Universitext

Ana Agore

A First
Course in
Category
Theory
Universitext

Series Editors
Nathanaël Berestycki, Universität Wien, Vienna, Austria
Carles Casacuberta, Universitat de Barcelona, Barcelona, Spain
John Greenlees, University of Warwick, Coventry, UK
Angus MacIntyre, Queen Mary University of London, London, UK
Claude Sabbah, École Polytechnique, CNRS, Université Paris-Saclay, Palaiseau,
France
Endre Süli, University of Oxford, Oxford, UK
Universitext is a series of textbooks that presents material from a wide variety
of mathematical disciplines at master’s level and beyond. The books, often well
class-tested by their author, may have an informal, personal, or even experimental
approach to their subject matter. Some of the most successful and established books
in the series have evolved through several editions, always following the evolution
of teaching curricula, into very polished texts.
Thus as research topics trickle down into graduate-level teaching, first textbooks
written for new, cutting-edge courses may find their way into Universitext.
Ana Agore

A First Course in Category

Theory
Ana Agore
Institute of Mathematics
Romanian Academy
Bucharest, Romania
Vrije Universiteit Brussel
Brussels, Belgium

ISSN 0172-5939 ISSN 2191-6675 (electronic)

Universitext
ISBN 978-3-031-42898-2 ISBN 978-3-031-42899-9 (eBook)
https://doi.org/10.1007/978-3-031-42899-9

Mathematics Subject Classification: 18-01

This work was supported by Fonds Wetenschappelijk Onderzoek, Belgium and Romanian Ministry of
Education and Research, CNCS/CCCDI-UEFISCDI.

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland
AG 2023
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Paper in this product is recyclable.

To my parents
Preface

Categories were first considered in 1945 in a paper by S. Eilenberg and S. Mac Lane
[21] with the purpose of formalizing the concept of “natural transformation”, which
was informally used at that time in many papers from various fields, especially in
algebraic topology. The initial theory introduced in [21] developed rapidly, allowing
for several new mathematical disciplines to arise, as was the case, for example, with
homological algebra. Category theory is based on the idea that many mathematical
properties can be described using diagrams of arrows of different types. Working
in this very general setting allows for a better understanding of the common
constructions and patterns in mathematics and leads to a unified treatment of similar
concepts across different mathematical structures. An early and notable example can
be found in [19], where group cohomology, Lie algebra cohomology and associative
algebra cohomology are recast as derived functors in a suitable module category.
Over the years, category theory has become a universal language allowing
mathematicians to achieve important advancements by exchanging ideas and tech-
niques between seemingly unrelated domains. Using very abstract definitions
that capture the idea behind a certain concept in universal terms rather than its
isolated properties, purely categorical techniques have found their way into most
mathematical areas.
Nowadays category theory is an indispensable tool for doing research not only in
various areas of pure mathematics such as algebraic topology, homological algebra,
algebraic geometry and functional analysis, but also in theoretical computer science
(e.g., the development of algorithms, automata theory), physics (e.g., electrical
circuits), chemistry (e.g., chemical interactions), biology (e.g., biological systems)
and medicine (e.g., genetics). We refer to [52] for an approach to category theory
in the spirit of the applied sciences and to [29] for applications to the cognitive
sciences.
The purpose of this book is to provide students with no prior exposure to
categorical reasoning with an accessible source from which to learn the basic
material. The fundamentals of category theory are clearly and thoroughly covered
with the aim of leaving the reader able to confidently use categorical techniques as
well as to easily explore and understand more advanced topics.

vii
viii Preface

The book is based on my lecture notes from the graduate course on category
theory that I have taught at Vrije Universiteit Brussel. Additional fully worked
examples and complete proofs have been added with the purpose of making the
material suitable for self-study. Although the reader is expected to be at the advanced
undergraduate level, some background and full references are provided throughout
the book. The prerequisites include familiarity with group theory, rings, modules
and topological spaces, as well as a basic understanding of set theory. As opposed
to the standard category theory monographs by S. Mac Lane [35] and F. Borceux
[8–10], and the more recent ones [5, 34, 47, 48], which are more encyclopedic
in nature and oriented toward researchers rather than students, the present book
serves as a first introduction to the field. The excellent monographs [1, 2, 5, 8–
10, 12, 13, 22, 23, 25, 30, 34, 35, 46–48] have been used when preparing these notes
and have influenced the approach and the development of certain topics.
The first chapter introduces the fundamental concepts needed in the sequel.
Important notions such as (sub)categories, functors, natural transformations, repre-
sentable functors, which form the backbone of category theory, are well illustrated
by many familiar examples. A concise description of the duality principle, a crucial
reasoning process in category theory, is also presented. The first important result we
present is Yoneda’s lemma, which allows us to embed any (locally small) category
into a category of functors on that category. This generalizes the well-known group
theory result called Cayley’s theorem, stating that any group is isomorphic to a
subgroup of a symmetric group.
The second chapter treats the general theory of limits and colimits. Both are
very general concepts which arise in various forms in all fields of mathematics.
We introduce them gradually, starting with some special cases which might be
familiar to the reader such as: (co)products, (co)equalizers, pullbacks and pushouts.
A variety of detailed examples are included to illustrate the newly introduced
concepts. (Co)products and (co)equalizers are not only important special cases of
(co)limits but also generic in the sense that all (co)limits can be constructed out of
these two special cases. Certain types of functors are considered in connection to the
existence of (co)limits. The existence of (co)limits in several important categories
such as functor categories or comma categories is investigated in detail as well.
The third chapter deals with one of the most important notions in category
theory: adjoint functors. Several descriptions of adjoint functors are presented
and the theory is illustrated by a wide range of examples from various areas of
mathematics. Many important constructions in mathematics are shown to be part
of an adjunction, including for instance the classical free constructions present in
algebra, localizations in ring theory or Stone–Čech compactifications of topological
spaces. Important related concepts such as equivalence of categories, (co)reflective
subcategories or localization of categories are also investigated and well illustrated
by a plethora of detailed examples. Deeper connections with the concepts introduced
in the previous chapters are emphasized. For instance, (co)limits and representable
functors are equivalently described by means of adjoint functors. Going beyond
what is usually covered by an introductory text in category theory, the book ends
with a more advanced topic, the adjoint functor theorem. More precisely, two
Preface ix

variations of this celebrated theorem, namely Freyd’s Adjoint Functor Theorem and
the Special Adjoint Functor Theorem, are considered. They provide different kinds
of necessary and sufficient conditions for a functor to admit a left or a right adjoint.
I would like to take this opportunity to thank my coauthors as well as my
colleagues and students from Vrije Universiteit Brussel for everything I have learned
from them. My warmest thanks to Gigel Militaru for teaching me category theory
when I was a student and for the many wise suggestions he made after reading
a first draft of this book as well as to Alexandru Chirvasitu for the countless
illuminating discussions. I am very grateful to the Springer editors, especially
to Rémi Lodh, and to the anonymous referees for their comments and advice
which greatly improved the book’s presentation. During the preparation of this
manuscript, my work was supported at different stages by FWO (Fonds voor
Wetenschappelijk Onderzoek—Flanders) and a grant of Ministry of Research,
Innovation and Digitization, CNCS/CCCDI–UEFISCDI, project number PN-III-P4-
ID-PCE-2020-0458.

Bucharest, Romania Ana Agore

Brussels, Belgium
June 2023
Contents

1 Categories and Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Set Theory. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Categories: Definition and First Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Special Objects and Morphisms in a Category . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Some Constructions of Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.5 Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.6 Isomorphisms of Categories. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
1.7 Natural Transformations: Representable Functors . . . . . . . . . . . . . . . . . . . 41
1.8 The Duality Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
1.9 Functor Categories. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
1.10 Yoneda’s Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
1.11 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2 Limits and Colimits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
2.1 (Co)products, (Co)equalizers, Pullbacks and Pushouts . . . . . . . . . . . . . . 83
2.2 (Co)limit of a Functor. (Co)complete Categories . . . . . . . . . . . . . . . . . . . . 105
2.3 (Co)limit as a Representing Pair . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
2.4 (Co)limits by (Co)equalizers and (Co)products . . . . . . . . . . . . . . . . . . . . . . 121
2.5 (Co)limit Preserving Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
2.6 (Co)limits in Comma Categories. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
2.7 (Co)limits in Functor Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
2.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
3 Adjoint Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
3.1 Definition and Generic Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
3.2 Adjoints Via Free Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
3.3 Galois Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
3.4 More Examples and Properties of Adjoint Functors . . . . . . . . . . . . . . . . . 159
3.5 The Unit and Counit of an Adjunction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
3.6 Another Characterisation of Adjoint Functors . . . . . . . . . . . . . . . . . . . . . . . 183
3.7 (Co)reflective Subcategories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
3.8 Equivalence of Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199

xi
xii Contents

3.9 Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214

3.10 (Co)limits as Adjoint Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
3.11 Freyd’s Adjoint Functor Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
3.12 Special Adjoint Functor Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
3.13 Representable Functors Revisited. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
3.14 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
4 Solutions to Selected Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
4.1 Chapter 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
4.2 Chapter 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
4.3 Chapter 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
Frequently Used Notations

.∅ empty set
.N set of natural numbers
.Z ring of integers
.Q field of rational numbers
.R field of real numbers
.A\B set-theoretic difference
.ℵ0 cardinality of .N (and .Z)
.|X| cardinality of the set X
.⊆ inclusion
.f|X restriction of a map f to a subset X of its domain
.⊗ tensor product
.H  G H is a normal subgroup of G
.Gab abelianization of the group G
.G(M) universal enveloping group of the monoid M
.kerf kernel of the morphism f
.Imf image of the morphism f
.S
−1 X localization of the ring (module) X by the multiplicative set S
.CX discrete category on the set X
PO(.X, ) category associated to the pre-ordered set .(X, )
Set category of sets
FinSet category of finite sets
Grp category of groups
SiGrp category of simple groups
Ab category of abelian groups
Div category of divisible groups
Rng category of rings
Ring category of unitary rings
c
.Ring category of commutative unitary rings
Field category of fields
R.M category of left R-modules
.MR category of right R-modules

xiii
xiv Frequently Used Notations

Top category of topological spaces

Haus category of Hausdorff topological spaces
KHaus category of compact Hausdorff topological spaces
PreOrd category of pre-ordered sets and order preserving maps
Poset category of partially ordered sets and order preserving maps
.HomC (A, B) hom set in the category .C between A and B
.HomC hom bifunctor
.HomC (C, −) covariant hom functor
.HomC (−, C) contravariant hom functor
Cat category of small categories
.Fun(I, C) category of functors from the small category I to the category .C
.(F ↓ G) comma category of the functors .F : A → C, .G : B → C
.Cone(F, C) set of cones on the functor F with vertex C
.Cocone(F, C) set of cocones on the functor F with vertex C
.Cone(F, −) cone functor
.Cocone(F, −) cocone functor
.lim F limit of the functor F
.lim limit functor
.colim F colimit of the functor F
.colim colimit functor
.F ∼=G naturally isomorphic functors
.Nat(F, G) class of all natural transformations between the functors F and
G
α∗β
. Godement product of the natural transformations .α and .β
F G
. F is left adjoint to G (or G is right adjoint to F )
Chapter 1
Categories and Functors

1.1 Set Theory

We start by setting very briefly the set theory model that will be assumed to hold
throughout. The main issue that arises is that most categories of interest have as
objects all sets, all groups, all topological spaces, etc. Therefore, a proper definition
of a category which includes the examples mentioned above is not possible in the
classical Zermelo–Fraenkel set theory. One way to get around this issue is by using
the von Neumann–Bernays–Gödel (NBG) set theory which introduces, in addition
to sets, the notion of a class to play the role of these “big sets” consisting of all sets,
all groups, etc. More precisely, the connection between sets and classes is given by
the so-called limitation of size axiom:
A class is a set if and only if it is not bijective with the class of all sets.

To conclude, we can use the word class to designate any collection of mathematical
objects; all sets are obviously classes.
The NBG axioms are in fact a conservative extension of the ZFC axioms.
Therefore, all statements about sets which can be proved in NBG hold in ZFC as
well. An important axiom included in NBG which will be used in many places in
the sequel is the following form of the axiom of choice:
We can choose an element from each of any class of nonempty sets.

Moreover, the following consequence of the NBG axioms will be intensively used
throughout: if A is a set and B is a subclass of A then B is a set. A detailed account
of the NBG set theory axioms can be found, for instance, in [51] or in the Appendix
of [46].

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 1

A. Agore, A First Course in Category Theory, Universitext,
https://doi.org/10.1007/978-3-031-42899-9_1
2 1 Categories and Functors

1.2 Categories: Definition and First Examples

We start by providing the definition of a category and many examples which may
already be familiar to the reader.
Definition 1.2.1 A category .C consists of the following data:
(1) a class Ob .C whose elements A, B, .C, . . . are called objects;
(2) for every pair of objects A, B, a (possibly empty) set .HomC (A, B), whose
elements are called morphisms from A to B. An element .f ∈ HomC (A, B) will
be denoted by .f : A → B; A and B are called the domain and the codomain of
f , respectively;
(3) for every triple of objects A, B, C, a composition law

HomC (A, B) × HomC (B, C) → HomC (A, C)

.

. (f, g) → g ◦ f ;

(4) for every object A, a morphism .1A ∈ HomC (A, A), called the identity on A
such that the following axioms hold:
(i) associative law: given morphisms .f ∈ HomC (A, B), .g ∈ HomC (B, C),
.h ∈ HomC (C, D) the following equality holds:

h ◦ (g ◦ f ) = (h ◦ g) ◦ f ;
.

(ii) identity law: given a morphism .f ∈ HomC (A, B) the following identities hold:

1B ◦ f = f = f ◦ 1A .
.

A category .C whose class of objects Ob .C is a set is called a small category.

Furthermore, a category will be called finite if it contains only finitely many
morphisms.
Note that, in certain references such as [35], the hom-sets are allowed to be
classes. In that framework, categories as in Definition 1.2.1 are called locally small.

Examples 1.2.2
(1) Any set X can be made into a small category, called the discrete category on
X and denoted by .CX , as follows:

Ob CX = X
.


∅ if x = y
HomCX (x, y) = , for every x, y ∈ X.
{1x } if x = y
1.2 Categories: Definition and First Examples 3

The only possible compositions of morphisms are .1x ◦ 1x = 1x for all .x ∈

X. Throughout, we denote by .n the discrete category on a set with .n ∈ N
elements. For .n = 0 we obtain the empty category with no objects and no
morphisms.
(2) More generally, any pre-ordered set1 .(X, ) defines a small category
PO(.X, ) as follows:

Ob PO(X, ) = X
.


∅ if x  y
HomPO(X, ) (x, y) = , for every x, y ∈ X,
{ux, y } if x  y

where .ux, y denotes the unique morphism from x to y. The composition of

morphisms is given by the rule .uy, z ◦ ux, y = ux, z , while the identity on any
.x ∈ X is .ux, x .

Moreover, following the same idea, we can further generalize this example
to the level of classes. For instance, consider the class of all sets pre-ordered
by inclusion; we obtain a category denoted by .Set(⊆) which has the class of
all sets as objects and for all sets A, B we have

∅ if A  B
HomSet(⊆) (A, B) =
. ,
{uA, B } if A ⊆ B

where .uA, B denotes the unique morphism from A to B. Composition of

morphisms is defined as in the case of pre-ordered sets.
(3) A monoid .(M, ·) can be seen as a small category .M with a single object
denoted by .∗ and the set of morphisms .HomM (∗, ∗) = M. The composition
of morphisms in .M is given by the multiplication of M and the identity on .∗
is just the unit .1M . In particular, using the same idea, any group can be made
into a category.
(4) The category Set of sets has the class of all sets as objects while .HomSet (A, B)
is the set of all functions from A to B. Composition is given by the usual
composition of functions and the identity on any set A is the identity map .1A .
Set is not a small category.
(5) FinSet is the category whose objects are finite sets, and .HomFinSet (A, B) is
just the set of all functions between the two finite sets A and B. FinSet is also
not a small category; this can be easily seen by noticing that .{X} is a singleton
and therefore a finite set for every set X. However, as opposed to Set, the
category FinSet satisfies the following property: by choosing2 exactly one
object from each isomorphism class of finite sets together with all functions

1A set X is called pre-ordered if is endowed with a binary relation . which is reflexive and
transitive.
2 Recall that the axiom of choice is assumed to hold.
4 1 Categories and Functors

between them we obtain a category, called a skeleton of FinSet, which is small.

Categories such as FinSet which admit a small skeleton are called essentially
small and will be treated in more detail in Sect. 3.8. Set is not an essentially
small category as we will see in Example 3.8.15, (2).
(6) Consider RelSet to be the category defined as follows:

Ob RelSet = Ob Set
.

HomRelSet (A, B) = P(A × B) = {f | f ⊆ A × B}, for every A, B ∈ Ob Set.

The composition of morphisms in RelSet is defined as follows: given .f ⊆

A × B and .g ⊆ B × C we consider

g ◦ f = {(a, c) ∈ A × C | ∃ b ∈ B such that (a, b) ∈ f and (b, c) ∈ g}.

.

Finally, the identity is defined as .1A = {(a, a) | a ∈ A}. RelSet is called the
category of relations.
(7) Grp is the category of groups, where Ob Grp is the class of all groups while
.HomGrp (A, B) is the set of all group homomorphisms from A to B. Similarly,

Mon denotes the category of monoids with monoid homomorphisms between

them.
(8) SiGrp denotes the category of simple3 groups with group homomorphisms
between them.
(9) Ab is the category of abelian groups with group homomorphisms between
them. Throughout, we use multiplicative notation for the group structure on
an arbitrary group and additive notation for abelian groups.
(10) Div is the category of divisible4 groups with group homomorphisms between
them.
(11) Rng is the category of rings with ring homomorphisms between them.
(12) Ring (resp. .Ringc ) is the category of (resp. commutative) unitary rings with
unit preserving ring homomorphisms between them.
(13) Field is the category of fields5 with field homomorphisms between them.
(14) For a ring with unity R, we denote by .R M the category of left R-modules
with morphisms between two R-modules given by R-linear functions. The
category of right R-modules .MR can be defined analogously. In particular,
if K is a field then .K M (resp. .K Mf d ) denotes the category of vector spaces
(resp. finite-dimensional vector spaces) over K.

3 A non-trivial group is called simple if its only normal subgroups are the trivial group and the

group itself.
4 An abelian group .(G, +) is called divisible if for every positive integer n and every .g ∈ G, there

exists an .h ∈ G such that .nh = g.

5 Throughout, a field means a commutative division ring.
1.2 Categories: Definition and First Examples 5

Furthermore, if R is commutative and .S ⊂ R is a multiplicative subset of

R,6 then .R MS−aut stands for the category of left R-modules on which S acts as
an automorphism, i.e., for all .s ∈ S and .M ∈ ObR MS−aut the multiplication
map .μs : M → M, .μs (m) = sm is invertible; the morphisms between two
such objects are given by R-linear functions.
(15) For a field K, we denote by .AlgK the category of unital and associative
K-algebras together with algebra homomorphisms between them. Similarly,
c
.Alg
K stands for the category of unital, associative and commutative K-
algebras.
(16) Top is the category of topological spaces where Ob Top is the class of all
topological spaces while .HomTop (A, B) is the set of continuous functions
between A and B. .Top∗ stands for the category of pointed topological spaces,
that is, the objects are pairs .(A, a0 ) where A is a topological space and .a0 ∈ A
while the morphisms between two such pairs .(A, a0 ) and .(B, b0 ) are just
continuous functions .f : A → B such that .f (a0 ) = b0 .
(17) Haus is the category of Hausdorff topological spaces where Ob Haus is the
class of all Hausdorff topological spaces while .HomHaus (A, B) is the set
of continuous functions between A and B. Similarly, KHaus denotes the
category of compact Hausdorff topological spaces.
(18) PreOrd is the category whose objects are pre-ordered sets and the morphisms
between two pre-ordered sets are order preserving maps.7 Similarly, Poset
is the category whose objects are partially ordered sets8 and the morphisms
between two partially ordered sets are order preserving maps.
(19) For a field K, we denote by .MatK the category whose object class is the set of
natural numbers .N. The morphisms in .MatK between two objects m, .n ∈ N
are all .n × m matrices with entries in K and the composition of morphisms is
given by matrix multiplication:

HomMatK (m, n) × HomMatK (n, p) → HomMatK (m, p)

.

(A, B) → BA.

The identity morphism on any .n ∈ N is given by the .n × n identity matrix,

where the .0 × 0 identity matrix is by definition the zero matrix. Furthermore,
by convention, if either m or n is zero, we have a unique .n × m matrix called
a null matrix. .

Remark 1.2.3 Notice that although we sometimes work with categories whose
objects are sets, morphisms in the sense of Definition 1.2.1 need not be functions.
This situation is best illustrated in Example 1.2.2, (6).

6S is called a multiplicative subset of the ring R if .1R ∈ S and for all s, .s ∈ S we have .ss ∈ S.
7 Given two pre-ordered sets .(X, X ) and .(Y, Y ), a map .f : X → Y is called order preserving
if .x X y implies .f (x) Y f (y).
8 A set X is called partially ordered if is endowed with a binary relation . which is reflexive,

antisymmetric and transitive.

6 1 Categories and Functors

Definition 1.2.4 Let .C, .C be two categories. We shall say that .C is a subcategory
of .C if the following conditions are satisfied:
(i) Ob .C ⊆ Ob .C, i.e., any object of .C is an object of .C;
(ii) .HomC (A, B) ⊆ HomC (A, B) for every A, .B ∈ Ob C ;
(iii) the composition of morphisms in .C is induced by the composition of
morphisms in .C;
(iv) the identity morphisms in .C are identity morphisms in .C.
Moreover, .C is said to be a full subcategory of .C if for every pair .(A, B) of objects
of .C we have

HomC (A, B) = HomC (A, B).

.

Examples 1.2.5 (1) The category FinSet is a full subcategory of Set.

(2) The category Ab is a full subcategory of Grp.
(3) The category .R MS−aut is a full subcategory of .R M.
(4) The category Haus is a full subcategory of Top.
(5) Ring is a subcategory of Rng but not a full subcategory as not all morphisms in
Rng between unitary rings are unit preserving.
(6) Set is a subcategory of RelSet but not a full subcategory since not every subset
of a cartesian product defines a function. .

1.3 Special Objects and Morphisms in a Category

The notions of monomorphism and epimorphism which we will introduce next

are generalizations to arbitrary categories of the familiar injective and surjective
functions from Set.
Definition 1.3.1 Let .C be a category and .f ∈ HomC (A, B).
(1) f is called a monomorphism if for any .g1 , .g2 ∈ HomC (C, A) such that .f ◦ g1 =
f ◦ g2 we have .g1 = g2 ;
(2) f is called an epimorphism if for any .h1 , .h2 ∈ HomC (B, C) such that .h1 ◦ f =
h2 ◦ f we have .h1 = h2 ;
(3) f is called an isomorphism if there exists an .f ∈ HomC (B, A) such that .f ◦
f = 1B and .f ◦ f = 1A . In this case we say that A and B are isomorphic
objects.
Although in Set monomorphisms (resp. epimorphisms) coincide with injective
(resp. surjective) functions, this is no longer true in an arbitrary category whose
objects and morphisms are sets and functions, respectively, as we will see in
Example 1.3.2, (4) and (5).
1.3 Special Objects and Morphisms in a Category 7

Examples 1.3.2
(1) In each of the categories Set, Grp, Ab, .R M monomorphisms coincide with
the injective homomorphisms, while in Top and KHaus monomorphisms
coincide with the injective continuous maps. We will only prove here that
monomorphisms in Set coincide with injective functions. Indeed, suppose
.f : A → B is an injective map and g, .h : C → A are such that .f ◦ h = f ◦ g.
   
Then, we have .f h(c) = f g(c) for any .c ∈ C and since f is injective we
get .h(c) = g(c) for any .c ∈ C, i.e., .g = h as desired.
Assume now that .f : A → B is a monomorphism and let a, .a ∈ A be such
that .f (a) = f (a ). We denote by .ia : {∗} → A, respectively .ia : {∗} → A,
the maps given by .ia (∗) = a, ia (∗) = a . This implies that .f ◦ ia = f ◦ ia
and since f is a monomorphism we obtain .ia = ia . Therefore .a = a and f is
indeed injective.
Note that the proof above can be carried over verbatim to the categories Top
and KHaus by simply considering on .{∗} the indiscrete topology.9
(2) Similarly, in each of the categories Set, Grp, Ab, .R M epimorphisms coincide
with the surjective homomorphisms, while in Top and KHaus epimorphisms
coincide with the surjective continuous functions. We prove here that epi-
morphisms in Set and Grp coincide with surjective functions and surjective
homomorphisms, respectively.
Consider first an epimorphism .f : A → B in Set. Define g, .h : B → {0, 1}
as follows:

1, if b ∈ f (A)
.g(b) = , h(b) = 1, for all b ∈ B.
0, if b ∈
/ f (A)

Then, for any .b ∈ B, we have .(h ◦ f )(b) = 1 = (g ◦ f )(b) and since f is an

epimorphism we obtain .g = h. This shows that the image of f is the entire B,
as desired.
Next we look at epimorphisms in Grp. Let .f : G → H be an epimorphism
in Grp and denote by .K = Im (f ) the image of f . Assume that .K = H . If
K is a normal subgroup of H we can form the quotient group .H /K. Consider
now two group homomorphisms .π , .u : H → H /K, where .π is the canonical
projection and u is the trivial homomorphism defined by .π(h) = hK, .u(h) = K
for all .h ∈ H . Obviously .u ◦ f = π ◦ f and since f is an epimorphism we
obtain .u = π . Therefore, .K = H which contradicts our assumption. Hence, K
cannot be a normal subgroup of H . In particular, note that the index of K in H
is at least 3, otherwise K would be a normal subgroup of H ([49, Proposition
2.62]). This allows us to choose three distinct right cosets K, Kx and Ky, for
some x, .y ∈ H . Let .S(H ) be the set of permutations on H and define .σ ∈ S(H )

9 The topology consisting only of the set itself and the empty set is called the indiscrete topology.
8 1 Categories and Functors

as follows:
⎧
⎨ kx, if h = ky
.σ (h) = ky, if h = kx , k ∈ K.
⎩
h, if h ∈
/ Kx ∪ Ky

Consider now the group homomorphisms .ψ, .ξ : H → S(H ) defined as follows

for all t, .h ∈ H :

ψ(t)(h) = th, ξ(t)(h) = σ −1 ◦ ψ(t) ◦ σ (h).

.

First, we show that .ψ ◦ f = ξ ◦ f . Indeed, for any .g ∈ G and .h ∈ H, we have
ψ(f (g))(h) = f (g)h and respectively .ξ(f (g))(h) = σ −1 ◦ ψ f (g) ◦ σ (h).
.

Keeping in mind that .f (g) ∈ K for all .g ∈ G, if .h = ky for some .k ∈ K,

we have

ξ(f (g))(ky) = σ −1 ◦ ψ(f (g))(kx) = σ −1 (f (g)kx) = f (g)ky = ψ(f (g))(ky).

.

Similarly, if .h = kx for some .k ∈ K, we obtain

ξ(f (g))(kx) = σ −1 ◦ ψ(f (g))(ky) = σ −1 (f (g)ky) = f (g)kx = ψ(f (g))(kx).

.

Finally, .h ∈
/ Kx ∪ Ky yields

ξ(f (g))(h) = σ −1 ◦ ψ(f (g))(h) = σ −1 (f (g)h) = f (g)h = ψ(f (g))(h).

.

To conclude, we have proved that .ψ ◦ f = ξ ◦ f and since f is an epimorphism

in Grp we obtain .ψ = ξ . However, this is not true as we have

ψ(y −1 )(x) = y −1 x,
.

ξ(y −1 )(x) = σ −1 ◦ ψ(y −1 ) ◦ σ (x) = σ −1 ◦ ψ(y −1 )(y) = σ −1 (1H ) = 1H .

Clearly .y −1 x = 1H as Kx and Ky are distinct cosets. Therefore .ψ = ξ and

we have reached a contradiction. We can now conclude that .K = H and f is
surjective.
Conversely, assume now that .f : A → B is a surjective map in Set and let
.t1 , .t2 : B → C be two maps such that .t1 ◦ f = t2 ◦ f . As f is surjective, for

any .b ∈ B there exists some .a ∈ A such that .f (a) = b and we obtain

   
t1 (b) = t1 f (a) = t2 f (a) = t2 (b),
.

i.e., .t1 = t2 , which shows that f is an epimorphism in Set. Note that the
argument above can be used verbatim for the categories Grp, Ab, .R M in order
to show that surjective morphisms are epimorphisms.
1.3 Special Objects and Morphisms in a Category 9

(3) In each of the categories Set, Grp, Ab, .R M, isomorphisms coincide with the
bijective homomorphisms. In Top, isomorphisms are exactly the homeomor-
phisms, i.e., continuous bijections whose inverses are also continuous.
(4) In the category Div of divisible groups, the quotient map .q : Q → Q/Z is
obviously not injective but it is a monomorphism. Indeed, let G be another
divisible group and f , .g : G → Q be two morphisms of groups such that .q◦f =
q ◦ g. Denoting .f − g by h we obtain .q ◦ h = 0. Now for any .x ∈ G we have
.q(h(x)) = 0 and thus .h(x) ∈ Z. Suppose there exists some .x0 ∈ G such

that .h(x0 ) = 0. We can assume without loss of generality that .h(x0 ) ∈ N\{0}.
Since we are working with divisible groups, we can find some .y0 ∈ G such that
.x0 = 2h(x0 )y0 . Applying h to the above equality we obtain

h(x0 ) = 2h(x0 )h(y0 ),

.

which is an obvious contradiction since .h(x) ∈ Z for all .x ∈ G and .h(x0 ) = 0.

Hence we get .h = 0, which implies that .f = g. This proves that q is indeed a
monomorphism in Div.
(5) In the category .Ringc of unitary commutative rings, the inclusion .i : Z → Q
is obviously not surjective but it is an epimorphism. Indeed, let R be another
commutative ring together with two ring morphisms f , .g : Q → R such that
.f ◦ i = g ◦ i. Consider now .z ∈ Z\{0}; then we have .1 = f (1) = f (z)f (1/z)

and therefore .f (1/z) = 1/f (z). Similarly we can prove that .g(1/z) = 1/g(z)
and since f and g coincide on .Z we get .f (1/z) = g(1/z). Now for any .z ∈ Z
we have

f (z /z) = f (z )f (1/z) = g(z )g(1/z) = g(z /z).

.

Therefore .f = g, which implies that i is an epimorphism in .Ringc .

In a similar manner one can show that if R is a commutative ring with unity
and .(S −1 R, j ) is its localization with respect to the multiplicative set .S ⊂ R
then .j : R → S −1 R defined by .j (r) = 1r , for all .r ∈ R, is also an epimorphism
in .Ringc . We refer to [3, Chapter 11] for more details on localization of rings.
(6) It can be easily seen that the inclusion .i : Z → Q is a monomorphism in
the category .Ringc of unitary commutative rings and also an epimorphism by
Example 1.3.2, (5). Therefore, it provides an example of a morphism which is
both a monomorphism and an epimorphism but not an isomorphism.
(7) Let .(T , τ ) be a topological space such that .τ is different from the discrete
topology.10 Consider now the set T endowed with the discrete topology
.P(T ). Then the identity .IdT : (T , P(T )) → (T , τ ) is obviously bijective and

a continuous map between the two topological spaces. Therefore, .IdT is a

morphism in Top but not an isomorphism as the inverse map .IdT : (T , τ ) →
(T , P(T )) is obviously not continuous.

10 The topology consisting of all subsets of T is called the discrete topology on T .

10 1 Categories and Functors

(8) A rather special situation occurs in KHaus, the subcategory of Top con-
sisting of all compact Hausdorff topological spaces. As opposed to the cat-
egory of topological spaces, in KHaus any bijective continuous map .f ∈
HomKHaus (K, H ) is automatically an isomorphism. Indeed, it will suffice to
show that the inverse map .f −1 : H → K is continuous too. To this end, we
need to show that images of closed sets of K under f are closed in H ([39,
Theorem 18.1]). Consider U to be a closed subset of K; as K is compact it
follows that U is compact as well ([39, Theorem 26.2]). Moreover, as the image
of a compact space under a continuous map is compact ([39, Theorem 26.5]) we
obtain .f (U ) compact. Now recall that compact subspaces of Hausdorff spaces
are closed ([39, Theorem 26.3]). Therefore .f (U ) is closed, as desired.
(9) In the category PO(.X, ) associated to a partially ordered set .(X, ), any
isomorphism is an identity morphism. Indeed suppose .f : x → y is an
isomorphism; this implies that .x  y. If .g : y → x is the inverse of f then
we also have .y  x. Due to the antisymmetry of . we obtain .x = y. Therefore
.f : x → x must be the identity on x. .

It can be easily seen that an isomorphism is in particular a monomorphism and

an epimorphism. However, the converse is not necessarily true: a morphism that is
both a monomorphism and an epimorphism need not be an isomorphism, as we have
seen, for instance, in Example 1.3.2, (6). This motivates the following definition:
Definition 1.3.3 A morphism that is both a monomorphism and an epimorphism
is called a bimorphism. A category with the property that every bimorphism is an
isomorphism is called balanced.
Examples 1.3.4
(1) The inclusion .i : Z → Q is a bimorphism in the category .Ringc of unitary
commutative rings. Consequently, .Ringc is not a balanced category.
(2) The identity map .IdT : (T , P(T )) → (T , τ ) defined in Example 1.3.2, (7) is
obviously a bimorphism in Top but not an isomorphism. Therefore, Top is not
a balanced category.
(3) The categories Set, Grp, Ab, .R M are balanced. .

Definition 1.3.5 A category in which every morphism is an isomorphism is called

a groupoid.
Examples 1.3.6
(1) The category associated to a group as in Example 1.2.2, (3) is perhaps the first
obvious example of a groupoid.
(2) More generally, we can associate a groupoid to any category. Indeed, given
a category .C we consider its subcategory .Cgrp consisting of all objects and
all isomorphisms; in other words, when constructing .Cgrp we leave out all
morphisms of .C which are not isomorphisms. .Cgrp is usually called the core
groupoid of .C. .

For the remaining of this section we focus on various properties of objects.

1.3 Special Objects and Morphisms in a Category 11

Definition 1.3.7 Let .C be a category and .A ∈ Ob C.

(1) A is called an initial object if the set .HomC (A, B) has exactly one element for
each .B ∈ Ob C.
(2) A is called a final object if the set .HomC (C, A) has exactly one element for
each .C ∈ Ob C.
(3) If A is both an initial and a final object we say that A is a zero-object. A category
which admits a zero-object is called pointed.
Remark 1.3.8 Note that in a pointed category there exists a morphism between
any two objects. Indeed, if .C is a pointed category with zero-object .C0 and
A, B are two arbitrary objects of .C, then we have a unique morphism .0A ∈
HomC (A, C0 ) and also a unique morphism .0B ∈ HomC (C0 , B). By composing
the two aforementioned morphisms we obtain a morphism .0A, B = 0B ◦ 0A called
the zero-morphism from A to B.
Proposition 1.3.9 If A and B are initial (resp. final) objects in a category .C then
A is isomorphic to B.
Proof Since A is initial there exists a unique morphism .f : A → B and a unique
morphism from A to A which must be the identity .1A . The same applies for B: there
exists a unique morphism .g : B → A and a unique morphism from B to B, namely
the identity .1B . Now .g ◦ f ∈ HomC (A, A) and thus .g ◦ f = 1A . Similarly we get
.f ◦ g = 1B and we have proved that A and B are isomorphic. The statement about

final objects can be proved in a similar manner. 


Examples 1.3.10
(1) In the category Set of sets the initial object is the empty set while the final
objects are the singletons, i.e., the one-element sets .{x}. Thus Set has infinitely
many final objects and they are all isomorphic.
(2) The category Set of sets has no zero-objects. In the categories Grp, Ab and .R M
the trivial group, respectively trivial module, is the zero-object.
(3) The category Ring of unitary rings has the ring of integers .Z as its initial object
while the zero ring .{0} is its final object (note that .{0} becomes an object in
Ring by assuming that .1 = 0).
(4) The category Field of fields has neither an initial nor a final object since there
are no morphisms between fields of different characteristics.
(5) Let .(X, ) be a pre-ordered set and PO(.X, ) the associated category (see
Example 1.2.2, (2)). Then PO(.X, ) has an initial object if and only if .(X, )
has a least element (i.e., some element .0 ∈ X such that .0 ≤ x for any .x ∈
X). Similarly, PO(.X, ) has a final object if and only if .(X, ) has a greatest
element (i.e., some element .1 ∈ X such that .x ≤ 1 for any .x ∈ X).
(6) The category .Set(⊆) defined in Example 1.2.2, (2) has an initial object and no
final object. Indeed, note first that the empty set is the initial object of .Set(⊆)
as for any set X we have .∅ ⊂ X and therefore there exists a unique morphism
between .∅ and X, namely .u∅, X .
12 1 Categories and Functors

The category .Set(⊆) has no final object. Indeed, U being a final object in
Set(⊆) would imply that for all sets X there exists a unique morphism .uX, U .
.

Consequently, .X ⊆ U for all sets X, which is a contradiction to the power-set

axiom ([37, Axiom 3.12]). .

It is well known that for any object in Set one can define the notion of a subset.
The corresponding concept in an arbitrary category is called a subobject:
Definition 1.3.11 Let .C be a category and .C ∈ Ob C. An equivalence class
of monomorphisms with codomain C is called a subobject of C, where two
monomorphisms .f ∈ HomC (A, C) and .g ∈ HomC (B, C) are equivalent if there
exists an isomorphism .u ∈ HomC (A, B) such that .g ◦ u = f .
Note that if the category .C is not small then the subobjects of a given object C
might form a class rather than a set.
Examples 1.3.12
(1) In the category Set, the class of subobjects of a set X is in bijection with the
power set of X. Indeed, if we denote by .SO(X) the class of all subobjects of X,
the map .ψX : P(X) → SO(X) defined as follows is bijective:

ψX (Y ) = i
. Y, Y ⊆ X,

where .iY denotes the equivalence class of the inclusion monomorphism

iY : Y → X. First we show that .ψX is surjective. To this end, we prove
.

that any monomorphism .f : V → X whose image is the subset Y of X

belongs to the equivalence class of the inclusion monomorphism .iY . Recall
that monomorphisms in Set coincide with injective functions and denote
by .u : V → Y the map obtained by restricting the range of f to Y , i.e.,
.u(v) = f (v) for all .v ∈ V . Then u is obviously a bijection and moreover we

Y = f, as desired.
have .iY ◦ u = f , which shows that .i
We are left to prove that .ψX is also injective. Consider now Y , .Z ⊆ X
Y = i
such that .i Z . Hence, there exists a bijective map .u : Y → Z such that
.iZ ◦ u = iY . This comes down to .u(y) = y for all .y ∈ Y and therefore .Y ⊆ Z.

In the same manner, .iZ = iY ◦ u−1 leads to .Z ⊆ Y . To conclude, we proved

that .Y = Z, as desired.
Similarly one can see that in Grp subobjects of a group G are in bijection
with the subgroups of G.
(2) In the category Top the situation is somewhat different. Recall that the
subspaces of a topological space .(X, τ ) are pairs .(Y, τY ) where .Y ⊆ X
and .τY = {Y ∩ U | U ∈ τ } is the subspace topology on Y . On the
other hand, the subobjects of .(X, τ ) are in bijection with the topological
spaces .(Z, U) where .Z ⊆ X and .U is a topology finer than .τZ .11 If

11 Let .τ , .τ be two topologies on a set X. Then .τ2 is called finer than .τ1 (or .τ1 is called coarser
1 2
than .τ2 ) if .τ1 ⊆ τ2 .
1.3 Special Objects and Morphisms in a Category 13

we denote by .SO(X) the class of all subobjects of X and by .T(X) =

{(Z, U) | (Z, U) is a topological space, where Z ⊆ X and τZ ⊆ U}, then
the map .ψX : T(X) → SO(X) defined as follows is a bijection:

.ψX (Z, U) = i
Z, (Z, U) ∈ T(X),

where 
 .iZ denotes the equivalence class of the inclusion monomorphism
.iZ : Z, U → (X, τ ). Note that .iZ is obviously continuous as .U contains .τZ ,

the coarsest topology on Z for which the inclusion map is continuous.

We show first that .ψX is injective. To this end, consider .(Y, U), .(Z, V) ∈
T(X) such that .i Y = iZ . Hence, there exists an isomorphism .u : (Y, U) →
(Z, V) in Top such that .iZ ◦ u = iY . This implies that .u(y) = y for all .y ∈ Y
and therefore .Y ⊆ Z and .V ⊆ U. Furthermore, we also have .iZ = iY ◦ u−1
and since u is a homeomorphism (see Example 1.3.2, (3)) we obtain .Z ⊆ Y and
.U ⊆ V. Therefore, .Y = Z and .U = V, as desired.

We are left to show that .ψX is surjective. Consider .f : (Z, V) → (X, τ ) to

be a monomorphism in Top. By Example 1.3.2, (1) f is an injective continuous
map. Let .Y = Im(f ) and denote by .g : Z → Y the bijection induced by
restricting the range of f , i.e., .f = iY ◦ g, where .iY : Y → X denotes
the inclusion. Define the topology .Uf on Y by letting its open subsets be
the images of open subsets of Z under f , i.e., .Uf = {f (V ) | V ∈ V}.
Note that this is indeed a topology on Y as the injectivity of f implies that
.f (V1 ∩V2 ) = f (V1 )∩f (V2 ) for all .V1 , .V2 ∈ V ([37, Exercise 1.4]). Therefore,

.g : (Z, V) → (Y, Uf ) is a homeomorphism. The inclusion .iY : (Y, Uf ) →

(X, τ ) is obviously continuous. Indeed, let .U ∈ τ ; as .f : (Z, V) → (X, τ ) is

continuous, we have .f −1 (U ) = V ∈ V. We obtain:
 
iY−1 (U ) = {y ∈ Y | iY (y) ∈ U } = Y ∩ U = f {z ∈ Z | f (z) ∈ U }
.
 
= f f −1 (U ) = f (V ).

As the subspace topology .τY is the coarsest topology on Y for which the
inclusion map .iY is continuous, we have .τY ⊆ Uf . This shows that .(Y, Uf ) ∈
T(X). Furthermore, as .g : (Z, V) → (Y, Uf ) is a homeomorphism and
.f = iY ◦ g, we can conclude that .iY = f and therefore .ψX (Y, Uf ) = i
Y = f.
This shows that .ψX is also surjective, as desired.
(3) As opposed to Top, in KHaus subobjects of a given object K are in bijection
with the closed subsets of K. Indeed, if we denote by .SSc (K) and .SO(K) the
set of all closed subsets and respectively the class of all subobjects of K, the
map .ψK : SSc (K) → SO(K) defined as follows is a bijection:

ψK (Y ) = i
. Y, Y ∈ SSc (K),

where 
 .iY denotes the equivalence class of the inclusion monomorphism
.iY : Y, τY → (K, τ ) and .τY is the subspace topology on Y .
14 1 Categories and Functors

Consider first Y , .Z ⊆ K such that .i Y = i Z . Therefore, we have an

isomorphism .u : (Y, τY ) → (Z, τZ ) in KHaus such that .iZ ◦ u = iY , which
implies, as in the case of the category Top, that .Y = Z. This shows that .ψK is
injective.
Consider now a monomorphism .f : (Z, V) → (X, τ ) in KHaus. In
particular f is injective and if we denote .Im(f ) by Y then the induced map
.u : (Z, V) → (Y, τY ) obtained by restricting the range of f is obviously

bijective. Moreover, u is also continuous; indeed, if .W ∈ τY then there exists

.U ∈ τ such that .W = U ∩ Y and we obtain

u−1 (W ) = u−1 (U ∩ Y ) = {z ∈ Z | u(z) ∈ U ∩ Y } = {z ∈ Z | f (z) ∈ U ∩ Y }

.

= {z ∈ Z | f (z) ∈ U } = f −1 (U ) ∈ V.

Now recall that any subspace of a Hausdorff space is also Hausdorff;

this shows that .(Y, τY ) is a Hausdorff space. It can be easily seen, as in
Example 1.3.2, (8) that u is in fact an isomorphism in KHaus. We are left to
show that .(Y, τY ) is a closed subspace of .(X, τ ). First note that since .(Z, V) is
in particular a compact space, its image under the continuous map f is compact
([39, Theorem 26.5]) as well. Therefore .(Y, τY ) is a compact subspace in the
Hausdorff space .(K, τ ) and is closed by [39, Theorem 26.3]. To summarize, we
have an isomorphism .u : (Z, V) → (Y, τY ) in KHaus such that .iY ◦ u = f ,
which shows that .i Y = f and therefore .ψK is also surjective.
.

Similarly, we can define quotient objects:

Definition 1.3.13 Let .C be a category and .C ∈ Ob C. An equivalence class of
epimorphisms with domain C is called a quotient of C, where two epimorphisms
.f ∈ HomC (C, A) and .g ∈ HomC (C, B) are equivalent if there exists an
isomorphism .u ∈ HomC (A, B) such that .u ◦ f = g.
As in the case of subobjects, if the category .C is not small then the quotients of a
given object C might form a class, not a set.
Examples 1.3.14
(1) In the category Set, the quotients of a set X are in bijection with the set of all
equivalence relations on X. Indeed, if we denote by .E(X) and .Q(X) the set of
equivalence relations and respectively the quotient objects on X, then the map
.ψX : Q(X) → E(X) defined as follows is a bijection:

ψX (f ) = Ef ,
. for all epimorphisms f ∈ HomSet (X, Y ),

where .Ef is the equivalence relation on X induced by f . Recall that .x Ef x if

and only if .f (x) = f (x ). We check first that .ψX is well-defined. Consider two
epimorphisms .f ∈ HomSet (X, Y ) and .g ∈ HomSet (X, Z) such that .f = g.
1.3 Special Objects and Morphisms in a Category 15

Hence, there exists an isomorphism .u ∈ HomSet (Y, Z) such that .u ◦ f = g. As

u is a set bijection, we clearly have .f (x) = f (x ) if and only if .g(x) = g(x ),
which shows that .Ef = Eg , as desired.
Next, we prove that .ψX is injective. To this end, let .f ∈ HomSet (X, Y )
and .g ∈ HomSet (X, Z) be two epimorphisms such that .Ef = Eg . Since f is
surjective (see Example 1.3.2, (2)) we can define a map .u ∈ HomSet (Y, Z) by
 
u f (x) = g(x) for all x ∈ X.
. (1.1)

Then .Ef = Eg implies, in particular, that if .f (x) = f (x ) then .g(x) = g(x ).

This shows that u is well-defined. Furthermore, it follows from (1.1) and the
surjectivity of g (Example 1.3.2, (2)) that u is also surjective. We show that u
is injective as well. Indeed, let y, .y ∈ Y be such that .u(y) = u(y ). As f is
surjective, we can find .xy , .xy ∈ X such that .f (xy ) = y and .f (xy ) = y .
Therefore we have .u f (xy ) = u f (xy ) and by (1.1) we obtain .g(xy ) =
g(xy ). Now .Ef = Eg implies that .f (xy ) = f (xy ) i.e., .y = y . Thus u is a
bijection, which proves that .f = g and therefore .ψX is injective.
We are left to show that .ψX is surjective. Indeed, consider an equivalence
relation E on X. We will show that .E = Eπ , where .π : X → X/E, .π(x) =  x
for all .x ∈ X, and .X/E denotes the set of equivalence classes of X by E.
Assume first that .x Ex for some x, .x ∈ X. Then x and .x belong to the same
equivalence class in .X/E. This shows that .π(x) = π(x ) and therefore .x Eπ x .
Conversely, if .x Eπ x then .π(x) = π(x ), which implies that . x = x . Thus x
and .x are in the same equivalence class in .X/E, i.e., .x E x . Putting all of this
together, we have proved that .E = Eπ and .ψX (π ) = E.
(2) In the category Grp, the quotient objects of a group G are in bijection with
the set of normal subgroups of G. Indeed, if we denote by .N(G) and .Q(G)
the set of normal subgroups and respectively quotient objects of G, the map
.ψG : N(G) → Q(G) defined as follows is a bijection:

ψG (K) = πK ,
. for all normal subgroups K of G,

where .πK denotes the equivalence class of the canonical projection

πK : G → G/K.
.

We show first that .ψG is injective. Consider two normal subgroups K, .K of

G such that .πK = πK . Hence, there exists an isomorphism .u : G/K
 → G/K
in Grp such that .u ◦ πK = πK . Now .x ∈ K implies that .u πK (x) = 1
and since u is in particular injective we obtain that .πK (x) = 1. Thus .x ∈
K , which leads to .K ⊂ K . Furthermore, if .x ∈ K then .1 = u πK (x) =
πK (x). This implies that .x ∈ K and therefore .K ⊂ K. Hence .K = K , as
desired.
Consider now an epimorphism .f : G → H in Grp and consider .K =
ker(f ), which is a normal subgroup of H . We will show that f is equivalent
to .πK . This shows that .ψG (K) = f and therefore .ψG is surjective as well.
16 1 Categories and Functors

Indeed, as in particular we have .K ⊆ ker(f ), the universal property of the

quotient group yields a unique group homomorphism .u : G/K → H such that

u ◦ πK = f.
. (1.2)

The proof will be finished once we show that u is an isomorphism. To

start with, let .
x , .
y ∈ G/K be such that .u(
x ) = u(y ). Now (1.2) implies
that .f (x) = f (y) and therefore .xy −1 ∈ Ker(f ) = K. Thus . x = y
and u is injective. We are left to show that u is surjective as well. To
this end recall from Example 1.3.2, (2) that f is surjective as a conse-
quence of being an epimorphism in Grp. The surjectivity of f together
with (1.2) implies that u is surjective too and the proof is finished.
.

Categories for which the subobjects (resp. quotients) of any given object form a
set are particularly important.
Definition 1.3.15 A category .C is called well-powered if the subobjects of any
object form a set. Similarly, .C is called co-well-powered if the quotients of any
object form a set.
Example 1.3.16 The discussion in Examples 1.3.12 and 1.3.14 immediately
implies that Set and Grp are both well-powered and co-well-powered. Furthermore,
in light of Example 1.3.12, Top and KHaus are well-powered. .

1.4 Some Constructions of Categories

In this section we provide several methods of constructing new categories. The first
one relies on formally reversing the direction of the morphisms in the given category
leading to what is called the dual category.
Definition 1.4.1 Given a category .C, the dual (or opposite) category of .C, denoted
by .Cop , is defined as follows:
(i) .Ob Cop = Ob C;
(ii) .HomCop (A, B) = HomC (B, A); in order to avoid any confusion we write
.f
op : A → B for the morphism of .Cop corresponding to the morphism

.f : B → A of .C;

(iii) the composition map

◦op : HomCop (A, B) × HomCop (B, C) → HomCop (A, C)

.

is defined as follows:

g op ◦op f op = (f ◦ g)op , for all f op ∈ HomCop (A, B), g op ∈ HomCop (B, C);
.
1.4 Some Constructions of Categories 17

op
(iv) the identities are the same as in .C, i.e., .1C = 1C for all .C ∈ Ob C.
Examples 1.4.2
(1) Obviously .(Cop )op = C for any category .C.
(2) Let PO(.X, ) be the category associated to the pre-ordered set .(X, ). Then
.PO(X, )
op = PO(X, ), where . is the pre-order on X defined as follows:

.x  y if and only if .y  x. Indeed, this follows immediately by observing that

there exists a morphism in .PO(X, )op from x to y if and only if there exists a
morphism in .PO(X, ) from y to x. .

The dual category introduced above suggests that we can assign a dual to any
categorical concept. More precisely, the dual of a certain concept will be obtained
by considering this concept in the dual category. To illustrate this, we highlight
several dual concepts we have encountered so far.
Proposition 1.4.3 Let .C be a category and .f ∈ HomC (A, B).
(1) f is a monomorphism in .C if and only if .f op is an epimorphism in .Cop .
(2) I is an initial object in .C if and only if I is a final object in .Cop .
(3) f is an isomorphism in .C if and only if .f op is a isomorphism in .Cop .
Proof
op op op op
(1) Let .h1 , .h2 ∈ HomCop (A, C) be such that .h1 ◦op f op = h2 ◦op f op . In other
words, we have .f ◦ h1 = f ◦ h2 and the desired conclusion follows easily.
(2) If I is an initial object in .C then .HomC (I, C) has exactly one element for all
.C ∈ Ob C. Since .HomCop (C, I ) = HomC (I, C), the conclusion follows.

(3) f is an isomorphism if and only if there exists a morphism .g ∈ HomC (B, A)

such that .f ◦ g = 1B and .g ◦ f = 1A . This is equivalent to the existence of a
morphism .g op ∈ HomCop (A, B) such that .g op ◦op f op = 1B and .f op ◦op g op =
1A . Hence .f op is an isomorphism in .Cop .


To conclude, the notions of epimorphisms and monomorphisms are dual to each
other; similarly, initial objects are dual to final objects while the notion of an
isomorphism is self-dual. We will discuss the duality principle in more depth in
Sect. 1.8.
Next we introduce the product of two given categories. The construction can be
easily extended to any family of categories indexed by a set.
Definition 1.4.4 Let .C and .D be two categories. We define the product category
C × D as follows:
.
 
(i) .Ob C × D = Ob C × Ob D, i.e., the objects of .C × D are pairs of the form
.(C, D) with .C ∈ Ob C and .D ∈ Ob D;
 
(ii) .HomC×D (C, D), (C , D ) = HomC (C, C ) × HomD (D, D );
18 1 Categories and Functors

(iii) the composition map is defined as follows:

   
HomC×D (C, D), (C , D ) × HomC×D (C , D ), (C , D
. )
 
→ HomC×D (C, D), (C , D ) ,
(f , g ) ◦ (f, g) = (f ◦ f, g ◦ g),
 
for all (f, g) ∈ HomC×D (C, D), (C , D ) ,
 
(f , g ) ∈ HomC×D (C , D ), (C , D ) ;

(iv) .1(C, D) = (1C , 1D ).

Example 1.4.5 Let .G1 and .G2 be two groups and consider the associated cate-
gories (as described in Example 1.2.2, (3)) denoted by .G1 and .G2 , respectively.
Then the product category .G1 × G2 is given by the associated category of the direct
product of groups .G1 × G2 . .

Another way of constructing categories involves directed graphs. We start by

reviewing these first.
Definition 1.4.6 A graph consists of a class .V whose elements are called vertices
and for each pair .(A, B) ∈ V × V a set .E(A, B) whose elements are called edges.
A graph is called small if .V is a set.
A path in a graph is a non-empty finite sequence .(A1 , f1 , A2 , . . . , An ) of
vertices and edges succeeding one another such that the first and the last terms are
vertices and each edge .fi ∈ E(Ai , Ai+1 ). A path of the form .(A) is called the trivial
path on A.
Notice that every category is in particular a graph; this can be easily seen by
leaving aside the composition law of the given category and by forgetting which
morphisms are identities. Conversely, we will be able to construct a category out of
a graph as follows:
 
Definition 1.4.7 Let .G = V, E be a small graph.12 The free category on the
graph G, denoted by .G, is constructed by considering:
(i) .Ob G = V as class of objects;
(ii) .HomG (A, B) = P(A, B) the set of paths between the vertices A and B, for
any A, .B ∈ Ob G;
(iii) the composition of morphisms is given by concatenation of paths:

(An , fn , . . . , Am ) ◦ (A1 , f1 , . . . , An ) = (A1 , f1 , . . . , An , fn , . . . , Am );

.

(iv) the identity maps are given by the trivial paths on each object.

12 The smallness assumption on the graph G is needed in order for the paths between any two

vertices to form a set.

1.4 Some Constructions of Categories 19

Example 1.4.8 Let G be the oriented graph depicted below:

1 2

. 1 2 3

Then the free category .G on the graph G is given as follows:

.Ob G = {v1 , v2 , v3 },
HomG (v1 , v1 ) = {(v1 )}, HomG (v1 , v2 ) = {(v1 e1 v2 )},
HomG (v1 , v3 ) = {(v1 e1 v2 e2 v3 )},
HomG (v2 , v1 ) = ∅, HomG (v2 , v2 ) = {(v2 )}, HomG (v2 , v3 ) = {(v2 e2 v3 )},
HomG (v3 , v1 ) = ∅, HomG (v3 , v2 ) = ∅, HomG (v3 , v3 ) = {(v3 )}.


.

Definition 1.4.9 Let .C be a category. A diagram in .C is a graph whose vertices and

edges are objects and respectively morphisms of .C. A diagram in .C will be called
commutative if for each pair of vertices, every two paths between them are equal as
morphisms.
Examples 1.4.10 Let .C be a category and A, B, C, .D ∈ Ob C.
(1) The following diagram is commutative if and only if .g ◦ f = h:

f
A B
g
h

.
C

(2) The following diagram is commutative if and only if .g ◦ f = h ◦ k:

f
A B

k g

C D
. h
20 1 Categories and Functors

(3) The following diagram is commutative if and only if .h ◦ g ◦ f = k:

f
A B

k g

C D
. h


.

The last construction we introduce is that of a quotient category; it involves a

certain kind of equivalence relation on the class of all morphisms of a given category.
Definition 1.4.11 Let .C be a category. An equivalence relation .∼ on the class of
all morphisms . HomC (A, B) of .C is called a congruence if the following
A, B∈Ob C
conditions are fulfilled:
(i) if .f ∈ HomC (A, B) and .f ∼ f then .f ∈ HomC (A, B);
(ii) if .f ∼ f , .g ∼ g and the composition .g ◦ f exists then .g ◦ f ∼ g ◦ f .
In fact, a congruence on a given category can be built up from certain equivalence
relations on all hom sets of that category. We make this precise in the following:
Proposition 1.4.12 Defining a congruence relation on a given category .C is
equivalent to specifying for each pair of objects A, B, an equivalence relation .∼A,B
on .HomC (A, B) such that:
(1) if f , .g : A → B and .h : B → C are such that .f ∼A,B g then .h ◦ f ∼A,C h ◦ g;
(2) if .f : A → B and g, .h : B → C are such that .g ∼B,C h then .g ◦ f ∼A,C h ◦ f .
Proof Indeed, an equivalence relation on . HomC (A, B) satisfying (i) of Defi-
A, B∈C
nition 1.4.11 restricts to an equivalence relation on each .HomC (A, B). Furthermore,
(ii) of Definition 1.4.11 shows that the two conditions listed above are fulfilled by
considering .g = g and .f = f respectively.
Conversely, by putting together the equivalence relations .∼A,B for each pair of
objects A, B in .C we obtain an equivalence relation on the entire class of morphisms
. HomC (A, B) satisfying (i) of Definition 1.4.11. Assume now that .f ∼ f
A, B∈C
and .g ∼ g such that the composition .g ◦ f exists. Then f , .f : A → B, g, .g : B →
C and using conditions (1) and (2) we obtain

g ◦ f ∼A,C g ◦ f ∼A,C g ◦ f ,
.

which proves that (2) holds, as desired. 


We can now introduce the quotient category:
1.4 Some Constructions of Categories 21

Proposition 1.4.13 Let .∼ be a congruence on a category .C and denote by .f the

equivalence class of a morphism f of .C. Then .C/ ∼ defined below is a category
called a quotient category of .C:
(i) .Ob C/ ∼ = Ob C;
(ii) .HomC/∼ (A, B) = { f | f ∈ HomC (A, B)} for any A, .B ∈ Ob C/ ∼;
(iii) the composition map .HomC/∼ (A, B) × HomC/∼ (B, C) → HomC/∼ (A, C) is
defined as follows:

g ◦ f = g ◦ f , for all f ∈ HomC/∼ (A, B), g ∈ HomC/∼ (B, C);

. (1.3)

(iv) the identity on .A ∈ Ob C/ ∼ is .1A .

Proof Proposition 1.4.12 shows that .∼ induces a partition on each hom set
HomC (A, B) and therefore .HomC/∼ (A, B) is also a set. Furthermore, Defini-
.

tion 1.4.11, (ii) shows that the composition law in .C/ ∼ is well-defined. 

Examples 1.4.14
(1) Consider a group G and let .G be the associated category (as in Exam-
ple 1.2.2, (3)) whose unique object we denote by .∗. There is a bijection between
normal subgroups of G and congruence relations on .G. Furthermore, for a
normal subgroup N of G, the quotient category by the congruence relation
induced by N is the quotient group .G/N.
Indeed, suppose first that N is a normal subgroup of G and define the
following equivalence relation on .HomG (∗, ∗) = G:

.g ∼N h if and only if gh−1 ∈ N.

The equivalence relation defined above is in fact a congruence on G in the

sense of Definition 1.4.11. To this end, we show that the two conditions in
Proposition 1.4.12 are fulfilled. Indeed, let g, h, .t ∈ G be such that .g ∼N h.
Then .gh−1 ∈ N and since any normal subgroup is invariant under conjugation
we obtain

tg(th)−1 = tgh−1 t −1 ∈ N,
. gt (ht)−1 = gtt −1 h−1 = gh−1 ∈ N.

Therefore .tg ∼N th and .gt ∼N ht and we can conclude that .∼N is a

congruence on G. Obviously, the quotient category .C/ ∼N coincides with the
quotient group .G/N.
Consider now .∼ to be a congruence relation on the morphisms of .G and
let .N∼ = {xy −1 | x, y ∈ G such that x ∼ y}. Note that the reflexivity and
symmetry of .∼ imply that .1G ∈ N∼ and respectively .g −1 ∈ N∼ whenever
.g ∈ N∼ . Consider now g, .g ∈ N∼ , i.e., .g = xy
−1 and .g = zt −1 for some

x, y, z, .t ∈ G such that .x ∼ y and .z ∼ t. Using conditions (1) and (2) of

Proposition 1.4.12 we obtain .xy −1 ∼ 1G and .1G ∼ tz−1 . By transitivity of .∼
22 1 Categories and Functors

it follows that .xy −1 ∼ tz−1 and therefore .gg = xy −1 zt −1 = xy −1 (tz−1 )−1 ∈

N∼ . Hence .N∼ is a subgroup of G. Consider now .h ∈ G and .g ∈ N∼ with
.g = xy
−1 , x, .y ∈ G such that .x ∼ y. Then we have .hgh−1 = hxy −1 h−1 =

hx(hy) −1 ∈ N∼ where the last term belongs to .N∼ due to condition (1) of
Proposition 1.4.12. We have proved that .N∼ is a normal subgroup of G, as
desired.
In order to conclude that there is a bijection between the normal subgroups
of G and the congruence relations on .G we are left to show that for all normal
subgroups N of G and all congruence relations .∼ on .G we have

N∼N = N
. and ∼N∼ = ∼ .

To start with, we have

N∼N = {xy −1 | x, y ∈ G such that x ∼N y}

.

= {xy −1 | x, y ∈ G such that xy −1 ∈ N} = N

and our first claim is proved.

Assume now that .g ∼ h; then .gh−1 ∈ N∼ which implies that .g ∼N∼ h.
Conversely, .g ∼N∼ h gives .gh−1 ∈ N∼ = {uv −1 | u, v ∈ G such that u ∼ v}.
Hence .gh−1 = xy −1 for some x, .y ∈ G such that .x ∼ y. Now since .x ∼ y and
.y
−1 ∼ y −1 , Proposition 1.4.12, (2) implies .xy −1 ∼ yy −1 = 1. Putting all the

above together yields

g = (xy −1 )h ∼ 1h = h, i.e., g ∼ h.
.

To summarize, we have proved that for all g, .h ∈ G we have .g ∼ h if and only

if .g ∼N∼ h.
(2) On the category Top we consider the relation .∼, called homotopy, defined as
follows for all f , .g ∈ HomTop (X, Y ): .f ∼X,Y g if and only if there exists a
continuous map .F : X ×[0, 1] → Y satisfying .F (x, 0) = f (x) and .F (x, 1) =
g(x) for all .x ∈ X. We will prove that homotopy is a congruence on Top.
To start with, we show that for all topological spaces X and Y , .∼X,Y is
an equivalence relation on .HomTop (X, Y ). Indeed, first note that for any .f ∈
HomTop (X, Y ) we have .f ∼X,Y f by considering the continuous map .F : X×
[0, 1] → Y defined by .F (x, t) = f (x). Furthermore, if .f ∼X,Y g then there
exists a continuous map .F : X × [0, 1] → Y such that .F (x, 0) = f (x) and
.F (x, 1) = g(x) for all .x ∈ X. Then, the continuous map .G : X × [0, 1] → Y

defined by .G(x, t) = F (x, 1 − t) shows that .g ∼X,Y f .

Finally, assume that .f ∼X,Y g and .g ∼X,Y h. Thus, there exist two
continuous maps .F : X × [0, 1] → Y and .G : X × [0, 1] → Y such that
for all .x ∈ X we have

F (x, 0) = f (x),
. F (x, 1) = g(x), G(x, 0) = g(x), G(x, 1) = h(x).
1.4 Some Constructions of Categories 23

Since .F (x, 1) = G(x, 0) = g(x) we can consider the map .H : X×[0, 1] → Y

defined as follows:

F (x, 2t), if 0  t  1/2
.H (x, t) =
G(x, 2t − 1), if 1/2  t  1.

The pasting lemma13 implies that H is continuous. Moreover, we have

H (x, 0) = F (x, 0) = f (x),

. H (x, 1) = G(x, 1) = h(x), for all x ∈ X.

Therefore, we obtain .f ∼X,Y h and we can conclude that .∼X,Y is an

equivalence relation on .HomTop (X, Y ). We are left to prove that the conditions
in Proposition 1.4.12 also hold. To this end, let f , .f ∈ HomTop (X, Y ) and g,
.g ∈ HomTop (Y, Z).

If .f ∼X,Y f , then there exists a continuous map .F : X × [0, 1] → Y such

that

F (x, 0) = f (x),
. F (x, 1) = f (x).
 
We can now define .F : X × [0, 1] → Z by .F (x, t) = g F (x, t) . This yields

   
F (x, 0) = g F (x, 0) = g f (x) = (g ◦ f )(x),
.
   
F (x, 1) = g F (x, 1) = g f (x) = (g ◦ f )(x),

which shows that .g ◦ f ∼X,Z g ◦ f . Assume now that .g ∼Y,Z g and

consequently there exists a continuous map .G : Y × [0, 1] → Z such that

G(x, 0) = g(x),
. G(x, 1) = g (x).
 
Define .G : X × [0, 1] → Z by .G (x, t) = G f (x), t . This yields
   
.G (x, 0) = G f (x), 0 = g f (x) = (g ◦ f )(x),
   
G (x, 1) = G f (x), 1 = g f (x) = (g ◦ f )(x),

which shows that .g ◦ f ∼X,Z g ◦ f . We can now conclude that the homotopy
relation .∼ is a congruence on Top. The resulting quotient category .Top/ ∼ is
called the homotopy category and will be denoted by HTop. For a thorough
introduction to homotopy theory, one of the cornerstones of algebraic topology,
we refer the reader to [4]. .

13 The pasting lemma: Let .X = A B, where A and B are closed in X, and consider two
continuous maps .f : A → Y , .g : B → Y . If .f (x) = g(x) for all .x ∈ A B, then f and g

f (x), if x ∈ A
combine to give a continuous function .h : X → Y , defined by .h(x) = (see [39,
g(x), if x ∈ B
Theorem 18.3]).
24 1 Categories and Functors

1.5 Functors

Functors are structure preserving maps which will be used to relate different
categories in the way morphisms do with objects.
Definition 1.5.1 Let .C and .D be two categories. A covariant functor (respectively
contravariant functor) .F : C → D consists of the following data:
(1) a mapping .A → F (A) : Ob C → Ob D;
(2) for each pair of objects A, .B ∈ Ob C, a mapping

f → F (f ) : HomC (A, B) → HomD (F (A), F (B))

.

(respectively f → F (f ) : HomC (A, B) → HomD (F (B), F (A)))

subject to the following conditions:

(1) for every .A ∈ Ob C we have .F (1A ) = 1F (A) ;
(2) for every .f ∈ HomC (A, B), .g ∈ HomC (B, C) we have

.F (g ◦ f ) = F (g) ◦ F (f ) (respectively F (g ◦ f ) = F (f ) ◦ F (g)).

A functor .F : A × B → C defined on the product of two categories is called a

bifunctor (functor of two variables).
Throughout, the term functor will denote a covariant functor. Any reference to
contravariant functors will be explicitly stated.
Remark 1.5.2 Note that the image14 of a functor need not be a category. Indeed,
consider the following two categories .C and .D:

1C 1 1C 2 1C 3 1C 4

f g
: C1 C2 C3 C4

1D 1 1D 2 1D 3

h k
: D1 D2 D3

. k h

14 The image of a functor (or values of a functor as defined in [23]) .F : C → D consists of a class
.{F (C) | C ∈ Ob C} together with all sets .{F (f ) | f ∈ HomC (A, B)} for any A, .B ∈ Ob C.
1.5 Functors 25

The image of the functor .F : C → D defined below is not a category:

F (C1 ) = D1 , F (C2 ) = F (C3 ) = D2 , F (C4 ) = D3 ,

.

F (f ) = h, F (g) = k, F (1C1 ) = 1D1 , F (1C2 ) = F (1C3 ) = 1D2 , F (1C4 ) = 1D3 .

Indeed, the morphisms h and k are contained in the image of F while their
composition .k ◦ h is not. However, if F is injective on objects then it can be easily
seen that its image is indeed a category.
Examples 1.5.3
(1) If .C is a subcategory of .C we can define the inclusion functor .I : C → C
which sends every object as well as every morphism to itself. If .C = C then I
is just the identity functor .1C on .C.
(2) Let .∼ be a congruence on a category .C and .C/ ∼ the corresponding quotient
category. Then, we can define a quotient functor .Π : C → C/ ∼ as follows
for all .C ∈ Ob C and .f ∈ HomC (A, B):

Π (C) = C,
. Π (f ) = f ,

where .f denotes the equivalence class of the morphism f of .C. For all .f ∈
HomC (A, B) and .g ∈ HomC (B, C) we have

(1.3)
Π (g ◦ f ) = g ◦ f = g ◦ f = Π (g) ◦ Π (f ),
.

which shows that .Π is indeed a functor.

(3) If .F : C → D and .G : D → E are two functors, we can define a new functor
.G ◦ F : C → E by pointwise composition, i.e., .(G ◦ F )(C) = G F (C)
 
and .(G ◦ F )(f ) = G F (f ) for any .C ∈ Ob C and f morphism in .C. It is
straightforward to check that .G ◦ F respects compositions and identity maps.
For brevity, the pointwise composition of functors will sometimes be denoted
by juxtaposition, i.e., we write GF instead of .G ◦ F .
If F , G are both covariant or contravariant functors, then the pointwise
composition defined above yields a covariant functor. On the other hand, if one
of the functors is covariant and the other one contravariant then their pointwise
composition is a contravariant functor.
(4) For any category .C we can define a functor .OC : C → Cop which sends each
object to itself and a morphism .f ∈ HomC (C, C ) to the opposite morphism
.f
op ∈ Hom op (C , C). The functor .O is obviously contravariant.
C C
It can be easily seen that a functor .F : C → D is contravariant if and only
if .F ◦ OCop : Cop → D (or .OD ◦ F : C → Dop ) is a covariant functor.
(5) Let .C and .D be two categories and .D0 ∈ Ob D a fixed object. The constant
functor at .D0 , denoted by .D0 , assigns to every object .C ∈ Ob C the object
.D0 and to every morphism f in .C the identity morphism .1D0 .
26 1 Categories and Functors

(6) Let .C, .D be two categories and consider .C × D to be the product category as
defined in Definition 1.4.4. We can define two projection functors as follows:

pC : C × D → C, pC (C, D) = C, pC (f, g) = f,
.

pD : C × D → D, pD (C, D) = D, pD (f, g) = g
 
for all .(C, D) ∈ Ob C × D and .(f, g) ∈ HomC×D (C, D), (C , D ) .
(7) If I is a small discrete category, then a functor .F : I → C is uniquely defined
by a family of objects .(Ci )i∈I indexed by I . More precisely, such a functor is
completely determined by the images of each object .i ∈ Ob I , say .Ci . Indeed,
note that for all .i ∈ Ob I the image of the identity morphism .1i is forced to be
.1Ci . If I is the empty set, then there exists a unique functor .F : I → C, called

the empty functor.

