do ■
Exercise 1.39 Show that a small category I is filtered if and only if

1. I ̸= ∅,
2. ∀i, j ∈ I, ∃i → k, j → k,
3. ∀u, v : i → j, ∃c : j → k / c ◦ u = c ◦ v.
Exercise 1.40 Show that a category with filtered colimits and finite colimits (resp.
finite coproducts) has all colimits (resp. coproducts).
Definition 1.4.3 An ind-object “ lim Xi ” of a category C is a filtered diagram (Xi )i∈I .

−→
They form a category Ind(C) with
Hom(“ lim Xi ”, “ lim Yj ”) = lim lim Hom(Xi , Yj ).

−→ −→ ←− −→
j∈J i∈I
The dual notion is that of a pro-object “ lim Xi ”and they form a category Pro(C) :=
←−
Ind(C op )op .
The notion of ind-object lives somehow between the notion of a diagram and the
notion of a limit. One can show that we get an equivalent category by considering
only directed sets instead of filtered categories (difficult).
Example The category of profinite sets (pro-objects of the category of finite sets)
is equivalent to the category of compact Hausdorff totally disconnected spaces.
Exercise 1.41 Show that the obvious functor C ,→ Ind(C) is fully faithful.
1.4.2 Preservation of limit

Any functor F : C → D provides by composition a functor
F I : C I → DI , D 7→ F (D) := F ◦ D
We shall usually simply write F instead of F I so that F (Xi , fα ) = (F (Xi ), F (fα )).
Definition 1.4.4 If D is a commutative diagram in C, then a functor F : C → D is
said to preserve (or commute with) the limit of D, if
F (lim D) ≃ lim F (D),

←− ←−
Of course, it is assumed here that the limit of D exists and it implies that the
limit of F (D) also exists. Also, there exists an obvious analogue for colimits. Be
careful that a (contravariant) functor F : C op → D preserves a limit when it turns a
colimit in C into limit in D.
Exercise 1.42 Let F : C → D be a functor.
1. Show that if F preserves all (finite) products and all kernels, then F preserves
all (finite) limits.
2. Show that if F preserves all fibered products and the final object, then F
preserves all finite limits.
Analogues for colimits.
Exercise 1.43 Show that a functor that preserves filtered colimits and finite
colimits (resp. finite coproducts) preserves all colimits (resp. coproducts).
1.4 Applications 27
Exercise 1.44 Show that the forgetful functor Top → Set preserves all limits and
colimits and that the functor Ab → Set preserves all limits (but not colimits).
Exercise 1.45 Show that, if F preserves fibered products, then F preserves

monomorphisms (and dual).
Proposition 1.4.5 A representable functor F : C → Set preserves all limits.
Proof. We may assume that F = hX with X ∈ C. It is then sufficient to prove that

if D is a commutative diagram in C, we have a sequence of bijections
hX (lim D) ≃ Hom(X, lim D) ≃ Hom(X, D)

←− ←−
≃ Hom({0}, hX (D)) ≃ Hom({0}, lim hX (D)) ≃ lim hX (D).

←− ←−
Only the middle one needs to be checked by hand: if we write D =: (Xi , fα ), then,
giving a morphism {0} → hX (D) is equivalent to give a compatible family of maps
{0} → Hom(X, Xi ), or in other words, to give for each i ∈ I, a morphism gi : X → Xi
such that fα ◦ gi = gj , or finaly a morphism X → D. ■
There is no dual statement here and the notion of a limit plays a special role.
When applied to the functor Hom, we have the following formulas
Hom(X, lim Yi ) ≃ lim Hom(X, Yi ) et Hom(lim Xi , Y ) ≃ lim Hom(Xi , Y ).

←− ←− −→ ←−
Definition 1.4.6 A functor is said to be left exact (resp. right exact) if it preserves
all finite limits (resp. colimits). Il is said to be exact if it is both left and right
exacta .
a
We should say exact, coexact and biexact respectively.
Examples 1. The forgetful functor Top → Set is exact, as well as the functor
Set → Top that endows a set with the discrete topology, but the functor
Set → Top that endows a set with the coarse topology is only left exact.
2. The forgetful functor Ab → Set is left exact but not right exact, and the free
abelian group functor Set → Ab is right exact but not left exact.
Exercise 1.46 Show that the obvious functor C ,→ Ind(C) is exact and preserves
all limits.
1.4.3 Projective/Injective
Definition 1.4.7 An object X of a category C is said to be projective if hX preserves
epimorphisms. The dual notion is that of an injective object.
It means that, if Z ↠ Y is an epimorphism (resp. Y ,→ Z is a monomorphism)

in C, then following map is surjective:
Hom(X, Z) ↠ Hom(X, Y ) (resp. Hom(Z, X) ↠ Hom(Y, X)).

In other words, any diagram
>Z ZO
?
X / Y (resp. Y / X )
can be completed with the dotted arrow. When X is projective (resp. injective),
any epimorphism Y ↠ X (resp. monomorphism X ,→ Y ) has a section (resp. a
retraction). Also, if X → Y has a retraction and Y is projective, then X also is
projective (analog for injective).
Examples 1. In Set, all objects are projective and injective.
2. In Ab, projective objects are free abelian groups and injective objects are
divisible groups (for example Q and Q/Z).
3. In R−Mod, projective objects are direct factors of free R-modules (Z/2 is
projective – but not free – over Z/6).
4. We shall show that the projective objects of the category of compact Haus-
dorff spaces are the Stonean (meaning extremally disconnected) spaces (or
equivalently the retracts of free compact Hausdorff spaces).
Exercise 1.47 Show that a coproduct of projectives is projective (analog for

injective).
Exercise 1.48 Assume C admits fibred products and that epimorphisms are uni-
versal (see definition 3.2.11 below). Show that X is projective if and only if any
epimorphism Y ↠ X has a section.
Definition 1.4.8 A category C has enough projectives if, given any X ∈ C, there
exists a projective Y and an epimorphism Y ↠ X (dual notion : enough injective).
Examples 1. The category Set has enough projectives and injectives.

2. The category R−Mod has enough projectives and injectives.
3. We will show that the category of compact Hausdorff spaces has enough
projectives.
Definition 1.4.9 An object X of a category C is said to be finitely presented (or

compact) if hX preserves filtered colimits.
In other words,
lim Hom(X, Yi ) ≃ Hom(X, lim Yi )
−→ −→
when (Yi ) is filtered. Or, in more down to earth terms, any morphism X → lim Yi
−→
factors through some Yi .
Examples 1. A set is finitely presented if and only if it is finite.
2. A topological space is finitely presented if and only if it is finite discrete.
3. An abelian group is finitely presented (resp. finitely presented projective) if
and only if it is finitely generated (resp. free of finite rank).
4. A topological space is compact (in the usual sense) if and only if X is a finitely
presented object of Open(X).
1.4 Applications 29
1.4.4 Algebraic structure

Definition 1.4.10 Let C be a category with finite products (a cartesian category).
A monoid of C is an object G endowed with a multiplication map µ : G × G → G
and a unit map ϵ : 1 → G making commutative the following diagrams:
µ×IdG ϵ×IdG
G×G×G / G×G and G / G×G
IdG ×µ µ IdG ×ϵ µ
µ µ
G×G / G G×G / G.
A morphism of monoids G → G′ of C is a morphism f : G → G′ in C making the

following commutative:
f ×f ′ f
G×G / G′ × G′ and G_ / ′
?G
ϵ ϵ′
µ µ′
f

G / G′ 1
It is a group if there exists an inversion map ι : G → G making commutative
(ι,IdG )
G / G×G
$
(IdG ,ι) 1 µ
ϵ
µ #
G×G / G.
Il is abelian if µ ◦ σ = µ if σ denotes the map that exchanges factors in G × G.
Monoids (resp. groups, resp. abelian groups) of C make a category Mon(C) (resp.
Gr(C), resp. Ab(C)). We shall concentrate on abelian groups.
Example An abelian group of the category Set is nothing but a usual abelian
group. An abelian group of Top is a topological abelian group (with continuous
multiplication and continuous inversion).
Exercise 1.49 A bialgebra is a monoid of the category opposite to the category of

k-algebras. Show that k[t] (resp. k[t, 1/t]) endowed with t 7→ t ⊗ 1 + 1 ⊗ t (resp.
t 7→ t ⊗ t) is a bialgebra.
Exercise 1.50 Define the notion of a ring A in a category C with finite products
as well as the notion of a G-object, an A-modules or a k-algebra. We shall denote
by Rng(C), G-Set(C), A-Mod(C) and k-Alg(C) these categories.
Exercise 1.51 Show that, if all limits exist in C, then the same holds in Ab(C)
and they are preserved by the obvious forgetful functor Ab(C) → C.
Exercise 1.52 Let F : C → C ′ be a functor between categories with finite products.
Show that
1. if F preserves finite products, then F induces a functor Ab(F ) : Ab(C) →
Ab(C ′ ),
2. if F preserves all limits, so does Ab(F ).
1.4.5 Localization
Definition 1.4.11 The localization of a small category C with respect to a set of
morphisms W is a categorya ho(C) which is universal for functors C → D sending
W to isomorphisms in D.
a
ho stands for homotopy category.
It means that there exists a functor γ : C → ho(C) sending W to isomorphisms

such that, given any category D, the functor
γ −1 : Hom(ho(C), D) → Hom(C, D)
induces an isomorphism with the full subcategory of functors sending W is isomor-
phisms.
Proposition 1.4.12 The localization ho(C) of C with respect to W always exists.
Proof. (Sketch) We may assume that W contains all isomorphisms in C. Then, the
objects of ho(C) are the objects of C and morphisms are finite chains
W W W
X = X0 ← X1 → X2 ← · · · → Xn−1 ← Xn → Xn+1 = Y
up to equivalence. ■
Definition 1.4.13 A category C admits right calculus of fractions with respect to a

set of morphisms W if
1. W contains all identities and is stable under composition,
2. given any f : X → Y in C and φ : Y ′ → Y in W , there always exists
f ′ : X ′ → Y ′ and φ′ : X ′ → X in W with φ ◦ f ′ = f ◦ φ′ ,
3. given any f, g : X → Y ′ in C and φ : Y ′ → Y in W such that φ ◦ f = φ ◦ g,
there exists φ′ : X ′ → X in W such that f ◦ φ′ = g ◦ φ′ .
Proposition 1.4.14 If a small category C admits right calculus of fraction with

respect to W , then ho(C) is the category having the same objects as C and
Homho(C) (X, Y ) = lim HomC (X ′ , Y ).

−→
X ′ →X∈W
Proof. (sketch) By definition, morphisms and composition are described, up to

equivalence, by the following diagram
X ′′
W

)
X′ Y′
W W
)/
X /) Y Z.
1.5 Adjoint 31
It is then a matter of checking the various properties. ■
Exercise 1.53 Show that if C admits right calculus of fraction with respect to W ,
then the functor Q : C → ho(C) is exact.
1.5 Adjoint
1.5.1 Definition
Definition 1.5.1 A functor F : C → C ′ is adjoint to a functor G : C ′ → C if there
exists a natural isomorphism
∀X ∈ C, X ′ ∈ C ′ , Hom(F (X), X ′ ) ≃ Hom(X, G(X ′ )).
The dual notion is that of a coadjoint so that G is coadjoint to F if and only if

F is adjoint to G. We may then write F ⊣ G or F : C ⇆ C ′ : G but usually simply
make explicit the natural isomorphism.
Examples 1. The forgetful functor Top → Set has both an adjoint (discrete
topology) and a coadjoint (coarse topology):
C(X disc , Y ) ≃ F(X, Y ) et F(X, Y ) ≃ C(X, Y coarse ).
2. The forgetful functor Ab → Set has an adjoint (free abelian group):
Hom(Z · X, M ) ≃ F(X, M ).
Exercise 1.54 Write explicitly what it means for ΦX,X ′ : Hom(F (X), X ′ ) ≃
Hom(X, G(X ′ )) and its inverse to be natural.
Solution. Given g : Y → X, g ′ : X ′ → Y ′ , then

∀f : F (X) → X ′ , ΦY,Y ′ (g ′ ◦ f ◦ F (g)) = G(g ′ ) ◦ ΦX,X ′ (f ) ◦ g and
∀f ′ : X ′ → G(X), Φ−1 ′ ′ −1
Y,Y ′ (G(g ) ◦ f ◦ g) = g ◦ ΦX,X ′ (f ) ◦ F (g). ■
Exercise 1.55 Show that most forgetful and inclusion functors we have already
met have an adjoint (and sometimes a coadjoint) and make them explicit.
Exercise 1.56 Show that the functor X 7→ X × Y from Set to itself is adjoint to
the functor Z 7→ F(Y, Z):
F(X × Y, Z) ≃ F(X, F(Y, Z)).
Analogue for Cat, Ab ?
Exercise 1.57 Show that if both F1 ⊣ G and F2 ⊣ G, then F1 ≃ F2 (and dual).
Solution. Both F1 (X) and F2 (X) represent the same functor X ′ 7→ Hom(X, G(X ′ ))
and there exists therefore a isomorphism F1 (X) ≃ F2 (X) which is easily seen to be
natural. ■
Exercise 1.58 Show that if F1 : C ⇆ D : G1 and F2 : D ⇆ E : G2 then,
F2 ◦ F1 ⊣ G1 ◦ G2 .
1.5.2 Unit and counit

Proposition 1.5.2 A functor F : C → C ′ is adjoint to G if and only if there
exists α : IdC ⇒ G ◦ F and β : F ◦ G ⇒ IdC ′ such that βF ◦ F (α) = IdF and
G(β) ◦ αG = idG .
Proof. Assume first that there exists a natural isomorphism
ΦX,X ′ : Hom(F (X), X ′ ) ≃ Hom(X, G(X ′ )).
We may then set αX := ΦX,F (X) (IdF (X) ) and, dually, βX ′ := Φ−1
G(X ′ ),X ′ (IdG(X ) ). Then,
′
we have
βF (X) ◦ F (αX ) = Φ−1

G(F (X)),F (X) (IdG(F (X)) ) ◦ F (ΦX,F (X) (IdF (X) ))
−1
= ΦX,F (X) (ΦX,F (X) (IdF (X) ))
= IdF (X)
and symetrically. Conversely, given f : F (X) → X ′ , we set

α G(f )
X
ΦX,X ′ (f ) : X −→ G(F (X)) −→ G(X ′ )
and define dually for f ′ : X → G(X ′ )

F (f ′ ) β ′
ΨX,X ′ (f ′ ) : F (X) −→ F (G(X ′ )) −→
X
X ′.
Then, we have
(ΨX,X ′ ◦ ΦX,X ′ )(f ) = βX ′ ◦ F (G(f )) ◦ F (αX ) = f ◦ βF (X) ◦ F (αX ) = f
and symetrically. ■
Definition 1.5.3 The morphisms α and β are called unit and counit (or else
adjunction morphisms).
Exercise 1.59 Assume F ⊣ G with unit α and counit β. Show that G is faith-
ful (resp. fully faithful) if and only if αX is always a monomorphism (resp. an
isomorphism). Analogue for G ?
Solution. Let us consider composite map
Hom(Y, X) → Hom(F (Y ), F (X)) ≃ Hom(Y, G(F (X)))
where the first one is f 7→ F (f ) and the second one is the adjunction ΦY,F (X) . We
have
ΦY,F (X) (F (f )) = αY ◦ G(F (f )) = f ◦ αX = hαX (f ).

1.5 Adjoint 33
Thus we see that G is faithful (resp. fully faithful) if and only hαX injective (resp.
bijective) for all X and all Y . This means that αX is a monomorphism (resp. an
isomorphism) for all X.
Now, we have Gop ⊣ F op and the unit for this adjunction is β op . Moreover, βX op
′
is a monomorphism (resp. an isomorphism) if and only if βX ′ is an epimorphism

(resp. an isomorphism). Therefore, G is faithful (resp. fully faithful) if and only if
Gop is faithful (resp. fully faithful) if and only if βX ′ is an epimorphism (resp. an
isomorphism) for all X ′ . ■
Exercise 1.60 Describe the adjunction morphisms in all the examples studied so
far. Deduce in each case faithfulness or full faithfulness of the functors.
Exercise 1.61 Show that, if a small category C has copowers, then all representable
functors F on C have an adjoint. Recover the fact that F preserves all limits.
Solution. We may assume that F = hX , consider the functor
Set → C, I 7→ X (I) ,
define unit I → Hom(X, X (I) ) and counit X (Hom(X,Y )) → Y and check the properties.
■
1.5.3 Adjoint and limit

Proposition 1.5.4 In a category C, all limits on I exist if and only if the functor
X 7→ X has a coadjoint which is then given by D 7→ lim D.
←−
Proof. We have indeed a natural isomorphism in X given by
Hom(X, D) ≃ Hom(X, lim D)

←−
and it is sufficient to show that it is also natural in D. This immediately follows
from the universal property of limits. ■
Exercise 1.62 Show that any adjunction F ⊣ G extends to an adjunction on

diagrams on I:
Hom(F (D), E) ≃ Hom(D, G(E)).
Theorem 1.5.5 A functor that admits an adjoint (resp. a coadjoint) preserves all
limits (resp. colimits).
Proof. Assume that F : C → D is adjoint to a functor G and that we are given a

diagram D in D. If X ∈ C, then there exists a sequence of natural isomorphisms
Hom(X, G(lim D)) ≃ Hom(F (X), lim D) ≃ Hom(F (X), D)

←− ←−
= Hom(F (X), D) ≃ Hom(X, G(D)) ≃ Hom(X, lim G(D)).
←−
It follows that G(lim D) and lim G(D) represent the same functor and are therefore
←− ←−
naturally isomorphic. ■
As a consequence, we see that limits (resp. colimits) commute with limits (resp.
colimits).
There exists a partial converse which is called Freyd adjunction theorem:
Theorem 1.5.6 If D is a small complete category, then a functor G : D → C which
preserves all limits has an adjoint (and dual).
Proof. (Sketch) It is sufficient to set
F X := lim Y. ■
←−
X→GY
One can replace the smallness condition on D by the solution-set condition: given
any X in C, there exists a set of morphisms X → GYi such that any morphism
X → GY factors through some GYi .
Example As we already noticed, the forgetful functor Top → Set preserves all
limits and all colimits and the forgetful functor Ab → Set preserves all limits.
Exercise 1.63 Assume F is an adjoint to a fully faithful G. Show that if D′ is a

diagram in C ′ and X = lim G(D′ ), then X ′ := F (X) = lim D′ and X ≃ G(X ′ ).
←− ←−
Exercise 1.64 Show that the forgetful functors on Mon, Gr, Rng, G-Set, A-Mod
and k-Alg preserve all limits.
Exercise 1.65 Study the exactness of various forgetful and inclusion functors as
well as their adjoints.
Exercise 1.66 Show that if all limits on I exist in C, then all limits on I exist in
C J and that, if D ∈ C I×J (≃ (C J )I ), then
!
∀j ∈ J, lim D = lim Dj .
←− ←−
i j i
Analogue for colimits.
Proposition 1.5.7 Filtered colimits of sets are exact.
Proof. The point is to show that if I is a directed set, then the functor
lim : SetI → Set

−→
is exact. It is sufficient to treat the cases of a final object, a product of two objects
or a kernel. We shall also use the fact that colimits of diagrams are computed
term-wise (see exercice 1.66). Clearly, lim {0} = {0}. We consider now two families
−→
of morphisms (uij : Xi → Xj ) and (vij : Yi → Yj ) defined for i < j with u`jk ◦uij = uik
and vjk ◦ vij = vik when i < j < k. Recall that one can write lim Xi = Xi / ∼ and
−→
we will denote by xi the class of xi ∈ Xi (and we will do the same for other colimits
on I). Recall also that, for all i < j, we have uij (xi ) = xi and that we may therefore
1.5 Adjoint 35
always replace any xi with some xj ∈ Xj when j > i without changing its class. By
considering colimits on the projections, we have the following morphisms
lim(Xi × Yi ) → lim Xi and lim(X × Yi ) → lim Yi .

−→ −→ −→ i −→
The universal property of products provides us with a map
lim(Xi × Yi ) → lim Xi × lim Yi , (xi , yi ) 7→ (xi , y i ).

−→ −→ −→
and we have to show that it is bijective. Assume that (xi , y i ) = (x′j , y ′j ). Since
I is directed, there exists k ≥ i, j and we may assume that i = j = k so that
(xi , yi ) = (x′i , yi′ ). This shows injectivity. Assume now given xi ∈ Xi and xj ∈ Yj .
After replacement of i and j by k ≥ i, j, we may assume that i = j = k. Surjectivity
follows. This takes care of the product of two sets and we will treat the case of
kernels in the same way. We give ourselves two families of maps (fi , gi : Xi → Yi )
compatible with uij ’s and vij ’s. The universal property of kernels (and the explicit
description of colimits of sets) provides us with an inclusion
lim ker(fi , gi ) ⊂ ker(lim fi , lim gi )

−→ −→ −→
We need to show that this is an equality. We give ourselves xi ∈ lim Xi such that
−→
fi (xi ) = gi (xi ). As usual, there exists k ≥ i, j and we may thererefore assume that
i = j = k. ■
Concretely, the proposition states that, if I is a filtered category, J is a finite
category and (Xi,j ) is a diagram of sets based on I × J, then we have
lim lim Xij ≃ lim lim Xij .

−→ ←− ←− −→
i j j i
Exercise 1.67 Show that filtered colimits are exact in Ab, etc. and that they
commute with the forgetful functors (one may show the second assertion first).
Exercise 1.68 Show that Ab satisfies AB6 which means that filtered colimits
commute with products:
Y Y
lim Mij ≃ lim Mij
−→ Q−→
j∈J ij ∈Ij j∈J Ij j∈J
when each Ij is filtered.
1.5.4 Reflective subcategory

Definition 1.5.8 A full subcategory C ′ ⊂ C is said to be reflective if the inclusion
functor has an adjoint, called reflection. The dual terminology is coreflective.
Examples 1. The category of abelian groups is a reflective subcategory of the

category of all groups.
2. The category of groups is both a reflective and a corefllective subcategory of
the category of all monoids.
3. We shall show that the category of compact Hausdorff spaces is a reflective

subcategory of the category of all topological spaces.
4. The category of sets is isomorphic to the reflective (resp. coreflective) subcate-
gory of discrete (resp. chaotic) topological spaces.
Exercise 1.69 Show that if C ′ is a reflective subcategory of C, then any diagram

D′ in C ′ that has a limit (resp. a colimit) in C has a limit (resp. a colimit) in C ′ .
Show also that C ′ is stable under limits that exist in C. Finally, show that, if a
colimit in C of objects of C ′ is an object of C ′ , then this is a colimit in C ′ .
Exercise 1.70 Show that a functor F : C → C ′ is a reflection if and only if

there exists a natural morphism αX : X → F (X) for X ∈ C such that αX is an
isomorphism when X ∈ C ′ .
Solution. The subcategory C ′ is reflective with reflection F if and only if there

exists a natural morphism αX : X → F (X) for X ∈ C and a natural isomorphism
βX : F (X) ≃ X when X ∈ C ′ such that βF (X) ◦ F (αX ) = IdF (X) when X ∈ C
and βX ◦ αX = idX when X ∈ C ′ . This is clearly equivalent to our condition with
βX := αX −1
when X ∈ C ′ . ■
Exercise 1.71 Show that if C ′ is a coreflective subcategory of C, then Mon(C ′ ) is

a full subcategory of Mon(C). Same thing with Gr, Ab, etc.
Exercise 1.72 Show that, if filtered colimits exist in C, then C is a reflective

subcategory of Ind(C) with adjoint “ lim Xi ” 7→ lim Xi .
−→ −→
1.5.5 Kan extension (optional)

Definition 1.5.9 Let p : C → C ′ be a functor between small categories. The (left)
Kan extension of a functor F : C → D along p is a functor p!F which is universal
for all functors G : C ′ → D and natural transformations F ⇒ p−1 G := G ◦ p.
In other words, p!F represents the functor G 7→ Hom(F, G ◦ p) on the category

Hom(C ′ , D). It means that p!F : C ′ → D is endowed with a natural transformation
α : F ⇒ p−1 p! F such that, given any natural transformation γ : F ⇒ p−1 G, there
exists a unique natural transformation γe : p! F ⇒ G such that γ = p−1 (e
γ ) ◦ α:
C
F / +3 p−1 G.
>D (not commutative) and F 5=
p
p! F ⇒G

′
C p−1 p! F
There exists the dual notion of a right Kan extension p∗ F with γ : p−1 G ⇒ F this
time.
Example 1. A diagram D : I → C has colimit X if and only if the constant

1.5 Adjoint 37
functor 1 → C, 0 7→ X is the Kan extenstion of D along the projection I → 1:

D / /Y.
I ?C and D >
X→Y

1 X
2. A functor F : C → D between small categories has a coadjoint G if and only if

the Kan extension of IdC along F exists and F ◦ F! IdC = F !F , in which case
G = F! IdC :
C F /7 +3 G5= ′ ◦ F +3
>C D, IdC and F H3; ◦ F.
G
F
F ◦G
D G◦F F ◦G◦F
Proposition 1.5.10 Let p : C → C ′ be functor between small categories. Then the

functor
p−1 : Hom(C ′ , D) → Hom(C, D), G 7→ G ◦ p.
has an adjoint p! if and only if all Kan extensions along p with values in D exist
(and dual).
Proof. Follows immediately from the definition. ■
Proposition 1.5.11 If all colimits exist in D, then the Kan extension of F : C → D

along p : C → C ′ always exists (and dual).
Proof. (Sketch) We set (p! F )(X ′ ) := limp(X)→X ′ F (X) and check. ■

−→
Exercise 1.73 Show that if C is small, all colimits exist in D and F : C ,→ C ′ is

fully faithful, then F ≃ p!F ◦ p (and dual).
2. Topology
2.1 Compact Hausdorff space

2.1.1 Compact/Hausdorff space
Definition 2.1.1 A topological space X is said to be
• T1 if all points are closed,
• T2 if any two distinct points have disjoint neighborhoods,
• T4 if any two disjoint closed subsets have disjoint neighborhoods.
One also say for Fréchet for T1 , Hausdorff for T2 and normal 1 for T2 + T4 (or
equivalently T1 + T4 ).
Examples 1. A discrete space is normal.

2. A finite space is Fréchet if and only if it is discrete (in which case it is normal).
3. The space N = N ∪ {∞}, in which Z is closed if and only if it is finite or
contains +∞, is Hausdorff.
4. The Sierpiński space {s, η}, in which s is the only closed point, is T4 but not
normal2 .
Exercise 2.1 Show that a topological space X is Hausdorff it and only if the
diagonal ∆X ⊂ X × X is closed.
Exercise 2.2 Show that any subspace of a Hausdorff space is Hausdorff. Show
that a an equivalence relation on a topological space is closed if and only if the
quotient is Hausdorff. Show that a non empty product of topological spaces is
Hausdorff if and only if each of them is Hausdorff. Show that any limit (but not
1
Be careful that T4 and normal are sometimes reversed.
2
But it is T0 -space (or Kolmogorov).
40 Chapter 2. Topology
colimit) of Hausdorff spaces is Hausdorff.
Exercise 2.3 Show that the category of Hausdorff topological spaces is a reflective
subcategory of the category of all topological spaces.
Solution. Follows from Freyd’s adjunction theorem (the solution set is obtained by
considering any Hausdorff topology on a quotient X/ ∼). ■
Exercise 2.4 Let π : X ↠ Y be a continuous closed surjective map. Show that if

X is T1 (resp. T4 , resp. normal), then Y is T1 (resp. T4 , resp. normal).
Solution. The T1 case is obvious. Let, for i = 1, 2, Fi be disjoint closed subsets

of Y . Then π −1 (Fi ) are disjoint closed subsets of X. If X is T4 , then there exists
disjoint open neighborhoods Ui of π −1 (Fi ). It follows that (π(Uic ))c are disjoint open
neighborhoods of Fi . And Y is also T4 . ■
Exercise 2.5 Show that a normala space X is “locally closed”: any neighborhood
of x ∈ X contains a closed neighborhood.
a
Or more generally regular.
Lemma 2.1.2 — Urysohn. A topological space X is T4 if and only if, given two non-
empty closed disjoint subsets A and B, there exists a continuous map f : X → R
such that f (A) = 0 and f (B) = 1.
Proof. Classical. ■
Definition 2.1.3 A topological space S is said to be compact a if any open covering

Ui has a refinement S = Ui with J finite.
S S
S= i∈I i∈J
a
Some people say quasi-compact to insist on the fact that S may not be Hausdorff.
Alternatively, if
T a family of closed subsets (Fi )i∈I satisfies ̸ ∅ for all
T
i∈J Fi =
finite J ⊂ I, then i∈I Fi ̸= ∅.
Examples 1. A finite space (such as the Sierpiński space) is compact (but not
Hausdorff in général).
2. A discrete space is compact if and only if it is finite (in which case it is compact
Hausdorff).
3. The space N is compact Hausdorff.
Exercise 2.6 Show that a closed subspace of a compact space is compact. Show
that the image of a compact space by a continuous map is compact. Show that
a compact subspace of a Hausdorff space is closed. Show that a finite disjoint
union of compact (Hausdorff) spaces is compact (Hausdorff).
Exercise 2.7 Let R be an equivalence relation on a compact topological space S.

Show that S/R is compact. Assume now that S is also Hausdorff. Show that S/R
2.1 Compact Hausdorff space 41
is compact Hausdorff if and only if R ⊂ S × S is closed if and only if S ↠ S/R is

a closed map.
Solution. The first assertion follows from the fact that the image of a compact
topological space by a continuous map is always compact. We also know that an
equivalence relation is closed if and only if the quotient is Hausdorff. Also, if S
is compact Hausdorff and π : S ↠ S/R is closed, then this is a closed continuous
surjective map and S is normal. Then, we know that S/R is normal and therefore
Hausdorff. Finally, if S is compact and S/R is Hausdorff, then π is closed because
any closed subset of S is compact and any compact subset of S/R is closed. ■
Exercise 2.8 Show that, for a continuous map π : S → T of compact Hausdorff

spaces, the following are equivalent
1. π is a quotient map by an equivalence relation,
2. π is a cokernel,
3. π is an epimorphism,
4. π is surjective.
Solution. This is a straightforwards circular proof. ■
Exercise 2.9 Show that a compact Hausdorff space is normal.
Theorem 2.1.4 — Tykhonov. Any product of compact spaces is compact.
Proof. Classical. ■
The following result will also follow from proposition 2.1.7 below:
Exercise 2.10 Show that any limit of compact Hausdorff spaces is compact Haus-
dorff.
Solution. Thanks to Tykhonov theorem, it is sufficient to show that the kernel of two
maps f, g : S → T of compact Hausdorff spaces is compact Hausdorff, or equivalently,
closed. But ker(f, g) is the inverse image of the diagonal T ⊂ T × T , which is closed
since T is Hausdorff, by the continuous map S → T × T, x 7→ (f (x), g(x)). ■
We shall need later the following lemma
Lemma 2.1.5 If f : S ↠ T is a continuous surjection of compact Hausdorff spaces,
then there exists a minimal closed subset S ′ of S such that the restriction of f to
S ′ is surjective.
Proof. We use Zorn’s lemma. Thus, we give ourselves a decreasing family Si of closed
subsets of S that for all i ∈ I, the restriction of f to Si is surjective. If y ∈ T , then
f −1 (y) is closer and meets any of the Si which are closed. Since f −1 (y) is a compact
subset of S, it also meets necessarily S ′ := i∈I Si .
T
■
2.1.2 Stone-Čech compactification

Definition 2.1.6 The Stone-Čech compactification of a topological space X is the
closure βX of the image of the map
X → [0, 1]C(X,[0,1]) , x 7→ (f (x))f ∈C(X,[0,1]) .
It is hard in general (even when X is discrete infinite) to describe of βX.

