ENCYCLOPEDIA OF MATHEMATICS AND ITS APPLICATIONS 50 Re Tee PVA To BASIC CATEGORY THEORY FRANCIS BORCELX10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40. 41. 42. 43. 45. 46. 47. 48. 49. 50. ENCYCLOPEDIA OF MATHEMATICS AND ITS APPLICATIONS LA Santalo Integral Geometry and Geometric Probability GE Andrews The theory of partitions R J McEliece The Theory of Information and Coding: A Mathematical Framework for Communication W Miller, Jr Symmetry and separation of variables D Ruelle Thermodynamic Formalism: The Mathematical Structures of Classical Equilib- rium Statistical Mechanics : H Minc Permanents FS Roberts Measurement Theory with Applications to Decisionmaking, Utility, and the Social Sciences L C Biedenharn and J D Louck Angular Momentum in Quantum Physics: Theory and Applications L C Biedenharn and J D Louck The Racah-Wigner Algebra in Quantum Theory JD Dollard and CN Friedman Product Integration with Applications in Quantum Theory WB Jones and W J Thron Continued Fractions: Analytic Theory and Applications NF G Martin and J W England Mathematical Theory of Entropy G A Baker, Jr and P Graves-Morris Padé Approzimants, Part I, Basic Theory G A Baker, Jr and P Graves-Morris Padé Approzimants, Part II, Extensions and Appli- cations E G Beltrametti and G Cassinelli_ The Logic of Quantum Mechanics G D James and A Kerber The Representation Theory of Symmetric Groups M Lothaire Combinatorics on Words HO Fattorini The Cauchy Problem G G Lorentz, K Jetter and S D Riemenschneider Birkhoff Interpolation R Lidl and H Niederreiter Finite Fields WT Tutte Graph Theory IR Bastida Field Extensions and Galois Theory JRCannon The One-Dimensional Heat Equation S Wagon The Banach-Tarski Paradox A Salomaa Computation and Automata N White (ed) Theory of Matroids N H Bingham, C M Goldie and J L Teugels Regular Variations P P Petrushev and V A Popov Rational Approzimation of Real Functions N White (ed) Combinatorial Geometries M Phost and H Zassenhaus Algorithmic Algebraic Number Theory J Aczel and J Dhombres Functional Equations in Several Variables M Kuczma, B Choczewski and R Ger Iterative Functional Equations RV Ambartzumian Factorization Calculus and Geometric Probability G Gripenberg, S-O Londen and O Staffans Volterra Integral and Functional Equations G Gasper and M Rahman Basic Hypergeometric Series E Torgersen Comparison of Statistical Experiments A Neumaier Interval Methods for Systems of Equations N Korneichuk Ezact Constants in Approzimation Theory R Brualdi and H Ryser Combinatorial Matric Theory N White (ed) Matroid applications S Sakai Operator Algebras in Dynamical Systems W Hodges Model Theory H Stahl and V Totik General Orthogonal Polynomials R Schneider Convex Bodies G Da Prato and J Zabezyk Stochastic Equations in Infinite Dimensions A Bjorner et al Oriented Matroids G Edgar and L Sucheston Stopping Times and Directed Processes C Sims Computation with Finitely Presented Groups TW Palmer C*-algebras I F Borceux Handbook of Categorical Algebra 1, Basic Category TheoryENCYCLOPEDIA OF MATHEMATICS AND ITS APPLICATIONS Handbook of Categorical Algebra 1 Basic Category Theory Francis Borceux Département de Mathématique Université Catholique de Louvain CAMBRIDGE § UNIVERSITY PRESSPublished by the Press Syndicate of the University of Cambridge The Pitt Building, Trumpington Street, Cambridge CB2 1RP 40 West 20th Street, New York, NY 10011-4211, USA 10 Stamford Road, Oakleigh, Melbourne 3166, Australia © Cambridge University Press 1994 First published 1994 Printed in Great Britain at the University Press, Cambridge A catalogue record for this book is available from the British Library Library of Congress cataloguing in publication data ISBN 0 521 44178 1 hardback TAGto Sammy Eilenberg, mathematician and friendContents Preface to volume 1 Introduction to this handbook 1 11 1.2 1.3 1.4 1.5 1.6 L7 1.8 1.9 1.10 1.11 2 21 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 2.10 2.11 The language of categories Logical foundations of the theory Categories and functors Natural transformations Contravariant functors Full and faithful functors Comma categories Monomorphisms Epimorphisms Isomorphisms The duality principle Exercises Limits Products Coproducts Initial and terminal objects Equalizers, coequalizers Pullbacks, pushouts Limits and colimits Complete categories Existence theorem for limits Limit preserving functors Absolute colimits Final functors vii page x xiii 10 15 18 20 23 27 31 33 36 38 38 44 47 48 51 56 59 60 63 66 69viii 2.12 2.13 2.14 2.15 2.16 2.17 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 4.1 4.2 4.3 44 4.5 46 4.7 48 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 Contents Interchange of limits Filtered colimits Universality of colimits Limits in categories of functors Limits in comma categories Exercises Adjoint functors Reflection along a functor Properties of adjoint functors The adjoint functor theorem Fully faithful adjoint functors Reflective subcategories Epireflective subcategories Kan extensions Tensor product of set-valued functors Exercises Generators and projectives Well-powered categories Intersection and union Strong epimorphisms Epi-mono factorizations Generators Projectives Injective