(8) Let G, H be two groups and .G, respectively .H the corresponding associated
categories (see Example 1.2.2, (3)). Then, defining a functor .G → H is the
same as providing a group homomorphism .G → H .
(9) For any category .C we can define the bifunctor .HomC (−, −) : Cop × C → Set
as follows:
 
HomC (A, B) = HomC (A, B) ∈ Ob Set for all (A, B) ∈ Ob Cop × C ;
.
 
if (f op , g) ∈ HomCop ×C (A, B), (C, D) = HomCop (A, C) × HomC (B, D)
then HomC (f op , g) : HomC (A, B) → HomC (C, D) is defined by
HomC (f op , g)(h) = g ◦ h ◦ f, for all h ∈ HomC (A, B).
op
Indeed, for any .h ∈ HomC (A, B) we have .HomC (1A , 1B )(h) = 1B ◦h◦1A =
h, which shows that .HomC (−, −) respects identities.
Furthermore, we have
   
HomC (r op , t) ◦ (f op , g) (h)= HomC (f ◦ r)op , t ◦ g (h)= t ◦ g ◦ h ◦ f ◦ r
.

= HomC (r op , t)(g ◦ h ◦ f ) = HomC (r op , t) ◦ HomC (f op , g)(h)

 
for all .(f op , g) ∈ HomCop ×C (A, B), (C, D) = HomCop (A, C) ×
HomC (B, D) and .(r op , t) ∈ HomCop ×C (C, D), (E, F ) = HomCop (C, E)×
HomC (D, F ).  
Hence .HomC (r op , t) ◦ (f op , g) = HomC (r op , t) ◦ HomC (f op , g) and
.HomC (−, −) is indeed a functor, called the hom bifunctor.

(10) Given a category .C and a fixed object .C ∈ Ob C, we can define two functors,
one of them being covariant and the other one contravariant, called the hom
1.5 Functors 27

functors. Indeed, define .HomC (C, −), .HomC (−, C) : C → Set as follows:

(i) HomC (C, A) = HomC (C, A) ∈ Ob Set for all A ∈ Ob C;

.

if f ∈ HomC (A, B) then HomC (C, f ) : HomC (C, A) → HomC (C, B)

is defined by HomC (C, f )(g) = f ◦ g, for all g ∈ HomC (C, A).
(ii) HomC (A, C) = HomC (A, C) ∈ Ob Set for all A ∈ Ob C;
if f ∈ HomC (A, B) then HomC (f, C) : HomC (B, C) → HomC (A, C)
is defined by HomC (f, C)(g) = g ◦ f, for all g ∈ HomC (B, C).

In certain cases, the hom sets .HomC (A, B) can inherit some extra structure
from the objects of .C, as can be seen in the following examples. This simple
observation is the main idea behind the concept of an enriched category. For
the precise definition and further details on enriched category theory we refer
the reader to [33].
(11) If X, .Y ∈ Ob Top then the set of continuous maps .HomTop (X, Y ) can be
endowed with the so-called compact-open topology,15 which will turn out to
be particularly important when dealing with adjoint functors in Chap. 3. We
check first that if .f ∈ HomTop (Y, Z) then .HomTop (X, f ) : HomTop (X, Y ) →
HomTop (X, Z) is a continuous map with respect to the compact-open topol-
ogy. Indeed, let .K ⊆ X be a compact subset, .U ⊆ Z an open sub-
set and .W (K, U ) a sub-basis open set of the compact-open topology on
.HomTop (X, Z). As .f
−1 (U ) is an open subset of Y , we have

 
HomTop (X, f )−1 W (K, U )
.

= {t ∈ HomTop (X, Y ) | HomTop (X, f )(t) ⊆ W (K, U )}

= {t ∈ HomTop (X, Y ) | HomTop (X, f )(t)(K) ⊆ U }
= {t ∈ HomTop (X, Y ) | (f ◦ t)(K) ⊆ U }
= {t ∈ HomTop (X, Y ) | t (K) ⊆ f −1 (U )}
 
= W K, f −1 (U ) .

Since continuity of a map need only be checked on a sub-basis of the codomain

(see the discussion in [39, page 103]), we can conclude that .HomTop (X, f )
is continuous, as desired. Hence, we have a functor .HomTop (X, −) : Top →
Top.

15 The compact-open topology on .HomTop (X, Y ) is the topology generated by the sub-basis
.W (K, U ) = {f ∈ HomTop (X, Y ) | f (K) ⊆ U }, where .K ⊆ X is compact and .U ⊆ Y is
open.
28 1 Categories and Functors

(12) If A, .B ∈ Ob Ab then .HomAb (A, B) has an abelian group structure given by

.(f + g)(a) = f (a) + g(a) (1.4)

for all f , .g ∈ HomAb (A, B) and .a ∈ A. Moreover, if .f ∈ HomAb (B, C), it

can be easily seen that .HomAb (A, f ) is a group homomorphism. Therefore,
for any .A ∈ Ob Ab we have a functor .HomAb (A, −) : Ab → Ab.
(13) Let K be a field. If M, .N ∈ Ob K M then .HomK M (M, N ) has the abelian
group structure given as in (1.4) and a K-vector space structure defined by
scalar multiplication, i.e., given .k ∈ K, .f ∈ HomK M (M, N ) define the linear
map kf as follows:

(kf )(m) = kf (m),

. for all m ∈ M.

Therefore, any .M ∈ Ob K M yields a functor .HomK M (A, −) : K M → K M.

(14) Consider the field K as an object in .K M and denote the corresponding
contravariant hom functor .HomK M (−, K) by .(−)∗ : K M → Set. That is,
we denote .HomK M (V , K) by .V ∗ and .HomK M (u, K) by .u∗ . Given a vector
space V , the set .V ∗ of linear maps from V to K can be endowed with a vector
space structure as follows: for any f , .g ∈ V ∗ and a, .b ∈ K the linear map
.af + bg is defined by .(af + bg)(v) = af (v) + bg(v). Furthermore, given a

linear map .u : V → W it can be easily seen that .u∗ : W ∗ → V ∗ defined by

∗ ∗
.u (w) = w ◦ u for all .w ∈ W is also a linear map. Therefore, the functor
∗
.(−) maps into the category .K M and is called the dual space (contravariant)

functor.
(15) By composing the dual space functor with itself we obtain a covariant functor
denoted by .(−)∗∗ : K M → K M and called the double dual space functor.
We only point out for further use that if .u : U → V is a linear map then
.u
∗∗ : U ∗∗ → V ∗∗ is defined by

u∗∗ (φ) = φ ◦ u∗ for all φ : U ∗ → K.

. (1.5)

(16) The cartesian product bifunctor .−×− : Set×Set → Set is defined as follows
for all X, .Y ∈ Ob Set and .f ∈ HomSet (A, C), .g ∈ HomSet (B, D):

(− × −)(A, B) = A × B;
.

(− × −)(f, g) = f × g : A × B → C × D,
 
where .(f × g)(a, b) = f (a), g(b) for all .a ∈ A, .b ∈ B.
(17) For any set X we can define the corresponding cartesian product functor .− ×
X : Set → Set as follows for all Y , .Z ∈ Ob Set and .f ∈ HomSet (Y, Z):

(− × X)(Y ) = Y × X;
.

(− × X)(f ) = f × 1X : Y × X → Z × X,
1.5 Functors 29

 
where .(f × 1X )(y, x) = f (y), x for all .y ∈ Y , .x ∈ X. Similarly we can
define the cartesian product functor .X × − : Set → Set.
Exactly as in the case of the hom functor, in certain cases, when the
given sets are endowed with some extra structures, the corresponding cartesian
product inherits this structure.
(18) Any .X ∈ Ob Top defines a functor .− × X : Top → Top, where for all .Y ∈
Ob Top we consider on .Y × X the product topology.16
(19) Similarly, any .G ∈ Ob Grp defines a functor .− × G : Grp → Grp, where for
all .H ∈ Ob Grp the group structure on .H × G is defined component-wise.
(20) Let K be a field and for simplicity denote the tensor product over K by .⊗
(i.e., .⊗ = ⊗K ). For any .X ∈ Ob K M we can define a functor .− ⊗ X : K M →
K M, called the tensor product functor, as follows:

(− ⊗ X)(M) = M ⊗ X, for all M ∈ Ob K M;

.

(− ⊗ X)(f ) = f ⊗ 1X , for all f ∈ HomOb K M (M, N ),

where .f ⊗ 1X : M ⊗ X → N ⊗ X is the K-linear map defined by

(f ⊗ 1X )(m ⊗ x) = f (m) ⊗ x and the K vector space structure on .M ⊗ X
.

is given by .k(m ⊗ x) = km ⊗ x for all .k ∈ K, .m ∈ M and .x ∈ X. For more

details on the tensor product of vector spaces (or modules) we refer the reader
to [45].
(21) For any set X we denote by .P(X) = {Y | Y ⊆ X} the power set of X. We can
define two power set functors .P : Set → Set and respectively .P c : Set → Set,
the first one being covariant and the second one contravariant, as follows:

i) P : Set → Set, P (A) = P(A) ∈ Ob Set, for all A ∈ Ob Set;

.

if f ∈ HomSet (A, B) then P (f ) : P(A) → P(B) is defined by

P (f )(U ) = f (U ), for all U ⊆ A.
ii) P c : Set → Set, P c (A) = P(A) ∈ Ob Set, for all A ∈ Ob Set;
if f ∈ HomSet (A, B) then P c (f ) : P(B) → P(A) is defined by
P c (f )(V ) = f −1 (V ), for all V ⊆ B.

(22) The so-called forgetful functors are functors which forget (some of) the
structure on objects of the domain category. For instance the categories in
Example 1.2.2, (7)–(13) allow for a forgetful functor to the category Set of
sets, which sends the objects of that category to the underlying set, and the

16 Let .(X ) be a family of topological

i i∈I   spaces indexed by the set I and consider the product of
the underlying sets . i∈I Xi , (πi )i∈I . The topology generated by the sub-basis .S = β∈I Sβ
is called the product topology, where .Sβ = {πβ−1 (Uβ ) | Uβ open in Xβ } ([39, Definition, page
114]).
30 1 Categories and Functors

homomorphisms to the underlying mappings between the underlying sets.

Similarly, we have many other examples of forgetful functors which forget
only some of the structure in objects of the domain category:

Grp → Set, forgets about the group structure;

.

Top → Set, forgets about the topological structure;

Rng → Ab, forgets about the product;
Ab → Grp, forgets about the commutativity;
Ring → Rng, forgets about the unit;
Top∗ → Top, forgets about the base point;
Haus → Top, forgets about the Hausdorff property.

(23) For a group G we denote by .[G, G] its commutator subgroup; in other words
.[G, G] is the subgroup of .G generated by all elements of the form .xyx
−1 y −1 ,

where x, .y ∈ G. The corresponding quotient group .Gab = G/[G, G]

is obviously an abelian group called the abelianization of G. Furthermore,
if .f ∈ HomGrp (G, H ) is a group homomorphism and .πG : G → Gab ,
.πH : H → Hab denote the corresponding projections, we have .[G, G] ⊆

ker(πH ◦ f ). Hence, the universal property of the quotient group .Gab yields
a unique group homomorphism .fab : Gab → Hab which makes the following
diagram commutative:

πG
G Gab

f ab
πH f

.
Hab (1.6)

This construction allows us to define a functor .F : Grp → Ab, called the

abelianization functor, as follows:

F (G) = Gab , for all G ∈ Ob Grp;

.

F (f ) = fab , for all f ∈ HomGrp (G, H ).

 
(24) For a topological space X we denote by . X, iX the corresponding Stone–
Čech compactification, where .X is a compact Hausdorff topological space
while .iX : X → X is a continuous function (see [39, Section 38] for further
details). Given .f ∈ HomTop (X, Y ), the universal property of the Stone–
Čech compactification yields a unique .f ∈ HomKHaus (X, Y ) such that the
1.5 Functors 31

following diagram is commutative:

iX
X X i.e., f iX iY f.

f
iY f

.
Y (1.7)

Now we can define a functor .S : Top → KHaus, called the Stone–Čech

compactification functor, as follows:

S(X) = X, for all X ∈ Ob Top;

.

S(f ) = f , for all f ∈ HomTop (X, Y ).

 
(25) Given a monoid .(M, ·), we denote by . G(M), iM the universal enveloping
group of M (also called the group completion or the Grothendieck group of
M), where .G(M) is a group while .iM : M → G(M) is a homomorphism of
monoids (see [6, Section 4.11]).
For any .f ∈ HomMon (M, N ), the universal property of the universal
enveloping group yields a unique .f ∈ HomGrp (G(M), G(N )) such that the
following diagram is commutative:

iM
M G(M) i.e., f iM iN f.

f
iN f

.
G(N) (1.8)

The functor .G : Mon → Grp defined below is called the universal enveloping
group functor:

G(M) = G(M), for all M ∈ Ob Mon;

.

G(f ) = f , for all f ∈ HomMon (M, N ).

(26) Given a monoid .(M, ·), we denote by .U (M) the set of all invertible elements
of M. Furthermore, if .f ∈ HomMon (M, N ), it can be easily seen that
.f|U (M) ⊆ U (N ). This gives rise to a functor .U : Mon → Grp defined as

follows:

U(M) = U (M), for all M ∈ Ob Mon;

.

U(f ) = f|U (M) , for all f ∈ HomMon (M, N ).

32 1 Categories and Functors

(27) Similarly, given a ring .(R, +, ·) we denote by .U (R) the set of invertible
elements of .(R, ·). This yields a functor .U : Ring → Grp defined as follows:

U(R) = U (R), for all R ∈ Ob Ring;

.

U(f ) = f|U (R) , for all f ∈ HomRing (R, R ),

where .f|U (R) denotes the restriction of f to the subset .U (R) of R.

(28) Let R be a ring, .M ∈ Ob MR and .N ∈ Ob R M. We can define the functor
.BilM, N : Ab → Ab of R-bilinear maps as follows for all .A ∈ Ab and .f ∈

HomAb (A, B):17

BilM, N (A) = {α : M × N → A | α is R−bilinear};

.

BilM, N (f )(u) = f ◦ u, u ∈ BilM, N (A).

Note that each .BilM, N (A) can be made into an abelian group as in (1.4), i.e.,
for all .α, .β ∈ BilM, N (A) and .m ∈ M, .n ∈ N define

(α + β)(m, n) = α(m, n) + β(m, n).

.

(29) Let
 −1R be a commutative ring with unity, S a multiplicative subset−1
of R, and
. S R, j the corresponding localization ring, where .j : R → S R is the
ring homomorphism defined by .j (r) = 1r , for all .r ∈ R. Furthermore, if
.M ∈ Ob R M, we denote by .(S
−1 M, ϕ ) the corresponding localization
M
module, where .S M ∈ Ob S −1 R M and .ϕM : M → S −1 M is the R-module
−1

homomorphism defined by .ϕM (m) = m1 , for all .m ∈ M. We refer the reader

to [3, Chapter 12] for more details on localization of modules.
For any .f ∈ HomR M (M, N ), the universal property of the localization
module ([3, Theorem 12.3]) yields a unique .f ∈ HomS −1 R M (S −1 M, S −1 N)
such that the following diagram is commutative:

M
M S 1M i.e., f f.
M N

f
N f
S 1N
. (1.9)

17 .α : M × N → A is called R-bilinear if for all .r ∈ R, m, .m ∈ M and n, .n ∈ N we have

.α(m + m , n) = α(m, n) + α(m , n),

α(m, n + n ) = α(m, n) + α(m, n ),
α(mr, n) = α(m, rn).
1.5 Functors 33

The functor .L : R M →S −1 R M defined below is called the localization with

respect to S:

L(M) = S −1 M, for all M ∈ Ob R M;

.

L(f ) = f , for all f ∈ HomR M (M, N ).

 
(30) For a topological space X we denote by . H (X), qX the corresponding
Hausdorff quotient, where .H (X) is a Hausdorff topological space while
.qX : X → H (X) is a continuous function (see [44] for more details).

Given .f ∈ HomTop (X, Y ), the universal property of the Hausdorff

quotient (as stated, for instance, in [40, page 81]) yields a unique .f ∈
HomHaus (H (X), H (Y )) such that the following diagram is commutative:

qX
X H (X) i.e., f iX iY f.

f
qY f

.
H (Y ) (1.10)

Now we can define a functor .H : Top → Haus, called the Hausdorff quotient
functor, as follows:

H(X) = H (X), for all X ∈ Ob Top;

.

H(f ) = f , for all f ∈ HomTop (X, Y ).

 
(31) For an arbitrary ring R we denote by . D(R), jR the corresponding Dorroh
extension, where .D(R) is a unitary ring while .jR : R → D(R) is a morphism
in Rng.18
Given .f ∈ HomRng (R, S), the universal property of the Dorroh extension
(see [18, Theorem 3.1.1])19 yields a unique .h ∈ HomRing (D(R), D(S)) such

18 The Dorroh extension of an arbitrary ring R is defined as .D(R) = Z × R with componentwise

addition and multiplication given for all .(n1 , r1 ), .(n2 , r2 ) ∈ Z × R by .(n1 , r1 )(n2 , r2 ) =
(n1 n2 , n1 r2 + n2 r1 + r1 r2 ) and .jR : R → D(R) is given by .jR (r) = (0, r) for all .r ∈ R (we
refer to [15, Theorem 2.12] for more details).
19 For any .f ∈ Hom
Rng (R, S), where .S ∈ Ob Ring, there exists a unique .f ∈ HomRing (D(R), S)
such that .h ◦ jR = f .
34 1 Categories and Functors

that the following diagram is commutative:

jR
R D(R) i.e., f jR jS f.

f
jS f

.
D(S) (1.11)

Now we can define a functor .D : Rng → Ring, called the Dorroh extension
functor, as follows:

D(R) = D(R), for all R ∈ Ob Rng;

.

D(f ) = f , for all f ∈ HomRng (R, S).

(32) For any .u ∈ HomRingc (R, S) we can define a functor .Fu : S M →R M, called
restriction of scalars, as follows for all .M ∈ Ob S M and .f ∈ HomS M (M, N ):

Fu (M) = M, where M ∈ Ob R M via rm = f (r)m, for all r ∈ R, m ∈ M;

.

Fu (f ) = f.

(33) Given a pre-ordered set .(X, X ), we denote by .(X, AX ) the Alexandroff (or
Alexandrov) topology20 on X with respect to the pre-order .X . Furthermore,
if .f : (X, X ) → (Y, Y ) is order preserving then .f : (X, AX ) →
(Y, AY ) is continuous. Indeed, let .U ⊆ Y be an open subset, .x ∈ f −1 (U )
and .x ∈ X such that .x X x . As f is order preserving, we have
.f (x) X f (x ) and since .f (x) ∈ U we obtain .f (x ) ∈ U . Therefore,

.x ∈ f
−1 (U ), which shows that .f −1 (U ) ⊆ X is an open set. This yields a

functor .F : PreOrd → Top defined as follows:

F (X, X ) = (X, AX );

.

F (f ) = f.

(34) A contravariant functor .F : Top → Poset can be defined as follows for all
topological spaces .(X, τ ), .(Y, γ ) and continuous maps .f : (X, τ ) → (Y, γ ):

. F (X, τ ) = {U ⊆ X | U is open in τ };
F (f ) = f −1 ,

20 Given a pre-ordered set .(X, X ), the Alexandroff topology on X with respect to .X is defined
by considering a subset U of X to be open if .x X x and .x ∈ U imply .x ∈ U .
1.5 Functors 35

where .{U ⊆ X | U is open in τ } is partially ordered by set inclusion and

f −1 : {V ⊆ Y | V is open in γ } → {U ⊆ X | U is open in τ } denotes the
.

preimage of f . Indeed, note that since .f : (X, τ ) → (Y, γ ) was assumed to

be a continuous map we have .f −1 (V ) ∈ τ for any .V ∈ γ .
(35) Given two pre-ordered sets .(X, ) and .(Y, ), a functor between the
corresponding induced categories .PO(X, ) and .PO(Y, ), respectively, as
defined in Example 1.2.2, (2) is nothing but an order preserving function
between .(X, ) and .(Y, ). Consequently, a functor between .PO(X, )op
and .PO(Y, ) is an order-reversing function between .(X, ) and .(Y, ). .
Having defined a functor as a type of morphism between categories, it is only
natural to consider the following category:
Proposition 1.5.4 The small categories and functors between them constitute a
category which we will denote by Cat, called the category of small categories.
Proof Given two functors .F : C → D and .G : D → E we obtain a new functor .G ◦
F : C → E by pointwise composition. The composition law is obviously associative
and the identity functor on a category is an identity for this composition.
Finally, if .C and .D are small categories, i.e., .Ob C and .Ob D are sets, then
.HomCat (C, D) is also a set. 

Proposition 1.5.5 Let .F : A × B → C be a bifunctor. Then for any .A ∈ Ob A
there exists a functor .FA : B → C, called the right associated functor with respect
to A, defined as follows for all B, .B ∈ Ob B and .f ∈ HomB (B, B ):

FA (B) = F (A, B),

. FA (f ) = F (1A , f ).

Similarly, for any .B ∈ Ob B there exists a functor .F B : A → C, called the left

associated functor with respect to B, defined as follows for all A, .A ∈ Ob A and
.f ∈ HomA (A, A ):

F B (A) = F (A, B),

. F B (f ) = F (f, 1B ).

Proof We only show the first assertion. To this end, for any .B ∈ Ob B we
have .FA (1B ) = F (1A , 1B ) = F (1(A, B) ) = 1F (A, B) , where the last equality
holds because F is a functor. Thus .FA respects identities. Furthermore, for any
.f ∈ HomB (B, B ) and .g ∈ HomB (B , B ) we have

 
FA (g ◦ f ) = F (1A , g ◦ f ) = F (1A , g) ◦ (1A , f ) = F (1A , g) ◦ F (1A , f )
.

= FA (g) ◦ FA (f ).

We have proved that .FA respects compositions as well and the proof is now finished.


36 1 Categories and Functors

Examples 1.5.6
(1) For any object C in a category .C, the hom functor .HomC (C, −) is the right
associated functor of the Hom bifunctor with respect to C. Similarly, the
contravariant hom functor .HomC (−, C) is the left associated functor of the Hom
bifunctor with respect to C.
(2) For any set X, the cartesian product functors .X × − and .− × X are the right
and respectively the left associated functors with respect to X of the cartesian
bifunctor. .

1.6 Isomorphisms of Categories

We start by introducing the following notions, which are weaker than isomorphism
but very useful.
Definition 1.6.1 Let .F : C → D be a functor and for all A, .B ∈ Ob C consider the
following induced mapping:
 
FA, B : HomC (A, B) → HomD F (A), F (B) , f → F (f ).
. (1.12)

(1) The functor F is called faithful if the mappings .FA, B are injective for all A,
.B ∈ Ob C.

(2) The functor F is called full if the mappings .FA, B are surjective for all A, .B ∈
Ob C.
(3) The functor F is called fully faithful if the mappings .FA, B are bijective for all
A, .B ∈ Ob C.
(4) The functor F is called essentially surjective if each object .D of D is
isomorphic to an object of the form .F (C) for some .C ∈ Ob C.
Examples 1.6.2
(1) The inclusion functor is automatically faithful. If the subcategory is full then
the inclusion functor is also full.
(2) The inclusion functor .I : Ab → Grp is fully faithful.
(3) The quotient functor .Π : C → C/ ∼ is always full, where .∼ is a congruence on
the category .C and .C/ ∼ denotes the corresponding quotient category.
(4) The quotient functor .Π : Top → HTop is full but not faithful.
(5) The inclusion functor .I : Cgrp → C is faithful but not full unless .C itself is
a groupoid, where .Cgrp is the core groupoid of the category .C constructed in
Example 1.3.6, (2).
(6) Given .f ∈ HomRingc (A, B), the restriction of scalars functor .Ff : B M →
A M defined in Example 1.5.3, (32) is obviously faithful. Furthermore, if f is
an epimorphism in .Ringc then the corresponding restriction of scalars functor
is also full ([53, Proposition XI.1.2]). Indeed, let M, .N ∈ Ob B M and .u ∈
1.6 Isomorphisms of Categories 37

HomA M (Ff (M), Ff (N )), i.e., for all .a ∈ A and .m ∈ M we have

 
u f (a)m = f (a)u(m).
. (1.13)

Note that throughout this example all module actions will be denoted by
juxtaposition and we will see B as an A-module via f , i.e., .ab = f (a)b for
all .a ∈ A and .b ∈ B.
Now let .m ∈ M and consider the map .v : B ⊗A B → N defined for all b,
.b ∈ B as follows:

v(b ⊗A b ) = b u(bm).
. (1.14)

It can be easily seen that v is well-defined; indeed, for instance we have

  (1.14)   (1.13)
v(ab ⊗A b ) = v f (a)b ⊗A b
. = b u f (a)bm = f (a)b u(bm)

(1.14) 
= v b ⊗A f (a)b = v(b ⊗A ab ).

Moreover, consider .α, .β ∈ HomRingc (B, B ⊗A B) defined for all .b ∈ B by

.α(b) = b ⊗A 1B , β(b) = 1B ⊗A b.

This shows that for all .a ∈ A we have

   
α f (a) = f (a) ⊗A 1B = a1B ⊗A 1B = 1B ⊗A a1B = 1 ⊗A f (a) = β f (a) .
.

As f was assumed to be an epimorphism, it follows that .b ⊗A 1B = 1B ⊗A b in

B ⊗A B for all .b ∈ B. This gives .u(bm) = v(b ⊗A 1B ) = v(1B ⊗A b) = bu(m).
.

We can now conclude that u is in fact a morphism in .B M and therefore .Ff is

full.
(7) The forgetful functor .U : Grp → Set is faithful but not full as not any function
between two given groups is a group homomorphism.
(8) The functor .U : Ring → Grp defined in Example 1.5.3, (27) is neither full nor
faithful. To start with, recall that .U (Z) = Z2 and .U (Fp ) = Zp−1 for any prime
number p, where .Fp denotes the field with p elements and .Zn is the group of
integers modulo n. Then, given an odd prime number p, the following induced
map is not surjective:

UZ, Fp : HomRing (Z, Fp ) → HomGrp (Z2 , Zp−1 ).

.

This follows easily by noticing that since .Z is the initial object in Ring, the
set .HomRing (Z, Fp ) has only one element while the cardinality of the set
.HomGrp (Z2 , Zp−1 ) is .gcd(2, p − 1) = 2 (see [27]).
38 1 Categories and Functors

In order to show that .U is not faithful either, consider the polynomial ring
k[X] over a field k and recall that .U (k[X]) = k\{0}. Then, the following
.

induced map is not injective:

UF2 [X], F2 [X] : HomRing (F2 [X], F2 [X]) → HomGrp ({1}, {1}),
.

where .{1} denotes the trivial group. Indeed, the set .HomGrp ({1}, {1}) obviously
has only one element while the cardinality of .HomRing (F2 [X], F2 [X]) is at
least two. .

Definition 1.6.3 A category .C is said to be concrete if there exists a faithful functor

F : C → Set.
.

Example 1.6.4 The categories FinSet, Grp, Ab, Rng, Ring, Top, .R M are all
concrete categories due to the existence of forgetful functors from any of the above
categories to Set, which are obviously faithful. .

In fact, we have a lot more examples of concrete categories, as can be seen from
the next result which, as noted in [6, Theorem 7.5.6], resembles Cayley’s theorem
from group theory.
Theorem 1.6.5 Any small category is concrete.
Proof For any small category .C we construct a faithful functor .F : C → Set as
follows. Given .C ∈ Ob C and .u ∈ HomC (C, C ) we define

.F (C) = {(Y, α) | Y ∈ Ob C, α ∈ HomC (Y, C)} ∈ Ob Set;

F (u) : F (C) → F (C ), F (u)(Y, α) = (Y, u ◦ α).

It is straightforward to see that F defined above is a functor; we only point out

that .F (C) is a set due to the fact that .C is a small category. We will prove that it
is faithful. Indeed, if .u1 , .u2 ∈ HomC (C, C ) such that .F (u1 ) = F (u2 ) we obtain
.(Y, u1 ◦ α) = (Y, u2 ◦ α) for any .(Y, α) ∈ F (C). Now for .(C, 1C ) ∈ F (C) we get

.u1 = u2 , as desired. 

Not every category admits a faithful functor to Set. The interested reader may
find such an example in [24], where it is shown that the homotopy category of
pointed spaces is not concrete.
Definition 1.6.6 A functor .F : C → D is called an isomorphism of categories if
there exists another functor .G : D → C such that .F ◦ G = 1D and .G ◦ F = 1C .
In this case we say that the categories .C and .D are isomorphic and G is called
the inverse of F . A contravariant isomorphism of categories is called an anti-
isomorphism of categories.
Let .F : C → D be a functor and for all A, .B ∈ Ob C consider the induced
mapping .FA, B defined in (1.12). If F is an isomorphism of categories with inverse
.G : D → C then each .FA, B is a bijective map with inverse given by .GF (A), F (B) ,
1.6 Isomorphisms of Categories 39

where
 
GF (A), F (B) : HomD F (A), F (B) → HomC (A, B), GF (A), F (B) (g) = G(g)
.

 
for all .g ∈ HomD F (A), F (B) .
In particular, an isomorphism of categories takes initial (resp. final) objects to
initial (resp. final) objects. Indeed, it follows from the above discussion that there is
a bijection between the sets .HomC (A, G(D)) and respectively .HomD (F (A), D);
hence, if A is an initial object in .C then .F (A) is an initial object in .D.
Examples 1.6.7
(1) The forgetful functor .F : Z M → Ab is an isomorphism of categories. Indeed,
the inverse of F is the functor .G : Ab → Z M defined by .G(M) = M, .G(u) =
u, where .M ∈ Ab has a left .Z-module structure as follows:
⎧
⎪
⎪
⎪
⎪  +m+
m · · · + m if t > 0
⎪
⎪
⎨ t times
.t · m = 0M if t = 0 , for every t ∈ Z, m ∈ M.
⎪
⎪
⎪ −m
⎪
⎪
⎪  − m
− · · · − m if t < 0
⎩
−t times

(2) Let R be a ring and denote by .R op the opposite ring.21 Then we have an
isomorphism of categories .F : R M → MR op given by

F (M) = M ∈ MR op via m ∗ r = rm, for all m ∈ M, r ∈ R;

.

F (u) = u,

with the inverse constructed in the same manner.

(3) The categories Set and .Setop are not isomorphic. Indeed, recall that .Set has
one initial object, namely the empty set, and infinitely many final objects, the
singletons. Therefore, in .Setop we have infinitely many initial objects and one
final object. The conclusion now follows since a potential isomorphism between
the two categories should take initial (final) objects to initial (final) objects.
(4) Let .C be an arbitrary category. The opposite functor .OC : C → Cop is an anti-
isomorphism of categories. .

Definition 1.6.8 Let .F : C → D be a functor.

(1) We say that F preserves a property P of morphisms if whenever f has the
property P in .C so does .F (f ) in .D.

21 In .R op we have .(R op , +) = (R, +) and the multiplication is given by .r ·op r = r r, for all r,
.r ∈ (R op , +).
40 1 Categories and Functors

(2) Similarly, F reflects a property P of morphisms if whenever .F (f ) has the

property P in .D so does f in .C.
Proposition 1.6.9 The following hold:
(1) Any functor preserves isomorphisms.
(2) Any fully faithful functor reflects isomorphisms.
(3) Any fully faithful functor reflects initial and final objects.
(4) Any faithful functor reflects monomorphisms and epimorphisms.
Proof
(1) Let .f ∈ HomC (A, B) be an isomorphism and .g ∈ HomC (B, A) its inverse.
Hence, we have .f ◦ g = 1B and .g ◦ f = 1A . Applying the functor .F : C → D
to these identities we obtain .F (f ) ◦ F (g) = 1F (B) and .F (g) ◦ F (f ) = 1F (A) ,
i.e., .F (f ) is an isomorphism.
(2) Let .F : C → D be a fully faithful functor and .f ∈ HomC (A, B) such that
.F (f ) is an isomorphism in .D. Then, there exists an .h ∈ HomD (F (B), F (A))

such that

F (f ) ◦ h = 1F (B) and h ◦ F (f ) = 1F (A) .

.

Since F is full we can find .g ∈ HomC (B, A) such that .F (g) = h. Therefore,
the above identities come down to

1F (B) = F (f ) ◦ F (g) = F (f ◦ g) and 1F (A) = F (g) ◦ F (f ) = F (g ◦ f ).

.

Now F is faithful and .F (1A ) = 1F (A) , respectively .F (1B ) = 1F (B) , yield

f ◦ g = 1B and .g ◦ f = 1A , as desired.
.

(3) Let .F : C → D be a fully faithful functor and I , .C ∈ Ob C. We have the

following bijection of sets:

HomC (I, C) ∼
. = HomD (F (I ), F (C)).

If .F (I ) is an initial object in .D then .1 = |HomD (F (I ), F (C))| for any .C ∈

Ob C and the above isomorphism gives .1 = |HomC (I, C)|, i.e., I is an initial
object in .C. The last statement follows in a similar manner.
(4) Let .F : C → D be a faithful functor and .f ∈ HomC (A, B) such that .F (f ) is
a monomorphism in .D. Consider now .C ∈ Ob C and g, .h : C → A such that
.f ◦ g = f ◦ h. Applying F to this equality gives .F (f ) ◦ F (g) = F (f ) ◦ F (h).

Since .F (f ) is a monomorphism this implies that .F (g) = F (h) and by the

faithfulness of F we get .g = h as desired. A similar argument proves that F
reflects epimorphisms as well.


1.7 Natural Transformations: Representable Functors 41

Examples 1.6.10
(1) The forgetful functor .U : Grp → Set reflects isomorphisms. Indeed, recall that
a group homomorphism is an isomorphism if and only if it is a bijection.
(2) The forgetful functor .U : Top → Set does not reflect isomorphisms, as can be
easily seen from Example 1.3.2, (7). .

Corollary 1.6.11 Let .C be a concrete category, .F : C → Set the corresponding

faithful functor and .f ∈ HomC (A, B).
(1) If .F (f ) is injective then f is a monomorphism in .C.
(2) If .F (f ) is surjective then f is an epimorphism in .C.
Proof 1) We have already proved that monomorphisms in Set coincide with
injective maps (see Example 1.3.2, (1)). Since .F (f ) is a monomorphism in Set
and F is faithful it follows by Proposition 1.6.9, (4) that f is also a monomorphism.
The second statement follows by similar arguments. 

Remark 1.6.12 Each of the following categories Grp, Ab, Rng, Ring, .Ringc , Div,
.R M, Top allow for a forgetful functor into Set. Hence, we can conclude that in the

above mentioned categories all injective maps are monomorphisms and respectively
all surjective maps are epimorphisms. However, the converse is not necessarily true,
as can be seen from Example 1.3.2, (4) and (5) respectively.

1.7 Natural Transformations: Representable Functors

Natural transformations are in some sense morphisms between functors, as we will

see in Proposition 1.9.1. The naturality of a certain transformation is meant to be
understood in the sense that its definition does not depend on any arbitrary choices
such as choosing a basis, a set of generators, etc.
Definition 1.7.1 Let F , .G : C → D be two functors. A natural transformation
α : F → G consists of a family of morphisms .αC : F (C) → G(C) in .D, indexed
.

by .C ∈ Ob C, such that for every morphism .f ∈ HomC (C, C ) the following

diagram is commutative:

αC
F (C) G(C) i.e., α C F(f) G( f ) α C .

F( f ) G( f )

F (C ) G(C )
αC
. (1.15)

If all components .αC are isomorphisms then .α : F → G is called a natural

isomorphism. In this case we say that the functors F and G are naturally isomorphic
42 1 Categories and Functors

and we use the notation .F ∼

= G. We denote by .Nat(F, G) the class of all natural
transformations between the functors F and G.
Examples 1.7.2
(1) Let .G1 and .G2 be two groups, .G1 , respectively .G2 , the corresponding categories
(see Example 1.2.2, (3)) and F , .H : G1 → G2 two functors. We denote by .fF ,
respectively .fH , the morphisms of groups from .G1 to .G2 corresponding to the
two functors F and H (see Example 1.5.3, (8)). Then .Nat(F, H ) is in bijective
correspondence with the set .{g ∈ G2 | fH (g ) = gfF (g )g −1 , for all g ∈
G1 }.
To this end denote .Ob G1 by .{∗} and .Ob G2 by .{}. Since .F (∗) =  and
.H (∗) = , any natural transformation .ϕ : F → H is completely determined by

a morphism .ϕ∗ :  →  in .G2 (i.e., an element of the group .G2 ) which makes
the following diagram commute for all morphisms .t : ∗ → ∗ in .G1 (i.e., an
element of the group .G1 ):

Having in mind that the composition of morphisms in .G2 is given by the

multiplication of the group .G2 , we can conclude that the natural transformations
from F to H are in bijection with elements .g ∈ G2 such that for any .g ∈ G1
we have .fH (g )g = gfF (g ). Hence, we have an isomorphism of sets between
.Nat(F, H ) and .{g ∈ G2 | fH (g )g = gfF (g ), for all g ∈ G1 }, as desired.

(2) The concept of a natural transformation, respectively a natural isomorphism, is

very well illustrated by looking at double duals of vector spaces. Recall that
a classical algebraic result states that any finite-dimensional vector space is
naturally isomorphic to its double dual. Loosely speaking, this means precisely
that by putting together the isomorphisms .V → V ∗∗ for all finite-dimensional
vector spaces V , we obtain a natural isomorphism in the sense of the above
definition. In more rigorous categorical terms, this comes down to defining a
natural transformation between the identity functor on the category of vector
spaces .K M and the double dual functor introduced in Example 1.5.3, (15). To
this end, let .η : 1K M → (−)∗∗ be the natural transformation whose components
.ηV : V → V
∗∗ for a given vector space V are the K-linear maps defined as

follows:

ηV (v)(f ) = f (v) for all v ∈ V and f ∈ V ∗ .

. (1.16)
1.7 Natural Transformations: Representable Functors 43

We will show that for any .u ∈ HomK M (V , W ) the following diagram is

commutative:

ηV
V V

u u

W W
.
ηW

Indeed, for all .v ∈ V and .f ∈ W ∗ we have

 ∗∗  (1.5)    
.u ◦ ηV (v) (f ) = ηV (v) ◦ u∗ (f ) = ηV (v) u∗ (f )
(1.16)    
= u∗ (f )(v) = f ◦ u (v) = f u(v)
(1.16)   
= ηW u(v) (f ) = (ηW ◦ u)(v) (f ).

This shows that .η : 1K M → (−)∗∗ is indeed a natural transformation.

Furthermore, if we restrict the two functors, the identity functor and the
double dual functor, to the category of finite-dimensional vector spaces .K Mf d
we obtain the natural isomorphism .η : 1K Mf d → (−)∗∗ mentioned in the
beginning.
(3) In contrast, the identity functor .1K Mf d and the dual functor .(−)∗ : K Mf d →
KM
fd
introduced in Example 1.5.3, (14) are not naturally isomorphic. In fact,
strictly speaking, the question itself of whether there are natural transformations
between the aforementioned functors does not make sense, as the dual functor
is contravariant while the identity functor is covariant.
(4) For any functor .F : C → D we have a natural transformation  .1F : F → F

called the identity natural transformation defined by .1F = 1F (C) C∈Ob C .
(5) If F , G, .H : C → D are functors and .α : F → G, .β : G → H are natural
transformations then we can define a new natural transformation .β ◦α : F → H
by the formula

.(β ◦ α)C : F (C) → H (C), (β ◦ α)C = βC ◦ αC for all C ∈ Ob C. (1.17)

The composition defined above is called the vertical composition of natural

transformations.
(6) If F , .G : C → D are functors and .α : F → G is a natural isomorphism then
.α
−1 : G → F defined by

 −1
αC−1 = αC
. for all C ∈ Ob C

is also a natural isomorphism, called the inverse natural transformation.

44 1 Categories and Functors

(7) Let F , .G : C → D, .H : D → E be two functors and .α : F → G a natural

transformation as in the picture below:

. G

Then we can define a new natural transformation .H α : H F → H G as follows:

(H α)C = H (αC ) for all C ∈ Ob C,

.

called the whiskering of the natural transformation .α on the right by H .

Similarly, let .K : B → C, F , .G : C → D be functors and .α : F → G a
natural transformation as in the picture below:

K
α

. G

We can define a new natural transformation .αK : F K → GK as follows:

(αK)B = αK(B) for all B ∈ Ob B,

.

called the whiskering of the natural transformation .α on the left by the functor
K.
Furthermore, if .α is a natural isomorphism then both .H α and .αK are natural
isomorphisms. Indeed, the first assertion follows by Proposition 1.6.9, (1) while
the second one is an easy consequence of .α itself being a natural isomorphism.
.

The examples above describing the whiskering of a natural transformation are

both special cases of a more general construction called the Godement product or
horizontal composition of natural transformations.
1.7 Natural Transformations: Representable Functors 45

Proposition 1.7.3 Let F , .G : C → D, H , .K : D → E functors and .α : F → G,

β : H → K two natural transformations as depicted below:
.

F H

. G K

Then .β ∗ α : H F → KG defined as follows for all .C ∈ Ob C:

. (1.18)

is a natural transformation called the Godement product of .α and .β.

Proof To start with, the naturally of .β applied to the morphism .αC : F (C) → G(C)
yields the following commutative diagram:

Therefore, we have .(β ∗ α)C = βG(C) ◦ H (αC ) = K(αC ) ◦ βF (C) for all .C ∈ Ob C.
Let .f ∈ HomC (C, C ); showing that .β ∗ α is a natural transformation comes down
to proving the commutativity of the following diagram:

. (1.19)
46 1 Categories and Functors

To this end, the naturality of .α and functoriality of H render the following diagram
commutative:

(1.20)

Moreover, the commutativity of the following diagram follows by the naturality of

β:
.

(1.21)

Putting all the above together yields

(1.20)
βG(C ) ◦ H (αC ) ◦ H F (f ) = βG(C ) ◦ H G(f ) ◦ H (αC )
.

(1.21)
= KG(f ) ◦ βG(C) ◦ H (αC ),

which proves that (1.19) holds and the proof is now finished. 


As we will see in Chap. 3, naturality in the case of bifunctors turns out to be

important when dealing with adjoint functors. For this reason, we conclude our
discussion on natural transformations with the following result, which shows that
naturality for bifunctors is equivalent to the naturality of both the left and the right
associated functors as defined in Proposition 1.5.5.
Proposition 1.7.4 Let F , .G : A×B → C be two bifunctors. A family of morphisms
α(A, B) : F (A, B) → G(A, B) in .C, indexed by .(A, B) ∈ Ob (A × B), form
.

a natural transformation .α : F → G if and only if the following conditions are

fulfilled:
(1) for all .A0 ∈ Ob A, the family of morphisms .(αA0 )B : FA0 (B) → GA0 (B) in .C,
indexed by .B ∈ Ob B, form a natural transformation .αA0 : FA0 → GA0 , where
.(αA0 )B = α(A0 ,B) ;
1.7 Natural Transformations: Representable Functors 47

(2) for all .B0 ∈ Ob B, the family of morphisms .(α B0 )A : F B0 (A) → GB0 (A) in .C,
indexed by .A ∈ Ob A, form a natural transformation .α B0 : F B0 → GB0 , where
.(α 0 )A = α(A,B0 ) .
B

 that .α : F → G
Proof Assume first  is a natural transformation; therefore, for all
(f, g) ∈ HomA×B (A, B), (A , B ) the following diagram is commutative:
.

(1.22)

Now considering .A = A and .f = 1A in (1.22) yields .α(A,B ) ◦ F (1A , g) =

G(1A , g) ◦ α(A,B) . The last compatibility is equivalent to .(αA )B ◦ FA (g) =
GA (g) ◦ (αA )B , which in turn implies that the following diagram is commutative
for all .g ∈ HomB (B, B ):

. (1.23)

This shows that .αA : FA → GA is a natural transformation for all .A ∈ Ob A.

Similarly, by setting .B = B and .g = 1B in (1.22)
 we obtain .α(A ,B) ◦ F (f,
 1B ) =
G(f, 1B )◦α(A,B) . Consequently, we have . α B A ◦F B (f ) = GB (f )◦ α B A , which
implies the commutativity of the following diagram for all .f ∈ HomB (A, A ):

. (1.24)

Hence .α B : F B → GB is a natural transformation for all .B ∈ Ob B.

Conversely, assume that the conditions 1) and 2) in the statement hold; thus for
all .f ∈ HomB (A, A ) and .g ∈ HomB (B, B ), the compatibilities
 (1.23) and (1.24)
are fulfilled. Consider .(u, v) ∈ HomA×B (X, Y ), (X , Y ) ; we have
48 1 Categories and Functors

 
G(u, v) ◦ α(X,Y ) = G (u, 1Y ) ◦ (1X , v) ◦ α(X,Y )
.

= G(u, 1Y ) ◦ G(1X , v) ◦ α(X,Y )

= GY (u) ◦ GX (v) ◦ α(X,Y )

 
= GY (u) ◦ GX (v) ◦ αX Y
(1.23)
= GY (u) ◦ (αX )Y ◦ FX (v)

= GY (u) ◦ (α Y )X ◦ FX (v)
(1.24)  Y
= α X ◦ F Y (u) ◦ FX (v)
= α(X ,Y ) ◦ F (u, 1Y ) ◦ F (1X , v)
= α(X ,Y ) ◦ F (u, v).
 
This shows that for all .(u, v) ∈ HomA×B (X, Y ), (X , Y ) the following diagram
is commutative:

and we can conclude that .α : F → G is a natural transformation, as desired. 


We are now ready to introduce an important concept: representable functors.
Definition 1.7.5 Let .C be a category. We say that a functor .F : C → Set is
representable if there exist .C ∈ Ob C and a natural isomorphism .F ∼
= HomC (C, −).
Similarly, a contravariant functor .G : C → Set is representable if there exist
.C ∈ Ob C and a natural isomorphism .G ∼ = HomC (−, C). In this case, C is called
the representing object of F .
More precisely, Definition 1.7.5 reads as follows: .F : C → Set is representable
if and only if there exist an object .C ∈ Ob C and a family of isomorphisms
 
. αA : HomC (C, A) → F (A) in Set (set bijections) such that for any
A∈Ob C
f ∈ HomC (X, Y ) the following diagram is commutative:
.
1.7 Natural Transformations: Representable Functors 49

Note that the representing object of a functor will prove to be unique up to

isomorphism as a consequence of Yoneda’s lemma (see Proposition 1.10.5). This
explains why the object C in Definition 1.7.5 is referred to as the representing object
of F .
Examples 1.7.6
(1) The forgetful functor .U : Grp → Set is representable and the representing
object is the abelian group .(Z, +). Indeed, for any .X ∈ Ob Grp and any .g ∈
HomGrp (Z, X) we define .αX : HomGrp (Z, X) → U (X) = X by .αX (g) =
g(1). Note that each .αX is a bijection as any group homomorphism .g : Z →
X is uniquely defined by its value in 1. The above diagram is now obviously
commutative for any .f ∈ HomGrp (X, Y ):


Indeed,
 for any .g ∈ HomGrp (Z, X) we have .αY ◦ HomGrp (Z, f )(g) = αY f ◦
g = (f ◦ g)(1) and .f ◦ αX (g) = f (g(1)).
(2) The forgetful functor .U : Top → Set is representable and the representing
object is any singleton topological space .{x0 } (with the discrete topology).
To this end, for any .X ∈ Ob Top and any .h ∈ HomTop ({x0 }, X) we define
.αX : HomTop ({x0 }, X) → U (X) = X by .αX (h) = h(x0 ) ∈ X. Each .αX is

obviously bijective. Furthermore, it can be easily seen that the above diagram is
commutative for any .f ∈ HomTop (X, Y ):

.
50 1 Categories and Functors

Indeed, for any .h ∈ HomTop ({x0 }, X) we have

f ◦ αX (h) = f (h(x0 ))
.
 
αY ◦ HomTop ({x0 }, f )(h) = αY f ◦ h = (f ◦ h)(x0 ).

Therefore, .f ◦ αX = αY ◦ HomTop ({x0 }, f ), as desired.

(3) The constant functor .Fx0 : C → Set which sends every object of .C to
the singleton .{x0 } and every morphism in .C to the identity map on .{x0 } is
representable if and only if .C has an initial object. Moreover, in this case the
representing object is the initial object. Indeed, suppose the functor .Fx0 is
represented by .I ∈ Ob C. Then, for any .C ∈ Ob C we have an isomorphism
in Set denoted by .αC : HomC (I, C) → {x0 }. This implies that .HomC (I, C)
has exactly one element for any .C ∈ Ob C, i.e., I is initial in .C.
(4) The covariant power-set functor .P : Set → Set is not representable. To this
end, assume that .P is representable; consider A to be the representing object
and .τ : HomSet (A, −) → P the corresponding natural isomorphism. Let
.{∗} be an arbitrary singleton. Since .τ is a natural isomorphism we obtain a

bijective map .τ : HomSet (A, {∗}) → P({∗}). This leads to a contradiction since
.|HomSet (A, {∗})| = 1 and .|P({∗})| = 2. Therefore, .P is not representable. .

The following result provides an important criterion for deciding whether a

functor is representable or not.
Proposition 1.7.7 Let .F : C → Set be a functor. Then F is representable if and
only if there exists a pair .(A, a) with .A ∈ Ob C and .a ∈ F (A) satisfying the
following property: for any other pair .(B, b) with .B ∈ Ob C and .b ∈ F (B) there
exists a unique morphism .f ∈ HomC (A, B) such that .F (f )(a) = b. In this case
.(A, a) is called a representing pair.

Proof Suppose first that F is representable, i.e., there exists a natural isomorphism
.ϕ : HomC (A, −) → F for some .A ∈ Ob C. In particular we have a bijection of sets
.ϕA : HomC (A, A) → F (A) and we denote .ϕA (1A ) ∈ F (A) by a. We will show

that .(A, a) is a representing pair. Indeed, consider .B ∈ Ob C and .b ∈ F (B). As

before, we have a bijective map .ϕB : HomC (A, B) → F (B) so there exists a unique
.f ∈ HomC (A, B) such that .ϕB (f ) = b. We are left to prove that .F (f )(a) = b.

Since .ϕ is a natural transformation, the following diagram is commutative:

.
1.7 Natural Transformations: Representable Functors 51

−1
i.e., .ϕB ◦ HomC (A, f ) = F (f ) ◦ ϕA . This yields .F (f ) = ϕB ◦ HomC (A, f ) ◦ ϕA
and we obtain
−1
F (f )(a) = ϕB ◦ HomC (A, f ) ◦ ϕA
. (a)
= ϕB ◦ HomC (A, f )(1A )
 
= ϕB f ◦ 1A
= ϕB (f ) = b.

Assume now that .(A, a) is a representing pair. Let .ψ : HomC (A, −) → F be the
natural isomorphism defined as follows for any .B ∈ Ob C:

ψB : HomC (A, B) → F (B), ψB (f ) = F (f )(a), f ∈ HomC (A, B).

.

The property assumed to be satisfied by .(A, a) implies that each such map .ψB is
bijective. The proof will be finished once we show that .ψ is a natural transformation,
i.e., for any .g ∈ HomC (B, C) the following diagram is commutative:

Indeed, for any .h ∈ HomC (A, B) we have

   
F (g) ψB (h) = F (g) F (h)(a) = F (g ◦h)(a) = ψC (g ◦h) = ψC ◦HomC (A, g)(h),
.

as desired. This finishes the proof. 


Example 1.7.8 Let A be a group and .H  A a normal subgroup. Consider the
functor .F : Grp → Set defined as follows:

F (G) = {f ∈ HomGrp (A, G) | H ⊆ kerf },

.

F (u)(g) = u ◦ g,

for any .G ∈ Ob Grp and any .u ∈ Hom  Grp (G, G ), .g ∈ F (G). Then F is
representable and . A/H, π : A → A/H is the representing pair, where .π is
the canonical projection. Indeed, consider another pair .(G, f : A → G) with
52 1 Categories and Functors

G ∈ Ob Grp and .f ∈ F (G):

.

π
A A/H

f
f
.
G

Since .H ⊆ kerf , the universal property of the quotient group .A/H yields a unique
f ∈ HomGrp (A/H, G) such that the above diagram is commutative, i.e., .f ◦π = f .
.

The last equality is equivalent to .F (f )(π ) = f and the desired conclusion now
follows from Proposition 1.7.7. .

1.8 The Duality Principle

The dual category allows us not only to define a dual notion for every concept but
also to state a dual result for any theorem. This new result requires no proof and
is obtained by reversing all morphisms and consequently the order of composition
in the given theorem. Indeed, if a given statement T is valid in any category then
the dual statement .T op is also valid in any category. This can be easily seen by
noticing that proving the statement .T op in a category .C is equivalent to proving the
statement T in the category .Cop , which is assumed to be valid. To illustrate this
duality principle we look at the following statement proved in an equivalent form in
Proposition 1.3.9:
When it exists, the initial object of a category is unique up to isomorphism.

By simply applying the duality principle we obtain the following dual statement:
When it exists, the final object of a category is unique up to isomorphism.

A certain care is required, however, when dealing with statements which involve
functors. More precisely, in the process of dualizing these statements, all categories
are replaced by their duals, and all morphisms are reversed, while functors .C → D
are not reversed but replaced by dual functors .Cop → Dop as defined below.
Proposition 1.8.1 Let .F : C → D be a functor. There exists a functor .F op : Cop →
Dop , called the dual functor, defined as follows for any C, .D ∈ Ob Cop and any
.f
op ∈ Hom op (C, D):
C

F op (C) = F (C),
. F op (f op ) = F (f )op .

Furthermore, .(F op )op = F and if .G : D → E is another functor we have

(G ◦ F )op = Gop ◦ F op ,
. (1.25)
1.8 The Duality Principle 53

where the functor composition in (1.25) is defined as in Example 1.5.3, (3).

op op
Proof For any .C ∈ Ob Cop we have .F op (1C ) = F (1C )op = 1F (C) and
therefore .F op preserves unit morphisms. Furthermore, for all .f op ∈ HomCop (C, D),
.g
op ∈ Hom op (D, E) we have
C
   
F op g op ◦op f op = F op (f ◦ g)op = F (f ◦ g)op
.
 op
= F (f ) ◦ F (g) = F (g)op ◦op F (f )op ,

which shows that .F op is indeed a functor.

Furthermore, we have
   
(G ◦ F )op (C) = (G ◦ F )(C) = G F (C) = Gop F op (C)
.
 
= Gop ◦ F op (C) and
 op  op  
(G ◦ F )op (f op ) = (G ◦ F )(f ) = G(F (f )) = Gop F (f )op
   
= Gop F op (f op ) = Gop ◦ F op (f op ).

Therefore, (1.25) holds as well. 


Consider now the following statement proved in Proposition 1.6.9, (4):
Any faithful functor reflects monomorphisms.

We have already established that a morphism f of a given category .C is a

monomorphism if and only if .f op is an epimorphism in .Cop . Moreover, it can be
easily seen that a functor F is faithful (resp. full) if and only if the dual functor .F op
is faithful (full). Therefore, the duality principle shows that the following statement
also holds:
Any faithful functor reflects epimorphisms.

In light of the duality principle, any statement about covariant functors has a
correspondent for contravariant functors obtained by replacing the given functor
.F : C → D by .F ◦ OCop : C
op → D (or .O ◦ F : C → Dop ). For example, in order
D
to obtain the contravariant version of Proposition 1.7.7 we consider .G : C → Set
to be a contravariant functor. Then .G ◦ OCop : Cop → Set is a covariant functor
and Proposition 1.7.7 implies that .G ◦ OCop : Cop → Set is representable if and
only if there exists a pair .(A, a) with .A ∈ Ob Cop and .a ∈ G ◦ OCop (A)
satisfying the following property: for any other pair .(B, b) with .B ∈ Ob Cop and
.b ∈ G ◦ OCop (B) there exists a unique .f
op ∈ Hom op (A, B) such that
C
.G ◦ OCop (f )(a) = b. Since .Ob C = Ob C and .HomCop (A, B) = HomC (B, A)
op op

the statement concerning contravariant functors is the following:

Corollary 1.8.2 A contravariant functor .G : C → Set is representable if and only
if there exists a pair .(A, a) with .A ∈ Ob C and .a ∈ G(A) satisfying the following
54 1 Categories and Functors

property: for any other pair .(B, b) with .B ∈ Ob C and .b ∈ G(B) there exists a
unique .f ∈ HomC (B, A) such that .G(f )(a) = b.
Finally, the dual of a natural transformation is defined as follows:
Proposition 1.8.3 Let F , .G : C → D be two functors and .ψ : F → G a natural
transformation. There exists a natural transformation .ψ op : Gop → F op , called the
opposite or dual natural transformation, defined as follows for all .C ∈ Ob C:
 op   op
. ψ C = ψC (1.26)

and any natural transformation between .Gop and .F op appears as the dual of some
natural transformation between F and G. Furthermore, .ψ is a natural isomorphism
if and only if .ψ op is a natural isomorphism.
Proof For any .f op ∈ HomCop (C, C ), the naturality of .ψ renders the following
diagram commutative:

. (1.27)

We will show that this implies the naturality of .ψ op . Indeed, (1.27) takes the
following equivalent forms:

ψC ◦ F (f ) = G(f ) ◦ ψC
.
 op  op
⇔ ψC ◦ F (f ) = G(f ) ◦ ψC
⇔ F (f )op ◦op (ψC )op = (ψC )op ◦op G(f )op
   
⇔ F op f op ◦op (ψC )op = (ψC )op ◦op Gop f op ,

which shows that the following diagram is commutative:

and therefore .ψ op : Gop → F op is a natural transformation.

1.8 The Duality Principle 55

The second claim follows by noticing that for any natural transformation we have
(ψ op )op = ψ.
. 

Numerous examples of duality arguments will be used in proofs throughout the
book.

Comma Categories

Comma categories are constructed using two functors with the same codomain. We
will see many instances of comma categories at work later on when dealing with
(co)limits or adjoint functors.
Theorem 1.8.4 Any two functors .F : A → C and .G : B → C define a comma
category denoted by .(F ↓ G) as follows:
(1) the objects are triples .(A, f, B) consisting of two objects .A ∈ Ob A, .B ∈ Ob B
and a morphism .f ∈ HomC (F (A), G(B));
(2) a morphism in .(F ↓ G) from .(A, f, B) to .(A , f , B ) is a pair .(a, b), where
.a ∈ HomA (A, A ), .b ∈ HomB (B, B ) such that the following diagram is

commutative:

. (1.28)

(3) the composition law in .(F ↓ G) is that induced by the composition laws of .A
and .B, i.e.:

.(a, b) ◦ (a , b ) = (a ◦ a , b ◦ b );

(4) the identities are .1(A, f, B) = (1A , 1B ).

Proof First note that for any .(A, f, B) ∈ Ob (F ↓ G), the following diagram is
trivially fulfilled:

.
56 1 Categories and Functors

Furthermore, if .(A, f, B), .(A , f , B ), .(A , f , B ) ∈ Ob (F ↓ G) and .(a, b),

(a , b ) are morphisms in .(F ↓ G) between .(A, f, B) and .(A , f , B ) and respec-
.

tively .(A , f , B ) and .(A , f , B ), the following diagrams are commutative:

Since F and G are functors and as a consequence they respect composition, we

obtain the commutativity of the following diagram:



Examples 1.8.5
(1) Let .A and .B be discrete categories with only one object, say .Ob A = {A} and
.Ob B = {B}. As we have seen in Example 1.5.3, (7), defining two functors

.F : A → C and .G : B → C comes down to choosing two objects C and .C in

.C. Then, the comma category .(F ↓ G) is isomorphic to the discrete category on

the set .HomC (C, C ). Indeed, the objects of the corresponding comma category
.(F ↓ G) are triples .(A, f, B), where .f ∈ HomC (C, C ). The morphisms in

.(F ↓ G) between two objects .(A, f, B) and .(A, f , B) are pairs .(a, b) where

.a ∈ HomA (A, A) and .b ∈ HomB (B, B). Now since the only morphisms in

.A and .B are identities on A and B respectively, we get .(a, b) = (1A , 1B ). So

the only morphisms in .(F ↓ G) are the identities on each object. Therefore
the functor from .(F ↓ G) to the discrete category on .HomC (C, C ) sending
any object .(A, f, B) to f and any morphism .(1A , 1B ) to the identity on f is
obviously an isomorphism of categories.
(2) Let .A = 1 be a discrete category with only one object, say A, and .B = C =
Top. Furthermore, consider the functors .F : 1 → Top, .G : Top → Top defined
as follows:

F (A) = {},
. F (1A ) = 1{} , G = 1Top ,
1.8 The Duality Principle 57

where .{} denotes a singleton set regarded as a topological space in the obvious
way. Then, the comma category .(F ↓ G) is isomorphic to the category of
pointed topological spaces .Top . To this end, the objects of the corresponding
comma category .(F ↓ G) are triples .(A, f, X), where .X ∈ Ob Top and
.f ∈ HomTop ({}, X). The morphisms in .(F ↓ G) between two objects

.(A, f, X) and .(A, g, Y ) are pairs .(a, b), where .a ∈ HomA (A, A) and

.b ∈ HomTop (X, Y ) such that diagram (1.28) is commutative. Now since the

only morphism in .A is the identity on A, we get .(a, b) = (1A , b), where

.b ∈ HomTop (X, Y ) fulfils the following compatibility:

. (1.29)

Therefore, the functor .H : (F ↓ G) → Top defined as follows for all

(A, f, X), .(A, g, Y ) ∈ Ob(F ↓ G) and all morphisms .(1A , b) in .(F ↓ G)
.

between .(A, f, X) and .(A, g, Y ):

 
H (A, f, X) = X, f () ,
. H (1A , b) = b,

is an isomorphism of categories. Indeed, note that the commutativity of (1.29)

implies that b is indeed a morphism in .Top between .(X, f ()) and .(Y, g()).
Furthermore, since any morphism .f ∈ HomTop ({}, X) is uniquely determined
by .f () we can conclude that H is an isomorphism of categories. .

In what follows we write down, for further use, two important special cases of
comma categories:
Corollary 1.8.6 Let .F : A → C and .G : B → C be two functors.
(1) If .A is the discrete category with only one object and .F : A → C is the constant
functor at .C0 ∈ Ob C then the comma category .(F ↓ G) = (C0 ↓ G) is
isomorphic to the category defined as follows:
(i) the objects are pairs .(f, B), where .B ∈ Ob B, .f ∈ HomC (C0 , G(B));
(ii) a morphism in .(C0 ↓ G) between two objects .(f, B) and .(f , B )
is a morphism .b ∈ HomB (B, B ) such that the following diagram is
58 1 Categories and Functors

commutative:

(iii) the composition of morphisms is given by the composition in .B and the

identities are .1(f, B) = 1B for all .B ∈ Ob B.
(2) Similarly, if .B is the discrete category with only one object and .G : B → C is the
constant functor at .C0 ∈ Ob C, then the comma category .(F ↓ G) = (F ↓ C0 )
is defined as follows:
(i) the objects are pairs .(A, f ), where .A ∈ Ob A, .f ∈ HomC (F (A), C0 );
(ii) a morphism in .(F ↓ C0 ) between two objects .(A, f ) and .(A , f )
is a morphism .a ∈ HomA (A, A ) such that the following diagram is
commutative:

(iii) the composition of morphisms is given by the composition in .A and the

identities are .1(A, f ) = 1A for all .A ∈ Ob A.
If we specialize the previous result further by considering the non-trivial functor
to be the identity then we obtain the so-called slice and coslice categories:
Corollary 1.8.7 Let .C be a category and .C0 ∈ Ob C.
(1) The category denoted by .(C0 ↓ C), called the category of objects under .C0 or
the coslice category with respect to .C0 , is defined as follows:
(i) the objects are pairs .(f, C), where .C ∈ Ob C, .f ∈ HomC (C0 , C);
(ii) a morphism in .(C0 ↓ C) between two objects .(f, C) and .(f , C )
is a morphism .h ∈ HomC (C, C ) such that the following diagram is
commutative:

C0 i.e., h f f
f f

C C
. h
1.8 The Duality Principle 59

(iii) the composition of morphisms is given by the composition in .C and the

identities are .1(f, C) = 1C for all .C ∈ Ob C.
(2) Similarly, .(C ↓ C0 ) is called the category of objects over .C0 or the slice
category with respect to .C0 and is defined as follows:
(i) the objects are pairs .(C, f ), where .C ∈ Ob C, .f ∈ HomC (C, C0 );
(ii) a morphism in .(C ↓ C0 ) between two objects .(C, f ) and .(C , f )
is a morphism .h ∈ HomA (C, C ) such that the following diagram is
commutative:
h
C C i.e., f h f

f f

.
C0

(iii) the composition of morphisms is given by the composition in .C and the

identities are .1(C, f ) = 1C for all .C ∈ Ob C.
As we already mentioned, various properties of functors can be characterized in
terms of certain related comma categories. One such property is representability:

 1.8.8  A functor .F : C → Set is representable if and only if the comma

Proposition
category . {} ↓ F has an initial object, where .{} denotes a singleton set.
Proof Assume first that .(A, a) is a representing pair of F , where .A ∈ Ob C and
a ∈ F (A). We will show that the pair .(fa , A), where .fa ∈ HomSet ({},
.
 F (A))

is defined by .fa () = a,
 is the initial
 object of the comma category . {} ↓ F .

Indeed, let .(g, B) ∈ Ob {} ↓ F , where .B ∈ Ob C and .g ∈ HomSet ({}, F (A)),

and consider .g() = b ∈ F (B). By Proposition 1.7.7, there exists a unique .u ∈
HomSet (A, B) such that .F (u)(a) = b. This is equivalent to u being the unique
morphism such that .F (u)◦fa () = g(), i.e., the following diagram is commutative:

 
In other words, u is the unique morphism in . {} ↓ F between .(fa , A) and .(g, B),
as desired.  
Conversely, let .(f, A) be the initial object of . {} ↓ F , where .A ∈ Ob C and
.f ∈ HomSet ({}, F (A)), and consider .a = f (). Then, .(A, a) is the representing

pair of F . Indeed, note that if .B ∈ Ob C and .b ∈ F (B) then .(fb , B) is an object in

. {} ↓ F , where .fb ∈ HomSet ({}, F (A)) is defined by .fb () = b. As .(f, A) is
60 1 Categories and Functors

 
theinitial object
 of . {} ↓ F , there exists a unique morphism .u : (f, A) → (fb , B)
in . {} ↓ F such that .F (u) ◦ fa = g. To conclude, there exists a unique morphism
.u ∈ HomSet (A, B) such that .F (u)(a) = b and therefore .(A, a) is the representing

pair of F by Proposition 1.7.7. 


Next we discuss certain properties of comma categories which are of interest for
the material developed in the sequel.
Lemma 1.8.9 Let .F : A → B be a functor and .A a small category. Then, for all
B ∈ Ob B, the comma-category .(F ↓ B) is small.
.

Proof The objects of .(F ↓ B) are of the form .(A, f ), where .A ∈ Ob A

and .f ∈ HomB (F (A), B). Now since .A is small, .Ob A is a set and
 conse-
quently . A∈Ob A Ob A × Hom(F (A), B) is also a set. As . A∈Ob A Ob A ×
Hom(F (A), B) contains .Ob(F ↓ B), we can conclude that .(F ↓ B) is a small
category. 

Proposition 1.8.10 Given two functors .F : A
op
→ C and .G : B → C, we have an
isomorphism of categories between . F ↓ G and . Gop ↓ F op .
 In particular,
op  for any .C ∈ Ob C we have an isomorphism of categories between
. C ↓ G and . Gop ↓ C .
Proof Throughout the proof we will freely  use the notation .◦op to denote the
op
composition of morphisms in .C , .D or . F ↓ G . It will, however, be obvious
op op

from the context which composition is used.

   
Note that the objects of the category . Gop ↓ F op are of the form . B, f op , A ,
where

.B ∈ Ob Bop = Ob B, A ∈ ObAop = Ob A and

f op ∈ HomCop (Gop (B), F op (A)) = HomCop (G(B), F (A)).
 
This also shows that .f ∈ HomC (F (A), G(B)). Therefore, we have . B, f op , A ∈
     op
Ob Gop ↓ F op if and only if . A, f, B ∈ Ob F ↓ G .   
op op
Furthermore,
 op a morphism
 between
 twoobjects . B1 , f1 , A1 and . B2 , f2 , A2
in . G ↓ F op is a pair . bop , a op , where .bop ∈ HomBop (B1 , B2 ) and
.a
op ∈ HomAop (A1 , A2 ) such that the following diagram is commutative:

(1.30)
1.8 The Duality Principle 61

In particular, we have .a ∈ HomA (A2 , A1 ) and .b ∈ HomB (B2 , B1 ). Further-

more, (1.30) can be equivalently written as follows:
op op
.F op (a op ) ◦op f1 = f2 ◦op Gop (bop )
op op
⇔ F (a)op ◦op f1 = f2 ◦op G(b)op
 op  op
⇔ f1 ◦ F (a) = G(b) ◦ f2
⇔ f1 ◦ F (a) = G(b) ◦ f2 .

This shows that (1.30) is commutative if and only if the following diagram is
commutative:

. (1.31)
 
Hence we have .(a, b) ∈ Hom  (F ↓G) (A2 , f2 , B2 ), (A1 , f1 , B1 ) . In fact, all the
above shows that . bop , a op ∈ HomGop ↓F op  (B1 , f1 , A1 ), (B2 , f2 , A2 if
op op
 
and only if .(a, b) ∈ Hom(F ↓G) (A2 , f2 , B2 ), (A1 , f1 , B1 ) .  
We
 opcan now  define two functors .H : (F ↓ G)op → Gop ↓ F op and
.T : G ↓ F op → (F ↓ G)op as follows:
   
. H (A, f, B) = B, f op , A ,
(A, f, B) ∈ Ob (F ↓ G)op ,
   
H (a, b)op = bop , a op ,
 
(a, b)op ∈ Hom(F ↓G)op (A1 , f1 , B1 ), (A2 , f2 , B2 ) ,
 
T (B, f op , A) = (A, f, B),
 
(B, f op , A) ∈ Ob Gop ↓ F op ,
 
T (bop , a op ) = (a, b)op ,
 
(bop , a op ) ∈ HomGop ↓F op  (B1 , f1 , A1 ), (B2 , f2 , A2 .
op op
62 1 Categories and Functors

The above discussion shows that both H and T are well-defined. We will show that
they are indeed functors. To start with, for all .A ∈ Ob A and .B ∈ Ob B we have
     op op 
H 1(A, f, B) = H (1A , 1B )op = 1B , 1A = 1B, f op , A ,
op
.

   op op 
T 1B, f op , A = T (1B , 1A ) = (1A , 1B )op = 1(A, f, B) .
op

op ∈ Hom
 
Consider now two morphisms  .(a, b) (F ↓G)op (A 1 , f1 , B1 ), (A2 , f2 , B2 )
and .(c, d)op ∈ Hom(F ↓G)op (A2 , f2 , B2 ), (A3 , f3 , B3 ) . We have
   op   op 
H (c, d)op ◦op (a, b)op = H (a, b) ◦ (c, d)
. = H a ◦ c, b ◦ d
       
= (b ◦ d)op , (a ◦ c)op = d op ◦op bop , cop ◦op a op = d op , cop ◦ bop , a op
   
= H (c, d)op ◦ H (a, b)op .