Proposition 2.1.7 The category of compact Hausdorff spaces is a reflective subcate-
gory of the the category of all topological spaces with Stone-Čech compactification
as reflection.
Proof. We have to prove that Stone-Čech compactification X 7→ βX is adjoint to

inclusion. It follows from Tykhonov’s theorem that βX is compact Hausdorff. Also
the canonical map X → βX is continuous. In order to see this, it is sufficient to
show that the map X → [0, 1]C(X,[0,1]) is continuous but this may be checked on some
fixed factor f : X → [0, 1] which is continuous by definition. Let us now show that
the construction is functorial. Any continuous map ϕ : X → Y will induce a map
C(Y, [0, 1]) → C(X, [0, 1]), g 7→ ϕ∗ g := g ◦ ϕ
which, in turn will induce a map
[0, 1]C(X,[0,1]) → [0, 1]C(Y,[0,1]) , (xf ) 7→ (xϕ∗ g ).
In order to show that this map is continuous, it is sufficient again to consider the
projection on some fixed factor:
[0, 1]C(X,[0,1]) → [0, 1], (xf ) 7→ xϕ∗ (g) .
This is continuous by definition of the the product topology. Moreover, the diagram
X / [0, 1]C(X,[0,1])
ϕ

Y / [0, 1]C(Y,[0,1])
is clearly commutative. It follows that the closure of the image of X is sent into the
closure of the image of Y and we do get a continuous map βϕ : βX → βY . Finally, it
follows from Urysohn’s lemma that, when S is compact Hausdorff, the natural map
S → βS is injective: if x ̸= x′ ∈ S, then there exists a continuous map f : S → [0, 1]
such that f (x) = 0 and f (x′ ) = 1. It is therefore a homeomorphism (a continuous
bijection between compact Hausdorff spaces is always a homeomorphism).
Our assertion therefore follows from exercise 1.70. ■
As a consequence of proposition 2.1.7, any limit (in the category of topological
spaces) of compact Hausdorff spaces is compact Hausdorff. In particular, a topological
monoid (resp. group, resp. abelian group) which is compact Hausdorff is the same
thing as monoid (resp. group, resp. abelian group) in the category of compact
Hausdorff spaces. There also exists colimits obtained by applying β to the colimit
of topological spaces. When we write lim Xi , we always mean the colimit in the
−→
category of topological spaces, even if all Xi ’s are compact Hausdorff. We shall then
write β lim Xi if we ever need to consider the colimit in the category of compact
−→
Hausdorff spaces.
It is important to remember that there exists a natural bijection
C(βX, S) ≃ C(X, S)
when S is compact Hausdorff.
2.2 Projective objects

2.2.1 Free compact Hausdorff space
Definition 2.2.1 A free compact Hausdorff space is a topological space F homeo-
morphic to the Stone-Čech compactification of a discrete space I.
There is a natural bijection C(F, S) ≃ S I when S is compact Hausdorff. In

other words, the functor I 7→ βI disc is adjoint to the forgetful functor from compact
Hausdorff spaces to sets. One may then call the set I a basis for F : it is equivalent
to give a continuous map F → S or a family (ti )i∈I of elements of S. When F ↠ S
is surjective, we may call (ti )i∈I a set of generators for S.
Examples 1. A compact Hausdorff space has a finite number of generators if

and only if it is finite (discrete).
2. The free compact Hausdorff space with basis N is homeomorphic to the set F of
maximal filters on N with the topology generated by the UA := {F ∈ F, A ∈ F}
for A ⊂ N.
Let us recall that F ⊂ P(I) is a (proper) filter on a set I if 0) ∅ ∈ / F, 1)

∀J1 , J2 ∈ F, J1 ∩ J2 ∈ F and 2) ∀J ⊂ J ′ , J ∈ F ⇒ J ′ ∈ F. For example, if i ∈ I,
then F := {J, i ∈ J} is a maximal filter.
Exercise 2.11 Show that the free compact Hausdorff space on a set I is the set F
of maximal filters on I with the topology generated by the UJ := {F ∈ F, J ∈ F}
for J ⊂ I.
Exercise 2.12 Show that, if S, T are two (free) compact Hausdorff spaces, then
S ⊔ T is the coproduct in the category of (free) compact Hausdorff spaces.
Proposition 2.2.2 1. Any compact Hausdorff space is a quotient of a free compact

Hausdorff space.
2. A free compact Hausdorff space is a projective object in the category of
compact Hausdorff spaces.
Proof. 1. If S is a compact Hausdorff space, then there exists a commutative

diagram
βS disc / βS
O O
≃
S disc //S
and therefore a natural continuous surjection F := βS disc ↠ S.

2. Assume given a free compact Hausdorff space F , a continuous surjection of
compact Hausdorff spaces T ↠ S and continuous map F → S. We have
F ≃ βI where I is a discrete space and the composite map I → βI ≃ F → S
lifts to a (automatically) continuous map I → T . Since T is compact Hausdorff,
it extends uniquely to F ≃ βI → T . ■
In particular, we see that the category of compact Hausdorff spaces has enough
projectives.
Definition 2.2.3 A diagram F ′ ⇒ F → S of compact Hausdorff spaces is called a
free presentation if both F, F ′ are free and both F ↠ S and F ′ ↠ F ×S F are
surjective.
Equivalently, the image of F ′ in F × F is (the graph of) an equivalence relation
R on F and F/R ≃ S. Then, automatically coker(F ′ ⇒ F ) ≃ S.
Corollary 2.2.4 Any compact Hausdorff space S has a free presentation. Actually,
any continuous surjection F → S from a free space extends to a free presentation.
■
2.2.2 Totally/Extremally disconnected space

Definition 2.2.5 Let X be a topological space. A subset of X is said to be clopen
if it is both open and closed. The space X is said to be connected if only ∅ and
X are clopen in X. The connected components of X are the maximal connected
subspaces.
Exercise 2.13 Show that the image of a connected space by a continuous map
is always connected. Show that the closure of a connected subset is connected.
Show that a connected component is closed (but not necessarily open). Show that
a clopen subset is a (disjoint) union of connected components.
Proposition 2.2.6 If S is compact Hausdorff, then the connected component of

x ∈ S is the intersection of all clopen subsets of S containing x.
Proof. The connected component of x is always contained in the intersection C of all

clopen subsets of S containing x. It remains to show that C is connected. Assume
that C = F ∪ G is the disjoint union of two closed subsets of C. Since C is closed in
S, then F and G are also closed in S. Since S is compact Hausdorff, it is normal and
there exist disjoint open subsets U, V ⊂ S with F ⊂ U and G ⊂ V . The complement
K of U ∪ V in S is a compact subset and K ∩ C = ∅. Since a finite intersection of
clopen is always clopen, there must exist a clopen H of S containing x such that
K ∩ H = ∅ (because K is compact). In other words, we have H ⊂ U ∪ V . Since U

and V are disjoint open subsets of S, then H ∩ U and H ∩ V are clopen in H and
therefore also in S. We may assume x ∈ F (or else, it is in G) so that x ∈ H ∩ U .
Then, C ⊂ H ∩ U so that C ∩ V = ∅ which implies that G = ∅. ■
Definition 2.2.7 A topological space X is said to be totally (resp. extremally)

disconnected if its connected components are the points (resp. if the closure of
any open subset is open). If, moreover, X is compact Hausdorff, then it is called
a Stone (resp. Stonean) space.
Examples 1. Q is totally disconnected but not extremally disconnected.

2. The Sierpiński space {s, η} is extremally disconnected but not totally discon-
nected.
3. The space N is Stone (but not Stonean).
4. The space βN is Stonean.
5. If p is a prime, then
(∞ )
X
Zp =: i
ai p , ai ∈ Z ≃ lim Z/pn+1 Z
←−
i=0 n
is Stone (but not Stonean).

6. A Stonean space is metrizable if and only if it is finite discrete.
2.2.3 Stone space

We shall denote by π0 (X) the set of connected components of a topological space
X. There exists an obvious surjection C : X → π0 (X) that sends a point x to its
connected component C(x) and we endow π0 (X) with the quotient topology. We
obtain a functor X 7→ π0 (X).
Exercise 2.14 Show that the category of totally disconnected spaces is a reflective
subcategory of the category of all topological spaces with reflection π0 .
Solution. The canonical map C : X → π0 (X) is an isomorphism if and only if X is

totally disconnected. Our assertion therefore follows from exercise 1.70. ■
As a consequence, any limit of totally disconnected spaces is totally disconnected
(and there also exists colimits obtained by applying π0 to the limit of the topological
spaces).
Exercise 2.15 Show that a subspace of a totally disconnected space is totally
disconnected.
Exercise 2.16 Show that the category of Stone spaces is a reflective subcategory
of the category of compact Hausdorff spaces. Actually, π0 provides a reflection.
Solution. The point is to show that, if S is compact Hausdorff, then π0 (S) is Hausdorff.
Let C and C ′ be two distinct connected components of S. Since C ′ is compact, and
clopen are stable under finite intersection, it follows from proposition 2.2.6 that there
exists a clopen U such that C ⊂ U and C ′ ∩ U = ∅. Then, U is a union of connected
components. In other words, we have U = π0−1 (V ) where V is a (necessarily) clopen

neighborhood of C ∈ π0 (S) and by construction, C ′ ∈
/ V. ■
As a consequence, any limit of Stone spaces is Stone (and there also exists colimits
obtained by applying successively β and π0 to the colimit of topological spaces).
Definition 2.2.8 A profinite space is a limit of finite discrete topological spaces.
Exercise 2.17 Show that a profinite space is a directed limit of finite discrete
topological spaces with surjective transitions maps.
Solution. If S = limi∈I Si and J ⊂ I is finite, we set SJ := limi∈J Si . Then,

←− ←−
S = limJ SJ . The result also follows from the proof of proposition 2.2.9 below. ■
←−
When we write S = limi∈I Si , we shall usually implicitly assume that I is a
←−
directed set and the maps Si → Sj are surjective when j ≤ i.
Proposition 2.2.9 A topological space is Stone if and only if it is profinite.
Proof. First of all, a profinite space is a Stone space as a limit of Stone spaces. Assume
conversely that` S is a Stone space. Let E be the set finite families E ⊂ Open(S) \ {∅}
such that S = U ∈E U . If E ∈ E, then there exists a continuous surjection S → E,
that sends x to U if x ∈ U and they are compatible with inclusion E ′ ⊂ E.
Therefore, there exists a continuous map S → lim E. Since S is compact, if (UE )E∈E
←−
is a compatible system, then ∩E∈E UE ̸= ∅: indeed, this is clearly true for a finite
family and the UE ’s are closed. It follows that the map is surjective. Now, since S is
a Stone space, if x ∈ S, then {x} it the intersection of all clopen containing x. It
follows that the map is injective: if x ̸= y, there exists U clopen such that x ∈ U
and y ∈/ U and we may choose E = {U, U c }. A bijective continuous map of compact
Hausdorff spaces is a homeomorphism. ■
Exercise 2.18 Show that, if S is a profinite space with transition maps πi : S → Si ,

then the Ux,i = πi−1 (πi (x)) for various i and x ∈ S form a basis of open subsets
stable under intersection.
Solution. By definition of the inverse limit topology, the topology of S is generated
by the πi−1 (U ) when i runs through I and U runs through a basis of open subsets
of Si . Since the points form a basis of open subsets of Si , we get exactly the Ux,i ’s.
Morevoer, we have Ux,j ⊂ Ux,i for i ≤ j, Ux,i = Uy,i if πi (x) = πi (y) and Ux,i ∩Uy,i = ∅
if πi (x) ̸= πi (y). ■
Exercise 2.19 Show that, if S is a Stone space, then any open covering has a finite
disjoint clopen refinement.
Solution. It follows from proposition 2.2.9 and exercise 2.17 that S = limi∈I Sk is
←−
a directed limit of finite discrete spaces and we shall denote by πi : S → Si the
projection. If x ∈ S, then the clopen subsets Ux,i := πi−1 (πi (x)) with i ∈ I form
a basis of open neighborhoods of x. Since S is compact, our covering has a finite
refinement by open subsets of the form Uxk ,ik with k = 1, . . . , r. Since I is directed,
there exists i ≥ ik for i = 1, . . . , r. Then, for a ∈ Si , we set Ua = πi−1 (a). This is a
finite disjoint clopen covering of S. This is also a refinement of our covering: if x ∈ Ua ,

then there exists k such that x ∈ Uxk ,ik and therefore Ua = Ux,i ⊂ Ux,ik = Uxk ,ik . ■
Exercise 2.20 Show that the category of profinite sets is equivalent to the category
of profinite spaces.
Recall that a boolean algebra is a commutative ring A such that all a ∈ A are
idempotent (a2 = a).
Exercise 2.21 Prove Stone representation theorem: the opposite of the category
of boolean algebras is equivalent to the category of Stone spaces.
Solution. One endows S(A) := HomRng (A, F2 ) ⊂ FA2 with the induced topology (and
F2 with the discrete topology). A quasi-inverse is given by sending S to the set
A(S) of all clopen subsets of S with the operations a + b := (a ∪ b) \ (a ∩ b) and
ab = a ∩ b. ■
2.2.4 Stonean space

Exercise 2.22 Show that if X is extremally disconnected and U, V are disjoint
open subsets, then U and V are also disjoint.
Solution. We have U ⊂ V c which is closed and therefore U ⊂ V c so that V ⊂ (U )c

which is also closed because U is open and therefore V ⊂ (U )c . ■
Alternatively, if T = F ∪ G is a closed covering, then T = F̊ ∪ G̊.

Lemma 2.2.10 An extremally disconnected Hausdorff space X is totally discon-
nected.
Proof. If C is the connected component of x ∈ X and x ̸= y ∈ X, then there exists

two disjoint open subsets U and V such that x ∈ U and y ∈ V . Since X is extremally
disconnected, U and V also are disjoint. Moreover, C ⊂ U which is clopen containing
x. It follows that y ̸∈ C. ■
Theorem 2.2.11 — Gleason. The projective objects of the category of compact

Hausdorff spaces are the Stonean spaces.
Proof. Assume first S is a projective compact Hausdorff space. Let U be an open

subset of S and F its complement. Then, the continuous surjection p : U ⊔ F ↠ S of
compact Hausdorff spaces has a continuous section s. Necessarily, s(U ) ⊂ U . Since
s is continuous, s(U ) ⊂ U . It follows that U = s−1 (U ) is open.
For the converse, thanks to exercise 1.48, it is sufficient to show that any continuous
surjection f : T ↠ S from a compact Hausdorff space to a Stonean space has a
section. Thanks to lemma 2.1.5, we may assume that the restriction of f to any
closed subset T ′ ⊊ T is never surjective. We shall then show that f : T ≃ S is a
homeomorphism and it is sufficient to prove that f is injective. Assume xi ∈ T , for
i = 1, 2, have same image y ∈ S and pick up disjoint neighborhoods Ui of xi . Since
f is closed, Si := f (Uic ) is closed and since f is surjective, S = S1 ∪ S2 . Since S is
extremally disconnected, we also have S = S̊1 ∪ S̊2 . We may assume that y ∈ S̊1 and
we set T ′ := (U1 ∩ f −1 (S̊1 ))c . By construction, the restriction of f to T ′ is surjective

and T ′ ⊊ T . Contradiction. ■
Corollary 2.2.12 A topological space is a Stonean space if and only if it is a retract

of a free compact Hausdorff space. ■
In particular, a free compact Hausdorff space is extremally disconnected.
“deformation”
2.3 Compactly generated space

The standard reference is Lewis 78 but you may also consider Strictland.
2.3.1 Definition
Definition 2.3.1 A topological space X is
1. locally compact Hausdorff if it is Hausdorff and any point has a compact
neighborhood,
2. compactly (Hausdorff ) generated (also called a k-space) if it is a colimit of
compact Hausdorff spaces.
Proposition 2.3.2 If X ↠ Y is a quotient map and Z is locally compact, then

X × Z ↠ Y × Z is also a quotient map.
Proof. Classic. ■
Exercise 2.23 Show that, for a topological space X, the following conditions are
equivalent:
1. X is compactly generated.
2. X is a quotient of a disjoint union of compact Hausdorff spaces.
3. X is a quotient of a locally compact Hausdorff space.
4. A subset Y of X is open (resp. closed) when, given any continuous map
f : S → X from a compact Hausdorff space, f −1 (Y ) is open (resp. closed)
in S.
5. A map X → Y is continuous when, given any continuous map S → X
from a compact Hausdorff space, the composite map S → X → Y is also
continuous.
6. X = lim S when S → X runs through all continuous maps from a compact
−→
S→X
Hausdorff space.
Solution. This is a straightforwards circular proof. ■
Examples 1. A locally compact Hausdorff space is compactly generated (this

applies to topological manifolds).
2. A sequential space (a subset is closed if and only if it is stable under convergent
sequences) is compactly generated (this applies to metric spaces).
3. If I is uncountable, then RI and ZI are not compactly generated.
Exercise 2.24 Sow that, if X is compactly generated and Y locally compact
Hausdorff, then X × Y is compactly generated.
Definition 2.3.3 A subset Y of a topological space X is said to be k-open (resp.

k-closed ) if for all continuous map f : S → X from a compact Hausdorff space,
f −1 (Y ) is open (resp. closed) in S. The corresponding topology on X is the called
the k-topology (or compactly generated topology).
We shall denote by kX the underlying set of X endowed with the k-topology.

Exercise 2.25 Show that the k-topology is indeed a topology and that kX = lim S
−→
S→X
when S runs through all compact Hausdorff spaces.
Proposition 2.3.4 The subcategory of compactly generated spaces is coreflective

with coreflection k.
Proof. The assignment X 7→ kX is functorial, the identity is a natural continuous

map kX → X which is a homeomorphism when X itself is compactly generated.
Our assertion therefore follows from exercise 1.70. ■
As a consequence, compactly generated spaces are stable under all colimits

(and consequently quotients). We shall use the properties kkX = kX when X
is any topological space and C(kX, kY ) = C(kX, Y ) if Y is another topological
space. Also, if X any topological space and Y is locally compact Hausdorff, then
k(X × Y ) = kX × Y .
2.3.2 Compact-open topology

Definition 2.3.5 Let X and Y be two topological spaces. If S → X is a continuous
map from a compact Hausdorff space and V an open subset of Y , we denote by
WS,V is the set of all continuous maps f : X → Y such that the image of the
composite map S → X → Y is contained in V . The compact-open topology a on
C(X, Y ) is the topology generated by all WS,V .
a
This is compatible with the usual definition when X is (weak) Hausdorff.
We shall systematically endow C(X, Y ) with the compact-open topology.

Exercise 2.26 Show that if X is discrete, then there exists a homeomorphism
C(X, Y ) ≃ Y X (with the product topology).
Exercise 2.27 Show that if S is compact Hausdorff and X is (semi-) metric

(complete), then the topology of C(S, X) is (semi-) metric (complete) with d(f, g) =
supx∈S d(f (x), g(x)).
Solution. We start from f ∈ WK,U with K ⊂ S compact and U ⊂ X open. Denote

by F the complement of U in X. Since f (K) is compact and f (K) ∩ F = ∅, then
d(F, K) = ϵ > 0 and B(f, ϵ− ) ⊂ WK,U . Conversely, given B(f, ϵ− ), there exists for
all x ∈ S a compact
Sr neighborhood of Kx of x such that f (Kx ) ⊂ B(f (x), ϵ− ). We
can Twrite S = i=1 Kxi . Then, if we set Ki = Kxi and Ui := B(f (x), ϵ− ), we have
f ∈ ri=1 WKi ,Ui ⊂ B(f, ϵ− ). ■
Note in particular, C(S, X) is compactly generated3 when S is compact and X

semi-metric.
Theorem 2.3.6 If X, Y, Z are three topological spaces, then there exists a homeo-
morphism
kC(k(X × Y ), Z) ≃ kC(kX, C(kY, Z)).
Proof. It is sufficient to check that the bijection
F(X × Y, Z) ≃ F(X, F(Y, Z)).
induces a bijection
C(k(X × Y ), Z) ≃ C(X, C(Y, Z)) (2.1)
for compactly generated spaces and then apply Yoneda lemma. This is done by hand
(see theorem 5.9.8 in [Bro06] for example). ■
The category of compactly generated spaces is cartesian closed (meaning that 2.1
holds). More generally:
Corollary 2.3.7 The functor X 7→ k(X ×Y ) is adjoint to the functor Z 7→ kC(kY, Z)

on the category of compactly generated spaces X. ■
As a consequence, the functor X 7→ k(X × Y ) (resp. Z 7→ kC(kY, Z)) commutes
with colimits (resp. limits) of compactly generated spaces. In particular, colimits are
universal (see definition 3.2.11 below) in the category of compactly generated spaces
(this is not true in the category of all topological spaces).
Exercise 2.28 Show that if X, Y are two topological spaces, then k(kX × Y ) =
k(X × Y ).
Solution. This follows from the uniqueness of the adjoint. ■
Exercise 2.29 Show that, if X is compactly generated and Y is locally compact

Hausdorff, then
C(X × Y, Z) ≃ C(X, C(Y, Z)) ≃ C(Y, C(X, Z)).
2.3.3 Weak Hausdorff space

Definition 2.3.8 A topological space X is said to be weak Hausdorff (also called
an h-space) if, for all morphism f : S → X with S compact Hausdorff, f (S) is
closed in X.
Clearly, any subspace of a weak Hausdorff space is also weak Hausdorff. Also, if
X is weak Hausdorff, then so is kX. A compactly generated weak Hausdorff space is
also called an hk-space.
3
It is usually quite hard to tell if C(X, Y ) is compactly generated.
Exercise 2.30 Show that Hausdorff implies weak Hausdorff implies Fréchet.
Exercise 2.31 Show that, if X is a weak Hausdorff topological space, then, for all
morphism f : S → X with S compact Hausdorff, f (S) is compact Hausdorff.
Solution. If S ′ ⊂ S, is a closed subset, it is compact Hausdorff and therefore f (S ′ )

is closed. The map S ↠ f (S) is closed surjective and S is normal. The assertion
therefore follows from exercise 2.4. ■
Exercise 2.32 Show that, if X is weak Hausdorff, then Y ⊂ X is k-closed (resp.

k-open) if and only if for all compact Hausdorff S ⊂ X, S ∩ Y is open (resp.
closed).
Exercise 2.33 1. Show that if S → X and S ′ → X are two morphisms from a

compact Hausdorff space to a weak Hausdorff space, then S ×X S ′ is compact
Hausdorff.
2. Show that compact Hausdorff subspaces of a weak Hausdorff space are stable
under finite unions.
Solution. The second assertion follows from the first one which we now prove. Since
X is weak Hausdorff and S1 ⊔ S2 is compact Hausdorff, then the image T of S1 ⊔ S2
in X is a compact Hausdorff subset and therefore S1 ×X S2 = S1 ×T S2 is compact
Hausdorff. ■
Exercise 2.34 Show that, if X is weak Hausdorff, then the diagonal X ⊂ X × X

is k-closed. Show that the converse holds if X is compactly generated.
Exercise 2.35 Show that if Y is weak Hausdorff and f, g : X → Y are two

continuous maps, then ker(f, g) ⊂ X is k-closed.
Solution. This is the inverse image of the diagonal along the continuous map (f, g) :
X →Y ×Y. ■
Exercise 2.36 Show that, if Y is weak Hausdorff, then kC(X, Y ) is also weak
Hausdorff.
Exercise 2.37 Show that an equivalence relation R on a compactly generated
space X is k-closed if and only if X/R is weak Hausdorff.
If X is a compactly generated topological space, we shall denote by hX := X/R

where R is the smallest closed equivalence relation on X.
Exercise 2.38 Show that the category of hk-spaces is a reflective subcategory of
the category of all k-spaces with reflection X 7→ hX.
It follows that any limit of compactly generated weak Hausdorff spaces, endowed
with the k-topology, is weak Hausdorff.
Exercise 2.39 Show that an open (resp. a closed) subspace of an hk-space is an
hk-space.
Proposition 2.3.9 A filtered limit X = limi∈I Xi of hk-spaces under closed inclusion

−→
maps is an hk-space and all Xi ’s are closed in X.
Proof. We may assume that I is a directed set. If i, j ∈ I, then there exists k ∈ I

such that i, j ≤ k, and we set
Rij := Xi ×Xk Xj = ker(Xi × Xj ⇒ Xk )
(which does not

` depend on k). Since Xk is weak Hausdorff, Rij is k-closed. Now, if
we set Y := i∈I Xi , then
a a
R := Rij ⊂ (Xi × Xj ) = Y × Y
i,j i,j
is an equivalence relation on Y and we have X = Y /R. Since colimits are universal in

the category of compactly generated spaces, we have a homeomorphism k(Y × Y ) ≃
i,j k(Xi × Xj ). It follows that R is k-closed and this implies that X is weak
`
Hausdorff. Finally, if i, j ∈ I, then there exists k ∈ I such that i, j ≤ k and we
assumed that Xi is closed in Xk . It follows that Xi ∩ Xj is closed in Xj . Thus, if we
denote by p : Y ↠ X the quotient map, we see that p−1 (Xi ) is closed in Y , and this
means that Xi is closed in X. ■
Exercise 2.40 Let X = limn∈N Sn be a countable filtered colimit of compact

−→
Hausdorff spaces under inclusion maps. Then, any continuous map S → X from
a compact Hausdorff space factors through some Sn .
Solution. Since X is weak Hausdorff, we can assume that Sn ⊂ Sn+1 for all n ∈ N.
We may also replace S with its image in X. Then, we have S = limn∈N S ∩ Sn and
−→
we may finally assume that X = S and S = limn∈N Sn . We have to show that there
−→
exists n such that S = Sn . Otherwise, we may assume that Sn ⊊ Sn+1 for all n ∈ N
and pick a point xn ∈ Sn+1 \ Sn up. If T ⊂ {xn }n∈N , then T ∩ Sn is (finite) closed in
Sn for all n ∈ N. Since S has the colimit topology, then T is closed in S. This shows
that {xn }n∈N is discrete in S compact and therefore finite. Contradiction. ■
3. Sites and topos
We shall now explain Grothendieck’s splendid theory of topos.
Recall that we are supposed to work in a fixed universe. Unfortunately, if we are

given two categories C and D, then the set of all functors F : C → D does not make
a (locally small) category in general because the set of all natural transformations
between two of them is not always small. This is the case however when C is small,
but then, the new category Hom(C, D) is not small in general, and the process
cannot be iterated. The simplest solution is to enlarge our universe as needed. We
shall do that informally and not worry much anymore about set-theoretic issues.
3.1 Presheaf
We fix a category C.
3.1.1 Defintion
Definition 3.1.1 A presheaf (of sets) on C is a (contravariant) functor T : C op → Set.
A morphism of presheaves is a natural transformation between them.
Examples 1. A presheaf on a category C is the same thing as a diagram of sets

on C .
op
2. A presheaf on a preordered set (I, ≤) is given by a family of sets Ti , together

with a compatible family of “restriction” maps Tj → Ti for i < j.
3. If X is a topological space, then a presheaf T on Open(X) (also called a
presheaf on X) is the following data :
(a) a set T (U ) for any open subset U of X, and
(b) compatible restriction maps
∀U ′ ⊊ U, T (U ) → T (U ′ ), s 7→ s|U ′ .
54 Chapter 3. Sites and topos
4. For fixed set E, we can consider the constant presheaf EC on a category C that
sends any X to E and any f to IdE . When C = Open(X), we shall write EX .
5. For fixed X ∈ C, we can consider the presheaf
hX : Y 7→ Hom(Y, X)
on C. Recall that a presheaf is said to be representable if it is isomorphic to

hX .
We shall denote by Cb := Hom(C op , Set) the category of all presheaves on C. If T

is a presheaf on C and f : Y → X any morphism, then we shall write f −1 := T (f ) :
T (X) → T (Y ). For s ∈ T (X), we may also write s|Y := f −1 (s).
Exercise 3.1 Show that, for a set E and a presheaf T , we have the adjunction
(where 1Cb denotes a final object)
Hom(EC , T ) ≃ HomSet (E, Hom(1Cb, T )) (≃ Hom(1Cb, T )E ).
Solution. We may assume that 1Cb is the constant presheaf associated to 1 := {0}
(which is clearly a final object). There exists a natural bijection
E ≃ Hom(1, E) ≃ Hom(1Cb, EC ), e 7→ eC .
Composition therefore provides us with a natural map
Hom(EC , T ) → Hom(E, Hom(1Cb, T )), φ 7→ (e 7→ φe := φ ◦ eC ).
By construction, if X ∈ C, we have φe,X (0) = φX (e) which implies that the map is
injective and provides a candidate for an inverse. It is however necessary to check
that ϕ will be a morphism of presheaves but if f : Y → X, we have f ∗ (φX (e)) =
f ∗ (φe,X (0)) = φe,Y (0) = φY (e). ■
Exercise 3.2 Show that a presheaf on Top is equivalent to the following data :
1. a presheaf TX (its realization) on each topological space X, and
2. a compatible family of morphisms TX 7→ fb∗ TY for all continuous map
f : Y → X, where fb∗ TY is the presheaf on X defined by
fb∗ TY (U ) = TY (f −1 (U )).
Examples 1. We have 0 b ≃ 1 and 1b ≃ Set.

2. If G is a monoid, then G ≃ G−Set.
b
3. If X is a topological space, then Open(X)
\ is the « usual » category of presheaves
(of sets) on X.
3.1 Presheaf 55
3.1.2 Yoneda embedding
Theorem 3.1.2 All limits and colimits exist in Cb and, if X ∈ C, then the functor
Cb → Set, Γ(X, T ) := T 7→ T (X)
preserves all limits and colimits.
Proof. Follows from exercises 1.66 and 1.37. ■
Examples 1. A morphism T → T ′ of presheaves of sets is a monomorphism

(resp. an epimorphism) in Cb if and only if all T (X) → T ′ (X) are injective (resp.
surjective).
2. If φ : T → T ′ is any morphism of presheaves of sets, then im φ exists in Cb and
we have for all X ∈ C, (im φ)(X) = Im φX (and dual).
3. Filtered direct limits are exact in C.
b
Proposition 3.1.3 — Yoneda embedding. The functor
ょ : C ,→ C,
b X 7→ hX
is fully faithful and preserves all limits.
Proof. It follows from Yoneda lemma that, if X, Y ∈ C, then
Hom(X, Y ) = hY (X) ≃ Hom(hX , hY ).
The other assertion follows from the theorem. ■

In other words, the category C can be identified with a full subcategory of Cb but
this embedding does not preserve colimits in general. However, since it is left exact,
it preserves any kind of algebraic structure.
Exercise 3.3 Show that there exists a fully faithful functor
Ind(C) ,→ C,
b “ lim Xi ” 7→ lim hXi .
−→ −→
A presheaf T isomorphic to lim hXi is said to be ind-representable.
−→
3.1.3 Equivalence relation

Definition 3.1.4 1. An equivalence relation on T ∈ Cb is a diagram R ⇒ T such
that, for all X ∈ C, the map R(X) → T (X) × T (X) is a bijection onto (the
graph of) an equivalence relation.
2. An equivalence relation on X ∈ C is a diagram R ⇒ X in C such that
hR ⇒ hX is an equivalence relation on hX .
3. An equivalence relation R on X ∈ C is said to be effective if
R = X ×X X with X = coker(R ⇒ X).