cogenerators Exercises Categories of fractions Graphs and path categories Calculus of fractions Reflective subcategories as categories of fractions The orthogonal subcategory problem Factorization systems The case of localizations Universal closure operations The calculus of bidense morphisms Exercises 72 75 84 87 91 93 96 96 106 108 114 118 120 122 128 130 132 132 133 136 147 150 163 167 174 176 176 181 190 193 209 218 227 235 248Contents 6 Flat functors and Cauchy completeness 6.1 Exact functors 6.2 Left exact reflection of a functor 6.3 Flat functors 6.4 The relevance of regular cardinals 6.5 The splitting of idempotents 6.6 The more general adjoint functor theorem 6.7 Exercises 7 Bicategories and distributors 7.1 2-categories 7.2 2-functors and 2-natural transformations 7.3 Modifications and n-categories 7.4 2-limits and bilimits 7.5 Lax functors and pseudo-functors 7.6 Lax limits and pseudo-limits 7.7 Bicategories 7.8 Distributors 7.9 Cauchy completeness versus distributors 7.10 Exercises 8 Internal category theory 8.1 Internal categories and functors 8.2 Internal base-valued functors 8.3 Internal limits and colimits 8.4 Exercises Bibliography Index ix 250 250 253 260 267 271 277 279 281 281 287 290 293 296 300 302 308 314 323 325 325 328 333 339 341 343Preface to volume 1 Volume 1 of the Handbook of category theory is concerned with those notions and techniques which turn out to appear quite naturally in most developments of category theory, independently of additional structures or properties one requires from the categories involved in the study. Any book on category theory must say a word on the non-obvious problems concerned with the logical foundations of the theory. We men- tion both the axiom system of classes and that of universes, and later we freely use the presentation which fits best the problem we study. We chose not to dwell on foundational questions as long as the development of the theory does not really depend on them. Another general principle in this volume is to develop the general notions from more accessible special cases, for which we have given a large supply of examples. This is by no means the most economical way of developing the theory, but we hope inexperienced readers will appreciate our pedagogical choice. Of course we start with the basic vocabulary of categories, functors, natural transformations, monomorphisms, epimorphisms, isomorphisms. The analogies between monomorphisms and epimorphisms, covariant and contravariant functors, lead to the famous duality principle which is, with the Yoneda lemma, one of the key results of the first chapter. Starting with the notions of products, coproducts, equalizers, coequal- izers, pullbacks, and so on, we reach the general notion of limit of a func- tor and prove the corresponding existence theorem. We devote special attention to some classes of colimits, like absolute ones, filtered ones and universal ones. We also study limits in categories of functors and dwell on the case of colimits of set-valued functors. Adjoint functors are the next fundamental notion of categorical al- gebra. We prove the classical general and special existence theorems inPreface to volume 1 xi terms of limits and particularize our study to the case of reflective sub- categories. We treat separately the case of Kan extensions, which is a fruitful example of an adjunction which cannot be deduced in its general form from the previous existence theorems. The special adjoint functor theorems refer explicitly to notions like subobject (equivalence class of monomorphisms), generator and cogen- erator. We decided to group in a separate chapter a specific study of the various kinds of monomorphisms, epimorphisms (plain, extremal, strong, regular) and correspondingly for generators, cogenerators (plain, strong, regular, dense). This is also the opportunity for studying epi- mono factorizations as well as notions related to monomorphisms or epimorphisms, like injective and projective objects. We pay special at- tention to the case of injective cogenerators, whose existence is very often related to rather deep specific theorems. Chapter 5 is probably the most difficult one in this first volume. It is essentially concerned with rather sophisticated methods for constructing reflective subcategories. We start with the problem of “formally inverting an arrow g: A——>C of a category” (categories of fractions) and relate this with the problem of “uniquely extending a morphism f: A—~B along the arrow g: A——>C” (the orthogonal subcategory problem). Un- der good assumptions, we prove by transfinite induction the reflectivity of the subcategory of those objects “orthogonal to a set of mappings”. We relate this to the general notion of (€,. M)-systems of factorization. We discuss the special case where the reflection is left exact (the “local- izations” ) and relate it to the notions of universal closure operation and bidense morphism. The rest of the book is essentially devoted to some generalizations or special cases of the basic notions of chapters 1, 2 and 3. When @ is a category with finite limits, set-valued functors on @ which Preserve finite limits (the “left exact functors”) admit