Finally, given two morphisms

 op op   
∈ HomGop ↓F op  (B1 , f1 , A1 ), (B2 , f2 , A2 and
op op
.u ,v
 op op   
∈ HomGop ↓F op  (B2 , f2 , A2 ), (B3 , f3 , A3
op op
r ,s

we have
   
T (r op , s op ) ◦ (uop , v op ) = T (r op ◦op uop ), (s op ◦op v op
.
 
= T (u ◦ r)op , (v ◦ s)op
 op  op
= v ◦ s, u ◦ r = (v, u) ◦ (s, r) = (s, r)op ◦op (v, u)op
   
= T (r op , s op ) ◦op T (uop , v op ) .

The proof is now finished as H and T are obviously inverses to each other. 


1.9 Functor Categories

As the name suggests, functor categories have functors as objects and natural
transformations as morphisms. The next result shows that this is indeed a category
under certain conditions.
Proposition 1.9.1 Let I and .C be two categories. If I is a small category then the
functors from I to .C and the natural transformations between them as morphisms
1.9 Functor Categories 63

form a category, called a functor category, which we denote by .Fun (I, C). If .C is
also small then .Fun (I, C) is small.
Proof The composition of natural transformations is given by the vertical composi-
tion as defined in Example 1.7.2, (5). This composition law is obviously associative
and the identity at each functor F is just the identity natural transformation defined
in Example 1.7.2, (4).  
Note that . i∈Ob I HomC F (i), G(i)
 is a union
 of sets indexed by another set,
namely .Ob I . Hence . i∈Ob I HomC F (i), G(i) is a set as well. Finally, note that
for any two functors F , .G : I → C, a natural transformation .η : F → G is
determined by a map
  
η : Ob I →
. HomC F (i), G(i) .
i∈Ob I

This shows that the class of all

 natural transformations
 between F and G is a subset
of .Ob I × i∈Ob I Hom C F (i), G(i) . Therefore, the natural transformations
between any two functors F , .G : I → C form a set and the proof is now finished.


Examples 1.9.2
(1) For any small category .C, a functor .F : Cop → Set is called a (set-valued)
presheaf on .C and the functor category .Fun(Cop , Set) is known as the category
of presheaves on .C.
(2) Let .2 denote the discrete category with two objects, say .∗ and .. Then, for any
category .C the functor category .Fun(2, C) is isomorphic to the product category
.C × C. Indeed, first note that any functor .F : 2 → C is uniquely determined by

two objects .CF , .DF ∈ Ob C such that

F (∗) = CF , F () = DF , F (1∗ ) = 1CF , F (1 ) = 1DF .

.

Consider now a natural transformation .η : F → H , where F , .H : 2 → C are

functors. Then .η is uniquely determined by two morphisms in .C:

η ∗ : CF → CH , η  : D F → D H .
.

Note that since there are no non-identity morphisms in .2 we have no non-trivial

naturality diagram to impose additional conditions for the two morphisms .η∗
and .η . Therefore, the functor .V : Fun(2, C) → C × C defined below is an
obvious isomorphism of categories:

V (F ) = (CF , DF ), V (η) = (η∗ , η ),

.

where F , .H : 2 → C are functors and .η : F → H is a natural transformation.


.
64 1 Categories and Functors

The isomorphisms in a functor category can be easily characterized:

Proposition 1.9.3 Let I be a small category, .C an arbitrary category and F ,
G : I → C two functors. Then a natural transformation .η : F → G is an
.

isomorphism in .Fun(I, C) if and only if .η is a natural isomorphism.

Proof Assume first that .η is an isomorphism in .Fun(I, C), i.e., there exists another
natural transformation .ξ : G → F such that .η ◦ ξ = 1G and .ξ ◦ η = 1F . Hence, for
any .i ∈ Ob I we have .ηi ◦ ξi = 1G(i) and .ξi ◦ ηi = 1F (i) . This proves that .ηi is an
isomorphism for all .i ∈ Ob I and therefore .η is a natural isomorphism.
Conversely, suppose now that .η is a natural isomorphism. In particular,
.ηi : F (i) → G(i) is an isomorphism in .C for all .i ∈ Ob I . We are left to show

that .η−1 : G → F assembled from the components .ηi−1 , .i ∈ Ob I , is a natural

transformation. To this end consider .f ∈ HomI (i, j ); the naturality of .η yields the
following commutative diagram:

i.e., .ηj ◦ F (f ) = G(f ) ◦ ηi . Since each .ηi is an isomorphism, we also have .F (f ) ◦

ηi−1 = ηj−1 ◦ G(f ). Therefore, the following diagram is commutative:

which proves that .η−1 is indeed a natural transformation. 


Some of the properties of a given category are inherited by all its corresponding
functor categories, as can be seen in the following:
Proposition 1.9.4 Let I be a small category and .C a pointed category. Then the
functor category .Fun(I, C) is also pointed.
Proof Let .C0 be the zero object of .C. We will show that .C0 : I → C, the constant
functor at .C0 (see Example 1.5.3, (5)), is the zero object of the functor category
.Fun(I, C). To this end, consider another functor .F : I → C. We will show that

there exists a unique natural transformation .ψ : C0 → F and a unique natural

transformation .ϕ : F → C0 .
1.9 Functor Categories 65

Indeed, for all .i ∈ Ob I , we define .ψi to be the unique morphism in .C from .C0 to
.F (i) (recall that .C0 is, in particular, an initial object in .C). We are left to show that
.ψ as defined above is a natural transformation. Given any .f ∈ HomI (i, j ), both .ψj

and .F (f ) ◦ ψi are morphisms in .C from .C0 to .F (j ). As .C0 is an initial object in .C

we obtain .ψj = F (f ) ◦ ψi . Hence, the following diagram is commutative:

which shows that .ψ is indeed a natural transformation between .C0 and F .

Furthermore, .ψ is the unique natural transformation between the aforementioned
functors as for each .i ∈ Ob I there exists a unique morphism between .C0 and .F (i).
Consider now .ϕi to be the unique morphism in .C from .F (i) to .C0 , .i ∈ Ob I
(recall that .C0 is, in particular, a final object in .C). Then, for any .f ∈ HomI (i, j ),
both .ϕi and .ϕj ◦ F (f ) are morphisms in .C from .F (i) to .C0 . As .C0 is the final object
in .C, we obtain .ϕi = ϕj ◦ F (f ). This shows that the following diagram commutes
and therefore .ϕ is a natural transformation between F and .C0 :

ϕ is the unique natural transformation which can be defined between the two
.

functors above as for each .i ∈ Ob I there exists a unique morphism between .F (i)
and .C0 . 

Theorem 1.9.5 Let .B, .C be two categories with .B small. Then we have an
isomorphism of categories between .Fun(Bop , Cop ) and .Fun(B, C)op .
Proof Throughout this proof, for any morphism .α : F → G in .Fun(B, C)
(i.e., natural transformation) we denote by .α op : G → F the corresponding
opposite morphism in .Fun(B, C)op (Definition 1.4.1) while .α op : Gop → F op
stands for the opposite natural transformation as defined in Proposition 1.8.3.
op
Moreover, .◦op denotes the composition in .Fun(B, C)op and .◦C (resp. .◦C ) stands
for the composition in the category .C (resp. .Cop ). Finally, we use .• to denote
the composition in .Fun(Bop , Cop ) and the unadorned .◦ for the composition in
.Fun(B, C).
66 1 Categories and Functors

 
Define a functor .T : Fun(B, C)op → Fun Bop , Cop as follows:

T (F ) = F op ,
. T (α op ) = α op (1.32)

for all functors F , .G : B → C and natural transformations .α : F → G. Note

that T is well-defined as .α op : G → F in .Fun(B, C)op and .α op : Gop → F op .
Furthermore, consider three functors F , G, .H : B → C and let .α : F →
G, .β : G → H be natural transformations. Then, for all .B ∈ Ob B we
have
  (1.26)  op  op op
. (β ◦ α)op B
= (β ◦ α)B = βB ◦C αB = (αB )op ◦C (βB )op
(1.26)  op  op   (1.17)
= α B ◦C β op B = (α op • β op )B ,

which leads to:

(β ◦ α)op = α op • β op .
. (1.33)

We obtain:
    (1.33)    
T α op ◦op β op = T (β ◦ α)op = (β ◦ α)op = α op • β op = T α op • T β op
.

and therefore T is indeed a functor. We are left to construct the inverse func-
tor of T . To this end, consider .S : Fun Bop , Cop → Fun(B, C)op defined
by

S(U ) = U op ,
. S(β) = (β op )op (1.34)

for all functors U , .V : Bop → Cop and all natural transformations .β : U → V .

Note that .β op : V op → U op is a natural transformation between the functors .U op ,
.V
op : B → C and therefore .(β op )op is a morphism in .Fun(B, C)op . Consider now

three functors U , V , .W : Bop → Cop and natural transformations .α : U → V ,

.β : V → W .

We aim to show that .S(β • α) = S(β) ◦op S(α). To start with, note
that for any .B ∈ Ob B we have .αB ∈ HomCop (U (B), V (B)) and therefore
op
.αB = α
B for some morphism .α B ∈ HomC (V (B), U (B)). Similarly, .βB =
op
β B for some morphism .β ∈ HomC (W (B), V (B)). This leads to the follow-
ing:
   op op  op op op
.(β • α)op B = (β • α)B = (βB ◦C αB )op = βB ◦C αB op
 op op
= (αB ◦C βB )op = αB ◦C βB = (αB op )op ◦C (βB )op
op op  
= αB ◦C βB = (α op )B ◦C (β op )B = α op ◦ β op B .
1.9 Functor Categories 67

Hence, we have:

(β • α)op = α op ◦ β op .
. (1.35)

Putting everything together, we obtain:

 op (1.35)  op op op  op op op  op op
S(β •α) = (β •α)op
. = α ◦β = β ◦ α = S(β)◦op S(α),

as desired. Finally, we will show that T and S are inverses to each other. To this
end, given two functors F , .G : B → C and a natural transformation .α : F → G we
have
  (1.32) (1.34)
.S T (F ) = S(F op ) = (F op )op = F,
  (1.32) (1.34)  op
S T (α op ) ) = S(α op ) = (α op )op = α op .

Similarly, for all functors U , .V : Bop → Cop and natural transformations

.β : U → V we have

  (1.34) (1.32)
T S(U ) = T (U op ) = (U op )op = U,
.

  (1.34)   (1.32)
T S(β) = T (β op )op = = (β op )op = β

and the proof is now finished.



Lemma 1.9.6 Let I be a small category and .C an arbitrary category. If F ,
.G : I → C are two functors and .ψ : F → G is a natural transformation then
.ψ is a monomorphism in .Fun(I, C) if and only if the dual natural transformation

.ψ
op : F op → Gop is an epimorphism in .Fun(I op , Cop ).

Proof Indeed, using the duality principle, .ψ is a monomorphism in .Fun(I, C) if

and only if .ψ op is an epimorphism
  in .Fun(I, C)op . Furthermore, by Theorem 1.9.5
this is equivalent to .T ψ op = ψ op being an epimorphism in .Fun(I op , Cop ) and
the proof is now finished. 

There are various methods of constructing functors between functor categories.
For instance, if I is a small category and .C, .D are arbitrary categories then any
functor .F : C → D induces a functor between the corresponding functor categories
as follows:

F : Fun(I, C) → Fun(I, D), F (G) = F G, F (ψ)i = F (ψi )

. (1.36)

for all functors G, .G : I → C, all natural transformations .ψ : G → G and all

i ∈ Ob I .
.
68 1 Categories and Functors

Similarly, if .F : I → J is a functor between small categories and .C is an arbitrary

category, we can define another induced functor .F  between the corresponding
functor categories as follows:

F  : Fun(J, C) → Fun(I, C), F  (G) = GF, F  (ψ)i = ψF (i)

. (1.37)

for all functors G, .G : J → C, all natural transformations .ψ : G → G and all

i ∈ Ob I .
.

Finally, a functor .F : I → J between small categories induces by precomposi-

tion with F a new functor .− ◦ F : Fun(J, C) → Fun(I, C) defined for all functors
G, .G : J → C and all natural transformations .ψ : G → G by .ψ ◦ F = ψF .
Furthermore, for any functor category we can define an evaluation bifunctor as
follows:
Proposition 1.9.7 Given a small category I and an arbitrary category .C, there
exists a bifunctor .Ev : Fun(I, C) × I → C, called the evaluation bifunctor, defined
as follows for all i, .j ∈ Ob I , .f ∈ HomI (i, j ), all functors F , .G : I → C and all
natural transformations .η : F → G:

Ev(F, i) = F (i),
. Ev(η, f ) = G(f ) ◦ ηi .

Proof To start with, note that as .η : F → G is a natural transformation, the

following diagram commutes:

. (1.38)

Hence, we obtain .Ev(η, f ) = G(f ) ◦ ηi = ηj ◦ F (f ), which shows that .Ev is

well-defined.
Moreover, for all .i ∈ Ob I and all functors .F : I → C we have

Ev(1F , 1i ) = F (1i ) ◦ 1F (i) = 1F (i) ◦ 1F (i) = 1F (i) = 1Ev(F, i) .

.

We have proved that .Ev preserves identities. Consider now .f ∈ HomI (i, j ), .g ∈
HomI (j, k) and .η : F → G, .ξ : G → H , natural transformations, where F , G,
.H : I → C. We obtain:

 
Ev (ξ, g) ◦ (η, f ) = Ev(ξ ◦ η, g ◦ f ) = H (g ◦ f ) ◦ (ξ ◦ η)i
.

= H (g) ◦ H (f ) ◦ ξi ◦ ηi = H (g) ◦ ξj ◦ G(f ) ◦ ηi = Ev(ξ, g) ◦ Ev(η, f ),

1.9 Functor Categories 69

where in the fourth equality we used the naturality of .ξ . Hence .Ev preserves
compositions and is indeed a functor. 


In light of Proposition 1.5.5, we have two other functors induced by the

evaluation bifunctor .Ev. Indeed, given .i ∈ Ob I , the left associated functor at i
denoted by .Ev(−, i) : Fun(I, C) → C will be called the evaluation functor at i and
is defined as follows:

Ev(F, i) = F (i)
. and Ev(η, 1i ) = G(1i ) ◦ ηi = ηi (1.39)

for all functors F , .G : I → C and all natural transformations .η : F → G.

On the other hand, for any functor .F : I → C, the right associated functor
.Ev(F, −) : I → C coincides with F . Indeed, for all i, .j ∈ Ob I and all .f ∈

HomI (i, j ) we have

Ev(F, −)(i) = Ev(F, i) = F (i) and

.

Ev(F, −)(f ) = Ev(1F , f ) = F (f ) ◦ 1F (i) = F (f ).

Another functor which will later turn out to be important is the diagonal functor
defined as follows:
Proposition 1.9.8 Let I be a small category and .C an arbitrary category. There
exists a functor . : C → Fun(I, C), called the diagonal functor, defined as follows
for all C, .D ∈ Ob C and all .f ∈ HomC (C, D):

.(C) = C , (f ) = η : C → D ,

where .C : I → C is the constant functor at C and .η is the natural transformation

given by .ηi = f for all .i ∈ Ob I .
Proof Recall from Example 1.5.3, (5) that .C : I → C is the functor that sends
each object of the category I to C and each morphism of I to the identity .1C . To
start with, it is straightforward to see that for any .t ∈ HomI (i, j ) the following
diagram

.
70 1 Categories and Functors

comes down to
t
C D

1C 1D

C D
. t

which is obviously commutative. This shows that .η defined by .ηi = t for all .i ∈
Ob I is indeed a natural transformation and therefore . is well-defined.
Clearly, for any .C ∈ Ob C, we have .(1C ) = ξ , where .ξ : C → C is the
natural transformation defined by .ξi = 1C for all .i ∈ Ob I . Hence .ξ is the identity
natural transformation .1C and we have proved that . respects identities.
Consider now .f ∈ HomC (C, D) and .g ∈ HomC (D, E) and let .(f ) = μ,
.(g) = ν where .μ : C → D and .ν : D → E are natural transformations

defined by .μi = f and .νi = g for all .i ∈ Ob I . Furthermore, we have .(g ◦f ) = μ,

where .η : C → E is the natural transformation defined by .ηi = g ◦ f for all
.i ∈ Ob I . Therefore, .ηi = νi ◦ μi for all .i ∈ Ob I , which implies that the natural

transformations .η and .ν ◦ μ coincide. Consequently, .(g ◦ f ) = (g) ◦ (f ) and

we have proved that . is indeed a functor. 


1.10 Yoneda’s Lemma

Yoneda’s lemma is arguably one of the most important results contained in this
book. It allows us to embed any category into its category of presheaves by means of
a fully faithful functor and this approach opens the way to a plethora of applications.
Some important consequences of Yoneda’s lemma, which go beyond the scope of
this book, are mentioned at the end of the section.
Theorem 1.10.1 (Yoneda’s Lemma) Let .F : C → Set be a functor and .C ∈
Ob C. Then the natural transformations from .HomC (C, −) to F are in bijection
with the elements of the set .F (C) and the bijection is given for any natural
transformation .ϕ : HomC (C, −) → F as follows:
 
πC, F : Nat HomC (C, −), F → F (C), πC, F (ϕ) = ϕC (1C ) ∈ F (C).
. (1.40)
 
In particular, .Nat HomC (C, −), F is a set.
1.10 Yoneda’s Lemma 71

 
Proof Consider .τC, F : F (C) → Nat HomC (C, −), F , defined for every .x ∈
F (C) by:
 
τC, F (x) = hx where hx D (f ) = F (f )(x)
. (1.41)

for every .D ∈ Ob C and .f ∈ HomC (C, D). First we have to check that .hx is a
natural transformation. This comes down to proving that given a morphism .f ∈
HomC (A, B) the following diagram is commutative:

Indeed, for any .g ∈ HomC (C, A) we have

 
(hx )B HomC (C, f ) (g) = (hx )B (f ◦ g)
.

= F (f ◦ g)(x) = F (f ) ◦ F (g)(x)
= F (f )(hx )A (g).

Thus .hx is a natural transformation. The proof will be finished once we show that
.πC, F and .τC, F are inverses to each other. To start with, consider .x ∈ F (C). Then

we have
   
.(πC, F ◦ τC, F )(x) = πC, F τC, F (x) = πC, F hx
 
= hx C (1C ) = F (1C )(x)
= 1F (C) (x) = x.

Thus .πC, F ◦ τC, F = 1F (C) . Consider now a natural transformation .ϕ :

HomC (C, −) → F . We want to prove that .(τC, F ◦ πC, F )(ϕ) = ϕ. Indeed, as
.ϕ is a natural transformation, for every .D ∈ Ob C and every .f ∈ HomC (C, D) the

following diagram is commutative:

.
72 1 Categories and Functors

In particular, by evaluating the above diagram at .1C ∈ HomC (C, C) we obtain:

 
.F (f ) ϕC (1C ) = ϕD (f ). (1.42)

Hence for every .D ∈ Ob C and every .f ∈ HomC (C, D) we have

   
τC, F πC, F (ϕ) D (f ) = τC, F ϕC (1C ) D (f )
.

 (1.42)
= F (f ) ϕC (1C )) = ϕD (f ),

which implies .τC, F ◦ πC, F (ϕ) = ϕ and the proof is now complete. 

It turns out that the bijections .πC, F defined in (1.40) form a natural transforma-
tion in the variable C. Furthermore, if .C is a small category, then the bijections .πC, F
form a natural transformation in the variable F as well. The precise statement is the
following:
Theorem 1.10.2 Let .C be a category, .C ∈ Ob C and .F : C → Set a functor.
(1) If .G : C → Set is the functor defined as follows for all .C ∈ Ob C and .f ∈
HomC (C, C ):
 
G(C) = Nat HomC (C, −), F ,
.
   
G(f ) : Nat HomC (C, −), F → Nat HomC (C , −), F ,
 
G(f )(ψ) = ψ ◦ HomC (f, −), ψ ∈ Nat HomC (C, −), F

then .π−, F : G → F defined in (1.40) is a natural transformation.

(2) Assume .C is a small category. If .H : Fun(C, Set) → Set is the functor defined
as follows for all functors .F : C → Set and natural transformations .ψ : F →
F:
 
H (F ) = Nat HomC (C, −), F ,
.
   
H (ψ) : Nat HomC (C, −), F → Nat HomC (C, −), F ,
 
H (ψ)(ϕ) = ψ ◦ ϕ, ϕ ∈ Nat HomC (C, −), F

then .πC, − : H → EvC defined in (1.40) is a natural transformation, where .EvC

is the evaluation functor at C.
1.10 Yoneda’s Lemma 73

Proof
(1) Let .f ∈ HomC (C, C ); we need to prove the commutativity of the following
diagram:

 
Indeed, for all .ϕ ∈ G(C) = Nat HomC (C, −), F we have

  (1.40)    
.F (f ) ◦ πC, F (ϕ) = F (f ) ϕC (1C ) = ϕC (f ) = ϕC HomC (f, C )(1C )
     
= ϕ ◦ HomC (f, −) C (1C ) = πC , F ϕ ◦ HomC (f, −) = πC , F ◦ G(f ) (ϕ),

where the second equality holds because .ϕ : HomC (C, −) → F is a natural

transformation.
(2) Consider two functors F , .F : C → Set and a natural transformation .ψ : F →
F . The proof will be finished once we show the commutativity of the following
diagram:

 
To this end, let .ϕ ∈ H (F ) = Nat HomC (C, −), F ; we obtain

    (1.40)  
. EvC (ψ) ◦ πC, F (ϕ) = EvC (ψ) πC, F (ϕ) = EvC (ψ) ϕC (1C )
(1.39)  
= ψC ϕC (1C ) = (ψ ◦ ϕ)C (1C ) = πC, F (ψ ◦ ϕ)
   
= πC, F H (ψ)(ϕ) = πC, F ◦ H (ψ) (ϕ),

as desired.


Remark 1.10.3 The contravariant version of Yoneda’s lemma can be easily
obtained by replacing the category .C with .Dop in Theorem 1.10.1 and noticing
that the functors .HomDop (C, −) : Dop → Set and .HomD (−, C) : Dop → Set
coincide.
74 1 Categories and Functors

As a first consequence, we can describe all natural transformations between hom

functors.
Corollary 1.10.4 Let .C be a category and C, .D ∈ Ob C. Then:
(1) .α : HomC (C, −) → HomC (D, −) is a natural transformation if and only
if there exists a unique .φ ∈ HomC (D, C) such that .α = hφ , where
.h : HomC (C, −) → HomC (D, −) is the natural transformation defined as
φ

follows:
 φ
.h X (f ) = f ◦ φ, for all X ∈ Ob C and f ∈ HomC (C, X). (1.43)

(2) .β : HomC (−, C) → HomC (−, D) is a natural transformation if and only

if there exists a unique .δ ∈ HomC (C, D) such that .β = t δ , where
.t : HomC (−, C) → HomC (−, D) is the natural transformation defined as
δ

follows:
 δ
.t X (g) = δ ◦ g, for all X ∈ Ob C and g ∈ HomC (X, C). (1.44)

Proof

 for .F = HomC (D, −). This

(1) We apply the Yoneda lemma  gives a set bijection
.τ : HomC (D, C) → Nat HomC (C, −), HomC (D, −) defined for every .φ ∈

HomC (D, C) by:

 
τ (φ) = hφ , where hφ X (f ) = HomC (D, f )(φ) = f ◦ φ
.

for every .X ∈ Ob C and .f ∈ HomC (C, X). Hence, any natural transformation
between .HomC (C, −) and .HomC (D, −) is of the form .hφ for a unique .φ ∈
HomC (D, C), as desired.
(1) Follows from the contravariant version of Yoneda’s lemma.


Proposition 1.10.5 Let .C be a category and C, .D ∈ Ob C. Then C and D are
isomorphic if and only if the functors .HomC (C, −) and .HomC (D, −) are naturally
isomorphic.
Proof Suppose first that C and D are isomorphic objects in .C and let .φ ∈
HomC (D, C) be an isomorphism. As proved in Corollary 1.10.4, (1) we have a
natural transformation
 defined by .τ (φ) = hφ : HomC (C, −) → HomC (D, −),
where . h X (f ) = HomC (D, f )(φ) = f ◦ φ for all .X ∈ Ob C and .f ∈
φ
 
HomC (C, X). We will prove that . hφ X is a set bijection for every .X ∈ Ob C. To
 φ  
this end, we define . μ X : HomC (D, X) → HomC (C, X) by . μφ X (g) = g◦φ −1
   
for any .g ∈ HomC (D, X). We will see that . μφ X is the inverse of . hφ X . Indeed,
1.10 Yoneda’s Lemma 75

for any .g ∈ HomC (D, X) and any .f ∈ HomC (C, X) we have

 φ      
. h
X
◦ μφ X (g) = hφ X g ◦ φ −1 = g ◦ φ −1 ◦ φ = g,
 φ      
μ X ◦ hφ X (f ) = μφ X f ◦ φ = f ◦ φ ◦ φ −1 = f.

Hence, .τ (φ) : HomC (C, −) → HomC (D, −) is a natural isomorphism, as desired.

Conversely, let .α : HomC (C, −) → HomC (D, −) be a natural isomorphism.
Denote .αC (1C ) ∈ HomC (D, C) by u; we will prove that u is an isomorphism.
We start by pointing out that .αD : HomC (C, D) → HomC (D, D) is a bijection and
since .1D ∈ HomC (D, D) we can find .v ∈ HomC (C, D) such that .αD (v) = 1D .
Now as .α is a natural transformation, the following diagram is commutative:

αD
Hom (C, D) Hom (D, D)

(C, u) (D, u)

Hom (C, C) Hom (D, C)

.
αC

i.e., .HomC (D, u) ◦ αD = αC ◦ HomC (C, u). By evaluating this diagram at .v ∈

HomC (C, D) and using .αD (v) = 1D we obtain

HomC (D, u) ◦ αD ◦ v = αC ◦ HomC (C, u) ◦ v

.

⇔ u ◦ αD (v) = αC (u ◦ v)
⇔ u = αC (u ◦ v).

Since we also have .u = αC (1C ) and .αC is a set bijection, we get .u ◦ v = 1C .

Using again the fact that .α is a natural transformation the following diagram is
commutative:

Evaluating the above diagram at .1C ∈ HomC (C, C) and using .αD (v) = 1D we
obtain

HomC (D, v) ◦ αC ◦ 1C = αD ◦ HomC (C, v) ◦ 1C

.

⇔ v ◦ αC (1C ) = αD (v ◦ 1C )
⇔ v ◦ u = 1D ,
76 1 Categories and Functors

therefore v is the inverse of u and the proof is now finished. 


In light of the above result we obtain:
Corollary 1.10.6 Any two representing objects of a given representable functor
F : C → Set are isomorphic.
.

Proof Let .F : C → Set be a functor and let C, .D ∈ Ob C be representing objects

of F . Therefore we have two natural isomorphisms .α : F → HomC (C, −) and
.β : F → HomC (D, −). This shows that .β ◦ α
−1 : Hom (C, −) → Hom (D, −)
C C
is a natural isomorphism, where .α −1 is the natural transformation defined in
Example 1.7.2, (6). Now Proposition 1.10.5 implies that C and D are isomorphic
objects in .C. 

Definition 1.10.7 Given a small category .C, the functor .Y : C → Fun(Cop , Set)
which sends any .C ∈ Ob C to the hom functor .HomC (−, C)22 and any .f ∈
HomC (C, D) to the natural transformation .t f defined in (1.44) is called the Yoneda
embedding.
Corollary 1.10.8 For any small category .C, the Yoneda embedding functor
Y : C → Fun(Cop , Set) is fully faithful.
.

Proof Let C, .D ∈ Ob C; we need to prove that the mapping defined below is

bijective:
 
YC, D : HomC (C, D) → Nat HomC (−, C), HomC (−, D) ,
.

YC, D (f ) = Y (f ) = t f , f ∈ HomC (C, D).

This follows trivially from Corollary 1.10.4, (2). 


Example 1.10.9 Cayley’s theorem is a well-known result in group theory. It states
that any group G is isomorphic to a subgroup of the symmetric group on G (i.e.,
the group of all permutations on the set G). In what follows, we show that Cayley’s
theorem can be obtained as an easy consequence of Yoneda’s lemma. Indeed, recall
that any group .(G, ·) can be made into a category .G with one object, say ., and
.HomG (, ) = G as in Example 1.2.2, (3). Now Corollary 1.10.4, (2) of Yoneda’s

lemma gives a set bijection:

 
Y,  : HomG (, ) → Nat HomG (−, ), HomG (−, )
.
 
g → t g , where t g  (x) = g · x for all g, x ∈ G.
 
Note that . t g  : G → G is a bijective map or, equivalently, a permutation on the set
G. Furthermore, we have a group structure on the natural transformations .{t g | g ∈

22 Note that .HomC (−, C) : Cop → Set is a covariant functor and therefore it is an object of the
category of presheaves on .C.
1.11 Exercises 77

G} whose multiplication is given as follows: .t g1 ◦ t g2 = t g1 ·g2 for all .g1 , .g2 ∈ G.

With this multiplication, the set bijection .Y,  becomes a group homomorphism.
Indeed, for all .g1 , .g2 we have

Y,  (g1 · g2 ) = t g1 ·g2 = t g1 ◦ t g2 ,

.

as desired. Therefore Yoneda’s lemma gives a bijection between G and some

subgroup of permutations on G, as stated in Cayley’s theorem. .

The influence of Yoneda’s lemma in various areas of mathematics can hardly

be exaggerated. As we cannot possibly do justice to the vast and far reaching
applications of Yoneda’s lemma without exceeding the scope of this introductory
book, we only provide here a few scattered examples with pointers to the literature
for details. Among the areas where Yoneda’s lemma is a key ingredient in proving
fundamental results we mention algebraic geometry (e.g., in showing that the
category of all solution functors of a system of polynomial equations is equivalent
to the category of relative schemes that are of finite presentation over the base field
[11, Part B]), the theory of locally presentable and accesible categories (e.g., in
showing that given a small category .C, every hom-functor can be seen as a finitely
presentable object in the corresponding category of presheaves [1]), topos theory
(e.g., in showing the existence of a subobject classifier for an arbitrary category
of presheaves [36, Chapter I, §4]). Further applications of Yoneda’s lemma will be
highlighted in Example 2.7.4 in connection to the (co)completeness of the category
of presheaves on an arbitrary small category.

1.11 Exercises

1.1 Is Grp a subcategory of Set?

1.2 If A ∈ Ob C is an initial object and B ∈ Ob C is such that A and B are
isomorphic then B is also an initial object in C.
1.3 Consider the category P whose objects are triples (X, x, f ), where X ∈
Ob Set, x ∈ X, f : X → X is a function, and a morphism in P between two
objects (X, x, f ) and (Y, y, g) is a function θ : X → Y such that θ (x) = y
which renders the following diagram commutative:

θ
X Y

f g

X Y
. θ

Show that P has an initial object.

78 1 Categories and Functors

1.4 Let X be an object in C.

(a) If X is an initial object then any monomorphism m ∈ HomC (C, X) is an
isomorphism.
(b) If X is a final object then any epimorphism m ∈ HomC (X, C) is an
isomorphism.
1.5 Let C be a category and f ∈ HomC (A, B), g ∈ HomC (B, C). Prove that:
(a) if both f and g are monomorphisms (epimorphisms) then g ◦ f is a
monomorphism (epimorphism);
(b) if the composition g ◦ f is a monomorphism then f is a monomorphism;
(c) if the composition g ◦ f is an epimorphism then g is an epimorphism.
1.6 A morphism f ∈ HomC (A, B) is called a split monomorphism if there exists
some t ∈ HomC (B, A) such that t ◦f = 1A (i.e., f is left invertible). Dually, a
morphism f in C is called a split epimorphism if f op is a split monomorphism
in Cop .
(a) Show that any split monomorphism (resp. split epimorphism) is a
monomorphism (resp. epimorphism) and prove by a counterexample
that the converse does not hold.
(b) Show that f is both an epimorphism and a split monomorphism if and
only if f is an isomorphism. Dually, f is both a monomorphism and a
split epimorphism if and only if f is an isomorphism.
1.7 An epimorphism f ∈ HomC (A, B) is called a strong epimorphism if for any
commutative square:

f
A B h f m g,
g h

C D
. m

where m ∈ HomC (C, D) is a monomorphism, there exists a unique v ∈

HomC (B, C) which gives rise to two commutative triangles, i.e., m ◦ v = h
and v ◦ f = g. Dually, a morphism f in C is called a strong monomorphism if
f op is a strong epimorphism in Cop . Show that the following are equivalent:
(a) f is an isomorphism;
(b) f a strong epimorphism and a monomorphism;
(c) f a strong monomorphism and an epimorphism.
1.11 Exercises 79

1.8 Prove that any group regarded as a one-object category is isomorphic to its
opposite. Is the assertion still true for monoids?
1.9 Give examples to show that:
(a) arbitrary functors do not necessarily preserve or reflect monomorphisms
and epimorphisms;
(b) functors which reflect isomorphisms are not necessarily fully faithful.
1.10 Construct an example of:
(a) a full functor which is not surjective on objects/morphisms;
(b) a faithful functor which is not injective on objects/morphisms.
1.11 Is there a functor F : Grp → Grp such that F (G) is equal to the center of G
for all groups G?
1.12 Is there a functor F : Top → Set such that F (X) is equal to the set of
connected components of X for all topological spaces X?
1.13 Let F : C → D and G : D → C be functors such that GF is naturally
isomorphic to 1C . Prove that F is faithful.
1.14 Let F : C → D and G : D → C be two functors.
(a) If GF is faithful then F is faithful.
(b) If GF is full and G is faithful then F is full.
1.15 Let I be a small category and F : C → D a fully faithful functor. If
F : Fun(I, C) → Fun(I, D) is the induced functor defined in (1.36), show
that:
(a) F is fully faithful;
(b) if G, H : I → C are two functors such that F (G) and F (H ) are naturally
isomorphic, then G and H are naturally isomorphic.
1.16 Let U : Mon → Set be the forgetful functor and define U × U : Mon → Set
as follows:
(U × U )(M) = U (M) × U (M) for all M ∈ Ob Mon;
(U × U )(f )(m, m ) = (f (m), f (m )) for all f ∈ HomMon (M, N ), m,
m ∈ U (M).
Prove that γ : U × U → U , defined by γM = mM for all M ∈ Ob Mon, is
a natural transformation, where mM denotes the multiplication of the monoid
M.
1.17 Prove the interchange law between vertical and horizontal composition of
natural transformations, i.e., (δ ∗ γ ) ◦ (β ∗ α) = (δ ◦ β) ∗ (γ ◦ α), where
80 1 Categories and Functors

the natural transformations involved fit into the following picture:

F H

G I

. P Q

1.18 Let  G be a group and let G be the corresponding category. Describe

Nat 1G , 1G .
1.19 Let (X, ) and (Y, ) be two pre-ordered sets and PO(X, ), respectively
PO(Y, ), the associated categories.

 If F, G : PO(X, 
(a) Describe the functors between these two categories.
) → PO(Y, ) are two such functors, describe Nat F, G .
(b) Show that Fun PO(X, ), PO(Y, ) is isomorphic to a category asso-
ciated to a pre-ordered set.
1.20 Let V be a finite-dimensional vector space. Show that the functors
HomK M (V , −), V ∗ ⊗ − : K M → K M, are naturally isomorphic.
1.21 Prove that for any category C there exists a functor F : C → Set which is not
representable.
1.22 Prove that the contravariant power-set functor P c : Set → Set is repre-
sentable.
1.23 Let G ∈ Ob Grp. Prove that the functor FG : Grp → Set defined as follows
for any H ∈ Ob Grp, f ∈ HomGrp (G, H ):

FG (H ) = {f ∈ HomSet (G, H ) | f is an antimorphism of groups},

.

F (f )(g) = f ◦ g, g ∈ FG (H )

is representable.
1.24 An object I of a category C is called injective if for any u ∈ HomC (A, I )
and any monomorphism m ∈ HomC (A, B) there exists a morphism v ∈
HomC (B, I ) such that the following diagram is commutative:

.
1.11 Exercises 81

Dually, an object P of C is called projective if it is an injective object in Cop .

Describe the injective (resp. projective) objects in Ab and R M.
1.25 Let I , J be two smallcategories and C an arbitrary category. Prove that the
functor categories Fun I, Fun(J, C) and Fun(I × J, C) are isomorphic.
Chapter 2
Limits and Colimits

Limits and colimits are fundamental and unifying concepts in category theory. Many
seemingly unrelated constructions from different fields of mathematics such as the
free product of groups (with amalgamation), the tensor product of (co)algebras
or the direct sum of modules are in fact special instances of these very general
concepts.

2.1 (Co)products, (Co)equalizers, Pullbacks and Pushouts

In order to achieve a better understanding of (co)limits we will introduce them grad-

ually, starting with some generic special cases, namely (co)products, (co)equalizers,
pullbacks and pushouts.
 
Definition 2.1.1 Let I be a set and . Pi i∈I a family of objects in a category .C. A
 
product of the family .(Pi )i∈I is a pair . i∈I Pi , (pi )i∈I , where

(1) . i∈I Pi ∈ Ob C;
(2) .pj : i∈I Pi → Pj are morphisms in .C for all .j ∈ I ,
 
and for any other pair . P , (fi )i∈I where
(1) .P ∈ Ob C;
(2) .fj : P → Pj are morphisms in .C for all .j ∈ I ,

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 83

A. Agore, A First Course in Category Theory, Universitext,
https://doi.org/10.1007/978-3-031-42899-9_2
84 2 Limits and Colimits


there exists a unique morphism .f : P → i∈I Pi in .C such that the following
diagram commutes for all .j ∈ I :

P
fj
f

i I Pi Pj
.
pj

We say that .C is a category with (finite) products or that .C has (finite) products if
there exists a product in .C for any (finite)
 family of objects. If I is a finite set, then
we also write .P1 × . . . × Pn instead of . i=1, n Pi .

 Coproducts
 are the dual notion of products; that is, the coproduct of a family
.Xi i∈I of objects of a category .C is defined to be the product of the same family of
objects in the dual category .C op . This comes down to the following:
 
Definition 2.1.2 Let I be a set and . Qi i∈I a family of objects in a category .C. A
 
coproduct of the family .(Qi )i∈I is a pair . i∈I Qi , (qi )i∈I , where

(1) . i∈I Qi ∈  Ob C;
(2) .qj : Qj → i∈I Qi are morphisms in .C for all .j ∈ I ,
 
and for any other pair . Q, (fi )i∈I , where
(1) .Q ∈ Ob C;
(2) .fj : Qj → Q are morphisms in .C for all .j ∈ I ,

there exists a unique morphism .f : i∈I Qi → Q in .C such that the following
diagram commutes for all .j ∈ I :

qj
Qj i I Qi

f
fj

.
Q

We say that .C is a category with (finite) coproducts or that .C has (finite) coproducts
if there is a coproduct in .C for any (finite) family of objects.
Proposition 2.1.3 When it exists, the (co)product of a family of objects is unique
up to isomorphism.
Proof We start by proving the uniqueness up to isomorphism
 of the product. Let

C be a category and consider two products . P , (pi )i∈I , respectively . P , (pi )i∈I ,
.
   
in .C of the same family of objects . Pi i∈I . Since . P , (pi )i∈I is assumed to be
a product, there exists a unique .f ∈ HomC (P , P ) such that for any .j ∈ I we
2.1 (Co)products, (Co)equalizers, Pullbacks and Pushouts 85

have

p j ◦ f = pj .
. (2.1)
 
Similarly, as . P , (pi )i∈I is also a product of the same family of objects, we obtain
a unique .g ∈ HomC (P , P ) such that for any .j ∈ I we have

pj ◦ g = p j .
. (2.2)

P
pj
f
pj
1P P Pj

g
pj

.
P

By composing the equality in (2.2) with f on the right and using (2.1) we obtain

(2.1)
pj ◦ (g ◦ f ) = pj ◦ f = pj
. (2.3)
 
for any .j ∈ I . Applying Definition 2.1.1 to the pair . P , (pi )i∈I , seen both as
a product and as the other pair, yields a unique .h ∈ HomC (P , P ) such that
.pj ◦ h = pj for any .j ∈ I . By the uniqueness of h we must have .h = 1P .

Moreover, since by (2.3) the map .g ◦ f also fulfills the above identity we obtain
.g ◦ f = 1P . In the same manner we obtain .f ◦ g = 1
P and therefore P and .P are
isomorphic.
The assertion regarding  coproducts  follows
 by applying
 the duality principle.
Indeed, assume now that . Q, (qi )i∈I and . Q, (q i )i∈I are coproducts in .C of the
   op   op 
same family of objects . Qi i∈I . Then . Q, (qi )i∈I and . Q, (q i )i∈I are both
 
products in .C op of the family of objects . Qi i∈I . According to the above proof, there
exists an isomorphism .f op ∈ HomC op (Q, Q). Therefore, we have an isomorphism
.f ∈ HomC (Q, Q). 

Proposition 2.1.4 Let .C be a category such that any two objects admit a
(co)product. Then, any non-empty finite family of objects in .C admits a
(co)product.
Proof Let .{X1 , X2 , . . . , Xn } be a non-empty family of objects in .C. We use
induction on n to construct the product of this family. If .n = 1 then the family
.{X} has a product given by the pair .(X, 1X ). Indeed, if . Y, f ∈ HomC (Y, X)
86 2 Limits and Colimits

then the unique morphism which makes the following diagram commutative is
precisely f :

Y
f
f

X X
.
1X

For .n = 2 the conclusion follows by our assumption. Assume now that

any family with at most .k ≥ 2 objects admits a product and consider
the family .{X1 , X2 , . . . , Xk+1 } consisting of .(k + 1) objects. According
 inductive hypothesis, the family .{X1 , X2 , . . . , Xk } admits a product,
to the
say . Y, (πj )j =1, k , where .πj ∈ HomC (Y, Xj ) for all .j = 1, 2, . . . , k.
 
Furthermore, consider . X, (πX , πk+1 ) to be the product of the family
.{Y, Xk+1 }, where .πX ∈ HomC (X, Y ) and .πk+1 ∈ HomC (X, Xk+1 ). We

will show that . X, (π1 ◦ πX , π2 ◦ πX , . . . , πk ◦ πX , πk+1 ) is the product
of the family of objects .{X1 , X2 , . . . , Xk+1 }. To this end,  let .Z ∈ Ob  C
and .fj ∈ HomC (Z, Xj ), .j = 1, 2, . . . , (k + 1). As . Y, (πj )j =1, k is
the product of the family of objects .{X1 , X2 , . . . , Xk }, there exists a unique
.f ∈ HomC (Z, Y ) such that the following diagram is commutative for all
.i = 1, 2, . . . , k:

Z i.e., i f f i.
fi
f

Y Xi
. i (2.4)
 
Moreover, as . X, (πX , πk+1 ) is the product of the family .{Y, Xk+1 }, there
exists a unique .g ∈ HomC (Z, X) such that the following diagram is commuta-
tive:

(2.5)

Note that .g ∈ HomC (Z, X) is a morphism such that .πk+1 ◦ g = fk+1

    (2.5) (2.4)
and . πj ◦ πX ◦ g = πj ◦ πX ◦ g = πj ◦ f = fj for all .j =
1, 2, . . . , k. The proof will be finished once we show that g is the unique morphism
2.1 (Co)products, (Co)equalizers, Pullbacks and Pushouts 87

with these properties. Indeed, assume there exists a .g  ∈ HomC (Z, X) such
that
 
πk+1 ◦ g  = fk+1
. and πj ◦ πX ◦ g  = fj , for all j = 1, 2, . . . , k.
 
In particular, we have .πj ◦ πX ◦ g  = fj for all .j = 1, 2, . . . , k and by the
uniqueness of the morphism which makes (2.4) commutative we obtain .f = πX ◦g  .
Putting everything together, we obtain .πk+1 ◦ g  = fk+1 and .πX ◦ g  = f .
As g is the unique morphism for which (2.5) holds it follows that .g = g  , as
desired.
The claim concerning coproducts follows by duality. Indeed, assume that .C is a
category such that any two objects admit a coproduct. Then, in .C op any two objects
admit a product and, according to the above proof, any non-empty finite family of
objects in .C op admits a product. Therefore, any non-empty finite family of objects
in .C admits a coproduct, as desired. 

Examples 2.1.5
(1) The product of any family .(Xi )i∈I of objects
 in Set, where I is a set, is given
by the corresponding
  cartesian product
 . i∈I Xi together with the canonical
projections . πj : i∈I Xi → Xj j ∈I . More precisely, we have
 
. Xi = {(xi )i∈I | xi ∈ Xi }, πj : Xi → Xj ,
i∈I i∈I
 
πj (xi )i∈I = xj for all j ∈ I.

Indeed, consider .X ∈ Ob Set and a family of functions .pi  : X → Xi , .i ∈ I .

We will show that there exists a unique map .f : X → i∈I Xi such that
.πi ◦ f = pi for all .i ∈ I . To this end, define .f : X → i∈I Xi by .f (x) =
 
pi (x) i∈I for all .x ∈ X. Then, for all .i ∈ Iand .x ∈ X we obviously have
.πi ◦ f (x) = pi (x). Consider now .g : X → i∈I Xi to be another map such
that .πi ◦g = pi for all .i ∈ I . As .πi is just the projection on the i-th component,
and .πi ◦g(x) = pi (x) for all .i ∈ I and all .x ∈ X, we obtain .f = g, as desired.
(2) The product of any family of objects in Grp, Ab, .R M, Ring is given by the
cartesian product of the underlying sets endowed with componentwise opera-
tions together with  the canonical projections. For instance, in the category Grp
for
 any family . Gi
i∈I
of objects, the group structure on the cartesian product
. i∈I G i is defined as follows:
     
.(gi )i∈I · (hi )i∈I = (gi ·i hi )i∈I ,

where .·i denotes the group multiplication in .Gi . It can now be easily checked
as in the previous example that this is indeed the product in Grp of the family
. Gi
i∈I
.
88 2 Limits and Colimits

(3) The product of any family of objects in Top is given by the cartesian product
of the underlying sets endowed with the product topology together with the
canonical
 projections. Let  I be a set, .(Xi )i∈I a family of topological spaces
and . i∈I Xi , (πi )i∈I the product of the underlying sets. Then, each .πj
is obviously
 continuous as the product topology  is the coarsest topology
on . i∈I Xi for which all projections .πj : i∈I Xi → Xj are continuous.
Assume now that .X ∈ Ob Top and .pj ∈ HomTop (X, Xj ), .j ∈ I , is
a family of  continuous maps. It can be easily seen that the unique map
.f : X → i∈I Xi defined in Example 2.1.5, (1) is continuous too. Indeed,
recall that the product topology is generated by the sub-basis .S = j ∈I Sj ,
where .Sj = {πj−1 (Uj ) | Uj open in Xj } for all .j ∈ J , and therefore in
order to prove continuity of  f it will suffice to show that the inverse image
of each sub-basis element of . i∈I Xi is open [39, p. 103]. To this end, for all
 
−1 π −1 (U ) = (π ◦ f )−1 (U ) = p −1 (U ) and since .p is
.j ∈ I we have .f j j j j j
j j
continuous, the latter term is an open set, as desired.
(4) Products in Haus and KHaus, the categories of Hausdorff and compact
Hausdorff spaces respectively, are constructed as in Top. Indeed, it will suffice
to show that given a family of Hausdorff (resp. compact Hausdorff) spaces
.(Xi )i∈I , where I is a set, the product of the underlying spaces together with

the product topology defined in the previous examples is a Hausdorff (resp.

compact
 Hausdorff) space. To this end, let .x = (xi )i∈I , .y = (yi )i∈I ∈
i∈I X i such that .x = y. Thus, there exists a .t ∈ I such that .xt = yt and
since .Xt is a Hausdorff space we can  find two disjoint open spaces in .Xt such
that .xt ∈ U and .yt ∈ V . As .πt : i∈I Xi → Xt is a continuous map, both
−1 
.πt (U ) and .πt−1 (V ) are open subsets in . i∈I Xi such that .x ∈ πt−1 (U ) and
−1
.y ∈ πt (V ). Moreover, we have .πt−1 (U ) ∩ πt−1 (V ) = πt−1 (U ∩ V ) = ∅
([37, Exercise 1.5]). To conclude, we have found two disjoint neighborhoods
−1
.πt (U ) and .πt−1 (V ) of x and y respectively, which shows that . i∈I Xi is
indeed a Hausdorff space.
Furthermore, if .(Xi )i∈I is a family of compact  Hausdorff spaces,
Tychonoff’s theorem shows that the product topology . i∈I Xi is a compact
space as well [39, Theorem 37.3].
(5) Let PO.(X, ) be the category corresponding to the pre-ordered set .(X, )
(as defined in Example 1.2.2, (2)) and .(xi )i∈I a family of objects in PO.(X, )
indexed by a set I , i.e., .xi ∈ X for any .i ∈ I . If it exists, the product of this
family is a pair . p, (πi )i∈I , where .p ∈ X and .πi : p → xi are morphisms in
PO.(X, ). Thiscomes down to .p  xi for any .i ∈ I . Moreover, for any other
pair . q, (ui )i∈I , where .q ∈ X and .ui : q → xi are morphisms in PO.(X, ),
there exists a morphism .f : q → p. In other words, for any .q ∈ X such that
.q  xi for any .i ∈ I , we also have .q  p. Therefore, p is precisely the

infimum (if it exists) of the family .(xi )i∈I .

(6) The product of categories as defined in Definition 1.4.4 together with the
corresponding projection functors (Example 1.5.3, (6)) is the product in the
category Cat of small categories (Proposition 1.5.4).  Indeed, if .(Ci )i∈I
 is a
family of small categories indexed by the set I , then . i∈I Ci , (pCi )i∈I is the
2.1 (Co)products, (Co)equalizers, Pullbacks and Pushouts 89

product of this family in Cat. It can be easily seen that for any small category
.D and anyfamily of functors .UCi : D → Ci there exists a unique functor
.F : D → i∈I Ci such that .pCi ◦ F = UCi for all .i ∈ I ; more precisely, F is
defined as follows for all .D ∈ Ob D and .f ∈ HomD (D, D  ):
   
F (D) = UCi (D) i∈I ,
. F (f ) = UCi (f ) i∈I .

(7) The category SiGrp of simple groups does not admit products or coproducts.
Indeed,
 suppose  this category admits products and let H , K be simple groups.
Let . X, (p, q) be the product in SiGrp of H and K. In particular, X is a sim-
ple group and .p : X → H , .q : X → K are group homomorphisms. Consider
now the pair . H, (IdH , 0K ) where .IdH is the identity homomorphism on H
while .0K : H → K denotes the group homomorphism defined by .0K (h) = 1K
for all .h ∈ H . By Definition 2.1.1 there exists a unique homomorphism of
groups .f : H → X such that the following diagram is commutative:

From .p ◦ f = IdH it follows that f is injective and .q ◦ f = 0K implies

that .Im(f ) ⊆ ker(q) ⊆ X. Putting all this together we have .{1} = H ∼ =
Im(f ) ⊆ ker(q)  X and since .ker(q) is a normal subgroup of the simple
group X, we must have .ker(q) = X, i.e., .q = 0K . Next, we consider the
pair . K, (0H , IdK ) where .IdK is the identity homomorphism on K while
.0H : K → H denotes the group homomorphism defined by .0H (k) = 1H for

all .k ∈ K. Using again Definition 2.1.1 yields a unique homomorphism of

groups .g : K → X such that the following diagram is commutative:

Since .q = 0K the last equality gives .K = {1}, which is a contradiction. A

similar argument shows that SiGrp does not have coproducts either.
(8) The category Field does not have products or coproducts. Indeed, consider the
fields .Z2 and .Z3 of integers modulo 2, respectively 3 and let .(K, (i, j )) be
their coproduct in Field, where .i : Z2 → K and .j : Z3 → K are morphisms
of fields. Thus, in K we have both .1 + 1 = 0 and .1 + 1 + 1 = 0, which yields
.1 = 0, an obvious contradiction. Similarly one can prove that Field does not

have products either.

90 2 Limits and Colimits

(9) In Set, the coproduct of a family .(Xi )i∈I is just its disjoint union, i.e., the
union of the sets .Xi = Xi × {i}. Thus, the coproduct of the family .(Xi )i∈I
is the pair . i∈I X  i , (qi ) i∈I , where . i∈I Xi = {(x, i) | i ∈ I, x ∈ Xi }
and .qj : Xj → i∈I X i , .qj (x) = (x, j ) for all .j ∈ I . Indeed, given
a setQ together with a collection of  maps  .fj : Xj → Q, .j ∈ I , define
.f : X i → Q by considering .f (x, j ) = fj (x) for all .(x, j ) ∈ Xj ⊂
 i∈I

i∈I Xi . Since each .(x, j ) lies inside a unique copy of .Xj , the following map
is well-defined:

 
(10) The coproduct of any family . Xi , τi i∈I of  objects in Top is given by
the disjoint union of the underlying sets . i∈I Xi constructed in the
previous example endowed with the finest topology .τ for which all
maps .qj are continuous. Consider now .Q ∈ Ob Top and a family
of continuous maps  .fj : Xj → Q, .j ∈ I . We will show that the
unique map .f : i∈I Xi → Q defined in the previous example is
continuous. To this end, let .U ⊆ Q be an open set. Note that since .τ
is the finest topology for which all .qj are continuous, in order to show
 
that .f −1 (U ) is an open set it will suffice to prove that .qj−1 f −1 (U )
 
is open for all .j ∈ I . Indeed, we have .qj−1 f −1 (U ) = (f ◦
qj )−1 (U ) = fj−1 (U ) and since each .fj is continuous, the desired conclusion
follows.
(11) For certain categories, such as Grp, the coproducts are more complicated
than the products and the constructions do not rely on the ones performed
in Set. This is basically because unions do not usually preserve oper-
ations (for instance, the union of an arbitrary family of groups is not
necessarily a group). In group theory, the construction which gives the
coproducts is called
 the free product of groups (see [50, Chapter 11, p.
388]). Let . Gi i∈I be a family of groups, where I is a set, and denote by
 
. ∗i∈I Gi , (jk : Gk → ∗i∈I Gi , )k∈I the corresponding free product. Consider

now .G ∈ Ob Grp and a family of group homomorphisms .fk : Gk → G,

.k ∈ I . The definition of the free product yields a unique group homomorphism

.f : ∗i∈I Gi → G such that .f ◦ jk = fk for all .k ∈ I . Therefore,

   
. ∗i∈I Gi , (jk : Gk → ∗i∈I Gi , )k∈I is the coproduct of the family . Gi i∈I
in Grp.
(12) In the category Ab of abelian groups the coproduct is given by the direct
sum with componentwise multiplication law. More precisely, for any family
2.1 (Co)products, (Co)equalizers, Pullbacks and Pushouts 91

. (Ai )i∈I of abelian groups we have


. Ai = {(ai )i∈I | ai ∈ Ai , {i | ai = 0} is finite},
i∈I

qi0 : Ai0 → Ai , qi0 (a) = (ai )i∈I ,
i∈I

where .ai0 = a and .aj = 0 for all .j = i0 . Indeed, consider the pair
 
. H, (fi )i∈I , where H is an abelian group and .fj : Aj → H are group
homomorphisms
 for all .j ∈ I . Then, the unique homomorphism of groups
.f : i∈I iA → H which makes the following diagram commutative for all
.j ∈ I :

  
is given by .f (ai )i∈I = i∈I fi (ai ).1 Suppose now that .g : i∈I Ai → H
is another group homomorphism such that .g ◦ qj = fj for all .j ∈ I .
Then .(f − g) ◦ qj is the zero map from .Aj to H and thus the image
of .qj is contained  in .ker(f − g) for all .j ∈ I . Now observe that
any element in . i∈I Ai is a sum of finitely many elements  of the
form .qj (aj ). Therefore, since .ker(f − g) is a subgroup in . Ai , we
 i∈I
obtain that .ker(f − g) = i∈I Ai , i.e., .f = g and the proof is now
finished.
An analogous description of coproducts holds for the category .R M of
modules over a ring R.
(13) Let PO.(X, ) be the category corresponding to the pre-ordered set .(X, ).
Then, the coproduct of a family .(xi )i∈I , if it exists, is just its supremum.
.

Another important example of a limit is an equalizer.

Definition 2.1.6 An equalizer of the morphisms f , .g ∈ HomC (X, Y ) is a pair
(E, p), where
.

(1) .E ∈ Ob C;
(2) .p ∈ HomC (E, X) such that .f ◦ p = g ◦ p,

1 Note that the sum in the right-hand side contains only finitely many non-zero terms.
92 2 Limits and Colimits

and for any other pair .(E  , p ), where

(1) .E  ∈ Ob C;
(2) .p ∈ HomC (E  , X) such that .f ◦ p = g ◦ p ,
there exists a unique .u ∈ HomC (E  , E) which makes the following diagram
commute:

We say that .C is a category with equalizers or that .C has equalizers if any pair of
morphisms in .C with the same domain and codomain has an equalizer.
Next we introduce coequalizers, the dual notion of equalizers.
Definition 2.1.7 A coequalizer of the morphisms f , .g ∈ HomC (X, Y ) is a pair
(Q, q), where
.

(1) .Q ∈ Ob C;
(2) .q ∈ HomC (Y, Q) such that .q ◦ f = q ◦ g,
and for any other pair .(Q , q  ), where
(1) .Q ∈ Ob C;
(2) .p ∈ HomC (E  , X) such that .f ◦ p = g ◦ p ,
there exists a unique .v ∈ HomC (Q, Q ) which makes the following diagram
commute:

We say that .C is a category with coequalizers or that .C has coequalizers if any pair
of morphisms in .C with the same domain and codomain has a coequalizer.
At this point we only state the uniqueness up to isomorphism of (co)equalizers.
The proof relies on the same idea used in proving Proposition 2.1.3. Furthermore,
this uniqueness result will follow as a special case of Proposition 2.2.9.
Proposition 2.1.8 When it exists, the (co)equalizer of two morphisms is unique up
to isomorphism.
2.1 (Co)products, (Co)equalizers, Pullbacks and Pushouts 93

Proposition 2.1.9 If .(E, p) is the equalizer (resp. coequalizer) of the pair of

morphisms f , .g ∈ HomC (X, Y ) in a category .C, then p is a monomorphism (resp.
epimorphism).
Proof Consider .h1 , .h2 : E  → E such that .p ◦ h1 = p ◦ h2 := h. First notice that
.f ◦ h = f ◦ (p ◦ h1 ) = (f ◦ p) ◦ h1 = (g ◦ p) ◦ h1 = g ◦ (p ◦ h1 ) = g ◦ h. Since

p is the equalizer of f and g, there exists a unique morphism .u : E  → E such that

.p ◦ u = h.

Now notice that both maps .h1 , .h2 : E  → E fulfill the above equality. Due to the
uniqueness of u we obtain .u = h1 = h2 , as desired.
For the second part we use duality. Indeed, .(E, p) is the coequalizer of the pair
of morphisms f , .g : X → Y in category .C if and only if .(E, pop ) is the equalizer of
the pair of morphisms .f op , .g op : Y → X in the category .C op . The conclusion now
follows from the first part of the proof and Proposition 1.4.3, (1). 

Examples 2.1.10
(1) In Set the equalizer of two functions f , .g : X → Y is the pair .(E, i), where
.E = {x ∈ X | f (x) = g(x)} and .i : E → X is the canonical inclusion. Indeed,

consider f , .g ∈ HomSet (X, Y ) and suppose .j : E  → X is a morphism in Set

such that .f ◦ j = g ◦ j .

Then .Im(j ) ⊆ E and the unique map .u : E  → E such that .i ◦ u = j is given

by .u(e) = j (e) for all .e ∈ E  .
(2) In Grp, Ab, .R M the underlying set of the equalizer is constructed as in
the previous example. Consider for example the category Grp and let f ,
.g ∈ HomGrp (G, H ). Then .E = {x ∈ G | f (x) = g(x)} is a subgroup

of G with respect to the induced structure and the inclusion .i : E → G is

obviously a group homomorphism. Similar arguments can be used for the other
two categories Ab and .R M.
94 2 Limits and Colimits

(3) Let f , .g ∈ HomTop (X, Y ) and endow the set E constructed in the first example
with the subspace topology.2 This implies that the inclusion .i : E → X is a
continuous map and therefore .(E, i) is the equalizer of the pair of morphisms
.(f, g) in Top.

(4) Equalizers in Haus and KHaus are constructed as in Top. Indeed, assume first
that f , .g ∈ HomHaus (X, Y ) and consider .E = {x ∈ X | f (x) = g(x)} endowed
with the subspace topology. We only need to show that E is a Hausdorff space.
To this end, let .x1 , .x2 ∈ E such that .x1 = x2 . As X is a Hausdorff space, we can
find two disjoint neighborhoods of .x1 and .x2 in X, say .U1 and .U2 respectively.
By definition of the subspace topology, the sets .U1 ∩ E and .U2 ∩ E are open
in E and therefore neighborhoods in E of .x1 and .x2 , respectively. Furthermore,
we have .(U1 ∩ E) ∩ (U2 ∩ E) = (U1 ∩ U2 ) ∩ E = ∅, which proves that E is
Hausdorff.
Consider now u, .v ∈ HomKHaus (K, H ). We have already proved that E
together with the subspace topology is Hausdorff; we are left to show that E
is compact as well. It will suffice to prove that E is closed in X as any closed
subspace of a compact space is compact as well [39, Theorem 26.2]. To this end,
let .x0 ∈ X − E. As .f (x0 ) = g(x0 ) and Y is Hausdorff we can find two disjoint
open sets U and V such that .f (x0 ) ∈ U and .g(x0 ) ∈ V . Then .T = f −1 (U )
and .W = g −1 (V ) are open sets in X by continuity of f and g. Consequently
.S = T ∩ W is also an open set in X and .x0 ∈ S. Furthermore, S and E are

disjoint; indeed, if there exists some .x ∈ S ∩ E then we have .f (x) ∈ U ,

.g(x) ∈ V and .f (x) = g(x), which is an obvious contradiction since this would

imply .f (x) ∈ U ∩ V = ∅.
To summarize, we have proved that for each .x0 ∈ X − E there exists an open
set S such that .x0 ∈ S ⊂ X − E. This shows that .X − E is contained in the
interior3 of .X − E, which allows us to conclude that .X − E = Int(X − E).
Hence .X − E is an open subset of X and therefore E is a closed set, as desired.
(5) Let f , .g ∈ HomSet (X, Y ) be two functions. Consider .R = {(f (x), g(x)) | x ∈
 let .∼R be the equivalence relation on Y generated by R. Then the
X} and 4

pair . Y /∼R , π is the coequalizer of the maps f and g in Set, where .Y /∼R is
the set of equivalence classes of Y with respect to .∼R , while .π : Y → Y /∼R ,
.π(y) = y, for all .y ∈ Y , is the canonical projection. To start with, for any

.x ∈ X, we have .π ◦ f (x) = f (x) = g(x) = π ◦ g(x). Assume now that

.q : Y → Q is another map such that .q ◦ f = q ◦ g and define .v : Y /∼R → Q

by .v(y) = q(y) for all .y ∈ Y /∼R . First we will show that v is well-defined.
To this end, let y, .y  ∈ Y such that .y = y  . Then .y ∼R y  and this implies that

2 Let .(X, τ ) be a topological space and .Y ⊆ X a subset of X. Then .τY = {Y ∩ U | U ∈ τ } is a

topology on Y called the subspace topology [39, p. 88].
3 Recall that given a subset U of a topological space X, the interior of U , denoted by Int(U), is

defined as the union of all open sets contained in U [39, p. 95]. Moreover, we have .Int(U ) ⊂ U .
4 The equivalence relation generated by a binary relation R on a set Y (regarded as a subset of

.Y × Y ) is defined as the intersection of all equivalence relations on Y which contain R.

2.1 (Co)products, (Co)equalizers, Pullbacks and Pushouts 95

there exists some positive integer n and .y0 , .y1 , . . . , yn ∈ Y such that .y = y0 ,
.y  = yn and for any .i = 1, 2 . . . , n we have either .yi ∼ yi+1 or .yi+1 ∼ yi .
Furthermore, note that if .yi ∼ yi+1 then .yi = f (x0 ), .yi+1 = g(x0 ) for some
.x0 ∈ X, which implies .q(yi ) = q ◦ f (x0 ) = q ◦ g(x0 ) = q(yi+1 ). Therefore, we

obtain .q(y0 ) = q(y1 ) = . . . = q(yn ), which leads to .q(y) = q(y  ). This shows
that v is well-defined. Obviously we have .v ◦ π(y) = q(y) for all .y ∈ Y . If
  
.v : Y /∼R → Q such that .v ◦ π = q then we easily obtain .v (y) = v ◦ π(y) =


q(y) = v ◦ π(y) = v(y) for all .y ∈ Y . Thus .v = v , which completes the proof.
(6) Let f , .g ∈ HomTop (X, Y ) and let .π : Y → Y /∼R be the canonical projection
constructed in the previous example. Now we endow  .Y /∼R with the quotient

topology5 with respect to .π and we will show that . Y /∼R , π is the coequalizer
of the pair .(f, g) in Top. In particular, .π : Y → Y /∼R is a continuous map.
Consider now .Q ∈ Ob Top and another continuous map .q : Y → Q such that
.q ◦ f = q ◦ g. We only need to prove that the unique map .v : Y /∼R → Q

constructed in the previous example is continuous. To this end, let .U ⊆ Q be

an open subset. Note that since .Y /∼R is endowed with the finest topology that
−1
makes .π into a continuous map, in order to prove that
−1 −1
 .v (U ) ⊆ Y /∼R is an
open subset it will suffice to show that .π v (U ) ⊆ Y is an open set. To
 
this end, we have .π −1 v −1 (U ) = (π ◦ v)−1 (U ) = q −1 (U ) and since q is
continuous, the desired conclusion follows. 
(7) Let f , .g ∈ HomHaus (X, Y ) and consider . Y /∼R , π to be the coequalizer in
Top constructed in the previous example. However, as the quotient topology
does not behave well with respect to the Hausdorff property,6 the coequalizer
in Haus will be constructed by using the so-called Hausdorff quotient of a
topological space (we refer to [40, 44] for further details). To this end, if
we denote by .(H, u) the Hausdorff quotient of .Y /∼R , then .(H, q) is the
coequalizer in Haus of the pair .(f, g), where .q = u ◦ π .

Indeed, to start with, we have .q ◦ f = u ◦ π ◦ f = u ◦ π ◦ g = q ◦ g,

where in the second equality we used the fact that . Y /∼R , π is the coequalizer
in Top of .(f, g). Consider now another Hausdorff 
   
 space .Q and a morphism
.q ∈ HomHaus (Y, Q ) such that .q ◦f = q ◦g. As . Y /∼R , π is the coequalizer

5 Let X be a topological space, Y a set and .f : X → Y a surjective map. The quotient topology
with respect to f is the finest topology on Y such that f is continuous. In other words, .U ⊆ Y is
open if and only if .f −1 (U ) is open in X [39, p. 138].
6 See, for instance, [38, Examples 6.4.17, 6.4.19].
 96 2 Limits and Colimits

 
in Top of .(f, g), there exists a unique morphism .t ∈ HomTop Y /∼R , Q such
that .t ◦ π = q  . Now the universal property of the Hausdorff quotient (see [40])
yields a unique .v ∈ HomHaus (H, Q ) such that .v ◦ u = t. If we put everything
together we obtain .v ◦ q = v ◦ u ◦ π = t ◦ π = q  . We are left to show that v
is the unique morphism with this property. Indeed, assume there exists another
    
.v ∈ HomHaus (H, Q ) such that .v ◦ q = q ; then we have .(v ◦ u) ◦ π = q and


since t is the unique morphism with this property it follows that .v ◦ u = t and
therefore .v = v  .
(8) Let f , .g ∈ HomGrp (G, H ) and consider .H  = {f (x)g(x)−1 | x ∈ X} ⊆ H .
If we denote by .N = H  ⊆U H U the normal subgroup generated by .H  then
.(H /N, π ) is the coequalizer of the pair of morphisms .(f, g) in the category

Grp of groups, where .π : H → H /N is the canonical projection. Indeed, since

.f (x)g(x)
−1 ∈ H  ⊆ H for all .x ∈ G, we have .f  (x) = g(x) in .H /N, which
comes down to .π ◦f = π ◦g. Consider now   
.q ∈ HomGrp (H, Q ) such that .q ◦
 
 −1

f = q ◦ g. This yields .q f (x)g(x) = 1 for all .x ∈ G. Therefore, we have
  
.H ⊆ ker(q )  H and thus .N ⊆ ker(q ). Now from the universal property of
 
the quotient group .H /N we obtain a unique morphism .v ∈ HomGrp H /N, Q
such that the following diagram is commutative:

i.e., .(H /N, π ) is the coequalizer of the pair of morphisms .(f, g) in Grp.
(9) Let f , .g ∈ HomAb (A, B). Then the map .(f − g) : A → B defined by .(f −
g)(a) = f (a) − g(a) for all .a ∈ A is a morphism of groups and, therefore,
the set .N = {f (a) − g(a) | a ∈ A} is a subgroup of B. It can  be easily
 seen,
using the same arguments as in the previous example, that . B/N, π is the
coequalizer of the pair of morphisms .(f, g) in Ab. Coequalizers in .R M have a
similar description.
(10) Let G be a non-trivial group and .G the associated category as described in
Example 1.2.2, (3). If x, .y ∈ G such that .x = y then the pair of morphisms
.(x, y) does not admit a (co)equalizer in .G. Indeed, this follows easily by

noticing that there are no elements .z ∈ G such that .xz = yz or .zx = zy.
.

In a pointed category we can introduce an important special case of

(co)equalizers. First, recall from Remark 1.3.8 that in such categories there exists a
morphism, called the zero-morphism, between any two given objects.
2.1 (Co)products, (Co)equalizers, Pullbacks and Pushouts 97

Definition 2.1.11 Let .C be a pointed category and .f ∈ HomC (A, B). The
(co)equalizer of the pair of morphisms .(f, 0A, B ) is called the (co)kernel of f .
Example 2.1.12 By applying the above definition to the pointed categories Grp,
Ab and .R M, we recover the familiar algebraic kernel of a morphism defined as the
preimage of the neutral element. .

The last examples we provide before introducing (co)limits are pullbacks

together with their duals, called pushouts.
Definition 2.1.13 Let .C be a category and .f ∈ HomC (B, A), .g ∈ HomC (C, A). A
pullback of .(f, g) is a triple .(P , f  , g  ), where
(1) .P ∈ Ob C;
(2) .f  ∈ HomC (P , C), .g  ∈ HomC (P , B) such that .f ◦ g  = g ◦ f  ,
and for any other triple .(P  , f  , g  ), where
(1) .P  ∈ Ob C;
(2) .f  ∈ HomC (P  , C), .g  ∈ HomC (P  , B) such that .f ◦ g  = g ◦ f  ,
there exists a unique . ∈ HomC (P  , P ) such that .f  = f  ◦  and .g  = g  ◦ .
The complete picture is captured by the diagram below:

We say that .C is a category with pullbacks or that .C has pullbacks if any pair of
morphisms in .C with the same codomain has a pullback.
Definition 2.1.14 Let .C be a category and .f ∈ HomC (A, B), .g ∈ HomC (A, C).
A pushout of .(f, g) is a triple .(P , f  , g  ), where
(1) .P ∈ Ob C;
(2) .f  ∈ HomC (C, P ), .g  ∈ HomC (B, P ) such that .g  ◦ f = f  ◦ g,
and for any other triple .(P  , f  , g  ), where
(1) .P  ∈ Ob C;
(2) .f  ∈ HomC (C, P  ), .g  ∈ HomC (B, P  ) such that .g  ◦ f = f  ◦ g,
there exists a unique . ∈ HomC (P , P  ) such that .f  =  ◦ f  and .g  =  ◦ g  .
98 2 Limits and Colimits

The complete picture is captured by the diagram below:

We say that .C is a category with pushouts or that .C has pushouts if any pair of
morphisms in .C with the same domain has a pushout.
As in the case of (co)products and (co)equalizers, both pullbacks and pushouts
are unique up to isomorphism:
Proposition 2.1.15 When it exists, the pullback (resp. pushout) of two morphisms
is unique up to isomorphism.
Examples 2.1.16
(1) In Set the pullback of two morphisms .f : B → A, .g : C → A is given by the
triple .(B ×A C, π C , π B ), where .B ×A C = {(b, c) ∈ B × C | f (b) = g(c)}
and .π C : B ×A C → C, .π B : B ×A C → B are given by

π C (b, c) = c,
. π B (b, c) = b, for all (b, c) ∈ B ×A C.