In this case X/R := X is called the quotient of X by R.
Exercise 3.4 Show that, if X → Y is any morphism (resp. a regular epimorphism),

then R := X ×Y X (if it exists) is an (resp. an effective) equivalence relation on
X (resp. and X/R = Y ).
Examples 1. An equivalence relation on a set or a topological space is a usual

equivalence relation. It is always effective.
2. An equivalence relation in Cb is always effective (and any epimorphism in Cb is
always regular).
3.1.4 Slice category

Definition 3.1.5 If T is a presheaf of sets on a category C, then the slice category
C/T is defined as follows:
1. an object of C/T is a pair made of an object X of C and a section s ∈ T (X),
2. a morphism in C/T is a morphism f : X → X ′ in C such that T (f )(s′ ) = s.
There exists an obvious forgetful functor jT : C/T → C.

Exercise 3.5 1. Show that, if X ∈ C, then there exist an equivalence C/X ≃
C/hX .
2. Show that the diagram

C/T / Cb/T

ょ
/ Cb
C
is cartesian.
Exercise 3.6 1. Prove the density theorem: any presheaf T on a category C is
a colimit of representable presheaves. More, precisely:
T ≃ lim hX .
−→
X∈C/T
2. Show that if T is another presheaf, then Hom(T, T ′ ) ≃ limX∈C T ′ (X).

←− /T
Solution. By definition, giving a morphism limX∈C hX → T ′ amounts to giving a

−→ /T
compatible family of morphisms s′ : hX → T ′ indexed by (X, s) ∈ C/T . Thanks
to Yoneda’s lemma, this is equivalent to giving a compatible family of elements
s′ ∈ T ′ (X) indexed by elements s ∈ T (X). This is the definition of a morphism
T → T ′ . This shows the first asssertion and the second one follows from the fact
that the functor Hom preserves all limits. ■
Exercise 3.7 Show that a presheaf T on a category C is ind-representable if and

only if C/T is filtered if and only if T is left exact.
3.2 Site 57
3.2 Site
3.2.1 Topology
Definition 3.2.1 If C is a category and X ∈ C, then a subobject R of hX is called
a sieve of X. The inverse image of R under a morphism f : Y → X is the sieve
f −1 (R) := R ×hX hY of Y .
The set C/R is sometimes also called a sieve (this was actually the original
definition).
Exercise 3.8 Show that it is equivalent to give a sieve R of X or a set R of
morphisms X ′ → X such that, if X ′ → X belongs to R, then any precomposition
X ′′ → X ′ → X is still in R. Show that f −1 (R) then corresponds to the set of all
morphisms Y ′ → Y such that the composition Y ′ → Y → X is in R.
Since the notion of a sieve of X is stable under intersection, there is a notion of
sieve generated by a family (Xi → X)i∈I .
Exercise 3.9 Show that the sieve R of X generated by a family (fi : Xi → X)i∈I
is i∈I im hfi . In other words, Y → X is in R(Y ) if and only if it factors through
S
some Xi .
Examples 1. If (I, ≤) is a preordered set and x ∈ I, then it is equivalent to give

a sieve of x or a subset J of I such that
(a) ∀y ∈ J, y ≤ x,
(b) ∀y ∈ J, ∀z ∈ I, z ≤ y ⇒ z ∈ J.
2. If X is a topological space and U an is open subspace of X, then a sieve of U
corresponds to a family U open subsets U ′ of U such that, if U ′′ ⊂ U ′ , then
also U ′′ ∈ U.
Definition 3.2.2 A topology on a category C is the data of a set J(X) of cover-

ing sieves of X for each X ∈ C, such that:
1. ∀X ∈ C, hX ∈ J(X),
2. ∀R ∈ J(X), ∀f : Y → X, f −1 (R) ∈ J(Y ),
3. (∃R′ ∈ J(X), ∀f ∈ R′ (Y ), f −1 (R) ∈ J(Y )) ⇒ R ∈ J(X).
A site is a category C endowed with a topology.
We shall simply say that C is a site without mentioning explicitly the topology.
Exercise 3.10 Show that J(X) is a filter (on the ordered set of all sieves):
1. J(X) ̸= ∅,
2. if R, R′ ∈ J(X), then R ∩ R′ ∈ J(X),
3. if R ⊂ R′ and R ∈ J(X), then R′ ∈ J(X).
Solution. If f ∈ R(Y ), then in the first case, f −1 (R ∩ R′ ) = f −1 (R′ ) ∈ J(Y ), and in

the second case, f −1 (R′ ) = hY ∈ J(Y ). ■
Definition 3.2.3 A topologically generating set for a site C is a set G ⊂ C such that
any X ∈ C admits a covering sieve generated by a family (fi : Xi → X)i∈I with
Xi ∈ G.
For set-theoretic issues, it is necessary to always assume that there exists a small
topologically generating set. It makes it possible to replace J(X) with a small set.
If C is any category, then the various topologies on C are ordered by inclusion from
coarse (only hX covers X) to discrete (any sieve R covers X). Any intersection of
topologies is a topology and it follows that any set of sieves generates a topology. As
a consequence, any set of families (Xi → X)i∈I for various X generates a topology.
In practice, it is convenient to rely on the following:
Definition 3.2.4 A pretopology on a category C is the data of sets Cov(X) of
covering families (Xi → X)i∈I for all X ∈ C such that
1. any isomorphism X ′ → X is in Cov(X),
2. if (Xi → X)i∈I ∈ Cov(X) and f : Y → X is any morphism, then (Xi ×X Y →
Y )i∈I ∈ Cov(Y ),
3. if (Xi → X)i∈I ∈ Cov(X), and for each i ∈ I, (Xij → Xi )j∈Ii ∈ Cov(Xi ),
then (Xij → X)i∈I,j∈Ij ∈ Cov(X).
If C is endowed with a pretopology, we will consider it as a site with respect to

the topology generated by the set of Cov(X) for all X.
Examples 1. If X is a topological space, S

we turn Open(X) into a site by calling
a family (Ui ⊂ U )i∈I a covering if U = i∈I Ui .
2. One can turn Set into a site by S calling a family (Xi → X)i∈I a covering when
it is jointly surjective : X = i∈I f (Xi ).
3. We turn Top into a site by calling a a family (fi : Xi → X)i∈I a covering if it
is jointly surjective and each fi induces an isomorphism Xi ≃ f (Xi ) with an
open subset of X.
Exercise 3.11 Show that, if C is endowed with a pretopology, then a sieve of

X ∈ C is a covering sieve if and only if it contains a sieve generated by a covering
family.
Exercise 3.12 Show that, if C is a site with fibered products, then the set of all
families (Xi → X)i∈I that generate a covering sieve of X, is a pretopology that
generates the topology of C. This is called the maximal pretopology (of the site).
Be careful however that a family that generates a covering sieve is not necessarily
a covering family for the given pretopology.
3.2.2 Sheaf
Definition 3.2.5 A presheaf F : C op → Set on a site C is separated (resp. a sheaf )
if, for all X ∈ C and R ∈ J(X), the restriction map
Hom(hX , F) → Hom(R, F).
is injective (resp. bijective).

3.2 Site 59
Exercise 3.13 Show that a presheaf F is a sheaf if and only if for all X ∈ C and
all covering sieves R of X, we have
F(X) ≃ lim F(Y ).

←−
Y ∈C/R
Solution. Follows from exercise 3.6. ■

Exercise 3.14 Show that, if C is endowed with a pretopology, then a presheaf F is
a sheaf if and only if for all X ∈ C and all covering families (Xi → X)i∈I
!
Y Y
F(X) ≃ ker F(Xi ) ⇒ F(Xi ×X Xj ) .
i∈I i,j∈I
We shall denote by Ce the full subcategory of sheaves of sets on the site C and by
H : Ce ,→ Cb the inclusion functor.
Examples 1. A sheaf on a topological space X (meaning on Open(X)) is a

presheaf F that satisfies: given an open covering U = ∪Ui of an open subset of
X and a family of si ∈ F(Ui ) such that (si )|Uj = (sj )|Ui , there exists a unique
s ∈ F(U ) such that s|Ui = si .
2. A presheaf F on Top is a sheaf if and only for any topological space X, the
realization FX of F is a sheaf on X.
3. For the coarse topology on a category C, any presheaf is a sheaf and consequently
Ce = C.
b
4. The only sheaf for the discrete topology on a category C is the constant presheaf
1 and consequently Ce ≃ 1.
If C is a site, we shall consider the functor Ȟ defined on Cb by
Ȟ(T )(X) = lim Hom(R, T ). (3.1)

−→
R∈J(X)
There exists a natural map T → Ȟ(T ) given by
T (X) / Ȟ(T )(X)

≃ ≃

Hom(hX , T ) / limR∈J(X) Hom(R, T ).
−→
Lemma 3.2.6 If R ∈ J(X) and we give ourselves a map R → T , then the

corresponding diagram
R_ / T

hX / Ȟ(T )
is commutative.
Proof. We have to show that, for all Y ∈ C, the diagram
R(Y ) / T (Y )
_

hX (Y ) / Ȟ(T )(Y )
is commutative. It means that, for all g ∈ R(Y ) seen as a map g : hY → R, the

diagram
hY ×hX R / u /T
9R _
g
≃

hY / hX / Ȟ(T )
commutes (note that the left arrow is a monomorphism with a section and therefore
an isomorphism). We are therefore reduced to the case R = hX where commutativity
follows from the definitions. ■
Proposition 3.2.7 1. The functor Ȟ is left exact.

2. If T is a presheaf (resp. a separated presheaf), then Ȟ(T ) is separated (resp.
a sheaf).
3. A presheaf T is separated (resp. a sheaf) if and only if T → Ȟ(T ) is a
monomorphism (resp. an isomorphism).
Proof. The first assertion follows from the fact that both filtered colimits and Hom
are left exact.
In order to show that Ȟ(T ) is separated, we give ourselves two maps hX ⇒ Ȟ(T )
that coincide on some R ∈ J(X), and we show that, after possibly refining R, they
come from the same map R → T . We may first assume that they come from two
maps R ⇒ T . Then, thanks to lemma 3.16, both compositions R ⇒ T → Ȟ(T ) are
the same. It is now sufficient to prove that R′ := ker(R ⇒ T ) ∈ J(X) and then
replace R with R′ . Since R ∈ J(X), it is sufficient to show that for all f ∈ R(Y ),
f −1 (R′ ) = ker(hY ⇒ T ) ∈ J(Y ). But since both compositions hY ⇒ T → Ȟ(T ) are
the same, there exists S ∈ J(Y ) such that both compositions S ,→ hY ⇒ T are the
same. Necessarily, S ⊂ ker(hY ⇒ T ) ∈ J(Y ).
As an immediate consequence, if T ,→ Ȟ(T ) is a monomorphism, then T is also
separated as a sub-presheaf of a separated presheaf. Conversely, since filtered limits
are exact, if T is separated, then we have a monomorphism T ,→ Ȟ(T ).
We shall now prove that, if T is separated, then Ȟ(T ) is a sheaf. We give
ourselves a map R → Ȟ(T ) with R ∈ J(X). We let R′ := T ×Ȟ(T ) R ⊂ R and
we first show that R′ ∈ J(X). It is sufficient to prove that, for all f ∈ R(Y ),
f −1 (R′ ) = T ×Ȟ(T ) hY ∈ J(Y ). But this is clear since the map hY → Ȟ(T ) comes
from some S → T with S ∈ J(Y ) and then, necessarily, S ⊂ T ×Ȟ(T ) hY ∈ J(Y ). We
may therefore replace R with R′ . Then our map factors as R → T ,→ Ȟ(T ) which
provides a map hX → Ȟ(T ) by definition.
Finally, it is clear that T is sheaf if and only if T ≃ Ȟ(T ). ■
3.2 Site 61
Theorem 3.2.8 If C is a site, then C

e is a reflective subcategory of Cb with exact
reflection T 7→ Te called sheafification and H(Te) = Ȟ(Ȟ(T )).
Proof. Immediate consequence of proposition 3.2.7. ■

We have
∀T ∈ C,
b F ∈ C,
e Hom(Te, F) ≃ Hom(T, F)
(a morphism T → F extends uniquely to Te → F).
Corollary 3.2.9 1. Limits in Ce are computed in C.b

2. The functor T 7→ Te preserves all colimits and finite limits. ■
It is important however to emphasize the fact that an epimorphism of sheaves
F → G is not necessarily an epimorphism of presheaves. Some people say local
epimorphism in order to remove the ambiguity. More generally, a morphism of
presheaves T → T ′ is called a local epimorphism if Te → Te′ is an epimorphism (of
sheaves).
Exercise 3.15 Show that if C is a site and T, T ′ ∈ C,
b then
im(Te → Te′ ) = im(T

^ → T ′ ).
Exercise 3.16 Show that, if C is endowed with a pretopology, then a morphism of

sheaves F → G is an epimorphism if and only if, for all X ∈ C and all s ∈ G(X),
there exists a covering (Xi → X) such that for all i ∈ I, s|Xi belongs to the image
of F(Xi ) → G(Xi ).
Example 1. If C has the discrete (resp. the coarse) topology, then Te = T (resp.
Te = ∅C ).
2. If E is any set, then we may consider the constant sheaf E eC (sometimes still
denoted by E) associated to the constant presheaf EC .
Exercise 3.17 Show that, for a set E and a sheaf F on a site C, we have the
adjunction (where 1Ce denotes a final object)
eC , F) ≃ HomSet (E, Hom(1 e, F)) (≃ Hom(1 e, F))E ).

Hom(E C C
Solution. By definition, we have Hom(E eC , F) ≃ Hom(EC , F). By left exactness, 1 e

C
is the sheaf associated to 1Cb. It follows that we also have Hom(1Ce, F) ≃ Hom(1Cb, F).
We may then apply the previous result. ■
Definition 3.2.10 The sheaf associated to an object X of a site C is X := hf

X.
The functor C → C,
e X 7→ X is left exact but not necessary fully faithful. Also,
we have
∀X ∈ C, F ∈ C,
e Hom(X, F) ≃ F(X).
Exercise 3.18 Show that if E a set, then
eTop ≃ E disc
E
(and E
eTop (X) = E if and only if X is connected).
Solution. Notice first that 1 (with 1 = {0}) is the final object of Top.
g Now, given
any sheaf F, we have on the one hand
eTop , F) ≃ Hom(1, F)E ≃ F(1)E ,
Hom(E
and on the other

Y
Hom(E disc , F) ≃ F(E disc ) ≃ F({x}) ≃ F(1)E .
x∈E
Our assertion therefore follows from Yoneda’s lemma. ■
3.2.3 Properties
Definition 3.2.11 In a category C,
1. a colimit X = limi∈I Xi is said to be universal if for all morphisms X → Y
−→
and Y ′ → Y , we have X ×Y Y ′ ≃ lim(Xi ×Y Y ′ ),
−→
2. A (regular) epimorphism f : X ↠ Y is said to be universal if for all
morphism Y ′ → Y ` , f ×Y Y ′ is a (regular) epimorphism,
3. A coproduct X = i∈I Xi is said to be disjoint if
(a) for all i ∈ I, Xi ⊂ X,
(b) for all i ̸= j ∈ I, Xi ∩ Xj = ∅.
We may also say that an effective equivalence relation is universal if the quotient
map is a universal regular epimorphism.
Exercise 3.19 Show that if C is a site, then all limits and colimits exist in Ce and
1. colimits are universal,
2. filtered colimits are exact,
3. epimorphisms are regular and universal,
4. coproducts are disjoint,
5. equivalence relations are effective.
Solution. The properties hold in Set. Therefore, they hold in C.

b We may then use
sheafification which is exact and preserves colimits. The third property however
requires some care. One first shows that a morphism which is at the same time a
monomorphism and an epimorphism is automatically an isomorphism. Then, if we
are given an epimorphism G ↠ F, we consider the image presheaf F ′ and see that
f′ = F.
F ■
Exercise 3.20 Show that if C is a site and F, G ∈ C,

e then
im(F → G) = ker (G ⇒ G ⊔F G)
3.2 Site 63
in Ce (an dual). Show that any morphism in Ce is strict.
Exercise 3.21 Show that if T is a presheaf on a site C, then
Te ≃ lim X
−→
X∈C/T
(in fancy terms, T 7→ Te is the left Kan exension of X 7→ X).
Exercise 3.22 Show that G → F is an epimorphism (resp. an isomorphism)

of sheaves if and only if for all X → F, the morphism X ×F G → X is an
epimorphism (resp. isomorphism).
Solution. The condition is necessary since epimorphisms (resp. isomorphisms) are

universal. It is sufficient because colimits preserve epimorphisms (resp. isomorphisms)
and are universal. ■
Proposition 3.2.12 Let C be a site X ∈ C and R a sieve of X. Then the following

are equivalent:
1. R is a covering sieve,
2. R
e ≃ X,
3. X ≃ limY ∈C Y .
−→ /R
Proof. The composite map R ,→ hX → hf X = X extends uniquely to u : R → X.

e
On the other hand, if R ∈ J(X), then the map R → R e extends uniquely to hX → R
e
and then to v : X = hX → R.
f e We have v ◦ u = Id e (resp. u ◦ v = IdX ) because this
R
is the unique extension of IdR (resp. IdX ).
Assume now that R e ≃ X. In order to show that R ∈ J(X), it is sufficient
to show that there exists R′ ∈ J(X) such that, given any f ∈ R′ (Y ), we have
f −1 (R) ∈ J(Y ). Under the composite map hX (X) → hf X (X) = X(X) ≃ R(X),
e
the identity provides some s ∈ R(X).
e According to formula (3.1), it comes from a
morphism ϕ : R → Ȟ(R) for some R ∈ J(X). If f ∈ R′ (Y ), we may then consider
′ ′
its image in Ȟ(R)(Y ). It comes from a morphism S → R with S ∈ J(Y ). By

construction, S ⊂ f −1 (R) and therefore also f −1 (R) ∈ J(Y ).
The other equivalence then follows from exercise 3.21. ■
Corollary 3.2.13 A family (Xi → X)i∈I generates a covering sieve if and only if
the morphism X i ↠ X is an epimorphism (in C).
` e
i∈I
Proof. The sieve R generated by (fi : Xi → X)i∈I is characterized by the

` fact that
i∈I hXi ↠ R ⊂ hX is an epimorphism of presheaves. It follows that
`
i∈I X i ↠
R ⊂ X is also an epimorphism of sheaves. Our assertion therefore follows from
e
proposition 3.2.12. ■
Definition 3.2.14 A topology is subcanonical if any representable presheaf is a

sheaf. The canonical topology on a category C is the finest subcanonical topology.
Exercise 3.23 Show that a topology on C is subcanonical if and only if for all
X ∈ C, X = hX if and only if ょ factors through Ce if and only if X 7→ X is fully
faithful.
Exercise 3.24 Show that one may always replace a site C with a subcanonical site
C ′ in such a way that Ce′ ≃ C.
e
Exercise 3.25 Show that if C is a site and Ce is endowed with its canonical topology,
then Ce ≃ Ce
e
Solution. We may assume that the topology of C is subcanonical. It is then sufficient

to check that a quasi-inverse for the Yoneda embedding
ょ : Ce ,→ C, F 7→ hF
ee
is simply given by

op ょ

G op G
Ceop → Set 7→ C ,→ C → Set .
e
Clearly, if F ∈ Ce and X ∈ C, then

hF (hX ) = Hom(hX , F) = F(X).
For the converse, let us first show that if F ′ ∈ C,
e then (hX → F ′ )X∈C ′ generates a
/F
covering sieve for the canonical topology. Indeed, given any F ∈ C, e we have
hF (F ′ ) = Hom(F ′ , F) = Hom( lim hX , F) = lim Hom(hX , F) = lim hF (hX ).

−→ ←− ←−
X∈C/F ′ X∈C/F ′ X∈C/F ′
Therefore, if G ∈ Ce and we set F(X) = G(hX ), then we see that

e
G(F ′ ) = lim G(hX ) = lim F(X)

←− ←−
X∈C/F ′ X∈C/F ′
only depend on F (and then G ≃ hF ). ■
3.3 Topos
3.3.1 Pretopos
Definition 3.3.1 A pretopos is a category C such that
1. finite limits exist,
2. finite coproducts exist and are disjoint and universal,
Examples 1. If C is any site, then Ce is a pretopos.

2. The category of (finite) sets is a pretopos.
3. The category of compact Hausdorff spaces is a pretopos.
4. The category Top is not a pretopos (epimorphisms are not always regular).
3.3 Topos 65
Exercise 3.26 Show that, in a pretopos, any morphism factors uniquely (up to
a unique isomorphism) as an epimorphism followed by a monomorphism (an
epi-mono factorization).
Solution. Let f : X → Y be a morphism. We let R := X ×Y X and X := X/R =

π ι
coker(R ⇒ X). There exists a factorisation X ↠ X → Y and we shall prove that ι
is a monomorphism. If g, h : Z → X, we may consider the cartesian diagram
Ze π
e //Z
(e
g ,e
h) (g,h)

π×π/ /
X ×X X × X.
It is easily checked that if ι ◦ g = ι ◦ h, then f ◦ ge = f ◦ e

h. It follows that (e
g, e
h)
factors through R. This implies that g ◦ π e = h◦π e. Since, by construction, π
e is
a (regular) epimorphism, we obtain g = h. Assume now that there exists another
π′ ι′ π s
epi-mono factorization X ↠ X ′ ,→ Y . Then, necessarily, π ′ factors as X → X → X ′
and we must have ι = ι′ ◦ s:
π′ //
X X′
>> _
s
π ι′
.
ι / Y.
X
It follows that the diagram X ×X ′ X ⇒ X → X is commutative. Since π ′ is regular,

we have X ′ = coker(X ×X ′ X ⇒ X) and there exists necessarily a factorization
π′ t
X → X ′ → X. On easily checks that t is an inverse for s. ■
Exercise 3.27 Show that a pretopos is balanced : A morphism which is at the same
time a monomorphism and an epimorphism is automatically an isomorphism.
Exercise 3.28 Show that, in a pretopos, any morphism is strict.
In a pretopos, if X1 , X2 ⊂ X, their union is X1 ∪ X2 := im (X1

`
X2 → X).
Exercise 3.29 Show that, in a pretopos, if Y1 , Y2 ⊂ Y , and f : X → Y is any
morphism, then
f −1 (Y1 ∪ Y2 ) = f −1 (Y1 ) ∪ f −1 (Y2 ).
Actually, the set of subobjects of a given object of a pretopos is a bounded lattice 1

and pullback is a morphism 2 of bounded lattices.
1
A (pre-) ordered set with all finite limits and colimits.
2
It preserves finite limits and colimits.
3.3.2 Precanonical topology

Definition 3.3.2 The precanonical pretopology on a pretopos C is the pretopology
made of finite families (Xi → X)i∈I such that i∈I Xi ↠ X is an epimorphism.
`
Alternatively, a family (Xi → X)`

i∈I generates a covering sieve if and only if there
exists a finite J ⊂ I such such that i∈J Xi ↠ X is an epimorphism.
Exercise 3.30 Show that the precanonical topology is generated by epimorphisms
and finite disjoint unions.
Exercise 3.31 Show that the precanonical topology is subcanonical.
Solution. We have to show that, given any Y ∈ C and any covering family (Xi →
X)i∈I , then
!
Y Y
hY (X) ≃ ker hY (Xi ) ⇒ hY (Xi ×X Xj ) .
i∈I i,j∈I
It is sufficient to consider a finite disjoint union X `

= i∈I Xi or
Q an epimorphism
`
X → X. In the first case, this boils down to hY ( i∈I Xi ) = i∈I hY (Xi ) which
′
holds just because hY is left exact. In the second case, we have X = X ′ /R with
R := X ′ ×X X ′ and we have to show that
Hom(X ′ /R, Y ) ≃ ker(Hom(X ′ , Y )) ⇒ Hom(R, Y ))),
which is the very definition of a quotient. ■
Proposition 3.3.3 Let C be a pretopos endowed with the precanonical topology.

1. If X1 , . . . , Xr ∈ C, then
X1 ⊔ . . . ⊔ Xr ≃ X 1 ⊔ . . . ⊔ X r .
2. If R is an equivalence relation on X, then
X/R ≃ X/R.
Proof. 1. We proceed by induction. We have ∅ = ∅ and if F is a sheaf and

X, Y ∈ C, then
Hom(X ⊔ Y , F) = F(X ⊔ Y ) ≃ F(X) × X(Y )
= Hom(X, F) × Hom(Y , F) = Hom(X ⊔ Y , F).
2. The same kind of arguments also works for the other assertion. More precisely,
the sequence
F(X/R) → F(X) ⇒ F(R)
is exact and may be rewritten as
Hom(X/R, F) → Hom(X, F) ⇒ Hom(R, F). ■
Be careful however that the functor X 7→ X does not preserve finite colimits in
general.
3.3 Topos 67
Exercise 3.32 Show that, if C be a pretopos endowed with the precanonical topol-
ogy, then a morphism f : X → Y is a monomorphism (resp. epimorphism, resp.
isomorphism) if and only if f : X → Y is a monomorphism (resp. epimorphism,
resp. isomorphism).
Solution. The equivalence is true for isomorphisms because the topology is subcanon-
ical. The direct implication holds for monomorphisms since X 7→ X is left exact.
If f is an epimorphism, then it is regular and Y ≃ X/R so that Y = X/R ≃ X/R
which shows that f also is an epimorphism. In general, there exists an epi-mono
factorization X ↠ Z ,→ Y providing an epi-mono factorization X ↠ Z ,→ Y . If f
is a monomorphism then X = Z so that X = Z and f is a monomorphism. If f is
an epimorphism, then Z = Y so that Z = Y and f is an epimorphism. ■
3.3.3 Generator
Definition 3.3.4 A set S ⊂ C is a set of generators (or separators) for a category C
if the functor hG is faithful. When S = {G}, we say that G is a generator.
Q
G∈S
In down to earth terms, it means that if f1 ≠ g1 : X → Y , then there exists

g : G → X with G ∈ S such that f1 ◦ g ̸= f2 ◦ g.
Examples 1. 1 := {0} is a generator for Set.

2. A is a generator for A-Mod.
3. If C is a category, then C is a set of generators for C.
b
Exercise 3.33 Show that, if C is a site and S ⊂ C is a topologically generating

subset, then {X, X ∈ S} is a set of generators for C.
e
Exercise 3.34 Show that, if C has all coproducts, then S is a set of generators if
and only if we have for all X ∈ C an epimorphism
a
G ↠ X.
G∈S,f :G→X
3.3.4 Topos
Definition 3.3.5 A category T is a topos if
1. there exists a small set of generators,
2. finite limits exist,
3. coproducts exist and they are disjoint and universal,
4. equivalence relations are effective and universal.
Examples 1. The categories 1, Set and G-Set are topos.

2. If C is a category, then Cb is a topos.
3. If C is a site, then Ce is a topos.
4. The category of sheaves of sets on a topological space X is a topos.
5. The category of espaces étalés over a topological space X is a topos.
6. The category of condensed sets is a topos.
7. The category of compact Hausdorff spaces is not a topos.
Theorem 3.3.6 — Giraud. For a category T , the following are equivalent:

1. T is a topos,
2. T has a small set of generators and all sheaves for the canonical topology
are representable,
3. there exists a site C such that T ≃ C,
e
4. there exists a category C such that T is a reflective subcategory of Cb with
exact reflection.
Proof. For (1) ⇒ (2), given a sheaf F, we set X := limY ∈T Y . Then, one can
−→ /F
check that F ≃ hX . For (2) ⇒ (3), we can almost choose C = T with its canonical
3
topology. Also, it follows from exercises 3.19 and 3.33 that (3) ⇒ (1). We also proved
in proposition 3.16 that (3) ⇒ (4). For (4) ⇒ (1), it is sufficient to notice that C
b is
a topos and use the reflection. ■
If follows that, in a topos T , all limits and colimits exist, colimits are universal,
epimorphisms are regular and universal, filtered colimits are exact and equivalence
relations are effective. Actually, any formula that involves colimits and finite limits
that hold for sets also holds in T . In particular, a topos is a pretopos and therefore,
any morphism has a unique epi-mono factorization. Subobjects form a bounded
lattice and pulling back is a morphism of bounded lattices.
Corollary 3.3.7 If C is a small set of generators of a topos T and C has the induced
topology, then T ≃ C.e ■
3.3.5 Internal Hom

Unless otherwise specified, we will always consider a topos T as a site for its canonical
topology so that T ≃ Te . We will then identify X ∈ T with hX and with X and
write for example Y (X) = Hom(X, Y ).
The topos T is automatically endowed with the maximal pretopology (a covering
is a family that generates a covering sieve).
Exercise 3.35 Show that, in a topos T , a family (Xi → X)i∈I is a covering if and
only if the map Xi → X is an epimorphism.
`
i∈I
Example A family of maps (fi : Xi → X)i∈I is a covering in the topos Set (for the
canonical topology) if and only if it is jointly surjective:
∀x ∈ X, ∃i ∈ I, ∃xi ∈ Xi , fi (xi ) = x.
Exercise 3.36 Show that a presheaf T on a topos T is a sheaf if and only if it

preserves all limits.
3
In order to comply with our set-theoretic hypothesis, it is however necessary to take for C the
full subcategory generated by the small set of generators and endow it with topology induced by
the canonical topology of T .
3.3 Topos 69
Solution. If R is a covering sieve of X ∈ T , then it follows from proposition 3.2.12

that
X ≃ lim X ′ .
−
′
→
X ∈T/R
If T preserves limits, then
T (X) ≃ lim T (X ′ )
←
′
−
X ∈T/R
and this shows that T is a sheaf. The converse follows from proposition 1.4.5. ■
Proposition 3.3.8 If T is a topos and Y, Z ∈ T , then the presheaf
X 7→ Hom(X × Y, Z)
is representable by an object Hom(Y, Z) ∈ T .
Proof. It is sufficient to show that it is a sheaf but this follows from the fact that
colimits are universal in a topos. ■
As a consequence, a topos is cartesian closed : there exists a natural isomorphism
(currying)
Hom(X × Y, Z) ≃ Hom(X, Hom(Y, Z)) (≃ Hom(Y, Z)(X)).
In particular, for fixed Y , the functor X 7→ X × Y is adjoint to the functor Z 7→

Hom(Y, Z).
Example If F, G are two (pre-) sheaves on a topological space X, then Hom(F, G)(U ) =
Hom(F|U , G|U ) (with F|U (V ) := F(V ) for V ⊂ U ).
Exercise 3.37 Show that, in a topos,
Hom(X × Y, Z) ≃ Hom(X, Hom(Y, Z)).
Solution. It is sufficient to notice that, given any object T , we have a natural

isomorphism
Hom(T, Hom(X × Y, Z)) ≃ Hom(T × X × Y, Z))

≃ Hom(T × X, Hom(Y, Z)))
≃ Hom(T, Hom(X, Hom(Y, Z))). ■
Exercise 3.38 Show that, if 1 denote the final object of a topos, then
1. Hom(1, X) ≃ X,
2. Hom(X, Y ) = Hom(X, Y )(1).
The last assertion means that T is enriched over itself.