interesting prop- erties and characterizations: in particular, they are exactly the filtered colimits of representable functors. When @ is an arbitrary small cate- gory, this characterization serves as a definition for the notion of “flat functor”. Flat functors share most properties of left exact functors and will turn out to play a key role in a categorical approach of model theory (see chapter 5, volume 2, accessible categories). Another elementary notion which turns out to have unexpectedly rich applications is that of “splitting idempotents”; this happens to be equiv- alent to a rather weak completeness property: having just those limits or colimits which are preserved by all functors. This is called the Cauchyxii Preface to volume 1 completeness of the category, for reasons which will become apparent in chapter 6 of volume 2. Replacing “completeness” by “Cauchy complete- ness” and “preserving limits” by “being a-flat, for every regular cardinal a” yields amazingly enough a generalization of the adjoint functor theo- rem; this generalization now contains as a special case the Kan extension theorem of chapter 3. While a category has just objects and arrows, the category of cat- egories and functors can be provided with additional devices, namely natural transformations between functors. This leads to the richer no- tion of a 2-category, where besides objects and arrows one gives also “2-cells” between the arrows. There are corresponding enrichments of the notions of functor, natural transformation, limit, adjoint functors, and so on. Now in category theory, many constructions are defined uniquely... up to isomorphism! This results in very often obtaining isomorphisms where one would have expected equalities. So it is not unusual to reach a situation where a composite f o(goh) (whatever f,9,h and the com- position are) is just isomorphic to (f og) oh, not necessarily equal. Taking this seriously, one gets the notion of a bicategory: one has ob- jects, arrows and 2-cells, but in various axioms “equalities” are replaced by “isomorphic 2-cells”. A basic example of a bicategory is that of small categories, distributors and natural transformations between them: in category theory, a distributor is to a functor what, in set theory, a rela- tion is to a function. Every functor turns out to be a distributor with a right adjoint, and the converse holds when working with Cauchy com- plete categories. In the same spirit as bicategories, one can “relax” the notions of functor, natural transformation, limit, colimit, working now “up to an isomorphic 2-cell” or even “up to an arbitrary 2-cell”. We conclude this first volume with an elementary study of internal categories. While a small category has a set of objects and a set of arrows, together with some operations “source”, “target”, “unit”, “composition” given by mappings, one is now interested in replacing set by object of a category @ and mapping by arrow of the category @. This is the notion of “category internal to @”. We study the corresponding notions of internal functors and internal limits or colimits.Introduction to this handbook My concern in writing the three volumes of this Handbook of categorical algebra has been to propose a directly accessible account of what — in my opinion — a Ph.D. student should ideally know of category theory before starting research on one precise topic in this domain. Of course, there are already many good books on category theory: general accounts of the state of the art as it was in the late sixties, or specialized books on more specific recent topics. If you add to this several famous original papers not covered by any book and some important but never published works, you get a mass of material which gives probably a deeper insight in the field than this Handbook can do. But the great number and the diversity of those excellent sources just act to convince me that an integrated presentation of the most relevant aspects of them remains a useful service to the mathematical community. This is the objective of these three volumes. if The first volume presents those basic aspects of category theory which are present as such in almost every topic of categorical algebra. This cludes the general theory of limits, adjoint functors and Kan exten- ions, but also quite sophisticated methods (like categories of fractions orthogonal subcategories) for constructing adjoint functors. Special ention is also devoted to some refinements of the standard notions, Cauchy completeness, flat functors, distributors, 2-categories, bicat- jories, lax-functors, and so on. The second volume presents a selection of the most famous classes of “structured categories”, with the exception of toposes which appear volume 3. Fhe first historical example is that of abelian categories, | which we follow by its natural non-additive generalizations: the regular and exact categories. Next we study various approaches to “categories of models of a theory”: algebraic categories, monadic categories, locally xiiixiv Introduction to this handbook presentable and accessible categories. We introduce also enriched cat- egory theory and devote some attention to topological categories. The