First note that .f ◦ π B = g ◦ π C . Consider now .P  ∈ Ob Set and .f  ∈

HomSet (P  , C), .g  ∈ HomSet (P  , B) such that .f ◦ g  = g ◦ f  . Then
   
.(g (x), f (x)) ∈ B ×A C for all .x ∈ P and we can define . : P → B ×A C
 
by .(x) = (g (x), f (x)). Moreover, we have

π C ◦ (x) = f  (x),
.

π B ◦ (x) = g  (x).

As .π B and .π C are the restrictions to .B ×A C of the projections on B and C,

respectively, . is obviously the unique morphism which renders the diagram
2.1 (Co)products, (Co)equalizers, Pullbacks and Pushouts 99

below commutative:

(2) Let .f ∈ HomGrp (B, A), .g ∈ HomGrp (C, A) and consider the set .B ×A C
constructed in the previous example endowed with the group structure given
by the direct product. Then, the two projections .π C : B ×A C → C, .π B : B ×A
C → B defined in the previous example are group homomorphisms and .(B×A
C, π C , π B ) is the pullback of the pair of morphisms .(f, g). Indeed, let .P  ∈
Ob Grp, .f  ∈ HomGrp (P  , C), .g  ∈ HomGrp (P  , B) such that .f ◦ g  =
g◦f  . The only thing left to check is that the map . : P  → B ×A C defined in
the previous example is a group homomorphism. To this end, for all x, .y ∈ P 
we have .(xy) = (g  (xy), f  (xy)) = (g  (x), f  (x))(g  (y), f  (y)) =
(x)(y). Furthermore, .(1P ) = (g  (1P ), f  (1P )) = (1B , 1C ) and the
desired conclusion follows.
(3) Let .f ∈ HomTop (B, A), .g ∈ HomTop (C, A) and consider the set .B ×A C
constructed in the first example. Furthermore, we see .B × C as a topological
space with respect to the product topology and endow  .B ×A C ⊆ B × C
with the subspace topology. Then . B ×A C, π C , π B is the pullback in Top
of the pair of morphisms .(f, g). Indeed, note first that both .π B = πB ◦ i
and .π C = πC ◦ i are compositions of continuous maps and are therefore
continuous, where .i : B×A C → B×C is the inclusion map and .πB , .πC denote
the projection maps. Consider now .P  ∈ Ob Top and .f  ∈ HomTop (P  , C),
   
.g ∈ HomTop (P , B) such that .f ◦ g = g ◦ f . We are left to show that the

map . : P → B ×A C defined in the first example is continuous. To this end,
recall that the topology on .B ×A C is generated by the sets .π −1 −1
B (U ), .π C (V )
−1 −1
for all open sets .U ⊆ B and .V ⊆ C. In other words, the sets .π B (U ), .π C (V ),
where .U ⊆ B and .V ⊆ C are open sets, form a sub-basis for the topology
on .B ×A C. Therefore, in order to show that . is a continuous map it will
suffice to show that the preimage of any of these sets through . is an open set.
Indeed, for all open sets .U ⊆ B, .V ⊆ C we have

−1 ◦ π −1
.
−1
B (U ) = (π B ◦ ) (U ) = g
−1
(U ),

−1 ◦ π −1 −1
C (V ) = (π C ◦ ) (V ) = f
−1
(V ).

The conclusion now follows easily as both .f  and .g  are continuous maps.
100 2 Limits and Colimits

(4) Assume R is a commutative ring and .R M is the category of left modules over
R. Let .f ∈ HomR M (B, A), .g ∈ HomR M (C, A) and consider the submodule
.B ×A C = {(b, c) ∈ B × C | f (b) = g(c)} of the direct product .B × C. Then
 
. B ×A C, π C , π B is the pullback in .R M of the pair of morphisms .(f, g),

where .π C : B ×A C → C and .π B : B ×A C → B are the restrictions of the

canonical projections. It can be easily checked, as in the previous examples,
that .f ◦π B = g ◦π C . Consider now .P  ∈ Ob R M and .f  ∈ HomR M (P  , C),
   
.g ∈ Hom M (P , B) such that .f ◦ g = g ◦ f . We are left to show that the
R

map . : P → B ×A C defined as follows is R-linear:

(x) = (g  (x), f  (x)),

. x ∈ P .

Indeed, since both maps .f  and .g  are R-linear, for all r, .s ∈ R and x, .y ∈ P 
we have
 
(rx + sy) = g  (rx + sy), f  (rx + sy)
.
 
= rg  (x) + sg  (y), rf  (x) + sf  (y)
   
= rg  (x), rf  (x) + sg  (y), sf  (y)
   
= r g  (x), f  (x) + s g  (y), f  (y)
= r(x) + s(y),

as desired.
(5) Let .f ∈ HomSet (A, B), .g ∈ Hom Set (A, C) and consider the disjoint union
of the sets B and C, denoted by .B C = {(b, 0) | b ∈ B} ∪ {(c,  1) | c ∈ C},
together with the corresponding inclusion maps .j0 : B → B C, .j1 : C →

B C defined by .j0 (b) = (b,  0), .j1 (c) = (c, 1) for all .b ∈ B, .c ∈ C. Define
     
.R = { (f (a), 0), (g(a), 1) | a ∈ A} ⊆ B C × B C and let .∼R be

the equivalence relation on .B C generated    R. Then the pushout of the
by
pair .(f, g) is 
given
 by the quotient    C /∼R together with the maps
set . B

.f : C → B C /∼R , .g  : B → B C /∼R defined as follows:

f  = π ◦ j1 ,
. g  = π ◦ j0 ,
   
where .π : B C → B C /∼R is the canonical projection. Indeed, first
note that for all .a ∈ A we have .g  ◦ f (a) = π(f (a), 0) = π(g(a), 1) =
f  ◦ g(a). Consider now .P  ∈ Ob Set and .f  ∈ HomSet (C,  
 P ), .g  ∈
  
HomSet (B, P ) such that .g ◦ f = f ◦ g and define .ψ : B C → P as
follows:

f  (c) if x = (c, 1) for some c ∈ C,

ψ(x) =
. (2.6)
g  (b) if x = (b, 0) for some b ∈ B.
2.1 (Co)products, (Co)equalizers, Pullbacks and Pushouts 101

  
  
 
Then  .a ∈ A we have .ψ f (a), 0 = g f (a) = f g(a) =
 for all
 1 . The universal
ψ g(a), property of the quotient set yields a unique map
. : B C /∼R → P  such that . ◦ π = ψ. Furthermore, for all .b ∈ B and
.c ∈ C we have

 ◦ f  (c) =  ◦ π ◦ j1 (c) = ψ(c, 1) = f  (c),

.

 ◦ g  (b) =  ◦ π ◦ j0 (b) = ψ(b, 0) = g  (b).

  
Assume there exists another map . : B C /∼R → P  such that . ◦ f  =
 
f  and .◦g  = g  . Then, for all .b ∈ B we obtain . (b, 0) = ◦π(b, 0) =
 
 ◦ π ◦ j0 (b) =  ◦ g  (b) = g (b)=   
 ◦ π ◦ j0 (b) =  (b, 0) , where .(b, 0)
denotes the class of .(b, 0) in . B C /∼R . Similarly, one can easily see that
   
 1) =  (c, 1) for all .c ∈ C and therefore
we also have . (c,   .  = .
(6) Let .f ∈ HomTop (A, B), .g ∈ HomTop (A, C) and let . B C /∼R be the
set constructed
 in
 the previous
 example together
 with the projection map
.π : B C → B C /∼R . We endow .B C with the finest topology 
for which the inclusion maps .j0 : B → B C  and .j1 : C → B C are
continuous maps and we consider on . B C /∼R the quotient topology.
Furthermore, note that both .f  = π ◦ j1 and .g  = π ◦ j0 are continuous as 
  
compositions of continuous maps. We will show that . B C /∼R , (f  g  )
is the pushout of the pair .(f, g). Indeed, consider .P  ∈ Ob Top and .f  ∈
HomTop (C, P  ), .g  ∈ HomTop (B, P  ) such  that .g  ◦ f = f  ◦ g. We
are left to show that the unique map . : B C /∼R → P  such that
   
. ◦ f = f and . ◦ g = g defined in the previous example is continuous.

  
To this end, let .U ⊆ P be an open set. Note that since . B C /∼R
is endowed with the finest topology that  makes
 .π into a continuous map,

in order to prove that .−1 (U ) ⊆ B C /∼R is an open subset it will

  
suffice to show that .π −1 −1 (U ) ⊆ B C is an open subset. We have
 
−1 −1 (U ) = ( ◦ π )−1 (U ) = ψ −1 (U ), where .ψ is the map defined
.π
−1

in (2.6). Our claim will be proved  once we show that .ψ (U ) ⊆ B C is
an open subset. Indeed, since .B C is endowed with the finest topology for
which the canonical inclusions .j0 and .j1 are continuous maps, we only need
 
to show that .j0−1 ψ −1 (U ) ⊆ B and .j1−1 ψ −1 (U ) ⊆ C are open subsets.
   
We have .j0−1 ψ −1 (U ) = (ψ ◦ j0 )−1 (U ) = (g  )−1 (U ) and .j1−1 ψ −1 (U ) =
(ψ ◦ j1 )−1 (U ) = (f  )−1 (U ) and the desired conclusion now follows as both
 
.f and .g are continuous maps.

(7) Let .f ∈ HomGrp (A, B), .g ∈ HomGrp (A, C) and let .B  C be the
free product of the groups B and C together with the corresponding group
homomorphisms .iB : B → B  C and .iC : C → B  C. Furthermore, we let

.N = {iB (f (a))
−1 i (g(a)) | a ∈ A} and denote by N the normal subgroup of
C

.B  C generated by .N . Then the quotient group .(B  C)/N together with

the group homomorphisms .f  : C → (B  C)/N, .g  : B → (B  C)/N

102 2 Limits and Colimits

is the pushout of the pair .(f, g), where .f  = π ◦ iC , .g  = π ◦ iB and

.π : B  C → (B  C)/N is the canonical projection. Indeed, having in mind

that .N  ⊆ N, we have
   
(f  ◦ g)(a) = π iC (g(a)) = π iB (f (a)) = (g  ◦ f )(a) for all a ∈ A.
.

Consider now .P  ∈ Ob Grp and .f  ∈ HomGrp (C, P  ), .g  ∈ HomGrp (B, P  )

such that .g  ◦ f = f  ◦ g. The definition of the free product of groups ([50,
Chapter 11, p. 388]) yields a unique group homomorphism . : B  C → P 
such that the following diagram is commutative:

(2.7)
We show first that .N  ⊆ ker . Indeed, for all .a ∈ A we have
 −1
  −1  
. iB (f (a)) iC (g(a)) =  iB (f (a)  iC (g(a)
= g  (f (a))−1 f  (g(a)) = 1.

Therefore, .N ⊆ ker  and the universal property of the quotient groups yields
a unique group homomorphism . : (B  C)/N → P  such that . ◦ π = .
We obtain

 ◦ f  =  ◦ π ◦ iC =  ◦ iC = f  ,
.

 ◦ g  =  ◦ π ◦ iB =  ◦ iB = g  .

We are left to show that . is the unique group homomorphism with this
property. Assume . : (B  C)/N → P  is another group homomorphism
such that . ◦ f  = f  and . ◦ g  = g  . This leads to . ◦ π ◦ iC = f 
and . ◦ π ◦ iB = g  and since . is the unique group homomorphism which
makes diagram (2.7) commute, we obtain . ◦ π = . Now by the universal
property of the quotient groups, . is the unique group homomorphism with
this property and therefore . = .
(8) Assume R is a commutative ring and .R M is the category of left modules over
R. Let .f ∈ HomR M (A, B), .g ∈ HomR M (A, C) and consider  the submodule
.S = {(f (a), −g(a)) | a ∈ A} of .B × C. Then the triple . (B × C)/S, f , g
 

is the pushout in .R M of the morphisms above, where .(B × C)/S denotes

the quotient module corresponding to S and .f  : C → (B × C)/S, .g  : B →
2.1 (Co)products, (Co)equalizers, Pullbacks and Pushouts 103

(B × C)/S are given, for any .b ∈ B and .c ∈ C, as follows:

f  (c) = (0, c),

. g  (b) = (b, 0).
     
Indeed, since . f (a), 0 − 0, g(a) = f (a), −g(a) ∈ S, we get
      
. f (a), 0 = 0, g(a) and thus .g ◦ f = f ◦ g. Consider now .P ∈ Ob R M

and .f ∈ HomR M (C, P ), .g ∈ HomR M (B, P ) such that .g ◦ f = f  ◦ g.

    

The map defined for all .(b, c) ∈ B × C as follows:

χ : B × C → P ,
. χ (b, c) = g  (b) + f  (c),

is a morphism in .R M and moreover, .χ (S) = 0 since we have

 
χ f (a), −g(a) = g  (f (a)) − f  (g(a)) = 0
.

for all .a ∈ A. Now the universal property of the quotient module

 yields
 a
unique morphism . ∈ HomR M (B × C)/S, P  such that . (b, c) =
g  (b) + f  (c) for all .(b, c) ∈ B × C. Moreover, we have

  
.( ◦ g )(b) =  (b, 0) = g (b),
 
( ◦ f  )(c) =  (0, c) = f  (c),

i.e., the following diagram is commutative:

We are left to prove the uniqueness of .. Let .Υ : (B × C)/S → P  such that
   
.Υ ◦ g = g and .Υ ◦ f = f . To this end, for all .(b, c) ∈ B × C, we have

     
.Υ (b, c) = Υ (b, 0) + Υ (0, c)

= Υ ◦ g  (b) + Υ ◦ f  (c)
= g  (b) + f  (c)
 
=  (b, c) .

(9) Let PO.(X, ) be the category corresponding to the pre-ordered set .(X, )
and a, b, .c ∈ X such that .a ≤ b and .a ≤ c. If it exists, the pushout of the
above maps is some element .p ∈ X satisfying:
104 2 Limits and Colimits

• .b ≤ p and .c ≤ p;
.

• for any .x ∈ X such that .b ≤ x and .c ≤ x we have .p ≤ x.

.

In other words, if it exists, the pushout of the maps .b ≤ a and .c ≤ a is given

by the supremum of b and c. Similarly, if it exists, the pullback of two maps
above is given by the infimum of b and c.
(10) Let .M = End(X) denote the endomorphisms on the set .X = {x, y}. More
precisely, .M = {1X , τ, ψx , ψy } is a monoid with respect to the composition
of maps, where .τ , .ψx , .ψy : M → M are defined as follows:

τ (x) = y, τ (y) = x,
.

ψx (x) = ψx (y) = x,
ψy (x) = ψy (y) = y.

If .M denotes the category associated to the monoid M as in Example 1.2.2, (3)

then the pair of morphisms .(ψx , ψy ) does not have an equalizer in .M.
op op
Consequently, in .Mop the pair of morphisms .(ψx , ψy ) does not have a
coequalizer. .

The next two results highlight the different connections existing between
monomorphisms (resp. epimorphisms) and pullbacks (resp. pushouts) and will
be useful in the sequel.
Proposition 2.1.17 Let .C be a category.

(1) Let . X, u ∈ HomC (X, A), v ∈ HomC (X, C) be the pullback of the pair of
morphisms .f ∈ HomC (A, B) and .g ∈ HomC (C, B). If g is a monomorphism
then u is also a monomorphism. 
(2) Let . Y, u ∈ HomC (B, Y ), v ∈ HomC (C, Y ) be the pushout of the pair of
morphisms .f ∈ HomC (A, B) and .g ∈ HomC (A, C). If g is an epimorphism
then u is also an epimorphism.
Proof .(1) Let .Y ∈ Ob C and assume there exists .α, .β ∈ HomC (Y, X) such that
u ◦ α = u ◦ β. In particular, this leads to .f ◦ u ◦ α = f ◦ u ◦ β and since .f ◦ u = g ◦ v
.

we also have .g ◦ v ◦ α = g ◦ v ◦ β.

.
2.2 (Co)limit of a Functor. (Co)complete Categories 105

As g is a monomorphism, we obtain .v ◦ α = v ◦ β. Furthermore, this implies

.f ◦ u ◦ α = g ◦ v ◦ β. Since .(X, u, v) is the pullback of the pair .(f, g), there exists
a unique morphism .γ ∈ HomC (Y, X) such that .u ◦ γ = u ◦ α and .v ◦ β = v ◦ γ . As
both morphisms .α and .β satisfy this property we obtain .α = β, which shows that u
is indeed a monomorphism. 
.(2) By duality . Y, u
op ∈ Hom op (Y, B), v op ∈ Hom op (Y, C) is the pullback
C C
in .C op of the pair of morphisms .(f op , g op ). If g is an epimorphism in .C then .g op is
a monomorphism in .C op and by .1) we obtain that .uop is also a monomorphism in
.C . Therefore, u is an epimorphism in .C, as desired. 

op

Proposition 2.1.18 Let .C be a category and .f ∈ HomC (A, B).

(1) f is a monomorphism if and only if .(A, 1A , 1A ) is the pullback of the pair of
morphisms .(f, f ).
(2) f is an epimorphism if and only if .(B, 1B , 1B ) is the pushout of the pair of
morphisms .(f, f ).
Proof .(1) Assume first that f is a monomorphism and let u, .v ∈ HomC (P , A) such
that .f ◦ u = f ◦ v. As f is a monomorphism it follows that .u = v and therefore
we have a unique morphism in .HomC (P , A), namely u, which makes the following
diagram commutative:

This shows that .(A, 1A , 1A ) is the pullback of the pair of morphisms .(f, f ).
Conversely, assume that .(A, 1A , 1A ) is the pullback of the pair of morphisms
.(f, f ) and let u, .v ∈ HomC (P , A) such that .f ◦ u = f ◦ v. Hence, we have a

unique morphism .w ∈ HomC (P , A) such that .1A ◦ w = u and .1A ◦ w = v. This

shows that .u = v = w and therefore f is a monomorphism.
.(2) Follows easily by duality; indeed, applying .1) to the morphism .f
op gives the

desired conclusion. 


2.2 (Co)limit of a Functor. (Co)complete Categories

Following the general pattern induced by the previous constructions (i.e.,

(co)products, (co)equalizers and pullbacks/pushouts) we can now introduce the
106 2 Limits and Colimits

concepts of limit and colimit, which unify all the above. We start by introducing the
following:
Definition 2.2.1 Let .F : I → C be a functor.7 A cone on F consists of the
following:
(1) .C ∈ Ob C;
(2) for every .i ∈ Ob I , a morphism .si ∈ HomC (C, F (i)),
such that for any morphism .d ∈ HomD (i, j ), the following diagram is commuta-
tive:

The object C is called the vertex of the cone. If I is a small category and .C ∈ Ob C,
we denote by
   
.Cone(F, C) = { si ∈ HomC (C, F (i)) | C, (si )i∈Ob I is a cone on F }
i∈Ob I

the set8 of cones on F with vertex C.

The dual notion to a cone is called a cocone:

Definition 2.2.2 Let .F : I → C be a functor. A cocone on F consists of the
following:
(1) .C ∈ Ob C,
(2) for every .i ∈ Ob I , a morphism .ti ∈ HomC (F (i), C),
such that for any morphism .d ∈ HomD (i, j ), the following diagram is commuta-
tive:

The object C is called the vertex of the cocone. If I is a small category and .C ∈ Ob C,
we denote by

7 Thecategory I will almost always be considered small.

8 Note
 that .Cone(F, C) is indeed a set since .Ob I is a set and .Cone(F, C) ⊆
i∈Ob I HomC (C, F (i)).
2.2 (Co)limit of a Functor. (Co)complete Categories 107

   
Cocone(F, C) = { ti ∈ HomC (F (i), C) i∈Ob I | C, (ti )i∈Ob I is a cocone on F }
.

the set of cocones on F with vertex C.

The two notions are dual to each other in the following precise sense:
 
Lemma 2.2.3  Letop.F : I → C be a functor. Then . L, (pi )i∈ObopI isopa coneopon F if
and only if . L, (pi )i∈Ob I is a cocone on the dual functor .F : I → C .
 
Proof Let . L, (pi )i∈Ob I be a cone on F . Then for all .d ∈ HomI (i, j ) we have
op op op
.F (d) ◦ pi = pj . Equivalently, this means that .p
i ◦  F (d ) = pj for all
op op

op ∈ Hom op (j, i), which shows that . L, (p op ) op
.d I i i∈Ob I is a cocone on .F . The
op
converse follows by replacing F by .F in the argument above. 

A much more elegant description of (co)cones can be obtained by using natural
transformations:
Proposition 2.2.4 Let .F : I → C be a functor.
 
(1) The pair . C, (si )i∈Ob I is a cone on F if and only if .ηC : C → F defined by
i = si , for all .i ∈ Ob I , is a natural transformation.
.η
C
 
(2) The pair . C, (ti )i∈Ob I is a cocone on F if and only if .C η : F → C defined
by .C ηi = ti , for all .i ∈ Ob I , is a natural transformation.
Proof (1) .ηC : C → F is a natural transformation if and only if the following
diagram is commutative for all .f ∈ HomD (i, j ):

. (2.8)

The commutativity of the above diagram

 comes down to .F (f ) ◦ si = sj , which
means precisely that . C, (si )i∈Ob I is a cone on F .
(2) .C η : F → C is a natural transformation if and only if the following diagram
is commutative for all .f ∈ HomD (i, j ):

. (2.9)
108 2 Limits and Colimits

The commutativity of  the above diagram

 comes down to .tj ◦ F (f ) = ti , which
means precisely that . C, (ti )i∈Ob I is a cocone on F . 

Definition 2.2.5 Let .F : I → C be a functor.
   
(1) A morphism between two cones . C, (si )i∈Ob I and . C, (ri )i∈Ob I on F is a
morphism .f ∈ HomC (C, C) such that the following diagram is commutative
for all .i ∈ Ob I :

The cones on F together with morphisms between them as defined above form
a category denoted by .C(F ).    
(2) A morphism between two cocones . C, (ti )i∈Ob I and . C, (ui )i∈Ob I on F is a
morphism .f ∈ HomC (C, C) such that the following diagram is commutative
for any .i ∈ Ob I :

The cocones on F together with morphisms between them as defined above

form a category denoted by .CO(F ).
Under some assumptions, the category of (co)cones is in fact isomorphic to a
certain comma category.
Theorem 2.2.6 Let I be a small category and .F : I → C a functor. If .TF : 1 →
Fun (I, C) denotes the constant functor at F and . : C → Fun(I, C) is the diagonal
functor, then:
(1) the category .C(F ) of cones on F is isomorphic to the comma category .( ↓
TF );
(2) the category .CO(F ) of cocones on F is isomorphic to the comma category
.(TF ↓ ).
Proof Throughout, we denote by .∗ the unique object of the discrete category .1.
.(1) Recall from Theorem 1.8.4 that the objects of the comma-category .( ↓
TF ) are triples .(C, α, ∗), where .C ∈ Ob C and .α : C → F is  a natural
transformation. Proposition 2.2.4, (1) implies that the pair . C, (αi )i∈Ob I is a cone
on F . Furthermore, a morphism between two objects .(C, α, ∗) and .(C  , α  , ∗) is a
2.2 (Co)limit of a Functor. (Co)complete Categories 109

pair .(f, 1∗ ) consisting of a morphism .f ∈ HomC (C, C  ) and the identity morphism
.1∗ on .∗ such that the following diagram commutes:

where .1F denotes the identity natural transformation on the functor F . This leads to
the following equality between natural transformations: .α  ◦ (f ) = α. Hence, for

all .i ∈ Ob I we
 have .αi ◦ f = αi i.e.,
 f is a morphism in .C(F ) between the cones
. C, (αi )i∈Ob I and . C , (α )i∈Ob I . We can now define two functors .U : C(F ) →
i
( ↓ TF ) and .V : ( ↓ TF ) → C(F ) as follows:

It is now straightforward to see that .U ◦ V = 1( ↓TF ) and .V ◦ U = 1C (F ) . Hence

the categories .( ↓ TF ) and .C(F ) are isomorphic.
.(2) Similarly, the objects of the comma-category .(TF ↓ ) are triples .(∗, α, C),
where .C ∈ Ob C and .α : F →  C is a natural transformation. By Proposi-
tion 2.2.4, (2) the pair . C, (αi )i∈Ob I is a cocone on F . Furthermore, a morphism
between two objects .(∗, α, C) and .(∗, α  , C  ) is a pair .(1∗ , f ) consisting of a
morphism .f ∈ HomC (C, C  ) and the identity morphism .1∗ on .∗ such that the
following diagram commutes:

where .1F denotes the identity natural transformation on the functor F . This leads
to the following equality between natural transformations: . (f ) ◦ α = α  . Hence,
for all .i ∈ Ob 
 I we have .f ◦ αi = αi , i.e., f is a morphism in .CO(F ) between
the cocones . C, (αi )i∈Ob I and . C , (αi )i∈Ob I . We can now define two functors
.U : CO(F ) → (TF ↓ ) and .V : (TF ↓ ) → CO(F ) as follows:
 
U C, (αi )i∈Ob I = (∗, α, C),
. U (f ) = (1∗ , f ),
 
V (∗, α, C) = C, (αi )i∈Ob i , V (1∗ , f ) = f.
110 2 Limits and Colimits

It is straightforward to see that .U ◦ V = 1(TF ↓ ) and .V ◦ U = 1CO(F ) , which shows

that the categories .( ↓ TF ) and .CO(F ) are isomorphic. 

We can now introduce (co)limits:
Definition 2.2.7 Let .F : I → C be a functor.
(1) A limit for the functor F is a final object in the category of cones on F ,
i.e.,
 a cone on F denoted by . lim F, (pi )i∈Ob I such that for any other cone
. C, (si )i∈Ob I on F there exists a unique morphism .f ∈ HomC (C, lim F )
such that the following diagram is commutative for any .i ∈ Ob I :

A category .C has (small/finite) limits if any functor .F : I → C has a limit for

any (small/finite) category I . We say that a category .C is complete if it has small
limits.
(2) A colimit for the functor F is an initial object in the category  of cocones
on F , i.e., a cocone  on F denoted
 by . colim F, (qi )i∈Ob I such that for
every other cocone . C, (ti )i∈Ob I on F there exists a unique morphism .f ∈
HomC (colim F, C) such that the following diagram is commutative for any
.i ∈ Ob I :

A category .C has (small/finite) colimits if any functor .F : I → C has a colimit

for any (small/finite) category I . We say that a category .C is cocomplete if it has
small colimits.
In the sequel we will consider solely small (co)limits. Dealing with limits of
functors .F : I → C for arbitrary categories I leads to set-theoretical issues which
are far beyond our scope. The interested reader is referred to [51].
We can now recover all previously introduced special cases of (co)limits: initial
and final objects, (co)products, (co)equalizers, pullbacks and pushouts.
Examples 2.2.8
(1) Consider the empty functor .φ from the empty category to .C. Then the limit of
.φ, if it exists, is just the final object in .C. Indeed, note that a cone on the empty

functor is just an object of .C. Furthermore, any morphism between two such
2.2 (Co)limit of a Functor. (Co)complete Categories 111

cones are just morphisms in .C between the corresponding objects. Therefore,

the category of cones on the empty functor is just the category .C and the limit
is a final object of .C. Analogously, the colimit of .φ, if it exists, is just the initial
object in .C.
(2) Take I to be a small discrete category. Then a functor .F : I → C is
essentially
 nothing but a family  of objects .(Ci )i∈Ob I in .C. A cone on F is a
pair . C, (ui : C → Ci )i∈Ob I and since I is a discrete category, no further
constrains are imposed on the family of morphisms .(ui )i∈Ob I . Now the limit
of F , if it exists, is just a product in .C of the family .(Ci )i∈Ob I . Similarly, the
colimit of F is just a coproduct in .C of the family .(Ci )i∈I .
(3) Consider a category I with two objects .A1 and .A2 and four morphisms .1A1 ,
.1A2 , u, v, where u, .v ∈ HomI (A1 , A2 ) and define the functor .F : I → C as

follows:

F (A1 ) = X,
. F (A2 ) = Y, F (u) = f, F (v) = g.

A
 cone on F consists of an object  .C ∈ Ob C and morphisms
.s ∈ HomC (C, X), t ∈ HomC (C, Y ) such that the following diagrams
are commutative:

As the morphism t is uniquely

 determined by s, a cone on F comes down to a
pair . C, s ∈ HomC (C, X) such that .f ◦ s = g ◦ s.
Therefore, the (co)limit of the functor F defined above, if it exists, is nothing
but the (co)equalizer of the pair of morphisms f , .g : X → Y in .C.
(4) Consider a category I with three objects .A1 , .A2 , .A3 and five morphisms .1A1 ,
.1A2 , .1A3 , u, v, where .u ∈ HomI (A3 , A1 ), .v ∈ HomI (A3 , A2 ) and define the

functor .F : I → C as follows:

F (A1 ) = X,
. F (A2 ) = Y, F (A3 ) = Z,
F (u) = f ∈ HomC (Z, X), F (v) = g ∈ HomC (Z, Y ).

A
 cocone on F consists of an object C together with
 morphisms
.s ∈ HomC (X, C), t ∈ HomC (Y, C), l ∈ HomC (Z, C) such that the
112 2 Limits and Colimits

following diagrams are commutative:

As the morphism l is uniquely determined by s and t, a cocone on F comes

down to a triple . C, s ∈ HomC (X, C), t ∈ HomC (Y, C) such that .s ◦ f =
t ◦ g. The colimit of the functor F defined above, if it exists, is nothing but the
pushout of the pair of morphisms .f : Z → X, .g : Z → Y in .C.
In a similar manner pullbacks can be obtained as a special case of limits.
Consider a category J with three objects .B1 , .B2 , .B3 and five morphisms .1B1 ,
.1B2 , .1B3 , u, v, where .u ∈ HomI (B1 , B3 ), .v ∈ HomI (B2 , B3 ) and define the

functor .F : J → C as follows:

F (B1 ) = X,
. F (B2 ) = Y, F (B3 ) = Z,
F (u) = f, F (v) = g.

It can be easily seen that the limit of the functor F defined above, if it exists, is
nothing but the pullback of the pair of morphisms .f : X → Z, .g : Y → Z in
.C. .

The uniqueness up to isomorphism of initial and final objects in a category proved

in Proposition 1.3.9 implies:
Proposition 2.2.9 When it exists, the (co)limit of a functor is unique up to
isomorphism.
Remark 2.2.10 In light of Example 2.2.8, the uniqueness results stated in Propo-
sitions 2.1.3, 2.1.8 and 2.1.15 can now be easily derived from Proposition 2.2.9.
.

Example 2.2.11 The category .Set(⊆) defined in Example 1.2.2, (2) is obviously
not complete, as we have already established in Example 1.3.10, (6) that it has
no final objects (i.e., the empty functor from the empty category to .Set(⊆) does
not have a limit). However, .Set(⊆) is cocomplete. To this  end, consider a functor
.H : I → Set(⊆) where I is a small category. Then . Z, (uH (i), Z )i∈Ob I is the

colimit of H , where .Z = i∈Ob I H (i). Recall that for all sets A, B such that
.A ⊆ B, we denote by .uA, B the unique morphism in .Set(⊆) from A to B. With this
 
in mind, it is straightforward
 to see that . Z,
 (uH (i), Z )i∈Ob I is a cocone on H .
Assume now that . W, (uH (i), W )i∈Ob I is another cocone on H . In particular,
as we have morphisms .uH (i), W : H (i) →  W in .Set(⊆), it follows that .H (i) ⊆ W
for all .i ∈ Ob I . This implies that .Z = i∈Ob I H (i) ⊆ W and we have a unique
morphism in .Set(⊆) between Z and W , namely .uZ, W . Furthermore, for all .i ∈ Ob I
2.2 (Co)limit of a Functor. (Co)complete Categories 113

we have .uZ, W ◦ uH (i), Z = uH (i), W . This shows that .uZ, W is the unique morphism
in .Set(⊆) which makes the following diagram commutative for all .i ∈ Ob I :

 
Hence, . Z, (uH (i), Z )i∈Ob I is the colimit of H . .
 
Lemma  2.2.12 Let .G : I → C be a functor and let . L, i∈Ob I be
 (pi )op  a cone on G.
Then . L, (pi )i∈Ob I is the limit of G if and only if . L, (pi )i∈Ob I is the colimit
of the dual functor .Gop : I op → C op .
 
Proof Assume that . L, (pi )i∈Ob I is the limit of G and, furthermore, consider a
 op 
cocone . M, qi ∈ HomC op (Gop (i), M) i∈Ob I on .Gop . Lemma 2.2.3 implies
    
that . M, qi ∈ HomC (M, G(i)) i∈Ob I is a cone on G and since . L, (pi )i∈Ob I
is its limit, there exists a unique .f ∈ HomC (M, L) such that the following diagram
is commutative for all .i ∈ I :

. (2.10)
op op
In particular, this implies that we also have .f op ◦op pi = qi for all .i ∈ I , i.e., the
following diagram is commutative:

Moreover, the uniqueness of the morphism .f op which makes the above diagram
commutative follows from the uniqueness of the morphism
 which makes
 diagram
op
(2.10) commutative. Therefore, we can conclude that . L, (pi )i∈Ob I is the colimit
of .Gop . 

As an easy consequence of Theorem 2.2.6 we obtain the following:
Corollary 2.2.13 Let I be a small category and .F : I → C a functor. Then:
(1) F has a limit of and only if the comma category .( ↓ TF ) has a final object;
114 2 Limits and Colimits

(2) F has a colimit if and only if the comma category .(TF ↓ ) has an initial
object.
We record here, for further use, the following useful results which generalize
Proposition 2.1.9:
Proposition 2.2.14 Let .F : I → C be a functor and .C ∈ Ob C.
 
(1) If . lim F, (pi )i∈Ob I is the limit of F and f , .g ∈ HomC (C, lim F ) such that
.pi ◦ f = pi ◦ g for all .i ∈ Ob I then .f = g.
 
(2) If . colim F, (qi )i∈Ob I is the colimit of .F : I → C and f , .g ∈
HomC (colim F, C) such that .f ◦ qi = g ◦ qi for all .i ∈ Ob I then .f = g.
 
Proof (1) To start with, note that . C, (pi ◦ f )i∈Ob I is a cone onthe functor F , i.e.,
.F (d) ◦ pi ◦ f = pj ◦ f for any .d ∈ HomI (i, j ). Indeed, since . lim F, (pi )i∈Ob I

is the limit of the functor F and in particular a cone on F , the following diagram is
commutative:

i.e., .F (d) ◦ pi = pj and the conclusion

 follows by simply
 composing the last
equality on the right by f . Now since . lim F, (pi )i∈Ob I is the limit of the functor
F there exists a unique morphism .h ∈ HomC (C, lim F ) which makes the following
diagram commutative:

As both morphisms f , .g ∈ HomC (C, lim F ) render the above diagram commuta-
tive, the desired conclusion follows. 
op
(2) Lemma 2.2.12 implies that . colim F, (qi )i∈Ob I is the limit of the functor
.F
op : I op → C op and moreover, we have .q ◦op f op = q op ◦op g op for all .i ∈ Ob I ,
op
i i
where .◦op denotes the composition in .C op . The first part of the proof implies .f op =
g op and therefore .f = g, as desired. 

Lemma 2.2.15
 Let F , .G : J → C be two  functors,
 where J is a small category,
 and
denote by . L, (pj : L → F (j ))j ∈Ob J and . C, (qj : F (j ) → C)j ∈Ob J the limit
and colimit, respectively, of F . If .α : F → G is a natural transformation, then:
2.2 (Co)limit of a Functor. (Co)complete Categories 115

 
(1) . L, (αj ◦ pj : L → G(j ))j ∈Ob J is a cone on G. Furthermore,
 if .α is a natural
isomorphism then . L, (αj ◦ pj : L → G(j ))j ∈Ob J is the limit of G.
 
(2) If .α is a natural isomorphism then . C, (qj ◦ αj−1 : G(j ) → C)j ∈Ob J is the
colimit of G.
Proof Given that in most of the previous proofs we have worked mainly with limits,
here we will prove the second
 assertion concerning colimits. 
(2) First we show that . C, (qj ◦ αj−1 : G(j ) → C)j ∈Ob J is a cocone on G.
To this end, consider .d ∈ HomJ (i, l); we need to prove the commutativity of the
following diagram:

. (2.11)
 
Since . C, (qj : F (j ) → C)j ∈Ob J is in particular a cocone on F , the following
diagram is commutative:

. (2.12)

Furthermore, as .α is a natural transformation, the following diagram is commutative

as well:

. (2.13)

Putting all the above together yields

ql ◦ αl−1 ◦ G(d) = ql ◦ F (d) ◦ αi−1 = qi ◦ αi−1 ,

(2.13) (2.12)
.
116 2 Limits and Colimits

 
which proves that (2.11) holds and therefore . C, (qj ◦ αj−1 : G(j ) → C)j ∈Ob J is
  

indeed a cocone . C , (tj : G(j ) → C )j ∈Ob J
  on G. Consider now another cocone

on G. Then . C , (tj ◦ αj : F (j ) → C  )j ∈Ob J is a cocone on F . Indeed, for all
.d ∈ HomJ (i, l) we have

(2.13)
tl ◦ αl ◦ F (d) = tl ◦ G(d) ◦ αi = ti ◦ αi ,
.

 
where in the last equality we used the fact that . C  , (tj : G(j ) → C  )j ∈Ob J is a
cocone on G.  
Now since . C, (qj : F (j ) → C)j ∈Ob J is the colimit of F , there exists a unique

.f ∈ HomC (C, C ) such that the following diagram is commutative for all .j ∈ Ob J :

Thus .f ∈ HomC (C, C  ) is the unique morphism such that .f ◦ (qj ◦ αj−1 ) = tj ,
i.e., the unique morphism which makes the following diagram commutative for all
.j ∈ Ob J :

 
This proves that . C, (qj ◦ αj−1 : G(j ) → C)j ∈Ob J is the initial object in the
category of cocones on G, as desired.
.(1) To start with, note that .α
op : Gop → F op is also a natural isomorphism
 −1
by Proposition 1.8.3. Furthermore, . α op : F op → Gop is  again a natural
op
isomorphism (see Example 1.7.2, (6)). Lemma 2.2.12 implies that . L, (pj )j ∈Ob J
op
is the colimit of .F op and using part .(2) proved above we obtain that . L, (pj ◦op

op op op
αj )j ∈Ob J is the colimit of .Gop . Now since we have .pj ◦op αj = (αj ◦ pj )op ,
the conclusion follows by Lemma 2.2.12. 

2.2 (Co)limit of a Functor. (Co)complete Categories 117

(Co)limit as a Functor

It turns out that taking (co)limits9 yields a functor:

Theorem 2.2.16 Let I be a small category and .C a complete category. Then
lim : Fun(I, C) → C defined below is a functor:
.

lim (F ) = lim F,
. lim (α) = α
for all functorsF, G : I → C and all natural transformations α : F → G,
   
where . lim F, (pi )i∈Ob I and . lim G, (si )i∈Ob I are the limits of F and G respec-
tively and .α ∈ HomC (lim F, lim G) is the unique morphism which makes the
following diagram commute for all .i ∈ Ob I :

Proof Before
 going into the proof,
 we point out that in light of Lemma 2.2.15, (1)
the pair . lim F, (αi ◦ pi )i∈Ob I is a cone on G. Therefore, the unique morphism .α ∈
HomC (lim F, lim G) which makes the diagram above commutative for all .i ∈ Ob I
exists by virtue of Definition 2.2.7.
Now let .F : I → C be a functor and .1F the identity natural transformation. Then,
.lim(1F ) is the unique morphism in .HomC (lim F, lim F ) such that the following

diagram is commutative for all .i ∈ Ob I :

lim

As .1lim F makes the same diagram commutative we obtain .lim(1F ) = 1lim F . This
leads to .lim(1F ) = 1lim(F ) , as desired.
Consider now functors F , G, .H : I → C and natural transformations .α : F → G,
.β : G → H . The proof will be finished once we show that .lim(β ◦ α) = lim(β) ◦
     
lim(α). To this end, let . lim F, (pi )i∈Ob I , . lim G, (si )i∈Ob I , . lim H, (ti )i∈Ob I
be the limits of F , G and H , respectively. Recall that .β ◦ α, .β, .α are the unique

9 Note that defining the (co)limit functor requires an arbitrary choice of a limit for each functor.

This is always possible as we assumed the axiom of choice to hold.

118 2 Limits and Colimits

morphisms which make the following diagrams commutative for all .i ∈ Ob I :

i.e., ti ◦ β ◦ α = (β ◦ α)i ◦ pi , .
. (2.14)
si ◦ α = αi ◦ pi , . (2.15)
ti ◦ β = βi ◦ si . (2.16)

Putting everything together yields

(2.16) (2.15)
ti ◦ β ◦ α = βi ◦ si ◦ α = βi ◦ αi ◦ pi = (β ◦ α)i ◦ pi for all i ∈ Ob I.
.

Now since .β ◦ α is the unique morphism for which (2.14) holds, we obtain .β ◦ α =
β ◦ α, and the proof is now complete. 

For the sake of completeness we record below the dual result concerning colimits
and leave the proof to the reader:
Theorem 2.2.17 Let I be a small category and .C a cocomplete category. Then
colim : Fun(I, C) → C defined below is a functor:
.

colim(F ) = colim F,
. colim(α) = α
for all functorsF, G : I → C and all natural transformations α : F → G,
   
where . colim F, (qi )i∈Ob I and . colim G, (ti )i∈Ob I are the colimits of F and G
respectively and .α ∈ HomC (colim F, colim G) is the unique morphism which
makes the following diagram commute for all .i ∈ Ob I :

.
2.3 (Co)limit as a Representing Pair 119

2.3 (Co)limit as a Representing Pair

In this section we show that the (co)limit of a functor arises as the representing pair
of a certain functor. Throughout, I is a small category and .F : I → C is a functor.
We can define two functors, .Cocone(F, −) : C → Set and .Cone(F, −) : C →
Set, as follows:

Cocone(F, C) = Cocone(F, C),

.

Cocone(F, f ) : Cocone(F, C) → Cocone(F, D),

   
Cocone(F, f ) (ti )i∈Ob I = f ◦ ti i∈Ob I ,
Cone(F, C) = Cone(F, C),
Cone(F, f ) : Cone(F, D) → Cone(F, C),
   
Cone(F, f ) (si )i∈Ob I = si ◦ f i∈Ob I

for all C, .D ∈ Ob C, .f ∈ HomC (C, D), .(ti )i∈Ob I ∈ Cocone(F, C) and

(si )i∈Ob I ∈ Cone(F, D).
.

Note that .Cocone(F, −) : C → Set is a covariant functor while .Cone(F, −) : C

→ Set is contravariant.
Lemma 2.3.1 Let .F : I → C be a functor and I a small category. Then
 
. C, ti ∈ HomC (C, F (i)) is the representing pair of the contravariant
i∈Ob I
 op  
functor .Cone(F, −) if and only if . C, ti ∈ HomC op (F (i), C) i∈Ob I is the
representing pair of the functor .Cocone(F op , −).
 op  
Proof Assume that . C, ti ∈ HomC op (F (i), C) i∈Ob I is the representing pair
 
of the functor .Cocone(F op , −) : C op → Set and . ui ∈ HomC (C  , F (i))i∈Ob I ∈
Cone(F, C  ), where .C  ∈ Ob C. Then Lemma 2.2.3 implies that . ui i∈Ob I ∈
op

Cocone(F , C ) and using Proposition 1.7.7 we obtain a unique .f op ∈
op

HomC op (C, C  ) such that

 op   op 
Cocone(F op , f op ) (ti )i∈Ob I = (ui )i∈Ob I .
.

   
This comes down to . ti ◦f i∈Ob I = ui i∈Ob I . In other words, there exists a unique

   
.f ∈ HomC (C , C) such that .Cone(F, f ) (ti )i∈Ob I = (ui )i∈Ob I and it follows
  
by Corollary 1.8.2 that . C, ti ∈ HomC (C, F (i)) i∈Ob I is the representing pair
of the contravariant functor .Cone(F, −) : C → Set.
The converse follows by similar arguments and is left to the reader. 

120 2 Limits and Colimits

The colimit of F , if it exists, is the representing pair of the functor

Cocone(F, −); similarly, the limit of F , if it exists, is the representing pair of
.

the functor .Cone(F, −).

Theorem 2.3.2 Let I be a small category and .F : I → C a functor. Then:
(1) a cocone on F is the colimit of F if and only if is the representing pair of the
functor .Cocone(F, −) : C → Set;
(2) a cone on F is the limit of F if and only if is the representing pair of the
contravariant functor .Cone(F, −) : C → Set.
  
Proof .(1) Assume first that . C, ti ∈ HomC (F (i), C) i∈Ob I is the colimit of F .
We define a natural transformation .α : HomC (C, −) → Cocone(F, −) as follows
for all .D ∈ Ob C and .f ∈ HomC (C, D):
 
αD (f ) = f ◦ ti i∈Ob I ∈ Cocone(F, D).
.

In order to prove that .α is indeed a natural transformation, consider .u ∈

HomC (D, D  ); we need to show the commutativity of the following diagram:

Hom Cocone

. (2.17)

Indeed, for all .f ∈ HomC (C, D) we have

   
Cocone(F, u) ◦ αD (f ) = Cocone(F, u) (f ◦ ti )i∈Ob I = u ◦ f ◦ ti i∈Ob I
.

= αD  (u ◦ f ) = αD  ◦ HomC (F, u)(f ),

which shows that diagram (2.17) is indeed commutative. We are left to show

that each .αD is an isomorphism. To this end, recall that since . C, ti ∈
   
HomC (F (i), C) i∈Ob I is the colimit of F , for any . vi i∈Ob I ∈ Cocone(F, D)
there exists a unique .v ∈ HomC (C, D) such that the following diagram is
commutative for all .i ∈ Ob I :

. (2.18)
2.4 (Co)limits by (Co)equalizers and (Co)products 121

We can now define a map .βD : Cocone(F, D) → HomC (F, D) by

.βD (vi )i∈Ob I = v, where .v ∈ HomC (C, D) is the unique morphism which
makes diagram (2.18) commutative. Then, for all . vi i∈Ob I ∈ Cocone(F, D) and
.f ∈ HomC (C, D) we have

     
αD ◦ βD (vi )i∈Ob I = αD (v) = v ◦ ti i∈Ob I = vi i∈Ob I ,
.
 
βD ◦ αD (f ) = βD (f ◦ ti )i∈Ob I = f,

which shows that .βD is the inverse of .αD . Therefore, .α is a natural isomorphism.
This shows that the functor .Cocone(F, −) : C → Set  is representable, as desired.
 
Suppose now that . C, ti ∈ HomC (F (i), C) i∈Ob I is the representing pair of
 
the functor .Cocone(F, −) : C → Set and consider .D ∈ Ob C and . vi i∈Ob I ∈
Cocone(F, D). Using Proposition 1.7.7, there exists a unique .f ∈ HomC (C, D)
such that
   
Cocone(F, f ) (ti )i∈Ob I = vi i∈Ob I .
.

   
This comes down to . f ◦ ti i∈Ob I = vi i∈Ob I . In other words, there exists a unique
.f ∈ HomC (C, D) such that the following diagram commutes for all .i ∈ Ob I :

 
Therefore, . C, (ti )i∈Ob I is the colimit of F . 
 
(2) By Lemma 2.2.12, a cone . C, si ∈ HomC (C, F (i)) i∈Ob I on F is the
 op  
limit of F if and only if . C, si ∈ HomC op (F (i), C) i∈Ob I is the colimit of the
 op  
dual functor .F op . Using .1) it follows that . C, si ∈ HomC op (F (i), C) i∈Ob I
 op  
is the colimit of .F op if and only if . C, si ∈ HomC op (F (i), C) i∈Ob I is the
representing pair of the functor .Cocone(F op , −) : C op→ Set and by Lemma 2.3.1
 
this is equivalent to . C, si ∈ HomC (C, F (i)) i∈Ob I being the representing pair
of the functor .Cone(F, −) : C → Set. 


2.4 (Co)limits by (Co)equalizers and (Co)products

(Co)products and (co)equalizers are perhaps the most important among the special
cases of (co)limits. This is due to the fact that all small (co)limits can be constructed
122 2 Limits and Colimits

out of (co)products and (co)equalizers. We start by presenting an example which

hints at this construction.
Example
  .F : I → Set be a functor, where I is a small category. Then:
2.4.1 Let
.lim F, (pk )k∈ObI is the limit of F , where

lim F = {(xk )k∈ObI ∈
. F (k) | F (f )(xi ) = xj for all f ∈ HomI (i, j )},
k∈ObI

pj : lim F → F (j ), pj ((xk )k∈ObI ) = xj , for all j ∈ Ob I.

 
We start by proving that . lim F, (pk )k∈ObI is a cone on F . Indeed, for any .f ∈
HomI (i, j ) we have .F(f )(xi ) = xj for
 all .(xk)k∈Ob I ∈ lim F , which can be written
equivalently as .F (f ) pi ((xk )k∈ObI ) = pj (xk )k∈ObI . Thus we obtain .F (f ) ◦
pi = pj , as desired.  
Assume now that . C, (sk )k∈Ob I is another cone on F , i.e., .C ∈ Ob Set and .sk ∈
Hom (C, F (k)) such that .F (f ) ◦ si = sj for all .f ∈ HomI (i, j ). Now recall that
 Set   
. F (k), (pk )k∈ObI is the product in Set of the family . (F (k))k∈ObI ; hence,
k∈ObI 
there exists a unique morphism .g : C → F (k) in Set such that the following
k∈Ob I
diagram is commutative for all .j ∈ ObI :

 
We are left to prove that .Img ⊆ lim F . To this end, notice that .g(c) = sk (c) k∈Ob I
for all .c ∈ C. Moreover, since .F (f ) ◦ si = sj for any .f ∈ HomI (i, j ) we get
.F (f ) si (c) = sj (c) for all .c ∈ C. Thus .g(c) = sk (c) ∈ lim F for all
k∈Ob I
.c ∈ C. .

The previous example shows that the limit of a functor .F : I →  Set, where
I is a small category, can be constructed as a subset of the product . i∈ObI F (i).
This suggests that equalizers are being used in order to construct the limit. The next
theorem shows that this method can be generalized to arbitrary categories allowing
for the construction of small limits out of products and equalizers.
Theorem 2.4.2 A category .C is (co)complete if and only if it has (co)products and
(co)equalizers.
2.4 (Co)limits by (Co)equalizers and (Co)products 123

Proof We will only prove the assertion regarding completeness and leave the (dual)
one about cocompleteness to the reader. Obviously, if a category is complete then
it has products and equalizers, as shown in Example 2.2.8, (2) and (3). Conversely,
assume .C is a category with products and equalizers and let .F : I → C be a functor,
where I is a small category. For any morphism f in I we will denote by .d(f )
the domain of f and by .c(f ) the codomain of f ; in other words we have .f ∈
HomI (d(f ), c(f )). We start by constructing
 the products in .C of the families of
objects .(F (i))i∈ObI and . F (c(f )) f ∈Hom (d(f ), c(f )) , respectively:
I

   
. F (i), (ui )i∈Ob I , F (c(f )), (vc(f ) )f ∈HomI (d(f ), c(f )) .
i f

 
As . f F (c(f )), (vc(f ) )f ∈HomI (d(f ), c(f )) is the product in .C of the fam-
 
ily of objects . F (c(f )) f ∈Hom (d(f ), c(f )) , there exists a unique morphism
  I
.α : i F (i) → f F (c(f )) in .C such that the following diagram is commutative
for all .g ∈ HomI (d(g), c(g)):

. (2.19)
 
Similarly, there exists a unique morphism .β : i F (i) → f F (c(f )) in .C such
that the following diagram is commutative for all .g ∈ HomI (d(g), c(g)):

(2.20)
124 2 Limits and Colimits

Now consider .(L, l) to be the equalizer in .C of the pair of morphisms .(α, β). The
complete picture is captured by the following diagram:

 
We willprove that . L, (pi = ui ◦ l)i∈Ob I is the limit of the functor F . First we
prove that . L, (pi )i∈Ob I is a cone on F . Indeed, if .r ∈ HomI (d(r), c(r)) we have

(2.20) (2.19)
F (r) ◦ pd(r) = F (r) ◦ ud(r) ◦ l = vc(r) ◦ β ◦ l = vc(r) ◦ α ◦ l = uc(r) ◦ l = pc(r) .
.

   
Moreover, consider another cone . M, (qi )i∈ObI on F . Since . i F (i), (ui )i∈Ob I
is the product in .C 
of the family of objects .(F (i))i∈ObI , there exists a unique
morphism .q  : M → i F (i) in .C such that for any .j ∈ Ob I we have

. (2.21)

Now for any .r ∈ HomI (d(r), c(r)) we have

(2.19) (2.21)
vc(r) ◦ α ◦ q  = uc(r) ◦ q  = qc(r) = F (r) ◦ qd(r)
.

(2.21) (2.20)
= F (r) ◦ ud(r) ◦ q  = vc(r) ◦ β ◦ q  ,
 
where in the third equality we used the fact that . M, (qi )i∈Ob I is a cone on F .
Therefore we have .vc(r) ◦ α ◦ q  = vc(r) ◦ β ◦ q  for any .r ∈ HomI (d(r), c(r))
and according to Proposition 2.2.14, (1) we obtain .α ◦ q  = β ◦ q  . Since .(L, l) is
the equalizer of the pair of morphisms .(α, β) in .C we obtain a unique morphism
.q : M → L such that

l ◦ q = q .
. (2.22)
2.4 (Co)limits by (Co)equalizers and (Co)products 125

It turns out that q is the unique morphism in .C which makes the following diagram
commute for all .j ∈ Ob I :

(2.22) (2.21)
Indeed, for any .j ∈ Ob I we have .pj ◦q = uj ◦l ◦ q = uj ◦q  = qj . Finally,
we are left to prove the uniqueness of q. To this end, assume .q ∈ HomC (M, L)
is another morphism such that .pj ◦ q = qj for all .j ∈ Ob I . Hence, we obtain

.uj ◦ l ◦ q = qj for all .j ∈ Ob I and since .q is the unique morphism in .C which

makes diagram (2.21) commute, we obtain .l ◦ q = q  . Now the uniqueness of the

morphism in .C for which (2.22) holds implies .q = q, as desired. 

Examples 2.4.3
(1) Products and equalizers for the categories Set, Grp, Ab, .R M, Top are
constructed in Examples 2.1.5 and 2.1.10, respectively. Hence, Theorem 2.4.2
shows that all these categories are complete.
(2) Similarly, coproducts and coequalizers for the categories Set, Grp, Ab, .R M,
Top are constructed in Examples 2.1.5 and 2.1.10, respectively. In light of
Theorem 2.4.2 we can conclude that all these categories are cocomplete. .

The next example gives a precise description of colimits in .Set.

Example
 2.4.4 Let I  be a small category and .F : I → Set a functor. Let
. i∈I F (i), (qi )i∈Ob I be the coproduct in .Set of the family
 . F (i)
i∈Ob Set
as
described in Example 2.1.5, (9) and let R be the relation on . i∈I F (i) defined as
follows:

(x, i) R (y, j ) if and only if F (f )(x) = y, for some f ∈ HomI (i, j ).

.

(2.23)

R is obviously reflexive and transitivebut not necessarily symmetric. We denote

by .∼R the equivalence relation on . i∈I F (i) generated by R. Then, the pair
.(colim F, (ri )i∈Ob I ) defined below is the colimit of F :

 
colim F =
. F (i) ,
∼R
i∈I

rj : F (j ) → colim, rj = π ◦ qj , for all j ∈ Ob I,

126 2 Limits and Colimits

 
where . i∈I F (i) denotes the quotient set by the equivalence relation .∼R and
∼R 
 
.π : i∈Ob I F (i) → i∈I F (i) is the associated quotient function.
∼R
We start by proving that .(colim F, (ri )i∈Ob I ) is a cocone on F . To this end, let
.d ∈ HomI (i, j ); then, for any .x ∈ F (i) we have

(2.23)
rj ◦ F (d)(x) = π ◦ qj ◦ F (d)(x)=π ◦ (F (d)(x), j ) = π(x, i)=π ◦ qi (x)=ri (x),
.

i.e., the following diagram is commutative:

Hence
 .(colim F, (ri )i∈Ob I ) is indeed a cocone on F . Consider now another cocone
  
.C, (ti )i∈Ob I on F . Definition 2.1.2 yields a unique .ψ ∈ HomSet i∈I F (i), C
which renders the following diagram commutative for all .j ∈ Ob I :

. (2.24)

Let .(x, i), .(y, j ) ∈ i∈Ob I F (i) such that .(x, i) ∼R (y, j ). Thus we have either
.(x, i) R (y, j ) or .(y, j ) R (x, i). In the first case, there exists some .f ∈ HomI (i, j )

such that .F (f )(x) = y. Then, we obtain

(2.24) (2.24)
ψ(x, i) = ψ ◦ qi (x) = ti (x) = tj ◦ F (f )(x) = tj (y) = ψ ◦ qj (y) = ψ(y, j )
.

 
where in the third equality we used the fact that . C, (ti )i∈Ob I is a cocone on F .
On the other hand, if .(y, j ) R (x, i) there exists some .g ∈ HomI (j, i) such that
.F (g)(y) = x. This leads to

(2.24) (2.24)
ψ(y, j ) = ψ ◦ qj (y) = tj (y) = ti ◦F (g)(y) = ti (x) = ψ ◦qi (x) = ψ(x, i),
.

 
where the third equality follows from the fact that . C, (ti )i∈Ob I is a cocone on F .
Putting all together, we proved that .(x, i) ∼R (y, j ) implies .ψ(x,
 i) = ψ(y, j ).

Therefore, by the universal property of the quotient set . i∈I F (i) there exists
∼R
2.5 (Co)limit Preserving Functors 127

 
a unique map .ϕ : i∈I F (i) → C such that the following diagram commutes:
∼R

. (2.25)
 
Now it can be easily seen that .ϕ : i∈I F (i) → C is the unique morphism
∼R
which makes the following diagram commutative for all .i ∈ Ob I :

Indeed, for all .i ∈ Ob I we have

(2.25) (2.24)
ϕ ◦ ri = ϕ ◦ π ◦ qi = ψ ◦ qi = = ti .
.


.

2.5 (Co)limit Preserving Functors

Definition 2.5.1 A functor F : C → D preserves (small) limits/colimits

 when
 for
every functor G : I → C, where I is a (small)
 category, if L, (p i i∈ObI is the
)
limit/colimit of G then F (L), (F (pi ))i∈ObI is the limit/colimit of F G.
Lemma 2.5.2 Let F : C → D be a functor. Then F preserves limits if and only if
the dual functor F op : C op → Dop preserves colimits.
Proof Assume F preserves limits and let I be a small category. If G : I → C op
 op 
is a functor whose colimit we denote by C, qi ∈ HomC op (G(i), C) i∈Ob I
  
then using Lemma 2.2.12 we obtain that C, qi ∈ HomC (C, G(i)) i∈Ob I
is the limit of Gop: I op → C. As F preserves limits, it follows that
 
F (C), F (qi ) i∈Ob I is the limit of F Gop . Using again Lemma 2.2.12 we obtain
 
op   op (1.25) op op op
that F (C), F op (qi ) i∈Ob I is the colimit of F Gop = F G =
op op
F G. This shows that F preserves colimits, as desired.
128 2 Limits and Colimits

Assume now that F op preserves colimits and let H : I →

 C be a functor whose
 
limit we denote by L, pi ∈ HomC (L, H (i)) i∈Ob I . Then Lemma 2.2.12
 op  
implies that L, pi ∈ HomC op (H (i), L) i∈Ob I is the colimit of H op .
 op 
As F op is colimit preserving we obtain that F (L), (F op (pi ))i∈Ob I is the
(1.25)
 = (F H ) . Again by Lemma 2.2.12 it follows that
op op op
colimit of F H
F (L), (F (pi ))i∈Ob I is the limit of F H . Therefore, F is limit preserving and
the proof is now finished. 

As a consequence of Theorem 2.4.2 we obtain the following:
Proposition 2.5.3 Let C and D be two categories such that C is (co)complete. A
functor F : C → D preserves small (co)limits if and only if it preserves (co)products
and (co)equalizers.
Proof Theorem 2.4.2 proves that the limit of a functor H : I → C, for a small
category I , can be constructed as the equalizer of certain morphisms between
two products of some families of objects in C. As F preserves both products and
equalizers we can conclude that it preserves limits. 

Example 2.5.4 Let BilM, N : Ab → Ab be the functor defined in Exam-
ple 1.5.3, (28). We will show that BilM, N preserves limits. In light of
Proposition 2.5.3 it will suffice to show that it preserves products and equalizers.
 be a family 
To this end, let (Ai )i∈I of abelian groups, where
 I is a set, and
consider its product i∈I A i , (p j : A
i∈I  i → A )
j j ∈I in Ab. Recall from
Example 2.1.5, (2) that the underlying set of i∈I Ai is just the cartesian product
of the Aj s while pj is the j -th projection. We aim to prove that

   
. BilM, N ( Ai ), (BilM, N (pj ) : BilM, N ( Ai ) → BilM, N (Aj ))j ∈I
i∈I i∈I
 
is the product in Ab of the family BilM, N (Ai ) i∈I . Denote by
  
. BilM, N (Ai ), (πj : BilM, N (Ai ) → BilM, N (Aj ))j ∈I
i∈I i∈I
 
the product in Ab of the family  BilM, N (Ai ) i∈I . Again by Example 2.1.5, (2) we
know that the underlying set of i∈I BilM, N (Ai ) is the cartesian product  of the
 s and π is the j -th projection. Now define ψ : Bil
Bil
 M, N (Aj ) j   
M, N ( i∈I Ai) →
i∈I Bil M, N (Ai ) by ψ(α) = p i ◦ α i∈I
for all α ∈ BilM, N ( i∈I A i ). It can be
easily seen that for all i ∈ I we have pi ◦ α ∈ BilM, N (Ai ), which shows  that ψ
is well-defined. Furthermore, ψ is bijective. Indeed, if α, β ∈ BilM, N ( i∈I Ai )
such that ψ(α) = ψ(β), we obtain pi ◦ α = pi ◦ β for all i ∈ I . Now Propo-
 (1) implies α = β, which shows that ψ is injective. Consider now
sition 2.2.14,
(ui )i∈I ∈ i∈I BilM, N (Ai ), where uj ∈ BilM, N (Aj ) for all j ∈ I . Then ψ(u) =
(ui )i∈I , where u ∈ BilM, N ( i∈I Ai ) is defined by u(m, n) = (ui (m, n))i∈I for
2.5 (Co)limit Preserving Functors 129

all m ∈ M, n ∈ N, which shows thatψ is also surjective.

 To conclude, we have an
isomorphism of groups ψ : BilM, N ( i∈I Ai ) → i∈I BilM, N (Ai ) which makes
the following diagram commutative for all j ∈ I :


Indeed, for all α ∈ BilM, N ( i∈I Ai ) and j ∈ I , we have
 
(πj ◦ ψ)(α) = πj (pi ◦ α)i∈I = pj ◦ α = BilM, N (pj )(α).
.

Now Proposition 2.1.3 allows us to conclude that

  
. BilM, N ( Ai ), (BilM, N (pj ))j ∈I
i∈I
 
is the product in Ab of the family BilM, N (Ai ) i∈I , as desired.
Next we show that BilM, N preserves equalizers. To this end, let f , g ∈
HomAb (A, B) and consider the equalizer (E, i) of the pair (f, g) in Ab, i.e.,
E = {a ∈ A | f (a) = g(a)} and i : E → A is the inclusion morphism (see
Example 2.1.10, (2)). Moreover, denote by (Q, j ) the equalizer in Ab of the pair of
morphisms BilM, N (f ), BilM, N (g) ∈ HomAb (BilM, N (A), BilM, N (B)). We have

Q = {u ∈ BilM, N (A) | BilM, N (f )(u) = BilM, N (g)(u)},

.

while j : Q → BilM, N (A) denotes the inclusion. Now define ϕ : BilM, N (E) → Q
by ϕ(v) = i ◦ v for all v ∈ BilM, N (E). Consider w ∈ Q, i.e., w : M × N → A
is a bilinear map such that f ◦ w = g ◦ w. Then, for all m ∈ M, n ∈ N we
have f (w(m, n)) = g(w(m, n)), which implies w(m, n) ∈ E. If we denote by w
the map obtained from w by restricting its codomain to E, we have ϕ(w) = i ◦
w = w. Hence ϕ is surjective and is also trivially injective as i is a monomorphism
by Proposition 2.1.9. Moreover, ϕ is the unique group morphism which makes the
following diagram commutative:

.
130 2 Limits and Colimits

Indeed, for all u ∈

 BilM, N (E) we have Bil  M, N (i)(u) = i ◦ u = j ◦i ◦ u = j ◦ϕ(u).
This
 shows that Bil M, N(E), Bil M, N (i) is the equalizer of the pair of morphisms
BilM, N (f ), BilM, N (g) and therefore the functor BilM, N preserves equalizers as
well. 
The following lemma will be useful in the sequel:
 
Lemma 2.5.5 Let F : C → D and G : I →C be two functors. If X, (si )i∈Ob I is
a (co)cone on G, then F (X), (F (si ))i∈Ob I is a (co)cone on F G : I → D.
Proof We will only prove the statement regarding cones. To this end consider f ∈
HomI (i, j ). The proof will be finished once we show that the following diagram is
commutative:

(2.26)
 
Indeed, as X, (si )i∈Ob I is a cone on G, the following diagram is commutative:

. (2.27)

Now it is straightforward to see that (2.26) holds true just by applying F to the
identity (2.27). 

One of the most important examples of functors which preserve limits are the
hom functors.
Theorem 2.5.6 Let C be a category and C ∈ Ob C.
(1) The hom functor HomC (C, −) : C → Set preserves all existing small limits.
(2) The contravariant hom functor HomC (−, C) : C → Set maps existing small
colimits to small limits.

Proof (1) Consider

 a functor G : I → C, where I is a small category, and
L, (pi )i∈Ob I its limit. The proof will be finished once we show that
 
. HomC (C, L), (HomC (C, pi ))i∈Ob I
2.5 (Co)limit Preserving Functors 131

 functor HomC (C, G(−)) : I → Set.

is the limit of the  To start with, by Lemma 2.5.5
we obtain that HomC (C, L), (HomC (C, pi ))i∈Ob I is a cone on HomC (C, G(−)).
Consider now another cone  M, (qi )i∈Ob I on HomC (C, G(−)), where M ∈
Ob Set and qi ∈ HomSet M, HomC (C, G(i)) for all i ∈ ObI . Hence, the
following diagram is commutative for any f ∈ HomI (i, j ):

Hom

Hom
.

 
Therefore, for all m ∈ M we have HomC (C, G(f )) qi (m) = qj (m), which leads
to G(f ) ◦ qi (m) = qj (m). This implies that for each m ∈ M, C,(qi (m))i∈Ob I 
is a cone on G, where qi (m) ∈ HomC (C, G(i)) for all i ∈ Ob I . As L, (pi )i∈Ob I
is the limit of G, it yields a unique morphism q(m) ∈ HomC (C, L) such that the
following diagram is commutative for all i ∈ Ob I :

Putting all this together we have defined a function q : M → HomC (C, L) (i.e.,
a morphism in Set) satisfying HomC (C, pi ) ◦ q = qi for any i ∈ Ob I , i.e., the
following diagram commutes:

Hom

Furthermore, the uniqueness of q with this property follows from that of the q(m)’s.
(2) Showing that the contravariant hom functor maps existing small colimits to
small limits follows the strategy used in the proof above and is left to the reader. 

Definition 2.5.7 A functor F : C → D reflects (small) limits/colimits when for
every
 functor G : I → C, where I is a (small) category, and every cone/cocone
L, (p i )i∈Ob I  on G, if F (L), (F (pi ))i∈Ob I is the limit/colimit of F G, then
L, (pi )i∈Ob I is the limit/colimit of G.
132 2 Limits and Colimits

An important class of (co)limit reflecting functors is the class of fully faithful

functors:
Theorem 2.5.8 A fully faithful functor F : C → D reflects small (co)limits.
Proof Let G : I → C be a functor where I is a small category  and consider a
cone L, (pi )i∈Ob I  on G such that
 F (L), (F (p i )) i∈Ob I is the
 limit of F ◦G.
We will prove that L, (pi )i∈Ob I is the limit of G. Indeed, if M, (qi)i∈Ob I is
another cone on G then Lemma 2.5.5 implies that F (M), (F (qi ))i∈Ob I is a cone
on F ◦ G. Therefore, we have a unique morphism f ∈ HomD (F (M), F (L)) such
that the following diagram is commutative for all i ∈ Ob I :

. (2.28)

Since F is fully faithful there exists a unique morphism f ∈ HomC (M, L) such
that F (f ) = f . Then (2.28) comes down to F (qi ) = F (pi ) ◦ F (f ) and since F
is faithful we obtain qi = pi ◦ f for all i ∈ Ob I , i.e., the following diagram is
commutative:

We are left to prove that f is the unique morphism which makes the above diagram
commutative. To this end, assume that g ∈ HomC (M, L) is another morphism such
that qi = pi ◦ g for all i ∈ Ob I . This implies F (pi ) ◦ F (g) = F (qi ) for all
i ∈ Ob I and since f is the unique morphism which makes diagram (2.28) commute
we obtain F (g) = f . Now recall that we also have F (f ) = f and since F is
faithful we arrive at g = f , as desired. Therefore L, (pi )i∈Ob I is a final object in
the category of cones on G, as desired.
The dual statement will be settled as usual by the duality principle. Indeed,
 H : I → C be a functor, where
let  I is a small category,
 and consider a cocone
Q, (qi )i∈Ob I on H such that F (Q), (F (qi))i∈Ob I is the colimit of F ◦ H .
op
In particular, by Lemma 2.2.3, Q, (qi )i∈Ob I is a cone on H op . Lemma 2.2.12
 op  (1.25)
implies that F (Q), (F op (qi ))i∈Ob I is the limit of (F ◦ H )op = F op ◦ H op .
Since F op is obviously also fully faithful and by
 the first part of
 the proof any fully
op
faithful functor reflects limits, we obtain that Q, (qi  )i∈Ob I is the limit of H op .
By applying Lemma 2.2.12 once more it follows that Q, (qi )i∈Ob I is the colimit
of H . This shows that F is colimit reflecting, as desired. 

2.5 (Co)limit Preserving Functors 133

Proposition 2.5.9 Let F : C → D be a (co)limit preserving functor. If C is

(co)complete and F reflects isomorphisms then F also reflects small (co)limits.
Proof Assume first that F is limit preserving, isomorphisms reflecting and C is
complete. Consider
 a functor G : I → C, where I is a small category,
 and let
M, (qi )i∈Ob I be a cone on G such that F (M), (F  (q ))
i i∈Ob I is the limit of
F ◦ G. The proof will be finished once we show that M, (qi )i∈Ob I is the limit of
 According to the completeness assumption on C the functor G has a limit, say
G.
L, (pi )i∈Ob I . Thus, there exists a unique morphism f ∈ HomC (M, L) such that
the following diagram is commutative for all i ∈ Ob I :

. (2.29)

In particular, this implies the following:

F (pi ) ◦ F (f ) = F (qi ), for all i ∈ Ob I.

.