Exercise 3.39 Show that, if C is a site, T ∈ Cb and F ∈ C,

e then Hom e(Te, F) =
C
HomCb(T, F).
Solution. For X ∈ C, we have the sequence of isomorphisms
HomCe(Te, F)(X) = Hom(X, HomCe(Te, F))

= Hom(X × Te, F)
= Hom(h^X × T , F)
= Hom(hX × T, F)
= Hom(hX , HomCb(T, F))
= HomCb(T, F)(X). ■
3.3.6 Quasi-compact/separated
See for example Lurie.
Definition 3.3.9 An object X of a site C is said to be quasi-compact if, given any
family (Xi → X)i∈I that generates a covering sieve, there exists a finite subset J
of I such that the sieve generated by (Xi → X)i∈J is also a covering.
In particular,
` an object X of a topos T is quasi-compact if and only ` if, given an
epimorphism i∈I Xi ↠ X, there exists a finite subset J of I such that i∈J Xi ↠ X
is an epimorphism. Actually, quasi-compactness may always be checked in a topos:
Exercise 3.40 Show that an object X of a site C is quasi-compact if and only if X
is quasi-compact (for the canonical topology).
Solution. Direct implication is clear. For the converse,`consider an epimorphism

i Fi → X. There exists for each Fi an epimorphism j X ij → Fi . Then, there
`
exists for each X ij → X a covering (Xijk → Xij )k and morphisms Xijk → X in C
giving rise to X ijk → X ij → F → X. We may then pick up a finite number of
i, j, k. ■
Exercise 3.41 Show that if X has a finite covering by quasi-compacts objects,

then X is quasi-compact.
Proposition 3.3.10 A sheaf F on a pretopos C (for the precanonical topology) is

quasi-compact if and only if there exists an epimorphism X ↠ F with X ∈ C.
Proof. It follows from exercise 3.34 that there exists an epimorphism i∈I X i ↠ F
`
with X`i ∈ C. If F is quasi-compact, we may assume that I is finite and set
X := i∈I Xi . For the converse, we may assume that F = X. It follows from the
very definition of the topology that X is quasi-compact. Or, thanks to exercise 3.40,
equivalently, that X is quasi-compact. ■
Definition 3.3.11 An object X of a topos T is said to be quasi-separated if given

any Y → X and Z → X with Y and Z quasi-compacts, then Y ×X Z is also
quasi-compact.
3.3 Topos 71
Exercise 3.42 1. Show that a subobject of a quasi-separated object is quasi-
separated.
2. Show that a coproduct of quasi-separated objects is quasi-separated.
3. Show that a filtered colimit under monomorphisms of quasi-separated objects
is quasi-separated.
Solution. 1. Assume X is quasi-separated and X ′ ⊂ X. We give ourselves Y → X

and Z → X with Y and Z quasi-compacts. Then, Y ×X ′ Z = Y ×X Z is also
quasi-compact.
2. Assume that i∈I Xi . If Y → X is any morphism, then we have
`
X =
Y = i∈I Yi with Yi = Y ×X Xi . In other words, the family (Yi ,→ Y )i∈I is a
`
covering. Therefore, if Y is `
quasi-compact, we can then replace I with a finite
subset J and we have Y = i∈J Yi . Moreover, a summand of a quasi-compact
is always quasi-compact - as we shall show below - so that each Yi is quasi-
compact. Of course,`if Z → X is another morphism with Z quasi-compact, we
can also write Z = i∈J Zi and we may assume that this is the same finite J.
It is then formal to check that
a
Y ×X Z ≃ Yi ×Xi Zi .
i∈J
If we assume that all Xi are quasi-separated, then (Yi ×Xi Zi → Y ×X Z)i∈J is

a finite covering by quasi-compact objects and it follows that Y ×X Z is also
quasi-compact. It remains to show that a summand of a quasi-compact is itself
quasi-compact. But if we are given a covering (Xi → X)i∈I and we know that
X ⊔ Y is quasi-compact, we can then consider the covering made of the Xi ’s
and Y of X ⊔ Y . It has a finite refinement and we are done.
3. If X = limi∈I Xi is any colimit, then the corresponding morphism i∈I Xi ↠ X
`
−→
is an epimorphism (as usual, this follows formally from the analogous assertion in
Set). In other words, the family (Xi → X)i∈I is a covering. In particular, when
X is quasi-compact, there exists a finite subset J of I such that i∈J Xi ↠ X
`
is an epimorphism and therefore X = limi∈J Xi . In the case of a filtered colimit,
−→
if k is any cocone for J in I, then we will have X = Xk . Not assuming X quasi-
compact anymore, let Y → X be a morphism with Y quasi-compact. Then,
Y = limi∈I Yi with Yi = Y ×X Xi . If the colimit is filtered, then there exists k
−→
such that Y = Yk . In other words, there exists a factorization Y → Xk → X of
the original morphism. If Z → Y is another morphism with Z quasi-compact,
then there exists also a factorization Z → Xk → X and we may assume that
this is the same k since I is filtered. Finally, if we assume that Xk ⊂ X, then
we have Y ×X Z = Y ×Xk Z which is quasi-compact if we assume that Xk is
quasi-separated.
■
Theorem 3.3.12 If C is a pretopos (endowed with the precanonical topology),

then the functor X 7→ X induces an equivalence between C and the category of
quasi-compact quasi-separated sheaves on C.
Proof. Since the topology is subcanonical, the functor is fully faithful and it easily
follows from proposition 3.3.10 that X is always quasi-compact quasi-separated.
Assume conversely that F is quasi-compact quasi-separated and let X ↠ F (resp.
X ′ ↠ X ×F X) be an epimorphism. The composite morphism
X ′ ↠ X ×F X ,→ X × X ≃ X × X
Is an epi-mono factorization. It comes from a morphism X ′ → X × X in C that has

an epi-mono factorization X ′ ↠ R ,→ X × X. It follows that X ′ ↠ R ,→ X × X
is also an epi-mono factorization. By uniqueness, R ≃ X ×F X is an equivalence
relation with quotient F. It follows that R also is an equivalence relation and that
F = X/R ≃ X/R. ■
3.4 Morphism of topos (optional)

3.4.1 Definition
Definition 3.4.1 A morphism of topos f : T −→ T ′ is a couple of functors
f −1 : T ′ −→ T , f∗ : T −→ T ′

with f −1 exact and adjoint to f∗ .

We say that f is an embedding of topos when f∗ is fully faithful. Note that if f is
a morphism of topos, then both f −1 and f∗ preserve algebraic structures.
Examples 1. If C is a site, then there exists an embedding of topos i : Ce ,→ Cb

given by i−1 (T ) = Te and i∗ F = F.
2. There exists a unique morphism of topos p : T → Set. We have p∗ F = F(1T )
and p−1 E = E eC .
Exercise 3.43 Show that, if E = limi∈I Ei in Set, then E

eC = lim E e .
i∈I iC
−→ −→
Exercise 3.44 Show that if f : T −→ T ′ is a morphism of topos, then f −1 E

eT ′ =
eT .
E
Exercise 3.45 Show that if f : T −→ T ′ is a morphism of topos, then
f∗ Hom(f −1 X ′ , Y ) ≃ Hom(X ′ , f∗ Y ).
One defines in the obvious way the composition of two morphisms of topos so
that we may consider the topos as the objects of a category.
3.4.2 Presheaves
Theorem 3.4.2 If g : C → C ′ is any functor, then the functor
gb−1 : Cb′ / C,
b T′ / T ′ ◦ g.
has an adjoint gb! (resp. a coadjoint gb∗ ) : Cb → Cb′ .
Proof. Follows from proposition 1.5.10. ■
As a consequence, gb! preserves all colimits, gb−1 preserves all limits and colimits and
gb∗ preserves all limits. In particular, both gb−1 and gb∗ preserve algebraic structures.
Note also that we can recover the original functor g from the equality hg(X) = g! hX .
Corollary 3.4.3 The functor g : C → C ′ induces a morphism of topos g

b : Cb → Cb′
given by gb−1 and gb∗
Proof. The existence of gb! makes sure that gb−1 is left exact. ■
Exercise 3.46 Any X ∈ C may be seen as a functor X : 1 → C giving rise to a

morphism of topos
b : Set → C.
X b
Make X
b! , X
b −1 and X
b∗ explicit.
Solution. First of all, we have X

b −1 (T ) = T (X). Next, (X b! (E))(Y ) = E if there
exists Y → X and ∅ otherwise. Finally, (X b∗ (E))(Y ) = E if there exists X → Y and
{0} otherwise. ■
Examples 1. If f : Y → X is a continuous map, it induces a functor
(g = ) f −1 : Open(X) → Open(Y ),
and by composition, a functor
g −1 = ) fb∗ : Open(Y
(b \ ) → Open(X),
\
that has adjoint and coadjoint
g! = ) fb−1
(b g∗ = ) fb! : Open(X)
and (b \ → Open(Y
\ ).
2. Explicitly, the adjoint functors are given by
fb−1 (T )(V ) = lim T (U ), fb∗ (T )(U ) = T (f −1 (U )),

−→−1
V ⊂f (U )
and fb! (T )(V ) = lim T (U ).

←−
f −1 (U )⊂V
3. There exists a morphism of topos fb : Open(Y

\ ) → Open(X).
\
By analogy, we may also denote a functor as f −1 := g : C → C ′ (even if this is

not an inverse image), and consequently write:
fb−1 := gb! , fb∗ := gb−1 , fb! := gb∗ ,
so that now hf −1 (X) = fb−1 (hX ).

Exercise 3.47 Let X be a topological space. One can see a point x ∈ X as a
continuous map 1 → X. Make x b−1 , x
b∗ and x
b! explicit. If T is a presheaf, then
b T is called the stalk of T at x. Show that T is Q
Tx := x −1
a separated presheaf if
and only if for all open subset U of X, the map T (U ) → x∈U Tx is injective.
3.4.3 Morphisms of sites

Definition 3.4.4 If C and C ′ are two sites, then a functor f −1 := g : C → C ′ is said
to be continuous if the functor
fb∗ : Cb′ → C,
b T ′ 7→ T ′ ◦ f −1
preserves sheaves.
We shall then denote by fe∗ : Ce′ → Ce (or f∗ for short) the induced functor.
Exercise 3.48 Show that if C and C ′ are two sites and if f −1 : C → C ′ is a
continuous functor, then the functor
^
fe−1 : Ce → Ce′ , F 7→ fb−1 (F)
is adjoint to f∗ .
Since we always have f −1 (X) = fe−1 (X), we may simply write f −1 instead of fe−1
when there is no ambiguity.
Definition 3.4.5 A morphism of sites is a continuous functor f −1 : C → C ′ such
that fe−1 is (left) exact.
Equivalently, it means that
f := (f −1 , f∗ ) : Ce′ → Ce
is a morphism of topos. Note that there exists a commutative diagram of morphisms

of topos
fb
CbO ′ / CbO
? ?
f
Ce′ / C.
e
Examples 1. If we endow two topos T and T ′ with their canonical topology,

then a morphism of sites f : T ′ → T is nothing but a morphism of topos.
2. If f : Y → X is a continuous map of topological spaces, then the inverse image

functor f −1 defines a morphism of sites and therefore a morphism of topos
^ ) → Open(X).
f : Open(Y ^
If G is a sheaf on Y , we have for any open subset U of X,
f∗ G(U ) = G(f −1 (U )).
And if F is a sheaf on X, then f −1 F is the sheafification of
V 7→ lim F(U ).
−→
f (V )⊂U
Exercise 3.49 Make explicit the morphism of topos coming from the final map
p : X → 1 when X is a topological space. Show that if f : Y → X is a continuous
map, then E eX .
eY = f −1 E
Solution. We have p∗ F = Γ(X, F) and p−1 E = E

eX . With q : Y → 1, we have
q = p ◦ f and therefore E
eY = q E = f p E = f −1 E
−1 −1 −1 eX . ■
Exercise 3.50 Make explicit the morphism of topos coming from a point x : 1 → X
when X is a topological space.

E if x ∈ U
Solution. We have (x∗ E)(U ) = and x−1 F = Fx . ■
∅ if x ∈
/U
Exercise 3.51 Show that if T is a presheaf on a topological space X and x ∈ X,

then Tex = Tx .
Exercise 3.52 Show that, if C has fibered products and f −1 : C → C ′ is a functor

between two sites which is left exact and preserves covering families, then f −1 is
morphism of sites.
Definition 3.4.6 A functor g : C ′ → C between two sites is said to be cocontinuous

if gb∗ preserves sheaves.
Exercise 3.53 Show that if g is cocontinuous, then the induced functor g∗ : Ce′ → Ce
extends uniquely to a morphism of topos g := (g −1 , g∗ ) : Ce → Ce′ .
Exercise 3.54 Show that, if g : C ′ → C be a functor between two sites and for any
X ′ ∈ C ′ , any covering family of g(X ′ ) is the image of a covering family of X ′ , then
g is cocontinuous.
3.4.4 Induced topology

Definition 3.4.7 If C ′ is a site and f −1 : C → C ′ is any functor, then the induced
topology on C is the finest topology on C making f −1 : C → C ′ continuous.
Exercise 3.55 Show that if C is a site, then the topology of C is induced by the
(canonical) topology of C.
e
If C is a site and T ∈ C,
b we can consider the forgetful functor jT : C/T → C and
endow C/T with the induced topology.
Exercise 3.56 Show that a sieve R of s ∈ T (X) is a covering in C/T if and only if
jT ! R is a covering sieve of X in C.
b
Unfortunately, the adjoint jT ! is not left exact in general and we do not get a
morphism of sites. However:
Exercise 3.57 Show that, if C is a site and T ∈ C,b then jT is also cocontinuous
and there exists therefore a morphism of topos
jT : Cf
/T → C.
e
Exercise 3.58 Show that, if C is a site and T ∈ C,

b then Cf
/T ≃ C/Te . Show
e
that jT ! (resp. jT−1 ) corresponds to the forgetful functor Ce/Te → Ce (resp. to
F 7→ (F × Te → Te).
Exercise 3.59 Show that, in a topos T , we have

−1 −1 −1
jX Hom(X, Y ) ≃ Hom(jX Y, jX Z) and
−1
Hom(jX! X, Y ) ≃ jX∗ Hom(Y, jX Z).
Exercise 3.60 Show that, in a topos T , we have

−1
Hom(X, Y ) ≃ jX∗ jX Y and
Hom(X, Y )(Z) = Hom(jZ−1 X, jZ−1 Y ).
3.4.5 Topological spaces

If X is a topological space, we may then consider the site Top/X of all topological
spaces over X. The inclusion map
Open(X) ,→ Top/X
is continuous, cocontinuous and left exact giving rise to two morphisms of topos
φX
/ ^
Top/X o Open(X)
^
ψX
with φX ◦ ψX = Id and φX∗ = ψX

−1
.
Any continuous map f : Y → X will provide a commutative diagram of topos
φY
Top
^ / ^ )
Open(Y
/Y
j f

φX
Top
^ / Open(X).
^
/X
Let X be any topological space. If F is a sheaf on Top/X and Y is a topological

space over X, then the realization of F on Y is
FY := φY ∗ F/Y .
Any morphism f : Z → Y over S will induce a morphism
αf : f −1 FY → FZ
between the realizations.

Exercise 3.61 Show that, giving a sheaf F on Top/X is equivalent to giving the
set of all FY and compatible morphisms αf : f −1 FY → FZ .
We shall call a sheaf F on Top/X crystalline if all the maps αf are isomorphisms
f −1 FY ≃ FZ .
Exercise 3.62 Show that realization F 7→ FX induces an equivalence between
crystalline sheaves on Top/X and sheaves on X.
Exercise 3.63 Show that the category Et(X) is equivalent to the category of
crystalline sheaves on Top/X , and consequently to the category of sheaves on X.
This shows that Et(X) is a topos because Et(X) ≃ Open(X).

^
Exercise 3.64 Show that a sheaf on a topological space X is quasi-compact (for
the canonical topology) if and only if the corresponding espace étalé is compact.
4. Condensed sets
Again, we shall not worry much about set-theoretical issues and work in a fixed
universe. Note however that there exists an unconditional theory of condensed sets
(that does not depend on the choice of a universe). There also exists a theory of light
condensed sets.
4.1 Condensed set

4.1.1 Compact Hausdorff spaces
We shall denote by CHaus the category of compact Hausdorff spaces and continuous
maps. We shall also write • for the one point space (final object).
Proposition 4.1.1 In CHaus,
1. all limits and colimits exist,
2. finite coproducts are disjoint and universal,
Proof. Recall that Top has all limits and colimits and that they are computed in
Set. Moreover, Set is a topos and therefore, all the above properties are satisfied. In
particular, it follows from proposition 2.1.7 that CHaus has all limits and colimits
with limits computed in Top. Moreover, any colimit in Top of compact Hausdorff
spaces which is itself compact Hausdorff is also the colimit in CHaus. This applies
in particular to finite coproducts, epimorphisms and closed equivalence relations. It
follows that finite coproducts are disjoint, epimorphisms are universal and universal
and that equivalence relations are effective. Finally, a continuous surjective map
between compact Hausdorff spaces is a quotient map and therefore. In other words,
an epimorphism is regular. ■
80 Chapter 4. Condensed sets
Corollary 4.1.2 The category CHaus is a pretopos (see definition 3.3.1). ■
We may also stress out the fact that a continuous map is a monomorphism (resp.
an epimorphism, an isomorphism) if and only if it is injective (resp. surjective, resp.
bijective). Any monomorphism (resp. epimorphism) is regular: any continuous
injective (resp. surjective) map is a homeomorphism onto a closed subspace (resp. a
quotient map).
4.1.2 Definition
We shall systematically endow CHaus with its precanonical topology: a family
(Si → S)i∈I of continuous maps of compact Hausdorff spaces is a covering if and only
if I is finite and i∈I Si ↠ S is surjective. Recall that this topology is subcanonical.
`
Definition 4.1.3 A condensed set is a sheaf of sets on CHaus.

We shall denote by Cond := CHaus
^ the category of condensed sets.
Examples 1. We shall see that, if X is any topological space, then S 7→ X(S) :=

C(S, X) defines a condensed set.
2. If X is a topological space, then X(N) is the set of convergent sequences
(xn )n∈N , together with a specified limit x∞ .
3. There exists a unique condensed set Q such that Q(S) = C(S, R)/C(S, Rdisc )
when S is Stonean.
Exercise 4.1 Show that a presheaf of sets X on CHaus is a condensed set if and
only if
1. given any compact Hausdorff spaces S1 , . . . , Sr ,
X(S1 ⊔ . . . ⊔ Sr ) ≃ X(S1 ) × · · · × X(Sr ) and (4.1)
2. given any closed equivalence relation R on a compact Hausdorff space S,
X(S/R) ≃ ker (X(S) ⇒ X(R)) (4.2)
The first condition can be reduced to
X(∅) ≃ {0} and X(S ⊔ S ′ ) = X(S) × X(S ′ )
for compact Hausdorff spaces S, S ′ . Also, the second one may be rephrased by saying
that, if f : S ′ ↠ S is a continuous surjection, then the sequence
X(S) → X(S ′ ) ⇒ X(S ′ ×S S ′ )
is (left) exact.
Theorem 4.1.4 The category Cond is a topos.
Proof. Follows from theorem 3.3.6. ■

4.1 Condensed set 81
Corollary 4.1.5 In the category Cond, all limits and colimits exist and
1. colimits are universal,
2. filtered colimits are exact,
4. equivalence relations are effective. ■
More generally, any formula that involves colimits and finite limits that holds for
sets also holds for condensed sets. Also that any morphism has a unique epi-mono
factorization and subobjects form a bounded lattice (and pulling back is a morphism
of bounded lattices).
The topos Cond will be endowed with its canonical topology (a sheaf is a
representable presheaf).
4.1.3 Free compact Hausdorff spaces

Let us consider now the subcategory FCHaus of free compact Hausdorff spaces and
continuous maps. Any compact Hausdorff space is a quotient of a free compact
Hausdorff space and any such quotient may be extended to a free presentation. Also,
any free compact Hausdorff space is a projective object of the category of compact
Hausdorff spaces, or equivalently a Stonean space. Any surjection from a compact
Hausdorff space to a Stonean space (and, in particular, to a free compact Hausdorff
space) has a section.
Exercise 4.2 Show that finite disjoint unions define a pretopology on FCHaus.
We shall systematically endow FCHaus with this topology: a family (Fi → F )i∈I
of continuous
` maps of free compact Hausdorff spaces is a covering if and only if I is
finite and i∈I Fi ≃ F is bijective.
Exercise 4.3 Show that a presheaf of sets X on FCHaus is a sheaf if and only if
it preserves products (condition (4.1) above).
Proposition 4.1.6 Inclusion induces an equivalence of categories
^ ≃ FCHaus.
Cond := CHaus ^
Proof. Clearly, any sheaf X on CHaus will restrict to a sheaf on FCHaus. Conversely,
if X be a sheaf on FCHaus and S any compact Hausdorff space, we set
X(S) := lim X(F )

←−
F →S
where F runs through all free compact Hausdorff spaces over S. If
F′ // F // S (4.3)
is a free presentation (meaning that both F ↠ S and F ′ ↠ F ×S F are surjective),

then there exists a canonical map
X(S) → ker (X(F ) ⇒ X(F ′ )) . (4.4)

Conversely, if x lives in the right-hand side and F ′′ is a free compact Hausdorff

space, then any continuous map F ′′ → S will lifts to some f : F ′′ → F and we
may consider f −1 (x) ∈ X(F ′′ ). If we are given two such liftings f1 , f2 : F ′′ → F ,
then they induce a map F ′′ → F ×S F that lifts to a map f ′ : F ′′ → F ′ . This
implies that f1−1 (x) = f2−1 (x) and we have defined a map in the other direction.
It is straightforward so check that this is an inverse for the map in (4.4), which
is therefore bijective. Note in particular that, if F ↠ S is surjective with F free,
then necessarily X(S) ⊂ X(F ). We need to show that our presheaf X is a sheaf.
Condition (4.1) follows from the fact that a disjoint union of free presentations is a
free presentation and it remains to check condition (4.2). If R is a closed equivalence
relation on a compact Hausdorff spaces S, then there exists a free compact Hausdorff
space F (resp. F ′ ) and a continuous surjection F ↠ S (resp. F ′ ↠ F ×S/R F ). The
commutative diagram
F′ // F ×S/R F // F // S/R

R S ×S/R S // S // S/R
provides a commutative diagram
X(S/R) / X(F ) // X(F ′ )

O O
? ?
X(S/R) / X(S) // X(R).
Since F is free, the morphism F ↠ S/R has a section and the upper sequence is
therefpre left exact. The same holds for the bottom one. ■
Our assertion means that any sheaf on FCHaus extends uniquely to a condensed
set. We will say for short that the sheaf is a condensed set.
There exists analogous statements with many intermediate categories such as
Stone or Stonean spaces. More precisely, the category of Stone spaces is endowed
with the pretopology made of jointly surjective maps (like CHaus) and the category
of Stonean spaces with the pretopology made of disjoint unions (like FCHaus). This
was the original approach to the theory.
Proposition 4.1.7 A morphism X → Y of condensed sets is an epimorphism if
and only if, for all free compact Hausdorff space F , the map X(F ) → Y (F ) is
surjective.
Proof. It is equivalent to prove this property in FCHaus.^ Let p : X ↠ Y be an

epimorphism, ` F a free Hausdorff space and y ∈ Y (F ). Then there exists a disjoint
covering F = ni=1 Fi and,
Q for all i = 1, . . . , n, an xi ∈ X(Fi ) such that p(xi ) = y|Fi .
If we set x = (xi )i=1 ∈ X(Fi ) = X(F ), we will have p(x) = y.
n
■
4.2 Topological space and condensed set

4.2.1 Associated condensed set
Lemma 4.2.1 If X is any topological space, then the presheaf
h
X : CHausop ,→ Topop −→
X
Set, S 7→ C(S, X)
is a condensed set.
Proof. Clearly X(∅) = C(∅, X) = {∅} and
X(S ⊔ S ′ ) = C(S ⊔ S ′ , X) = C(S, X) × C(S, X) = X(S) × X(S ′ ).
It remains to show that the sequence
C(S/R, X) → C(S, X) ⇒ C(R, X)
is exact when R is a closed equivalence relation on S. But if f : S → X is a continuous
map which is compatible with R, then it factors uniquely through S/R. ■
Proposition 4.2.2 1. If S1 , . . . , Sr are compact Hausdorff spaces, then
S1 ⊔ . . . ⊔ Sr ≃ S 1 ⊔ . . . ⊔ S r .
2. If R is a closed equivalence relation on a compact Hausdorff space S, then
S/R ≃ S/R.
Proof. This was shown in proposition 3.3.3. ■

Be careful that the functor S 7→ S does not preserve finite colimits of compact
Hausdorff spaces in general. However, a morphism f : S → T of compact Hausdorff
spaces is a injective (resp. surjective, resp. bijective) if and only if f : S → T is a
monomorphism (resp. an epimorphism, resp. an isomorphism).
Exercise 4.4 Show that, if F is a free compact Hausdorff space, then F is a
projective condensed set. Show that the category Cond has enough projectives.
Solution. We know from proposition 4.1.7 that, if Y ↠ X is an epimorphism of

condensed sets, then Y (F ) → X(F ) is surjective. Thanks to Yoneda lemma, it
exactly means that Hom(F , Y ) → Hom(F , X) is surjective. This shows that F is
projective. The second assertion follows from the fact that a coproduct of projectives
is projective. ■
Let X be a condensed set. If S is a compact Hausdorff space, then there exists a
natural (Yoneda) bijection
Hom(S, X) ≃ X(S).
In other words, any f ∈ X(S) may be seen as a morphism f : S → X. It provides
us with a map
f• : S ≃ S(•) → X(•),
Definition 4.2.3 If X is a condensed set, then underlying topological space is X(•)
endowed with the finest topology such that
• Given any (free) compact Hausdorff space S and any f ∈ X(S), the map f•
is continuous.
Exercise 4.5 Show that if X is a condensed set, then a subset Y of X(•) is
open (resp. closed) if and only if given any (free) compact Hausdorff S and any
f ∈ X(S), f•−1 (Y ) is open (resp. closed) in S.
Let us denote by kTop the full subcategory of compactly generated topological

spaces. We know that it is a coreflective subcategory of Top with coadjoint X 7→ kX.
Theorem 4.2.4 The functor
Top → Cond, (resp. kTop → Cond) X 7→ X
is faithful (resp. fully faithful) with adjoint X 7→ X(•).
Proof. Let us first show that, if X is a topological space, then X(•) = kX. First of
all, as sets, we have X(•) = C(•, X) ≃ X. Moreover, if S is a compact Hausdorff
space, then X(S) ≃ C(S, X). Under this bijection, an f ∈ X(S) corresponds to the
composite map
f•
S → X(•) ≃ X.
Our claim then follows from the very definitions of the topology of kX and X(•)
and we can move to the main statement.
Since the coreflection X 7→ kX is clearly faithful, it is sufficient to show that, if
we are given a condensed set X and topological space Y , then the natural map
Hom(X, Y ) → C(X(•), Y ), ϕ 7→ ϕ• (4.5)
is bijective. Assume given a continuous map φ : X(•) → Y . Then, if S is a compact

Hausdorff space and g ∈ X(S), the composite map
g• φ
ϕS (g) : S ≃ S(•) → X(•) → Y
is continuous. It provides a compatible family of maps ϕS : X(S) → C(S, Y ), or

equivalently a morphism ϕ : X → Y . One easily checks that this is an inverse to the
the map in (4.5). ■
It follows that the functor X 7→ X preserves all limits. Actually, we can identify
the category of compactly generated spaces with a reflective subcategory of the
category of all condensed sets (with all the pleasant consequences). Let us also
mention that there are factorizations
k
Top → kTop ,→ Cond, and Cond → kTop ,→ Top
and in particular, kX = X if X is any topological space.

Exercise 4.6 Show that the functor
Set → Cond, E 7→ E := E disc
is fully faithful and that
Hom(E, X) ≃ X(•)E .
Show that E is naturally isomorphic to the constant sheaf E

e associated to the
set E.
Solution. The first assertion is obtained by composition. More precisely, the functor
E 7→ E disc is fully faithful (any map between discrete topological spaces is continuous),
E disc is compactly generated and the functor X 7→ X is fully faithful on compactly
generated spaces. Now, we know that (still using the fact that E disc is compactly
generated)
Hom(E disc , X) ≃ C(E disc , X(•)) ≃ F(E, X(•)) ≃ X(•)E .
By left exactness, we know that • is a final object of Cond. Then, we know from
the previous result that there exists a natural isomorphism
e X) ≃ Hom(•, X)E ≃ X(•)E .
Hom(E,
The last assertion therefore follows from Yoneda’s lemma. ■
Exercise 4.7 Show that if X = i∈I Xi is an open covering, then (X i → X)i∈I is

S
a covering.
Solution. In the case X = S is compact Hausdorff, it follows from exercise 2.5 that
we can replace our open covering with a finite compact covering and we are done. In
general, any morphism S → X where S is a compact Hausdorff space has the form
f and we can apply the first case to the covering S = i∈I f −1 (Xi ). It means that
S
(S ×X X i → S)i∈I
is a covering and we can apply exercise 3.22. ■
Exercise 4.8 Show that, if X is a topological space and Y → X a morphism from

a condensed set, then the presheaf
U 7→ Y (U ) := Hom/X (U , Y )
(the Hom in Cond/X ) is a sheaf on X.
Exercise 4.9 Show that if X = i∈I Xi , then X = i∈I X i .

` `
Solution. Using exercise 3.22, we may assume that X is compact. ■

Proposition 4.2.5 Let π : X → Y be a continuous map. Assume Y is weak

Hausdorff and any compact Hausdorff subset of Y is contained is some π(S) with
S compact Hausdorff. Then, π : X → Y is an epimorphism.
Proof. It is sufficient to show that if F is a free compact Hausdorff space, then

the map X(F ) → Y (F ), or equivalently C(F, X) → C(F, Y ), is surjective. But if
g : F → Y is a continuous map, then g(F ) is compact Hausdorff and there exists
therefore a compact Hausdorff subset S ⊂ X such that g(F ) ⊂ π(S). Since F is a
projective object of the category of compact Hausdorff spaces, the map F → π(S)
induced by g lifts to some continuous map F → S and we can compose with the
inclusion map S ,→ X. ■
Exercise 4.10 Show that the hypothesis is satisfied when X locally compact
Hausdorff, Y is weak Hausdorff and π is open surjective.
Solution. Pick up for each x ∈ X a compact neighborhood Sx of x. Since π is open,

π(Sx ) is a neighborhood of π(x). Since π is surjective, Y = ∪x∈X π(Sx ). Any compact
subset of Y is contained in a finite union ∪ri=1 π(Sxi ) and we set S := ∪ri=1 Sxi . ■
4.2.2 Internal Hom

Since Cond is a topos, it is cartesian closed : there exists an internal Hom and a
natural isomorphism
∀X, Y, Z ∈ Cond, Hom(X × Y, Z) ≃ Hom(X, Hom(Y, Z)).
Recall from theorem 2.3.6 that the category of compactly generated spaces is also
cartesian closed. Actually, if X, Y, Z are three topological spaces, then there exists a
homeomorphism
kC(k(X × Y ), Z) ≃ kC(kX, C(kY, Z)).
The adjunction in theorem 4.2.4 can be enriched as follows:

Proposition 4.2.6 If X is a condensed set and Y a topological space, then there
exists a natural isomorphism of condensed sets
C(X(•), Y ) ≃ Hom(X, Y ).
Proof. We shall first do the case X = S with S compact Hausdorff so that X(•) = S.
If T is compact Hausdorff, then
Hom(S, Y )(T ) ≃ Hom(T , Hom(S, Y ))

≃ Hom(T × S, Y )
≃ Hom(T × S, Y )
≃ C(T × S, Y )
≃ C(T, C(S, Y ))
≃ C(S, Y )(T ).
In general, we obtain for S compact Hausdorff

Hom(X, Y )(S) ≃ Hom(S, Hom(X, Y ))
≃ Hom(X, Hom(S, Y ))
≃ Hom(X, C(S, Y ))
≃ C(X(•), C(S, Y ))
≃ C(S, C(X(•), Y ))
≃ C(X(•), Y )(S). ■
If X and Y are two condensed sets, then
Hom(X, Y ) = Hom(X, Y )(•)
inherits the structure of a compactly generated space. Then, if X is a condensed set
and Y is a topological space, there exists a homeomorphism
kC(X(•), Y ) ≃ Hom(X, Y ).
In particular, if X and Y are two topological spaces, then there exists a homeomor-
phism
kC(kX, Y ) ≃ Hom(X, Y ).
Exercise 4.11 Show that if E is any set and X a topological space, then
Hom(E, X) ≃ kX E .
Corollary 4.2.7 The functor
Cond → kTop, X 7→ X(•)
preserves finite products:
∀X, Y ∈ Cond, (X × Y )(•) = k(X(•) × Y (•)).
Proof. We use uniqueness of the adjoint. If X, Y are two condensed sets and Z is
any topological space, then
C(k(X(•) × Y (•)), Z) ≃ C(X(•), C(Y (•), Z))
≃ Hom(X, C(Y (•), Z))
≃ Hom(X, Hom(Y, Z))
≃ Hom(X × Y, Z)
= C((X × Y )(•), Z). ■
It is important to notice that the result does not hold anymore if we replace
kTop with Top. However, there exists a projection formula
S × X(•) ≃ (S × X)(•)
when X is any condensed set and S is a (locally) compact Hausdorff.
4.2.3 Quasi-compact/separated condensed set

Proposition 4.2.8 1. A condensed set X is quasi-compact if and only if there
exists an epimorphism S ↠ X with S (free) compact Hausdorff.
2. The functor S 7→ S induces an equivalence between compact Hausdorff spaces
and quasi-compact quasi-separated (quasi-compact quasi-separated for short)
condensed sets.
Proof. This was shown in proposition 3.3.10 and theorem 3.3.12 (use also proposition
2.2.2). ■
Lemma 4.2.9 If X is a weak Hausdorff topological space, then X is quasi-separated.
Proof. We may assume that X is compactly generated. For i = 1, 2, let Xi → X

be a morphism with Xi quasi-compact. There exists a compact Hausdorff space Si
and an epimorphism S i ↠ Xi . We may then consider the epimorphism S1 ×X S2 =
S 1 ×X S 2 ↠ X1 ×X X2 . Since X is weak Hausdorff, then S1 ×X S2 is compact
Hausdorff. ■
Lemma 4.2.10 A condensed set X is quasi-separated if and only if X = limi∈I Si

−→
as a filtered colimit of compact Hausdorff spaces under inclusion maps.
Proof. Thanks to exercise 3.42, only the direct implication needs a proof. There
exists an isomorphism X ≃ limi∈I T i with Ti compact Hausdorff. Thus, if we denote
−→
by Xi the image of T i in X, we have X = limi∈I Xi . As an image of T i , Xi is quasi-
−→
compact. But it is also quasi-separated because X is assumed to be quasi-separated.
It follows that Xi ≃ S i with Si compact Hausdorff. We may then replace the family
(Si ) with the family of SJ = ∪i∈J Si with J finite to get a directed set. ■
In other words, there exists an equivalence between the subcategory of Ind(CHaus)
of ind-objects “ lim Si ”with injective transition maps and the category of quasi-
−→
compact quasi-separated condensed sets.
Exercise 4.12 Show that if X = lim Sn is a countable filtered colimit of compact
−→
Hausdorff spaces under inclusion maps, then X is quasi-separated and X = lim S n .
−→
Solution. This follows from exercise 2.40. More precisely, if we are given a compact
Hausdorff space T , then an element of X(T ) is a continuous map T → X and it
comes from a morphism T → Sn which is an element of S n (T ). ■
Lemma 4.2.11 If X is a quasi-separated condensed set, then X(•) is weak Haus-

dorff.
Proof. We can write X = limi∈I Si as a filtered limit of inclusions of compact
−→
Hausdorff spaces and it follows that X(•) = limi∈I Si . Our assertion therefore follows
−→
from proposition 2.3.9. ■
Proposition 4.2.12 A compactly generated space X is weak Hausdorff if and only

if X is quasi-separated.
Proof. Follows from lemmas 4.2.9 and 4.2.11. ■
Exercise 4.13 Show that if X is a quasi-separated condensed set, then the canonical
map X → X(•) is a monomorphism.
5. Commutative algebra
5.1 Additive category

5.1.1 Pre-additive category
Definition 5.1.1 A pre-additive category (also called Ab-category)a is a category C
endowed with a factorization of the Hom functor:
Hom : C op × C / Set
O
(
Ab.
a
This is a particular instance of the notion of an enriched category.
In other words, we require that for all M, N ∈ C, Hom(M, N ) is endowed with

the structure of an abelian group and that for all M, N, P ∈ C, composition
Hom(M, N ) × Hom(N, P ) / Hom(M, P )
(f, g) / g◦f
is bilinear. In particular, End(M ) becomes a ring. We shall simply say that the
category is pre-additive (and not mention the factorization through the category of
abelian groups).
Examples 1. The categories Ab and A−Mod are pre-additive.

2. If C is any category with finite products, then Ab(C) is pre-additive. This
applies in particular to the category AbTop of topological abelian groups.
3. The category MatA is pre-additive.
92 Chapter 5. Commutative algebra
4. The category A associated to the multiplicative monoid of a ring A, is a

pre-additive category. Any pre-additive category with exactly one object has
this form.
5. The categories Set, Mon, Gr or Rng cannot be endowed with the structure of
a pre-additive category.
If C is a pre-additive category, we will actually consider the functor Hom as a

functor with values in Ab. Consequently, if M ∈ C, we will write
hM : C → Ab and hM : C op → Ab.
If C is a pre-additive category, then C op also. Same for C I if I is any small

category.
Definition 5.1.2 A functor F : C → D between two pre-additive categories is
additive if for any M, N ∈ C, the map
Hom(M, N ) → Hom(F (M ), F (N ))
is a group homomorphism.
The composite of two additive functors is additive. If F is an additive functor,
so is F op . And so is F I if I is a category.
Examples 1. If C is any pre-additive category, then the functors hM and hM are

additive.
2. If A is a ring, then the functors HomA and ⊗A are additive.
5.1.2 Additive category

Definition 5.1.3 Let C be a pre-additive category.
1. An M ∈ C is a zero object if End(M ) = 1 (meaning IdM = 0M ).
2. An M ∈ C is a direct sum of two objects M1 and M2 (in which case we write
M = M1 ⊕ M2 ) if there exists
pk : M → Mk , ik : Mk → M, k = 1, 2 such that
p1 ◦ i1 = IdM1 , p2 ◦ i2 = IdM2 and i1 ◦ p1 + i2 ◦ p2 = IdM .
Both notions are autodual in the sense that the property is satisfied in C if and
only if it is satisfied in C op .
Exercise 5.1 Let C be a pre-additive category.
1. Show that an M ∈ C is a zero object if and only if it is a final object (and
dual).
2. Show that an M ∈ C is a direct sum of M1 and M2 if and only if it is a
product of M1 and M2 with projections p1 and p2 (and dual).
More generally, if the coproduct of a family (Mi )i∈I exists in a pre-additive

category C, it is denoted by ⊕i∈I Mi and called a direct sum.
5.1 Additive category 93
Definition 5.1.4 An additive category is a pre-additive category with a zero object
and all direct sums.
Equivalently, it means that all finite products or all finite sums exist (and then
they both exist and are equal).
If C is an additive category then C op is also an additive category, and so is C I if I
is a small category.
Examples 1. The categories Ab, A−Mod and MatA are additive.

2. If C is any category with finite products, then Ab(C) is additive.
3. If A is a non-zero ring, then the category A is not an additive category.
Exercise 5.2 Show that if C is an additive category, then the factorization of Hom
through Ab is unique.
Exercise 5.3 Show that a functor between two additive categories is additive if
and only if it preserves all direct sums and the zero object - or equivalently all
finite products (and dual).
Exercise 5.4 Show that if a functor F between two additive categories is adjoint to
a functor G, then both functors are additive and there exists a natural isomorphism
of abelian groups
Hom(F M, N ) ≃ Hom(M, GN )).
5.1.3 Exact sequence

Let C be an additive category.
Definition 5.1.5 1. The kernel of a morphism f : M → N is ker f := ker(f, 0)
if it exists. The cokernel coker f of f is the kernel of f in C op (if it exists).
2. A sequence
f
0 → M ′ → M → M ′′
is said to be left exact if the sequence

f
/ /
M′ M / M ′′
0
is left exact (M ′ ≃ ker f ). A sequence M ′ → M → M ′′ → 0 is right exact if

it is left exact in C op (M ′′ ≃ cokerf ).
3. A short exact sequence is a sequence
ι π
0 → M ′ → M → M ′′ → 0
which is exact both on the left and on the right. We shall also say that M
is an extension of M ′′ by M ′ .
Exercise 5.5 In the the category AbHaus of Hausdorff topological abelian groups,
show that
1. if f : M → N is any continuous homomorphism, then ker f = f −1 (0) with
the induced topology and cokerf = N/f (M ) with the quotient topology.
2. a continuous map is a kernel (resp. cokernel) if and only if it is closed
injective (resp. open surjective).
ι π
3. a short sequence 0 → M ′ → M → M ′′ → 0 is exacta if and only if it is exact
as a sequence of abelian groups with ι closed and π open.
a
Sometimes called strict exact in order to insist on the fact that this is not just exact as a
sequence of abelian groups.
Exercise 5.6 Show that left (resp. right, resp. short) exact sequences in C form an
additive subcategory of C 3 .
Definition 5.1.6 A short exact sequence is said to split if it is isomorphic to
i1 p2
0 / M1 / M1 ⊕ M2 / M2 / 0.
An extension is said to be trivial if it splits.
Exercise 5.7 Show that a short exact sequence
p
0 / M′ i / M / M ′′ / 0
splits if and only if p has a section (and dual).
5.2 Abelian category

5.2.1 Definition
Definition 5.2.1 A pre-abelian category is an additive category C where any mor-
phism has both a kernel and a cokernel.
Equivalently, a pre-additive category C is pre-abelian if and only if all finite limits
and all finite colimits exist in C.
Examples 1. The categories Ab, A−Mod and MatA are pre-abelian.

2. The categories AbTop and AbHaus are pre-abelian.
If C is pre-abelian, so are C op and C I if I is a small category.

Definition 5.2.2 An abelian category is a pre-abelian category satisfying one of the
following equivalent properties:
1. Every monomorphism is regular and dual.
2. If i : N ,→ M is a monomorphism, then N ≃ ker(M ↠ coker(i)) and dual.
3. Any morphism f : M → N factors uniquely up to an isomorphism as an
epimorphism followed by a monomorphism.
5.2 Abelian category 95
4. If f : M → N is a morphism, then
ker(N → coker f ) ≃ coker(ker f → M ).

In an abelian category, it is common to say injection and surjection instead of
monomorphism and epimorphism.
Exercise 5.8 Show that all the above conditions are equivalent.
Exercise 5.9 Show that, in an abelian category C, we have for any morphism
f : M → N,
im f := ker(N → coker f ) (and dual).
Show that any morphism is strict.
Examples 1. The category A−Mod is abelian.

2. The category Op(k-Mod) of k-modules endowed with an operator is abelian.
3. If T is a topos, then Ab(T ) is abelian (see theorem 5.3.2 below).
4. The category AbCHaus is abelian.
5. The category AbTop is not abelian however because the identity Zdisc → Zcoarse
is not strict.
6. The category AbHaus is not abelian either because the map ℓ1 (R) → ℓ2 (R) is
not strict.
7. The category MatZ is not abelian because 2 is a monomorphism which is not
regular.
f g
Definition 5.2.3 A sequence M ′ → M → M ′′ is said to be exact (in M ) if
Im(f ) = ker(g).
Exercise 5.10 Show that a sequence
p
0 / M′ i / M / M ′′ / 0
is a short exact sequence (resp. left exact, resp. right exact) if and only if it is
exact in M ′ , M, M ′′ (resp. M ′ , M , resp. M, M ′′ ).
If C is an abelian category, then C op is also an abelian category, as well as C I if I

is any small category.
Exercise 5.11 Show that a functor F : C → D between two abelian categories is
left exact if and only if it is additive and preserves left exact sequences (and dual).
Exercise 5.12 Let D be an additive (resp. abelian) category. Show that, if a fully
faithful (resp. and exact) functor C ,→ D has an adjoint or a coadjoint, then C
also is additive (resp. abelian).
We shall not prove the next result which is useful to reduce many general
statements on abelian categories to the case of a category of A-modules:
Theorem 5.2.4 — Freyd-Mitchell. If C is a small abelian category, then there exists

a ring A and a fully faithful exact functor C ,→ A−Mod. ■
5.2.2 Grothendieck category

We now introduce Grothendieck axioms:
Definition 5.2.5 A category C is:
1. AB1 : pre-abelian.
2. AB2 : abelian.
3. AB3 : AB2 with all colimits (AB3* : dual).
4. AB4 : AB3 and direct sums are exact (AB4* : dual).
5. AB5 : AB4 and filtered colimits are exact (AB5* : dual).
6. AB6 : AB5 and filtered colimits commute with products (AB5* : dual).
Examples 1. The category Ab satisfies AB6 and AB4*.

2. The category AbTop satisfies AB1.
3. The category AbCHaus ≃ Abop satisfies AB4 and AB6* (Pontryagin duality).
4. If C is any category, then AbCb satisfies AB6 and AB4*.
5. If T is a topos, then AbT satisfies AB5 and AB3* (theorem 5.3.2).
6. It is not true however that AbT satisfies AB6 or AB4* in general.
7. There is no category satisfying AB5 and AB5* besides {0}.
Definition 5.2.6 A Grothendieck category is an AB5 category that has a generator.
Examples 1. A-Mod is a Grothendieck category.

2. One can show that if C is an abelian category, then Ind(C) is a Grothendieck
category.
3. If T is a topos, then Ab(T ) is a Grothendieck category (theorem 5.3.2).
Exercise 5.13 Show that if C is an AB5 category, then G is a generator if and

only if, for all M ∈ C, there exists an epimorphism G(I) ↠ M .
Proposition 5.2.7 A Grothendieck category is automatically AB3*.
Proof. To do. ■
Proposition 5.2.8 A Grothendieck category has enough injectives.
Proof. To do. ■
5.3 Abelian sheaf

5.3.1 Defintiion/Properties
Definition 5.3.1 1. A presheaf on a category C with values in a category D is a
(contravariant) functor T : C op → D. A morphism of presheaves is a natural
transformation.
2. If C is a site, then a sheaf F : C op → D is a presheaf such that, for all
Y ∈ D, the presheaf of sets
X 7→ Hom(Y, F(X))
is a sheaf.
One can extend many former results from sheaf theory to this situation but
we shall concentrate on the case of sheaves with values in Ab and say sheaf of
abelian groups or abelian sheaf or even abelian group on C. We shall denote by
C(Ab)
b := Hom(C op , (Ab) (resp. C(Ab)
e ⊂ C(Ab))
b the category of presheaves (resp.
the full subcategory of sheaves) of abelian groups on C. Note that a presheaf is a
particular case of a sheaf where we endow C with the coarse topology. When there is
no risk of ambiguity, we shall write
HomZ (M, N ) := HomC(Ab)

e (M, N ) = HomC(Ab)
b (M, N ).
By composition, the forgetful functor Ab → Set provides a forgetful functor C(Ab)

b →
C that sends a presheaf of abelian groups to the underlying presheaf of sets.
b
Exercise 5.14 Show that a presheaf of abelian groups M on a site C is a sheaf if
and only if the underlying presheaf of sets is a sheaf.
Solution. By definition, the presheaf M is a sheaf if and only if, for all covering
sieves R of X ∈ C and for all N ∈ A, we have
!
HomZ (N, M(X)) ≃ lim HomZ (N, M(Y )) ≃ HomZ N, lim M(Y )
←− ←−
Y ∈C/R Y ∈C/R
By Yoneda lemma, this is equivalent to require that
M(X) ≃ lim M(Y )

←−
Y ∈C/R
(in Ab) and we know that the forgetful functor preserves all limits. ■
As a consequence, there also exists a forgetful functor C(Ab)

e → Ce that sends a
sheaf of abelian groups to the underlying sheaf of sets.
Exercise 5.15 Show that if C is a site, then there exists an equivalence of categories
C(Ab)
e ≃ Ab(C).
e
Solution. It follows from exercise 5.17 that the global section functor preserves
products. Therefore, if M ∈ Ab(C)e and X ∈ C, then M(X) is a usual abelian group
and this is clearly functorial. Conversely, if M ∈ C(Ab),

e then the obvious family of
maps
µX : M(X)×M(X) → M(X), ϵX : {0} → M(X) and ιX : M(X) → M(X)
define a structure of abelian group on the underlying sheaf of sets of M. ■

We shall identify these categories: a sheaf of abelian groups on C is the same
thing as an abelian group in the topos C.
e In particular, it only depends on the topos
and not on the site itself.
Exercise 5.16 Show that if C is a site, then C(Ab)
e is a reflective subcategory
of C(Ab) with exact (additive) reflection M 7→ M which is compatible with
b f
underlying sheaves and presheaves of sets.
Solution. Since sheafification is (left) exact, this easily follows from theorem 3.2.8. ■
Exercise 5.17 Show that if C is a site and X ∈ C then the functor
C(Ab)
e → Ab, (resp. C(Ab)
b → Ab) M 7→ Γ(X, M) := M(X).
is additive and preserves all limits (resp. limits and colimits).
Solution. In the presheaf case, this is shown as in theorem 3.1.2 and we may then
use exercise 5.16. ■
5.3.2 Abelian group

Exercise 5.18 Show that if T is a topos, then the forgetful functor T (Ab) → T
has an adjointa X 7→ Z · X.
a
One also sometimes write Z[X].
Solution. We can write T = C. e We know that the forgetful functor Ab → Set as an

adjoint E 7→ Z · E := Z . It follows from exercise 1.62 that the forgetful functor
(E)
p
C(Ab)
b → Cb as an adjoint T 7→ Z · T (p for presheaf). If F is a sheaf, we may then
p
define Z · F as the sheafification of Z · F. ■
It means that there exists a natural isomorphism
HomZ (Z · X, M ) ≃ Hom(X, M ) = M (X)
for X ∈ T and M ∈ T (Ab). In particular, if C is a site, X ∈ C and M ∈ C(Ab),

e
then
HomZ (Z · X, M) ≃ M(X).
Example If E is a set, then Z · E (E) . In particular, Z · e

e ≃ Zg 1 ≃ Z.
e
Exercise 5.19 Show that, if C is any category, then Z · hX is projective in C.

b
Solution. The functor M 7→ HomZ (Z · hX , M) ≃ M(X) is exact on presheaves. ■
Exercise 5.20 Show that is S is a set of generators of a topos T , then {Z·X, X ∈ S}

is a set of generators for T (Ab).
Theorem 5.3.2 If T is a topos, then T (Ab) is a Grothendieck category.
Proof. In the case T = C, b this is may be checked component by component (details

left to the reader). In general, we already know that T (Ab) ≃ Ab(T ) is pre-abelian
and we shall now write T = C. e In order to see that T (Ab) is abelian, it remains
to check that if u : M → N is any morphism, then coker(ker(u)) ≃ ker(coker(u)).
Denote by ι : C(Ab)
e ,→ C(Ab)
b the inclusion functor and π : C(Ab)
b ↠ C(Ab)
e the
exact reflection. We have coker(ker(ι(u))) ≃ ker(coker(ι(u))) in C(Ab) and we apply
b
π which is exact and satisfies π ◦ i = Id. The same kind of argument shows that
filtered direct limits are exact. The fact that all limits and colimits exist follows
directly from the fact that C(Ab)
e is a reflexive subcategory of C(Ab).
b Finally, since
T has a small set of generators, it follows from exercise 5.20 that T (Ab) has a small
set of generators S. Then ⊕M ∈S M is a generator. ■
5.3.3 Internal Hom and tensor product

Let us first remark that, if T is a topos, X ∈ T and M ∈ T (Ab), then the sheaf
Y 7→ Hom(X, M )(Y ) ≃ Hom(X × Y, M ) ≃ HomZ (Z · (X × Y ), M )
is a sheaf of abelian groups.
Exercise 5.21 Show that if T is a topos and M, N ∈ T (Ab), then the presheaf
X 7→ HomZ (M, Hom(X, N ))
on T is representable by an HomZ (M, N ) ∈ T (Ab).
Solution. Thanks to exercise 3.36, it is sufficient to notice that the presheaf commutes
will all limits. ■
As a consequence, there exists a natural isomorphism
HomZ (M, Hom(X, N )) ≃ Hom(X, HomZ (M, N )) ≃ HomZ (M, N )(X).
Exercise 5.22 Show that, if C is a site and M, N ∈ C(Ab),

e then HomC(Ab)
e (M, N ) =
HomC(Ab)
b (M, N ).
Solution. For X ∈ C, we have

HomC(Ab)
e (M, N )(X) = HomZ (M, HomCe(X, N ))
= HomZ (M, HomCb(hX , N ))
= HomC(Ab)
b (M, N )(X). ■
Exercise 5.23 Show that, if T is a topos and M, N ∈ T (Ab), then HomZ (M, N ) =
HomZ (M, N )(1).
Actually, T (Ab) is enriched over itself.

Exercise 5.24 Show that if T is a topos, X ∈ T and M ∈ T (Ab), then
HomZ (Z · X, M ) ≃ Hom(X, M ).
Exercise 5.25 Show that if T is a topos and M, N ∈ T (Ab), then the functor
P 7→ HomZ (M, HomZ (N, P ))
is representable by an M ⊗Z N ∈ T (Ab).
Solution. It is sufficient to consider the case T = Cb and then apply sheafification.

This in turn blows down to the analog assertion in the category of usual abelian
groups. ■
As a consequence, T (Ab) is a closed symmetric monoidal category: there exists
a natural isomorphism
HomZ (M ⊗Z N, P ) ≃ HomZ (M, HomZ (N, P )).
In particular, for fixed N , the functor M 7→ M ⊗Z N is adjoint to the functor

P 7→ HomZ (N, P ).
Exercise 5.26 Show that, if M, N are two abelian sheaves on a site C, then
M ⊗Z N is the sheafification of the presheaf
p
M ⊗Z N : X 7→ M(X) ⊗Z N (X).
Exercise 5.27 Show that

1. ∀M, N ∈ T (Ab), M ⊗Z N ≃ N ⊗Z M ,
2. ∀M, N, P ∈ T (Ab), (M ⊗Z N ) ⊗Z P ≃ M ⊗Z (N ⊗Z P ),
3. ∀M ∈ T (Ab), M ⊗Z Z · 1 ≃ M .
Exercise 5.28 Show that there exists a natural isomorphism
Exercise 5.29 Show that Z · X ⊗Z Z · Y ≃ Z · (X × Y ).
Solution. Follows from Yoneda’s Lemma since
HomZ (Z · X ⊗Z Z · Y, N ) ≃ HomZ (Z · X, HomZ (Z · Y, N )

≃ Hom(X, Hom(Y, N )
≃ Hom(X × Y, N )
≃ Hom(Z · (X × Y ), N ). ■
Exercise 5.30 Show that, if we set M · X := M ⊗Z Z · X, then
HomZ (M, N )(X) ≃ HomZ (M · X, N )
and
HomZ (M · X, N ) ≃ HomZ (M, Hom(X, N )) = Hom(X, HomZ (M, N )).
Definition 5.3.3 An abelian group P of a topos T is said to be flat if the functor

M 7→ P ⊗Z M is exact.
Example A usual abelian group is flat if and only if it is torsion free.
Exercise 5.31 Show that if T is a topos and X ∈ T , then Z · X is flat.
Solution. It is sufficient to consider the case T = Cb and then apply sheafification.

This reduces to the case of ordinary abelian groups and a free abelian group is
torsion-free. ■
6. Condensed abelian groups
6.1 Condensed abelian group

6.1.1 Definition
Definition 6.1.1 A condensed abelian group is an abelian group in the category of
condensed sets.
They form a category AbCond which maybe also denoted CondAb thanks to:
Proposition 6.1.2 The following are equivalent:
1. The category of condensed abelian groups.
2. The category of sheaves of abelian groups on the site Cond of condensed
sets.
3. The category of sheaves of abelian groups on the site CHaus of compact
Hausdorff spaces.
4. The category of sheaves of abelian groups on the site FCHaus of free compact
Hausdorff spaces.
Proof. This follows from the definition and proposition 4.1.6 thanks to exercise
5.15. ■
The category CondAb is also equivalent to the category of sheaves of abelian
groups on the site of Stone or Stonean spaces. We shall identify all these categories.
The category CondAb is automatically a Grothendiek category but we shall do a lot
better (theorem 6.1.4 below).
Exercise 6.1 Show that a presheaf M of abelian groups on FCHaus is a condensed
abelian group if and only if it preserves finite products: for all free compact
104 Chapter 6. Condensed abelian groups
Hausdorff spaces F1 , . . . , Fr , we have

r
! r
a M
M Fj ≃ M (Fj ) .
j=1 j=1

Equivalently: M (∅) = 0 and M (F ⊔ F ′ ) ≃ M (F ) ⊕ M (F ′ ) for F, F ′ ∈ FCHaus.
Lemma 6.1.3 If F is a free compact Hausdorff space (more generally a Stonean

space), then the functor
AbCond → Ab, M 7→ Γ(F, M )
preserves all limits and colimits.
Proof. Thanks to exercise 5.17, it suffices to show that if (Mi )i∈I is diagram of
sheaves of abelian groups on FCHaus, then its colimit limpi∈I Mi in the category of
−→
presheaves of abelian groups is automatically a sheaf. We give ourselves free compact
Hausdorff spaces F1 , . . . , Fr and we compute
! r ! r
!
p a a
lim Mi Fj = lim Mi Fj
−→ −→
i∈I j=1 i∈I j=1
r
M
= lim Mi (Fj )
−→
i∈I j=1
Mr
= lim Mi (Fj )
−→
j=1 i∈I
r
!
M p
= lim Mi (Fj ) . ■
−→
j=1 i∈I
6.1.2 Grothendieck category

Theorem 6.1.4 The category CondAb is a Grothendieck category satisfying axioms
AB6 and AB4*.
Proof. Thanks to lemma 6.1.3, this follows from the analog assertion in Ab. ■
Corollary 6.1.5 The category CondAb is an abelian category and

1. there exists a a generator,
2. all limits and colimits exist,
3. all direct sums and products are exact,
4. filtered colimits are exact and commute with products. ■
Lemma 6.1.6 If F is a free compact Hausdorff space (more generally Stonean),

then Z · F is a finitely presenteda projective condensed abelian group.

a
Also called compact.
Proof. It is means that the functor1
CondAb → Ab, M 7→ HomAb (Z · F , M )
preserves epimorphisms and filtered colimits. This follows from proosition 6.1.3 since
HomAb (Z · F , M ) ≃ Hom(F , M ) ≃ M (F ) = Γ(F, M )
and the functor actually commutes with all colimits. ■
Proposition 6.1.7 The category CondAb is generated by finitely presented projec-

tive condensed abelian groups. In particular, it has enough projectives.
Proof. Since Cond is generated the family of F where F is a free compact Hausdorff
space, we know from exercise 5.20 that CondAb is generated by all Z · F . ■
The category CondAb is a closed symmetric monoidal category (in particular, it
is enriched over itself):
Proposition 6.1.8 There exists two bifunctors HomZ and ⊗Z on CondAb with
natural isomorphisms of abelian groups
HomZ (M, N )(•) ≃ HomZ (M, N )
and
Proof. This is exercises 5.23 and 5.25. ■

Note that HomZ (M, N ) = HomZ (M, N )(•) is naturally a compactly generated
topological space. Be careful however that this is not a topological abelian group
because compactly generated spaces are not closed under products of topological
spaces.
6.2 Topological abelian groups

6.2.1 Associated condensed group
Proposition 6.2.1 There exists a faithful functor that preserves all limits:
AbTop → AbCond, M 7→ M
It is even fully faithful on compactly generated abelian groups.
Proof. Formally follows from theorem 4.2.4. ■

1
We may use Ab or Ens indifferently.
Be careful however that there is no obvious adjoint because the functor X 7→ X(•)
does not preserve products. In the same way, if M is a topological abelian group,
then kM does not get the structure of a topological abelian group in general.
If M, N are two topological abelian groups, then we denote by CZ (M, N ) ⊂
C(M, N ) the subspace of continuous homomorphisms. Note that CZ (M , N ) =
HomZ (M, N ) when M is compactly generated.
Proposition 6.2.2 If M, N are two topological abelian groups with M compactly
generated, then there exists a natural isomorphism of condensed abelian groups
CZ (M, N ) ≃ HomZ (M , N ).
Proof. Let S be a compact Hausdorff space. It follows from proposition 6.2.1 that
HomZ (M , N )(S) ≃ Hom(S, HomZ (M , N ))

≃ HomZ (M , Hom(S, N ))
≃ HomZ (M , C(S, N ))
≃ CZ (M, C(S, N )).
On the other hand, we have
CZ (M, N )(S) ≃ Hom(S, CZ (M, N )) ≃ C(S, CZ (M, N )).
Our isomorphism is then induced by the natural bijection
C(M, C(S, N )) ≃ C(M × S, N ) ≃ C(S, C(M, N ))
coming from theorem 2.3.6. ■
ι π
Proposition 6.2.3 Let 0 → M ′ → M → M ′′ → 0 be an exact sequence of topological
abelian groups with M ′′ weak Hausdorff. Assume that for all compact Hausdorff
K ′′ ⊂ M ′′ , there exists a compact Hausdorff K ⊂ M such that K ′′ ⊂ π(K). Then,
the sequence of condensed abelian groups 0 → M ′ → M → M ′′ → 0 is also exact.
Proof. Since left exactness is automatic, this follows from proposition 4.2.5. ■
Exercise 6.2 Show that the hypothesis is satisfied if M is locally compact.
Exercise 6.3 Show that if (Mi )i∈I is a family of abelian groups, then
!disc
M M
Mi ≃ Midisc .
i∈I i∈I
6.2.2 Locally compact abelian groups

The standard reference is Morris 79.
Exercise 6.4 Show that the category of locally compact Hausdorff abelian groups
is pre-abelian but not abelian.
Solution. The kernel of f : M → N is the usual kernel with the induced topology.
The cokernel is the quotient N/f (M ) with the quotient topology. The category is
not abelian because the identity Rdisc → R is not strict. ■
Exercise 6.5 Show that if N is a closed subgroup of a locally compact Hausdorff

abelian group M , then M/N ≃ M /N .
We shall denote by T := {z ∈ C× , ∥z∥ = 1} and
M ∗ := CZ (M, T)
the Pontryagin dual of a topological abelian group M .