 
Since F is a limit preserving functor then F (L), (F (pi ))i∈Ob I is also a limit of
F ◦ G. Exactly as in the proof of Proposition 2.1.3, one can show that there exists a
unique isomorphism g ∈ HomD (F (M), F (L)) such that the following diagram is
commutative for all i ∈ Ob I :

Hence F (f ) = g is an isomorphism
 in D. Our assumption implies that f is an
isomorphism in C and thus M, (qi )i∈Ob I is also a limit of G, as desired.
 let H : I →
For the dual statement,  C be a functor,where I is a small category,

and consider a cocone Q, (qi )i∈Ob I on H such  that F op(Q), (F (qi ))i∈Ob
 I is the
op
colimit of F ◦ H . Lemma 2.2.12 implies that F (Q), (F (qi ))i∈Ob I is the limit
(1.25)  op 
of (F ◦ H )op = F op ◦ H op . Furthermore, by Lemma 2.2.3, Q, (qi )i∈Ob I is
a cone on H op while Lemma 2.5.2 implies that F op is limit preserving. As F op
is
 obviously also
 isomorphism reflecting, the first part of the proof
 implies that
op
Q, (qi )i∈Ob I is the limit of H op . Now Lemma 2.2.12 shows that Q, (qi )i∈Ob I
is the colimit of H . 

134 2 Limits and Colimits

Examples 2.5.10
(1) The forgetful functor U : Top → Set preserves products and equalizers (see
Examples 2.1.5 and 2.1.10). Therefore U preserves small limits by Proposi-
tion 2.5.3. Similar arguments show that the forgetful functor U : Grp → Set is
also limit preserving.
(2) The category Ab is complete and the inclusion functor I : Ab → Grp
preserves preserves products and equalizers (see Examples 2.1.5 and 2.1.10).
Therefore I preserves small limits by Proposition 2.5.3. Furthermore, I reflects
isomorphisms and, according to Proposition 2.5.9, I also reflects small limits.


2.6 (Co)limits in Comma Categories

In this section we discuss the (co)completeness of comma categories. Throughout,

F : A → C, .G : B → C are two functors and .(F ↓ G) denotes the corresponding
.

comma category (see Theorem 1.8.4). We consider the forgetful functors .U : (F ↓

G) → A and .V : (F ↓ G) → B defined as follows for all .(A, f, B) ∈ Ob (F ↓ G)
and all morphisms .(a, b) in .(F ↓ G):

.U (A, f, B) = A, U (a, b) = a,
V (A, f, B) = B, V (a, b) = b.

Lemma 2.6.1 There exists a natural transformation .α : F U → GV between the

functors .F U , .GV : (F ↓ G) → C given by .α(A, f, B) = f , for all .(A, f, B) ∈
Ob (F ↓ G).
Proof Let .(a, b) : (A, f, B) → (A , f  , B  ) be a morphism in .(F ↓ G). We need
to show that the following diagram is commutative:

.
2.6 (Co)limits in Comma Categories 135

Indeed, note that the above diagram simplifies to the following:

and the latter diagram is obviously commutative as a consequence of .(a, b) being a

morphism in .(F ↓ G). 

Let I be a category and .H : I → (F ↓ G) a functor. We will use the following
notation for all .i ∈ Ob I :
 
.H (i) = U H (i), αH (i) , V H (i) , (2.30)

where .α is the natural transformation defined in Lemma 2.6.1.

We record a new useful result in the following:
Lemma 2.6.2 Let .H : I → (F  ↓ G) be a functor,
 where I is a small category. If
A is a complete category and . L, (pi )i∈Ob I is the limit of .U H : I → A, then the
.

pair . F (L), (αH (i) ◦ F (pi ))i∈Ob I is a cone on .GV H : I → C.

Proof Let .t ∈ HomI (i, j ) and let .H (t) = (aij , bij ), where
   
.(aij , bij ) : U H (i), αH (i) , V H (i) → U H (j ), αH (j ) , V H (j )

 
is a morphism in .(F ↓ G). This  implies that .aij ∈ HomA U H (i), U H (j ) and
.bij ∈ HomB V H (i), V H (j ) such that the following diagram is commutative:

. (2.31)

Furthermore, as .(L, (pi )i∈Ob I ) is, in particular, a cone on .U H : I → A, the

following diagram is commutative as well:

. (2.32)
136 2 Limits and Colimits

The proof will be finished once we show the commutativity of the following
diagram:

Indeed, we have

GV H (t) ◦ αH (i) ◦ F (pi ) = G(bij ) ◦ αH (i) ◦ F (pi )

.

(2.31)
= αH (j ) ◦ F (aij ) ◦ F (pi )
(2.32)
= αH (j ) ◦ F (pj ),

as desired. 

Lemma 2.6.3 Let .H : I → (F ↓ G) be a functor, where I is a small category.
   
If . (A, f, B), (pi , qi ) i∈Ob I is a (co)cone on H then . A, (pi )i∈Ob I and
 
. B, (qi )i∈Ob I are (co)cones on .U H : I → A and .V H : I → B, respectively.

Proof We only prove the statement concerning cones. To this end, consider .t ∈
HomI (i, j ) and let .H (t) = (aij , bij ), where .(aij , bij ) ∈ Hom(F ↓G) (H (i), H (j )).
In particular, we have

aij ∈ HomA (U H (i), U H (j )) and bij ∈ HomB (V H (i), V H (j )).

.

The proof will be finished once we show that the following diagrams commute:

. (2.33)

. (2.34)
2.6 (Co)limits in Comma Categories 137

  
Now since . (A, f, B), (pi , qi ) i∈Ob I is a cone on H , we have .H (t)◦(pi , qi ) =
(pj , qj ), which componentwise comes down to (2.33) and (2.34). 

We can now state and prove the main result of this section:
Theorem 2.6.4 Let .F : A → C and .G : B → C be two functors.
(1) If .A and .B are complete categories and G preserves small limits then .(F ↓ G)
is also complete and both forgetful functors .U : (F ↓ G) → A and .V : (F ↓
G) → B preserve small limits.
(2) If .A and .B are cocomplete categories and F preserves small colimits then .(F ↓
G) is also cocomplete and both forgetful functors .U : (F ↓ G) → A and
.V : (F ↓ G) → B preserve small colimits.

Proof Throughout, we use the notation introduced in (2.30). Let I be a small

category and .H : I → (F ↓ G) a functor.
.(1) We need to show that H has a limit. Recall that both categories .A and .B are

complete
 and therefore
 the functors .UH : I → A and .V H : I → B have limits,
say . L, (pi )i∈Ob I and . M, (qi )i∈Ob I respectively, where:

L ∈ Ob A,
. pi ∈ HomA (L, U H (i)),
M ∈ Ob B, qi ∈ HomB (M, V H (i)).
 
Using Lemma 2.6.2, we have that . F (L), (αH (i) ◦ F (pi ))i∈Ob I is a cone on
.GV H : I → C, where .αH (i) ◦ F (pi ) ∈ HomC (F (L), GV H (i)). Furthermore,
 
. G(M), (G(qi ))i∈Ob I is the limit of the functor .GV H : I → C as G is limit
preserving. Hence, there exists a unique morphism .h ∈ HomC (F (L), G(M)) which
makes the following diagram commutative for all .i ∈ Ob I :

(2.35)

Note that, in particular, we have .(L, h, M) ∈ Ob (F ↓ G). Furthermore, (2.35)

implies the commutativity of the following diagram:

.
138 2 Limits and Colimits

which proves that .(pi , qi ) : (L, h, M) → H (i) is in fact a morphism in the comma
category .(F ↓ G), for all .i ∈ Ob I . 
 
We will show that . (L, h, M), (pi , qi ) i∈Ob I is the limit of the functor H .
  
We start by proving that . (L, h, M), (pi , qi ) i∈Ob I is a cone on H . To this end,
let .t ∈ HomI (i, j ) and let .H (t) = (aij , bij ), where .(aij , bij ) : H (i) → H (j )
is a morphism in .(F ↓ G). Now observe that  the commutativity
  of the following

diagram is trivially implied by the fact that . L, (pi )i∈Ob I and . M, (qi )i∈Ob I are
in particular cones on .U H and .V H , respectively:

  
Thus, . (L, h, M), (pi , qi ) i∈Ob I is indeed a cone on H . Consider now another
  
cone . (L, h, M), (pi , qi ) i∈Ob I on H . Since .(pi , qi ) : (L, h, M) → H (i) is a
morphism in .(F ↓ G) for all .i ∈ Ob I , the following diagram is commutative:

(2.36)
   
Furthermore, Lemma 2.6.3 implies that . L, (pi )i∈Ob I and
 . M, (qi )i∈Ob I are
cones on .U H and .V H respectively. Since . L, (pi )i∈Ob I and . M, (qi )i∈Ob I
are the limits of .U H and .V H , respectively, there exist two unique morphisms
.u ∈ HomA (L, L) and .v ∈ HomB (M, M) such that the following diagrams are

commutative for all .i ∈ Ob I :

. (2.37)
2.6 (Co)limits in Comma Categories 139

. (2.38)

 
We will show that .(u, v) is a morphism in .(F ↓ G) from . L, h, M to .(L, h, M),
i.e., the following diagram is commutative:

. (2.39)
 
As . G(M), (G(qi ))i∈Ob I is the limit of the functor .GV H : I → C, using
Proposition 2.2.14, (1) it will suffice to show that the following holds for all
.i ∈ Ob I :

G(qi ) ◦ h ◦ F (u) = G(qi ) ◦ G(v) ◦ h.

.

Indeed, we have

(2.35)
G(qi ) ◦ h ◦ F (u) = αH (i) ◦ F (pi ) ◦ F (u) = αH (i) ◦ F (pi ◦ u)
.

(2.37)
= αH (i) ◦ F (pi )
(2.36)
= G(qi ) ◦ h
(2.38)
= G(qi ◦ v) ◦ h = G(qi ) ◦ G(v) ◦ h,

as desired. Moreover, in light of (2.37) and (2.38), .(u, v) is obviously the unique
morphism in .(F ↓ G) which makes the following diagram commute for all .i ∈
Ob I :

.
140 2 Limits and Colimits

  
which shows that . (L, h, M), (pi , qi ) i∈Ob I is indeed the limit of the functor
H . Finally, note that both functors U and V are obviously limit preserving.
.(2) We use the duality principle. As .A
op and .B op are complete categories and

.F
op : Aop → C op preserves limits (Lemma 2.5.2) we obtain, by applying .1), that

the comma-category .(Gop ↓ F op ) is complete. Now by Proposition 1.8.10 we have

an isomorphism of categories between .(Gop ↓ F op ) and .(F ↓ G)op and therefore
.(F ↓ G)
op is complete as well. This shows that .(F ↓ G) is cocomplete, as desired.



As an easy consequence of the previous result we have:
Corollary 2.6.5
(1) Let .B be a complete category and .G : B → C a functor which preserves small
limits. Then, for all .C0 ∈ Ob C, the category .(C0 ↓ G) is complete.
(2) Let .A be a cocomplete category and .F : A → C a functor which preserves
small colimits. Then, for all .C0 ∈ Ob C, the category .(F ↓ C0 ) is cocomplete.
Proof Recall from Corollary 1.8.6 that the category .(C0 ↓ G) is obtained as
a special case of the comma-category .(F ↓ G) by considering .A to be the
discrete category with one object while the functor F is the object .C0 of .C. Note
that the discrete category with one object is obviously complete and the desired
conclusion now follows from Theorem 2.6.4, (1). Similarly, .(2) follows from
Theorem 2.6.4, (2). 

As the identity functor on any category preserves all small (co)limits, we obtain:
Corollary 2.6.6 Let .C be a category and .C0 ∈ Ob C.
(1) If .C is complete then the coslice category .(C0 ↓ C) is also complete.
(2) If .C is cocomplete then the slice category .(C ↓ C0 ) is also cocomplete.
Proof .(1) Follows by considering .B = C and .F = 1C in Corollary 2.6.5, (1). Simi-
larly, .(2) can be obtain by specializing .A = C and .F = 1C in Corollary 2.6.5, (2).



2.7 (Co)limits in Functor Categories

Functor categories form another class of categories which behave well with respect
to (co)limits. In what follows I and J are small categories and .C is an arbitrary
category. We start by introducing the following induced functors:
Lemma 2.7.1 Let .F : I → Fun(J, C) be a functor. Then, we have
2.7 (Co)limits in Functor Categories 141

(1) each .j ∈ Ob J induces a functor .Fj : I → C defined as follows for all i,

.k ∈ Ob I and .t ∈ HomI (i, k):

Fj (i) = F (i)(j ),
. Fj (t) = F (t)j , (2.40)

where .F (t)j denotes the j -component of the natural transformation

F (t) : F (i) → F (k);
.

(2) each .f ∈ HomJ (j, s) induces a natural transformation .Ff : Fj → Fs defined

for all .i ∈ Ob I by .(Ff )i = F (i)(f ).
Proof .(1) Given .j ∈ Ob J , we show first that .Fj preserves identity morphisms.
Since F itself is a functor and it preserves identities, for all .i ∈ Ob I we have
 
.Fj (1i ) = F (1i )j = 1F (i) j = 1F (i)(j ) = 1Fj (i) .

Furthermore, for any .u ∈ HomI (i, k) and .v ∈ HomI (k, l) we have

 
Fj (v ◦ u) = F (v ◦ u)j = F (v) ◦ F (u) j = F (v)j ◦ F (u)j = Fj (v) ◦ Fj (u),
.

where the second equality holds because F is a functor. This proves that .Fj is a
functor.
.(2) Let .t ∈ HomI (i, r). The proof will be finished once we show the
commutativity of the following diagram:

(2.41)

Note that .F (t) is a natural transformation between the functors .F (i), .F (r) : J → C.
Writing down the naturality of .F (t) for the morphism .f ∈ HomJ (j, s) yields the
following commutative diagram:

(2.42)
142 2 Limits and Colimits

Clearly, the commutativity of the diagram (2.42) implies the commutativity of

(2.41), as desired. 

We are now ready to prove the main result of this section, which states that,
under some conditions, the (co)limit of a functor .F : I → Fun(J, C) is constructed
pointwise. Loosely speaking, this means that the (co)limit of F will be constructed
by putting together the (co)limits of the functors .Fj , for each .j ∈ Ob J , as defined
in (2.40).
Theorem 2.7.2 Let .F : I → Fun(J, C) be a functor and for each .j ∈ Ob J
consider the functor .Fj : I → C defined in (2.40). If for all .j ∈ Ob J the functor
.Fj has a (co)limit then F has a (co)limit as well and this (co)limit is computed

pointwise.
Proof We start by proving the claim concerning limits. For each .j ∈ Ob J , let
 j 
. Lj , (p : Lj → Fj (i))i∈Ob I be the limit of the functor .Fj : I → C, where .Lj ∈
i
Ob C. We will construct
 a functor
 .L : J → C together with a family of natural

transformations . L → F (i) i∈Ob I which will form the limit of F . First we define
the functor L on objects by .L(j ) = Lj , for all .j ∈ Ob J . In order to define L on
morphisms we show first that for any .f ∈ HomJ (j, s), the pair
  
j
. Lj , (Ff )i ◦ pi : Lj → Fs (i) i∈Ob I

is a cone on .Fs : I → C. Indeed, we aim to prove that for any .t ∈ HomI (i, k) the
following diagram is commutative:

(2.43)

To this end, the naturality of .Ff applied to t yields the following commutative
diagram:

(2.44)
2.7 (Co)limits in Functor Categories 143

 j 
Furthermore, as . Lj , (pi : Lj → Fj (i))i∈Ob I is in particular a cone on .Fj , the
following diagram is commutative:

. (2.45)

Putting all the above together yields

j j (2.44) j
Fs (t) ◦ (Ff )i ◦ pi = F (t)s ◦ F (i)(f ) ◦ pi
. = F (k)(f ) ◦ F (t)j ◦ pi
(2.45) j j
= F (k)(f ) ◦ pk = (Ff )k ◦ pk

  
j
and we have proved that . Lj , (Ff )i ◦ pi : Lj → Fs (i) i∈Ob I is indeed a cone
 
on .Fs . As . Ls , (pis : Ls → Fs (i))i∈Ob I is the limit of .Fs , there exists a unique
morphism, denoted by .L(f ), which makes the following diagram commutative for
all .i ∈ Ob I :

. (2.46)

We can now define a functor .L : J → C by setting .L(j ) = Lj while for all

.f ∈ HomJ (j, s), .L(f ) is the unique morphism which makes (2.46) commutative.
However, we still need to show that .L : J → C is indeed a functor. To start
with, we prove that L preserves identities. As .F (i) : J → C is a functor for any
.i ∈ Ob I , we have .F (i)(1j ) = 1F (i)(j ) and .L(1j ) is the unique morphism such that
j j
.p ◦ L(1j ) = 1F (i)(j ) ◦ p . Clearly this implies .L(1j ) = 1L(j ) , as desired. Consider
i i
now .f ∈ HomJ (j, s) and .g ∈ HomJ (s, l). Then .L(f ), .L(g) and .L(g ◦ f ),
respectively, are the unique morphisms such that for all .i ∈ Ob I we have

j
pis ◦ L(f ) = F (i)(f ) ◦ pi , .
. (2.47)
pil ◦ L(g) = F (i)(g) ◦ pis , . (2.48)
j
pil ◦ L(g ◦ f ) = F (i)(g ◦ f ) ◦ pi . (2.49)
144 2 Limits and Colimits

Hence, for all .i ∈ Ob I the following holds:

(2.48)
pil ◦ L(g) ◦ L(f ) = F (i)(g) ◦ pis ◦ L(f )
.

(2.47) j j
= F (i)(g) ◦ F (i)(f ) ◦ pi = F (i)(g ◦ f ) ◦ pi ,

where in the last equality we used the fact that .F (i) is a functor. This shows that
.L(g) ◦ L(f ) fulfills (2.49), which implies that .L(g) ◦ L(f ) = L(g ◦ f ). Hence
.L : J → C is a functor and therefore an object in .Fun(J, C). Next, for each .i ∈ Ob I ,
j
we define a natural transformation .pi : L → F (i) by .(pi )j = pi for all .j ∈ Ob J .
 j 
In order to prove that the family of morphisms . pi : L(j ) → F (i)(j ) j ∈Ob J indeed
form a natural transformation we need to show the commutativity of the following
diagram for all .f ∈ HomJ (j, s):

The commutativity of the above diagram follows from the commutativity of

(2.46). Hence, .pi : L → F (i) is a natural transformation for any .i ∈ Ob I and

therefore a morphism in the category .Fun(J, C). We will prove that . L, pi : L →
 
F (i) i∈Ob I is the limit of F . We start by showing that the above pair is a cone
on F . Indeed, let .t ∈ HomI (i, k); we need to prove that the following diagram is
commutative:

. (2.50)
j j
By the commutativity of diagram (2.45) we have .F (t)j ◦ pi = pk for all .j ∈ Ob J .
This implies that the natural transformations .F (t)◦pi and .pk are equal and therefore
diagram (2.50) is commutative, as desired. 
 
Consider now another cone . H, qi : H → F (i) i∈Ob I on F , where .H : J →
2.7 (Co)limits in Functor Categories 145

C is a functor, .qi : H → F (i) is a natural transformation for each .i ∈ Ob I and

j
we denote .(qi )j by .qi for all .j ∈ Ob J . Note for further use that the naturality
of .qi applied to a morphism .f ∈ HomJ (j, s) yields the following commutative
diagram:

(2.51)

  
Moreover, as . H, qi : H → F (i) i∈Ob I is a cone on F , the following diagram
is commutative for any .t ∈ HomI (i, k):

Therefore, for any .j ∈ Ob J , we have

j j
Fj (t) ◦ qi = qk ,
. (2.52)
 j  
which implies that . H (j ), qi : H (j ) → Fj (i) i∈Ob I is a cone on .Fj . Recall
 j  
now that . L(j ), pi : L(j ) → Fj (i) i∈Ob I is the limit of .Fj , so there exists
 
a unique morphism .ξj ∈ HomC H (j ), L(j ) such that the following diagram is
commutative for all .i ∈ Ob I :

. (2.53)

The proof will be finished once we show that the morphisms .ξi , .i ∈ Ob I form
a natural transformation .ξ : H → L or, equivalently, that .ξ is a morphism in the
category .Fun(J, C). To this end, we are left to prove the commutativity of the
146 2 Limits and Colimits

following diagram for any .f ∈ HomJ (j, s):

. (2.54)

By Proposition 2.2.14, (1) it is enough to show that for all .i ∈ Ob I we have

pis ◦ L(f ) ◦ ξj = pis ◦ ξs ◦ H (f ). Indeed, we have
.

To conclude, .ξ is a natural transformation and therefore a morphism in the

category .Fun(J, C). Furthermore, by (2.53), .ξj is the unique morphism satisfying
j j
.p ◦ ξj = q for all .i ∈ Ob I . Therefore, .ξ is the unique natural transformation for
i i
which the following diagram is commutative:

  
which shows that . L, pi : L → F (i) i∈Ob I is indeed the limit of F . 


Corollary 2.7.3 If .C is a (co)complete category then the functor category

Fun(J, C) is also (co)complete, for all small categories J .
.

Proof Let .F : I → Fun(J, C) be a functor. The (co)completeness assumption on

C ensures the existence of a (co)limit for any functor .Fj : I → C. The desired
.

conclusion now follows from Theorem 2.7.2. 


Example 2.7.4 An important consequence that can be derived from Corollary 2.7.3
concerns the category of presheaves on an arbitrary small category .C as defined
in Example 1.9.2, (1). More precisely, as Set is both complete and cocomplete
(see Examples 2.4.1 and 2.4.4) we can conclude that the category of presheaves
.Fun(C
op , Set) on any small category .C is also complete and cocomplete.

Furthermore, this allows us to embed any arbitrary small category .C into the
(co)complete category of presheaves on .C through the Yoneda embedding functor
.Y : C → Fun(C , Set) (see Definition 1.10.7). As the category of presheaves
op
2.7 (Co)limits in Functor Categories 147

on .C inherits most properties of Set, the aforementioned embedding creates the

appropriate setting for using set theoretical tools for studying the category .C or,
in certain situations, even to replace .C by .Fun(C op , Set). For further important
applications of this approach we refer the reader to [1, 32]. .

We end this section with a characterization of monomorphisms (resp. epimor-

phisms) in functor categories, where the fact that (co)limits in the aforementioned
categories are computed pointwise is crucial.
Proposition 2.7.5 Let I and .C be two categories with I small, F , .G : I → C
functors and .ψ : F → G a natural transformation.
(1) If .C is a category with pullbacks then .ψ is a monomorphism in the functor
category .Fun(I, C) if and only if .ψi : F (i) → G(i) is a monomorphism in .C,
for all .i ∈ Ob I .
(2) If .C is a category with pushouts then .ψ is an epimorphism in the functor
category .Fun(I, C) if and only if .ψi : F (i) → G(i) is an epimorphism in .C,
for all .i ∈ Ob I .
Proof .(1) Assume first that .ψi : F (i) → G(i) is a monomorphism for all .i ∈ Ob I
and consider natural transformations .α, .β : H → F such that .ψ ◦ α = ψ ◦ β, where
.H : I → C is a functor. This implies .ψi ◦ αi = ψi ◦ βi for all .i ∈ Ob I . Since

.ψi : F (i) → G(i) is a monomorphism, we obtain .αi = βi for all .i ∈ Ob I .

Conversely, assume that .ψ is a monomorphism in the functor category .Fun(I, C).

Consider .i0 ∈ Ob I , .C ∈ Ob C and f , .g ∈ HomC (C, F (i0 )) such that .ψi0 ◦
f = ψi0 ◦ g. We know from Proposition 2.1.18, (1) that .(F, 1F , 1F ) is the
pullback of .(ψ, ψ) in .Fun(I, C) and since limits in the functor category .Fun(I, C)
are computed pointwise we obtain that .(F (i0 ), 1F (i0 ) , 1F (i0 ) ) is the pullback of
.(ψi0 , ψi0 ) in .C. Therefore, we have a unique .u ∈ HomC (C, F (i0 )) such that the

following diagram is commutative:

Hence we have .1F (i0 ) ◦ u = g and .1F (i0 ) ◦ u = f , which shows that .f = g, as
desired.
(2) Follows easily by the duality principle and Lemma 1.9.6. Indeed, assume
.C is a category with pushouts and .ψ is an epimorphism in the functor category

.Fun(I, C). By Lemma 1.9.6, this is equivalent to .C

op being a category with

pullbacks and .ψ being a monomorphism in the functor category .Fun(I op , C op ).

op
148 2 Limits and Colimits

By the first part of the proof, the last statement is equivalent to .(ψ op )i : F op (i) →
Gop (i) being a monomorphism in .C op for all .i ∈ Ob I . As .(ψ op )i = (ψi )op , this is
the same as .ψi : F (i) → G(i) being an epimorphism in .C for all .i ∈ Ob I . 


2.8 Exercises

2.1 Consider the forgetful functor U : Ringc → Set. Decide if U is:

(a) faithful; (b) representable; (c) epimorphism preserving.

.

2.2 Let C be a category with a final object T and binary products. Prove that there
is a natural isomorphism between the identity functor 1C on C and the product
functor − × T : C → C.
2.3 Let C be a category, f , g ∈ HomC (A, B) and (E, e) the equalizer of the pair
(f, g). Then the following are equivalent:
a. f = g;
b. e is an epimorphism;
c. e is an isomorphism;
d. (A, 1A ) is the equalizer of (f, g).
2.4 Let C be a category and F : C → Set a functor. Prove that
a. if F is representable then it preserves monomorphisms;
b. if F is contravariant representable then it maps epimorphisms to monomor-
phisms.
2.5 Describe binary products in the following categories:
a. PO(P(X), ⊆), where X is a non-empty set and ⊆ denotes the inclusion of
sets;
b. PO(N, |), where | denotes the usual divisibility relation on N.
2.6 Consider the following diagram in an arbitrary category C:

such that the following hold:

h ◦ f = h ◦ g,
. h ◦ v = 1C , g ◦ u = 1B , f ◦ u = v ◦ h.

In this case, the pair (C, h) is called the split coequalizer of (f, g). Prove that
(C, h) is the coequalizer of the pair of morphisms (f, g). State and prove the
dual statement.
2.8 Exercises 149

2.7 Let F : C → D be a functor and suppose that the following diagram is a split
coequalizer in C:

 
Show that F (C), F (h) is the coequalizer of the pair of morphisms
 
F (f ), F (g) .10 State and prove the dual statement.
2.8 Let C be a complete category, f , g ∈ HomC (X, Y ) and let (X × Y, (pX , pY ))
denote the product in C of X and Y . Show that if (E, p) is the equalizer of
the pair (f, g) then (E, p, p) is the pullback of the pair (f , g), where f ,
g ∈ HomC (X, X × Y ) are the unique morphisms such that the following
hold:

pX ◦ f = 1X , pY ◦ f = f, pX ◦ g = 1X , pY ◦ g = g.
.

2.9 Let C be a category,

 I a set, (Ai )i∈I a family of objects in C and define the
functor F = i∈I HomC (Ai , −) : C → Set as follows:

F (C) = i∈I Hom C (Ai, C),
11

F (u) (ηi )i∈I = u ◦ ηi i∈I

for all C, D ∈ Ob C, u ∈ HomC (C, D) and ηi ∈ HomC (Ai , C), i ∈ I . Prove
that F is representable if and only if the family (Ai )i∈I has a coproduct in the
category C.
2.10 Let C be a category. Prove that
a. if C has an initial object and pushouts then C has binary coproducts and
coequalizers;
b. if C has a final object and pullbacks then C has binary products and
equalizers.
2.11 Let C be a category. Prove that
a. if C has binary coproducts and coequalizers then C has pushouts;
b. if C has binary products and equalizers then C has pullbacks.
2.12 Let C be a category. Prove that
a. if C has binary coproducts, coequalizers and an initial object then C has
finite colimits;
b. if C has binary products, equalizers and a final object then C has finite
limits.

10 In other words, split (co)equalizers are preserved by any functor.

11

i∈I HomC (Ai , C) denotes the product in the category Set of sets.
150 2 Limits and Colimits

2.13 Let C be a category. Prove that

a. if C has an initial object and pushouts then C has finite colimits;
b. if C has a final object and pullbacks and then C has finite limits.
2.14 A morphism e ∈ HomC (X, Y ) is called a regular monomorphism if there exist
Z ∈ Ob C and f , g ∈ HomC (Y, Z) such that (X, e) is the equalizer of (f, g).
Dually, a morphism q ∈ HomC (X, Y ) is called a regular epimorphism if q is
a regular monomorphism in C op . Show that
a. any split epimorphism (resp. monomorphism) is a regular epimorphism
(resp. monomorphism);
b. any regular epimorphism (resp. monomorphism) is a strong epimorphism
(resp. monomorphism).
2.15 An epimorphism f ∈ HomC (A, B) is called an extremal epimorphism if
it does not factor through any proper subobject of B (i.e., if f = g ◦ h
and g is a monomorphism then g must be an isomorphism). Show that in
a balanced category any epimorphism is extremal. Formulate and prove the
dual statement.
2.16 Give an example to show that a category with pullbacks (resp. pushouts) does
not necessarily have equalizers (resp. coequalizers).
2.17 Let C be a category and f , g ∈ HomC (A, B). Show that if the triple (C, u, u)
is the pushout of (f, g) then the pair (C, u) is the coequalizer of (f, g).
2.18 Consider the following commutative diagram in an arbitrary category C and
assume that the right-side square is a pullback:

Show that the left-side square is a pullback if and only if the outer rectangle is
a pullback.
2.19 Let G1 and G2 be two groups with a common subgroup H and i : H → G1 ,
j : H → G2 the inclusion morphisms.
 
a. Show that G1 ∗H G2 , f  , g  is the pushout of (i, j ), where G1 ∗H G2
denotes the free product with amalgamated subgroup and f  : G2 → G1 ∗H
G2 and g  : G1 → G1 ∗H G2 are its corresponding group morphisms.
b. Describe the pushout of (i, j ) when H (resp. G1 ) is the trivial group.
2.20 Let C be a category with equalizers and F : C → D a functor which preserves
equalizers and reflects isomorphisms. Prove that F is faithful.
2.21 Let I and C be two arbitrary categories and assume that I is small and has an
initial object. Show that any functor F : I → C has a limit.
2.8 Exercises 151

2.22 Let (X, ) be a pre-ordered set and PO(X, ) the corresponding category.
Describe (co)limits in PO(X, ).
2.23 Let G be a group and G the corresponding category. Is G (co)complete?
2.24 Let PO(Z, ) be the category corresponding to the poset (Z, ), where  is
the usual ordering on the integers. Decide if the identity functor Id : PO(Z, )
→ PO(Z, ) has a (co)limit.
2.25 Decide if the following categories are (co)complete: Grp, Ab, Top, Ring,
R M, Field. Describe (co)limits whenever they exist.
Chapter 3
Adjoint Functors

Adjoint functors were first defined by Kan ([31]) in the 50s, motivated by homo-
logical algebra ([19, 28]). Nowadays they are present in most fields of mathematics,
as will be shown in the forthcoming examples. The terminology was inspired by
adjoint operators, whose definition is somewhat similar to the correspondence in
Definition 3.1.1.

3.1 Definition and Generic Examples

We start by introducing the main characters of this chapter: adjoint functors.

Definition 3.1.1 An adjunction consists of a pair of functors .F : C → D,
G : D → C and for any .X ∈ Ob C, Y ∈ Ob D a bijective map
.

θX, Y : HomD (F (X), Y ) → HomC (X, G(Y ))

.

which is natural in both variables. In this case, we say that F is left adjoint to G or
equivalently that G is right adjoint to F and the notation .F  G is used to designate
such a pair of adjoint functors.
Unpacking the above naturality assumption in the two variables comes down
to the following: for any .Y ∈ Ob D, θ−,Y is a natural isomorphism between
the (contravariant) functors .HomD (F (−), Y ) and .HomC (−, G(Y )) and for any
.X ∈ Ob C, θX,− is a natural isomorphism between the (covariant) functors

.HomD (F (X), −) and .HomC (X, G(−)). In particular, this amounts to the com-

mutativity of the following diagrams for all .f ∈ HomC (X  , X) and .g ∈

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 153
A. Agore, A First Course in Category Theory, Universitext,
https://doi.org/10.1007/978-3-031-42899-9_3
154 3 Adjoint Functors

HomD (Y, Y  ):

. (3.1)

. (3.2)

Definition 3.1.1 can also be stated in an equivalent manner in terms of bifunctors.

Indeed, recall from Example 1.5.6, (1) that the right and the left associated functors
(as defined in Proposition 1.5.5) of the Hom bifunctor .HomC (−, −) : Cop × C →
Set with respect to an object C in .C are precisely .HomC (C, −) and .HomC (−, C),
respectively. Furthermore, Proposition 1.7.4 shows that for bifunctors naturality
is equivalent to the naturality of both the left and the right associated functors.
Therefore, an adjunction between two functors .F : C → D and .G : D → C
can be defined equivalently as a natural isomorphism .θ between the bifunctors
.HomD (F
op (−), −) : Cop × D → Set and .Hom (−, G(−)) : Cop × D → Set,
C
where .F : Cop → Dop denotes the dual functor as defined in Proposition 1.8.1.
op

3.2 Adjoints Via Free Objects

Most of the categories we have considered so far are categories of sets endowed with
some extra structure (e.g., groups, rings, vector spaces, algebras, topological spaces
etc.) which allow for various forgetful functors: for example, from Grp to Set, from
.K M to Set, from .AlgK (or .Alg ) to .K M. It turns out that all the forgetful functors
c
K
mentioned above do have left adjoints and this phenomenon can be explained by
the existence of the so-called free objects; to be more precise, we have a free group
(vector space) on any set, a free algebra (i.e., the tensor algebra) on any vector
space and a free commutative algebra (i.e., the symmetric algebra) on any vector
space. These free objects, together with their universal property, will be the main
ingredients in the construction of the left adjoints for the aforementioned forgetful
3.2 Adjoints Via Free Objects 155

functors. We consider below the case of the forgetful functor from Grp to Set, but
the same strategy works in general.
Example 3.2.1 Let .U : Grp → Set be the forgetful functor. We will see that U has
a left adjoint .F : Set → Grp called the free group functor. More precisely, F is
constructed as follows:
• for any .X ∈ Ob Set, define .F (X) = F X, the free group on the set X;1
• given .f ∈ HomSet (X, Y ), define .F (f ) : F X → F Y by .F (f ) = f ,
where .f is obtained from the universal property of the free group F X, i.e.,
.f ∈ Hom
Grp (F X, F Y ) is the unique group homomorphism which makes the
following diagram commute:

. (3.3)

where .iX and .iY are the inclusion maps.

Let .X ∈ Ob Set and .G ∈ Ob Grp. We will prove that there is a bijection

θX,G : HomGrp (F X, G) → HomSet (X, U (G)), given by θX,G (v) = v ◦ iX

.

for any .v ∈ HomGrp (F X, G). The inverse of .θX,G , denoted by .ψX,G , is defined
as follows:

ψX,G : HomSet (X, U (G)) → HomGrp (F X, G),

.

ψX,G (u) = u, for all u ∈ HomSet (X, U (G)),

where .u ∈ HomGrp (F X, G) is the unique group homomorphism which makes the

following diagram commute:

. (3.4)

1A group G containing X as a subset is called the free group on X if for every group .G and
every function .f : X → G , there exists a unique group homomorphism .ψ : G → G such that
.f = ψ ◦ iX , where .iX : X → G is the inclusion map ([49, Section 5.5]).
156 3 Adjoint Functors

Indeed, for any .v ∈ HomGrp (F X, G) we have

ψX,G ◦ θX,G (v) = ψX,G (v ◦ iX ) = v ◦ iX ,

.

where .v ◦ iX is the unique group homomorphism which makes the following

diagram commute:

Since v makes the above diagram commutative we get .ψX,G ◦ θX,G (v) = v. On the
other hand, if .u ∈ HomSet (X, U (G)), we have

θX,G ◦ ψX,G (u) = θX,G (u) = u ◦ iX ,

.

where .u is the unique group homomorphism which makes diagram (3.4) commute.
Thus .u ◦ iX = u and we obtain .θX,G ◦ ψX,G (u) = u, as desired.
Finally we check that the isomorphism .θ is natural in both variables. First,
fix .G ∈ Ob Grp and consider .f ∈ HomSet (X , X). We need to prove the
commutativity of the following diagram:

i.e., HomSet (f, U (G)) ◦ θX, G = θX , G ◦ HomGrp (F (f ), G).

.

To this end, consider .r ∈ HomGrp (F X, G); we have

HomSet (f, U (G)) ◦ θX, G (r) = HomSet (f, U (G))(r ◦ iX )

.

= r ◦ iX ◦ f
(3.3)
= r ◦ f ◦ iX = r ◦ F (f ) ◦ iX
= θX , G (r ◦ F (f )) = θX ,G ◦ HomGrp (F (f ), G)(r).
3.3 Galois Connections 157

Finally, fix .X ∈ Ob Set and consider .g ∈ HomGrp (G, G ). We are left to prove
that the following diagram is commutative:

i.e., HomSet (X, U (g)) ◦ θX, G = θX, G ◦ HomGrp (F (X), g).

.

Let .t ∈ HomGrp (F (X), G); we have

.HomSet (X, U (g)) ◦ θX, G (t) = U (g) ◦ t ◦ iX

= g ◦ t ◦ iX
= θX, G (g ◦ t) = θX, G ◦ HomGrp (F (X), g)(t).

. 

3.3 Galois Connections

Another class of generic examples of adjoint functors can be obtained from pre-
ordered sets regarded as categories (see Example 1.2.2, (2)). The general context
is the following: .(X, ) and .(Y, ) are two pre-ordered sets and we consider the
corresponding induced categories .PO(X, ) and .PO(Y, ), respectively. More-
over, functors between such categories are nothing but order-preserving functions
between the underlying pre-ordered sets, as we have seen in Example 1.5.3, (35).
If .F : PO(X, ) → PO(Y, )op and .G : PO(Y, )op → PO(X, ) are two
functors then F is left adjoint to G if and only if for all .x ∈ X and .y ∈ Y we have

.F (x)  y in Y if and only if x  G(y) in X. (3.5)

Indeed, recall that the hom sets in any category induced by a pre-ordered set have
at most one element. Therefore, condition (3.5) can be equivalently expressed as
a bijection between .HomPO(Y, )op (F (x), y) and .HomPO(X, ) (x, G(y)). This
bijection is trivially natural as we have at most one element in each hom set.
A pair of adjoint functors as above is called a Galois connection from .(X, )
to .(Y, ). Important examples of Galois connections can be found all across the
158 3 Adjoint Functors

mathematical landscape. For instance, these include the classical Galois correspon-
dence for field extensions as well as the correspondence between algebraic sets and
radical ideals in algebraic geometry.
An important example concerns the Galois correspondence for field extensions.
Example 3.3.1 Throughout this example, .K ⊆ L is a field extension and for any
field .S ⊆ L we define

Gal(L/S) = {ψ ∈ Aut (L) | ψ(x) = x, for all x ∈ S} and G = Gal(L/K).

.

Furthermore, let

A = {S ⊆ L subfield | K ⊆ S ⊆ L} and
.

B = {H ⊆ G | H subgroup in G}.

Then .(A, ⊆) and .(B, ⊆) are pre-ordered sets, where by a slight abuse of notation
we use “.⊆” to denote both inclusions. We can now construct a Galois connection
from .(A, ⊆) to .(B, ⊆) as follows for all .S ∈ A and .H ∈ B:

F : (A, ⊆) → (B, ⊆)op , F (S) = Gal(L/S),

.

G : (B, ⊆)op → (A, ⊆), G(H ) = Fix(H ) = {l ∈ L | τ (l) = l for all τ ∈ H }.

First note that if .S ∈ A then any .ψ in .Gal(L/S) fixes all elements of S and, in
particular, those of K, which shows that .Gal(L/S) is a subgroup of .Gal(L/K) = G;
thus .F (S) ∈ B. Similarly, if H is a subgroup of G, and all elements of G fix all
elements of K, it follows that .K ⊆ Fix(H ); we can now conclude that .G(H ) ∈
A.
Moreover, if .S ⊆ S  and .ψ ∈ Gal(L/S  ) then .ψ(x) = x for all .x ∈ S  and, in
particular, the same holds for all .x ∈ S. This shows that .Gal(L/S  ) ⊆ Gal(L/S)
and therefore F is order-preserving, i.e., a functor between the corresponding
categories. Now if .H ⊆ H  and .l ∈ Fix(H  ) then .τ (l) = l for all .τ ∈ H 
and, in particular, the same holds for all .τ ∈ H . Therefore, .Fix(H  ) ⊆ Fix(H )
and G is also order-preserving. We are left to show that (3.5) also holds. Indeed
for all .S ∈ A and .H ∈ B we have .G(H ) = Fix(H ) ⊇ S if and only if
.τ (y) = y for all .τ ∈ H and .y ∈ S if and only if .H ⊆ F (S), as desired.

.

In fact, the above bijective correspondence is not specific to the theory of

fields and can be generalized by replacing the automorphism group in the previous
example by an arbitrary group. To this end, let G be a group and . : G × X →
X a group action on a set X. If we consider .A = P(X) and .B = {H ⊆
G | H subgroup in G}, then .(A, ⊆) and .(B, ⊆) are pre-ordered sets, where again
we use “.⊆” to denote both inclusions. We can now define a Galois connection from
3.4 More Examples and Properties of Adjoint Functors 159

(A, ⊆) to .(B, ⊆) as follows for all .Y ∈ A and .H ∈ B:

.

F : (A, ⊆) → (B, ⊆)op , F (Y ) = {h ∈ G | h y = y for all y ∈ Y },

.

G : (B, ⊆)op → (A, ⊆), G(H ) = {x ∈ X | h x = x for all h ∈ H }.

First, note that the properties of a group action imply that the set

.{h ∈ G | h y = y for all y ∈ Y }

is in fact a subgroup of G and therefore .F (Y ) ∈ B for all .Y ∈ A. Furthermore, if

.Y ⊆ Y  and .h ∈ F (Y  ) then .h y = y for all .y ∈ Y  and, in particular, for all .y ∈ Y ;
this implies that .F (Y  ) ⊆ F (Y ) and therefore, F is a well-defined functor from
 
.(A, ⊆) to .(B, ⊆) . Similarly, if .H, H ∈ B with .H ⊆ H and .x ∈ G(H ) then
op 

.h
 
x = x for all .h ∈ H and, in particular, for all .h ∈ H . Hence .G(H ) ⊆ G(H )
and G is a well-defined functor from .(B, ⊆)op to .(A, ⊆). Finally, the two functors
fulfill condition (3.5); indeed, given .Y ∈ A and .H ∈ B we have .F (Y ) ⊇ H if and
only if .h y = y for all .y ∈ Y and .h ∈ H if and only if .Y ⊆ G(H ). To conclude,
we have proved that the functors F and G form a Galois connection.

3.4 More Examples and Properties of Adjoint Functors

We start this section with more examples of adjoint functors, spanning various fields.
Examples 3.4.1
(1) For any non-empty set X, the functor .− × X : Set → Set has a right adjoint
given by .HomSet (X, −) : Set → Set. Indeed, for all .Y, Z ∈ Ob Set, define
 
θY, Z : HomSet (Y × X, Z) → HomSet Y, HomSet (X, Z) ,
.

θY, Z (f )(y)(x) = f (y, x),

for all .f ∈ HomSet (Y × X, Z), and all .y ∈ Y, x ∈ X. Then .θY, Z is bijective

with inverse given as follows:
 
ψY,. Z : HomSet Y, HomSet (X, Z) → HomSet (Y × X, Z),
 
ψY, Z (g)(y, x) = g(y) (x),
 
for all .g ∈ HomSet Y, HomSet (X, Z) , y ∈ Y and .x ∈ X.
160 3 Adjoint Functors

Indeed, we have
   
.ψY, Z ◦ θY, Z (f )(y, x) = θY, Z (f )(y) (x) = f (y, x),
   
θY, Z ◦ ψY, Z (g)(y)(x) = ψY, Z (g)(y, x) = g(y) (x).

We are left to show the commutativity of diagrams (3.1) and (3.2). To start with,
let .f ∈ HomSet (Y  , Y ); we need to show the commutativity of the following
diagram:

For any .t ∈ HomSet (Y × X, Z), y  ∈ Y  , x ∈ X we have

   
.HomSet f, HomSet (X, Z) ◦ θY, Z (t) (y  )(x)
   
= θY, Z (t) ◦ f (y  )(x) = t f (y  ), x
 
= t ◦ (f × 1X )(y  , x) = θY  , Z ◦ t ◦ (f × 1X ) (y  )(x)
   
= θY  , Z ◦ HomSet f × 1X , Z (t) (y  )(x)

and the commutativity of (3.1) is proved. Consider now .g ∈ HomSet (Z, Z  );

we are left to show the commutativity of the following diagram:

Indeed, if .t ∈ HomSet (Y × X, Z), y ∈ Y , .x ∈ X we have

 
.θY, Z  ◦ HomSet (Y × X, g)(t) (y)(x) = θY, Z  (g ◦ t)(y)(x) = g ◦ t (y, x)
   
= g ◦ θY, Z (t)(y)(x) = HomSet Y, HomSet (X, g) ◦ θY, Z (t) (y)(x),

which proves that (3.2) is also commutative.

(2) Let K be a field and denote the tensor product over K simply by .⊗ (i.e., .⊗ =
⊗K ). If .X ∈ Ob K M, then the tensor product functor .− ⊗ X : K M → K M (see
3.4 More Examples and Properties of Adjoint Functors 161

Example 1.5.3, (20)) has a right adjoint given by .HomK M (X, −) : KM → KM

(Example 1.5.3, (13)). Indeed, for all .Y, Z ∈ Ob K M define
 
θY, Z : HomK M (Y ⊗ X, Z) → HomK M Y, HomK M (X, Z) ,
.

θY, Z (f )(y)(x) = f (y ⊗ x) for all f ∈ HomK M (Y ⊗ X, Z), y ∈ Y, x ∈ X.

The inverse of .θY, Z is given as follows:

 
ψY, Z : HomK M Y, HomK M (X, Z) → HomK M (Y ⊗ X, Z),
.

 
ψY, Z (g)(y ⊗ x) = g(y) (x),
 
for all .g ∈ HomK M Y, HomK M (X, Z) , y ∈ Y and .x ∈ X.
Showing that the maps defined above are indeed K-linear is straightforward
using the properties of the tensor product while proving the commutativity of
the diagrams (3.1) and (3.2) goes very much along the lines of the previous
example and is left to the reader.
(3) If X is a locally compact and Hausdorff topological space then the
functor .− × X : Top → Top (see Example 1.5.3, (18)) is left adjoint to
.Hom
Top (X, −) : Top → Top (see Example 1.5.3, (11)) as proved, for instance,
in [12, Chapter 5] in a more general setting. Recall that for any .Y ∈ Ob Top
we consider on .HomTop (X, Y ) the compact-open topology while .Y × X is
endowed with the product topology.
In the first example above we proved that for any .Y, Z  ∈ Ob Set we have 
a set bijection between .HomSet (Y × X, Z) and .HomSet Y, HomSet (X, Z) .
We will show that this bijection induces
 a continuous bijective
 map between
.Hom
Top (Y × X, Z) and .Hom
Top Y, Hom Top (X, Z) with respect to the
previously mentioned topologies. To this end, for all Y , .Z ∈ Ob Top define
 
θY, Z : HomTop (Y × X, Z) → HomTop Y, HomTop (X, Z) ,
.

θY, Z (f )(y)(x) = f (y, x) for all f ∈ HomTop (Y × X, Z), y ∈ Y, x ∈ X.

In order to show that .θY, Z is well-defined we need to  prove that if .f ∈

HomTop (Y × X, Z) then .θY, Z (f ) = f ∈ HomTop Y, HomTop (X, Z) .
Consider .W (K, V ) to be a sub-basic open set of .HomTop (X, Z), i.e., .K ⊆ X
is a compact subset, .V ⊆ Z is an open subset and

W (K, V ) = {f ∈ HomTop (X, Z) | f (K) ⊆ V }.

.

−1  
If .f W (K, V ) = ∅ then .f is obviously continuous, as desired. Assume now
−1   −1  
that .f W (K, V ) = ∅ and consider .y ∈ f W (K, V ) . Therefore, we
have .f (y) ∈ W (K, V ) and we obtain .f (y)(k) = f (y, k) ∈ V for all .k ∈ K.
162 3 Adjoint Functors

Since .f : Y ×X → Z is continuous, it follows that .f −1 (V ) = {y}×K ⊆ Y ×X

is an open subset. Now the tube lemma2 implies that there exist open subsets
.Uy ⊆ Y and .Ty ⊆ X such that

{y} × K ⊆ Uy × Ty ⊆ f −1 (V ).
.

−1  
This shows that .y ∈ Uy ⊆ f W (K, V ) and therefore we have

−1   
f
. W (K, V ) = Uy .
−1
 
y∈f W (K, V )

−1  
We can now conclude that .f W (K, V ) is an open set as a union of open
sets. To summarize, we proved that if .f : Y × X → Z is continuous then
.f = θY, Z (f ) : Y → Hom
Top (X, Z) is continuous as well.
Consider now .ψY, Z , the inverse of .θY, Z , given as follows:
 
ψY, Z. : HomTop Y, HomTop (X, Z) → HomTop (Y × X, Z),
 
ψY, Z (g)(y, x) = g(y) (x),
 
for all .g ∈ HomTop Y, HomTop (X, Z) , y ∈ Y and .x ∈ X.
We are left to show that .ψY, Z is well-defined, i.e., if .g : Y →
HomTop (X, Z) is continuous then .g = ψY, Z (g) : Y × X → Z is continuous
as well. We start by showing that the evaluation map .ev : HomTop (X, Z) ×
X → Z defined by .ev(g, x) = g(x) is continuous at every point3
.(g, x) ∈ Hom
Top (X, Z) × X. Indeed, if .V ⊆ Z is an open set such that
.ev(g, x) = g(x) ∈ V then the continuity of g implies that .g
−1 (V ) ⊆ X
−1
is an open subset such that .x ∈ g (V ). As X is Hausdorff and locally
compact4 we can find an open subset .U ⊆ X whose closure  .U is compact
and .x ∈ U ⊆ U ⊆ g −1 (V ). Therefore, we have .g(x) ∈ g U ⊆ V , which
shows that .W (U , V ) × U ⊆ HomTop (X, Z) × X is an open subset such that
 
.(g, x) ∈ W (U , V ) × U and .ev W (U , V ), U ⊆ V . Hence .ev is a continuous

map. Now, as we assumed .g : Y → HomTop (X, Z) to be continuous, the

2 The tube lemma: Let A and B compact subspaces of X and Y , respectively, and let N be an open

set in .X × Y containing .A × B. Then, there exist open subsets U and V in X and Y , respectively,
such that .A × B ⊆ U × V ⊆ N (see [39, Lemma 26.8 and exercise 9 on page 171]).
3 Let .f : A → B be a map between two topological spaces. We say that f is continuous at a

point .x ∈ A if, for each neighborhood V of .f (x), there is a neighborhood U of x whose closure
.f (U ) ⊂ V . The map f is continuous if and only if is continuous at every point .x ∈ A ([39,
Theorem 18.1]).
4 Recall that if X is a Hausdorff space then X is locally compact if and only if given .x ∈ X and

given a neighborhood U of x, there is a neighborhood V of x whose closure .V is compact and

.V ⊆ U ([39, Theorem 29.2]).
3.4 More Examples and Properties of Adjoint Functors 163

following composition is also continuous:

Furthermore, for all .(y, x) ∈ Y × X we have

 
ev ◦ (g × 1X )(y, x) = ev g(y), x = g(y)(x) = ψY, Z (g)(y, x).
.

We can now conclude that .ψY, Z (g) = ev ◦ (g × 1X ) is indeed continuous, as

desired.
For a more general and comprehensive account of the various topologies that
can be defined on a set of continuous maps and their behaviour with respect to
the above adjunction, we refer to [12, Chapter 5].
(4) The inclusion functor .I : Ab → Grp is right adjoint to the abelianization
functor .F : Grp → Ab defined in Example 1.5.3, (23). Indeed, for any
.A ∈ Ob Ab and .G ∈ Ob Grp, consider .θG, A : Hom
Ab (F (G), A) →
HomGrp (G, I (A)) defined by

θG, A (f ) = f ◦ πG , for all f ∈ HomAb (F (G), A),

.

where .πG : G → Gab is the canonical projection. Furthermore, given a group

homomorphism .g ∈ HomGrp (G, I (A)), since A is an abelian group it can
be easily seen that .[G, G] ⊆ ker(g), where .[G, G] denotes the commutator
subgroup. Therefore, the universal property of the quotient group .Gab yields a
unique group homomorphism .h ∈ HomAb (Gab , A) such that .h ◦ πG = g. This
shows that .θA, G is bijective for all .A ∈ Ob Ab and .G ∈ Ob Grp. We are left
to show that .θ defined above makes diagrams (3.1) and (3.2) commutative. To
this end, let .f ∈ HomGrp (G , G) and .A ∈ Ob Ab; we start by showing the
commutativity of the following diagram:

For any .t ∈ HomAb (Gab , A) we have

HomGrp (f, A) ◦ θG, A (t) = HomGrp (f, A)(t ◦ πG )

.

(1.6)
= t ◦ πG ◦ f = t ◦ fab ◦ πG
= θG , A (t ◦ fab ) = θG , A ◦ HomAb (fab , A)(t),
164 3 Adjoint Functors

which shows that (3.1) is commutative. Consider now .g ∈ HomAb (A, A ) and
.G ∈ Ob Grp. We are left to show the commutativity of the following diagram:

Indeed, for any .t ∈ HomAb (Gab , A) we have

HomGrp (G, g) ◦ θG, A (t) = HomGrp (G, g)(t ◦ πG ) = g ◦ t ◦ πG

.

= θG, A (g ◦ t) = θG, A ◦ HomAb (Gab , g)(t).

Therefore, (3.2) is commutative and we have proved that F is left adjoint to I .

(5) Let .1 be the discrete category with one object denoted by .. For any category .C
we can define a unique functor .T : C → 1. It can be easily seen that the functor
T has a left (resp. right) adjoint if and only if .C has an initial (resp. final) object.
Indeed, .L : 1 → C is a left adjoint for T if and only if for any .C ∈ Ob C
C : HomC (L(), C) → Hom1 (, T (C)) which
l
there exists a bijective map .θ,
is natural in both variables. As .Hom1 (, T (C)) = {1 }, .θ, l
C is a bijection
for any .C ∈ Ob C if and only if .HomC (L(), C) has one element for any
.C ∈ Ob C. This means precisely that .L() is the initial object of .C.

Similarly, .R : 1 → C is a right adjoint for T if and only if for any .C ∈ Ob C

r : Hom (T (C), ) → Hom (C, R()) which
there exists a bijective map .θC,  1 C
is natural in both variables. As .Hom1 (T (C), ) = {1 }, .θC, r
 is a bijection
for any .C ∈ Ob C if and only if .HomC (C, R()) has one element for any
.C ∈ Ob C. This means precisely that .R() is the final object of .C. .

We continue with further properties of adjoint functors. First we look at

compositions of adjoint functors.
Proposition 3.4.2 Consider the functors .F : A → B, G : B → A such that .F 
G and .H : B → C, K : C → B such that .H  K. Then .H F  GK.
Proof We have the following natural isomorphisms for all .A ∈ Ob A and .C ∈
Ob C: .HomC (H F (A), C) ≈ HomB (F (A), K(C)) ≈ HomA (A, GK(C)). 

Any pair of adjoint functors induces an adjunction between the opposite functors
as follows:
Theorem 3.4.3 Let .F : C → D and .G : D → C be two functors. Then .F  G if
and only if .Gop  F op .
3.4 More Examples and Properties of Adjoint Functors 165

Proof Assume that .F  G. Then, for any .Y ∈ Ob C and .X ∈ Ob D we

have a bijective map .θY, X : HomD (F (Y ), X) → HomC (Y, G(X)) which is
natural in both variables. Consider now the map .θ X, Y : HomCop (Gop (X), Y ) →
HomDop (X, F op (Y )) defined for all .t op ∈ HomCop (Gop (X), Y ) by
   −1 op
θ X, Y t op = θY,
.
X (t) . (3.6)

The map defined above is bijective with inverse given by

   op
. ξX, Y uop = θY, X (u) (3.7)

for all .uop ∈ HomDop (X, F op (Y )) Indeed, for all .t op ∈ HomCop (Gop (X), Y ) and
.u
op ∈ Hom op (X, F op (Y )) we have
D
   −1  op
θ X, Y ◦ ξX, Y (uop ) = θ X, Y θY, X (u)op = θY,
.
X θ Y, X (u) = uop ,
 op    −1 op
−1
ξX, Y ◦ θ X, Y (t op ) = ξX, Y θY, X (t) = θY, X θ Y, X (t) = t op .

We are left to show that .θ is natural in both variables. To this end, consider first
g op ∈ HomDop (D  , D). The naturality of .θ in the first variable comes down to
.

showing the commutativity of the following diagram:

i.e., HomDop (g op , F op (C)) ◦ θ D, C = θ D  , C ◦ HomCop (Gop (g op ), C).

. (3.8)

Since we have already proved that each .θ D, C is invertible with inverse .ξD, C it will
suffice to show that the following holds:

ξD  , C ◦ HomDop (g op , F op (C)) = HomCop (Gop (g op ), C) ◦ ξD, C .

.

To this end, given .uop ∈ HomDop (D, F op (C)), we have

 
ξD  , C ◦ HomDop (g op , F op (C))(uop ) = ξD  , C uop ◦op g op
.
 
= ξD  , C (g ◦ u)op
 op
= θC, D  (g ◦ u)
166 3 Adjoint Functors


(3.2) op
= G(g) ◦ θC, D (u)
 op
= θC, D (u) ◦op G(g)op
 op 
= HomCop (Gop (g op ), C) θC, D (u)

= HomCop (Gop (g op ), C) ◦ ξD, C (uop ),

which shows that (3.8) indeed holds.

Next we show that .θ is also natural in the second variable. Consider .f op ∈
HomCop (C, C  ). The naturality of .θ in the second variable comes down to showing
the commutativity of the following diagram:

i.e., θ D, C  ◦ HomCop (Gop (D), f op ) = HomDop (D, F op (f op )) ◦ θ D, C .

. (3.9)

Relying again on the fact that .θ D, C is invertible with inverse .ξD, C it will suffice to
show that the following holds:

HomCop (Gop (D), f op ) ◦ ξD, C = ξD, C  ◦ HomDop (D, F op (f op )).

.

Indeed, given .uop ∈ HomDop (D, F op (C)), we have

 op 
HomCop (Gop (D), f op ) ◦ ξD, C (uop ) = HomCop (Gop (D), f op ) θC, D (u)
.

 op
= f op ◦op θC, D (u)
 op
= θC, D (u) ◦ f
(3.1)
 op
= θC  , D (u ◦ F (f ))
 
= ξD, C  (u ◦ F (f ))op
 
= ξD, C  F (f )op ◦op uop
= ξD, C  ◦ HomDop (D, F op (f op ))(uop ).

Hence (3.9) holds and this shows the naturality of .θ in the second variable.
3.4 More Examples and Properties of Adjoint Functors 167

Conversely, assume now that .Gop  F op . Then, as proved above, we have

.F
op op  Gop op , which comes down to .F  G, as desired. 

The next result gives a necessary condition for the existence of adjoints: if a
left (resp. right) adjoint of a functor F exists, then F has to preserve small limits
(resp. colimits). This is, however, not a sufficient condition for a functor to admit an
adjoint, as we will see in Sect. 3.11.
Theorem 3.4.4 Consider the functors .F : A → B, G : B → A such that .F  G.
Then F preserves small colimits while G preserves small limits.
Proof We start by showing that G preserves all existing small limits of .B. Consider
the natural isomorphism .θ : HomB (F (−), −) → HomA (−, G(−)) corresponding
to the adjunction .F  G. Let  I be a small category and .H : I → B a functor
whose limit we denote by . L, (pi : L → H (i))i∈Ob I . We will prove that
 
. G(L), (G(pi ) : G(L) → GH (i))i∈Ob I
  limit of .GH : I → A. To start
is the
with, . G(L), (G(pi ) : G(L) →  GH (i)) i∈Ob I is a cone onGH by Lemma 2.5.5.
Consider now another cone . A, (qi : A → GH (i))i∈Ob I on GH . Since the map
.θA, H (i) : Hom (F (A), H (i)) → HomA (A, GH (i)) is a bijection, there exists a
B
unique morphism .ri ∈ Hom (F (A), H (i)) such that .θA, H (i) (ri ) = qi . We will
 B 
prove that . F (A), (ri : F (A) → H (i))i∈Ob I is a cone on H , i.e., for any .d ∈
HomI (i, j ) we have .H (d) ◦ ri = rj . To this end, the naturality of .θ renders the
following diagram commutative:

i.e., HomA (A, GH (d)) ◦ θA, H (i) = θA, H (j ) ◦ HomB (F (A), H (d)).
. (3.10)
 
Moreover, since . A, (qi : A → GH (i))i∈Ob I is a cone on GH the following
diagram is commutative:

. (3.11)
168 3 Adjoint Functors

Now by evaluating (3.10) at .ri we obtain

HomA (A, GH (d)) ◦ θA, H (i) (ri ) = θA, H (j ) ◦ HomB (F (A), H (d))(ri )
.

⇔ HomA (A, GH (d))(qi ) = θA, H (j ) (H (d) ◦ ri )

⇔ GH (d) ◦ qi = θA, H (j ) (H (d) ◦ ri )
(3.11)
⇔ qj = θA, H (j ) (H (d) ◦ ri ) ⇔ rj = H (d) ◦ ri .
 
Thus . F (A), (ri )i∈Ob I is a cone on H . Hence, there exists a unique morphism
.f ∈ Hom (F (A), L) such that the following diagram is commutative for all .i ∈
B
Ob I :

. (3.12)

Denote .θA,L (f ) ∈ HomA (A, G(L)) by g. We are left to prove that the following
diagram is commutative for all .i ∈ Ob I :

. (3.13)

Using again the naturality of the bijection .θ we obtain the following commutative
diagram for all .i ∈ Ob I :

i.e., HomA (A, G(pi )) ◦ θA, L = θA, H (i) ◦ HomB (F (A), pi ).

. (3.14)
3.4 More Examples and Properties of Adjoint Functors 169

By evaluating (3.14) at .f ∈ HomB (F (A), L) we obtain

HomA (A, G(pi )) ◦ θA, L (f ) = θA, H (i) ◦ HomB (F (A), pi )(f )

.

⇔ G(pi ) ◦ g = θA, H (i) (pi ◦ f )

(3.12)
⇔ G(pi ) ◦ g = θA, H (i) (ri ) ⇔ G(pi ) ◦ g = qi

for all .i ∈ Ob I . Hence the diagram (3.13) is indeed commutative.

Assume now that there exists another morphism .g ∈ HomA (A, G(L)) such
that .G(pi ) ◦ g = qi for all .i ∈ Ob I . Since .θA, L : HomB (F (A), L) →
HomA (A, G(L)) is bijective, there exists a unique morphism .f ∈ HomB (F (A), L)
such that .θA, L (f ) = g. Now by evaluating (3.14) at .f we arrive at

HomA (A, G(pi )) ◦ θA, L (f ) = θA, H (i) ◦ HomB (F (A), pi )(f )

.

⇔ G(pi ) ◦ g = θA, H (i) (pi ◦ f )

⇔ qi = θA, H (i) (pi ◦ f )

(3.12)
for any .i ∈ Ob I . Therefore, we have .pi ◦ f = ri = pi ◦ f for all .i ∈ Ob I . By
Proposition 2.2.14, (1) this implies .f = f and consequently .g = g, as desired.
The second part of the theorem follows easily by duality. Indeed, if .F  G
then Theorem 3.4.3 implies that we also have .Gop  F op . According to the above
proof, .F op preserves all existing limits. Now using Lemma 2.5.2 we obtain that F
preserves colimits, as desired. 

Theorem 3.4.4 can be very useful in ruling out the existence of left/right adjoints
for certain functors, as shown in the following examples:
Examples 3.4.5
(1) The forgetful functor .F : Ab → Set does not preserve coproducts. Therefore,
by Theorem 3.4.4 it does not have a right adjoint.
(2) Consider now the inclusion functor .I : Ring → Rng. As .Z is an initial object
in .Ring but not in .Rng we can conclude by Theorem 3.4.4 that it does not admit
a right adjoint.
(3) The forgetful functor .U : Field → Set does not have a left adjoint. Indeed,
if .F : Set → Field is a left adjoint to U then by Theorem 3.4.4, F needs
to preserve colimits. In particular, this would imply the existence of an initial
object in Field, which contradicts Example 1.3.10, (4). .
170 3 Adjoint Functors

3.5 The Unit and Counit of an Adjunction

Our next result gives an important equivalent description of adjoint functors in terms
of two natural transformations called the unit and the counit of the adjunction.
Theorem 3.5.1 Let .F : C → D and .G : D → C be two functors. Then F is left
adjoint to G if and only if there exist two natural transformations

.η : 1C → GF, ε : F G → 1D

such that for all .C ∈ Ob C and .D ∈ Ob D we have

 
1F (C) = εF (C) ◦ F ηC .
. (3.15)
 
1G(D) = G εD ◦ ηG(D) (3.16)

In this case .η and .ε are called the unit and the counit of the adjunction, respectively.
Proof Suppose first that .F  G and let .θ : HomD (F (−), −) → HomC (−, G(−))
be the corresponding natural isomorphism. For each .C ∈ Ob C and .D ∈ Ob D we
have the following bijective maps:

.θC, F (C) : HomD (F (C), F (C)) → HomC (C, GF (C)),

θG(D), D : HomD (F G(D), D) → HomC (G(D), G(D)).
  −1  
Now define .ηC = θC, F (C) 1F (C) : C → GF (C) and .εD = θG(D), D 1G(D) :
F G(D) → D. We are left to prove (3.15) and (3.16) as well as the naturality
of .η and .ε. We start by proving (3.15); indeed, if we consider the commutative
diagram (3.1) for .X = GF (C), X = C, Y = F (C) and .f = ηC ∈
HomC (C, GF (C)) we obtain

From the commutativity of the above diagram applied to

. εF (C) ∈ HomD (F GF (C), F (C))

3.5 The Unit and Counit of an Adjunction 171

we obtain

HomC (ηC , GF (C)) ◦ θGF (C), F (C) (εF (C) )

.

= θC, F (C) ◦ HomD (F (ηC ), F (C))(εF (C) )

 
⇔ θGF (C), F (C) (εF (C) ) ◦ ηC = θC, F (C) εF (C) ◦ F (ηC )
 
⇔ 1GF (C) ◦ ηC = θC, F (C) εF (C) ◦ F (ηC )
−1
⇔ θC, F (C) (ηC ) = εF (C) ◦ F (ηC )

⇔ 1F (C) = εF (C) ◦ F (ηC ) i.e., (3.15) holds.

Similarly, by considering the commutative diagram (3.2) for .X = G(D), Y =

F G(D), Y  = D and .g = εD ∈ HomD (F G(D), D) we get

The commutativity of the above diagram applied to

1F G(D) ∈ HomD (F G(D), F G(D))

.

yields

θG(D), D ◦ HomD (F G(D), εD )(1F G(D) )

.
 
= HomC G(D), G(εD ) ◦ θG(D), F G(D) (1F G(D) )

⇔ θG(D), D ◦ HomD (F G(D), εD ) ◦ 1F G(D) = G(εD ) ◦ ηG(D)

⇔ θG(D), D (εD ) = G(εD ) ◦ ηG(D)

⇔ 1G(D) = G(εD ) ◦ ηG(D) i.e., (3.16) holds.

Finally, we move on to proving that .η and .ε are natural transformations. First we will
collect some compatibilities using the commutativity of the diagrams (3.1) and (3.2).
Setting .X = C, X = C  and .Y = F (C) in (3.1) yields the following commutative
172 3 Adjoint Functors

diagram for all .f ∈ HomC (C  , C):

From the commutativity of the above diagram applied to .1F (C) we get

HomC (f, GF (C)) ◦ θC, F (C) (1F (C) ) = θC  , F (C) ◦ HomD (F (f ), F (C))(1F (C) )
.
 
⇔ HomC (f, GF (C)) ◦ ηC = θC  , F (C) F (f )
 
i.e., ηC ◦ f = θC  , F (C) F (f ) .
. (3.17)

On the other hand, setting .X = C  , Y = F (C  ) and .Y  = F (C) in (3.2) yields the

following commutative diagram for all .g ∈ HomD (F (C  ), F (C)):

By applying the commutativity of the above diagram to .1F (C  ) we obtain

HomC (C  , G(g)) ◦ θC  , F (C  ) (1F (C  ) ) = θC  , F (C) ◦ HomD (F (C  ), g)(1F (C  ) ),

.

. i.e., G(g) ◦ ηC  = θC  , F (C) (g). (3.18)

Next we use the commutativity of the diagram (3.1) for .X = G(D), X =

G(D  ) and .Y = D. It comes down to the following commutative diagram for all

.f ∈ HomC (G(D ), G(D)):

From the commutativity of the above diagram applied to .εD we get

3.5 The Unit and Counit of an Adjunction 173

θG(D  ), D ◦ HomD (F (f ), D)(εD ) = HomC (f, G(D)) ◦ θG(D), D (εD )

.

⇔ θG(D  ), D (εD ◦ F (f )) = 1G(D) ◦ f,

−1
. i.e., εD ◦ F (f ) = θG(D  ), D (f ). (3.19)

Finally, we use the commutativity of the diagram (3.2) for .X = G(D  ), Y = D  and
 
.Y = D. It yields the following commutative diagram for all .g ∈ HomD (D , D):

The commutativity of the above diagram applied to .εD  gives

θG(D  ), D ◦ HomD (F G(D  ), g)(εD  ) = HomC (G(D  ), G(g)) ◦ θG(D  ), D  (εD  )

.

⇔ θG(D  ), D (g ◦ εD  ) = G(g) ◦ 1G(D  ) ,

−1  
i.e., g ◦ εD  = θG(D
.  ), D G(g) . (3.20)

We are now in a position to prove that .η and .ε are natural transformations. Indeed,
the naturality of .η comes down to proving the commutativity of the following
diagram for all .h ∈ HomC (C  , C):

To this end we have

(3.17)   (3.18)
ηC ◦ h = θC  , F (C) F (h) = GF (h) ◦ ηC  ,
.

where in the second equality we used (3.18) for .g = F (h). Thus .η is a natural
transformation.
The naturality of .ε comes down to proving the commutativity of the following
diagram for all .t ∈ HomD (D  , D):
174 3 Adjoint Functors

which can be proved using (3.19) and respectively (3.20)

(3.19) −1   (3.20)
.εD ◦ F G(t) = θG(D  ), D G(t) = g ◦ εD  .

Note that the first equality follows by applying (3.19) for .f = G(t).
Assume now that there exist two natural transformations .η : 1C → GF and
.ε : F G → 1D such that (3.15) and (3.16) are fulfilled for any .C ∈ Ob C and

.D ∈ Ob D. Define the following maps

.θC, D : HomD (F (C), D) → HomC (C, G(D)), θC, D (u) = G(u) ◦ ηC ,

ϕC, D : HomC (C, G(D)) → HomD (F (C), D), ϕC, D (v) = εD ◦ F (v),

for any .u ∈ HomD (F (C), D) and .v ∈ HomC (C, G(D)). First we will prove that
θC, D and .ϕC, D are inverses to each other for any .C ∈ Ob C and .D ∈ Ob D. To start
.

with, we note for further use that the naturality of .η and .ε imply the commutativity
of the following diagrams for all .u ∈ HomD (F (C), D) and .v ∈ HomC (C, G(D)):

. (3.21)

. (3.22)

Now, we have
   
.θC, D ◦ ϕC, D (v) = θC, D εD ◦ F (v) = G εD ◦ F (v) ◦ ηC
= G(εD ) ◦ GF (v) ◦ ηC
3.5 The Unit and Counit of an Adjunction 175

(3.21)
= G(εD ) ◦ ηG(D) ◦ v
(3.16)
= v,
   
ϕC, D ◦ θC, D (u) = ϕC, D G(u) ◦ ηC = εD ◦ F G(u) ◦ ηC
= εD ◦ F G(u) ◦ F (ηC )
(3.22)
= u ◦ εF (C) ◦ F (ηC )
(3.15)
= u.