Exercise 6.6 Show that Fourier transform
R × R → T, (x, y) 7→ e2iπxy
induces an isomorphism R ≃ R∗ . Show that Z ≃ T∗ and T ≃ Z∗ .
Solution. If f ∈ R∗ , then
R ϵ f is continuous and f (0) = 1. It follows that there exists
ϵ > 0 such that δ := 0 f (t)dt ̸= 0. Since f is a homomorphism, we have
Z y+ϵ Z ϵ Z ϵ
f (t)dt = f (y + t)dt = f (y)f (t)dt = δf (y).
y 0 0
R y+ϵ
It follows that f (y) = δ −1 y f (t)dt, and in particular, f is differentiable. Since
f is a homomorphism, we have f (y + h) − f (y) = f (y)(f (h) − f (0)). Dividing by
′
h and taking limit provides f ′ (y) = f ′ (0)f (y) and therefore f (y) = Cef (0)y . Since
f (0) = 1, we have C = 1 and since |f (1)| = 1, we have f ′ (0) ∈ Ri. We can therefore
write f (y) = e2iπxy for a unique x ∈ R. This implies that R ≃ R∗ . The isomorphism
Z ≃ T∗ follows immediately (by composition with the covering R ↠ T) and T ≃ Z∗
is trivial. ■
Theorem 6.2.4 — Pontryagin-van Kampen. The functor M 7→ M ∗ := CZ (M, T) is

a self-equivalence (an equivalence which is self-adjoint) on the category of locally
compact Hausdorff abelian groups.
Proof. To do. ■
It means that, if M is a locally compact Hausdorff abelian group, then M ∗ also
and (M ∗ )∗ = M .
Exercise 6.7 Show that if 0 → M ′ → M → M ′′ → 0 is a (strict) exact sequence
of locally compact abelian groups, then the sequence 0 → M ′′∗ → M ∗ → M ′∗ → 0
is also exact.
Solution. There exists an exact sequence of topological abelian groups 0 → N →
M → M/N → 0. Since N is closed, M/N is Hausdorff. Moreover, M is locally
compact. Then, we know that the sequence 0 → N → M → M/N → 0 is exact. It
means that M/N ≃ M /N . ■
Exercise 6.8 Show that Pontryagin duality reduces to usual duality on finite
dimensional real Banach spacesa .
a
A real Banach space is locally compact if and only if it is finite dimensinal.
Exercise 6.9 Show that Pontryagin duality induces an equivalence between dis-
crete abelian groups and compact Hausdorff abelian groups. Show that torsion
corresponds to Stonea and torsion free corresponds to connected.
a
Profinite.
Proof. If M is a discrete abelian group, then there exists a surjective homomorphism

Z(I) ↠ M and, by duality, a closed embedding M ∗ ,→ TI which shows that M ∗ is
compact. Conversely, if U ⊂ T is defined by Im(z) > 0, then CZ (M, T) ∩ C(M, U ) =
{0} and it easily follows that M ∗ is discrete. As a consequence, we see that the dual
of a finite abelian group is also a finite abelian group. Now, M is torsion if and only
if M = lim Mi with Mi finite if and only if M ∗ = lim Ni with Ni finite if and only if
−→ ←−
M ∗ is Stone. Also, M is not torsion free if and only if there an injection M ′ ,→ M
with M ′ finite not trivial if and only if there exists a surjection M ∗ ↠ N ′ with N ′
finite not trivial if and only if M ∗ is not connected. ■
Exercise 6.10 Show that any compact Hausdorff abelian group M has a two terms
resolution (a short exact sequence)
0 → M → M0 → M1 → 0
with M0 , M1 connected.
Exercise 6.11 Show that if M is a locally compact Hausdorff abelian group, then
M ∗ ≃ HomZ (M , T) and M ∗ ≃ HomZ (M , T).
Solution. Since M is locally compact, it is compactly generated and therefore
M ∗ = CZ (M, T) ≃ HomZ (M , T).
It follows that
M ∗ ≃ M ∗ (•) ≃ HomZ (M , T)(•) ≃ HomZ (M , T). ■
Theorem 6.2.5 Any locally compact Hausdorff abelian group M has an open
subgroup of the form V × K where V is a finite dimensional real Banach space
and K is a compact Hausdorff abelian group.
Proof. To do. ■
In other words, there exists an exact sequence
0→V ×K →M →D →0
with D discrete.
7. Cohomology (optional)
7.1 Complex
7.1.1 Definition
Let C be an additive category.

Definition 7.1.1 1. A (long) sequence in C is a commutative diagram on the
ordered set (Z, ≤):
/ dn−1 / dn / /
··· K n−1 Kn K n+1 ···
2. If dn ◦ dn−1 = 0 for each n ∈ Z, we call K • a (cochain) complex. The dual

notion is that of chain complex. Then, one usually writes Kn := K −n and
dn := d1−n .
3. A complex K • is said to be bounded below (resp. bounded above, resp.
bounded ) if K n = 0 for n << 0 (resp. it is bounded below in C op , resp. if it
is bounded both above and below).
Exercise 7.1 Show that the following hold:

1. The complexes of C form an additive subcategory C(C) of C (Z,≤) .
2. Limits and colimits in C(C) are computed argument by argument.
3. The equivalence (Z, ≤) ≃ (Z, ≥) induces an equivalence C(C op ) ≃ C(C)
between chain complexes and cochain complexes.
We shall denote by C+ (C), C− (C) and Cb (C) the categories of complexes that
are respectively bounded below, bounded above and bounded. All the coming
development has an equivalent with +, − and b.
112 Chapter 7. Cohomology (optional)
Exercise 7.2 Show that if K• is a (semi-) simplicial object and we set dn :=
(−1)i dni , then K• becomes a chain complex.
Pn−1
i=0
Example If X is any topological space, we can consider the simplicial set S• (X) =
hX ◦ ∆• so that Sn (X) = Homcont (∆n , X). We may then consider the simplicial
(resp. cosimplicial) group
C• (X) := Z · S• (X) (resp. C • (X) := ZS• (X) )
and see it as a chain (resp. cochain) complex.
Exercise 7.3 1. Show that the inclusion 1 = {0} ,→ (Z, ≤) induces a functor
C ,→ C(C), M 7→ [M ]
(so that [M ]0 = M and [M ]n = 0 otherwise) which is fully faithful and

commutes with all limits and colimits.
2. Show that 2 ,→ (Z, ≤) induces a functor
f f
Mor(C) ,→ C(C), (M → N ) 7→ [M → N ].
3. Same thing with left, right or short exact sequences.
7.1.2 Homotopy
Definition 7.1.2 Two morphisms f, g : K • → L• are said to be homotopic if there
exists a family of sn : K n → Ln−1 :
/ dn−1 / dn / / ···
··· K n−1 Kn K n+1
sn−1 sn sn+1
{ { { dn
/ dn−1 / Ln / Ln+1 / ···
··· Ln−1
such that for all n, we have
f n − g n = sn+1 ◦ dn + dn−1 ◦ sn .
We shall then write s : f ∼ g.
Exercise 7.4 Show that morphisms that are homotopic to 0 form a subgroup
of HomC(C) (K • , L• ). Show that composition on both sides with a morphism
homotopic to 0 always gives a morphism homotopic to 0.
Exercise 7.5 Show that homotopy is an equivalence relation compatible with

composition on both sides.
Definition 7.1.3 The category K(C) of complexes up to homotopy is the category

that has the same objects as C(C) and
HomK(C) (K • , L• ) = HomC(C) (K • , L• )/{homotopic to 0}

7.1 Complex 113
with the induced composition. A morphism of complexes that becomes an isomor-

phism in K(C) is called a homotopy equivalence. A complex that becomes 0 in
K(C) is said to be homotopically trivial.
In other words, a diagram
K•
h / M •
=
f g
!
L•
in C(C) is commutative in K(C) if it commutes up to homotopy: there exists a

homotopy s : h ∼ g ◦ f :
Exercise 7.6 1. Show that a morphism of complexes f : K • → L• is a homotopy
equivalence if and only if there exists a morphism g : L• → K • such that
g ◦ f ∼ Id and f ◦ g ∼ Id.
2. Show that a complex K • is a homotopy trivial if and only if IdK• ∼ 0K • .
Exercise 7.7 1. Show that K(C) is an additive category and that the obvious
functor C(C) → K(C) is additive.
2. Show that any additive functor F : C → C ′ induces a functor
F : K(C) → K(C ′ ).
On defines K+ (C), K− (C) and Kb (C) in the same way (may also be seen as full
subcategories).
7.1.3 Mapping cone

Definition 7.1.4 The mapping cone of a morphism of complexes f : K • → L• is
the complex M (f )• with

n n+1 n n −dn+1 0
M (f ) := K ⊕L and d := .
fn dn
The kth shift K • [k] of a complex K • is defined by K • [k]n := K n+k endowed with
(−1)k dn+k .
Exercise 7.8 Show that there exists a short exact sequence of complexes
g h
0 → L• → M (f )• → K • [1] → 0

Id
with g := and h = 0 Id .

0
Definition 7.1.5 A triangle in K(C) is a diagram
+
K • → L• → M • → K • [1] (or K • → L• → M • → for short).
A morphism of triangles is a commutative diagram (in K(C))
K• / L• / M• / K • [1]
u v w u[1]

K ′• / L′• / M ′• / K ′• [1].
The triangle is said to be distinguished if it is isomorphic (in K(C)) to

f g h
K • → L• → M (f )• → K • [1].
Exercise 7.9 Show that if F is an additive functor and K • → L• → M • → K • [1]

is distinguished, then F K • → F L• → F M • → F K • [1] is also distinguished.
Proposition 7.1.6 A triangle
f g h
K • → L• → M • → K • [1]
is distinguished if and only if the triangle

g h −f [1]
L• → M • → K • [1] → L• [1]
is distinguished.
Proof. In order to prove the direct implication, it is sufficient to show that there
exists a commutative diagram in K(C):
g −f [1]
L• / M (f )• h / K • [1] / L• [1]
≃
g
L• / M (f )• / M (g)• / L• [1].
We can then simply set

   
−f n 0 0 Id
ϕn :=  Id  , ψ n := and sn :=  0 0 0  .

0 Id 0
0 0 0 0
The diagram is clearly commutative, ψ ◦φ = Id and s : φ◦ψ ∼ Id since Id−ϕn ◦ψ n =

sn+1 ◦ dn + dn−1 ◦ sn . For the converse implication, it is equivalent to show that
K • [1] → L• [1] → M • [1] → K • [2] is distinguished and we may apply twice the
previous result. ■
Exercise 7.10 Show that K • = K • → [0] → K • [1] is distinguished.
Proposition 7.1.7 1. Any morphism f : K • → L• in K(C) can be extended to a

(not unique) distinguished triangle.
7.1 Complex 115
2. Any commutative diagram

f
K• / L•
u v
f′

K ′• / L′•
in K(C) can be extended to a (not unique) morphism of distinguished

triangles.
Proof. The first assertion is clear. For the second one, we can write v n ◦f n −f ′n ◦un =
sn+1 ◦ dn + dn−1 ◦ sn and set
n+1
u 0
= M (f )n → M (f ′ )n . ■
sn+1 v n
In other words, the mapping cone is “almost” functorial in K(C).

Lemma 7.1.8 If we are given a morphism of distinguished triangles
f g
K• / L• / M• h / K • [1]
0 0 w 0
f g
K• / L• / M• h / K • [1]
in K(C), then w2 = 0 (in K(C)).
Proof. We may assume that M • = M (f )• and write

w11 w12
w= .
w21 w22

w11
Since ∼ 0 and ∼ 0, we have

w21 w22
w21

′ 0 w12 ′′ w11 w12
w ∼ w := and w ∼ w := .
0 w22 0 0
It follows that w2 ∼ w′′ ◦ w′ = 0. ■
Proposition 7.1.9 In a morphism of distinguished triangles (in K(C))
f
K• / L• / M• / K • [1]
u v w u[1]
f′

K ′• / L′• / M ′• / K ′• [1],
if u and v are homotopy equivalences, so is w.

Proof. We may assume that M • = M (f )• and that M ′• = M (f ′ )• . After composing

both u and v with an inverse in K(C) and extending to a morphism of distinguished
triangles, we are reduced to the case u = Id and v = Id. Then, we have a morphism
of distinguished triangles
f
K• / L• / M• / K • [1]
0 0 w−Id 0
f
K• / L• / M• / K • [1].
If follows from lemma 7.1.8 that (w − Id)2 = 0 in K(C). Thus, 2Id − w is an inverse
for w. ■
As a consequence, the mapping cone is “almost” unique (up to a homotopy
equivalence) in K(C).
Corollary 7.1.10 In a distinguished triangle
f +
K • → L• → M • →,
f is a homotopy equivalence if and only if M • is homotopically trivial.
Proof. It is sufficient to contemplate the following diagram:

f
K• / L• / M• / K • [1] ■
f f [1]

L• L• / [0] / L• [1].
7.1.4 Cohomology
Let A be an abelian category.
Exercise 7.11 Show that C(A) also is abelian.
Definition 7.1.11 The n-th cohomology of a cochain complex K • of A is
Hn (K • ) := Zn (K • )/Bn (K • )
with Zn (K • ) := ker dn and Bn (K • ) := im dn−1 . The n-th homology Hn (K• ) of a

chain complex K• is its (1 − n)-th cohomology in Aop .
Examples 1. We have Hn (K • [k]) = Hn+k (K • ).

2. If M ∈ A, then H0 ([M ]) = M and Hn ([M ]) = 0 otherwise.
f f
3. We have H0 ([M → N ]) = ker f and H1 ([M → N ]) = coker f (and 0 otherwise).
4. A short exact sequence 0 → M ′ → M → M ′′ → 0 has 0 cohomology every-
where.
5. If X is a topological space, then
•
Hsing n n
n (X) := Hn (C• (X)) and Hsing (X) := H (C (X)).
7.1 Complex 117
Exercise 7.12 Show that Hn is functorial in K • .
Exercise 7.13 Show that if K • is a complex, then the sequence
0 → Hk → cokerdk−1 → Zk+1 → Hk+1 → 0
est exact (everywhere).
Exercise 7.14 Show that the functor
Hn : C(A) → A
is additive but not exact in general.
Definition 7.1.12 If C is an additive category, then an additive functor F : K(C) →

A is said to be cohomological if, whenever K • → L• → M • → K • [1] is distin-
guished, then F K • → F L• → F M • is exact in the middle.
Exercise 7.15 Show that, if F : K(C) → A is a cohomological functor and

K • → L• → M • → K • [1] a distinguished triangle, then there exists a long exact
sequence
· · · → F K • [n] → F L• [n] → F M • [n] → F K • [n + 1] → · · ·
Proposition 7.1.13 For fixed K • , the functor L• 7→ HomK(C) (K • , L• ) is cohomolog-

ical (and dual).
f g
Proof. Let L′• → L• → L′′• → L′• [1] be a distinguished triangle. If K • → L′• is any
morphism, then there exists a morphism of triangles
K• K• / 0 / K • [1]

L′• / L• / L′′• / L′• [1].
This shows that K • → L′′• is homotopic to zero. Conversely, if we are given a

morphism K • → L• such that the composite K • → L• → L′′• is homotopic to zero,
then there exists such a morphism of triangles which provides a suitable morphism
K • → L′• . ■
Proposition 7.1.14 — Snake lemma. If

p
0 / M′ i /M / M ′′ /0
f′ f f ′′
j q

0 / N′ / N / N ′′ /0
is a commutative diagram with exact lines, then there exists a natural (long) exact
sequence
0 → ker f ′ → ker f → ker f ′′ → coker f ′ → coker f → coker f ′′ → 0.
Proof. This is a tedious diagram chasing1 . ■
Theorem 7.1.15 Any short exact sequence 0 → K • → L• → M • → 0 of cochain

complexes gives rise to a natural long exact sequence
· · · → Hn (K • ) → Hn (L• ) → Hn (M • ) → Hn+1 (K • ) → · · ·
Proof. This is a consequence of the snake lemma. ■
Corollary 7.1.16 The functor H n : K(A) → A is cohomological.
Proof. If we are given a distinguished triangle K • → L• → M • → K • [1], we may

assume that M • = M (f )• and the sequence 0 → L• → M • → K • [1] → 0 is therefore
exact. It follows from the theorem that Hn (K • ) → Hn (L• ) → Hn (M • ) is exact in
the middle. ■
7.1.5 Quasi-isomorphism
Definition 7.1.17 A quasi-isomorphism of complexes f : K • → L• is a morphism
such that Hn (f ) is an isomorphism for all n ∈ Z. An acylic complex K • is a
complex such that Hn (K • ) = 0 for all n ∈ Z.
Note that being quasi-isomorphic is not a symmetric relation. Also, a complex is

acyclic if and only if [0] is quasi-isomorphic to K • if and only if K • is quasi-isomorphic
to [0].
+
Exercise 7.16 Show that if K • → L• → M • → is a distinguished triangle, then f
is a quasi-isomorphism if and only if M • is acyclic.
Exercise 7.17 Show that, if
K• / L• / M•
+ /
u v w

K ′• / L′• / M ′•
+ /
is a morphism of distinguished triangles and u, v are quasi-isomorphism, then w

is also a quasi-isomorphism.
Exercise 7.18 1. Show that I ∈ A is injective if and only if any monomorphism

I ,→ M has a retraction if and only if hI is exact if and only if any extension
by I is trivial (and dual).
2. Show that if 0 → M ′ → M → M ′′ → 0 is an exact sequence with M ′
injective, then M is injective if and only if M ′′ is (and dual).
1
With a lot of fun - assume A = A−Mod to make it easier.
7.1 Complex 119
3. Show that, if a functor G has an exact adjoint, then G preserves injective

objects (and dual).
We denote by I (resp. P) the full subcategory of injective (resp. projective)
objects.
Lemma 7.1.18 If K • is acyclic and I • ∈ C+ (I) then any morphism f : K • → I •
is homotopic to zero (and dual).
Proof. 2 Since I • is bounded below, we may assume that we have built a homotopy
sk up to k = n. In particular, we have f n−1 = sn ◦ dn−1 + dn−2 ◦ sn−1 . Thus, if we
set g n := f n − dn−1 ◦ sn , we have
g n ◦ dn−1 = f n ◦ dn−1 − dn−1 ◦ sn ◦ dn−1

= dn−1 ◦ f n−1 − dn−1 ◦ sn ◦ dn−1
= dn−1 ◦ (f n−1 − sn ◦ dn−1 )
= dn−1 ◦ dn−2 ◦ sn−1
= 0.
In other words gn is zero on Z n (K • ). Since K • is acyclic, its means that g n factors

through Z n+1 (K • ). Since I n is injective, it can be extended to sn+1 : K n+1 → I n
and we have g n = sn+1 ◦ dn so that
f n = sn+1 ◦ dn + dn−1 ◦ sn . ■
Proposition 7.1.19 If K • → L• is a quasi-isomorphism and I • ∈ C+ (I), then
HomK(A) (L• , I • ) = HomK(A) (K • , I • ) (and dual).
Proof. There exists a distinguished triangle K • → L• → M • → K • [1]. Applying the

cohomological functor HomK(A) (−, I • ) provides an exact sequence
HomK(A) (M • [−1], I • ) → HomK(A) (L• , I • ) → HomK(A) (K • , I • ) → HomK(A) (M • , I • ).
Since M • is acyclic, we can apply lemma 7.1.18. ■
Exercise 7.19 Show that a complex I • ∈ C+ (I) is acyclic if and only if it is

homotopically trivial (and dual).
Exercise 7.20 Show that I • , J • ∈ C+ (I) are quasi-isomorphic if and only if they
are homotopically equivalent (and dual).
Definition 7.1.20 If f : K • → L• is a quasi-isomorphism, we also say that L• is a

right resolutiona of K • . It is called an injective resolution if each Ln is injective.
a
Or replacement.
A left (resp. a projective) resolution is a right (resp. an injective) resolution in

A .
op
2
The proof in [Sta19, Tag 013R] is not correct.
Examples 1. A sequence 0 → M → K 0 → K 1 → · · · is exact if and only if

[M ] → K • is a right resolution.
2. Any abelian group M has a free (projective) resolution of length 2: a short
exact sequence 0 → L1 → L0 → M → 0 with L0 and L1 free.
Exercise 7.21 Show that if K • → I • and L• → J • are two injective resolutions in

C+ (A), then any morphism f : K • → L• extends to I • → J • (and dual).
Proposition 7.1.21 If A has enough injectives and K • ∈ C+ (A), then there exists
an injective resolution K • → I • made of injective maps (and dual).
Proof. We may assume that K n = 0 for n < 0 and, by induction, that there exists
an injective morphism of complexes with right exact lines
IO 0 / I1 / ··· / I n−2 / I n−1 / DO n / 0

O O O
? ? ? ? ?
K0 / K1 / ··· / K n−2 / K n−1 / Cn / 0
such that Hm (K • ) ≃ Hm (I • ) is an isomorphism for m < n − 1. Then, there exists

an injection E n := (K n ⊕ Dn )/C n ,→ I n into an injective. We obtain injective maps
K n ,→ I n and C n+1 ,→ Dn+1 on the cokernels and a morphism I n−1 → I n such that
Hn−1 (K • ) ≃ Hn−1 (I • ). This may be shown3 by playing around with the following
diagram:
I n−2 / I n−1 / / Dn / In // Dn+1 ■

O O O ; O O
" -
n
E
< c
? ? ? Q ?n
K n−2 / K n−1 / / Cn /1 K / / C n+1
7.2 Derived functor

7.2.1 Derived category
Let A be an abelian category.
Proposition 7.2.1 The category K(A) admits both left and right calculus of fractions
with respect to quasi-isomorphisms.
Proof. An identity is a quasi-isomorphism and quasi-isomorphisms are clearly stable

under composition. Assume now that we are given a morphism K • → L• and
a quasi-isomorphism L′• → L• . Then, there exists a morphism of distinguished
3
With a lot of fun - assume A = A−Mod to make it easier.
triangles (build N • first and then K ′• )
K ′ [−1] / K• / N• / K ′•

L′• / L• / N• / L′• [1].
Since L′• → L• is a quasi-isomorphism, N • is acyclic and therefore K ′ [−1] → K • is

also a quasi-isomorphism. The last condition is proved in the same way. The left
property also is shown with the same method. ■
Definition 7.2.2 The derived category D(A) of A is the localization of K(A) at

quasi-isomorphisms.
Thus, D(A) has the same objects as K(A) (or equivalently C(A)) and a morphism
K • to L• is a diagram
K ′•
"
K• L•
where the vertical map is a quasi-isomorphism. A morphism of complexes K • → L•

is a quasi-isomorphism if and only if it becomes an isomorphism in D(A). Actually,
D(A) is also the localization of C(A) at quasi-isomorphisms (even it this last category
does not admits right or left calculus of fraction).
One can also define D+ (A), D− (A) and Db (A) along the same lines. Moreover,
there exists a canonical embedding D+ (A) ,→ D(A) whose essential image is made
of complexes such that Hn (K • ) = 0 for n << 0 (and analogous statements with −
and b).
Definition 7.2.3 A triangle in D(A) is a diagram
K • → L• → M • → K • [1].
A morphism of triangles is a commutative diagram (in D(A))
K• / L• / M• / K • [1]
u v w u[1]

K ′• / L′• / M ′• / K ′• [1].
The triangle is said to be distinguished if it is isomorphic (in D(A)) to

f g h
K • → L• → M (f )• → K • [1].
f g
Proposition 7.2.4 If 0 → K • → L• → M • → 0 is an exact sequence of complexes,
f g
then there exists a distinguished triangle K • → L• → M • → K • [1] in D(A).
Proof. The composite maps M (f )n ↠ Ln ↠ M n defines a morphism of complexes

and there is an exact sequence
0 → M (IdK • ) → M • (f ) → M • → 0.
It follows from corollary 7.1.16 that M (IdK • ) is acyclic and then from theorem 7.1.15
that M • (f ) → M • is a quasi-isomorphism. ■
Proposition 7.2.5 If K • ∈ C(A) and I • ∈ C+ (I), then
HomK(A) (K • , I • ) = HomD(A) (K • , I • ) (and dual).
Proof. Formal consequence of proposition 7.1.19. ■
Theorem 7.2.6 Assume that A has enough injectives. Then there exists an equiva-
lence
K+ (I) ≃ D+ (A) (and dual).
Proof. Follows from propositions 7.1.21 and 7.2.5. ■
7.2.2 Derived functor

Let F : A → A′ be a functor between two abelian categories with A having enough
injectives (and dual).
Definition 7.2.7 The right derived functor a of F is the unique (up to isomorphism)
functor RF making commutative the following diagram:
K+ (A)
F / K+ (A′ )
O
?
K+ (I)
≃

+
D (A)
RF / D (A′ ).
+
a
This definition is usually only applied to left exact functors.
In practice, if K • → I • is an injective resolution, then RF K • ≃ F I • and this

can be made functorial. We will then set
∀n ∈ N, Rn F K • := Hn (RF K • ).
If M ∈ A, we may consider RF M := RF [M ] ∈ D+ (A′ ) and for all n ∈ Z, Rn F M ∈

A′ (and dual). One defines dually LF := RF op and Ln F K• := Hn (LF K• ) when A
has enough projectives.
Exercise 7.22 Show that if F is left exact, then RF is the (left) Kan extension of
F
the composite map K+ (A) → K+ (A′ ) → D+ (A′ ) along K+ (A) → D+ (A) (and
dual).
Theorem 7.2.8 Any short exact sequence
0 → K • → L• → M • → 0
in C+ (A) gives rise to a long exact sequence
· · · → Rn F K • → Rn F L• → Rn F M • → Rn+1 F K • → · · · (and dual).
Proof. We know from propositon 7.2.4 that there exists a distinguished triangle
K • → L• → M • → K • [1] in D(A). Thanks to theorem 7.2.6, we may assume that
all complexes are composed of injective objects and maps defined in C(A). We can
then apply F and then use corollary 7.1.16 ■
Exercise 7.23 Show that F is left exact (resp. exact) if and only if ∀M ∈
A, R0 F M ≃ F M (resp. RF M ≃ FM ) (and dual).
Solution. Let M → I • be an injective resolution. Assume first that F is left exact.

Then, since M ≃ H0 (I • ) ≃ ker(I 0 → I 1 ), we will have F M ≃ ker(F I 0 → F I 1 ) ≃
H0 (F I • ) = R0 F M . If F is exact, then F M ≃ F I • ≃ RF M . The converse follows
from theorem 7.2.8. ■
Definition 7.2.9 The n-th extension group of L• by K • is
Extn (K • , L• ) := HomD(A) (K • , L• [n]).
Example 1. We have Extn (M, N ) = 0 for n < 0,

2. we have Ext0 (M, N ) = Hom(M, N ),
3. Ext1 (M, N ) = Ext(M, N ) classifies extensions of N by M up to isomorphism,
4. we have Extn (M, N ) = 0 for n > 1 in Ab,
5. Ext(Z/nZ, Z/mZ) = Z/dZ with d = m ∧ n.
Proposition 7.2.10 If M ∈ A and K • ∈ C+ (A), then
Rn Hom(M, K • ) = Extn (M, K • ) (and dual).
Proof. It is sufficient to prove the assertion when K • [n] = I • ∈ C+ (I). Thanks to

proposition 7.2.5, we have to show that
HomK(A) (M, I • ) = H0 (Hom(M, I • )).
A quick inspection shows that
HomC(A) (M, I • ) = Z0 (Hom(M, I • ))
and f ∈ B0 (Hom(M, I • )) if and only if the corresponding map of complexes is

homotopic to zero. ■
Note in particular, that if A has enough injectives and projectives, then we

can use an injective resolution of N or a projective resolution of M to compute
Extn (M, N ).
Exercise 7.24 Show that (even if A does not have enough projectives), then
Extn (P, M ) = 0 for n ̸= 0 when P is projective.
Solution. If M → I • is an injective resolution, then Hom(P, M ) → Hom(P, I • ) also

and therefore Hom(P, M ) ≃ RHom(P, M ). ■
Exercise 7.25 Show that if M, N are two abelian groups, then Extn (M, N ) = 0
for n ̸= 0, 1 and Ext0 (M, N ) = Hom(M, N ). Show that Ext1 (Z/kZ, N ) ≃ N/kN .
Definition 7.2.11 An object M ∈ A is said to be (right) F -acyclic if F M = RF M

(i.e. F M = R0 F M and Rn F M = 0 for n ̸= 0).
Proposition 7.2.12 — Leray acyclicity. Show that if K • ∈ C+ (A) and each K n is

F -acyclic, then RF K • = F K • .
Proof. We may assume that K n = 0 for n < 0. Assume first that K • is bounded
and denote by K ′• := K 1 → K 2 → · · · . Then, there exists an exact sequence
0 → K ′• → K • → [K 0 ] → 0. It follows that there exists a morphism of distinguished
triangles
F K ′• / F K• / F K0 / F K ′• [1]

RF K ′• / RF K • / RF K 0 / RF K ′• [1].
in D(A). By induction, the middle arrow is also an isomorphism. In général, if we

set K ′• = K n+2 → K n+3 → · · · and K ′′• = K 0 → · · · → K n+1 (bounded), we get an
exact sequence 0 → K ′• → K • → K ′′• → 0. It follows that there exists a morphism
of long exact sequences
0 / H nF K •
≃ / H n F K ′′• / 0 ■
≃

0 / Rn F K •
≃ / Rn F K ′′• / 0.
As a consequence, if M → K • is a right resolution with each K n is F -acyclic,

then RF M = F K • and therefore
∀n ∈ N, Rn F M = H n (F K • ).
Exercise 7.26 Show that an object is injective if and only if it is F -acyclic for all
left exact functor F if and only if this is the case when F = Hom(M, −) for all
M ∈ A (and dual).
Corollary 7.2.13 Assume A′ also has enough injective, A′′ is another abelian
category and G : A′ → A′′ an additive functor. Assume that F I is G-acyclic
whenever I is injective. Then R(G ◦ F ) = RG ◦ RF .
Proof. Follows from Leray acyclicity. ■
Note that the condition is automatic if F has an exact adjoint.
7.2.3 Spectral sequence

Let A be an abelian category (with exact countable direct sums4 ).
Definition 7.2.14 A (decreasing) filtration on an object M ∈ A is diagram on
(Z, ≥) of subobjects of M :
M ⊃ · · · ⊃ Fn M ⊃ Fn+1 M ⊃ · · · ⊃ 0.
We shall always assume that the filtration is : n∈Z F M = M and

n
S
exhaustive
separated : n∈Z F n M = 0. It is said to be finite if F n M = M for n << 0 and
T
F n M = 0 for n >> 0
We shall denote this category by F(A) and write for each n ∈ Z, Grn M =
Fn M/Fn+1 M .
Exercise 7.27 Show that F(C(A)) ≃ C(F(A)).
Definition 7.2.15 A spectral sequence
Erp,q
0
⇒ H p+q
is
1. a family of complexes with p, q, r ∈ Z and r ≥ r0
drp−r,q+r−1 dp,q
· · · −→ Erp−r,q+r−1 −→ Erp,q −→
r
Erp+r,q−r+1 −→ · · ·
such that
p,q
Er+1 = H p,q (Er ) := ker(dp,q p−r,q+r−1
r )/Im(dr ),
2. a family of filtered objects H n for n ∈ Z such that
∀p, q ∈ N, Grp H p+q = Erp,q for r >> 0.
It is called a first quadrant spectral sequence if Erp,q = 0 unless p, q ∈ N.
Exercise 7.28 Represent E0 , E1 and E2 with p, q as coordinates.
Exercise 7.29 Show that, if r ≥ 1, then ∀p ̸= 0, Erp,q = 0 ⇒ ∀q ∈ Z, Er0,q ≃ H q

and if r ≥ 2, then ∀q ̸= 0, Erp,q = 0 ⇒ ∀p ∈ Z, Erp,0 ≃ H p .
4
For example a Grothendieck category.
Exercise 7.30 Show that, in a first quadrant spectral sequence, the sequence
0 → E21,0 → H 1 → E20,1 → E22,0
is exact.
Theorem 7.2.16 Let K • is a filtered complex. Assume K • is bounded below. Then

there exists a spectral sequence
E0p,q = Grp K p+q ⇒ Hp+q K • .
Proof. To do. ■
Exercise 7.31 Show that E1

p,q
= Hp+q (Grp K • ).
Definition 7.2.17 A bicomplex (in an additive category) is a complex of complexes.
In other words, a bicomplex is a diagram (K p,q , dp,q , d′p,q ) on (Z, ≤)2 such that
for all p, q ∈ Z,
dp+1,q ◦ dp,q = 0, d′p,q+1 ◦ d′p,q = 0 and d′p+1,q ◦ dp,q = dp,q+1 ◦ d′p,q .
Exercise 7.32 Show that if K •,• is a bicomplex, and we endow K n := K p,q

L
p+q=n
with d := ⊕p+q=n (d
n p,q
+ (−1) d p ′p,q
), then K is a complex.
•
It is called the simple complex associated to the bicomplex or the total complex
of the bicomplex.
Proposition 7.2.18 Let K •,• be a bicomplex. Assume that it is bounded below (in
both variables). Then there exists a spectral sequence
E0p,q = K p,q ⇒ Hp+q (K • ).
Proof. Apply theorem 7.2.16 with Fp K n := K p,q .

L
i+j=n,i≥p ■
Exercise 7.33 Show that E1 = Hq (K p,• ) and E2p,q = Hp (Hq (K •,• )).
p,q
Lemma 7.2.19 If K • is bounded below, then there exists a bounded below bicom-
plex I •,• with I •,q = 0 for q < 0 and a morphism of complexes K • → I •,0 such
that each K p → I p,• and H p (K • ) → H p (I •,• ) are injective resolutions.
Proof. To do. ■
This is called a Cartan-Eilenberg resolution.
Exercise 7.34 Show that if I • denotes the associated simple complex, then K • → I •
is also an injective resolution.
F
Proposition 7.2.20 Let A → A′ be an additive functor. Assume A has enough
injectives. If K • is bounded below, then there exists two spectral sequences

′
E1p,q = Rq F K p ⇒ Rp+q F K • and ′′
E2p,q = Rp F (H q (K • )) ⇒ Rp+q F K • .
Proof. Let I •,• be a Cartan-Eilenberg resolution of K • . Then, there exists a spectral

sequence
′
E1p,q = Hq (F I p,• ) ⇒ Hp+q (F I • ).
But we may also exchange the rôle of p and q and consider the spectral sequence
′′
E2p,q = Hq (F Hp (I •,• )) ⇒ Hp+q (F I • ). ■
F G
Corollary 7.2.21 Let A → A′ → A′′ be a sequence of additive functors. Assume A
and A′ have enough injectives and F I is G-acyclic whenever I is injective. If K •
is bounded below, then there exists a spectral sequence
E2p,q = Rp G(Rq F (K • )) ⇒ Rp+q (G ◦ F )(K • ).
Proof. We may assume that K • ∈ C+ (I) and apply the proposition to F K • . ■
Exercise 7.35 Show that, in the situation of the corollary, when F (resp. G) is
exact, there is an isomorphism
Rn G(F (K • )) ≃ Rn (G ◦ F )(K • ) (resp. G(Rn F (K • )) ≃ Rn (G ◦ F )(K • )).
7.3 Sheaf cohomology

7.3.1 Definition
Recall that, if C is a site and X ∈ C, then there exists a (left exact additive) functor
C(Ab)
e → Ab, M → Γ(X, M) := M(X).
Definition 7.3.1 If M• is a complex of abelian sheaves on a site C, then its nth

cohomology group on X ∈ C is
Hn (X, M• ) := Rn Γ(X, M• ).
Examples The above definition applies in particular to an abelian complex on a

topological space. We give (without details) a list of examples from geometry that
show that sheaf cohomology agrees with classical cohomology:
1. If X is a locally contractible topological space, then
Hn (X, Z) ≃ Hnsing (X).
2. If X is a differentiable manifold (and there exists a complex analog), then
Hn (X, R) ≃ Hn (X, Ω•X/R ) ≃ Hn (Ω• (X)) =: HndR (X/R).

3. If X is a smooth algebraic variety over C, we have (GAGA)
HndR (X/C) := Hn (X, Ω•X/C ) ≃ Hn (X an , C)
Note that, by definition, if f : Y → X is a morphism in C, there exists a canonical

maps
RΓ(X, M• ) → RΓ(Y, M• ) and Hn (X, M• ) → Hn (Y, M• ).
Cohomology can be computed in the topos:

Exercise 7.36 Show that RΓ(X, M• ) = RΓ(X, M• ) and therefore Hn (X, M• ) =
Hn (X, M• ).
Solution. Choosing an injective resolution I • , it is sufficient to show that Γ(X, I • ) =

Γ(X, I • ). But we know that for any sheaf M, we have Γ(X, M) = Γ(X, M) ■
Exercise 7.37 Show that in a topos T , we have RΓ(X, M • ) ≃ RHomZ (Z · X, M • )

and therefore Hn (X, M • ) ≃ ExtnZ (Z · X, M • ).
Exercise 7.38 Show that in a topos T , we have Hn •

≃ i∈I Hn (Xi , M • ).
` Q
i∈I X i , M
Solution. Choose an injective resolution, use left exactness of global section and the
fact that products are exact in abelian groups. ■
On a topos T (with enough projectives - but this not really necessary), we can
derive our functors HomZ and ⊗Z (on one side or the other) and obtain
RHomZ (M, N ) and M ⊗LZ N.
Cohomology is then usually denoted respectively by
ExtnZ (M, N ) and T orZn (M, N ).
Exercise 7.39 Show that, in a topos, if P is a flat abelian group and M is a

complex of abelian groups which is bounded below and acyclic, then P ⊗Z M is
also acyclic.
7.3.2 Simplicial method

If f : X0 → X is a morphism in a topos T , we may then consider the (semi-)
simplicial object
[n] 7→ Xn := Y ×X · · · ×X Y
| {z }
n+1
so that
/ /
/ / /
X• : / X ×X ×Y ×X Y / Y ×X Y / Y.
/
We shall also consider the augmented (semi-) simplicial object

/ /
/ / / /
X•+ : / X ×X ×Y ×X Y / Y ×X Y / Y X.
/
More generally, if X := (Xi → X)i∈I is a family of morphisms, we shall simply write
! !+
a a
X• = Xi and X•+ = Xi .
i∈I • i∈I •
Lemma 7.3.2 If X := (Xi → X)i∈I is a covering in T , then Z · X•+ is acyclic.
Proof. It is sufficient to consider the case of an epimorphism X0 ↠ X. When

T = Set, it reduces to X = {x} in which case, this is clear. It follows that this is also
true when T := Cb for any category C. Finally, if C is a site, then any epimorphism
in T := Ce is the sheafification of an epimorphism in C.
b ■
If X := (Xi → X)i∈I is a family of morphisms and M is an abelian group of T ,
then we set
Cˇ• (X , M ) := HomZ (Z · X• , M ), Cˇ• (X + , M ) = HomZ (Z · X•+ , M )
and
∀n ∈ N, Ȟn (X , M ) := Hn (Cˇ• (X , M )).
When X := (Xi → X)i∈I is a family of morphisms in a site C and X := (X i → X)i∈I ,
we shall write
Cˇ• (X , M ) := Cˇ• (X , M ), Cˇ• (X + , M ) := Cˇ• (X + , M ), Ȟn (X , M ) := Ȟn (X , M ).
Exercise 7.40 Show that if X := (Xi → X)i∈I is a family of morphisms in a site

C with fibered products and M a sheaf of abelian groups on C, then for n ≥ 0,
Y
Cˇn (X , M ) = Cˇn (X + , M ) = M (Xi0 ×X · · · ×X Xin ).
i0 ,...,in
Proposition 7.3.3 If a morphism X0 ↠ X in T has a section s and M is an abelian

group of T , then Cˇ• ((X0 → X)+ , M ) is homotopically trivial. The homotopy has
the form f 7→ f ◦ sn with sn : Xn → Xn+1 .
Proof. We consider for all n ≥ −1 the morphism

sn : s ×X Id : Xn → Xn+1
(so that s−1 = s). It induces a morphism
sn : Z · Xn → Z · Xn+1
and we have for all n ≥ −1, Idn = sn+1 ◦ dn + dn−1 ◦ sn . In other words, Z · X•+ is
homotopically trivial and so is Cˇ• ((X0 → X)+ , M ) = HomZ (Z · X•+ , M ). ■
If C is any category and we endow Cb with its canonical topology, we shall consider
RΓ(T, M ) for any presheaf of sets T and any presheaf of abelian group M .
Exercise 7.41 Show that if X ∈ C, then Hn (hX , M ) = 0 for n ̸= 0.
Proposition 7.3.4 If C is any category and R is the sieve generated by a family

X := (Xi → X)i∈I , then
RΓ(R, M) ≃ Cˇ• (X , M)
for all abelian presheaf M on C (and therefore Rn Γ(R, M) ≃ Ȟn (X , M)).
Proof. The family X := (hXi → R)i∈I is a covering in C. b It therefore follows from

lemma 7.3.2 that Z · hX• ≃ Z · R. This is a projective resolution and therefore
RΓ(R, M) ≃ RHomZ (Z · R, M)
≃ RHomZ (Z · hX• , M)
≃ HomZ (Z · hX• , M)
= Cˇ• (X , M). ■
7.3.3 Čech cohomology

Let C be a site with fibered products.
We consider the inclusion functor
H : C(Ab)
e ,→ C(Ab)
b
(so that H(M) denotes the sheaf M seen as a presheaf) and write for all n ∈ Z and
all complex M• of abelian sheaves, Hn (M• ) = Rn H(M• ).
Exercise 7.42 Show that if X ∈ C, then Γ(X, RH(M• )) = RΓ(X, M• ) and
therefore Γ(X, Hn (M• )) = Hn (X, M• ).
Solution. The functor H : C(Ab)

e ,→ C(Ab)
b has an right adjoint. Therefore, it
preserves injectives. Moreover, the functor M 7→ Γ(X, M) is exact on presheaves.
We can apply corollary 7.2.13. ■
Exercise 7.43 Show that if M is an abelian sheaf, then H^

n (M) = 0 for n ̸= 0.
Exercise 7.44 Show that if X is family of morphisms of C and M is a sheaf on C,

then Ȟn (X , H(M )) = Ȟn (X , M ).
Recall no that there exists a left exact functor (see (3.1))
Ȟ : C(Ab)
b → C(Ab),
b ∀X ∈ C, Ȟ(M)(X) = lim Hom(R, M)
−→
R∈J(X)
such that Ȟ(Ȟ(M)) = H(M).

f We shall write Ȟn (M• ) = Rn Ȟ(M• ).
Exercise 7.45 Show that Ȟ(Hn (M)) = 0 for n ̸= 0.
Definition 7.3.5 If M• is a complex of abelian presheaves on C, its nth Čech

cohomology group on X is
Ȟn (X, M• ) := Γ(X, Ȟn (M• )).
Exercise 7.46 Show that if X ∈ C and M is an abelian presheaf, then
Ȟn (X, M) = lim Hn (R, M) = lim Ȟn (X , M).

−→ −→
R∈J(X) X ∈Cov(X)
Exercise 7.47 Show that Ȟn ([0, 1], M ) = 0 for n =

̸ 0 if M is a constant abelian
group.
Solution. Any open covering of [0, 1] has a refinement of the form [0, 1] = rk=0 Ik
S
where Ik is an interval and Ik ∩ Ik−1 = Jk is also an interval for k = 1, . . . r (make a
pictutre). Then, the augmented Čech complex
M / M r+1 /Mr
(sk ) / (sk+1 − sk )
is acyclic. ■
We shall apply the above definition when M• is a complex of abelian sheaves so
that
Ȟn (M• ) := Ȟn (H(M• )) and Ȟn (X, M• ) := Ȟn (X, H(M• )).
Theorem 7.3.6 — Cartan-Leray. If X is a covering of X and M is an abelian sheaf,

then there exists a spectral sequence
E2p,q := Ȟp (X , Hq (M)) ⇒ Hp+q (X, M).
Proof. If R denote the sieve generated by X , then Γ(X, M) = Γ(R, H(M)). Since
H preserves injectives, the spectral sequence is obtained from corollary 7.2.21 and
proposition 7.3.4. ■
Corollary 7.3.7 Assume Hq (Xk , M) = 0 for all q ̸= 0 and k ∈ N. Then, there exists
an isomorphism
∀n ∈ N, Ȟn (X , M ) ≃ Hn (X, M).
Proof. Assume q ̸= 0. For all k ∈ N, we have Γ(Xk , Hq (M)) = Hn (X, M) = 0 and

therefore C(X
ˇ , Hq (M)) = 0 so that E2p,q = 0. ■
Corollary 7.3.8 If X ∈ C and M is an abelian sheaf, then there exists a spectral

sequence
E2p,q := Ȟp (X, Hq (M)) ⇒ Hp+q (X, M).
Proof. Filtered direct limits are exact and therefore preserve spectral sequences. ■
Exercise 7.48 Show that Ȟ1 (X, M) = H1 (X, M).
Solution. We consider the spectral sequence from corollary 7.3.8. It follows from
exercise 7.45 that E20,1 = 0. Our assertion therefore follows from exercise 7.30 ■
Definition 7.3.9 An abelian sheaf M on C is said to be acyclic if Hn (M) = 0 for

n ̸= 0.
Be careful that this definition depends on the category C and not only on the
topos C.
e
Proposition 7.3.10 An abelian sheaf M on C is acyclic if and only if Ȟn (M) = 0

for n ̸= 0.
Proof. It follows from corollary 7.3.8 and exercise 7.42 that there exists a spectral
sequence of presheaves
E2p,q := Ȟp (Hq (M)) ⇒ Hp+q (M).
The implication follows immediately. Conversely, it is sufficient to prove by induction

on n > 0 that for all p, q ∈ N such that 0 < p + q ≤ n, we have E2p,q = 0. If this is
the case and 0 < q ≤ n, then Hq (M) = 0, but then also E2p,q = 0 for all p ∈ N (and
0 < q ≤ n). Now, it follows from exercise 7.45 that E20,q = 0 for q ̸= 0 and from our
hypothesis that E2p,0 = 0 for p ̸= 0. Therefore, our assertion is satisfied for n = 1
(E20,1 = E21,0 = 0) and extends from n to n + 1 (E20,n+1 = 0). ■
Exercise 7.49 Show that the following are equivalent:

1. M is acyclic on C,
2. If X ∈ C, then Hn (X, M) = 0 for n ̸= 0,
3. If X is a covering family in C, then Ȟn (X , M) = 0 for n =
̸ 0 (equivalently
C(X , M) is an acyclic complex),
ˇ +
4. Iff X ∈ C, then Ȟn (X, M) = 0 for n ̸= 0.
Solution. (1) ⇔ (2) follows from exercise 7.42. Then, (1) ⇒ (3) follows from theorem
7.3.6. And (3) ⇒ (4) is obtained by taking the limit. Finally, (1) ⇔ (4) follows from
the proposition. ■
Note that this is also equivalent to C(X
ˇ + , M) being an acyclic complex.
If M is an abelian sheaf on a topological space X, we shall denote its cohomological
groups by Hnsheaf (X, M) or Hn (X, M) when there is no ambiguity.
Proposition 7.3.11 If X is a compact Hausdorff space, then
Ȟn (X, M) ≃ Hn (X, M).
Proof. Classic. ■
Exercise 7.50 Show that if S is a Stone space, then Hn (S, M) = 0 for n ̸= 0.
Solution. Since S is a compact Hausdorff, we can use Čech cohomology. We saw

in exercise 2.19 that any covering has a finite disjoint clopen refinement S. It is
therefore sufficient to show that Ȟn (S, M) = 0. But then, Cˇ• (S, M ) is concentrated
in degree 0. ■
Proposition 7.3.12 If X = limi∈I Xi is a filetered limit of compact Hausdorff spaces

←−
and M is a constant abelian group, then
Hn (X, M ) ≃ lim Hn (Xi , M ).

−→
i∈I
Proof. We can use Čech cohomology. If we denote by πi : X → Xi the projection,

then any covering of X has a refinement of the form πi−1 (X ) for some covering X of
some Xi . Since M is constant, we have
ˇ −1 (X ), M ) ≃ C(X
C(π ˇ , M ) so that Ȟn (π −1 (X ), M ) ≃ Ȟn (X , M ).
i i
We conclude with exercise 7.46. ■
7.4 Morphisms of topos (optional)

7.4.1 Morphism
Recall that a morphism of topos f = T → T ′ is a couple of adjoint functors
f −1 : T ′ → T (inverse image) and f∗ : T → T ′ (direct image) with f −1 exact.
Exercise 7.51 Show that a morphism of topos induces an adjunction on abelian
groups on both sides with exact inverse image.
f∗ HomZ (f −1 M, N ) = HomZ (M, f∗ N ) and f −1 (M ⊗Z N ) ≃ f −1 M ⊗Z f −1 N.
Exercise 7.53 Show that if f : T → T ′ is a morphism of topos, then f∗ preserves

injectives.
Exercise 7.54 Show that if f : T → T ′ is a morphism of topos, there exists

canonical maps
Hn (X ′ , M • ) → Hn (f −1 (X ′ ), f −1 M • ).
Solution. There exists a canonical map
Γ(X ′ , M ) → Γ(X ′ , f∗ f −1 M ) ≃ Γ(f −1 (X ′ ), f −1 M ).
Let M • → I • and f −1 I • → J • be two injective resolutions. Since f −1 is exact, the

composite map f −1 M • → J • is also an injective resolution and therefore
RΓ(X ′ , M • ) = Γ(X ′ , I • ) → Γ(f −1 (X ′ ), f −1 I • ) → Γ(f −1 (X ′ ), J • ) = RΓ(f −1 (X ′ ), f −1 M • )
f f′
Exercise 7.55 Show that if T → T ′ → T ′ is a sequence of morphisms of topos,
then Rf∗′ ◦ Rf∗ = R(f ′ ◦ f )∗ and there is a spectral sequence
E2p,q = Rp f∗′ (Rq f∗ (M • )) ⇒ Rp+q (f ′ ◦ f )∗ (M • ).
Solution. Our assertion follows from proposition 7.2.13 and corollary 7.2.21. ■
Exercise 7.56 Show that if f : T → T ′ is a morphism of topos and X ′ ∈ T ′ , then

there is a spectral sequence
E2p,q = H p (X ′ , Rq f∗ (M • )) ⇒ Hp+q (f −1 (X ′ ), M • ).
Proof. By adjunction, we have Γ(X ′ , f∗ M ) = Γ(f −1 (X ′ ), M ). ■
Proposition 7.4.1 If f : T → T ′ is a morphism of topos, then Rn f∗ M is the sheaf

associated to X ′ 7→ Hn (f −1 (X ′ ), M ).
Proof. The morphism fb : Tb → Tb induced on presheaves as well as sheafification are

^ ^
both exact. From f∗ M = fb∗ H(M ), we obtain Rf∗ M = fb∗ RH(M ). It follows that
Rn f∗ M is the sheaf associated to fb∗ Hn (M ). By adjunction, we have
Γ(X ′ , fb∗ Hn (M )) = Γ(f −1 (X ′ ), Hn (M )) = Hn (f −1 (X ′ ), M ). ■
7.4.2 Localization
Exercise 7.57 Show that if T is a topos and X ∈ T , then the functor jX
−1
:
Ab(T ) → Ab(T/X ) has an exact left adjoint jX! : Ab(T/X ) → Ab(T ).
Solution. In the case T = C,

b the functor is given by the explicit formula jS! M (X) =
s:X→S M (s). In general, one uses sheafification.
L
■
Be careful that the functors jX! on abelian groups and set are not compatible.
Exercise 7.58 Show that jX! Z = Z · X.
Solution. It follows from exercise 3.58 that jX! 1X = X where 1X = IdX denotes the
final object of T/X . Therefore, we have
−1
HomZ (jX! Z, M ) ≃ HomZ (Z, jX M)
−1
= Hom(1X , jX M )
= Hom(jX! 1X , M )
= Hom(X, M )
= HomZ (Z · X, M ). ■
Exercise 7.59 Show that Hn (1X , jX

−1
M • ) = Hn (X, M • ).
Solution. We have
−1
HomZ (Z, jX M ) = HomZ (jX! Z, M ) = HomZ (Z · X, M ).
Since jX
−1
has an exact adjoint, it preserves injectives and therefore
−1
RHomZ (Z, jX M • ) = RHomZ (Z · X, M • )
It is then sufficient to take cohomology on both sides. ■

−1
HomZ (M, N )(X) = HomZ (jX! jX M, N )
−1 −1 −1
= HomZ (jX M, jX N ) = HomZ (M, jX∗ jX N ).

−1 −1
jX! (M ⊗Z jX N ) = jX! M ⊗Z N, jX! (jX M ⊗Z N ) = M ⊗Z jX! N
−1
and M · X := M ⊗Z Z · X = jX! jX M.
7.4.3 Topological spaces

Recall that, if X is a topological space, then there exists various morphisms of topos
φX
g o
Top Top
^ /X o
/ Open(X).
^
jX
ψX
We shall call MX := φX∗ jX

−1
M the realization of an abelian sheaf M on Top. Then
we have a natural isomorphism
Hn (X, M • ) ≃ Hn (X, MX• ).
Actually, both functors jX−1

and ϕX∗ = ψX −1
induce equivalences on constant
abelian groups and we shall not make any difference in this case.
Proposition 7.4.2 If f ∼ g : X → Y are two homotopic continuous maps and M is

a constant abelian group, then the induced maps
Hn (Y, M ) → Hn (X, M ).
coincide.
Proof. Classic (see for example Schapira’s course on Algebra and Topology). ■
Corollary 7.4.3 If f : X → Y is a homotopy equivalence and M is a constant

abelian group, then
Hn (Y, M ) ≃ Hn (X, M ).
This applies in particular to a projection p : X × Y → Y when X is contractile.

8. Condensed cohomology (optional)
8.1 Cohomology
8.1.1 On Stonean spaces
If M is a condensed abelian group and X is a condensed set, we can consider the
cohomology groups Hn (X, M ). We may write Hncond (X, M ) in order to remove any
ambiguity. In the case X is a topological space (and M is a condensed abelian group),
then we may write Hn (X, M ) = Hn (X, M ). Also, if M is a topological abelian group
(and X a condensed set), we may simply write Hn (X, M ) := Hn (X, M ). Finally, we
may still denote by M the constant condensed abelian group associated to a usual
abelian group M .
Proposition 8.1.1 Condensed abelian groups are acyclic on Stonean spaces.
Proof. We saw in lemma 6.1.3 that the functor M 7→ Γ(S, M ) is exact on Stonean
spaces (and has therefore no higher cohomlogy). ■
In other words, we always have Hncond (S, M ) = 0 for n ̸= 0 when S is Stonean.
Using exercise exercise 7.49, this last result may also be deduced from the following:
Exercise 8.1 Show that, if M is a condensed abelian group and S := (Si ,→ S)ri=1
is a finite disjoint covering in CHaus, then Ȟn (S, M ) = 0 for n ̸= 0.
Solution. The canonical map i=1 Si → S, being an isomorphism, has a section. We

`
can then use proposition 7.3.3. ■
Unfortunately, the category of Stonean spaces does not have fibered products
and there is no Čech cohomology1 on this site. It will be necessary to rely on Stone
spaces.
1
There exists a workaround through the theory of hypercoverings.
138 Chapter 8. Condensed cohomology (optional)
Exercise 8.2 Show that condensed abelian groups are acyclic on discrete spaces,
and more generally on products E × S of a discrete space and a Stonean space.
Proof. Using exercises 7.38 and 4.6 and 3.43, we are reduced to the case of a
Stonean. ■
8.1.2 On Stone spaces

Lemma 8.1.2 Constant condensed abelian groups are acyclic on Stone spaces.
Proof. As shown in exercise 7.49, it is sufficient to prove that if M is an abelian

group and f : S0 ↠ S is a surjective map of Stone spaces, then the Čech complex
Cˇ• ((S0 → S)+ , M )
is acyclic (note that the case of a finite disjoint covering is taken care of by exercise
8.1). If S and S0 are finite, this follows from proposition 7.3.3. In general, we can
write S0 = limk S0k with S0k finite and we shall denote by S−1k the image S0k in S.
←−
Since M is discrete,
Cˇ• ((S0 → S)+ , M ) = lim Cˇ• ((S0k → S−1k )+ , M ).

−→
Since filtered direct limits are exact, we are done. ■
In other words, we always have Hncond (S, M ) = 0 for n =

̸ 0 when S is Stone and
M is a discrete abelian group.
Exercise 8.3 Show that if Q = R/Rdisc , then Q(S) := C(S, R)/C(S, Rdisc ) when S
is Stone.
Solution. There exists an exact sequence
0 → Γ(S, Rdisc ) → Γ(S, R) → Γ(S, Q) → H1cond (S, Rdisc ) = 0. ■
8.1.3 On locally compact spaces

We shall freely use here the notion of a morphism of topos.
Proposition 8.1.3 If X is a topological space, there exists a morphism of topos
^
cX : Cond/X → Open(X)
given by

Y 7→ U 7→ Y (U ) := Hom/X (U , Y ) and F 7→ F := lim U.
−→
U ⊂X,s∈F (U )
Proof. It follows from exercise 4.7 that the functors are well defined. Moreover,
inverse image is left exact because direct colimits are exact. Now, we have
Hom/X (F, Y ) = lim Hom/X (U , Y ).

←−
U ⊂X,s∈F (U )
In other words, a morphism F → Y is a compatible family of morphisms s : U → Y

over X for all open subset U of X and s ∈ F(U ). Conversely, a morphism F → Y is
a compatible family of maps F(U ) → Hom/X (U , Y ), s 7→ fs . This is the same thing
and it follows that there exists an adjunction
Hom/X (F, Y ) ≃ Hom(F, Y ). ■
Proposition 8.1.4 — Dyckson. If X is a locally compact Hausdorff space and M is

a discrete abelian group, then
Hncond (X, M ) = Hnsheaf (X, M ).
Proof. Since c−1

X X = X, there exists a spectral sequence
E2p,q = Hpsheaf (X, Rq cX∗ M ) ⇒ Hp+q

cond (X, M ).
Since cX∗ M = M , it is sufficient to show that Rn cX∗ M = 0 for n ̸= 0. This is the

sheaf associated to U 7→ Hncond (U, M ). Since X is locally compact, the stalk of this
sheaf at x ∈ X is
lim Hncond (U, M ) = lim Hncond (S, M )
−→ −→
x∈U x∈S
when U (resp. S) runs trough the open (resp. compact) neighbohoods of x. Thanks
to exercise 3.47, it is sufficient to show that this stalk is zero. Fix some compact
neighborhood K of x in X. Let f : K0 → K be a surjective map with K0 Stone. Let
S ⊂ K be a compact subset and S0 := f −1 (K0 ). Then, Sn is a Stone space for all
n ∈ N because a product of Stone, as well as a subspace of Stone is automatically
Stone. It follows from corollary 7.3.7 and lemma 8.1.2 that
Hncond (S, M ) = Ȟn (S• , M ).
Since filtered colimits are exact, we are reduced to the case S = {x} in which case
we can apply proposition 7.3.3. ■
8.2 Banach abelian groups

8.2.1 K-exactness
We recall that a semi-norm 2 on an abelian group M is a map M → R≥0 , s 7→ ∥s∥
satisfying ∥0∥ = 0, ∥s1 + s2 ∥ ≤ ∥s1 ∥ + ∥s2 ∥ and ∥ − s∥ = ∥s∥. The topology on M is
defined via the semi-distance δ(s1 , s2 ) = ∥s2 − s1 ∥. Semi-normed abelian groups form
a subcategory of the category of all topological abelian groups. Continuous homo-
morphisms u : M → N form a semi-normed abelian group for ∥u∥ := sups∈M ∥u(s)∥.
A semi-normed abelian group M is Hausdorff if and only if the semi-norm is a norm
: ∥s∥ = 0 ⇔ s = 0. A Banach abelian group is a complete normed abelian group.
Banach abelian groups form a reflexive subcategory of semi-normed abelian groups
with reflection M 7→ Mc.
2
It is actually sufficient to require that ∥ − ∥ : M → R satisfies ∥0∥ ≤ 0, ∥s1 + s2 ∥ ≤ ∥s1 ∥ + ∥s2 ∥
and ∥ − s∥ ≤ ∥s∥.
Definition 8.2.1 A complex M • of semi-normed abelian groups is said to be K-
bounded exact at M n for some K ∈ R if
∀s ∈ M n , ∀ϵ > 0, ∃s′ ∈ M n−1 , ∥s − dn−1 s′ ∥ ≤ K∥dn s∥ + ϵ.
The complex is said to be K-bounded acyclic if it is K-bounded exact at all M n .