Thus .θC, D and .ϕC, D are inverses to each other for any .C ∈ Ob C and .D ∈ Ob D.
We are left to prove that .θ is natural in both variables, i.e., diagrams (3.1) and (3.2)
are commutative. Indeed, let .f ∈ HomC (C  , C) and .u ∈ HomD (F (C), D); we
have
 
HomC (f, G(D)) ◦ θC, D (u) = HomC (f, G(D)) ◦ G(u) ◦ ηC
.

= G(u) ◦ ηC ◦ f
= G(u) ◦ GF (f ) ◦ ηC 
 
= θC  , D u ◦ F (f )
= θC  , D ◦ HomD (F (f ), D)(u),

where in the third equality we used the naturality of .η applied to f . Thus, (3.1)
holds.
Consider now .g ∈ HomD (D, D  ) and .u ∈ HomD (F (C), D). Then:

HomC (C, G(g)) ◦ θC, D (u) = G(g) ◦ G(u) ◦ ηC

.

= G(g ◦ u) ◦ ηC
= θC, D  ◦ HomD (F (C), g)(u).

This proves that (3.2) also holds and the proof is now finished. 

Examples 3.5.2
(1) Let .F : C → D be an isomorphism of categories with inverse .G : D → C. Then
.(F, G) and .(G, F ) are pairs of adjoint functors with unit and counit given

by the identity natural transformations. It is straightforward to check that the

compatibility conditions (3.15) and (3.16) are trivially fulfilled.
(2) Consider the pair of adjoint functors .F = −⊗X, .G = HomK M (X, −) : K M →
K M from Example 3.4.1, (2) where .⊗ = ⊗K . The unit and counit of the
176 3 Adjoint Functors

adjunction .F  G are given as follows for any .Y, Z ∈ Ob K M:

η : 1K M → HomK M (X, − ⊗ X),

. ε : HomK M (X, −) ⊗ X → 1K M ,

ηY : Y → HomK M (X, Y ⊗ X), ηY (y)(x) = y ⊗ x, y ∈ Y, x ∈ X,

εZ : HomK M (X, Z) ⊗ X → Z, εZ (f ⊗ x)

= f (x), f ∈ HomK M (X, Z), x ∈ X.

Indeed, for all .x ∈ X, y ∈ Y and .f ∈ HomK M (X, Y ) we have

 
.εY ⊗X ◦ (ηY ⊗ 1X )(y ⊗ x) = εY ⊗X ηY (y) ⊗ x
= ηY (y)(x) = y ⊗ x,
HomK M (X, εY ) ◦ ηHom (f )(x) = HomK M (X, εY )(f ⊗ x)
K M (X, Y )
= εY (f ⊗ x) = f (x),

which shows that (3.15) and (3.16) hold true.

(3) Similarly to the free group functor, we can construct the free module functor
.R : Set → R M as follows:

• for any .X ∈ Ob Set, define .R(X) = RX, the free module generated by X;5
• given .f ∈ HomSet (X, Y ), define .R(f ) : RX → RY by .R(f ) = f , where
.f is the unique homomorphism of R-modules which makes the following

diagram commute:

. (3.23)

where .(RX, iX ) and .(RY, iY ) are the free R-modules generated by X and
Y , respectively. .R is the left adjoint of the forgetful functor .U : R M → Set.
Indeed, the unit .η : 1Set → U R is defined for all .X ∈ Ob Set by .ηX = iX ,
where .iX : X → RX is the map corresponding to the free R-module on X
while the counit .ε : RU → 1R M is defined for all .M ∈ Ob R M as the unique
homomorphism of R-modules .εM : R(U (M)) → M such that the following

5 Thefree R-module generated by X is a pair .(RX, iX ), where RX is an R-module and .iX : X →

RX is a map, such that for any R-module M and any map .f : X → M there exists a unique
R-module homomorphism .g : RX → M such that .g ◦ iX = f (see [45, Definition 1.8]).
3.5 The Unit and Counit of an Adjunction 177

diagram is commutative:

. (3.24)

First, note that (3.23) implies in particular that for all .f ∈ HomSet (X, Y ) the
following diagram is commutative:

In other words, .η is a natural transformation. By applying U to (3.24) and

having in mind that .ηU (M) = iU (M) gives .U (εM ) ◦ ηU (M) = 1U (M) , which
shows that condition (3.16) is fulfilled. We are left to show that .ε is a natural
transformation such that (3.15) also holds. To this end, we have

(3.24) (3.23)
1RX ◦ iX = = εRX ◦ iU (RX) ◦ iX = εRX ◦ R(iX ) ◦ iX .
.

This shows that .1RX ◦ iX = εRX ◦ R(iX ) ◦ iX and in light of [45, Definition
1.8] we can conclude that .1RX = εRX ◦ R(iX ). Hence, (3.15) also holds.
Consider now .g ∈ HomR M (M, N ); the proof will be finished once we show
the commutativity of the following diagram:

By the same argument used in the above paragraph, it will suffice to show that
g ◦ εM ◦ iU (M) = εN ◦ RU (g) ◦ iU (M) . Indeed, we have
.

(3.23)
εN ◦ RU (g) ◦ iU (M) = εN ◦ iU (N ) ◦ U (g)
.

(3.24) (3.24)
= 1N ◦ g = g ◦ 1M = g ◦ εM ◦ iU (M) ,
178 3 Adjoint Functors

as desired. 
.

We record here for further use the following slightly more general version of the
compatibility conditions between the unit and counit of an adjunction:
Lemma 3.5.3 Let .F : C → D and .G : D → C be two functors such that .F 
G and consider the corresponding natural isomorphism .θ : HomD (F (−), −) →
HomC (−, G(−)). If .η and .ε are the unit and counit of this adjunction, then for all
.u ∈ HomD (F (C), D) and .v ∈ HomC (C, G(D)) we have

 
εD ◦ F θC, D (u) = u, .
. (3.25)
 −1 
G θC, D (v) ◦ ηC = v. (3.26)

Proof To start with, if we consider the commutative diagram (3.1) for .X =

G(D), X = C, Y = D and .f = θC, D (u) we get

From the commutativity of the above diagram applied to .εD ∈ HomD (F G(D), D)
we obtain
   
.HomC (θC, D (u), G(D)) ◦ θG(D), D (εD ) = θC, D ◦ HomD F θC, D (u) , D (εD )

  
⇔ θG(D), D (εD ) ◦ θC, D (u) = θC, D εD ◦ F θC, D (u)
  
⇔ 1G(D) ◦ θC, D (u) = θC, D εD ◦ F θC, D (u)

and the bijectivity of .θC, D shows that (3.25) indeed holds.

For the second identity, consider the commutative diagram (3.2) for .X = C, Y =
−1
F (C), Y  = D and .g = θC, D (v) ∈ HomD (F (C), D)

.
3.5 The Unit and Counit of an Adjunction 179

The commutativity of the above diagram applied to .1F (C) ∈ HomD (F (C),
F (C)) yields
−1
θC, D ◦ HomD (F (C), θC,
.
D (v))(1F (C) )
  −1 
= HomC C, G θC, D (v) ◦ θC, F (C) (1F (C) )
 −1 
⇔ v = G θC, D (v) ◦ ηC ,

which shows that (3.26) also holds. 


Corollary 3.5.4 Let .F : C → D and .G : D → C be two functors such that .F  G
and consider .η and .ε to be the unit and respectively the counit of this adjunction.
Then .εop and .ηop are the unit and respectively the counit of the adjunction .Gop 
F op .
Proof Indeed, by Theorem 3.5.1 the unit and counit of the adjunction .F  G are
defined as follows for all .C ∈ Ob C and .D ∈ Ob D:
  −1  
ηC = θC, F (C) 1F (C) , εD = θG(D),
.
D 1G(D) ,

where .θ denotes the natural isomorphism induced by the adjunction. Similarly, if .η,
ε, θ denote the unit, the counit and respectively the natural isomorphism induced
.

by the adjunction .Gop  F op , we have

  (3.6)   op
op −1 op
ηD = θ D, Gop (D) 1Gop (D) = θG(D),
.
D 1G(D) = εD ,
  (3.7)   op
−1 op op
ε C = θ F op (C), C 1F op (C) = θC, F (C) 1F (C) = ηC ,

and the proof is now finished. 


Our next result provides a way of inducing natural transformations between two
pairs of adjoint functors.
Theorem 3.5.5 Let .Fi : C → D and .Gi : D → C be functors such that .Fi 
Gi and denote by .θ i the corresponding natural isomorphism for all .i = 1, 2.
Then, for any natural transformation .α : G1 → G2 there exists a unique natural
transformation .α : F2 → F1 such that the following diagram is commutative for all
180 3 Adjoint Functors

C ∈ Ob C and .D ∈ Ob D:
.

D ◦ HomD (α C , D)(f ) = HomC (C, αD ) ◦ θC, D (f ),

2 1
i.e., θC,
. (3.27)

for all .f ∈ HomD (F1 (C), D).

Proof For all .C ∈ Ob C, we denote by .ψ C : HomD (F1 (C), −) → HomD (F2
(C), −) the natural transformation defined for all .D ∈ Ob D as the following
composition:
 2 −1
.ψD = θC, D ◦ HomC (C, αD ) ◦ θC,
C 1
D. (3.28)

We will show first that each .ψ C is indeed a natural transformation between

.HomD (F1 (C), −) and .HomD (F2 (C), −). To this end, we will show that the

following diagram is commutative for all .u ∈ HomD (D, D  ):

i.e., u ◦ ψD
C
(v) = ψD
C
 (u ◦ v) for all v ∈ HomD (F1 (C), D).

Indeed, this can be written equivalently as follows:

 2 −1    2 −1  
u ◦ θC,
. D αD ◦ θC,1
D (v) = θC, D  αD  ◦ θC,
1
D  (u ◦ v)
  −1  
⇔ θC,
2
D  u ◦ θ 2
C, D αD ◦ θ 1
C, D (v) = αD  ◦ θC,
1
D  (u ◦ v)

(3.2)
 −1  
⇔ G2 (u) ◦ θC,
2
D
2
θC, D αD ◦ θC,
1
D (v) = αD  ◦ θC,
1
D  (u ◦ v)

⇔ G2 (u) ◦ αD ◦ θC,
1
D (v) = αD  ◦ θC, D  (u ◦ v)
1

(3.2)
⇔ G2 (u) ◦ αD ◦ θC,
1
D (v) = αD  ◦ G1 (u) ◦ θC, D (v),
1
3.5 The Unit and Counit of an Adjunction 181

and the last equality holds because .α : G1 → G2 is a natural transformation, i.e.,

we have .G2 (u) ◦ αD = αD  ◦ G1 (u) for all .u ∈ HomD (D, D  ).
This shows that, for each .C ∈ Ob C, ψ C is indeed a natural transfor-
mation and by Corollary 1.10.4, (1) there exists a unique morphism .α C ∈
HomD (F2 (C), F1 (C)) such that .ψD C (f ) = f ◦ α for all .f ∈ Hom (F (C), D).
C D 1
In light of (3.28), .α C ∈ HomD (F2 (C), F1 (C)) is the unique morphism such that

D (f ◦ α C ) = αD ◦ θC, D (f ), for all f ∈ HomD (F1 (C), D),

2 1
θC,
. (3.29)

which proves that diagram (3.27) is commutative.

Next we show that .α : F2 → F1 defined by (3.27) is in fact a natural transforma-
tion, i.e., the following diagram is commutative for all .g ∈ HomC (C, C  ):

2
Given the bijectivity of each .θC, F1 (C  ) it will suffice to prove that the following
holds:
   
F1 (C  ) α C  ◦ F2 (g) = θC, F1 (C  ) F1 (g) ◦ α C .
2 2
θC,
.

To this end, we have

  (3.29)  
F1 (C  ) F1 (g) ◦ α C = αF1 (C  ) ◦ θC,
2 1
.θC, F1 (C  ) F1 (g)

(3.17)
= αF1 (C  ) ◦ ηC 1
 ◦g

 
= αF1 (C  ) ◦ θC1  , F1 (C  ) 1F1 (C  ) ◦ g
(3.29)
= θC2  , F1 (C  ) (α C  ) ◦ g
(3.1)  
= θC,
2
F1 (C  ) α C  ◦ F2 (g) ,

where .η1 and .θ 1 denote the unit and respectively the natural isomorphism induced
by the adjunction .F1  G1 . 

Recall that if I is a small category then any functor .F : C → D induces a
functor between the corresponding functor categories .F : Fun (I, C) → Fun (I, D)
as in (1.36). Our next result shows that any adjunction can be lifted to an adjunction
between the corresponding induced functors.
182 3 Adjoint Functors

Proposition 3.5.6 Let .F : C → D, G : D → C be two functors such that .F  G

and I a small category. Then .F  G , where .F : Fun (I, C) → Fun (I, D) and
.G : Fun (I, D) → Fun (I, C) are the corresponding induced functors.

Proof Let .η : 1C → GF and .ε : F G → 1D denote the unit and respectively

the counit of the adjunction .F  G. Consider now the natural transformations
.η : 1Fun (I, C) → G F and .ε : F G → 1Fun (I, D) defined for all functors .H : I →

C, K : I → D by the whiskering of .η on the left by H and respectively the

whiskering of .ε on the left by K (see Example 1.7.2, (7)). More precisely, we have

ηH = ηH, ε K = εK.
.

In order to show that .η and .ε are indeed natural transformations, consider two
natural transformations .α : H → H  and .β : K → K  , where .H, H  : I → C
and K, .K  : I → D are functors. First, by the naturality of .η and .ε, the following
diagrams are commutative for all .i ∈ Ob I :

(3.30)
To summarize, for all .i ∈ Ob I we obtain

  (1.36) (3.30)
. G F (α) ◦ ηH i
= GF (αi ) ◦ ηH (i) = ηH  (i) ◦ αi = (ηH  ◦ α)i ,
  (1.36) (3.30)
ε K  ◦ F G (β) i = εK  (i) ◦ F G(βi ) = βi ◦ εK(i) = (β ◦ ε K )i ,

which shows that .η and .ε are natural transformations. The proof will be finished
once we show that .η and .ε fulfill (3.15) and (3.16). To start with, note that since .η
and .ε fulfill (3.15) and (3.16), in particular the following hold for all .i ∈ Ob I :
   
1F H (i) = εF H (i) ◦ F ηH (i) ,
. 1GK(i) = G εK(i) ◦ ηGK(i) .

The above identities come down to the following:

   
1F (H ) = ε F (H ) ◦ F ηH ,
. 1G (K) = G ε K ◦ ηG (K) ,
3.6 Another Characterisation of Adjoint Functors 183

which shows precisely that .η and .ε fulfill (3.15) and (3.16) and the proof is now
finished. 


3.6 Another Characterisation of Adjoint Functors

A very useful characterization of an adjunction involving only the (co)unit is the

following:
Theorem 3.6.1 Let .F : C → D and .G : D → C be two functors. The following are
equivalent:
(1) .F  G;
(2) there exists a natural transformation .η : 1C → GF such that for any morphism
.f ∈ HomC (C, G(D)) there exists a unique morphism .g ∈ HomD (F (C), D)

which makes the following diagram commutative:

. (3.31)

(3) there exists a natural transformation .ε : F G → 1D such that for any morphism
.f ∈ HomD (F (C), D) there exists a unique morphism .g ∈ HomC (C, G(D))

which makes the following diagram commutative:

. (3.32)

As the notation suggests, the natural transformations .η and .ε are precisely the unit
and the counit, respectively, of the adjunction .F  G.
Proof We start by proving the equivalence between (1) and (2). Suppose first that
.F  G and let .θ be the corresponding natural isomorphism. We define the natural
transformation .η : 1C → GF as in the proof of Theorem 3.5.1, namely by .ηC =
θC,F (C) (1F (C) ) for any .C ∈ Ob C. Let .f ∈ HomC (C, G(D)); we will prove that
−1
.g = θ
C,D (f ) ∈ HomD (F (C), D) is the unique morphism in .D which makes
184 3 Adjoint Functors

diagram (3.31) commutative. Indeed, setting .X = C, Y = F (C) and .Y  = D in

(3.2) gives the following commutative diagram for all .u ∈ HomC (F (C), D):

By applying the commutativity of the above diagram to .1F (C) we obtain

HomC (C, G(u)) ◦ θC, F (C) (1F (C) ) = θC, D ◦ HomD (F (C), u)(1F (C) ),
.

i.e., G(u) ◦ ηC = θC, D (u).

. (3.33)

Thus, we have

(3.33) −1
G(g) ◦ ηC = θC,D (g) = θC,D ◦ θC,D
. (f ) = f,

−1
which shows that .g = θC,D (f ) ∈ HomD (F (C), D) makes diagram (3.31) com-
mutative. Assume now that there exists another morphism .g  ∈ HomD (F (C), D)
such that .G(g  ) ◦ ηC = f and let .f  = θC,D (g  ). Following the same steps as in
the argument above it can be easily seen that .G(g  ) ◦ ηC = θC,D (g  ) = f  . Our
assumption now implies that .f = f  and therefore, since .θC, D is a bijection, we
obtain .g = g  .
Assume now that .2) holds, i.e., for any .f ∈ HomC (C, G(D)) there exists a
unique morphism .g ∈ HomD (F (C), D) such that (3.31) is fulfilled. Given .C ∈
Ob C and .D ∈ Ob D we define the following map:

θC, D : HomD (F (C), D) → HomC (C, G(D)),

. θC, D (u) = G(u) ◦ ηC
(3.34)

for any .u ∈ HomD (F (C), D). Obviously, our assumption implies that .θC, D is a
set bijection for all .C ∈ Ob C and .D ∈ Ob D. The fact that .θ defined in (3.34) is
natural in both variables follows exactly as in the proof of Theorem 3.5.1.
Finally, we are left to show the equivalence between (1) and (3). Indeed, by
Theorem 3.4.3, .F  G if and only if .Gop  F op . By applying the equivalence
between (1) and (2) we obtain that .Gop  F op if and only if there exists a
natural transformation .η : 1Dop → F op Gop with the property that for any .f op ∈
HomDop (D, F op (C)) there exists a unique .g op ∈ HomCop (Gop (D), C) such that
3.6 Another Characterisation of Adjoint Functors 185

the following holds:

F op (g op ) ◦op ηD = f op .
. (3.35)

By Proposition 1.8.3, there exists a natural transformation .ε : F G → 1D such that

η = εop . Now it can be easily seen that (3.35) comes down to .εD ◦ F (g) = f .
.

Putting everything together, we obtain that .F  G if and only if there exists a

natural transformation .ε : F G → 1D such that for any .f ∈ HomD (D, F (C))
there exists a unique .g ∈ HomC (G(D), C) such that .εD ◦ F (g) = f . The proof is
now complete. 

As a straightforward consequence of Theorem 3.6.1 we have:
Corollary 3.6.2 Suppose .F : C → D and .G : D → C are two functors such that
F  G and let .η : 1C → GF , .ε : F G → 1D be the unit, respectively the counit of
.

the adjunction.
(1) If .g, g  ∈ HomD (F (C), D) such that .G(g) ◦ ηC = G(g  ) ◦ ηC then .g = g  .
(2) If .h, h ∈ HomC (C, G(D)) such that .εD ◦ F (h) = εD ◦ F (h ) then .h = h .
Proof
(1) Follows trivially from Theorem 3.6.1, (2) by considering .f = G(g  ) ◦ ηC . Then
both morphisms g and .g  make diagram (3.31) commutative, which implies .g =
g  . The second part follows in a similar manner by using Theorem 3.6.1, (3).


Examples 3.6.3
(1) The forgetful functor .U : Top → Set has both a left and a right adjoint. We
start by constructing the left adjoint functor .F : Set → Top which endows
each .X ∈ Ob Set with the discrete topology. We define a natural transformation
.η : 1
Set → U F by .ηX (x) = x for any .X ∈ Ob Set and .x ∈ X. Consider now
.f ∈ Hom
Set (X, U (Y )), where .Y ∈ Ob Top. According to Theorem 3.6.1, (2)
in order to prove that .F  U we need to find a unique morphism .g ∈
HomTop (F (X), Y ) such that the following diagram commutes:

To this end, it is enough to consider .g = f . Note that f is obviously continuous

since .F (X) is endowed with the discrete topology.
On the other hand, the right adjoint .G : Set → Top endows each .X ∈ Ob Set
with the indiscrete topology. We define a natural transformation .η : 1Top →
186 3 Adjoint Functors

−1
GU by .ηX (x) = x for all .X ∈ Ob Top and .x ∈ X. Now since .ηX (∅) =
−1
∅ and .ηX (G(X)) = G(X) = X we obtain that each .ηX is continuous.
Consider now .f ∈ HomTop (X, G(Y )), where .Y ∈ Ob Set. We aim to find
a unique morphism .g ∈ HomSet (U (X), Y ) such that the following diagram is
commutative:

As before, we set .g = f .
(2) The Stone–Čech compactification functor .S : Top → KHaus defined in Exam-
ple 1.5.3, (24) is left adjoint to the inclusion functor .I : KHaus → Top. Indeed,
let .i : 1Top → I S be the natural transformation defined for all topological
spaces X by the continuous map .iX : X → S(X) associated with the Stone–
Čech compactification of X. If .f ∈ HomTop (X, Y ) then .S(f ) is defined
by (1.7) as the unique morphism in .KHaus such that .I S(f ) ◦ iX = iY ◦ f .
Therefore, the following diagram is commutative for all .f ∈ HomTop (X, Y ):

This shows that i is indeed a natural transformation. Now recall that by

the universal property of the Stone–Čech compactification, for any .f ∈
HomTop (X, I (Z)), where .Z ∈ Ob KHaus, there exists a unique .g ∈
HomKHaus (S(X), Z) such that the following diagram is commutative:

Using Theorem 3.6.1, (2) we can conclude that .S is left adjoint to I , as desired.
(3) The Grothendieck group functor .G : Mon → Grp defined in Exam-
ple 1.5.3, (25) is left adjoint to the inclusion functor .I : Grp → Mon. Indeed,
let .i : 1Mon → I G be the natural transformation defined for all monoids M
3.6 Another Characterisation of Adjoint Functors 187

by the homomorphism of monoids .iM : M → I G(M) associated with the

Grothendieck group of M. Indeed, if .f ∈ HomMon (M, N ) then .G(f ) is
defined by (1.8) as the unique morphism in .Grp such that .I G(f ) ◦ iM = iN ◦ f .
Therefore, the following diagram is commutative for all .f ∈ HomMon (M, N ):

In particular, this shows that i is indeed a natural transformation. Now

using the universal property of the Grothendieck group, for any .f ∈
HomMon (M, I (N )), where .N ∈ Ob Grp, there exists a unique .g ∈
HomGrp (G(M), N ) such that the following diagram is commutative:

Using Theorem 3.6.1, (2) we can conclude that G is left adjoint to I , as desired.
(4) The functor .U : Mon → Grp defined in Example 1.5.3, (26) which assigns to
each monoid its group of invertible elements, is right adjoint to the inclusion
functor .I : Grp → Mon. Indeed, let .ε : I U → 1Mon be the natural
transformation defined for all .M ∈ Ob Mon by .εM : U (M) → M, εM = iM ,
where .iM denotes the inclusion map. If .f ∈ HomMon (M, N ), then for all
.m ∈ U (M) we have

   
.(f ◦ iM )(m) = f (m) = iN ◦ f|U (M) (m) = iN ◦ I U(f ) (m).

Therefore, the following diagram is commutative for all .f ∈ HomMon (M, N ):

which shows that .ε is a natural transformation. Now, if .f ∈ HomMon (I (G),

M), where .G ∈ Ob Grp, then .f (G) is also a group and therefore .f (G) ⊆
U (M). Hence .f ∈ HomGrp (G, U (M)) and f is the unique morphism which
makes the following diagram commutative:
188 3 Adjoint Functors

Using Theorem 3.6.1, (3) we can conclude that .U is right adjoint to I , as

desired.
(5) Let R be a commutative ring with unity and .(S −1 R, j ) its localization with
respect to the multiplicative set .S ⊂ R, where .j : R → S −1 R is the ring
homomorphism defined by .j (r) = 1r for all .r ∈ R. Then the localization
functor .L : R M → S −1 R M defined in Example 1.5.3, (29) is left adjoint
to the restriction of scalars functor .Fj : S −1 R M →R M induced by the ring
homomorphism j (see Example 1.5.3, (32)).
Throughout, if M is an R-module we denote by .(S −1 M, ϕM ) its localization
module at S. Consider the natural transformation .ϕ : 1R M → Fj L defined
for all R-modules M by the R-module homomorphism .ϕM : M → S −1 M
associated with the localization module .S −1 M. If .f ∈ HomR M (M, N ) then
.L(f ) is defined by (1.9) to be the unique morphism in .S −1 R M such that

.ϕN ◦ f = Fj L(f ) ◦ ϕM . Therefore, the following diagram is commutative

for all .f ∈ HomR M (M, N ):

In particular, this shows that .ϕ is indeed a natural transformation. Now recall

that by the universal property of the localization module ([3, Theorem 12.3]),
for any .f ∈ HomR M (M, Fj (N )), with .N ∈ Ob S −1 R M, there exists a unique
.g ∈ Hom (L(M), N ) such that the following diagram is commutative:
S −1 M
R

Using Theorem 3.6.1, (2) we can now conclude that .L is left adjoint to .Fj , as
desired.
3.6 Another Characterisation of Adjoint Functors 189

(6) The Hausdorff quotient functor .H : Top → Haus defined in Exam-

ple 1.5.3, (30) is left adjoint to the inclusion functor .I : Haus → Top. Indeed,
let .q : 1Top → I H be the natural transformation defined for all topological
spaces X by the continuous map .qX : X → H (X) associated with the Hausdorff
quotient of X. If .f ∈ HomTop (X, Y ) then .H(f ) is defined by (1.10) as the
unique morphism in Haus such that .qY ◦ f = I H(f ) ◦ qX . Therefore, the
following diagram is commutative for all .f ∈ HomTop (X, Y ):

This shows that q is indeed a natural transformation. Now recall that by the
universal property of the Hausdorff quotient, for any .f ∈ HomTop (X, I (Z)),
where .Z ∈ Ob Haus, there exists a unique .g ∈ HomHaus (H (X), Z) such that
the following diagram is commutative:

Using Theorem 3.6.1, (2) we can conclude that .H is left adjoint to I , as desired.
(7) The Dorroh extension functor .D : Rng → Ring defined in Example 1.5.3, (31)
is left adjoint to the inclusion functor .I : Ring → Rng. Indeed, let .j : 1Rng →
I D be the natural transformation defined for all rings R by the ring homo-
morphism .jR : R → D(R) associated with the Dorroh extension of R. If
.f ∈ Hom
Rng (R, S) then .D(f ) is defined by (1.11) as the unique morphism
in Ring such that .jS ◦ f = I D(f ) ◦ jR . Therefore, the following diagram is
commutative for all .f ∈ HomRng (R, S):

In particular, this shows that j is a natural transformation. Now recall that by the
universal property of the Dorroh extension, for any .f ∈ HomRng (R, I (T )),
190 3 Adjoint Functors

where .T ∈ Ob Ring, there exists a unique .g ∈ HomRing (D(R), T ) such that

the following diagram is commutative:

Using Theorem 3.6.1, (2) we can conclude that .D is left adjoint to I , as

desired. .

As another application of Theorem 3.6.1 we will show that, when they exist,
left/right adjoints are unique up to natural isomorphism.
Theorem 3.6.4 Any two left (right) adjoints of a given functor are naturally
isomorphic.
Proof Assume .F, F  : C → D are both left adjoint functors of .G : D → C. Then
there exist natural transformations .η : 1C → GF and .η : 1C → GF  satisfying
the conditions in Theorem 3.6.1, (2). Given .C ∈ Ob C, as .F   G and .ηC ∈
HomC (C, GF (C)), there exists a unique morphism .γC ∈ HomD (F  (C), F (C))
such that

G(γC ) ◦ ηC
. = ηC . (3.36)

 ∈ Hom (C, GF  (C)), there exists a unique morphism

Similarly, as .F  G and .ηC C
 
.γ ∈ HomD (F (C), F (C)) such that
C

G(γC ) ◦ ηC = ηC
.

. (3.37)

We will see that each .γC is an isomorphism with the inverse given precisely by
.γC . Indeed, using (3.36) and (3.37) we can easily see that .G(γC ◦γC )◦ηC  = η and
C
 
since we obviously also have .G(1F (C) ) ◦ ηC = ηC it follows by Corollary 3.6.2, (1)


that .γC ◦ γC = 1F  (C) . Similarly, one can prove that .γC ◦ γC = 1F (C) .
We are left to prove that .γ : F  → F is a natural transformation, i.e., for any

.f ∈ HomC (C, C ) the following diagram is commutative:

.
3.6 Another Characterisation of Adjoint Functors 191

Using Corollary 3.6.2, (1) it is enough to prove that the following holds:

G(F (f ) ◦ γC ) ◦ ηC
. = G(γC  ◦ F  (f )) ◦ ηC

. (3.38)

To this end, we use the naturality of .η and respectively .η ; that is, the commutativity
of the following diagrams:

. (3.39)

. (3.40)

Then, we have

 (3.36)
GF (f ) ◦ G(γC ) ◦ ηC
. = GF (f ) ◦ ηC
(3.39)
= ηC  ◦ f
(3.36) 
= G(γC  ) ◦ ηC  ◦f

(3.40)
= G(γC  ) ◦ GF  (f ) ◦ ηC

.

Therefore, (3.38) indeed holds. To summarize, we have proved that there exists a
natural isomorphism .γ : F  → F and the proof is now finished. 

Adjunctions can also be used to easily derive important properties of certain func-
torial constructions, as the following examples show. This includes, for instance, the
commutation of tensor products or localizations with direct sums of modules. All of
these are obtained by applying Theorem 3.4.4.
Example 3.6.5 Given a commutative ring R, for any .X ∈ ObR M and any family
(Mi )i∈I of R-modules we have the following isomorphisms of R-modules:
.

   
. ⊕i∈I Mi ⊗ X  ⊕i∈I Mi ⊗ X ,
 
S −1 ⊕i∈I Mi  ⊕i∈I S −1 Mi ,
192 3 Adjoint Functors

where .⊗ = ⊗R . Indeed, both statements are consequences of the fact that both
the tensor product functor .− ⊗ X : R M → R M and the localization functor
.L : R M →S −1 R M are left adjoints (see Example 3.4.1, (2) and Example 3.6.3, (5))

and therefore they preserve coproducts (see Example 2.1.5, (12)). .

We end this section with the following useful result:

Lemma 3.6.6 Let .F : C → D, G : D → C be two functors such that .F  G
and let .η : 1C → GF and .ε : F G → 1D be the unit and counit of the adjunction,
respectively.
(1) F is faithful if and only if .ηC is a monomorphism for all .C ∈ Ob C; G is faithful
if and only if .εD is an epimorphism for all .D ∈ Ob D;
(2) F is full if and only if .ηC is a split epimorphism for all .C ∈ Ob C; G is full if
and only if .εD is a split monomorphism for all .D ∈ Ob D;
(3) F is fully faithful if and only if the unit of the adjunction is a natural
isomorphism; G is fully faithful if and only if the counit of the adjunction is
a natural isomorphism.
Proof
(1) Assume first that F is faithful and let .f1 , f2 ∈ HomC (C  , C) such that .ηC ◦
f1 = ηC ◦ f2 . Applying F to the last identity yields .F (ηC ) ◦ F (f1 ) = F (ηC ) ◦
F (f2 ) and after composing on the left with .εF (C) and using (3.15) we obtain
.F (f1 ) = F (f2 ). As F is faithful we arrive at .f1 = f2 , which shows that .ηC is

indeed a monomorphism.
Conversely, assume .ηC is a monomorphism for all .C ∈ Ob C and let

.f1 , f2 ∈ HomC (C , C) such that .F (f1 ) = F (f2 ). Using (3.15) we obtain

.εF (C) ◦ F (ηC ◦ f1 ) = εF (C) ◦ F (ηC ◦ f2 ). Now Corollary 3.6.2, (2) implies

.ηC ◦ f1 = ηC ◦ f2 and since .ηC is a monomorphism we obtain .f1 = f2 .

Therefore, F is faithful.
The result concerning the functor G follows by duality. Indeed, by The-
orem 3.4.3 we have .Gop  F op and moreover, Corollary 3.5.4 shows that
the unit of this adjunction is precisely .εop . Since G is faithful if and only if
.G
op is faithful, the desired conclusion follows the first part of the proof and

Proposition 1.4.3, (1).

(2) Let F be a full functor and .C ∈ Ob C. Then .εF (C) ∈ HomD (F GF (C),
F (C)) and since F is full there exists a morphism .uC ∈ HomC (GF (C), C)
such that .εF (C) = F (uC ). We obtain

(3.15)
εF (C) ◦ F (ηC ◦ uC ) = εF (C) ◦ F (ηC ) ◦ F (uC ) = F (uC )
.

 
= εF (C) = εF (C) ◦ F 1GF (C)

Now Corollary 3.6.2, (2) implies .ηC ◦ uC = 1GF (C) , as desired.

Conversely, assume .ηC is a split epimorphism for all .C ∈ Ob C. Thus, there
exists a morphism .vC ∈ HomC (GF (C), C) such that .ηC ◦ vC = 1GF (C) .
3.6 Another Characterisation of Adjoint Functors 193

Let .g ∈ HomD (F (C), F (C  )). The naturality of .η applied to .vC renders the
following diagram commutative:

(3.41)
Therefore, for all .C ∈ Ob C, we obtain

(3.41) (3.16)
GF (vC ) ◦ ηGF (C) = ηC ◦ vC = 1GF (C) = G(εF (C) ) ◦ ηGF (C) .
.

Now Corollary 3.6.2, (1) implies .F (vC ) = εF (C) . Finally, the naturality of .ε
applied to g yields

(3.42)
Putting all this together we obtain

(3.15)
g = g ◦ 1F (C) = g ◦ εF (C) ◦ F (ηC )
.

(3.42)  
= εF (C  ) ◦ F G(g) ◦ F (ηC ) = F vC  ◦ G(g) ◦ ηC ,

which shows that F is full.

The result concerning the functor G follows again by duality. Indeed, by
Theorem 3.4.3 we have .Gop  F op and, moreover, Corollary 3.5.4 shows that
the unit of this adjunction is precisely .εop . Since G is full if and only if .Gop is
full, the desired conclusion follows from Exercise 1.6.
(3) Using (1) and (2), F is fully faithful if and only if .ηC is both a monomorphism
and a split epimorphism for all .C ∈ Ob C. Similarly, G is fully faithful if and
only if .εD is both an epimorphism and a split monomorphism and for all .D ∈
Ob D. The conclusion now follows from Exercise 1.6, (b).


194 3 Adjoint Functors

3.7 (Co)reflective Subcategories

This section is devoted to a special kind of adjunction, namely those for which one
of the functors involved is an inclusion.
Definition 3.7.1 A full subcategory .A of .B is called reflective if the inclusion
functor .I : A → B admits a left adjoint, called a reflector. Dually, a full subcategory
.A of .B is called coreflective if the inclusion functor admits a right adjoint, called a

coreflector.
We have already encountered many examples of such subcategories:
Examples 3.7.2
(1) Ab is a reflective subcategory of Grp, as shown in Example 3.4.1, (4).
(2) Grp is both a reflective and a coreflective subcategory of Mon as shown in
Example 3.6.3, (3) and (4).
(3) KHaus is a reflective subcategory of Top. The Stone–Čech compactification
provides the reflector, as shown in Example 3.6.3, (2).
(4) Haus is a reflective subcategory of Top. The Hausdorff quotient provides the
reflector, as shown in Example 3.6.3, (6).
(5) Ring is not a reflective subcategory of Rng. Indeed, note that although the inclu-
sion functor .I : Ring → Rng has a left adjoint, as shown in Example 3.6.3, (7)
the category Ring is not a full subcategory of Rng. .

One important feature of (co)reflective subcategories is that they behave well

with respect to (co)limits. The remaining of this section will be devoted to studying
their (co)completeness and providing explicit descriptions of the (co)limits.
Proposition 3.7.3 Let .A be a subcategory of .B.
(1) If .A is a reflective subcategory of the complete category .B then .A is itself
complete.
(2) If .A is a coreflective subcategory of the cocomplete category .B then .A is itself
cocomplete.
Proof
(1) Let .R : B → A be the reflector of the inclusion functor .I : A → B, i.e., .R  I .
Let J be a small category and .F : J → A a functor. Since .B is complete,
the functor .I F : J → B has a limit, which we denote  by . L, (pj : L →
I F (j ))j ∈Ob J . Since . L, (pj : L → I F (j ))j ∈Ob J is in particular a cone on
3.7 (Co)reflective Subcategories 195

I F , the following diagram is commutative for all .d ∈ HomJ (j, l):

. (3.43)

Let .η : 1B → I R be the unit of the adjunction .R  I (see Theorem 3.5.1).

By Theorem 3.6.1, (2) for any .pj ∈ HomB (L, I F (j )) there exists a
unique morphism .qj ∈ HomA (R(L), F (j )) such that the following diagram
commutes for all .j ∈ Ob J :

. (3.44)
 
We will prove first that . R(L), (qj : R(L) → F (j ))j ∈Ob J is a cone on F .
Indeed, for all .d ∈ HomJ (j, l) we have
    (3.44) (3.43) (3.44)  
I F (d)◦qj ◦ηL =I F (d)◦I qj ◦ ηL = I F (d) ◦ pj = pl = I ql ◦ ηL .
.

Now by Corollary 3.6.2, (1) we get .F (d) ◦ qj = ql , i.e.,

 
.R(L), (qj : R(L) → F (j ))j ∈Ob J
 
is a cone on F . We will show that . R(L), (qj )j ∈Ob J is in fact the limit of F .
 We start by showing that .ηL : L → I R(L)  is an isomorphism. Indeed, since
. I R(L), (I (qj ) : I R(L) → I F (j ))j ∈Ob J is a cone on I F by Lemma 2.5.5
 
and . L, (pj : L → I F (j ))j ∈Ob J is its limit, there exists a unique morphism
.f ∈ Hom (I R(L), L) such that the following diagram is commutative for all
B
.j ∈ Ob J :

. (3.45)
196 3 Adjoint Functors

(3.45) (3.44)
Thus, for any .j ∈ Ob J we have .pj ◦ f ◦ηL = I (qj )◦ηL = pj = pj ◦1L
and using Proposition 2.2.14, (1) we obtain

f ◦ ηL = 1L .
. (3.46)

On the other hand .ηL ◦ f ∈ HomB (I R(L), I R(L)) and since .I : A → B is

fully faithful, there exists a unique morphism .t ∈ HomA (R(L), R(L)) such
that

ηL ◦ f = I (t).
. (3.47)

Moreover, we have

(3.47) (3.46)
I (t) ◦ ηL = ηL ◦ f ◦ ηL = ηL = I (1R(L) ) ◦ ηL .
.

Using again Corollary 3.6.2, (1) we get .t = 1R(L) and hence .ηL ◦ f = 1I R(L) ,
so .ηL is an isomorphism, as desired.  
 Consider now another
 cone . L , (tj : L → F (j ))j ∈Ob J on F . Then

. I (L ), (I (tj )j ∈Ob J is a cone on I F . Therefore, there exists a unique mor-
phism .g ∈ HomB (I (L ), L) such that the following diagram is commutative
for all .j ∈ Ob J :

. (3.48)

Since we also have .pj = I (qj ) ◦ ηL for all .j ∈ Ob J , we obtain

I (qj ) ◦ ηL ◦ g = I (tj ).
. (3.49)

As .ηL ◦ g ∈ HomB (I (L ), I R(L)) and I is fully faithful, there exists a unique
morphism .h ∈ HomA (L , R(L)) such that .I (h) = ηL ◦g. Then (3.49) becomes
.I (qj ◦h) = I (tj ) and since I is fully faithful we get .qj ◦h = tj for all .j ∈ Ob J ,
3.7 (Co)reflective Subcategories 197

i.e., the following diagram is commutative:

The proof will be finished once we show that h is the unique morphism which
makes the above diagram commutative. Indeed, suppose there exists an .h ∈
HomA (L , R(L)) such that .qj ◦ h = tj for all .j ∈ Ob J . Then we also have
.I (qj ) ◦ I (h) = I (tj ) and using (3.44) and respectively (3.48) we get

−1
pj ◦ η L
. ◦ I (h) = pj ◦ g

−1
for all .j ∈ Ob J . Proposition 2.2.14, (1) implies .ηL ◦ I (h) = g and thus
.I (h) = ηL ◦ g. Since h is the unique morphism such that .I (h) = ηL ◦ g, we
 
get .h = h. This shows that . R(L), (qj )j ∈Ob J is indeed the limit of F and
therefore .A is a complete category.
(2) Let .A be a coreflective subcategory of a cocomplete category .B and denote by
.C : B → A the right adjoint of the inclusion functor .I : A → B and by .ε the

counit of this adjunction. Let J be a small category and .F : J → A a functor.

Then .Aop is obviously a subcategory of .Bop and the inclusion functor is
precisely .I op . Theorem 3.4.3 implies that .C op  I op and, moreover, as proved
in Corollary 3.5.4, .εop is the unit of this adjunction. Using the first part of the
proof, the limit of the functor .F op : J op → Aop is given by the pair
 
op
.C(L), (qj ∈ HomAop (C(L), F (j op )))i∈Ob J ,

op
where .qj is the unique morphism which makes the following diagram commu-
tative:

 
op
and . L, (pj ∈ HomBop (L, I op F op (j )))i∈Ob J is the limit of the functor
.I
op F op : J op → Bop . In other words, for all .j ∈ Ob J, q is the unique
j
morphism in .A such that .εL ◦ I (qj ) = pj . Now we can conclude by
198 3 Adjoint Functors

 
Lemma 2.2.12 that . C(L), (qj ∈ HomA (F (j ), C(L)))i∈Ob J is the colimit
of the functor F and therefore .A is cocomplete.


In light of Proposition 3.7.3 the next natural question we are led to consider con-
cerns the cocompleteness of reflective subcategories (and, dually, the completeness
of coreflective subcategories).
Proposition 3.7.4 Let .A be a subcategory of .B.
(1) If .A is a reflective subcategory of a cocomplete category .B then .A is also
cocomplete.
(2) If .A is a coreflective subcategory of a complete category .B then .A is also
complete.
Proof
(1) Let .I : A → B be the inclusion functor and .R : B → A the reflector. Let
.

.F : J → A be a functor, where J is a small category. Since .B is cocomplete,

the functor .I F : J → B has a colimit, which we denote by

 
. D, (qj : I F (j ) → D)j ∈Ob J .

R is left adjoint to I and by Theorem 3.4.4 it preserves colimits, so

 
.R(D), (R(qj ) : RI F (j ) → R(D))j ∈Ob J

is the colimit of the functor .RI F : J → A. By Lemma 3.6.6, (3) we know that
the counit .ε : RI → 1A of the adjunction .R  I is a natural isomorphism.
Therefore, the natural transformation .εF : RI F → F defined by

(εF )j = εF (j ) for all j ∈ Ob J

.

is also a natural isomorphism (Example 1.7.2, (7)). Now, in light of

Lemma 2.2.15, (2) we can conclude that . R(D), (R(qj ) ◦ εF−1(j ) : F (j ) →

R(D))j ∈Ob J is the colimit of F .
.(2) Let .I : A → B to be the inclusion functor and denote by .C : B → A the

coreflector and by .η the unit of this adjunction, which is a natural isomorphism

by Lemma 3.6.6, (3). Let .F : J → A be a functor, where J is a small category.
By Theorem 3.4.3 we also have an adjunction .C op  I op whose counit is
precisely .ηop . This shows, in particular, that .Aop is a reflective subcategory
of the cocomplete category .Bop with inclusion functor .I op . By part .1) proved
above, the colimit of the functor .F op is given by the pair
 
C op (D), (C op (qj ) ◦op (ηF−1op (j ) )op ∈ HomAop (F op (j ), C op (D)))j ∈Ob J ,
op
.
3.8 Equivalence of Categories 199

 op 
where . D, (qj )j ∈Ob J is the colimit of the functor .I op F op : J op → Bop . Now
 
Lemma 2.2.12 implies that . C(D), (ηF−1(j ) ◦ C(qj ))j ∈Ob J is the limit of the
functor F .



3.8 Equivalence of Categories

When studying categories which are practically the same, the first notion we usually
encounter is that of an isomorphism of categories, as introduced in Definition 1.6.6.
However, this concept turns out to be too strict, as there are many examples
of categories with similar properties (such as completeness, cocompleteness etc.)
which are not isomorphic. To express that two categories share many of the same
properties, a more suitable notion than isomorphism is the following:
Definition 3.8.1 A functor .F : C → D is called an equivalence of categories
and the category .C is said to be equivalent to .D if there exists another functor
.G : D → C such that we have natural isomorphisms .GF ∼ = 1C and .F G ∼ = 1D .
A contravariant functor .F : C → D for which .F : Cop → D is an equivalence of
categories is called a duality of categories.
Example 3.8.2 Given a field K, the category .MatK defined in Example 1.2.2, (19)
is equivalent to the category of finite-dimensional K-vector spaces .K Mf d . Indeed,
the functor .F : MatK → K Mf d defined below is an equivalence of categories:

F (n) = K n , the n-dimensional space of column vectors over K for all n ∈ N,

.

F (A) = MA , for all morphisms A : m → n in MatK , where MA : K m → K n is

given by MA (v) = Av for all v ∈ K m .

To this end, we choose a basis6 .BV for each finite dimensional vector space V
and we define a functor .G : K Mf d → MatK as follows:

G(V ) = dim(V ), for all finite-dimensional vector spaces V ,

.

G(α) = Uα , where Uα is the matrix of the linear map α : V → W with respect to

the chosen bases BV and BW of V and W respectively.

Throughout this example, by convention, the basis we consider on .K n will be

the standard basis. We start by showing that .GF = 1MatK . First, for any .n ∈ N we

6 Note that this is always possible due to the axiom of choice.

200 3 Adjoint Functors

have
   
GF (n) = G K n = dim K n = n = 1MatK (n).
.

Moreover, if .A : m → n is a morphism in .MatK , we have .GF (A) = G(MA ) =

UMA , where .UMA is the matrix of the linear map .MA : K m → K n given
by .MA (v) = Av with respect to the standard bases .{e1 , e2 , . . . , em } and
.{f1 , f2 , . . . , fn } of .K
m and .K n , respectively. Having in mind that the element

.ei (resp. .fj ) of the standard basis is the column vector in .K

m (resp. in .K n ) with

1 on the i-th (resp. j -th) position and zeros elsewhere for all n.i = 1, 2, . . . , m
(resp. .j = 1, 2, . . . , n), we obtain .MA (ei ) = Aei = j =1 aj i fj , where
 
.A = akl . This proves that .UMA = A, i.e., .GF (A) = A, as desired.
k=1, n, l=1, m
Hence, we have proved that .GF = 1MatK , which shows that, in particular, GF is
naturally isomorphic to .1MatK .
We are left to show that F G is naturally isomorphic to .1 f d . Consider
KM
.η : 1 f d → F G defined for any vector space V by .ηV : V → K
dim(V ) , η (v) =
KM
V
[v], where we denote by .[v] the (column) coordinate vector of v with respect to the
chosen basis of V . We claim that .η is a natural isomorphism. To start with, each
.ηV is clearly a linear bijection. We are left to check the naturality condition. To

this end, let .α : V → W be a morphism in .K Mf d and consider the chosen bases

.BV = {t1 , t2 , . . . , tm } and .BW = {w1 , w2 , . . . , wn } in V and W , respectively,

where .m = dim(V ), n = dim(W ). The proof will be finished once we show that
the following diagram is commutative:

. (3.50)
  
If we define .Uα = ukl k=1, n, l=1, m , then for any .v = m vi ti ∈ V we have
m m  n  ni=1  m 
.α(v) = v
i=1 i α(t i ) = v
i=1 i u w
j =1 j i j = j =1 vi uj i wj .

i=1
This shows that the j -th component of the column vector .[α(v)] is . m i=1 vi uj i ,
for all .j = 1, 2, . . . , n. Moreover, a similar straightforward computation shows
that the j -th component of .Uα [v] is . m i=1 uj i vi for all .j = 1, 2, . . . , n. Putting
everything together we have

[α(v)] = Uα [v].
. (3.51)

Therefore, for any .v ∈ V , we have

    (3.51)
.F G(α) ◦ ηV (v) = F G(α) [v] = MUα [v] = Uα [v] = α(v) = ηW ◦ α(v),
3.8 Equivalence of Categories 201

which proves the commutativity of diagram (3.50). 

.

Remark 3.8.3 The categories .MatK and .K M from the previous example are
fd

equivalent but not isomorphic. Indeed, this follows easily by noticing that .MatK is
a small category while .K Mf d has a class of objects.
Proposition 3.8.4 Let .A, B and .C be three categories. The following hold:
(1) any category is equivalent to itself;
(2) if .A is equivalent to .B then .B is equivalent to .A;
(3) if .A is equivalent to .B and .B is equivalent to .C then .A is equivalent to .C.
Proof
(1) Any category .A is equivalent to itself as the identity functor .1A : A → A is
obviously an equivalence of categories.
(2) Assume that the category .A is equivalent to .B and .F : A → B is the
equivalence functor. Then there exists another functor .G : B → A and two
natural isomorphisms .GF ∼ = 1A and .F G ∼ = 1B . This shows that G is also an
equivalence of categories and therefore .B is equivalent to .A.
(3) Assume that .A is equivalent to .B and .B is equivalent to .C. Then, we have two
pairs of functors and their corresponding natural isomorphisms

F : A → B, G : B → A,
. α : F G → 1B , β : GF → 1A ,
H : B → C, T : C → B, γ : H T → 1C , σ : T H → 1B .

Consider now the functors .H F : A → C and .GT : C → A and the following

natural transformations obtained by whiskering .α and .σ as in Example 1.7.2, (7)
both on the left and on the right:

.
202 3 Adjoint Functors

Note that the above natural transformations are in fact natural isomorphisms
since .αT and .σF are natural isomorphisms (Example 1.7.2, (7)) and all functors
preserve isomorphisms (Proposition 1.6.9, (1)). This gives rise to the following
natural isomorphisms:

and we can now conclude that .A is equivalent to .C.



We have the following very useful characterization of equivalences of categories:
Theorem 3.8.5 Let .F : C → D be a functor. The following are equivalent:
(1) F is an equivalence of categories;
(2) F is fully faithful and essentially surjective;
(3) there exists a right adjoint .G : D → C of F such that the unit and counit of the
adjunction are natural isomorphisms;
(4) there exists a left adjoint .G : D → C of F such that the unit and counit of the
adjunction are natural isomorphisms.
In particular, as the notation suggests, an equivalence of categories has a left adjoint
which is also a right adjoint and the unit and counit of these adjunctions are natural
isomorphisms.
Proof
(1) ⇒ (2) Since F is an equivalence of categories, there exists a functor .G : D →
.

C and two natural isomorphisms .η : 1C → GF and .ε : F G → 1D . To

start with, for any .D ∈ Ob D the morphism .εD ∈ HomD (F G(D), D)
is an isomorphism and therefore F is essentially surjective. Next we
prove that F is fully faithful. Let .h1 , h2 ∈ HomC (C, C  ) such that
.F (h1 ) = F (h2 ). Then we also have .GF (h1 ) = GF (h2 ). Moreover, the
3.8 Equivalence of Categories 203

naturality of .η renders the following diagrams commutative:

(3.52)

(3.53)
From (3.52) and (3.53) we obtain .ηC  ◦ h1 = ηC  ◦ h2 and since .ηC  is an
isomorphism we get .h1 = h2 , as desired. Similarly, using the naturality
of .ε it follows that G is faithful as well.
Consider now .C, C  ∈ Ob C and .g ∈ HomD (F (C), F (C  )). Now
define
−1 
f = ηC
.  ◦ G(g) ◦ ηC ∈ HomC (C, C ). (3.54)

We will prove that .F (f ) = g. Indeed, using the naturality of .η applied

to f , we obtain

−1 (3.52)−1 (3.54)
ηC
.  ◦ GF (f ) ◦ ηC = f = ηC  ◦ G(g) ◦ ηC .

Since .ηC and .ηC  are isomorphisms, we get .GF (f ) = G(g). As G is

faithful, the above equality comes down to .F (f ) = g, which proves that
F is full as well.
.(2) ⇒ (1) Assume now that F is fully faithful and essentially surjective. First

note that since F is fully faithful it reflects isomorphisms (see Propo-

sition 1.6.9, (2)) and therefore two objects C and .C  are isomorphic
in .C if and only if .F (C) and .F (C  ) are isomorphic in .D. Therefore,
as F is essentially surjective, for any .D ∈ Ob D there exists a unique
(up to isomorphism) .C ∈ Ob C such that .F (C)  D. Thus, for any
.D ∈ Ob D we can choose an object .G(D) ∈ Ob C and an isomorphism
7

.εD : F G(D) → D in .D. Now if .g ∈ HomD (D, D ) we have the



7 We assume that the axiom of choice holds.

204 3 Adjoint Functors

following morphism in .D:

−1 
εD
.  ◦ g ◦ εD : F G(D) → F G(D ).

Since F is fully faithful, there exists a unique morphism .G(g) ∈

−1
HomC (G(D), G(D  )) such that .F G(g) = εD  ◦g◦εD . The last equality
implies the commutativity of the following diagram:

. (3.55)

We will prove that G defined above is in fact a functor and .ε : F G →

1D is a natural transformation. Indeed, setting .D = D  and .g = 1D
yields
−1
εD
. ◦ 1D ◦ εD = 1F G(D) : F G(D) → F G(D)

and there exists a unique morphism .G(1D ) ∈ HomC (G(D), G(D))

such that .F G(1D ) = 1F G(D) = F (1G(D) ) where the last
equality holds because F is a functor. Since F is faithful we
get .G(1D ) = 1G(D) . Consider now .g ∈ HomD (D, D  ), g  ∈
HomD (D  , D  ) and the unique morphisms .G(g) ∈ HomC (G(D),
G(D  )), respectively .G(g  ) ∈ HomC (G(D  ), G(D  )) such that

−1 −1
F G(g) = εD
.  ◦ g ◦ εD and F G(g  ) = εD 
 ◦ g ◦ εD  .

This yields
  −1
F G(g  ) ◦ G(g) = εD
.

 ◦ g ◦ g ◦ εD . (3.56)

Now having in mind that there exists a unique morphism .G(g  ◦ g) ∈

−1
HomC (G(D), G(D  )) such that .F G(g  ◦ g) = εD 
 ◦ (g ◦ g) ◦ εD ,
 
it follows from (3.56) that .G(g ◦ g) = G(g ) ◦ G(g). Therefore, G is
indeed a functor. Now the commutativity of diagram (3.55) implies that
.ε is a natural transformation. Recall that every .εD is an isomorphism

and thus .ε is in fact a natural isomorphism. We are left to construct

a natural isomorphism .η : 1C → GF . For any .C ∈ Ob C we have
−1
.ε
F (C) ∈ HomD (F (C), F GF (C)), an isomorphism in .D. Since F
is fully faithful, there exists a unique .ηC ∈ HomC (C, GF (C)) such
that .F (ηC ) = εF−1(C) . Obviously, .ηC is an isomorphism for all .C ∈
3.8 Equivalence of Categories 205

Ob C since F reflects isomorphisms. We prove now that .η is a natural

transformation. To this end, let .f ∈ HomC (C, C  ); we need to prove
the commutativity of the following diagram:

(3.57)
By naturality of .ε applied to .F (f ) we have the following commuta-
tive diagram:

i.e., εF (C  ) ◦ F GF (f ) = F (f ) ◦ εF (C)
.

⇔ F GF (f ) ◦ εF−1(C) = εF−1(C  ) ◦ F (f )

⇔ F GF (f ) ◦ F (ηC ) = F (ηC  ) ◦ F (f )
 
⇔ F GF (f ) ◦ ηC = F (ηC  ◦ f ).

Since F is faithful we get .GF (f ) ◦ ηC = ηC  ◦ f , i.e., (3.57) is

commutative, as desired.
.(3) ⇒ (1) Obvious.

.(1) ⇒ (3) Suppose F is an equivalence of categories and let .G : D → C such

that .GF ∼ = 1C and .F G ∼ = 1D . We will prove that G is right

adjoint to F . Denote by .ε : F G → 1D the natural isomorphism
arising from the above equivalence. Thus, for any .C ∈ Ob C the mor-
phism .εF (C) ∈ HomD (F GF (C), F (C)) is an isomorphism. Therefore,
−1
.ε
F (C) ∈ HomD (F (C), F GF (C)) and since F is fully faithful (see
.(1) ⇒ (2)) there exists a unique morphism .ηC ∈ HomC (C, GF (C))

such that .F (ηC ) = εF−1(C) . Since F is fully faithful, it reflects iso-

morphisms; thus .ηC is also an isomorphism. Furthermore, one can
show exactly as in the proof of .(2) ⇒ (1) that .η is also a natural
transformation. In light of Theorem 3.5.1, the proof will be finished once
we show that (3.15) and (3.16) hold. To start with, for all .C ∈ Ob C we
206 3 Adjoint Functors

have

εF (C) ◦ F (ηC ) = εF (C) ◦ εF−1(C) = 1F (C) ,

. i.e., (3.15) is fulfilled.

−1
Consider now .D ∈ Ob D and .εD : D → F G(D). From the naturality
−1
of .ε applied to the morphism .εD we obtain the following commutative
diagram:

Therefore, since F is faithful, we have

.F G(εD ) ◦ εF−1G(D) = 1F G(D)

⇔ F G(εD ) ◦ F (ηG(D) ) = 1F G(D)

 
⇔ F (G(εD ) ◦ ηG(D) ) = F 1G(D)
⇔ G(εD ) ◦ ηG(D) = 1G(D) , i.e., (3.16) holds as well.

(3) ⇒ (4) Assume now that .G : D → C is a functor such that .F  G and the
.

unit .η : 1C → GF and counit .ε : F G → 1D of this adjunction are

natural isomorphisms. Then .G  F with unit .ε−1 : 1D → F G and
counit .η−1 : GF → 1C . Indeed, as .η and .ε are natural isomorphisms,
the compatibility conditions (3.15) and (3.16) imply that for all .C ∈
Ob C and .D ∈ Ob D we have

.εF−1(C) = F (ηC ), −1
ηG(D) = G(εD ).

Therefore, for all .C ∈ Ob C and .D ∈ Ob D we have

−1  −1 
.1G(D) = ηG(D) ◦ G εD ,
 −1 
1F (C) = F ηC ◦ εF−1(C) ,

which shows that the compatibility conditions (3.15) and (3.16) are
fulfilled for .ε−1 and .η−1 .
.(4) ⇒ (3) Follows in the same fashion as .3) ⇒ 4). The proof is now finished.



As an application of the previous theorem we will highlight an equivalence of
categories involving ring localizations.
3.8 Equivalence of Categories 207

Example 3.8.6 Let R be a commutative ring with unity and .(S −1 R, j ) its local-
ization at the multiplicative set .S ⊂ R. We will show, using Theorem 3.8.5, (2) that
the category .S −1 R M of modules over the localization ring .S −1 R is equivalent to
the category .R MS−aut of modules over R on which S acts as automorphisms (see
Example 1.2.2, (14)).
Indeed, consider the restriction of scalars functor .Fj : S −1 R M →R M induced
by the ring homomorphism .j : R → S −1 R as defined in Example 1.5.3, (32). First
note that since .j : R → S −1 R is an epimorphism in .Ringc (see Example 1.3.2,
(5)), the corresponding restriction of scalars functor .Fj is fully faithful, as proved
in Example 1.6.2, (6).
Furthermore, one can easily show that S acts as automorphisms on .Fj (M), for
any .S −1 R-module M. Indeed, it is straightforward to see that for all .s ∈ S the
inverse of the multiplication map .μs : M → M, μs (m) = sm is given by the R-
linear homomorphism .μ 1 , where the juxtaposition denotes the R-module structure
s
on .Fj (M) = M. This proves that the image of the restriction of scalars functor .Fj
is contained in the category .R MS−aut .
Therefore, we have a fully faithful functor .Fj : S −1 R M → R MS−aut . We are left
to show that .Fj is essentially surjective as well. To this end, let .M ∈ Ob R MS−aut .
Then M admits an .S −1 R-module structure defined for all .r ∈ R, s ∈ S and .m ∈ M
as follows:
r
.  m = rn, (3.58)
s
where the juxtaposition denotes the R-module structure on M and n is the unique
element of M such that .sn = m. Note that the existence and uniqueness of the
element n with this property is a consequence of M being an R-module on which
S acts as an automorphism. Moreover, if .M ∈ Ob S −1 R M with the .S −1 R-module
structure given in (3.58) then .Fj (M) has the R-module structure defined as follows
for all .r ∈ R and .m ∈ M:
r
j (r)  m =
.  m = rm,
1
i.e., it coincides with the initial R-module structure on M. This finishes the proof.

.

As stated in the beginning, equivalent categories share most of the important

properties:
Proposition 3.8.7 Let .C and .D be two equivalent categories. Then .C is
(co)complete if and only if .D is (co)complete.
Proof Given an equivalence of categories .F : C → D, by Theorem 3.8.5, (3) there
exists a right adjoint .G : D → C of F such that the unit .η : 1C → GF and
the counit .ε : F G → 1D of the adjunction are natural isomorphisms. Assume
first that .D is complete and let .H : J → C be a functor, where J is a small
208 3 Adjoint Functors

category.
 As .D is a complete category, the functor .F H : J → D has a limit, say
. L, (pj : L →  F H (j ))j ∈Ob J . Moreover, as G is right adjoint
 to F , Theorem 3.4.4
implies that . G(L), (G(pj ) : G(L) → GF H (j ))j ∈Ob J is the limit of the functor
.GF H : J → C. Now .η
−1 : GF → 1
C (as defined in Example 1.7.2, (6))
−1
is a natural isomorphism and consequently .ηH : GF H → H (as defined in
Example
 1.7.2, (7)) is also a natural isomorphism. Now
 Lemma 2.2.15, (1) implies
−1
that . G(L), (ηH (j ) ◦ G(pj ) : G(L) → H (j ))j ∈Ob J is the limit of H and therefore
the category .C is complete.
The statement concerning cocompleteness follows similarly using Theo-
rem 3.8.5, (4). 

Definition 3.8.8 A skeleton of a category .C is a full subcategory .C0 of .C such that
each object of .C is isomorphic to exactly one object of .C0 .
Example 3.8.9 A skeleton of a given category .C always exists; indeed, it can be
constructed by choosing8 an object from each isomorphism class of objects in .C
and considering the full subcategory of .C with this objects class. .

Moreover, as an easy consequence of Theorem 3.8.5 we obtain the following:

Corollary 3.8.10 A category is equivalent to any of its skeletons.
Proof Let .C0 be a skeleton of a given category .C. Then the inclusion functor
I : C0 → C is fully faithful and essentially surjective. The conclusion now follows
.

from Theorem 3.8.5, (2). 


Proposition 3.8.11 Let .C0 and .D0 be two skeletons of the categories .C and .D,
respectively. Then .C and .D are equivalent if and only if .C0 and .D0 are isomorphic.
Proof Assume first that the categories .C0 and .D0 are isomorphic; in particular, the
two categories are also equivalent. Moreover, recall that the categories .C0 and .C,
respectively .D0 and .D, are equivalent. Then Proposition 3.8.4, (2) and (3) show
that .C is equivalent to .D.
Conversely, assume now that .C and .D are equivalent. As noted before, .C0
and .C, respectively .D0 and .D, are also equivalent and using again Proposi-
tion 3.8.4, (2) and (3) we obtain that .C0 and .D0 are equivalent too. Thus, there
exist two functors .F : C0 → D0 , G : D0 → C0 and two natural isomorphisms
.ψ : F G → 1D , ϕ : GF → 1C . Hence, for each .C ∈ Ob C0 , we have
0 0
an isomorphism .ϕC : GF (C) → C. Since .C0 is a skeleton of .C, such an
isomorphism can only be an identity morphism and therefore .GF (C) = C.
Similarly, we obtain .F G(D) = D for all .D ∈ Ob D0 . Furthermore, if .f ∈
HomD0 (D1 , D2 ) = HomD0 (F G(D1 ), F G(D2 )), since F is in particular fully
faithful by Theorem 3.8.5, (2) there exists a unique .f  ∈ HomC0 (G(D1 ), G(D2 ))
such that .f = F (f  ). We can now define a functor .H : D0 → C0 as follows:

8 Recall that the axiom of choice is assumed to hold.

3.8 Equivalence of Categories 209

H (D) = G(D), D ∈ Ob D0 , .
. (3.59)
H (f ) = f  , f ∈ HomD0 (D1 , D2 ). (3.60)

It can be easily seen that H is a functor. Indeed, for all .D ∈ Ob D0 we have

1D = 1F G(D) = F (1G(D) ), which implies that .H (1D ) = 1G(D) = 1H (D) .
.

Furthermore, consider .f ∈ HomD0 (D1 , D2 ), g ∈ HomD0 (D2 , D3 ) and let .f  ∈

HomC0 (G(D1 ), G(D2 )), g  ∈ HomC0 (G(D2 ), G(D3 )) be the unique morphisms
such that .F (f  ) = f and .F (g  ) = g. Then .g  ◦ f  ∈ HomC0 (G(D1 ), G(D3 )) is
the unique morphism such that .g ◦ f = F (g  ◦ f  ) and we obtain

. H (g) ◦ H (f ) = g  ◦ f  = H (g ◦ f ).

This shows that H is a functor. The proof will be finished once we show that H is
the inverse of F . Indeed, for all .C ∈ Ob C0 and .D ∈ Ob D0 we have

(3.59) (3.59)
.F H (D) = F G(D) = D, H F (C) = GF (C) = C.

Furthermore, if .f ∈ HomD0 (D1 , D2 ) and .f  ∈ HomC0 (G(D1 ), G(D2 )) is the

unique morphism such that .F (f  ) = f , we obtain
  (3.60)
F H (f ) = F (f  ) = f.
.

Similarly, if we have .t ∈ HomC0 (C1 , C2 ) = HomC0 (GF (C1 ), GF (C2 )), then
  (3.60)
.H F (t) = t and the proof is now finished. 

In light of our previous result, loosely speaking, we can conclude that two
equivalent categories might differ only by the numbers of isomorphic copies of the
same object. Another important consequence is the following:
Corollary 3.8.12 The skeleton of a category is unique up to isomorphism.
Proof As proved in Proposition 3.8.4, (1) any category .C is trivially equivalent to
itself. Now any two skeletons of .C are isomorphic by Proposition 3.8.11. 

Categories with a small skeleton have been mentioned in passing in Exam-
ple 1.2.2, (5). We discuss them here in more detail.
Definition 3.8.13 A category is called essentially small if its skeleton is a small
category.
A useful characterization of essentially small categories is the following:
Proposition 3.8.14 A category is essentially small if and only if it is equivalent to
a small category.
Proof Consider .C to be a category equivalent to a small category .D. If .C0 and .D0
denote the skeleton of .C and .D, respectively, then in particular .D0 is also a small
210 3 Adjoint Functors

category. Now Proposition 3.8.11 implies that .C0 and .D0 are isomorphic categories
and therefore .C0 is small. This shows that .C is essentially small.
Conversely, if .C is essentially small then its skeleton .C0 is a small category and
the conclusion follows from Corollary 3.8.10. 

Examples 3.8.15
(1) The categories FinSet and .K Mf d are essentially small. Indeed, a skeleton of
FinSet is given by its full subcategory whose objects are the sets .n, for all
∅ if n = 0
.n ∈ N, where .n = . The latter category is obviously
{1, . . . , n} if n ∈ N\{0}
small.
Furthermore, Example 3.8.2 shows that the category .K Mf d is equivalent
to the small category .MatK defined in Example 1.2.2, (19). Hence, .K Mf d is
essentially small by virtue of Proposition 3.8.14.
(2) The categories Set, Grp, Ring, Top and .K M are not essentially small. We only
prove the assertion regarding the category Set and leave the others to the reader.
To this end, assume there exists a small skeleton .C of Set. Given that .Ob C is a
set we can consider .Ob C = {Xi | i ∈ I }, where I is a set and .Xi ∈ Ob Set for
all .i ∈ I . Now let .X = P( i∈I Xi ) be the power set of the coproduct of the
family of objects .(Xi )i∈I in Set. As .C is assumed to be the skeleton of Set, there
exists some .i0 ∈ I and an isomorphism in Set (i.e., a set bijection) between X
and .Xi0 . From Cantor’s theorem9 we have .| i∈I Xi | < |X|. Furthermore, as

. i∈I Xi is the union of the sets .Xi = Xi × {i} (see Example 2.1.5, (9)) we also
have .|Xi | ≤ | i∈I Xi | for all .i ∈ I . Putting all this together leads in particular
to .|Xi0 | < |X| and we have reached a contradiction as X was assumed to be
isomorphic to .Xi0 . Therefore, Set cannot have a small skeleton, as desired. .
We end this section with a generic example of a duality of categories. Let .C
and .D be two small categories. We denote by .FunL (C, D) the full subcategory
of .Fun(C, D) consisting of all functors from .C to .D which admit a left adjoint.
Similarly, .FunR (C, D) denotes the full subcategory of .Fun(C, D) consisting of all
functors from .C to .D which admit a right adjoint.
Theorem 3.8.16 For all small categories .C and .D, we have a duality of categories
between .FunR (C, D) and .FunL (D, C).
Proof We will define an equivalence of categories

H : FunL (D, C) → FunR (C, D)op

.

as follows. Given .G ∈ FunL (D, C), we choose a functor .F : C → D such that

F  G and define .H (G) = F .
.

9 Cantor’s theorem: For any set X we have .|X| < |P(X)| ([37, Theorem 2.21]).
3.8 Equivalence of Categories 211

Furthermore, if .G1 , G2 ∈ FunL (D, C) with corresponding left adjoint functors

.F1 , F2 : C → D and .α : G1 → G2 is a natural transformation then we define
.H (α) = α, where .α : F2 → F1 is the unique natural transformation which makes

diagram (3.27) commutative, as constructed in (the proof of) Theorem 3.5.5.

We show first that H is indeed a functor. To start with, let .G : D → C be a
functor whose chosen left adjoint we denote
 by
 F and consider .α to be the identity
natural transformation on G, i.e., .α = 1G(D) D∈Ob D . Then, for all .C ∈ Ob C, .α C
is the unique morphism such that .θC, D (f ◦ α C ) = θC, D (f ) holds for all morphisms
.f : F (C) → D. Obviously, this implies .α C = 1F (C) , which shows that H respects

identities.
Consider now functors .Gi : D → C and choose .Fi : C → D to be the
corresponding left adjoints, .i = 1, 2, 3. Moreover, denote by .θ i the natural
isomorphism induced by the adjunction .Fi  Gi , i = 1, 2, 3. If .α : G1 → G2
and .β : G2 → G3 are natural transformations and .H (α) = α, H (β) = β we aim to
show that .H (β◦α) = β ◦op α. This comes down to showing that for each .C ∈ Ob C,
the morphism . β ◦op α C = α C ◦β C is the unique one such that the following holds:

D (h ◦ α C ◦ β C ) = βD ◦ αD ◦ θC, D (h)
3 1
θC,
. (3.61)

for all .h ∈ HomD (F1 (C), D). To this end, recall that .α C and .β C are the unique
morphisms such that

D (f ◦ α C ) = αD ◦ θC, D (f ), .
2 1
.θC, (3.62)

D (g ◦ β C ) = βD ◦ θC, D (g)
3 2
θC, (3.63)

for all .f ∈ HomD (F1 (C), D), g ∈ HomD (F2 (C), D). Putting all the above
together yields

(3.62) (3.63)
.βD ◦ αD ◦ θC,
1
D (h) = βD ◦ θC, D (h ◦ α C ) = θC, D (h ◦ α C ◦ β C )
2 3

for all .h ∈ HomD (F1 (C), D). Therefore, H is indeed a functor.

The proof will be finished once we  show that H is essentially surjective and
fully faithful. To this end, let .F ∈ Ob FunR (C, D)op and consider a right adjoint
 
.G : D → C of F . Then .H (G) = F for some .F : C → D such that .F  G. By

 
Theorem 3.6.4 we have a natural isomorphism .F  F . Therefore .F  F = H (G),
which shows that H is essentially surjective.
Next
 we show  that H is fully faithful. To this end, consider .G1 , G2 ∈
Ob FunL (D, C) and the map

HG1 , G2 : HomFunL (D, C) (G1 , G2 ) → HomFunR (C, D)op (F1 , F2 )

.
212 3 Adjoint Functors

defined by

HG1 , G2 (γ ) = H (γ ) = γ for all natural transformations γ : G1 → G2 ,

.

where .Fi : C → D is the left adjoint of .Gi and denote by .θ i the natural isomorphism
corresponding to the adjunction .Fi  Gi , i = 1, 2.
Assume .α, β : G1 → G2 are natural transformations such that .HG1 , G2 (α) =
HG1 , G2 (β), i.e., .α = β. This implies .θC, 2 (f ◦ α ) = θ 2 (f ◦ β ) and
D C C, D C
consequently we have

αD ◦ θC,
.
1
D (f ) = βD ◦ θC, D (f ),
1
(3.64)

for all .f ∈ HomD (F1 (C), D). Considering

 1 −1
C = G1 (D) and f = θG
.
1 (D), D
(1G(D) )

in (3.64) yields .αD = βD for all .D ∈ Ob D. Hence .α = β and H is faithful.

We are left to show that H is full. Consider a natural transformation .μ : F2 → F1 ,
i.e., .μ ∈ HomFunR (C, D)op (F1 , F2 ). For all .D ∈ Ob D define

αD = θG
.
2
1 (D), D
1
(εD ◦ μG1 (D) ) ∈ HomD (G1 (D), G2 (D)),

where .ε1 denotes the counit of the adjunction .F1  G1 . We show first that the
morphisms .αD , D ∈ Ob D, form a natural transformation .α : G1 → G2 , i.e., for
all .r ∈ HomD (D, D  ) the following diagram is commutative:

(3.65)
To start with, note that .ε1 ◦ μG1 : F2 G1 → 1D is a natural transformation and
therefore the following diagram is commutative:

(3.66)
3.8 Equivalence of Categories 213

Therefore, we have
 1 
G2 (r) ◦ αD = G2 (r) ◦ θG
.
2
1 (D), D
εD ◦ μG1 (D)
(3.2)  
= θG
2
1 (D), D
 r ◦ εD ◦ μG1 (D)
1

(3.66)  1 
= θG
2
1 (D), D
 εD  ◦ μG1 (D  ) ◦ F2 G1 (r)

(3.1)  1 
= θG
2

1 (D ), D
 εD  ◦ μG1 (D  ) ◦ G1 (r)

= αD  ◦ G1 (r),

which shows that (3.65) indeed holds and hence .α is a natural transformation. The
proof will be finished if we show that .HG1 , G2 (α) = μ or, equivalently, that the
following holds for all .f ∈ HomD (F1 (C), D):

D (f ◦ μC ) = αD ◦ θC, D (f ).
2 1
θC,
.

To start with, recall that .μ : F2 → F1 is a natural transformation and therefore the

following diagram is commutative:

. (3.67)
 1  1
i.e., F1 θC, D (f ) ◦ μC = μG1 (D) ◦ F2 θC, D (f ).

Then, by the way we defined .α, we have

αD ◦ θC,
.
1
D (f ) = θG1 (D), D (εD ◦ μG1 (D) ) ◦ θC, D (f )
2 1 1

(3.1) 2
  1 
= θC, D εD 1
◦ μG1 (D) ◦ F2 θC, D (f )
(3.67) 2
   
= θC, D εD 1
◦ F1 θC,
1
D (f ) ◦ μC

(3.25)
= θC,
2
D (f ◦ μC ),

as desired. This concludes the proof. 


Examples of duality theorems abound in the mathematical landscape and are
often used to build bridges between different fields. We only mention here some
214 3 Adjoint Functors

of the most important ones: the category of compact topological abelian groups
is dual to the category of abelian groups (Pontryagin duality); the category of
commutative unital .C ∗ -algebras is dual to the category of compact Hausdorff
topological spaces (Gelfand-Naimark duality); the category of compact and totally
disconnected topological spaces10 is dual to the category of Boolean algebras (Stone
duality). For further details we refer the reader to [14, 43].

3.9 Localization

The idea of formally adjoining inverses in a systematic way, called localization,

exists for many algebraic structures such as rings or modules. A similar construction
can be performed in the general setting of category theory. Indeed, consider S to be
a class of morphisms in a category .C. The purpose of localization as first introduced
in [26] is to construct a new category .CS in which all morphisms in S became
invertible, while approximating the original category as closely as possible. The
precise definition is the following:
Definition 3.9.1 A localization of a category .C (or category of fractions as referred
to in ([8, Section 5.2]) by a class of morphisms S of .C is a category .CS together with
a functor .F : C → CS such that
(1) for any .s ∈ S, F (s) is an isomorphism in .CS ;
(2) if .G : C → D is a functor such that for all .s ∈ S, G(s) is an isomorphism in
.D, there exists a unique functor .H : CS → D such that the following diagram

is commutative:

. (3.68)

Theorem 3.9.2 Let .C be a category. Then there exists a localization of .C by any set
of morphisms S of .C.
Proof In order to construct the localization of .C by the set S we start by defining an
oriented graph .Γ as follows:
• the vertices of .Γ are the objects of .C;

10 A topological space that is compact and totally disconnected is called a Stone space.
3.9 Localization 215

• the edges of .Γ are the morphisms of .C (any morphism .f ∈ HomC (X, Y ) is seen

as an oriented edge ) together with the set .{xs | s ∈ S}, where

xs is an edge having the same vertices as s but the opposite orientation (i.e., if
.

s ∈ HomC (X, Y ) then

. ).
Two paths in the above graph will be called equivalent if one can be transformed
into the other by applying the following elementary operations a finite number of
times:

• if .f ∈ HomC (X, Y ) and .g ∈ HomC (Y, Z) then the path

can be replaced by the composition path ;

• if .s ∈ S, s ∈ HomC (X, Y ) then the path can be

replaced by the path ; similarly, the path can

be replaced by the path .
It is straightforward to see that this is an equivalence relation on the class of paths
of .Γ . We denote by .
γ the equivalence class of the path .γ . The localization category
.CS is defined as follows:

Ob CS = Ob C;
.

HomCS (X, Y ) = {
. γ | γ is a path in Γ from X to Y },11 for all .X, Y ∈ Ob C,
with the composition of morphisms in .CS induced by the concatenation of paths and
the identity maps given by the trivial paths. The functor .F : C → CS is defined as
follows:
F (X) = X, for all .X ∈ Ob C;
.

F (f ) = f, for all .f ∈ HomC (X, Y ).

.

Note that if .s ∈ S, .s ∈ HomC (X, Y ), then .F (s) = 

s has an inverse in .CS , namely
.xs , where . We are left to show that the pair .(CS , F ) satisfies the second
condition in Definition 3.9.1 as well. To this end, let .D be a category and .G : C → D
a functor such that .G(s) is an isomorphism for any .s ∈ S. Consider the functor
.H : CS → D defined as follows:

H (X) = G(X), for all .X ∈ Ob Cs = Ob C;

.

H (f) = G(f ), for all .f ∈ HomC (X, Y );

.

, for all .s ∈ S, .s ∈ HomC (Z, W ).

The way we defined the functor above ensures the commutativity of diagram (3.68)
as well as the uniqueness of H with this property. Indeed, if a functor H makes

11 This is obviously a set as a consequence of S being a set.

216 3 Adjoint Functors

diagram (3.68) commutative, then we have .H (X) = G(X) and .H (f) = G(f ),
for all .X ∈ Ob Cs = Ob C and .f ∈ HomC (X, Y ); furthermore, in order for  H to
be a functor and to respect compositions and identities, it should satisfy .H xs =
G(s)−1 , for all .s ∈ S.
We are left to prove that H is well-defined. To this end, consider two paths u
and v in .Γ such that . u =
v . Since the paths u and v are equivalent, we can turn
u into v after a finite number of elementary operations. Thus, it suffice to prove
that by applying H to each of these elementary operations we obtain equalities

in .D. Indeed, whenever in .CS we

obviously have

H (g
. g ) ◦ H (f) = H (
◦ f ) = G(g ◦ f ) = G(g) ◦ G(f ) = H ( g ◦ f).

Analogously, whenever in .CS it follows

that we have

xs ◦
H (
. s) = H ( s) = G(s)−1 ◦ G(s) = 1G(X) = G(1X ) = H (1X ).
xs ) ◦ H (

Therefore H is well-defined and the proof is now finished. 


Proposition 3.9.3 When it exists, the localization of a category .C by a class of
morphisms S of .C is unique up to isomorphism.
Proof Suppose .(CS , F ) and .(CS , F ) are two localizations of .C by S. Thus, there
exists a unique functor .G : CS → CS such that

G ◦ F = F.
. (3.69)

Similarly, as .(CS , F ) is also a localization of .C by S, there exists a unique functor

G : CS → CS such that
.

G ◦ F = F.
. (3.70)

By putting all this together we obtain

(3.70) (3.69)
F = G ◦ F = G ◦ G ◦ F = (G ◦ G) ◦ F.
. (3.71)

Applying Definition 3.9.1 to the pair .(CS , F ), seen both as a localization and as the
other pair, yields a unique functor .H : CS → CS such that .H ◦ F = F . By the
uniqueness of H we must have .H = 1CS . Moreover, since by (3.71) the functor
3.9 Localization 217

G ◦ G makes the same diagram commutative, we obtain .G ◦ G = 1CS .

.

. (3.72)

Similarly one can prove that .G ◦ G = 1C and therefore the categories .CS and .CS
S
are isomorphic, as desired. The proof is now finished. 

One of the situations when the localization of a category can be described, up to
equivalence of categories, even without assuming the localizing class of morphisms
to be a set, is that of reflective subcategories.
Theorem 3.9.4 Let .I : A → B be a reflective subcategory inclusion with reflector
R : B → A and denote by S the class of all morphisms s of .B such that .R(s) is an
.

isomorphism in .A. Then the localization of .B by S is equivalent to .A.

Proof Let .η : 1B → I R and .ε : RI → 1A be the unit and respectively the counit
of the adjunction .R  I . To start with, recall that by Lemma 3.6.6, (3) the counit
.ε : RI → 1A is a natural isomorphism. Moreover, for all .B ∈ ObB we have
(3.15)
.1R(B) = εR(B) ◦ R(ηB ) and given  that .εR(B) is an isomorphism it follows that
.R(ηB ) ∈ HomA R(B), RI R(B) is an isomorphism as well. Therefore, .ηB ∈ S

for all .B ∈ Ob B.
Define a category .BS as follows:
Ob BS = Ob B;
.

HomBS (B, B  ) = HomA (R(B), R(B  )) for all B, .B  ∈ Ob BS ,

.

with the composition of morphisms and identities given by those of .A.

First we prove that .(BS , F ) is the localization of .B with respect to S, where
F : B → BS is the functor defined as follows:
.

F (B) = B, for all .B ∈ Ob B;

.

F (f ) = R(f ), for all .f ∈ HomB (B, B  ).

.

Recall that S is the class of all morphisms s of .B such that .R(s) is an isomorphism
and therefore .F (s) is obviously an isomorphism for any .s ∈ S.
Consider now another functor .G : B → D such that .G(s) is an isomorphism
for all .s ∈ S. We need to find a functor .H : BS → D which makes the following
218 3 Adjoint Functors

diagram commutative:

. (3.73)

Having in mind that .ηB ∈ S for any .B ∈ ObB, it can be easily seen that a functor
H which makes the above diagram commute has the following property for all .B ∈
ObB:

(3.15)     −1
H (εR(B) ) = H R(ηB )−1 = H R(ηB )
.

  −1 (3.73)
= H F (ηB ) = G(ηB )−1 . (3.74)

Furthermore, for any morphism .f ∈ HomBS (B, B  ) = HomA (R(B), R(B  )), the
naturality of .ε : RI → 1A renders the following diagram commutative:

(3.75)
Therefore, for any .f ∈ HomBS (B, B  ) = HomA (R(B), R(B  )) we have

(3.15)  
.H (f ) = H f ◦ εR(B) ◦ R(ηB )

(3.75) 
= H εR(B  ) ◦ RI (f ) ◦ R(ηB )
 
= H εR(B  ) ◦ H RI (f ) ◦ H R(ηB )
(3.74)
= G(ηB  )−1 ◦ H F I (f ) ◦ H F (ηB )
(3.73) −1
= G(ηB  ◦ GI (f ) ◦ G(ηB ).

We define the functor .H : BS → D as follows:

H (B) = G(B), for all .B ∈ Ob BS ;
.

H (f ) = G(ηB  )−1 ◦ GI (f ) ◦ G(ηB ), for all .f ∈ HomBS (B, B  ).

.
3.9 Localization 219

The above discussion proves that H is the unique functor which might render dia-
gram (3.73) commutative. We are left to prove that indeed H makes diagram (3.73)
commute. To this end we will use the naturality of .η, i.e., the commutativity of the
above diagram for any .g ∈ HomB (B, B  ):

. (3.76)

Obviously, for any .B ∈ Ob BS we have .H ◦ F (B) = H (B) = G(B). Moreover, for

any .g ∈ HomB (B, B  ) we have

H ◦ F (g) = H (R(g)) = G(ηB  )−1 ◦ GI R(g) ◦ G(ηB )

.

= G(ηB  )−1 ◦ G(I R(g) ◦ ηB )

(3.76)
= G(ηB  )−1 ◦ G(ηB  ◦ g) = G(g).

Next we show that the category .BS is equivalent to .A. Indeed, consider the
functor .T : A → BS defined as follows:
T (A) = I (A), for all .A ∈ Ob A;
.

T (f ) = RI (f ), for all .f ∈ HomA (A, A ).

.

T is well-defined as for all .f ∈ HomA (A, A ), we have

   
.RI (f ) ∈ HomA RI (A), RI (A ) = HomBS I (A), I (A ) .

Furthermore, T is fully faithful as RI is naturally isomorphic to .1A via .ε. Indeed,

let .h1 , .h2 ∈ HomA (A, A ) such that .RI (h1 ) = RI (h2 ). The naturality of .ε renders
the following diagrams commutative for .i = 1, 2:

. (3.77)

Hence we obtain .h1 ◦ εA = h2 ◦ εA and since .εA is an isomorphism we get .h1 = h2

as desired. This shows that T is faithful.
220 3 Adjoint Functors

Consider now A, .A ∈ Ob A, .v ∈ HomA (RI (A), RI (A )) and define

−1
u = εA ◦ v ◦ εA
. ∈ HomC (A, A ). (3.78)

We will prove that .RI (u) = v. Indeed, using again the naturality of .ε we obtain

−1 (3.77) (3.78) −1
εA ◦ RI (u) ◦ εA
. = u = εA ◦ v ◦ εA .

Since .εA and .εA are isomorphisms we get .RI (u) = v and we have proved that T is
full.
Moreover, for any .B ∈ Ob BS we have an isomorphism

R(ηB ) ∈ HomA (R(B), RI R(B)) = HomBS (B, I R(B)) = HomBS (B, T R(B)),
.

and this shows that T is essentially surjective as well. Therefore, by Theo-

rem 3.8.5, (2) T is an equivalence of categories and the proof is now finished. 

Our next example connects ring and module localizations with the categorical
notion introduced in Definition 3.9.1. We will show that the category of modules
over the localized ring .S −1 R is equivalent to a localization, in the categorical sense,
of the category of modules over R.
Example 3.9.5 Let Rbe a commutative ring with unity, S a multiplicative subset
of R, and . S −1 R, j the corresponding localization ring. If .M ∈ Ob R M,
we denote by .(S −1 M, ϕM ) the corresponding localization module with respect
to S, where .S −1 M ∈ Ob S −1 R M and .ϕM : M → S −1 M is the R-module
homomorphism defined by .ϕM (m) = m1 , for all .m ∈ M. Throughout, .R MS−aut
stands for the category of left R-modules on which S acts as an automorphism (see
Example 1.2.2, (14)).
Consider now the inclusion functor .I : R MS−aut → R M and the functor
.L : R M → R MS−aut defined as follows for all R-modules M and .f ∈
HomR M (M, N ):

L(M) = S −1 M,
. L(f ) = f,
 x  f (x)
f: S −1 M → S −1 N, f = , for all x ∈ M, s ∈ S.
s s

Note that we see .S −1 M as an R-module via j and by [3, Proposition 12.1], the
multiplication map .μs : S −1 M → S −1 M is bijective for all .s ∈ S and therefore L
is well-defined.
We will show that L is left adjoint to the inclusion functor I . To this end, let
.ϕ : 1 → I L be the natural transformation defined for all R-modules M by
RM
the R-module homomorphism .ϕM : M → S −1 M associated with the localization
3.10 (Co)limits as Adjoint Functors 221

S −1 M. Indeed, if .f ∈ HomR M (M, N ) and .m ∈ M we have

.

   m  f (m)  
.f◦ ϕM (m) = f = = ϕN (f (m)) = ϕN ◦ f (m),
1 1
which shows that .ϕ is a natural transformation, as claimed.
Consider .u ∈ HomR M (M, I (N )), with .N ∈ Ob R MS−aut . Now let
 
.u : S
−1 M → N be defined for all .x ∈ M and .s ∈ S by .u x = y, where y
s
is the unique element of N such that .sy = u(x); note that since the multiplication
by s is a bijection on N, we have a unique such y. It is straightforward to see that
.u is a well-defined R-module homomorphism. Furthermore, .u : S
−1 M → N is

the unique R-module homomorphism such that .u ◦ ϕM = u. As both .S −1 M and

N are objects in .R MS−aut , which is a full subcategory of .R M, we obtain that .u
is a morphism in .R MS−aut as well. To summarize, we have a unique morphism
.u : S
−1 M → N in . MS−aut such that .I (u) ◦ ϕ = u and Theorem 3.6.1, (2) shows
R M
that L is left adjoint to the inclusion functor I . Therefore, .R MS−aut is a reflective
subcategory of .R M. Now Theorem 3.9.4 implies that .R MS−aut is equivalent to the
localization (in the sense of Definition 3.9.1) of the category .R M with respect to
the family of morphisms f in .R M for which .L(f ) is an isomorphism. Furthermore,
recall from Example 3.8.6 that .R MS−aut is also equivalent to .S −1 R M. We can now
conclude by Proposition 3.8.4, (3) that .S −1 R M is equivalent to a localization, in the
sense of category theory, of the category .R M. .

3.10 (Co)limits as Adjoint Functors

We start by recalling from Theorem 2.2.16 (resp. Theorem 2.2.17) that taking
(co)limits yields a functor. It turns out that in certain conditions this limit (resp.
colimit) functor has a left (resp. right) adjoint, namely the diagonal functor defined
in Proposition 1.9.8.
Theorem 3.10.1 Let I be a small category and .C an arbitrary category.
(1) The diagonal functor .Δ : C → Fun(I, C) has a right adjoint if and only if .C is
complete. In this case, the right adjoint is the limit functor .lim : Fun(I, C) →
C.
(2) The diagonal functor .Δ : C → Fun(I, C) has a left adjoint if and only
if .C is cocomplete. In this case, the left adjoint is the colimit functor
.colim : Fun(I, C) → C.

Proof
(1) Assume first that any functor .F : I → C, where I is a small category,
has a limit. We will define a bijective map .θ : HomFun(I, C) (Δ, −) →
HomC (−, lim), natural in both variables. To this end, for any .X ∈ Ob C,
222 3 Adjoint Functors

 
.F ∈ Ob Fun(I, C) and any natural transformation .α : Δ(X) → F , we define
.θX, F (α) = f , where .f : X → lim F is the unique morphism in .C which makes

the following diagram commutative for all .i ∈ Ob I :

. (3.79)
 
and . lim F, (pi : lim F → F (i))i∈Ob I denotes the limit of F .
First we prove that each map .θX, F is bijective. Indeed, consider two
natural transformations .α, .β ∈ HomFun(I, C) (Δ(X), F ) such that .θX, F (α) =
θX, F (β) = f . This implies that for all .i ∈ Ob I we have .pi ◦ f = αi and
.pi ◦ f = βi . Hence .αi = βi for all .i ∈ Ob I , which implies that the two natural

transformations .α and .β coincide. This shows that .θX, F is injective.

Furthermore, consider .f ∈ HomC (X, lim F ) and for all .i ∈ Ob I define
.αi ∈ HomC (X, F (i)) by .αi = pi ◦ f . We will show that the family
 
of morphisms . αi : Δ(X)(i) → F (i) i∈Ob I form a natural transformation
.α : Δ(X) → F . To this end, let .u ∈ HomI (i, j ); we will show that the

following diagram is commutative:

 
Indeed, recall that . lim F, (pi : lim F → F (i))i∈Ob I is in particular a cone on
F and therefore we have .F (u)◦pi = pj . This yields .F (u)◦αi = F (u)◦pi ◦f =
pj ◦ f = αj , which shows that the above diagram is indeed commutative.
Next we show that .θ is natural in both variables. First, consider .f ∈
HomC (X , X). We will prove the commutativity of the following diagram,
which ensures the naturality in the first variable:

i.e.,
. HomC (f, lim F ) ◦ θX, F = θX , F ◦ HomFun(I, C) (Δ(f ), F ).
(3.80)
3.10 (Co)limits as Adjoint Functors 223

To this end, let .α ∈ HomFun(I, C) (Δ(X), F ), i.e., .α : Δ(X) → F is a natural

transformation. We obtain

HomC (f, lim F ) ◦ θX, F (α) = t ◦ f,

.
 
θX , F ◦ HomFun(I, C) (Δ(f ), F )(α) = θX , F α ◦ Δ(f ) ,

where .t : X → lim F is the unique morphism in .C which makes the following

diagram commutative for all .i ∈ Ob I :

. (3.81)
 
Hence, we are left to show that .θX , F α ◦ Δ(f ) = t ◦ f . Having in mind the
way .θ was defined, this comes down to proving that .t ◦ f makes the following
diagram commutative for all .i ∈ Ob I :

(3.81)
Indeed, we have .pi ◦ t ◦f = αi ◦f for all .i ∈ Ob I and this shows that (3.80)
holds.
Consider now two functors F , .F  : I → C and .β ∈ HomFun(I, C) (F, F  ),


i.e., .β : F →
 F is a natural transformation. We denote by . lim F, (pi : lim F →

F (i))i∈Ob I and . lim F  , (si : lim F  → F  (i))i∈Ob I the limit of F and .F 

respectively. The naturality of .θ in the second variable comes down to proving
the commutativity of the following diagram:

i.e.,
. HomC (X, lim β) ◦ θX, F = θX, F  ◦ HomFun(I, C) (Δ(X), β).
(3.82)
224 3 Adjoint Functors

To this end, let .γ ∈ HomFun(I, C) (Δ(X), F ), i.e., .α : Δ(X) → F is a natural

transformation. We obtain

HomC (X, lim β) ◦ θX, F (γ ) = lim β ◦ t,

.
 
θX, F  ◦ HomFun(I, C) (Δ(X), β)(γ ) = θX, F  β ◦ γ = r,

where .t : X → lim F and .r : X → lim F  are the unique morphisms in .C which

make the following diagrams commutative for all .i ∈ Ob I :

(3.83)
We are left to show that .lim β ◦ t = r. To this end, recall that .lim β ∈
HomC (lim F, lim F  ) is the unique morphism in .C which makes the following
diagram commute for all .i ∈ Ob I :

. (3.84)

By Proposition 2.2.14, (1) we only need to show that .si ◦lim β ◦t = si ◦r for all
(3.84) (3.83) (3.83)
.i ∈ Ob I . Indeed, we have .si ◦ lim β ◦t = βi ◦pi ◦ t = βi ◦ γi = si ◦r,
as desired.
Assume now that the diagonal functor .Δ : C → Fun(I, C) has a right
adjoint, denoted by .R : Fun(I, C) → C. We will show that any func-
tor .F : I → C has a limit. To this end, let .ε : ΔR → 1Fun(I, C) and
.θ : HomFun(I, C) (Δ, −) → HomC (−, R) be the counit and respectively the

natural bijection induced by the adjunction .Δ  R. In particular, .εF : ΔR(F ) →

F is a natural transformation for any functor .F : I →
 C. Proposition 2.2.4, (1)
implies that . R(F ), (εF )i ) : R(F ) → F (i)i∈Ob I is a cone on F . We will
show that . R(F ), (εF )i ) : R(F ) → F (i)i∈Ob I is in fact the limit of F .
Indeed, consider another cone . X, (αi ) : X → F (i)i∈Ob I on F . Using again
Proposition 2.2.4, (1) we obtain that .α : ΔX → F , where .α = (αi )i∈Ob I , is a
natural transformation, i.e., .α ∈ HomFun(I, C) (ΔX , F ). Now Theorem 3.6.1, (3)
3.10 (Co)limits as Adjoint Functors 225

yields a unique morphism .g ∈ HomC (X, R(F )) such that the following
diagram is commutative:

. (3.85)

The above equality between the natural transformations .εF ◦ Δ(g) and .α comes
down to identities between the corresponding morphisms associated to each
.i ∈ Ob I . In light of Example 1.7.2, (5) we have .(εF )i ◦ g = αi for all .i ∈ Ob I

and therefore the following diagram is commutative:

The proof will be finished once we show that g is the unique morphism in .C
which makes the above diagram commutative. To this end, assume that .h ∈
HomC (X, R(F )) such that .(εF )i ◦ h = αi for all .i ∈ Ob I . This leads to
.εF ◦Δ(h) = α and the uniqueness of the morphism which makes diagram (3.85)

commutative implies .g = h, as desired.



Remark 3.10.2 Functors having a left adjoint which is also a right adjoint are
called Frobenius functors in the literature (see [17] for further details). The notion
was first introduced in [16] and is motivated by the following example coming from
ring theory: a ring extension .R → S is Frobenius (in the sense of [41]) if and only if
the corresponding restriction of scalars functor (Example 1.5.3, (32)) is Frobenius.
In light of Theorem 3.10.1, since for complete and cocomplete categories .C the
diagonal functor .Δ : C → Fun(I, C) has both a left and a right adjoint, it is natural
to ask when the two adjoints are naturally isomorphic (or, equivalently, when the
diagonal functor is Frobenius). This problem was considered in [20] and, given a
complete and cocomplete category .C, the small categories I for which the diagonal
functor is Frobenius are characterized. .

As a straightforward consequence of Theorem 3.10.1 we can easily conclude

using Theorem 3.4.4 that if .C is complete (resp. cocomplete) then the diagonal
functor preserves colimits (resp. limits). However, we will see that even without the
(co)completeness assumption on .C, the diagonal functor still preserves all existing
small (co)limits.
226 3 Adjoint Functors

Proposition 3.10.3 Let I be a small category and .C an arbitrary category. Then

the diagonal functor .Δ : C → Fun(I, C) preserves all existing small (co)limits.
Proof We only show that .Δ preserves limits; colimit preservation follows similarly
 let J be a small category and .G : J → C a
and is left to the reader. To this end,
functor whose limit we denote by. L, (pj : L → G(j ))j ∈Ob J . First note that, as
 
proved in Lemma 2.5.5, the pair . Δ(L), Δ(pj ) : Δ(L) → Δ(G(j )) j ∈Ob J is a
cone on .Δ ◦ G : J → Fun(I, C). Recall that each natural transformation .Δ(pj ) is
defined by . (Δ(pj ) i = pj for all .i ∈ Ob I .
   
Consider now another cone . U, qj : U → Δ(G(j )) j ∈Ob J on .Δ ◦ G, where
.U : I → C is a functor and .qj is a natural transformation for all .j ∈ Ob J . Hence,

for all .u ∈ HomJ (j, t), the following diagram is commutative:

. (3.86)
 
Therefore, for all .i ∈ Ob I we have . Δ(G(u)) i ◦ (qj )i = (qt )i , which comes down
to .G(u) ◦ (qj )i = (qt )i , i.e., the following diagram is commutative:

. (3.87)

Note that the commutativity of (3.87) implies that

   
.U (i), (qj )i : U (i) → G(j ) j ∈Ob J
 
is a cone on G. Since . L, (pj : L → G(j ))j ∈Ob J is the limit of G, for any .i ∈
Ob I , there exists a unique .gi ∈ HomC (U (i), L) such that the following diagram is
commutative for all .j ∈ Ob J :

. (3.88)
3.11 Freyd’s Adjoint Functor Theorem 227

The proof will be finished once we show that .g = (gi )i∈Ob I : U → Δ(L) is a
natural transformation. To this end, let .v ∈ HomI (i, s); we are left to prove the
commutativity of the following diagram:

. (3.89)

In light of Proposition 2.2.14, (1) it is enough to prove that for all .j ∈ Ob J we have
pj ◦ gi = pj ◦ gs ◦ U (v). Indeed, we have
.

(3.88) (3.88)
.pj ◦ gs ◦ U (v) = (qj )s ◦ U (v) = (qj )i = pj ◦ g i ,

where the second equality holds because .qj : U → Δ(G(j )) is a natural transfor-
mation. 


3.11 Freyd’s Adjoint Functor Theorem

Theorem 3.4.4 shows that right (left) adjoints preserve all existing small limits
(colimits). However, in general, small limit/colimit preservation alone does not
guarantee the existence of a left/right adjoint. Indeed, consider the unique functor
.T : Set(⊆) → 1, where .1 is the discrete category with one object and .Set(⊆)

is the category defined in Example 1.2.2, (2). Note that the category .Set(⊆) is
cocomplete by Example 2.2.11 and does not posses a final object as shown in
Example 1.3.10, (6). Therefore, T does not admit a right adjoint as can easily be
seen from Example 3.4.1, (5) while it trivially preserves small colimits.
In this section we will prove that limit/colimit preservation is part of a necessary
and sufficient condition which needs to be fulfilled by a functor in order to admit
a left/right adjoint. Let .G : D → C be a functor, .X ∈ Ob C and let .(X ↓ G) be
the comma category defined in Corollary 1.8.6, (1). We have an  obvious forgetful
functor .U : (X ↓ G) → D defined for any .(f, Y ) ∈ Ob X ↓ G and any morphism
h in .(X ↓ G) as follows:

U (f, Y ) = Y, U (h) = h.
.

Lemma 3.11.1 Let .G : D → C be a functor.

(1) The functor G admits a left adjoint if and only if for all .X ∈ Ob C the comma
category .(X ↓ G) has an initial object.
228 3 Adjoint Functors

(2) The functor G admits a right adjoint if and only if for all .X ∈ Ob C the comma
category .(G ↓ X) has a final object.
Proof
(1) Suppose first that G has a left adjoint .F : C → D and let .θ : HomD (F (−), −)
→ HomC (−, G(−)) be the natural isomorphism corresponding to the adjunc-
tion .F  G. Now consider .X ∈ Ob C and let .η : 1C → GF be the
unit of the adjunction. We will prove that . ηX , F (X) is the initial object
of the category .(X ↓ G). Let .(v, W ) be another object in .(X ↓ G),
i.e., .W ∈ Ob D and .v ∈ HomC (X, G(W )). To this end, we need to
find a unique morphism .f : ηX , F (X) → (v, W ) in .(X ↓ G), i.e.,
a morphism .f ∈ HomD (F (X), W ) such that the following diagram is
commutative:

. (3.90)

Recall from (the proof of) Theorem 3.5.1 that for all .u ∈ HomD (F (X), W )
−1
we have .G(u) ◦ ηX = θX, W (u). Now if we consider .f = θX, W (v) we
obtain

v = θX, W (f ) = G(f ) ◦ ηX ,
.

as desired. The uniqueness of f with this property follows from the bijectivity
of .θX, W .
Conversely, assume now that for each .X ∈ Ob C the comma category
.(X ↓ G) has an initial object, which we denote by .(uX , VX ), where
.VX ∈ Ob D and .uX ∈ HomC (X, G(VX )). Hence, for any .(f, Y ) ∈
Ob (X ↓ G) there exists a unique morphism .h : (uX , VX ) → (f, Y ) in
.(X ↓ G); in other words, for any .f ∈ HomC (X, G(Y )) there exists a

unique morphism .h ∈ HomD (VX , Y ) making the following diagram com-

mute:

. (3.91)

We define a functor .F : C → D on objects by .F (X) = VX for all .X ∈ Ob C.

Consider now .f ∈ HomC (X, X ); then .uX ◦ f ∈ HomC (X, G(F (X )))
3.11 Freyd’s Adjoint Functor Theorem 229

and, using (3.91), we define .F (f ) ∈ HomD (F (X), F (X )) to be the unique

morphism such that

GF (f ) ◦ uX = uX ◦ f.
. (3.92)

Obviously .F (1X ) = 1F (X) for all .X ∈ Ob C . Moreover, if .f ∈ HomC (X, X )

and .f  ∈ HomC (X , X ) then .F (f  ◦ f ) and .F (f  ) ◦ F (f ) are both
morphisms in the comma category .(X ↓ G) from .(uX , F (X)) to .(uX ◦
f  ◦ f, F (X )) so they must be equal as .(uX , F (X)) is the initial object
of .(X ↓ G). Hence F is a functor and furthermore, according to (3.92),
.u : 1C → GF is a natural transformation. To summarize, we have con-

structed a natural transformation .u : 1C → GF such that for any .f ∈

HomC (X, G(Y )) there exists a unique .h ∈ HomD (F (X), Y ) satisfying
.G(h) ◦ uX = f . Now Theorem 3.6.1, (2) implies that F is left adjoint to G, as

desired.
(2) Theorem 3.4.3 shows that .G : D → C admits a right adjoint if and only if
.G
op : Dop → Cop admits a left adjoint. We have already proved in .1) that

.G
op has a left adjoint if and only if for all .X ∈ Ob Cop the comma category

.(X ↓ G ) has an initial object. Furthermore, by Proposition 1.8.10 we have

op

an isomorphism of categories between .(X ↓ Gop )op and .(G ↓ X). Now
using Proposition 1.4.3, (2) we obtain that .(X ↓ Gop ) has an initial object
if and only if its opposite category, namely .(G ↓ X), has a final object.
By putting all the above together we obtain that G has a right adjoint if
and only if for all .X ∈ Ob C the comma category .(G ↓ X) has a final
object.


Definition 3.11.2 Let .C be a category.
(1) A family .(Ki )i∈I of objects of .C, where I is a set, is called a weakly initial set if
j
for any .C ∈ Ob C there exists a morphism .tC ∈ HomC (Kj , C) for some .j ∈ I .
(2) Dually, a family .(Wi )i∈I of objects of .C, where I is a set, is called a weakly
final set if it is a weakly initial set in .Cop ; that is, if for any .C ∈ Ob C there
j
exists a morphism .lC ∈ HomC (C, Wj ) for some .j ∈ I .
Lemma 3.11.3 Let .C be a category. Then:
(1) if .C is complete then .C has an initial object if and only if .C has a weakly initial
set;
(2) if .C is cocomplete then .C has a final object if and only if .C has a weakly final
set.
Proof
(1) Assume first that .C has an initial object I ; then .{I } is obviously a weakly initial
set.
230 3 Adjoint Functors

Conversely, let .(Ki )i∈I be a weakly initial set. As .C is complete and I is a
set we can consider the product . P , (πi : P → Ki )i∈I of the family of objects
.(Ki )i∈I . Notice that for each .C ∈ Ob C there exists at least one morphism

.uC ∈ HomC (P , C) given by the composition for some

.j ∈ I.
Consider now the category J with .Ob J = {P } and .HomJ (P , P ) =
HomC (P , P ) and let .(L, q : L → P ) be the limit of the inclusion functor
.F : J → C.

We will prove that L is the initial object of the category .C. Indeed, for
any .C ∈ Ob C there exists at least one morphism in .HomC (L, C) given
by the composition . Suppose now that we have two
such morphisms f , .g ∈ HomC (L, C) and consider .(E, e : E → L) to
be the equalizer of .(f, g). Since .E ∈ Ob C there exists a morphism .uE ∈
HomC (P , E) given by the composition for some .j ∈ I .
Thus .q ◦ e ◦ uE ∈ HomC (P , P ) and since .(L, q : L → P ) is in particular a
cone on F , the following diagram is commutative:

Thus we have .q ◦ e ◦ uE ◦ q = q = q ◦ 1L and by Proposition 2.2.14, (1) we

get .e ◦ uE ◦ q = 1L . This yields

f = f ◦ 1L = f ◦ e ◦ uE ◦ q = g ◦ e ◦ uE ◦ q = g ◦ 1L = g,
.

where in the third equality we used the fact that .(E, e : E → L) is the equalizer
of .(f, g). We have obtained .f = g and hence L is an initial object of .C.
(2) If .C is cocomplete then .Cop is complete and, as proved in 1), .Cop has an initial
object if and only if .Cop has a weakly initial set. Equivalently, .C has a final
object if and only if .C has a weakly final set.


We are now ready to state the main result of this section:
Theorem 3.11.4 (Freyd’s adjoint functor theorem) Let .G : D → C be a
functor.
(1) If .D is a complete category then G has a left adjoint if and only if G preserves
all small limits and for each .X ∈ Ob C the comma category .(X ↓ G) has a
weakly initial set.
3.11 Freyd’s Adjoint Functor Theorem 231

(2) If .D is a cocomplete category then G has a right adjoint if and only if G

preserves all small colimits and for each .X ∈ Ob C the comma category
.(G ↓ X) has a weakly final set.

Proof
(1) Suppose G has a left adjoint F . Then G is a right adjoint to F and by The-
orem 3.4.4 it preserves limits.
 Moreover,  by (the proof of) Lemma 3.11.1, (1)
for any .X ∈ Ob C, the pair . ηX , F (X) is an initial object in .(X ↓ G), where
.η : 1C → GF is the unit of the adjunction .(F, G).

Assume now that .G : D → C preserves small limits and for each .X ∈ Ob C

the comma category .(X ↓ G) has a weakly initial set. By Corollary 2.6.5, (1)
the category .(X ↓ G) is complete. Thus, from Lemma 3.11.3, (1) we
obtain that .(X ↓ G) has an initial object. The conclusion now follows by
Lemma 3.11.1, (1).
(2) The category .Dop is complete and by applying (1) for the functor .Gop : Dop →
Cop it follows that .Gop has a left adjoint if and only if .Gop preserves small limits
and for each .X ∈ Ob C the comma category .(X ↓ Gop ) has a weakly initial set.
Note also that by Theorem 3.4.3, G has a right adjoint if and only if .Gop has a
left adjoint. Furthermore, Proposition 1.8.10 shows that the comma categories
.(X ↓ G ) and .(G ↓ X)
op op are isomorphic and therefore a weakly initial set in

.(X ↓ G ) is a weakly final set in .(G ↓ X). Putting all of the above together
op

we can conclude that G has a right adjoint if and only if G preserves all small
colimits and for each .X ∈ Ob C the comma category .(G ↓ X) has a weakly
final set, as desired.


Theorem 3.11.4 can be stated in an equivalent form without the use of comma-
categories. To this end, we introduce the following:
Definition 3.11.5 Let .F : C → D be a functor and .D ∈ Ob D. Then:
(1) F satisfies the solution set condition with respect to D if there exists a set .UD
of objects of .C such that for any .C ∈ Ob C and any .f ∈ HomD (D, F (C)),
there exists an object .C  ∈ UD and morphisms .u ∈ HomC (C  , C), .g ∈
HomD (D, F (C  )) such that the following diagram is commutative:

(2) F satisfies the cosolution set condition with respect to D if there exists a set .WD
of objects of .C such that for any .C ∈ Ob C and any .f ∈ HomD (F (C), D),
there exists an object .C  ∈ WD and morphisms .u ∈ HomC (C, C  ), .g ∈
232 3 Adjoint Functors

HomD (F (C  ), D) such that the following diagram is commutative:

Example 3.11.6 If I is a small category and .C is a cocomplete category then the

 .Δ : C → Fun(I, C) satisfies the solution set condition with respect
diagonal functor
to all .G ∈ Ob Fun(I, C) . Indeed, as .G : I → C is a functor and .C is cocomplete, G

has a colimit, which we denote as usual by . colim G, (qi : G(i) → colim G)i∈Ob I .
Consider now the set .UG = {colim G}. If .C ∈ Ob C and .α ∈ HomFun(I, C) (G, ΔC )
 
then by Proposition 2.2.4, (2) we obtain
 that . C, (αi : G(i) → C)i∈Ob I is a
cocone on G. As . colim G, (qi )i∈Ob I is the colimit of G, there exists a unique
.u ∈ HomC (colim G, C) such that for all .i ∈ Ob I we have

u ◦ qi = αi .
. (3.93)

Furthermore, using again Proposition 2.2.4, (2) it follows that .β : G → Δcolim G

defined by .βi = qi for all .i ∈ Ob I is a natural transformation and, moreover, (3.93)
implies

Δ(u) ◦ β = α.
. (3.94)

To summarize, given .C ∈ Ob C and .α ∈ HomFun(I, C) (G, ΔC ), there exists

colim G ∈ UG and two morphisms
.

u ∈ HomC (colim G, C),

. β ∈ HomFun(I, C) (G, Δcolim G )

 .Δ : C →Fun(I, C) satisfies the solution set

such that (3.94) holds. This shows that
condition with respect to all .G ∈ Ob Fun(I, C) .
Similarly, it can be proved that if .C is a complete category then the diagonal
functor .Δ : C → Fun(I, C) satisfies the cosolution set condition. .

Theorem 3.11.7 (Freyd’s Adjoint Functor Theorem) Let .G : D → C be a

functor.
(1) If .D is a complete category then G has a left adjoint if and only if G preserves
all small limits and satisfies the solution set condition with respect to each .X ∈
Ob C.
(2) If .D is a cocomplete category then G has a right adjoint if and only if G
preserves all small colimits and satisfies the cosolution set condition with
respect to each .X ∈ Ob C.
3.11 Freyd’s Adjoint Functor Theorem 233

Proof
(1) Note that G satisfies the solution set condition with respect to X if and only
if the comma category .(X ↓ G) has a weakly initial set and the conclusion
follows by Theorem 3.11.4, (1). Indeed, assume first that G satisfies the solution
set condition, let .X ∈ Ob C and let .UX be the corresponding set as in
Definition 3.11.5, (1). Then .{(h, Y ) | h ∈ HomC (X, G(Y )), Y ∈ UX } is a
weakly initial set in .(X ↓ G). Conversely, if .K = {(t, Z) | Z ∈ Ob D, t ∈
HomC (X, G(Z))} is a weakly initial set in .(X ↓ G) then .UX = {D ∈
Ob D | there exists t ∈ HomC (X, G(Z)) such that (t, Z) ∈ K} is a set
which fulfills the condition in Definition 3.11.5, (1) and therefore G satisfies
the solution set condition.
(2) Similarly, G satisfies the cosolution set condition with respect to X if and only
if the comma category .(G ↓ X) has a weakly final set and Theorem 3.11.4, (2)
leads to the desired conclusion.


We end this section with some applications of the adjoint functor theorem. As
we will see, it can be used to show the existence of various free objects (such as free
groups, free algebras, free modules etc.) without explicitly constructing them.
Examples 3.11.8
(1) Let .U : Grp → Set be the forgetful functor. We will show that U admits a
left adjoint by using Freyd’s adjoint functor theorem. To start with, .Grp is
complete as shown in Example 2.4.3, (1) and U preserves small limits by
Example 2.5.10, (1). In order to conclude that U admits a left adjoint we
need to show that it satisfies the solution set condition with respect to each
.X ∈ Ob Set. To this end, for a given .X ∈ Ob Set we consider the class

.UX of all isomorphism classes of groups of cardinality less than or equal to

.λ = max.{ℵ0 , |X|}, where .ℵ0 denotes the cardinality of the set of all natural

numbers. First we show that .UX is in fact a set. Indeed, recall that there is
only a set of composition laws (in particular, of composition laws which are
group structures) on any given set. Thus, we can conclude that we have, up to
isomorphism, only a set of group structures on any set whose cardinality is at
most .λ. This shows that .UX , being a reunion of sets, is a set itself.
In order to show that .UX satisfies the conditions in Definition 3.11.5, (1)
we start by proving that if .(gx )x∈X is a set of elements of a group G then the
subgroup .GX of G generated by this set has cardinality at most .λ. This can be
easily seen by observing that the following map is surjective:
  
f:
. (X × Z)n → GX , f (x1 , e1 ), (x2 , e2 ), . . . (xn , en ) = gxe11 gxe22 . . . gxenn ,
n∈N
234 3 Adjoint Functors

where the domain of f is the coproduct of the family .{(X × Z)n }n∈N in Set. We
have
 
  
 n
.|GX |   (X × Z)  = |(X × Z)n |
 
n∈N n∈N

ℵ0 , if |X|  ℵ0
=  max{ℵ0 , |X|} = λ.
|X|, if |X|  ℵ0

Consider now .f ∈ HomSet (X, U (G)) for some .G ∈ Ob Grp and let .G
be the subgroup of G generated by the set .(f (x))x∈X . Then, according to the
above discussion, .G is a group of cardinality at most .λ and we can find a
group .H ∈ UX and a group isomorphism .t : H → G . Denote by .i : G → G
the inclusion map and consider .i ◦ t ∈ HomGrp (H, G) and .U (t −1 ) ◦ f ∈
HomSet (X, U (H )) such that the following diagram is commutative:

We can now conclude by Freyd’s adjoint functor theorem that .U : Grp → Set
has a left adjoint.
(2) Let R be a ring, .M ∈ Ob MR , .N ∈ Ob R M and .BilM, N : Ab → Ab
the functor defined in Example 1.5.3, (28). We will show, using Freyd’s
adjoint functor theorem, that .BilM, N admits a left adjoint. By Example 2.5.4,
.BilM, N preserves limits. We are left to show that .BilM, N satisfies the

solution set condition for each .A ∈ Ob Ab. To this end, for a given .A ∈ Ob Ab
we denote by .UA the class of all isomorphism classes of abelian groups of
cardinality less than or equal to .λ = max.{ℵ0 , |A| · |M × N |}. It follows easily,
as in the previous example, that .UA is in fact a set.
We will show that .UA satisfies the conditions in Definition 3.11.5, (1).
Indeed, let .B ∈ Ob Ab and .f ∈ HomAb (A, BilM, N (B)). For all .a ∈ A
let .Pa = {f (a)(m, n) | m ∈ M, n ∈ N} and let .B  be the abelian subgroup
of B generated by the set .P = ∪a∈A Pa . We start by proving that .B  has
cardinality at most .λ. This can be easily seen by observing that the following
map is surjective:

   n
.f: (P × Z)n → B  , f (x1 , k1 ), (x2 , k2 ), . . . (xn , kn ) = ki xi ,
n∈N i=1
3.11 Freyd’s Adjoint Functor Theorem 235

where the domain of f is the coproduct of the family .{(P × Z)n }n∈N in Set. We
have
   
    n 
  n 
.|B |   (P × Z)  =  (∪a∈A Pa ) × Z 
   
n∈N n∈N

ℵ0 , if |A| · |M × N|  ℵ0

|A| · |M × N|, if |A| · |M × N|  ℵ0
 max{ℵ0 , |A| · |M × N|} = λ.

Hence, there exists an abelian group .H ∈ UA and a group isomorphism

.t : H → B  . Now given the way we defined .B  , we obviously have .f ∈
HomAb (A, BilM, N (B  )). Summarizing, if we denote by .i : B  → B the
inclusion map, we have found two maps .i ◦ t ∈ HomAb (H, B) and
−1 )◦f ∈ Hom
.BilM, N (t
Ab (A, BilM, N (H )) such that the following diagram
is commutative:

Now Freyd’s adjoint functor theorem implies that .BilM, N : Ab → Ab has a

left adjoint.
(3) Let I be a small category and .Δ : Grp → Fun(I, Grp) the diagonal functor.
We show that .Δ admits a left adjoint by using Freyd’s adjoint functor theorem.
To start with, .Grp is a complete category by Example 2.4.3, (1) and the diagonal
functor preserves limits as shown in Proposition 3.10.3. We are left to show that
.Δ also fulfills the solution set condition. To this end, let .F : I → Grp be a

functor and let .Xi = F (i), for all .i ∈ Ob I . Consider .λ = | i∈Ob I Xj | and
let .UF be the class of isomorphism classes of groups of cardinality less than or
equal to .λ, where . i∈Ob I Xj is the coproduct in Set of the underlying sets of
the family of groups .(Xi )i∈I . As in the first example, .UF can be easily proved
to be a set. Now let .C ∈ Ob Grp and .ψ : F → ΔC be  a natural transformation.
Consider H to be the subgroup of C generated by . j ∈Ob I ψj (Xj ), where
.ψj : Xj → C, .j ∈ Ob I , are the group morphisms corresponding to the

natural transformation .ψ. In particular, we have .ψ : F → ΔH . Furthermore,

the following holds:
 
|H |  |
. ψj (Xj )|  | Xj | = λ.
j ∈Ob I i∈Ob I
236 3 Adjoint Functors

Therefore we can find a group .Gα ∈ UF and a group isomorphism .u : H →

Gα . Consider now the natural  transformation .τ : F → ΔGα defined for all
.i ∈ Ob I by .τi (x) = u ψi (x) , .x ∈ Xi . It can be easily seen that .τ is a natural

transformation; indeed, if .v ∈ HomI (i, j ) and .x ∈ Xi we have

     
. τj ◦ F (h) (x) = u ψj (F (h)(x)) = u ψi (x) = τi (x),

where the second equality holds because .ψ is a natural transformation. Now

define .t : Gα → C by .t (y) = u−1 (y) for all .y ∈ Gα . Note that for all .x ∈ Xi
we have
   
(t ◦ τi )(x) = t u(ψi (x)) = u−1 u(ψi (x)) = ψi (x).
.

This shows that the following diagram is commutative:

and therefore .UF is indeed a solution set. We can now conclude by Freyd’s
adjoint functor theorem that .Δ has a left adjoint. Furthermore, we can derive
the cocompleteness of .Grp from Theorem 3.10.1, (2). .

3.12 Special Adjoint Functor Theorem

In this section we show that under certain conditions on the domain category, a given
functor admits a left (right) adjoint if and only if it preserves small limits (colimits).
We start with some preparations.
Definition 3.12.1 Let .C be a category.
(1) We say that .C admits a set of generators (or separators) if there exists a set
of objects .{Si | i ∈ I } in .C with the property that for any two morphisms u,
.v ∈ HomC (A, B) such that .u = v there exists a morphism .g ∈ HomC (Sj , A),

for some .j ∈ I , satisfying .u ◦ g = v ◦ g. An object S of .C is called a generator

(or separator) if .{S} is a set of generators.
(2) Dually, we say that .C admits a set of cogenerators (or coseparators) if .Cop
admits a set of separators. More precisely, .C admits a set of cogenerators if
there exists a set of objects .{Ui | i ∈ I } in .C with the property that for any
two morphisms u, .v ∈ HomC (A, B) such that .u = v there exists a morphism
3.12 Special Adjoint Functor Theorem 237

h ∈ HomC (B, Uj ), for some .j ∈ I , satisfying .h ◦ u = h ◦ v. An object U of .C

.

is called a cogenerator (or coseparator) if .{U } is a set of cogenerators.

Examples 3.12.2
(1) In Set, any set with only one element is a generator while any set with at least
two elements set is a cogenerator. Indeed, let u, .v : X → Y be two functions
such that .u = v. Hence there exists an .x0 ∈ X such that .u(x0 ) = v(x0 ) and we
can define a map .g : {} → X by .g() = x0 . Therefore, we have a map g such
that .u(g()) = u(x0 ) = v(x0 ) = v(g()). This shows that .u ◦ g = v ◦ g and the
singleton .{} is a generator in Set. Similarly, one can show that any singleton
set is a generator in Top.
Next we look at cogenerators in Set. With the notations above, let .u(x0 ) = y 
and .v(x0 ) = y  where y, .y  ∈ Y and .y  = y  . We can now define a map
.h : Y → Z by

z , if y = y  ,
h(y) =
.
z , if y =
 y,

where Z is a set with at least two elements and .z = z . This leads to .h(u(x0 )) =
h(y  ) = z = z = h(y  ) = h(v(x
 0 )) and the desired conclusion follows.
(2) In Grp, the group of integers . Z, + is a generator. To this end, let u, .v ∈
HomGrp (G, H ) such that .u = v and consider .g0 ∈ G such that .u(g0 ) = v(g0 ).
Now define .g : Z → G by .g(k) = g0k for all .k ∈ Z, where the group structure on
G is considered to be multiplicative. It can be easily seen that g is a morphism
of groups and moreover .u(g(1)) = v(g(1)), which leads to .u ◦ g = v ◦ g, as
desired.
On the other hand, Grp has no cogenerators. To this end, assume there
exists a cogenerator U in Grp and let S be a simple group whose cardinality is
larger than that of U .12 Let .IdS , .0S ∈ HomGrp (S, S), where .IdS denotes the
identity morphism on S while .0S is defined by .0S (s) = 1S for all .s ∈ S.
As S is a simple group we have .IdS = 0S and since U is assumed to be
a cogenerator in Grp, we have a group homomorphism .f : S → U such
that:

f ◦ IdS = f ◦ 0S .
. (3.95)

Now .ker(f )  S and since S is a simple group we have either .ker(f ) =

S or .ker(f ) = {1S }. The first option is ruled out by (3.95) and
therefore we obtain .ker(f ) = {1S }. This shows that f is injective,
which implies that the cardinality of S is less than or equal to the

12 Such a group is, for instance, the projective special linear group .PSL(2, k), where k is the field
of rational functions over the complex numbers in .|U | variables, i.e., .k = C(Xu )u∈U (see [50,
Theorem 9.46]).
238 3 Adjoint Functors

cardinality of U , contradicting our hypothesis. Therefore, we have

reached a contradiction and we can conclude that Grp has no cogenera-
tors.
(3) In KHaus, the category of compact Hausdorff spaces, the unit interval .[0, 1]
is a cogenerator. Indeed, consider u, .v ∈ HomKHaus (H, K) such that
.u = v. Then, there exists an .h0 ∈ h such that .u(h0 ) = v(h0 ). As K is

in particular a Hausdorff space, there exist two disjoint neighborhoods .U0

and .V0 of .u(h0 ) and .v(h0 ), respectively. Now since any compact Hausdorff
space is normal13 ([39, Theorem 32.3]), we can apply Urysohn’s lemma14
to conclude that there exists a continuous map .f : K → [0, 1] such that
.f (x) = 0 for all .x ∈ U0 and .f (x) = 1 for all .x ∈ V0 . In particular, we
   
have .f u(h0 ) = 0 and .f v(h0 ) = 1, which shows that there exists an
.f ∈ Hom
KHaus (K, [0, 1]) such that .f ◦ u = f ◦ v. This shows that .[0, 1] is
a cogenerator in KHaus.
(4) If J is a small category, then the functor category .Fun(J, Set) has a set of
generators. Indeed, we will show that .{HomJ (j, −) | j ∈ Ob J } is a set
of generators. To this end, let F , .G ∈ Ob Fun(J, Set), i.e., F and G are
functors, and consider two natural transformations .α, .β : F → G such that
.α = β. Since .α = β, there exists .j0 ∈ Ob J and some .x0 ∈ F (j0 ) such

that .αj0 (x0 ) = βj0 (x0 ). Consider now .hx0 : Hom

 J (j0 , −) → F defined for
all .k ∈ Ob J and .f ∈ HomJ (j0 , k) by . hx0 k (f ) = F (f )(x0 ). According
 
.h 0 ∈ Nat HomJ (j0 , −), F and,
to (the proof of) Yoneda’s lemma we have x
 
consequently .α ◦ hx0 , .β ◦ hx0 ∈ Nat HomJ (j0 , −), G . Furthermore, we
have
   
.α ◦ hx0 j (1j0 ) = αj0 ◦ (hx0 )j0 (1j0 ) = αj0 F (1j0 )(x0 )
0
 
= αj0 1F (j0 ) (x0 ) = αj0 (x0 ).
 
A similar computation shows that . β ◦ hx0 j (1j0 ) = βj0 (x0 ). As .αj0 (x0 ) = 
   0 
βj0 (x0 ) we obtain . α ◦ hx0 j (1j0 ) = β ◦ hx0 j (1j0 ). Therefore, .α ◦ hx0 = 
0 0
β ◦ h , as desired.
x 0 .

Proposition 3.12.3 Let .S = {Xi | i ∈ I } be a set of objects of a category .C.

(1) Assume .C has products, S is a set of cogenerators and consider the product
 
. P , (pf )f ∈Hom (C, Xi ) of the family S, where the product consists of as
C
many copies of .Xi as there are morphisms in .HomC (C, Xi ). Then, for any
.C ∈ Ob C, the unique morphism .γC : C → P which makes the following

13 A topological space X is called normal if for each pair U , V of disjoint closed subsets of X there

exist disjoint open subsets of X containing U and V ([39, Section 31]).

14 Urysohn’s lemma: Let X be a normal space and U and V two disjoint subsets of X. If .[a, b] ⊂ R

is a closed interval then there exists a continuous map .f : X → [a, b] such that .f (x) = a for every
.x ∈ U and .f (x) = b for every .x ∈ V ([39, Theorem 33.1]).
3.12 Special Adjoint Functor Theorem 239

diagram commutative for all .i ∈ I and all .f ∈ HomC (C, Xi ):

. (3.96)

is a monomorphism.
(2) Dually, assume .C has coproducts, S is a set of generators and consider
the coproduct . Q, (qf )f ∈Hom (Xi , C) of the family S, where the coproduct
C
consists of as many copies of .Xi as there are morphisms in .HomC (Xi , C).
Then, for any .C ∈ Ob C, the unique morphism .ξC : P → C which makes the
following diagram commutative for all .i ∈ I and all .f ∈ HomC (Xi , C):

. (3.97)

is an epimorphism.
Proof
(1) Consider u, .v ∈ HomC (A, C) such that .γC ◦ u = γC ◦ v. This implies
.pf ◦ γC ◦ u = pf ◦ γC ◦ v for all .i ∈ I and all .f ∈ HomC (C, Xi ).

The commutativity of (3.96) leads to .f ◦ u = f ◦ v for all .i ∈ I and all

.f ∈ HomC (C, Xi ). As S is a set of cogenerators we obtain .u = v, as desired.

(2) Consider u, .v ∈ HomC (C, D) such that .u ◦ ξC = v ◦ ξC . This implies .u ◦ ξC ◦

qf = v ◦ ξC ◦ qf for all .i ∈ I and all .f ∈ HomC (Xi , C). The commutativity
of (3.97) leads to .u ◦ f = v ◦ f for all .i ∈ I and all .f ∈ HomC (Xi , C). As S
is a set of generators we obtain .u = v, as desired.


Having introduced the necessary concepts, we can now state the main result of
this section.
Theorem 3.12.4 (Special Adjoint Functor Theorem) Let .F : C → D be a
functor.
(1) Assume .C is a complete, well-powered category which admits a cogenerating
set. Then F admits a left adjoint if and only if F preserves small limits.
(2) Assume .C is a cocomplete, co-well-powered category which admits a generating
set. Then F admits a right adjoint if and only if F preserves small colimits.
240 3 Adjoint Functors

Proof
(1) If F admits a left adjoint then F is a right adjoint and it preserves limits by
Theorem 3.4.4.
Assume now that F preserves small limits and consider .D ∈ Ob D. In light
of Theorem 3.11.7, (1) it suffices to find a solution set for D. To this end,
consider a coseparating set .S = {Gi | i ∈ I } of .C and denote by
   
.P , pf i∈I
f ∈Hom (C, Gi )
C

the product of the family S, where the product consists of as many copies of .Gi
as there are elements in .HomC (C, Gi ), where C is a fixed object. Similarly, we
consider the product
   
.P  , qf i∈I
f ∈Hom
D (D, F (Gi ))

of the same family S, but this time consisting of as many copies of .Gi as there
are morphisms in .HomD (D, F (Gi )).
Let .UD = {T | T is a subobject of P  }. Note that .UD is in fact a set as .C is
well-powered.
Consider now .g ∈ HomD (D, F (C)). By Proposition 3.12.3, (1) the unique
morphism .αC ∈ HomC (C, P ) such that the following holds for all .i ∈ I and
all .f ∈ HomC (C, Gi )

. (3.98)

is a monomorphism.
   
As . P , pf i∈I is a product, there exists a unique morphism
f ∈Hom (C, Gi )
C

.βC : P → P such that the following diagram is commutative for all .i ∈ I and

all .f ∈ HomC (C, Gi ):

. (3.99)
3.12 Special Adjoint Functor Theorem 241

As .C is a complete category, the pair of morphisms .(αC , βC ) admits a

pullback, which we denote by .(S, μ, γ )

. (3.100)

Furthermore, since .αC is a monomorphism it follows by Proposition 2.1.17, (1)

that .γ is also a monomorphism and therefore we can assume without loss
of generality that .S ∈ UD . Indeed, if .S ∈ / UD , then there exists some
   
.S ∈ UD together with a monomorphism .γ : S → P and an isomorphism
  
.u ∈ HomC (S , S) such that .γ = γ ◦ u. Then, the triple .(S , γ ◦ u, μ ◦ u) is

also a pullback of the pair of morphisms .(α 

.S ∈ UD .
C , βC ) and
  
As F is assumed to be limit preserving, . F (P  ), F (qf ) i∈I
f ∈Hom (D, F (Gi ))
D
is the product of the family .{F (Gi ) | i ∈ I }, where the product consists of as
many copies of .Gi as there are morphisms in .HomD (D, F (Gi )). Hence, we
obtain a unique morphism .λ : D → F (P  ) such that the following diagram is
commutative for all .i ∈ I and all .h ∈ HomD (D, F (Gi )):

. (3.101)

Now
 using again the fact that F is
 limit preserving we obtain, in particular,

that . F (P ), F (pf ) i∈I is the product of the family .{F (Gi ) | i ∈
f ∈Hom (C, Gi )
C
I }, where the product consists of as many copies of .Gi as there are morphisms
in .HomC (C, Gi ). Therefore, for all .i ∈ I and all .f ∈ HomC (C, Gi ) we have

(3.98) 
(3.101) 
F (pf ) ◦ F (αC ) ◦ g = F (pf ◦ αC ) ◦ g = F (f ) ◦ g
. = F qF (f )◦g ◦ λ

(3.99)  
= F pf ◦ βC ◦ λ = F (pf ) ◦ F (βC ◦ λ.

Proposition 2.2.14, (1) implies that .F (αC ) ◦ g = F (βC ◦ λ. As F is
limit preserving it follows that .(F (S), F (μ), F (γ )) is the pullback of the
pair of morphisms .(F (αC ), F (βC )) and we obtain a unique morphism .g  ∈
HomD (D, F (S)) such that .F (μ) ◦ g  = g and .F (γ ) ◦ g  = λ. The complete
242 3 Adjoint Functors

picture is captured in the diagram below:

To conclude, we have proved that for any .C ∈ Ob C and .g ∈ HomD (D, F (C))
there exists some .S ∈ UD together with morphisms .μ ∈ HomC (S, C) and
 
.g ∈ HomD (D, F (S)) such that .F (μ) ◦ g = g. The desired conclusion now

follows by Theorem 3.11.7, (1).

(2) To start with, note that the category .Cop is complete, well-powered and admits
a coseparating set. By applying .1) for the functor .F op : Cop → Dop , we obtain
that .F op admits a left adjoint if and only if .F op preserves small limits. In light
of Lemma 2.5.2 and Theorem 3.4.3 it follows that F admits a right adjoint if
and only if F preserves small colimits, as desired.


Corollary 3.12.5 Let .C be a category.
(1) If .C is complete, well-powered and admits a coseparating set, then .C is also
cocomplete.
(2) If .C is cocomplete, co-well-powered and admits a separating set, then .C is also
complete.
Proof
(1) Consider the diagonal functor .Δ : C → Fun(I, C) which preserves small limits,
as proved in Proposition 3.10.3. Since the conditions in Theorem 3.12.4, (1) are
fulfilled it follows that .Δ admits a left adjoint. Now Theorem 3.10.1, (2) implies
that .C is cocomplete.
(2) The diagonal functor .Δ : C → Fun(I, C) also preserves small colimits, as
proved in Proposition 3.10.3. Since the conditions in Theorem 3.12.4, (2) are
fulfilled it follows that .Δ admits a right adjoint. Now Theorem 3.10.1, (1)
implies that .C is complete.


3.12 Special Adjoint Functor Theorem 243

Examples 3.12.6
(1) Let .U : KHaus → Top be the forgetful functor. Recall that KHaus, the
category KHaus of compact Hausdorff spaces, is well-powered, as shown in
Example 1.3.16, and has a cogenerator by Example 3.12.2, (3). Furthermore,
KHaus has products (Example 2.1.5, (4)) and equalizers (Example 2.1.10, (4)),
which shows that is complete by Theorem 2.4.2. As both products and
equalizers are constructed as in Top we can conclude by the Special Adjoint
Functor Theorem that U has a left adjoint.
(2) Let .F : K → J be a functor between small categories and consider the induced
functor .F  : Fun(J, Set) → Fun(K, Set) defined in (1.37)

F  (G) = GF, F  (ψ)k = ψF (k)

.

for all functors G, .H : J → Set and all natural transformations .ψ : G →

H . We will use the Special Adjoint Functor Theorem to prove that .F 
has a right adjoint. Indeed, note that the category .Fun(J, Set) has a set of
generators, as proved in Example 3.12.2, (4). Furthermore, Set is cocomplete by
Example 2.4.4 while Example 2.7.4 shows that .Fun(J, Set) is also cocomplete.
As Set is cocomplete, it follows from Proposition 2.7.5 that a morphism
.ψ : F → G in .Fun(J, Set) (i.e., a natural transformation) is an epimorphism if

and only if each .ψj : F (j ) → G(j ) is an epimorphism in Set for all .j ∈ Ob J .

Set is co-well-powered by Example 1.3.16, which shows that each .F (j ) has
only a set of quotients and since J is small we can conclude that F has a
set of quotients. Therefore, .Fun(J, Set) is co-well-powered. We are left to
 end, let .G : I → Fun(J, Set) and
show that .F 
 preserves colimits. To this
denote by . H, (qi : G(i) → H )i∈Ob I its colimit, where .H : J → Set is
a functor and .qi is a natural transformation
 for all .i ∈Ob I . The proof will

  
be finished once we show that . F (H ), F (qi ) i∈Ob I is the colimit of the
functor .F  ◦ G : I → Fun(K, Set). In light of Theorem 2.7.2 it is enough
 
to prove that for any .k ∈ Ob K the pair . F  (H )(k), F  (qi )k i∈Ob I is the
colimit of .(F  ◦ G)k : I → Set, where .(F  ◦ G)k denotes the induced functor
as defined in (2.40).  
 
We start by showing that . F  (H )(k), F  (qi )k i∈Ob I is a cocone on
.(F ◦ G)k : I → Set, i.e., for all .u ∈ HomI (l, t) the following diagram is


commutative:

. (3.102)
244 3 Adjoint Functors

 j 
Indeed, recall from Theorem 2.7.2 that . H (j ), (qi : Gj (i) → H (j ))i∈Ob I is
the colimit of the induced functor .Gj : I → Set and in particular a cocone on
j
.Gj , where .q = (qi )j for all .i ∈ Ob I and .j ∈ Ob J . Therefore, the following
i
diagram is commutative:

. (3.103)

Now note that since we have

F  (H )(k) = H (F (k)),
.

F  (ql )k = (ql )F (k) = qlF (k) ,

 
(F  ◦ G)k (u) = (F  ◦ G)(u) k = F  (G(u))k = G(u)F (k) ,
(F  ◦ G)k (l) = (F  ◦ G)(l)(k) = F  (G(l))(k) = G(l)(F (k)) = GF (k) (l),

it can be easily seen that the commutativity of (3.103) implies the commutativity
of (3.102).  
Consider now another cocone . Xk , (sik : (F  ◦ G)k (i) → Xk )i∈Ob I on the
functor .(F  ◦ G)k : I → Set. Hence, for all .u ∈ HomI (l, t) the following
diagram is commutative:

As .(F  ◦ G)k (u) = G(u)F (k) and .(F  ◦ G)k (i) = GF (k) (i), the commu-
tativity of the above diagram comes down to .G(u)F (k) ◦ slk = stk . There-
 
fore, . Xk , (sik : GF (k) (i) → Xk )i∈Ob I is a cocone on the induced functor
 F (k)
.GF (k) : I → Set. Now recall that the pair . H (F (k)), (q : GF (k) (i) →
 i
H (F (k)))i∈Ob I is the colimit of the functor .GF (k) . Thus, we have a unique
morphism .f ∈ HomSet (H (F (k)), Xk ) such that the following diagram is
3.13 Representable Functors Revisited 245

commutative for all .i ∈ Ob I :

To conclude, there exists a unique morphism .f ∈ HomSet (H (F (k)), Xk ) such

that for all .i ∈ Ob I we have

f ◦ F  (qi )k = sik .
.

   
This shows that . F  (H )(k), F  (qi )k i∈Ob I is the colimit of the functor
.F ◦ G : I → Fun(K, Set), as desired. .


3.13 Representable Functors Revisited

This section collects new representability criteria for certain classes of functors. The
first one refers to limit preserving functors.
Theorem 3.13.1 (Representability criterion) Let .C be a complete category and
F : C → Set a functor such that
.

(1) F preserves limits;

(2) there exists a set I , a family of objects .(Xi )i∈I in .C and for each .i ∈ I an
element .fi ∈ F (Xi ) such that for any .Y ∈ Ob C and any .g ∈ F (Y ) there exists
a morphism .ϕ ∈ HomC (Xi0 , Y ) for some .i0 ∈ I such that .F (ϕ)(fi0 ) = g.
Then F is representable, i.e., there exists .X ∈ Ob C and a natural isomorphism
.F ∼ = HomC (X, −).
 
Proof To start with, note that condition .2) implies that .{ fi∗ , Xi i∈I } is a weakly
 
initial set in the comma category . {} ↓ F , where .{} denotes a singleton set and
∗
.f ∈ Hom
∗
i Set ({}, F (Xi )) is defined by .fi () = fi ∈ F (Xi ) for all
 .i ∈ I . Since .C
is a complete category and F preserves limits, the comma category . {} ↓F is also
complete by Corollary 2.6.5, (1). Now Lemma 3.11.3, (1) implies that . {} ↓ F
has an initial object and therefore F is representable by Proposition 1.8.8. 

We state, for the sake of completeness, the contravariant version of the repre-
sentability criterion:
Theorem 3.13.2 (Representability criterion for contravariant functors) Let .D
be a cocomplete category and .G : D → Set a contravariant functor such that
246 3 Adjoint Functors

(1) G turns colimits into limits;

(2) there exists a set I , a family of objects .(Xi )i∈I in .D and for each .i ∈ I an
element .fi ∈ G(Xi ) such that for any .Y ∈ Ob D and any .g ∈ G(Y ) there exists
a morphism .ϕ ∈ HomD (Y, Xi0 ) for some .i0 ∈ I such that .G(ϕ)(fi0 ) = g.
Then F is representable, i.e., there exists .X ∈ Ob C and a natural isomorphism
F ∼
. = HomC (−, X).
Proof Consider the covariant functor .F = G ◦ ODop : Dop → Set. Note that .Dop
is a complete category and F preserves limits. Furthermore, condition (2) can be
rephrased as follows: there exists a set I and a family of objects .(Xi )i∈I in .Dop
and for each .i ∈ I an element .fi ∈ F (Xi ) such that for any .Y ∈ Ob Dop and any
.g ∈ F (Y ) there exists a morphism .ϕ
op ∈ Hom op (X , Y ) such that .F (ϕ)(f ) = g.
D i i
Thus, F fulfills all conditions in Theorem 3.13.1 and therefore F is representable.
This shows that G is a representable contravariant functor, as desired. 