In practice, we shall not use superscripts unless necessary and simply write d for
d .
n
Exercise 8.4 Let M • = limk Mk• a direct colimit of complexes of semi-normed

−→
abelian groups with isometric transitions maps. Show that, if each Mk• is K-
bounded exact in Mkn , then M • is K-bounded exact in M n .
Lemma 8.2.2 A complex M • of semi-normed abelian groups is K-bounded exact

at M n if and only if M
c• is K-bounded exact at M
cn .
Proof. We may clearly replace M • with its Hausdorff quotient and assume that
M• ⊂ M c• . For the implication, fix ϵ > 0 and s ∈ M
cn . There exists t ∈ M n such
that
ϵ
∥s − t∥ ≤
2(1 + K∥dn ∥)
Now, there exists s′ ∈ M n−1 such that ∥t − ds′ ∥ ≤ K∥dt∥ + ϵ/2. It follows that
∥s − ds′ ∥ ≤ ∥s − t∥ + ∥t − ds′ ∥
≤ ∥s − t∥ + K∥dt∥ + ϵ/2
≤ ∥s − t∥ + K∥ds∥ + K∥ds − dt∥ + ϵ/2
≤ K∥ds∥ + ∥s − t∥ + K∥dn ∥∥s − t∥ + ϵ/2
≤ ϵ.
Conversely, if s ∈ M n then there exists t′ ∈ M

cn−1 such that ∥s − dt′ ∥ ≤ K∥ds∥ + ϵ/2.
Then, there exists s′ ∈ M n−1 such that ∥s′ − t′ ∥ ≤ ϵ/(2∥dn−1 ∥) and therefore
∥s − ds′ ∥ ≤ ∥s − dt′ ∥ + ∥d(s′ − t′ )∥ ≤ K∥ds∥ + ϵ/2 + ∥dn−1 ∥∥s′ − t′ ∥ ≤ ϵ. ■
Proposition 8.2.3 If a complex M • of Banach abelian groups is K-bounded exact

at M n−1 and M n , then it is exact at M n .
Proof. Let s ∈ M n such that ds = 0. Then, there exists s′i ∈ M n−1 and s′′i ∈ M n−2
such that
1 1
∥s − ds′i ∥ ≤ and ∥s′i+1 − s′i − ds′′i ∥ ≤ K∥d(s′i+1 − s′i )∥ + .
2i+2 K 2i+1
We have
1
∥d(s′i − s′i+1 )∥ ≤ ∥s − ds′i+1 ∥ + ∥s − ds′i ∥ ≤ .
2i+1 K
Therefore, if we set t′i = s′i − j<i ds′′j , we see that

P
∥t′i+1 − t′i ∥ = ∥s′i+1 − s′i − ds′′i ∥

1
≤ K∥d(s′i+1 − s′i )∥ + .
2i+1
1 1
=≤ + .
2i+1 2i+1
1
≤ .
2i
It follows that t′i → t′ ∈ M n−1 . Now we have ds′i → s (by definition) and dt′i → dt′
bu continuity. Since ds′i = dt′i , it follows that dt′ = s. ■
Corollary 8.2.4 A K-acylcic complex M • of Banach abelian groups is acyclic. ■
8.2.2 On Stone spaces

Exercise 8.5 Show that if S is compact Hausdorff and M is a (semi-) normed
(resp. Banach) abelian group, then so is C(S, M ) and ∥f ∥ = supx∈S ∥f (x)∥. Show
that if S is a Stone space, then (the image of) a C(S, M disc ) is dense in C(S, M ).
a
The induced topology is not the compact-open topology.
Solution. The first assertion follows from exercise 8.5. Now, if f : S → M is a

continuous map and ϵ > 0, then the open covering −1
(B(s, ϵ− )) has a
S
S = s∈M f
finite disjoint clopen refinement S = i=1 Si . For each i = 1, . . . , r, there exists
`r
si ∈ M such that Si ⊂ f −1 (B(si , ϵ− )) and we set g(x) = si for x ∈ Si . Then,
∥f − g∥ < ϵ. ■
Lemma 8.2.5 If M is a semi-normed abelian group and f : S0 ↠ S is a continuous

surjective map of Stone spaces, then the augmented Čech complex
Cˇ• ((S0 → S)+ , M )
is 1-bounded acyclic.
Proof. We can write S0 = limk S0k with S0k finite and we shall denote by S−1k the
←−
image S0k in S. Then, we consider
lim Cˇ• ((S0k → S−1k )+ , M ) ≃ Cˇ• ((S0 → S)+ , M disc ) → Cˇ• ((S0 → S)+ , M ).
−→
The direct limit topology on Cˇ• ((S0 → S)+ , M disc ) coincides with the induced
topology (this is the sup-norm topology). Using exercise 8.5 and 8.2.2, we see that
it is sufficient to show that Cˇ• ((S0 → S)+ , M disc ) is 1-bounded acyclic (for this
topology). Now, thanks to lemma 8.4, we are reduced to the case S, S0 finite. In
this case, we know from proposition 7.3.3 that the complex Cˇ• ((S0 → S)+ , M ) is
homotopically trivial. Moreover, the homotopy has the form hn : f 7→ f ◦ kn for
some kn : Sn → Sn+1 and therefore ∥hn ∥ ≤ 1. Since Id = h ◦ d + d ◦ h, we finally
obtain that, if s ∈ M n , then ∥s − dhs∥ ≤ ∥ds∥. ■
Proposition 8.2.6 A Banach abelian group is acyclic on Stone spaces.
Proof. According to exercise 7.49, we have to show that if M is a Banach abelian

group and f : S0 ↠ S is a surjective map of Stone spaces, then the augmented Čech
complex
Cˇ• ((S0 → S)+ , M )
is acyclic. This follows from corollary 8.2.4 and lemma 8.2.5. ■
8.2.3 Banach spaces

Theorem 8.2.7 — Tietze. Let X is a normal topological space, K ⊂ X a compact
subset and V a real Banach space, then C(X, V ) ↠ C(K, V ) is surjective.
Proof. Sketch3 . When V is finite dimensional or V = ℓ∞ (R), this reduces to the case
V = R where this is a classical result in the spirit of Urysohn’s lemma. In general,
since a compact subset of a Banach space is separable, one may assume that V is
separable. Since all infinite dimensional separable Banach spaces are isomorphic, we
may assume that V = c0 (R) (space of null sequences). This is a Lipschitz retract to
ℓ∞ (R) where we know that the result holds. ■
Theorem 8.2.8 Real Banach spaces are acyclic on compact Hausdorff spaces.
Proof. Let V be a Banach space. Thanks to propositions 7.3.7 and 8.2.6, and
corollary 8.2.4, it is sufficient to show that, if S0 → S is a continuous surjection from
a Stone space, then the augmented Čech complex Cˇ• ((S0 → S)+ , V ) is 1-acyclic. We
fix some ϵ > 0. We denote by πn : Sn → S the canonical map and apply for all x ∈ S,
lemma 8.2.5 to π0−1 (x) → x. If f : Sn → V is a continuous map and fx denote its
restriction to πn−1 (x), then there exists gx : πn−1
−1
(x) → V such that (use exercise 8.6)
∥fx − dgx ∥ ≤ ∥dfx ∥ + ϵ/2 ≤ ∥df ∥ + ϵ/2.
By Tietze’s theorem, gx extends to gex on Sn−1 . By continuity, there exists a

neighborhood Vx of x such that
∥f − de
gx ∥πn−1 (Vx ) < ∥df ∥ + ϵ.
We can pick-up Vi := Vxi for i = 1, . . . , r that cover S and

P we write gi := gexi . Choose
a subordinated partition of unity {ϕi }ri=1 and set g = ri=1 (ϕi ◦ πn−1 )gi . Then,
r
X r
X r
X
f= (ϕi ◦ πn )f and dg = (ϕi ◦ πn−1 ◦ d)dgi = (ϕi ◦ πn )dgi
i=1 i=1 i=1
3
This is supposed to be a result of Dugundji but I have not been able to provide a reference.
It follows that
r
X
∥f − dg∥ = (ϕi ◦ πn )(f − dgi )
i=1
r
X
≤ ϕi ∥(f − dgi )∥πn−1 (Vi )
i=1
≤ ∥df ∥ + ϵ. ■
Exercise 8.6 Let π : Y → X be a closed continuous map and A ⊂ X. Then the

subsets of the form π −1 (U ) when U is a neighborhood of A in X form a basis of
neighborhoods of π −1 (A) in Y .
Solution. Let V be an open neighborhood of π −1 (A) and U := X \ π(Y \ V ). Then,

π −1 (A) ⊂ π −1 (U ) ⊂ V . ■
We used above the notion of subordinated partition of unity for an open cover
{Vi }i∈I of a topological space S: this
P is a family of continuous maps ϕi : S → [0, 1]
with support in Vi such that 1S = i∈I ϕi .
Exercise 8.7 Show that a topological space is paracompact Hausdorff (resp. nor-
mal) if and only if any open cover (resp. locally finite open cover) admits a
subordinated partition of unity.
8.3 Extensions of abelian groups

8.3.1 First computations
Lemma 8.3.1 If M, N are two condensed abelian groups and S is a Stonean space,
then
ExtnZ (M, N )(S) ≃ ExtnZ (M · S, N ).
Proof. Since S is Stonean, then Z · S is projective, which implies that the functor
HomZ (Z · S, −) is exact. We apply exercise 7.35 to the natural isomorphism
HomZ (M · S, N ) ≃ HomZ (Z · S, HomZ (M, N ))
and obtain
ExtnZ (M · S, N ) ≃ HomZ (Z · S, ExtnZ (M, N )) ≃ ExtnZ (M, N )(S). ■
Exercise 8.8 Show that, if X is a condensed set, N a condensed abelian group

and S is a Stonean space, then
ExtnZ (Z · X, N )(S) ≃ Hn (X × S, N ).
Exercise 8.9 Show that if S is a Stonean space and N a condensed abelian group,
then
RHomZ (Z · S, N ) ≃ N (S).
Solution. Follows from the fact that Z · S is projective. ■
Exercise 8.10 Show that if E is a discrete topological space and N a condensed

abelian group, then
RHomZ (Z · E, N ) ≃ N E .
Solution. Exercises 8.8 and 8.2 show that, if S is a Stonean space, then
ExtnZ (M, N )(S) ≃ Hn (E × S, N ) = 0
for n ̸= 0. ■
Exercise 8.11 Show that if M, N are two discrete abelian groups, then
RHomZ (M , N ) = RHomZ (M, N ) and M ⊗LZ N ≃ M ⊗LZ N
Solution. Since M is the constant sheaf associated to M , the functor M 7→ M is

exact on discrete abelian groups. Moreover, it follows from exercise 8.10 that constant
free abelian groups are acyclic for the functor HomZ (−, N ). We also know that free
abelian groups are flat. Our assertion therefore follows from proposition 6.2.2 (resp.
exercise 7.52) and corollary 7.35. ■
Exercise 8.12 Show that if S is a compact Hausdorff space and N a real Banach
space, then
RHomZ (Z · S, N ) ≃ C(S, N ).
Solution. It follows from exercise 8.8 and theorem 8.2.8 that, if T is Stonean, we
have
ExtnZ (Z · S, N )(T ) ≃ Hn (S × T , N ) = 0. ■
8.3.2 Breen-Deligne resolutions

The following was attributed to Deligne but no full proof was available before Scholze
gave one in the appendix to the fourth lecture on condensed mathematics:
Theorem 8.3.2 — Breen-Deligne. If M is an abelian group (of a topos), then there
exists a natural left resolution F (M )• → M with
rn
M
F (M )n = Z · M sn,i .
i=1
Proof. This is difficult and will not be proved here. ■

It is worth describing the lower part of the (augmented) complex:
· · · → Z · M2 ⊕ Z · M3 → Z · M2 → Z · M → M
We have
1. d0 : [s] 7→ s,
2. d1 :
[s1 , s2 ] 7→ −[s1 ] + [s1 + s2 ] − [s2 ],
[s1 , s2 ] 7→ [s1 , s2 ] − [s2 , s1 ]
3. d2 :
[s1 , s2 , s3 ] 7→ −[s2 , s3 ] + [s1 + s2 , s3 ] − [s1 , s2 + s3 ] + [s1 , s2 ].
Exercise 8.13 Show that this is indeed the lower terms of a resolution.
Exercise 8.14 Show that, if M, N are two abelian groups of a topos, then there
exists a natural spectral sequence
rn
M
E1p,q = Hq (M sp,i , N ) ⇒ Extp+q
Z (M, N )
i=1
Solution. Apply proposition 7.2.20 to Hom(−, N ) and the Breen-Deling resolution.

We have ExtqZ (Z · M sp,i , N ) = Hq (M sp,i , N ) ■
We shall need later the following (whose proof is in the same sprit as the proof of
Breen-Deligne theorem):
Lemma 8.3.3 If M is an abelian group (in a topos), then multiplication by p ∈ Z

on the Breen-Deligne resolution F (M )• and the map [p] induced by multiplication
by p on M are naturally homotopic.
Proof. Assume first that F (M )• is a projective resolution of M . Since p = [p] on M ,

then [p] = p in the derived category on F (M )• and it follows from proposition 7.2.5
(dual version) that [p] ∼ p on F (M )• . Consider now the topos T = Ab [ op and the
abelian group M := h of T . If s ∈ N then M = h . If N is a presheaf of abelian

Z s Zs
groups on Abop , then

s
HomZ (Z · M s , N ) = Hom(M s , N ) = Hom(hZ , N ) = N (Zs ).
Since limits and colimits are computed argument by argument on presheaves, Z·M s is
a projective abelian group of T . It follows that the assertion is true in this case. Now,
if M is any (usual) abelian group, we can specialize to M = hZ (M ). This extends to
any category of presheaves and then to any category of sheaves by sheafification. ■
Proposition 8.3.4 If M, N are two condensed abelian groups and S is a Stonean

space, then there exists a natural spectral sequence
rn
M
E1p,q = Hq (M sp,i × S, N ) ⇒ Extp+q
Z (M, N )(S)
i=1
Proof. Let F (M )• be the Breen-Deligne resolution of M . Proposition 7.2.20 applied

to HomZ (−, N ) and and F (M )• · S provides a spectral sequence
E1p,q = ExtqZ (F (M )p · S, N ) ⇒ Extp+q

Z (F (M )• · S, N ).
On the one hand, we have

rn
M rn
M
sp,i
F (M )p · S = Z·M ⊗Z Z · S = Z · (M sp,i × S)
i=1 i=1
and
ExtqZ (Z · (M sp,i × S), N ) = Hq (M sp,i × S, N ).
On the other hand, since Z · S is flat, the morphism
F (M )• · S = F (M )• ⊗Z Z · S → M ⊗Z Z · S ≃ M · S
is a quasi-isomorphism and it therefore follows from lemma 8.3.1 that
ExtnZ (F (M )• · S, N ) ≃ ExtnZ (M · S, N ) ≃ ExtnZ (M, N )(S). ■
8.3.3 Applications
Proposition 8.3.5 If M is a finite dimensional real Banach space and N a discrete
abelian group, then
RHomZ (M , N ) = 0.
Proof. Recall first from corollary 8.3.4 that, if S is a Stonean space, then there exists
a natural spectral sequence
rn
M
Z (M, N )(S)
i=1
and we want to show that the abutment is zero. Now, let q, s ∈ N. Thanks to
corollary 7.4.3 and proposition 8.1.4, since M s is contractible and M s × S is locally
compact Hausdorff, we have a natural isomorphism
Hn (M s × S, N ) ≃ Hn (S, N ).
We may therefore assume that M = 0 and we are done. ■

More generally, if V is an R-module4 and N a discrete abelian group, then
RHomZ (V, N ) = RHomR (V, RHomZ (R, N )) = RHomR (V, 0) = 0.
4
We haven’t discussed this matter.
Proposition 8.3.6 If M is a compact Hausdorff abelian group and N a real Banach

space, then
RHomZ (M , N ) = 0.
Proof. We consider again the spectral sequence

rn
M
Z (M, N )(S).
i=1
It follows from theorem 8.2.8 that E1p,q = 0 for q =

̸ 0 and we shall show that E2p,0 = 0,
or equivalently, that the complex of Banach spaces K • with
rn
M
K n := C(M sn,i × S, N )
i=1
is acyclic. Now, we proved lemma 8.3.3 that the maps 2 and [2] induced by multipli-
cation by 2 on N and M respectively are homotopic on the Breen-Deligne resolution:
2 − [2] = dh + hd. Let f ∈ K n such that df = 0. Then, 2f − [2]∗ f = dh∗n−1 f and
therefore f = 12 [2]∗ f + d( 12 h∗n−1 f ). By induction, we get
n
!
1 n∗ X 1 ∗
f = n [2 ] f + d h ([2k−1 ]∗ f ) .
k n−1
2 k=1
2
Since ∥[2]∗ ∥ ≤ 1 and ∥h∗n−1 ([2k−1 ]∗ f )∥ ≤ ∥h∗n−1 ∥∥f ∥ (so that the series below con-
verges), we finally obtain
∞
!
X 1 ∗
f =d h ([2k−1 ]∗ f ) .
k n−1
■
k=1
2
Exercise 8.15 Show that, if M and N are finite dimensional Banach spaces, then
RHomZ (M , N ) = HomR (M, N )
Solution. We may assume that M = N = R. Since RHomZ (T, R) = 0, we have
RHomZ (R, R) = RHomZ (Z, R) ≃ R. ■
We shall need below the following elementary fact:

Exercise 8.16 In any topos, there exists a natural map
M ⊗LZ RHomZ (P, N ) → RHomZ (RHomZ (M, P ), N ).
Solution. By adjunction, the identity
HomZ (M, P ) = HomZ (M, P ) (resp. HomZ (P, N ) = HomZ (P, N ))

provides a map
M ⊗Z HomZ (M, P ) → P (resp. P ⊗Z HomZ (P, N ) → N ).
From this, we deduce
M ⊗Z HomZ (M, P ) ⊗Z HomZ (P, N ) → P ⊗Z HomZ (P, N ) → N,
and by adjunction again
M ⊗Z HomZ (P, N ) → HomZ (HomZ (M, P ), N )
We can then derive. ■
Recall that we denote by T the circle (the one dimensional torus).
Proposition 8.3.7 If M, N are two discrete abelian groups, then
RHomZ (RHomZ (M , T), N ) ≃ M ⊗LZ N [−1].
Proof. We first define the map

M ⊗LZ N [−1] ≃ M ⊗LZ RHomZ (Z[−1], N )
→ M ⊗LZ RHomZ (T, N )
→ RHomZ (RHomZ (M , T), N ).
The first isomorphism comes from exercise 8.11, the second from the exact sequence
0→Z→R→T→0
(use exercise 6.5) and the last one from exercise 8.16. In order to prove that this
is an isomorphism, we may assume that M = Z · E is free. Then, it follows from
exercises 8.10 and 4.11 that
RHomZ (M , T) = Hom(E, T) = TE .
On the other hand, we have M ⊗LZ N ≃ (N · E)[−1] and we are therefore reduced to
showing that
RHomZ (TE , N ) ≃ N · E[−1].
The distinguished triangle
+
RHomZ (Z, N ) → RHomZ (R, N ) → RHomZ (T, N ) →
provides thanks to proposition 8.3.5 an isomorphism
RHomZ (T, N ) ≃ RHomZ (Z, N )[−1] ≃ N [−1].
The case where E is finite follows and we write now E = lim E ′ when E ′ runs through
−→
the finite subsets. Thanks to the natural spectral sequence of corollary 8.3.4 (and
the fact that filtered coimits are exact), we are reduced to show that for all s ∈ N,
Hq ((Ts )E × S, N ) = lim Hq ((Ts )J × S, N )
−→′
E
This follows from propositions 8.1.4 and 7.3.12. ■

Exercise 8.17 Show that RHomZ (M , N ) is given by
M \N Z R T
Z Z R T
R 0 R R
T Z[−1] 0 Z
8.3.4 Locally compact abelian groups

Proposition 8.3.8 If M is a locally compact Hausdorff abelian group, then
M ∗ ≃ RHomZ (M , T).
We also have
M ∗ ≃ RHomZ (M , Z)[1] (resp. M ∗ ≃ RHomZ (M , R))
if M is compact (resp. Banach).
Proof. Since M is an extension of a discrete abelian group by the sum of a finite

dimensional Banach space and a (connected) compact Hausdorff abelian group, it
is sufficient to consider the case where it is of one of these types: discrete abelian
group, finite dimensional Banach space or compact Hausdorff abelian group. Assume
first M is discrete. Then, there exists an exact sequence 0 → F ′ → F → M → 0
with F, F ′ free. The corresponding long exact sequence reads (use exercise 8.10)
0 → M ∗ → F ∗ → F ′∗ → Ext1Z (M , T) → 0
which implies that Ext1Z (M , T) = 0. In the case M is compact, we can first apply
the previous result to M ∗ so that
RHomZ (M ∗ , T) ≃ M .
Then propositions 8.3.7 (applied to M ∗ and Z) and 8.3.6 (applied to M and R)

provide
RHomZ (M , Z) ≃ M ∗ [−1] and RHomZ (M , R) = 0
which allows us to conclude. Finally, if M is a finite dimensional Banach space, we

have know from proposition 8.3.5 and exercise 8.15 that
RHomZ (M , Z) = 0 and RHomZ (M , R) = M ∗ . ■
Exercise 8.18 Show that if M is a connected locally compact Hausdorff abelian

group and N a discrete abelian group, then
∀n ̸= 1, ExtnZ (M , N ) = 0.
Solution. Any connected locally compact Haudorff abelian group is the direct sum of
a connected compact Haudorff abelian group and a finite dimensional Banach space.
Using proposition 8.3.5, we may therefore assume that M is compact. Since M is

connected, its Pontryagin dual M ∗ is then torsion free and proposition 8.3.7 provides
RHomZ (M , N ) ≃ M ∗ ⊗Z N [−1]. ■
Theorem 8.3.9 If M, N are two locally compact Hausdorff abelian groups, then
∀n ̸= 0, 1, ExtnZ (M , N ) = 0.
Proof. Since both M and N are an extension of a discrete abelian group by the sum
of a finite dimensional Banach space and a connected compact Hausdorff abelian
group, it is sufficient to consider the case where they are of one of these types: discrete
abelian group, finite dimensional Banach space or connected compact Hausdorff
abelian group. If M is discrete, there exists a two terms free resolution and the result
follows from exercise 8.10. If N is discrete, then this follows from exercise 8.18. The
case N is Banach is taken care of by propositions 8.3.6 and 8.15. So, we may assume
now that N is connected compact Hausdorff. There exists a two terms resolution of
′
the form 0 → N → TE → TE → 0. It is therefore sufficient to show that
∀n ̸= 0, ExtnZ (M , TE ) = ExtnZ (M , T)E = 0.
But this follows from proposition 8.3.8. ■
It is also possible to treat non locally compact groups:
RHomZ (RI , Z) = 0 and RHomZ (ZI , Z) = Z · I.
Solution. The first assertion follows from the fact that RI is an R-module but we
shall give a direct proof. Considering the first assertion, one can use Breen-Deligne
resolution for some Stonean space S and it is sufficient to show that
Hp ((RI )s × S, Z) = Hp (S, Z)
when s ∈ N. We may clearly assume s = 1. One can write RI = lim i∈I [−ni , ni ]
Q
−→
(for the compact-open topology) and therefore
!
Y
RΓ(RI × S, Z) = R lim RΓ [−ni , ni ] × S, Z
←−
i∈I
Now, we can use sheaf cohomology since i∈I [−ni , ni ]×S is compact. But i∈I [−ni , ni ]
Q Q
is acyclic and therefore
!
Y
R lim RΓ [−ni , ni ] × S, Z = R lim RΓ (S, Z) = RΓ (S, Z) .
←− ←−
i∈I
The second equality is then obtained from

RHom(TI , Z) = Z · I[−1]. ■
Note that
Hom(RI , Z) = kCZ (kRI , Z) and Hom(ZI , Z) = kCZ (kZI , Z)
so that CZ (kRI , Z) = 0 and kCZ (kZI , Z) = Z · I.
Bibliography
[AHS90] Adamek, Horst Herrlich, and George E. Strecker. Abstract and concrete
categories. The joy of cats. English. New York etc.: John Wiley & Sons,
Inc., 1990.
[Apa21] Sofia Marlasca Aparicio. Condensed Mathematics: The internal Hom of
condensed sets and condensed abelian groups and a prismatic construction
of the real numbers. 2021. arXiv: 2109.07816 [math.GN].
[AGV71] Michael Artin, Alexander Grothendieck, and Jean-Louis Verdier. Theorie
de Topos et Cohomologie Etale des Schemas I, II, III. Volume 269, 270,
305. Springer, 1971.
[BH19] Clark Barwick and Peter Haine. Pyknotic objects, I. Basic notions. 2019.
arXiv: 1904.09966 [math.AG].
[Bro06] Ronald Brown. Topology and groupoids. English. 3rd revised, updated
and extended ed. Bangor: Ronald Brown, 2006 (cited on page 50).
[Hai22] Peter J. Haine. Descent for sheaves on compact Hausdorff spaces. 2022.
arXiv: 2210.00186 [math.AT].
[Joh79] P. T. Johnstone. “On a topological topos”. English. In: Proc. Lond. Math.
Soc. (3) 38 (1979), pages 237–271 (cited on page 8).
[Joh82] Peter T. Johnstone. Stone spaces. English. Volume 3. Cambridge Univer-
sity Press, Cambridge, 1982.
[Joh02a] Peter T. Johnstone. Sketches of an elephant. A topos theory compendium.
I. English. Volume 43. Oxford: Clarendon Press, 2002.
[Joh02b] Peter T. Johnstone. Sketches of an elephant. A topos theory compendium.
II. English. Volume 44. Oxford: Clarendon Press, 2002.
152 BIBLIOGRAPHY
[KS90] Masaki Kashiwara and Pierre Schapira. Sheaves on manifolds. Volume 292.
With a chapter in French by Christian Houzel. Berlin: Springer-Verlag,
1990, pages x+512.
[KS06] Masaki Kashiwara and Pierre Schapira. Categories and sheaves. Vol-
ume 332. Springer-Verlag, Berlin, 2006, pages x+497 (cited on page 11).
[Mai21] Catrin Mair. Animated Condensed Sets and Their Homotopy Groups.
2021. arXiv: 2105.07888 [math.AT].
[Mit65] Barry Mitchell. Theory of categories. English. Volume 17. New York and
London: Academic Press, 1965.
[Mun00] James R. Munkres. Topology. English. 2nd ed. Upper Saddle River, NJ:
Prentice Hall, 2000.
[Pop73] N. Popescu. Abelian categories with applications to rings and modules.
English. Volume 3. Academic Press, London, 1973.
[RW77] Fred Richman and Elbert A. Walker. “Ext in pre-Abelian categories”.
English. In: Pac. J. Math. 71 (1977), pages 521–535.
[Rie16] Emily Riehl. Category theory in context. English. Mineola, NY: Dover
Publications, 2016.
[Shu08] Michael A. Shulman. Set theory for category theory. 2008. arXiv: 0810.
1279 [math.CT] (cited on page 11).
[Sta19] The Stacks project authors. The Stacks project. https://stacks.math.
columbia.edu. 2019 (cited on page 119).
[Yam22] Koji Yamazaki. Condensed Sets on Compact Hausdorff Spaces. 2022.
arXiv: 2211.13855 [math.CT].

Condensed Book

Uploaded by

Copyright:

Available Formats

Condensed Book

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Condensed Book

Uploaded by

Copyright:

Available Formats

An introduction to

Bernard Le Stum (April 3, 2024)

Réalisé en LATEX to partir du modèle Legrand Orange Book

1 Categories and functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3 Sites and topos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

6 Condensed abelian groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

7 Cohomology (optional) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

7.3 Sheaf cohomology 127

8 Condensed cohomology (optional) . . . . . . . . . . . . . . . . . . . 137

The right environment in order to perform linear algebra is that of an

Hom(X, Y ) × Hom(Y, Z) → Hom(X, Z), (f, g) 7→ g ◦ f

Examples 1. If G is a monoid, one can consider the category G with a unique

δni : [n − 1] → [n] (resp. σni : [n + 1] → [n])

The product C × C ′ of two categories C and C ′ is itself a category (everything is

Exercise 1.4 Show that, if G is a monoid, then Mor(G) = G and that, if X is a

2. an isomorphism is a morphism that has at the same time a section and a

Examples 1. A retract A of a topological space X is a subspace such that the

Proposition 1.1.4 If f is an isomorphism, then it has a unique section and a unique

Proof. If f ◦ g = idY and h ◦ f = idX , then

The unique section/retraction of an isomorphism f is called its inverse and

∀X, Y ∈ C ′ , HomC ′ (X, Y ) = HomC (X, Y ).

It is a wide subcategory if any object of C is in C ′ .

A subcategory becomes a category with the induced composition. A full subcate-

Examples 1. Ab is a full subcategory of Gr which itself is a full subcategory of

We will denote by Hom(C, C ′ ) the set of all functors C → C ′ . We will often

Examples 1. There exists a functor that forgets topology (and continuity)

3. There exists an inclusion functor Gr ,→ Mon and two functors G 7→ G× and

G 7→ Ggr := ⟨{xg }g∈G / {x−1

in the other direction.

Topop → Cat, X 7→ Ouv(X), u 7→ u−1 .

Exercise 1.7 What is a functor G → H between categories associated to monoïds

Solution. For example, the free monoid generated by a set X is G := {g : n →

If we are given two categories C and C ′ , then the projections C × C ′ on C and

C ,→ C × C ′ associated to a fixed object X ∈ C or X ′ ∈ C ′ .

Hom : C op × C → Set, (X, Y ) 7→ Hom(X, Y )

1.2.2 Natural transformation

We shall say natural isomorphism if all αX are isomorphisms.

Examples 1. We obtain a natural transformation by considering detA : GLn (A) →

Definition 1.2.3 A functor F : C → C ′ is

Hom(X, Y ) → Hom(F (X), F (Y )), f 7→ F (f )

is injective (resp. surjective, resp. bijective).

Theorem 1.2.4 A functor is an equivalence of categories if and only if it is fully

and we obtain by construction a natrual isomorphism α : IdC ′ ≃ F ◦ G. In particular,

Exercise 1.18 Show that, if k is a commutative ring, then Mat(k) := N, endowed

1.2.3 Representable functor

We shall also say that F is represented by X (and s).

Examples 1. If k is a commutative ring then k[t] represents the forgetful functor

Exercise 1.21 Let k be a commutative ring and f1 , . . . , fr ∈ k[t1 , . . . , tn ]. Show

that the functor that sends a commutative k-algebra A to the set

S(A) := {(a1 , . . . , an ) ∈ An / f1 (a1 , . . . , an ) = · · · = fr (a1 , . . . , an ) = 0}

of all solutions with values in A, is representable.

Exercise 1.22 Show that, if F is represented by both X and X ′ , then X ≃ X ′ .

Lemma 1.2.6 — Yoneda. If F : C → Set is any functor and X ∈ C, then there

Hom(hX , F ) ≃ F (X), α 7→ αX (IdX ).

Proof. Given s ∈ F (X), if Y ∈ C and f : X → Y , then we set αY (f ) := F (f )(s).

(αZ ◦ hX (g))(f ) = αZ (hX (g)(f )) = αZ (g ◦ f ) = F (g ◦ f )(s)

It only remains to check that we did define an inverse (exercise). ■

Proposition 1.2.7 A functor F : C → Set is represented by X ∈ C if and only if

is fully faithful. Deduce that C op (resp. C) is equivalent to the full subcategory