Our next result relates representability to adjoint functors.
Theorem 3.13.3 Let .F : C → D and .G : D → C be two functors. Then:
(1) G has a left adjoint if and only if all functors

. HomC (C, G(−)) : D → Set

are representable for all .C ∈ Ob C;

(2) F has a right adjoint if and only if all contravariant functors

HomD (F (−), D) : C → Set

.

are representable for all .D ∈ Ob D.

Proof
(1) Assume first that .F : C → D is left adjoint of G. Then, for all .C ∈
Ob C and .D ∈ Ob D we have a bijective map .θC, D : HomD (F (C), D) →
HomC (C, G(D)) which is natural in both variables. In particular, naturality
in the second variable implies that for all .C ∈ Ob C, we have a natural iso-
morphism between the functors .HomD (F (C), −) and .HomC (C, G(−)). This
shows precisely that the functor .HomC (C, G(−)) : D → Set is representable
and its representing object is .F (C).
Conversely, suppose that the functors .HomC (C, G(−)) : D → Set are
representable for all .C ∈ Ob C. We will show that for all .C ∈ Ob C, the
comma category .(C ↓ G) has an initial object. The conclusion will follow
by Lemma 3.11.1, (1). Indeed, as .HomC (C, G(−)) : D → Set is representable,
there exists .XC ∈ Ob D and a natural isomorphism .α : HomD (XC , −) →

HomC (C, G(−)). We will prove that the pair . αXC (1XC ), XC is the initial
object of the comma category .(C ↓ G). To this end, given .(g, D) ∈ Ob (C ↓
−1
G) we define .h ∈ HomD (XC , D) by .h = αD (g). The proof will be finished
3.13 Representable Functors Revisited 247

once we show that h is the unique morphism such that .G(h) ◦ αXC (1XC ) = g.
Recall that .α is a natural transformation and therefore the following diagram is
commutative:

The commutativity of the above diagram applied to the morphism .1XC ∈

HomD (XC , XC ) yields .G(h) ◦ αXC (1XC ) = αD (h), which comes down to

G(h) ◦ αXC (1XC ) = g,

.

as desired. We are left to show that h is the unique morphism with this
property. Indeed, assume there exists an .h ∈ HomD (XC , D) such that .G(h) ◦
αXC (1XC ) = g. Using again the naturality of .α this time for the morphism .h
yields .G(h) ◦ αXC (1XC ) = αD (h). It follows that .αD (h) = g and since .αD is
bijective we obtain .h = h, which finishes the proof.
(2) The second part follows in a similar manner. Indeed, if G is right adjoint
to F then for all .C ∈ Ob C and .D ∈ Ob D we have a bijective map
.θC, D : HomD (F (C), D) → HomC (C, G(D)) which is natural in both

variables. In particular, naturality in the first variable implies that for all .D ∈
Ob D, we have a natural isomorphism between the functors .HomD (F (−), D)
and .HomC (−, G(D)). Hence the functor .HomD (F (−), D) : C → Set is
representable and its representing object is .G(D).
Conversely, as the functors .HomD (F (−), D) : C → Set are representable
for all .D ∈ Ob D, there exists .YD ∈ Ob C and a natural isomorphism
.β : HomC (−, YD ) → HomD (F (−), D). Then, it can be easily proved that
 
. YD , βYD (1YD ) is the final object of the category .(F ↓ D). The conclusion

now follows by Lemma 3.11.1, (2).



We end this section with an example which shows the existence of an algebraic
object, namely the tensor product of modules, without explicitly constructing it.
Example 3.13.4 Let .BilM, N : Ab → Ab be the functor defined in Exam-
ple 1.5.3, (28). It was proved in Example 3.11.8, (2) that .BilM, N admits a left
adjoint. Now Theorem 3.13.3, (1) shows that the functor .HomAb (A, BilM, N (−)) :
Ab → Set is representable for all abelian groups A. In particular, using
Proposition 1.7.7, the representability of the functor .HomAb ({0}, BilM, N (−))
implies the existence of a representing pair, denoted by .(M ⊗R N, i). Hence,
.M ⊗R N is an abelian group and .i ∈ BilM, N (M ⊗R N ), i.e., .i : M ×N → M ⊗R N

is a bilinear map, such that for any other pair .(A, f ), where A is an abelian group
248 3 Adjoint Functors

and .f : M × N → A is a bilinear map, there exists a unique group homomorphism

g : M ⊗R N → A which makes the following diagram commutative:
.

This means precisely that .M ⊗R N is the tensor product of the R-modules M and
N (see [45, Definition 1.5]). .

It is worth to point out that the various adjoint functor theorems proved in this
chapter have many notable applications, most of them exceeding the purpose of
this introductory book. For instance, the following corollary to Freyd’s theorem
is presumedly “more widely known than the theorem itself ” as stated in [7]: any
functor between varieties of algebras which respects underlying sets has a left
adjoint ([7, Corollary 8.17]). For precise definitions and more details we refer the
reader to [7].
Furthermore, the adjoint functor theorems have been extended to various settings
allowing for important applications. We only mention here the case of triangulated
categories and an important consequence in algebraic geometry. A criterion for
a functor between triangulated categories to admit a right adjoint was proved by
building on a version of the representability theorem for triangulated categories.
More precisely, a triangulated functor between triangulated categories satisfying
certain technical conditions which commutes with arbitrary coproducts admits a
right adjoint. A notable application of the aforementioned criterion on the existence
of adjoints for triangulated functors is the Grothendieck duality theorem proved by
A. Neeman (see [42] for further details).

3.14 Exercises

3.1 Prove that the forgetful functor U : Field → Ring does not admit a right or a
left adjoint.
3.2 Decide if the forgetful functor U : A → Set admits a right adjoint, where A
is Grp, Ring or R M.
3.3 Let R be a commutative ring. Show that the forgetful functor U : R M → Ab
has both a left and a right adjoint.
3.4 If R is a commutative ring, show that the forgetful functor F : AlgR → R M
(forgetting the multiplicative structure) has a left adjoint.
3.5 Decide if the inclusion functor I : Ringc → Ring has a left or a right adjoint.
3.6 Let F : C → D and G : D → C be functors such that F  G. Prove that F
preserves epimorphisms and G preserves monomorphisms.
3.14 Exercises 249

3.7 Let F : C → D and G : D → C be functors such that F  G. Show that if

GF is fully faithful then F is fully faithful.
3.8 Let F : C → D and G : D → C be functors such that F  G, and let
η : 1C → GF and ε : F G → 1D be the unit and respectively the counit of
this adjunction. Then the following are equivalent:
(a) F (ηC ) is an isomorphism for all C ∈ Ob C;
(b) GF (ηC ) = ηGF (C) for all C ∈ Ob C;
(c) εF (C) is an isomorphism for all C ∈ Ob C;
(d) G(εF (C) ) is an isomorphism for all C ∈ Ob C.
3.9 Let H : C → D and F , G : D → C be functors such that F  H and H  G.
If η : 1D → H F is the unit of the adjunction F  H and ε : H G → 1D is the
counit of the adjunction H  G then ηD : D → H F (D) is an epimorphism
for every D ∈ Ob D if and only if εD : H G(D) → D is a monomorphism for
every D ∈ Ob D.
3.10 Let F : C → D and G : D → C be functors such that F  G, and let
η : 1C → GF and ε : F G → 1D be the unit and respectively the counit of
this adjunction. Show that the categories Iso(C) and Iso(D) are equivalent,
where Iso(C) denotes the full subcategory of C consisting of those objects
C ∈ Ob C for which ηC is an isomorphism and Iso(D) is the full subcategory
of D consisting of those objects D ∈ Ob D for which εD is an isomorphism.
3.11 Let F , F  : C → D, G, G : D → C be functors such that F  G and
F   G and suppose that I is a small category.
(a) If T : I → D is a functor which admits a limit and β : G → G is a
natural transformation such that βT (i) is an isomorphism in C for every
i ∈ Ob I , then βlim T is also an isomorphism in C.
(b) If H : I → C is a functor which admits a colimit and α : F → F  is a
natural transformation such that αH (i) is an isomorphism in D for every
i ∈ Ob I , then αcolim H is also an isomorphism in D.
3.12 Let OB : Cat → Set be the objects functor defined as follows for all small
categories C, D and all functors F : C → D:
(a) OB(C) = Ob C;
(b) OB(F ) = F : Ob C → Ob D.
Show that OB has both a left and a right adjoint.
3.13 Show that the inclusion functor I : Poset → PreOrd has a left adjoint.
3.14 Let X be a set and consider the cartesian product functor X × − : Set → Set.
Find the sets X for which the functor X × − admits a left adjoint.
3.15 Show that the inclusion functor I : Haus → Top does not admit a right
adjoint.
3.16 Let F : C → D and G : D → C be functors such that F  G.
(a) If the functor F  : C → D is naturally isomorphic to F then F   G.
(b) If the functor G : D → C is naturally isomorphic to G then F  G .
250 3 Adjoint Functors

3.17 Let F : C → D be an equivalence of categories. Prove that

(a) f ∈ HomC (C, C  ) is a monomorphism if and only if F (f ) is a
monomorphism;
(b) f ∈ HomC (C, C  ) is an epimorphism if and only if F (f ) is an
epimorphism;
(c) f ∈ HomC (C, C  ) is an isomorphism if and only if F (f ) is an
isomorphism.
3.18 Let F : C → D be a fully faithful functor. Show that C is equivalent to a full
subcategory of D.
3.19 Let R, S be two rings. Show that the product category R M × S M is equivalent
to the category R×S M.
3.20 Show that the ring R is a generator in the category R M of left R-modules.
3.21 Let C be a pointed category which admits (co)products. Then C has a
(co)generator if and only if it has a set of (co)generators.
3.22 Let F : C → D and G : D → C be functors such that F  G. Prove that if
G is faithful and S is a generator in C then F (S) is a generator in D.
3.23 Prove that an object S in a category C is a generator if and only if the functor
HomC (S, −) : C → Set is faithful. State and prove the dual.
3.24 Let C be a category and S a class of morphisms in C such that the localization
category CS exists. Then the localization of Cop byS op exists too and we have
op op
an isomorphism of categories between CS op and CS , where S op denotes
the class of all opposites of morphisms in S.
3.25 Let F : C → D and G : D → C be two functors such that F  G. Prove
that
(a) if F preserves monomorphisms then G preserves injective objects;
(b) if G preserves epimorphisms then F preserves projective objects.
Chapter 4
Solutions to Selected Exercises

4.1 Chapter 1

1.4 (a) Let u denote the unique morphism in HomC (X, C). Then m ◦ u ∈
HomC (X, X) = {1X } and therefore we have

m ◦ u = 1X .
. (4.1)

(4.1)
Furthermore, we have m ◦ u ◦ m = m and since m is a monomorphism
we obtain u ◦ m = 1C .
(b) This claim follows by the duality principle; indeed, applying a) for the
dual category Cop yields the desired claim. 

1.5 (b) Let u, v ∈ HomC (E, A) such that f ◦ u = f ◦ v. This implies g ◦ f ◦ u =
g ◦ f ◦ v and since g ◦ f is a monomorphism, we obtain u = v, as desired.
(c) Consider t, w ∈ HomC (C, D) such that t ◦ g = w ◦ g. This implies
t ◦ g ◦ f = w ◦ g ◦ f and since g ◦ f is an epimorphism, we obtain
t = w. 

1.6 (a) Let f ∈ HomC (A, B) be a split monomorphism and denote by t its left
inverse. If g1 , g2 ∈ HomC (A , A) such that f ◦ g1 = f ◦ g2 , then by
composing on the left with t we obtain g1 = g2 , which shows that f is a
monomorphism.
For the converse consider the group morphism f : Z2 → Z4 defined by
f (0) = 0̂ and f (1) = 2̂, where x and x̂ denote the residue classes modulo
2 and 4, respectively. It can be easily seen that f is a monomorphism;
to this end, let g1 , g2 ∈ HomGrp (G, Z2 ) such that f ◦ g1 = f ◦ g2 . If
there exists some x0 ∈ G such that g1 (x0 ) = g2 (x0 ) then we can assume
 that g1 (x0 ) = 0 and g2 (x0 ) = 1. This implies
without loss of generality
0̂ = f (0) = f g1 (x0 ) = f g2 (x0 ) = f (1) = 2̂, which is an obvious
contradiction. Therefore, f is a monomorphism. We are left to show that
f is not a split monomorphism. Indeed, assume that there exists a group

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 251
A. Agore, A First Course in Category Theory, Universitext,
https://doi.org/10.1007/978-3-031-42899-9_4
252 4 Solutions to Selected Exercises

morphism g : Z4 → Z2 such that g ◦ f = 1Z2 . This implies

   
g(0̂) = g f (0) = 0,
. g(2̂) = g f (1) = 1

and g(1̂) ∈ {0, 1}. If g(1̂) = 0, it follows that g(2̂) = 0, which is a

contradiction. Thus, we must have g(1̂) = 1, which leads to g(2̂) = 2 =
0, which is another contradiction. To conclude, we have proved that f is
not a split monomorphism, as desired.
(b) Assume f ∈ HomC (A, B) is an epimorphism and split monomorphism.
In particular, there exists a t ∈ HomC (B, A) such that t ◦ f = 1A . This
implies that f ◦ t ◦ f = f = 1B ◦ f and since f is an epimorphism we
obtain f ◦ t = 1B . Hence, f is an isomorphism. The converse is obvious.


1.7 Assume first that f is a strong epimorphism and a monomorphism. Then, the
following commutative square

f
A B

1A 1B

A B
. f

admits a unique g ∈ HomC (B, A) such that g ◦ f = 1A and f ◦ g = 1B . This

shows that f is an isomorphism.
Conversely, suppose now that f is an isomorphism and consider the
following commutative square:

f
A B h f u g,
g h

C D
. u

where u ∈ HomC (C, D) is a monomorphism. It can be easily seen that v =

g ◦ f −1 : B → C is the unique morphism in C such that v ◦ f = g and
u ◦ v = h. We have proved that (a) is equivalent to (b). Furthermore, as f is
an isomorphism in C if and only if f op is an isomorphism in Cop , it follows
that (a) is also equivalent to (c). 

1.8 Recall from Example 1.5.3, (8) that functors between categories associ-
ated to groups (in the sense of Example 1.2.2, (3)) are nothing but group
homomorphisms between the given groups. Furthermore, the opposite of a
category associated to a group is precisely the category associated to the
4.1 Chapter 1 253

opposite group1 . Furthermore, for any group G, we have a group isomorphism

θ : G → Gop given by θ (g) = g −1 for all g ∈ G. This provides the desired
isomorphism of categories between the category associated to the group G
and its opposite.
For the case of monoids, it will suffice to consider the monoid described in
Example 2.1.16, (10) as a counterexample. 

1.9 (a) The forgetful functor U : Ringc → Set does not preserve epimorphisms.
Indeed, recall from Example 1.3.2, (5) that the inclusion i : Z → Q is an
epimorphism in Ringc but not in Set as it is not a surjective map.
Consider now the abelianization functor F : Grp → Ab introduced in
Example 1.5.3, (23). Let A5 be the alternating group of degree 5 generated
by the 3-cycles (123), (124), (125). Then the inclusion i : C3 → A5 is
an injective map and therefore a monomorphism in Grp, where C3 is the
cyclic group generated by the 3-cycle (123). Now recall that F (An ) = {1}
for all n  5, while F (Cn ) = Cn for all n ∈ N∗ . Therefore F (i) : C3 →
{1} is obviously not a monomorphism in Ab.
(b) Let C be a category which is not a groupoid. Then, the inclusion
functor I : Cgrp → C obviously reflects isomorphisms without being
full, where Cgrp is the core groupoid of the category C as defined in
Example 1.3.6, (2). 

1.10 Consider the following two categories B and C:

1A 1B

f
A B

1C 1D 1C 1D

u v
.
C D C D

(a) The functor F : B → C defined below is full while the morphism v and
the object C  are not in its image:

F (A) = C, F (B) = D,
.

F (1A ) = 1C , F (1B ) = 1D , F (f ) = u.

1 The opposite of a group (G, ·) is another group denoted by (Gop , ·op ), where Gop = G and the

group structure is given by g ·op g  = g  · g, for all g, g  ∈ G.

254 4 Solutions to Selected Exercises

(b) The functor G : C → B defined as follows:

G(C) = G(C  ) = A, G(D) = G(D  ) = B,

.

F (1C ) = F (1C  ) = 1A , F (1D ) = F (1D  ) = 1B , G(u) = G(v) = f

is faithful although we have G(u) = G(v) = f and G(C) = G(C  ). 


1.11 Assume there exists a functor F : Grp → Grp such that F (G) = Z(G)
for all groups G, where Z(G) is the center of G. Consider the symmetric
group S3 on three letters and let H be the cyclic subgroup of S3 generated
by the 3-cycle x = (123). It can be easily seen that H has order 3, H is a
normal subgroup of S3 and the quotient S3 /H has order 2 and is isomorphic
to Z2 . Furthermore, we consider Z2 to be the subgroup of S3 generated by the
transposition y = (23). Consider now the following group morphisms:

i
. 2 S3 S 3 /H

where i : Z2 → S3 and π : S3 → S3 /H are the inclusion map and respectively

the quotient map. As i(Z2 ) ∩ H = {1S3 } it follows that π ◦ i is a group
isomorphism. In light of Proposition 1.6.9, (1) the following composition is a
group isomorphism too:

Since Z(Z2 ) = Z2 and Z(S3 ) = {1} it follows that the composition below is
an isomorphism

which is an obvious contradiction. Therefore, there is no functor F : Grp →

Grp such that F (G) = Z(G) for all groups G. 

1.12 Consider the functor F : Top → Set defined as follows for all topological
spaces X, Y and f ∈ HomTop (X, Y ) :

F (X) = {Xi | Xi connected component of X},

.

i ,
F (f ) : F (X) → F (Y ), F (f )(Xi ) = X

i denotes the connected component of Y which contains f (Xi ). Since

where X
the image of a connected space under a continuous map is connected ([39,
i is indeed a connected component of F (Y ) and therefore
Theorem 23.5]), X
4.1 Chapter 1 255

F (f ) is well-defined. Furthermore, F (1X ) is the identity on F (X) and if f ∈

HomTop  (X, Y), g ∈ HomTop (Y, Z) and Xi is a connected component of X
then g f (Xi ) is also connected. This shows that F (g ◦ f ) = F (g) ◦ F (f )
and therefore F is indeed a functor. 

1.14 (a) Let f , f  ∈ HomC (A, B) such that F (f ) = F (f  ). Then we also have
GF (f ) = GF (f  ) and since GF is faithful we obtain f = f  . This
shows that F is faithful.  
(b) Consider now g ∈ HomD F (A), F (B) . Then:
 
G(g) ∈ HomD GF (A), GF (B)
.

 
and since GF is full there exists some f ∈ HomC A, B such that
G(g) = GF (f ). As G is faithful we obtain g = F (f ), which shows
that F is full. 

1.15 (a) Let G, G : I → C be two functors; we need to show that the following
mapping is bijective:

Nat(G, G ) → Nat(F G, F G ), α → F α,
.

where α : G → G is a natural transformation and F α is the whiskering

of α on the right by F as defined in Example 1.7.2, (7). Suppose first that
α, α  : G → G are natural transformations such that F α = F α  . This
implies that for all i ∈ Ob I we have F (αi ) = F (αi ) and since F is
faithful we obtain αi = αi . This shows that α = α  and therefore F is
faithful.
Consider now γ ∈ Nat(F G, F G ); for all i ∈ Ob I we have

γi ∈ HomD (F G(i), F G (i))

.

and since F is fully faithful there exists a unique βi ∈ HomD (G(i), G (i))
 that γi = F (βi ). We are left to show that the family of morphisms
such
βi i∈Ob I form a natural transformation between the functors G and G .
To this end, consider f ∈ HomI (i, j ). Since γ : F G → F G is a natural
transformation, the following diagram is commutative:

.
256 4 Solutions to Selected Exercises

As γi = F (βi ) for all i ∈ Ob I , the commutativity of the above

diagram leads to F G (f ) ◦ F (βi ) = F (βj ) ◦ F G(f ) and since F is
faithful we obtain G (f ) ◦ βi = βj ◦ G(f ), i.e., the following diagram is
commutative:

This shows that β : G → G is indeed a natural transformation and,

moreover, we have F (β) = γ .
(b) As proved above, F is fully faithful and the desired conclusion now
follows by Proposition 1.6.9, (2). 

1.17 If C ∈ Ob C, then the naturality of β : H → I applied to the morphism
γC ∈ HomD (G(C), P (C)) yields the following commutative diagram:

. (4.2)

Therefore, for all C ∈ Ob C, we have

  (1.18)  
.(δ ◦ β) ∗ (γ ◦ α) C = (δ ◦ β)P (C) ◦ H (γ ◦ α)C
= δP (C) ◦ βP (C) ◦ H (γC ) ◦ H (αC )
(4.2)
= δP (C) ◦ I (γC ) ◦ βG(C) ◦ H (αC )
(1.18)
= (δ ∗ γ )C ◦ (β ∗ α)C
 
= (δ ∗ γ ) ◦ (β ∗ α) C .



4.1 Chapter 1 257

1.20 Let {vi | i = 1, n} be the basis of V over the field K and denote by {vi }i=1, n
the dual basis of V ∗ . Define two natural transformations as follows:

α : V ∗ ⊗ − → HomK M (V , −), αU (f ⊗ u)(v) = f (v)u,

.


n
β : HomK M (V , −) → V ∗ ⊗ −, βU (t) = vi∗ ⊗ t (vi ),
i=1

for all U ∈ Ob K M, u ∈ U , v ∈ V , f ∈ V ∗ and t ∈ HomK M (V , U ). It is

straightforward to check that α and β are indeed natural transformations and
that for all U ∈ Ob K M we have αU ◦ βU = 1HomK M (V, U) and βU ◦ αU =
1V ∗ ⊗ U . 

1.21 If C is the empty category then the empty functor from the empty category
to Set is obviously not representable. For non-empty categories C, consider
A ∈ Ob Set such that |A| = 2 and define F : C → Set as follows for all C,
D ∈ Ob C and f ∈ HomC (C, D):

F (C) = A,
. F (f ) = 1A .

Assume F is representable and let (X, x) be the representing pair, where

X ∈ Ob C and x ∈ F (X) = A. Consider now another pair (X , x  ) where
x  ∈ F (X ) = A and x  = x. By Proposition 1.7.7 there exists a unique
f ∈ HomC (X, X ) such that F (f )(x) = x  . On the other hand, we have
F (f )(x) = 1A (x) = x and x = x  . We have reached a contradiction and
therefore F is notrepresentable. 

1.25 Consider ψ : Fun I, Fun(J, C) → Fun(I × J, C) defined as follows for all
functors F , G : I → Fun(J, C) and all natural transformations η : F → G:

ψ(F ) = FF , ψ(η)(i, j ) = (ηi )j , for all (i, j ) ∈ Ob (I × J ),

.

where FF : I × J → C is the functor defined by

.FF (i, j ) = F (i)(j ), FF (u, v) = F (k)(v) ◦ F (u)j ,

 
for all (i, j ), (k, l) ∈ Ob (I × J ) and (u, v) ∈ HomI ×J (i, j ), (k, l) .

It can be easily checked that FF is indeed a functor. We will show that ψ is

an isomorphism of categories. To start with, we first show that ψ is indeed a
functor. We have
 
ψ(1F )(i, j ) = (1F )i j = (1F (i) )j = 1F (i)(j ) ,
.
258 4 Solutions to Selected Exercises

which shows that ψ respects identities. Furthermore, for any two natural
transformations η : F → G, ζ : G → H , where F , G, H : I → Fun(J, C)
are functors, and any (i, j ) ∈ Ob (I × J ), we have
 
ψ(ζ ◦ η)(i, j ) = ζ ◦ ηi j = (ζi ◦ ηi )j = (ζi )j ◦ (ηi )j
.

= ψ(ζ )(i, j ) ◦ ψ(η)(i, j )

 
= ψ(ζ ) ◦ ψ(η) (i, j ).

This shows that ψ respects compositions  as well andis therefore a functor.

Consider now ϕ : Fun(I × J, C) → Fun I, Fun(J, C) defined as follows for
all functors F , G : I × J → C and all natural transformations η : F → G:
 
ϕ(F ) = GF ,
. ϕ(η)i j = η(i, j ) , for all i ∈ Ob I, j ∈ Ob J,

where GF : I → Fun(J, C) is the functor defined by

GF (i) = F (i, −), GF (f ) = F (f, 1j ),

.

for all i, l ∈ Ob I, j ∈ Ob J and f ∈ HomJ (i, l).

It can be easily checked by a straightforward computation that both GF and ϕ

are functors. We will show that ψ and ϕ are inverses to each other. Indeed, for
all functors F : I → Fun(J, C) and all i ∈ Ob I , j ∈ Ob J we have
 
ϕ ψ(F ) (i)(j ) = FF (i, j ) = F (i)(j ).
.

Therefore, we have proved that (ϕ ◦ ψ)(F ) = F for all functors F : I →

Fun(J, C). Consider now a natural transformation η : F → G, where F ,
G : I → Fun(J, C) are functors. Then, for all i ∈ Ob I , j ∈ Ob J we have
    
.ϕ ψ(η) i = (ψ(η)(i, j ) = ηi j .
j
 
Hence we obtain ϕ ψ(η = η. To summarize, we have proved that ϕ ◦ ψ =
1FunI, Fun(J, C) . The proof will be finished once we show that ψ ◦ ϕ =
1Fun(I ×J, C) . Let H : I × J → C be a functor. Then for all (i, j ) ∈ Ob (I × J )
we have
 
ψ ϕ(H ) (i, j ) = ϕ(H )(i, j ) = H (i, j ),
.
4.2 Chapter 2 259

as desired. Furthermore, if η : H → G is a natural transformation, where H ,

G : I × J → C are functors, and (i, j ) ∈ Ob (I × J ), we obtain
   
ψ ϕ(η) (i, j ) = ϕ(η)i j = η(i, j ) .
.

Therefore ψ ◦ ϕ = 1Fun(I ×J, C) and the proof is finished. 



4.2 Chapter 2

2.3 (a) ⇒ (b) By assumption, (E, e) is the equalizer of (f, f ), so there exists
a unique w ∈ HomC (A, E) such that the following diagram is
commutative:

e f
E A B i.e., e 1A .
f

1A
. A (4.3)

Consider h1 , h2 ∈ HomC (A, B  ) such that h1 ◦ e = h2 ◦

e. Composing this equality with w on the right and using the
commutativity of diagram (4.3) yields h1 = h2 . This shows that
e is an epimorphism.
(b) ⇒ (c) In particular, we have f ◦ e = g ◦ e and, since e is an epimorphism,
we obtain f = g. Therefore, we have a unique w ∈ HomC (A, E)
such that diagram (4.3) is commutative. Furthermore, we have e ◦
(4.3)
(w ◦ e) = (e ◦ w) ◦ e = e and since e is an epimorphism we
obtain w ◦ e = 1E . Putting everything together it follows that w is
the inverse of e.
(c) ⇒ (d) As e is an isomorphism and f ◦e = g◦e we obtain, after composing
on the right with the inverse of e, that f = g. Now note that given
u ∈ HomC (C, A) we have a unique morphism in HomC (C, A),
namely u, which makes the following diagram commutative:

1A f
E A B
f

. C

(d) ⇒ (a) Since (A, 1A ) is the equalizer of (f, g) we have f ◦ 1A = g ◦ 1A

and therefore f = g, as desired. 

260 4 Solutions to Selected Exercises

2.6 Let t ∈ HomC (B, C  ) such that t ◦ f = t ◦ g. First we prove that t ◦ v makes
the following diagram commutative:
f h
A B C
g

.
C

Indeed, we have t ◦ v ◦ h = t ◦ f ◦ u = t ◦ g ◦ u = t. Assume now that there

exists another morphism w ∈ HomC (C, C  ) which makes the above diagram
commutative, i.e., w ◦ h = t. By composing this last equality on the right with
v and having in mind that h◦v = 1C , we obtain w = t ◦v. This shows that t ◦v
is the unique morphism which makes the above diagram commutative. 

2.7 First, we have

F (h) ◦ F (f ) = F (h ◦ f ) = F (h ◦ g) = F (h) ◦ F (g).

.

Consider now e ∈ HomC (F (B), E) such that e ◦ F (f ) = e ◦ F (g). We will

show that e◦F (v) is the unique morphism which makes the following diagram
commutative:

To this end, we have

e ◦ F (v) ◦ F (h) = e ◦ F (v ◦ h) = e ◦ F (f ◦ u) = e ◦ F (f ) ◦ F (u)

.

= e ◦ F (g) ◦ F (u) = e ◦ F (g ◦ u) = e.

Furthermore, suppose there exists a w ∈ HomC (F (C), E) such that the above
diagram is commutative, i.e., w ◦F (h) = e. By composing this equality on the
right with F (v) and having in mind that h ◦ v = 1C , we obtain w = e ◦ F (v).


2.8 Since (E, p) is the equalizer in C of (f, g) we have

f ◦ p = g ◦ p.
. (4.4)
4.2 Chapter 2 261

Therefore, we have

pX ◦ f ◦ p = 1X ◦ p = pX ◦ g ◦ p,
.

(4.4)
pY ◦ f ◦ p = f ◦ p = g ◦ p = pY ◦ g ◦ p.

Now using Proposition 2.2.14, (1) we obtain f ◦ p = g ◦ p.

Consider now u, v ∈ HomC (E  , X) such that f ◦ u = g ◦ v. Composing
this identity on the left with pX gives v = u. Similarly, by composing on the
left with pY yields f ◦u = g ◦v. Putting everything together we obtain f ◦u =
g ◦ u. As (E, p) is the equalizer in C of (f, g) we obtain a unique morphism
w ∈ HomC (E  , E) such that the following diagram is commutative:

p f
E X Y
g

.
E

This shows that w is the unique morphism which makes the following diagram
commutative:

E X
p
p g

X X Y
. f

Therefore, (E, p, p) is the pullback of (f , g), as desired. 


2.9 Proposition 1.7.7 shows that F is representable if and only if there exists a
representing
 pair (A, a), where A ∈ Ob C and a = (αi )i∈Ob I ∈ F (A) =
i∈I Hom C (Ai , A). Hence, F is representable
 if and only if for any C ∈ Ob C
and any x = (fi )i∈Ob I ) ∈ F (C) =  i∈I HomC (Ai , C), there exists a unique
f ∈ HomC (A, C) such that F (f ) (αi )i∈Ob I = (fi )i∈Ob I . In other words,
F is representable
 if and only if for any C ∈ Ob C and any x = (fi )i∈Ob I ) ∈
F (C) = i∈I HomC (Ai , C), there exists a unique f ∈ HomC (A, C) such
that f ◦ αi = fi for all i ∈  Ob I . Now observe that the last condition is
equivalent
  to A, (α i ) i∈Ob I being the coproduct of the family of objects
Ai i∈Ob I in C. 

262 4 Solutions to Selected Exercises

2.10 (a) Let I be the initial object of C and consider A, B ∈ Ob C. We first

construct the coproduct of A and B. As I is the initial object of C, we have
unique morphisms f ∈ HomC (I, A) and g ∈ HomC (I, B). Consider now
the pushout (C, qA , qB ) of (f, g). In particular, we have qA ◦f = qB ◦g.
We will prove that (C, qA , qB ) is the coproduct of A and B. To this end,
let u ∈ HomC (A, D) and v ∈ HomC (B, D). Since I is the initial object
of C we have a unique morphism from I to D and therefore we obtain
u ◦ f = v ◦ g. As (C, qA , qB ) is the pushout of (f, g), there exists a
unique morphism ∈ HomC (C, D) such that ◦qA = u and ◦qB = v.
This shows that (C, qA , qB ) is the coproduct of A and B.

f
I A
g qA

B C
qB

.
D

Next we construct coequalizers. Let α, β ∈ HomC (A, B) and consider

the coproduct (C, qA , qB ) of A and B in C. Thus, there exist unique
morphisms u, v ∈ HomC (C, B) such that the following diagrams are
commutative:
qA qB qA qB
A C B A C B

1B β 1B

.
B B

u ◦ qA = α,
. u ◦ qB = 1B , . (4.5)
v ◦ qA = β, v ◦ qB = 1B . (4.6)

Consider now (D, p, q) to be the pushout of (u, v). In particular, we

have q ◦ u = p ◦ v and by composing on the right with qA and using
(4.5) and (4.6) we obtain q ◦ α = p ◦ β. On the other hand, composing
the equality q ◦ u = p ◦ v on the right with qB and using again (4.5) and
(4.6) yields p = q. Putting everything together we have p ◦ α = p ◦ β.
We will show that (D, p) is the coequalizer of (α, β). Indeed, consider
4.2 Chapter 2 263

t ∈ HomC (B, D  ) such that t ◦ α = t ◦ β.

p
A B D

.
D (4.7)

We obtain

t ◦ u ◦ qA = t ◦ α = t ◦ β = t ◦ v ◦ qA ,
.

t ◦ u ◦ qB = t ◦ 1B = t ◦ v ◦ qB .

Therefore, we have (t ◦u) ◦qA = (t ◦v) ◦qA and (t ◦u) ◦qB = (t ◦v) ◦qB .
Now Proposition 2.2.14, (2) implies that t ◦ u = t ◦ v.

C B
p

t
B D
p

t
.
D

Since (D, p, p) is the pushout of (u, v), we obtain a unique w ∈

HomC (D, D  ) such that w ◦ p = t. This shows that w is the unique
morphism which makes (4.7) commutative and (D, p) is the coequalizer
of (α, β), as desired.
(b) Follows by duality.  

2.11 (a) Let f ∈ HomC (A, B), g ∈ HomC (A, C) and consider B ×C, (qB , qC )
to be the coproduct of B and C, where qB ∈ HomC (B, B × C) and
qC ∈ HomC (C, B × C). Furthermore, let (Q, q) be the coequalizer of
(qB ◦ f, qC ◦ g). In particular, we have q ◦ qB ◦ f = q ◦ qC ◦ g.
We will prove that (Q, q ◦ qC , q ◦ qB ) is the pushout of (f, g). Indeed,
let f  ∈ HomC (C, P ) and g  ∈ HomC (B, P ) such that

g  ◦ f = f  ◦ g.
. (4.8)
 
Since B × C, (qB , qC ) is the coproduct of B and C, there exists
a unique u ∈ HomC (B × C, P ) such that the following diagram is
264 4 Solutions to Selected Exercises

commutative:

qB qC
B / B ×C o C
EE
EE yy
EE u yyy
EE y
g E"  |yyy f 
P

i.e., u ◦ qB = g  , .
. (4.9)

u ◦ qC = f . (4.10)

We obtain
(4.9) (4.8) (4.10)
u ◦ qB ◦ f = g  ◦ f = f  ◦ g = u ◦ qC ◦ g.
.

Now, since (Q, q) is the coequalizer of (qB ◦ f, qC ◦ g), there exists a

unique v ∈ HomC (Q, P ) such that the following diagram is commuta-
tive:
qB f q
A B C Q
qC g

.
P (4.11)

Moreover, we obtain

(4.11) (4.10)
v ◦ q ◦ qC = u ◦ qC = f  ,
.

(4.11) (4.9)
v ◦ q ◦ qB = u ◦ qB = g  ,

which shows that the following diagram is commutative:

f
A B

g q qB

g
C Q
q qC

f
.
P
4.2 Chapter 2 265

We are left to show that v is the unique morphism which makes the above
diagram commutative. Indeed, suppose there exists a t ∈ HomC (Q, P )
such that t ◦ q ◦ qC = f  and t ◦ q ◦ qB = g  . This yields

(4.9)
t ◦ q ◦ qB = g  = u ◦ qB ,
.

(4.10)
t ◦ q ◦ qC = f  = u ◦ qC .

Now, Proposition 2.2.14, (2) implies t ◦ q = u and since v is the unique

morphism which makes diagram (4.11) commutative, we obtain t = v, as
desired.
(b) Follows by duality. 

2.12 (a) As shown in (the proof of) Proposition 2.1.4, any non-empty family of
objects in C admits a coproduct. Furthermore, the existence of an initial
object in C implies that an empty family of objects of C admits a coproduct
as well and therefore, C has all finite coproducts. Now showing that a
category with finite coproducts and coequalizers has all finite limits goes
much in the same fashion as the proof of Theorem 2.4.2.
(b) Follows by duality. 

2.13 (a) By exercise 2.10, (a), C has binary coproducts and coequalizers. Thus,
C has an initial object, binary coproducts and coequalizers and the
conclusion now follows from Exercise 2.12, (a).
(b) Follows by duality. 

2.14 (a) Let f ∈ HomC (A, B) be a split epimorphism and consider its right
inverse g ∈ HomC (B, A), i.e., f ◦ g = 1B . We will show that (B, f )
is the coequalizer of the pair of morphisms (g ◦ f, 1A ). Indeed, we have
f ◦ (g ◦ f ) = 1B ◦ f = f = f ◦ 1A . Moreover, if t ∈ HomC (A, B  )
such that t ◦ 1A = t ◦ (g ◦ f ), then u = t ◦ g ∈ HomC (B, B  )
is the unique morphism which makes the following diagram commuta-
tive:

1A f
A A B
g f
t g
t

.
B

(b) Let f ∈ HomC (A, B) be a regular epimorphism, i.e., we can

find two morphisms u, v ∈ HomC (C, A) such that (B, f ) is the
266 4 Solutions to Selected Exercises

coequalizer of the pair (u, v). Consider now the following commutative
square:

f
A B
g h

C D
. m

where m ∈ HomC (C, D) is a monomorphism. We have

m ◦ g ◦ u = h ◦ f ◦ u = h ◦ f ◦ v = m ◦ g ◦ v,
.

and since m is a monomorphism we obtain g ◦ u = g ◦ v.

Since (B, f ) is the coequalizer of the pair (u, v), there exists
a unique t ∈ HomC (B, C) such that the following diagram is
commutative:

f
C A B

t
g

.
C

Furthermore, we have m ◦ t ◦ f = m ◦ g = h ◦ f and since f is

an epimorphism we obtain m ◦ t = h. To conclude, we have a unique
morphism t ∈ HomC (B, C) such that t ◦ f = g and m ◦ t = h; this shows
that f is a strong epimorphism. 

2.15 Assume f is an epimorphism such that f = g ◦ h, where g is a monomor-
phism. Since f = g ◦ h is an epimorphism, Exercise 1.5, (c), shows that
g is also an epimorphism. Therefore, as C is a balanced category, f is an
isomorphism, as desired. 

2.16 Let G be a non-trivial group and let G be its associated category in the sense
of Example 1.2.2, (3) with Ob G = {•}. As noticed in Example 2.1.10, (10)
a pair of morphisms (x, y) in G such that x = y does not have an equalizer.
However, the category G admits pullbacks for any pair of morphisms (x, y).
To this end, we will show that the triple (•, y −1 , x −1 ) is the pullback of
(x, y), where x −1 and y −1 denote the inverse of the elements x and y,
4.2 Chapter 2 267

respectively, in the group G.

g
xg

h x− 1
y− 1 x
y
.

Indeed, it is straightforward to see that (•, y −1 , x −1 ) makes the above square

commutative. Furthermore, for any other g, h ∈ G such that xg = yh, the
unique morphism in G which makes the two triangles above commutative is
xg. 

2.20 Let A, B ∈ Ob C and f , g ∈ HomC (A, B) such that F  A, B (f ) = FA, B (g),
where FA, B : HomC (A, B) → HomD F (A), F (B) is the induced map
defined in (1.12). Then, we have F (f ) = F (g). As C has equalizers we
can consider the equalizer (E, e) of (f, g). Since F preserves equalizers
we obtain that (F (E), F (e)) is the equalizer of (F (f ), F (f )). Exercise 2.3
implies that F (e) is an isomorphism. Furthermore, F reflects isomorphisms,
which implies that e is also an isomorphism. Hence, by Exercise 2.3, we
obtain f = g, and therefore FA, B is injective, as desired. 

2.21 Let io ∈ Ob I be the initial object of I and for any j ∈ Ob I denote by uj the
unique morphism in I between i0 and j . Then the pair F (i0 ), (pj )j ∈Ob I
is a cone on F , where pj = F (uj ) ∈ HomC (F (i0 ), F (j )) for all j ∈ Ob I .
Indeed, if d ∈ HomI (i, j ) then we have uj , d ◦ ui ∈ HomI (i0 , j )
and since i0 is the initial object of I we obtain uj = d ◦ ui . Applying
 yields pj = F (d) ◦ pi , as  desired. Consider now another cone
F
C, (fj ∈ HomC (C, F (j )))j ∈Ob I on F . In particular, this implies that
the following diagram is commutative for all j ∈ Ob I :

.
268 4 Solutions to Selected Exercises

In other words, the morphism fi0 ∈ HomC (C, F (i0 )) makes the following
diagram commutative for all j ∈ Ob I :

The proof will be finished once we show that fi0 is the unique morphism with
this property. To this end, let g ∈ HomC (C, F (i0 ) such that pj ◦ g = fj for
all j ∈ Ob I . We obtain

g = F (1i0 ) ◦ g = F (ui0 ) ◦ g = pi0 ◦ g = fi0 ,

.

as desired. 

2.23 See Exercise 2.16. 


4.3 Chapter 3

3.1 Assume U : Field → Ring has a left adjoint L : Ring → Field. As noted in
Example 1.3.10, (3) Z is the initial object of Ring and Theorem 3.4.4 implies
that L(Z) is the initial object of Field. This contradicts Example 1.3.10, (4);
therefore U does not admit a left adjoint.
Assume now that U : Field → Ring has a right adjoint R : Ring → Field.
As noted in Example 1.3.10, (3) the zero ring is the final object of Ring and
Theorem 3.4.4 implies that R({0}) is the final object of Field. Again, this
contradicts Example 1.3.10, (4); therefore U does not admit a right adjoint.


3.3 The left adjoint L : Ab → R M is the tensor functor defined as follows:

L(A) = R ⊗ A,
. L(f )(r ⊗ a) = r ⊗ f (a)

for all A, B ∈ Ob Ab, f ∈ HomAb (A, B) and r ⊗ a ∈ R ⊗ A, where

for simplicity we denote ⊗Z by ⊗. Note that R ⊗ A ∈ Ob R M with the
left R-module structure given by r(r  ⊗ a) = rr  ⊗ a for all r, r  ∈ R and
a ∈ A. We use Theorem 3.5.1 to show that L  U . Indeed, consider the
natural transformations η : 1Ab → U L and ε : LU → 1R M defined as follows
4.3 Chapter 3 269

for all A ∈ Ob Ab and M ∈ Ob R M:

ηA : A → R ⊗ A,
. ηA (a) = 1R ⊗ a,
εM : R ⊗ M → M, εM (r ⊗ m) = rm.

It will suffice to show that (3.15) and (3.16) hold. To this end, for all r ⊗ a ∈
R ⊗ A we have

εR⊗A ◦ L(ηA )(r ⊗ a) = εR⊗A (r ⊗ ηA (a)) = εR⊗A (r ⊗ 1R ⊗ a)

.

= r(1R ⊗ a) = r ⊗ a = 1R⊗A (r ⊗ a),

i.e., (3.15) holds. Furthermore, for all m ∈ M we have

U (εM ) ◦ ηM (m) = U (εM )(1R ⊗ m) = 1R m = m = 1M (m),

.

which shows that (3.16) also holds. Thus, in light of Theorem 3.5.1, we obtain
that L  U , as desired.
Next, the right adjoint of U is the hom functor T : Ab → R M defined as
follows:

T (A) = HomZ (R, A),

. T (f )(g) = f ◦ g,

for all A ∈ Ob Ab, f ∈ HomAb (A, B) and g ∈ HomZ (R, A). Note that
HomZ (R, A) ∈ Ob R M with the left R-module structure given by (rf )(t) =
f (rt) for all r, t ∈ R and f ∈ HomZ (R, A). Again, we use Theorem 3.5.1
to show that U  R. Indeed, consider the natural transformations η : 1R M →
T U and ε : U T → 1Ab defined as follows for all A, B ∈ Ob Ab, M ∈
Ob R M, g ∈ HomZ (R, A) and m ∈ M:
 
ηM : M → HomZ R, U (M) ,
. ηM (m) = ψm : R → U (M),
ψm (r) = rm,
εA : HomZ (R, A) → A, εA (g) = g(1R ).

It will suffice to show that (3.15) and (3.16) hold. To this end, for all m ∈ M
we have
 
.εU (M) ◦ U (ηM ) (m) = εU (M) (ψm ) = ψm (1R ) = 1R m = m = 1U (M) (m),

and therefore (3.15) holds. Furthermore, for all g ∈ HomZ (R, A) and r ∈ R
we have εA ◦ ψg (r) = εA (rg) = (rg)(1R ) = g(r1R ) = g(r). Hence, we
obtain

εA ◦ ψg = g.
. (4.12)
270 4 Solutions to Selected Exercises

This leads to the following:

  (4.12)
.T (εA ) ◦ ηT (A) (g) = T (εA )(ψg ) = εA ◦ ψg = g,

i.e., T (εA ) ◦ ηT (A) = 1T (A) , which shows that (3.16) holds as well and we
obtain the desired adjunction. 

3.8 The equivalence (a) ⇔ (c) follows easily from (3.15).
(b) ⇒ (d) For all C ∈ Ob C we have

(3.16) b)
1GF (C) = G(εF (C) ) ◦ ηGF (C) = G(εF (C) ) ◦ GF (ηC ).
.

Thus, for all C ∈ Ob C we have

. 1GF (C) = G(εF (C) ) ◦ GF (ηC ). (4.13)

On the other hand, we also have

 
G F (ηC ) ◦ εF (C) ◦ ηGF (C) = GF (ηC ) ◦ G(εF (C) ) ◦ ηGF (C)
.

(3.16) b)
= GF (ηC ) = ηGF (C) .

Now Corollary 3.6.2, (1) implies that for all C ∈ Ob C we have F (ηC ) ◦
εF (C) = 1F GF (C) , and therefore

.GF (ηC ) ◦ G(εF (C) ) = 1GF GF (C) . (4.14)

(4.13) together with (4.14) imply that G(εF (C) ) is an isomorphism.

(d) ⇒ (b) If we assume G(εF (C) ) is an isomorphism, (3.15) implies
that its inverse is GF (ηC ). Furthermore, from (3.16) we obtain 1GF (C) =
G(εF (C) )◦ηGF (C) and therefore the inverse of G(εF (C) ) is ηGF (C) . This shows
that GF (ηC ) = ηGF (C) .
(a) ⇒ (d) Using (3.15), for all C ∈ Ob C we have

. 1GF (C) = G(εF (C) ) ◦ GF (ηC ). (4.15)

Since by Proposition 1.6.9, (1) any functor preserves isomorphisms, it follows

that GF (ηC ) is an isomorphism. Now (4.15) implies that G(εF (C) ) is an
isomorphism as well.
(d) ⇒ (a) We have already proved that d) implies the following for all
C ∈ Ob C:

GF (ηC ) = ηGF (C) .

. (4.16)
4.3 Chapter 3 271

Thus for all C ∈ Ob C we have

  (3.16)
G F (ηC ) ◦ εF (C) ◦ ηGF (C) = GF (ηC ) ◦ G(εF (C) ) ◦ ηGF (C) = GF (ηC )
.

(4.16)
= ηGF (C) = G(1F GF (C) ) ◦ ηGF (C) .

Now Corollary 3.6.2, (1) implies F (ηC ) ◦ εF (C) = 1F GF (C) . Using (3.15) we
obtain that F (ηC ) is indeed an isomorphism. 

3.9 Suppose first that ηD ∈ HomD (D, H F (D)) is an epimorphism for any d ∈
Ob D and let f , g ∈ HomD (D  , H G(D)) such that

.εD ◦ f = εD ◦ g. (4.17)

By Theorem 3.6.1, (1) there exists a unique f  ∈ HomC (F (D  ), G(D)) such

that H (f  )◦ηD  = f . Similarly, we have a unique g  ∈ HomC (F (D  ), G(D  ))
such that H (g  ) ◦ ηD  = g. Then (4.17) comes down to εD ◦ H (f  ) ◦ ηD  =
εD ◦ H (g  ) ◦ ηD  and since ηD  is an epimorphism we obtain εD ◦ H (f  ) =
εD ◦ H (g  ). Now Corollary 3.6.2, (2) implies f  = g  and therefore f = g.
The converse follows by duality. Indeed, by Corollary 3.5.4 we have Gop 
H with unit εop and H op  F op with counit ηop .
op 

3.10 For all objects C in Iso(C), ηC : C → GF (C) is an isomorphism and
by Proposition 1.6.9, (1) it follows that F (ηC ) is also an isomorphism.
Furthermore, (3.15) now implies that εF (C) is an isomorphism and therefore
the restriction of F to the subcategory Iso(C) of C, denoted by F , is a functor
with codomain Iso(D), i.e., F : Iso(C) → Iso(D). Similarly, using (3.16) this
time, it can be easily seen that the restriction of G to the subcategory Iso(D)
of D, denoted by G, is a functor with codomain Iso(C), i.e., G : Iso(D) →
Iso(C).
Furthermore, we can consider the natural transformations η : 1Iso(C) →
GF and ε : F G  → 1Iso(D) defined  by ηC = ηC and εD = εD for all
C ∈ Ob Iso(C) and D ∈ Ob Iso(D) . Then F and G form a pair of
adjoint functors with unit and counit given by η and ε respectively. As η and ε
are natural isomorphisms it follows, using Theorem 3.8.5, that the categories
Iso(C) and
 Iso(D) are equivalent.  

3.11 (a) Let lim T , (qi : lim T → T (i))i∈Ob I be the limit of T . The naturality
of β renders the following diagram commutative for all i ∈ Ob I :

(4.18)
272 4 Solutions to Selected Exercises

As G is the right adjoint of F , it preserves

 limits by Theorem 3.4.4.
Therefore, the pair G(lim T ), (G(qi ))i∈Ob I is the limit of GT : I → C.
Hence, there exists a unique ψ : G (lim T ) → G(lim T ) such that the
following diagram is commutative for all i ∈ Ob I :

(4.19)

Therefore, for all i ∈ Ob I we have

(4.19) (4.18)
G (qi ) = βT (i) ◦ G(qi ) ◦ ψ = G (qi ) ◦ βlim T ◦ ψ.
.

 
Now since G (lim T ), (G (qi ))i∈Ob I is the limit of G T : I → C,
Proposition 2.2.14, (1) implies that 1G (lim T ) = βlim T ◦ ψ. Furthermore,
for all i ∈ Ob I we have

= βT−1(i) ◦ G (qi ) ◦ βlim T = βT−1(i) ◦ βT (i) ◦ G(qi )

(4.19) (4.18)
G(qi ) ◦ ψ ◦ βlim T
.

= G(qi ).

Proposition 2.2.14, (1) implies ψ ◦ βlim T = 1G(lim T ) . Therefore, βlim T is

invertible and its inverse is equal to ψ.
(b) Follows by duality. 

3.13 First we construct a functor F : PreOrd → Poset which will turn out to be
the left adjoint of the inclusion functor. To start with, given a pre-ordered set
(P , ), we consider on P the relation ‘∼’ defined as follows for all x, y ∈ P :
x ∼ y if and only if x  y and y  x. A straightforward computation shows
that ‘∼’ is in fact an equivalence relation on P and we denote by P the set of
equivalence classes with respect to  ‘∼’ and
 by x the equivalence class in P of
some element x ∈ P . Moreover, P ,  is a partially-ordered set, where  is
defined as follows for all x, y ∈ P : x  y if and only if x  y. Note that  is
a well-defined relation on P ; indeed, if x = x  and y = y  we have x  x  ,
x   x, y  y  , y   y and if x  y then x   x  y  y  , which implies
x   y  , as desired.
Furthermore, if f : (P , ) → (Q, ) is a morphism in PreOrd (i.e., an
order preserving
  map f : P → Q), then we can define an order preserving
map f : P ,  → Q,  by f (x) = f (x) for all x ∈ P . We only show
that f is well-defined: if x, y ∈ P such that x = y then x  y and y  x
and since f is order preserving we obtain f (x) = f (y) or, equivalently, that
f (x) = f (y).
4.3 Chapter 3 273

We can now define a functor F : PreOrd → Poset as follows for all

pre-ordered sets (P , ), (Q, ) and order-preserving maps f : (P , ) →
(Q, ):
 
F (P , ) = P ,  ,
. F (f ) = f .

Next we use Theorem 3.6.1, (1) in order to show that F  I . To this end,
note that for all pre-ordered
  sets (P , ) we have an order-preserving map
πP : (P , ) → P ,  defined by πP (x) = x for all x ∈ P . Moreover,
π : 1PreOrd → I F defined for any pre-ordered set (P , ) by πP is a natural
transformation. Consider now two  pre-ordered sets (P , ), (Q, ) and a
morphism f : (P , ) → Q,  in PreOrd; the proof will be finished once
   
we show that there exists a unique morphism g : P ,  , → Q,  in Poset
   
such that I (g) ◦ πP = f . Define g : P ,  , → Q,  by g(x) = f (x)
for all x ∈ P . The only thing left to prove is that g is well-defined. Indeed,
if x = y then x  y and y  x and since f is order-preserving we obtain
f (x)f (y) and f (y)f (x). Now recall that  is a partial order on Q and
the anti-symmetry implies f (x) = f (y), as desired. 

3.14 We will show that, unless X is a singleton set, the cartesian product functor
X×− does not preserve products and, therefore, by virtue of Theorem 3.4.4, it
does not admit a left adjoint. Note that if X is a singleton set then the identity
functor is obviously the left adjoint of the corresponding cartesian product
functor.
 X be a set suchthat |X| = 1. Let Y , Z ∈ Ob Set and consider
Hereafter, let
their product Y × Z, (p1 , p2 ) in Set, as constructed in Example 2.1.5, (1)
i.e., Y × Z is the cartesian product of the two sets while p1 : Y × Z → Y and
p2 : Y × Z → Z denote the projections
 on the first and second
 component,
respectively. Furthermore, let X × Y × X × Z, (π1 , π2 ) be the product of
X × Y and X × Z in Set, as constructed in Example 2.1.5, (1) where π1 : X ×
Y × X × Z → X × Y , π2 : X × Y × X × Z → X × Z denote the projections
on X × Y and X × Z, respectively. Assume now that the cartesian product
functor X × − preserves products. Then, in light of Proposition 2.1.3, there
exists a unique isomorphism f : X × Y × Z → X × Y × X × Z in Set (set
bijection) such that the following diagram is commutative:

X × Y× Z
1X × p 1 1X × p 2

X × Y f X × Z

π1 π2

.
X × Y× X × Z
274 4 Solutions to Selected Exercises

It can be easily seen that any map f : X × Y × Z → X × Y × X × Z which

makes the above diagram commutative is given by f (x, y, z) = (x, y, x, z)
for all x ∈ X, y ∈ Y and z ∈ Z. The map f is obviously not surjective
whenever X has more than one element and we have reached a contradiction.
Hence, X × − does not preserve products. 

3.22 Let u, v ∈ HomD (D, D  ) such that u = v. As G is a faithful functor we
have G(u) = G(v) and since S is a generator in C, there exists a morphism
h ∈ HomC (S, G(D)) such that

G(u) ◦ h = G(v) ◦ h.
. (4.20)

By Theorem 3.6.1, (2) there exists a unique morphism w ∈ HomD (F (S), D)

such that G(w) ◦ ηS = h, where η : 1C → GF is the unit of the adjunction
F  G. Then (4.20) becomes

G(u ◦ w) ◦ ηS = G(v ◦ w) ◦ ηS .
. (4.21)

This shows that u ◦ w = v ◦ w and therefore F (S) is a generator in D. Indeed,

if u ◦ w = v ◦ w then we would have G(u ◦ w) ◦ ηS = G(v ◦ w) ◦ ηS , which
contradicts (4.21). 

3.25 Let I be an injective object in D and consider f ∈ HomC (A, G(I )) and
m ∈ HomC (A, B) a monomorphism. As I is an injective object in D and
F (m) is a monomorphism, there exists a morphism w ∈ HomD (F (B), I )
which makes the following diagram commutative:

. (4.22)

where ε : F G → 1D is the counit of the adjunction F  G. Furthermore, by

Theorem 3.6.1, (3) there exists a unique v ∈ HomC (B, G(I )) such that

εI ◦ F (v) = w.
. (4.23)

Putting everything together we obtain

(4.22) (4.23)
εI ◦ F (f ) = w ◦ F (m) = εI ◦ F (v) ◦ F (m) = εI ◦ F (v ◦ m).
.

Now Corollary 3.6.2, (2) implies that v ◦ m = f . In other words, we have

found a morphism v ∈ HomC (B, G(I )) which makes the following diagram
4.3 Chapter 3 275

commutative:

This shows that G(I ) is an injective object in C. The second claim follows by
duality. 

References

1. Adamek, J., Rosicky, J.: Locally Presentable and Accessible Categories. London Mathemat-
ical Society Lecture Note Series, vol. 189. Cambridge University Press, Cambridge (1994)
2. Adamek, J., Herrlich, H., Strecker, G.E.: Abstract and concrete categories. Pure and Applied
Mathematics. John Wiley & Sons, New York (1990)
3. Altman, A., Kleiman, S.: A Term of Commutative Algebra. Worldwide Center of Mathe-
matics, Cambridge (2013)
4. Arkowitz, M.: Introduction to Homotopy Theory. Universitext, Springer, New York (2011)
5. Awodey, S.: Category Theory. Oxford Logic Guides, vol. 52. Oxford University Press,
Oxford (2010)
6. Bergman, G.M.: An Invitation to General Algebra and Universal Constructions. Universi-
text. Springer, Cham (2015)
7. Bergman, G.M., Hausknecht, A.O.: Co-Groups and Co-Rings in Categories of Associative
Rings. Mathematical Surveys and Monographs, vol. 45. American Mathematical Society,
Providence (1996)
8. Borceux, F.: Handbook of Categorical Algebra. 1. Encyclopedia of Mathematics and Its
Applications, vol. 50. Cambridge University Press, Cambridge (1994)
9. Borceux, F.: Handbook of Categorical Algebra. 2. Encyclopedia of Mathematics and Its
Applications, vol. 51. Cambridge University Press, Cambridge (1994)
10. Borceux, F.: Handbook of Categorical Algebra. 3. Encyclopedia of Mathematics and Its
Applications, vol. 52. Cambridge University Press, Cambridge (1994)
11. Bosch, S.: Algebraic Geometry and Commutative Algebra. Universitext. Springer, London
(2022)
12. Bradley, T.-D., Bryson, T., Terilla, J.: Topology: A Categorical Approach. MIT Press,
Cambridge (2020)
13. Bucur, I., Deleanu, A.: Introduction to the Theory of Categories and Functors. Pure and
Applied Mathematics, vol. XIX. John Wiley & Sons, London-New York-Sydney (1968)
14. Burris, S., Sankappanavar, H.P.: A Course in Universal Algebra. Graduate Texts in Mathe-
matics, vol. 78. Springer-Verlag, New York-Berlin (1981)
15. Burton, D.M.: A First Course in Rings and Ideals. Addison-Wesley Series in Mathematics.
Addison-Wesley Publishing Company, Boston (1970)
16. Caenepeel, S., Militaru, G., Zhu, S.: Doi-Hopf modules, Yetter-Drinfel’d modules and
Frobenius type properties. Trans. Am. Math. Soc. 349, 4311–4342 (1997)
17. Caenepeel, S., Militaru, G., Zhu, S.: Frobenius and Separable Functors for Generalized
Module Categories and Nonlinear Equations. Lecture Notes in Mathematics, vol. 1787.
Springer-Verlag, Berlin (2002)

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 277
A. Agore, A First Course in Category Theory, Universitext,
https://doi.org/10.1007/978-3-031-42899-9
278 References

18. Card, E.F.: Ring Extensions With Identity Element. https://dalspace.library.dal.ca//handle/

10222/79584 (1994). Accessed 15 Aug 2023
19. Cartan, H., Eilenberg, S.: Homological Algebra. Princeton University Press, Princeton
(1956)
20. Chirvăsitu, A.: When is the diagonal functor Frobenius? Commun. Algebra 39, 1208–1225
(2011)
21. Eilenberg, S., Mac Lane, S.: General theory of natural equivalences. Trans. Am. Math. Soc.
58, 231–294 (1945)
22. Faith, C.: Algebra: Rings, Modules and Categories I. Grundlehren der mathematischen
Wissenschaften, vol. 190. Springer, Berlin/Heidelberg (1973)
23. Freyd, P.: Abelian categories. An introduction to the theory of functors. Harper’s Series in
Modern Mathematics. Harper & Row, Publishers, New York (1964)
24. Freyd, P.: Homotopy is not concrete. Reprints in Theory Appl. Categ. 6, 1–10 (2004)
25. Freyd, P., Scedrov, A.: Categories, Allegories. North-Holland Mathematical Library, vol. 39.
North-Holland Publishing, Amsterdam (1990)
26. Gabriel, P., Zisman, M.: Calculus of fractions and homotopy theory, Springer-Verlag, Berlin
(1967)
27. Gallian, J. A., Van Buskirk, J.: The number of homomorphisms from .Zn to .Zm . Am. Math.
Monthly 91, 196–197 (1984)
28. Gelfand, S.I., Manin, Y.I.: Methods of Homological Algebra. Springer Monographs in
Mathematics. Springer-Verlag, Berlin (2003)
29. Gomez-Ramirez, J.: A New Foundation for Representation in Cognitive and Brain Science.
Springer Series in Cognitive and Neural Systems, vol. 7. Springer Dordrecht (2014)
30. Herrlich, H., Strecker, G. E.: Category Theory. Sigma Series in Pure Mathematics, vol. 1.
Heldermann Verlag, Lemgo (2007)
31. Kan, D.M.: Adjoint functors. Trans. Am. Math. Soc. 87, 294–329 (1958)
32. Kashiwara, M., Schapira, P.: Categories and sheaves. Grundlehren der mathematischen
Wissenschaften, vol. 332. Springer, Berlin/Heidelberg (2006)
33. Kelly, G.M.: Basic Concepts of Enriched Category Theory. London Mathematical Society
Lecture Note Series, vol. 64. Cambridge University Press, Cambridge-New York (1982)
34. Leinster, T.: Basic Category Theory. Cambridge Studies in Advanced Mathematics, vol. 143.
Cambridge University Press, Cambridge (2014)
35. Mac Lane, S.: Categories for the Working Mathematician. Graduate Texts in Mathematics,
vol. 5. Springer-Verlag, New York (1998)
36. Mac Lane, S., Moerdijk, I.: Sheaves in Geometry and Logic. Universitext. Springer-Verlag,
New York (1994)
37. Moschovakis, Y.: Notes on Set Theory. Undergraduate Texts in Mathematics. Springer, New
York (2006)
38. Müger, M.: Topology for the Working Mathematician. https://www.math.ru.nl/~mueger/
topology.pdf (2022). Accessed 15 Aug 2023
39. Munkres, J.R.: Topology. Prentice Hall, Upper Saddle River (2000)
40. van Munster, B.: Hausdorffization and Homotopy. Am. Math. Mon. 124, 81–82 (2017)
41. Nakayama, T., Tsuzuku, T.: On Frobenius extensions. I. Nagoya Math. J. 17, 89-110 (1960)
42. Neeman, A.: Triangulated categories. Annals of Mathematics Studies, vol. 148. Princeton
University Press, Princeton (2001)
43. Negrepontis, J.W.: Duality in analysis from the point of view of triples. J. Algebra 19, 228–
253 (1971)
44. Osborne, M.S.: Hausdorffization and Such. Am. Math. Mon. 121, 727–733 (2014)
45. Pareigis, B.: Advanced Algebra. https://www.mathematik.uni-muenchen.de/~pareigis/
Vorlesungen/01WS/advalg.pdf (2002). Accessed 15 Aug 2023
46. Pareigis, B.: Categories and Functors. Pure and Applied Mathematics, vol. 39. Academic
Press, New York-London (1970)
47. Riehl, E.: Categorical Homotopy Theory. New Mathematical Monographs, vol. 24. Cam-
bridge University Press, Cambridge (2014)
References 279

48. Riehl, E.: Category Theory in Context. Aurora: Dover Modern Math Originals. Dover
Publications, Mineola (2016)
49. Rotman, J.J.: Advanced Modern Algebra. Prentice Hall, Upper Saddle River (2002)
50. Rotman, J.J.: An introduction to the Theory of Groups. Graduate Texts in Mathematics, vol.
148. Springer-Verlag, New York (1995)
51. Shulman, M.A.: Set Theory for Category Theory. https://arxiv.org/abs/0810.1279v2 (2008).
Accessed 15 Aug 2023
52. Spivak, D. I.: Category Theory for the Sciences. MIT Press, Cambridge (2014)
53. Stenström, B.: Rings of Quotients. Grundlehren der mathematischen Wissenschaften, vol.
217. Springer-Verlag, New York-Heidelberg (1975)
Index

A of finite sets, 3
Abelianization, 30 of groups, 4
functor, 30 of Hausdorff topological spaces, 5
Adjunction/adjoint functors, 153 of left R-modules, 5
Alexandroff topology, 34 of monoids, 4
Anti-isomorphism of categories, 38 of objects over C (slice category), 59
Associative law, 2 of objects under C (coslice category), 58
Axiom of choice, 1 of partially ordered sets, 5
of pointed topological spaces, 5
of pre-ordered sets, 5
B of presheaves, 63
Balanced category, 10 with (finite) products, 84
Bifunctor, 24 with pullbacks, 97
Bimorphism, 10 with pushouts, 98
of relations, 4
of right R-modules, 4
C of rings, 4
Cartesian product of sets, 3
bifunctor, 28 of simple groups, 4
functor, 29 of small categories, 35
Category, 2 of topological spaces, 5
of abelian groups, 4 of unitary rings, 4
of (commutative) algebras, 5 Cayley’s theorem, 76
of cocones, 108 Cocomplete category, 110
with coequalizers, 92 Cocone, 106
of commutative unitary rings, 4 Codomain of a morphism, 2
of compact Hausdorff topological spaces, 5 Coequalizer, 92
of cones, 108 Cogenerator (coseparator), 237
with (finite) coproducts, 84 Cokernel of a morphism, 97
defined by a pre-ordered set, 3 Colimit
of divisible groups, 4 functor, 110, 118
with equalizers, 92 preserving functor, 127
of fields, 4 reflecting functor, 131
with finite colimits, 110 Comma-category, 55
with finite limits, 110 Commutative diagram, 19

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 281
A. Agore, A First Course in Category Theory, Universitext,
https://doi.org/10.1007/978-3-031-42899-9
282 Index

Commutator subgroup, 30 Essentially

Compact-open topology, 27 small category, 209
Complete category, 110 surjective functor, 36
Composition law, 2 Evaluation bifunctor, 68
Concrete category, 38 Extremal
Cone, 106 epimorphism, 150
Congruence (on a category), 20 monomorphism, 150
Constant functor, 25
Continuity at a point, 162
Coproduct, 84 F
Coreflective subcategory, 194 Faithful functor, 36
Coreflector, 194 Final object, 11
Core groupoid of a category, 10 Finite category, 2
Cosolution set condition, 231 Forgetful functor, 30
Counit of an adjunction, 170 Free
Covariant/contravariant functor, 24 category on a graph, 18
Co-well-powered category, 16 group functor, 155
group on a set, 155
module functor, 176
D module on a set, 176
Diagonal functor, 69 product of groups, 90
Diagram, 19 Freyd’s adjoint functor theorem, 230
Discrete Frobenius functor, 225
category on a set, 2 Full
topology, 9 functor, 36
Divisible group, 4 subcategory, 6
Domain of a morphism, 2 Fully faithful functor, 36
Dorroh extension Functor
functor, 34 category, 63
of a ring, 33 preserving a property, 40
Double dual space functor, 28 reflecting a property, 40
Dual
(opposite) category, 16
functor, 52 G
of a natural transformation, 54 Galois connection, 157
space (contravariant) functor, 28 Gelfand-Naimark duality, 214
Duality Generator (separator), 236
of categories, 199 Godement product, 44
principle, 52 Graph, 18
Groupoid, 10

E H
Edges of a graph, 18 Hausdorff quotient functor, 33
Empty Hom
category, 3 bifunctor, 26
functor, 26 (contravariant) functor, 27
Epimorphism, 6 Homeomorphism, 9
Equalizer, 91 Homotopy, 22
Equivalence category, 23
of categories, 199 Horizontal composition of natural
relation generated by a binary relation, transformations (Godement
94 product), 44
Index 283

I O
Identity Objects of a category, 2
law, 2 Order preserving map, 5
morphism, 2
Image of a functor, 25
Inclusion functor, 25 P
Indiscrete topology, 7 Partially ordered set, 5
Initial object, 11 Pasting lemma, 23
Injective object, 80 Path (in a graph), 18
Interchange law between vertical and Pointed category, 11
horizontal composition of natural Pointwise composition of functors, 25
transformations, 80 Pontryagin duality, 214
Interior of a subset of a topological space, 94 Power set (contravariant) functor, 29
Inverse Precomposition functor, 68
of a functor, 38 Pre-ordered set, 3
natural transformation, 43 Presheaf, 63
Isomorphic Product, 83
categories, 38 category, 17
objects, 6 topology, 29
Isomorphism of categories, 38 Projection functor, 26
Projective object, 81
Pullback, 97
K Pushout, 97
Kernel of a morphism, 97

Q
L Quotient
Left adjoint, 153 category, 21
Left/right associated functor of a bifunctor, functor, 25
35 object, 14
Limit topology, 95
functor, 110, 117
preserving functor, 127
reflecting functor, 131 R
Limitation of size axiom, 1 R-bilinear map, 32
Localization Reflective subcategory, 194
of a category by a class of morphisms, 214 Reflector, 194
functor with respect to a multiplicative set, Regular
33 epimorphism, 150
Locally monomorphism, 150
compact topological space, 162 Representability criterion, 245
small category, 2 Representable functor, 48
Representing
object, 48
M pair of a functor, 50
Monomorphism, 6 Restriction of scalars functor, 34
Morphisms of a category, 2 Right adjoint, 153
Multiplicative subset of a ring, 5

S
N Set
Natural isomorphism, 42 of cogenerators (coseparators), 236
Natural transformation/isomorphism, 41 of generators (separators), 236
Neumann–Bernays–Gödel set theory, 1 Simple group, 4
284 Index

Skeleton of a category, 208 V

Small Vertex
category, 2 of a cocone, 107
colimit of a functor, 110 of a cone, 106
graph, 18 Vertical composition of natural
limit of a functor, 110 transformations, 43
Solution set condition, 231 Vertices of a graph, 18
Special adjoint functor theorem, 239
Split
epimorphism, 78 W
(co)equalizer, 148 Weakly
monomorphism, 78 final set, 229
Stone duality, 214 initial set, 229
Stone space, 214 Well-powered category, 16
Stone–Čech compactification functor, 31 Whiskering of a natural transformation by a
Strong functor, 44
epimorphism, 78
monomorphism, 78
Subcategory, 6 Y
Subobject, 12 Yoneda’s embedding, 76
Subspace topology, 94 Yoneda’s lemma, 70

T Z
Tensor product functor, 29 Zermelo–Fraenkel set theory, 1
Trivial path in a graph, 18 Zero-morphism, 11
Tube lemma, 162 Zero object, 11

U
Unit of an adjunction, 170
Universal enveloping group functor, 31