volume ends with the theory of fibred categories “a la Bénabou”. The third volume is entirely devoted to the study of categories of sheaves: sheaves on a space, a locale, a site. This is the opportunity for developing the essential aspects of the theory of locales and introducing Grothendieck toposes. We relate this with the algebraic aspects of vol- ume 2 by proving in this context the existence of a classifying topos for coherent theories. All these considerations lead naturally to the notion of an elementary topos. We study quite extensively the internal logic of toposes, including the law of excluded middle and the axiom of infinity. We conclude by showing how toposes are a natural context for defining sheaves. Besides a technical development of the theory, many people appreciate historical notes explaining how the ideas appeared and grew. Let me tell you a story about that. It was in July, I don’t remember the year. I was participating in a summer meeting on category theory at the Isles of Thorns, in Sussex. Somebody was actually giving a talk on the history of Eilenberg and Mac Lane’s collaboration in the forties, making clear what the exact contribution of the two authors was. At some point, somebody in the au- dience started to complain about the speaker giving credit to Eilenberg and Mac Lane for some basic aspect of their work which — he claimed — they borrowed from somebody else. A very sophisticated and animated discussion followed, which I was too ignorant to follow properly. The only things I can remember are the names of the two opponents: the speaker was Saunders Mac Lane and his opponent was Samuel Eilenberg. I was not born when they invented category theory. With my little story in mind, maybe you will forgive me for not having tried to give credit to anybody for the notions and results presented in this Handbook. Let me conclude this introduction by thanking the various typists for their excellent job and my colleagues of the Louvain-la-Neuve category seminar for the fruitful discussions we had on various points of this Handbook. I want especially to acknowledge the numerous suggestions Enrico Vitale has made for improving the quality of my work.Introduction to this handbook xv Handbook of categorical algebra Contents of the three volumes Volume 1: Basic category theory 1. The language of categories Limits Adjoint functors Generators and projectives Categories of fractions Flat functors and Cauchy completeness Bicategories and distributors Internal category theory epxAaaaey Volume 2: Categories and structures 1. Abelian categories Regular categories Algebraic categories Monadic categories Accessible categories Enriched categories Topological categories Fibred categories PAP VTE Sy Volume 3: Categories of sheaves 1. Locales Sheaves Grothendieck toposes The classifying topos Elementary toposes Internal logic of a topos The law of excluded middle The axiom of infinity Sheaves in a topos PEANEAh EN1 The language of categories 1.1 Logical foundations of the theory It is a common practice, when developing mathematics, to consider a statement involving “all groups” or “all topological spaces” .... For example we say that an abelian group A is projective when, for every surjective homomorphism of abelian groups f: B——>C and every group homomorphism g: A——>C, g factors through f (see diagram 1.1). This definition of “A being projective” starts thus with a list of universal quantifiers VBIVG Vi vo 7 This formula, from the point of view of set theory, creates a problem: the variables B and C are “running through something (= the collec- tion of all abelian groups) which is not a set”. This last fact is an easy consequence of the following well-known paradox. Proposition 1.1.1 There exists no set S such that reéS x isa set. Proof In other (bad) words: “the set of sets does not exist”! To prove this, let us assume such an S exists. Since x ¢ = is a formula of set theory T={clceS and c¢z} defines a subset T of S, thus in particular a set T. The law of excluded middle tells us that TeT o T€T. But from the definition of T itself we conclude that TeT=T¢T, 12 The language of categories A wo B—- C f Diagram 1.1 T¢TSTeT?, thus in both cases a contradiction. Category theory will in fact be handling all the time “the collection of all groups”, “the collection of all sets”, “the collection of all topological spaces”, and so on .... Therefore it is useful to pay some attention to these questions of “size” at the very beginning of this book. A first way to handle, in category theory, problems of this type is to assume the axiom on the existence of “universes”. Definition 1.1.2 A universe is a set U with the following properties (1) ceyandyeu => reU, (2) Leu and Vi € Ia; CEU > Ujer ti EU, (3) ceU>P(z) EU, (4) « €U and f:x—~+y surjective function > y €U, (5) Neu, where N denotes the set of finite ordinals and P(x) denotes the set of subsets of x. Notice some easy consequences of the definition. Proposition 1.1.3 (1) ce€U andyCr>yeU, (2) c EU andy EU = {z,y} EU, (3) c EU andyEU>zx ye, (4) ceU andyeUsrvel. Proof We prove (1) and leave the rest as an easy exercise. First of all 0 € Nand N € YU, thus 0 € U. Now if z € U and y C « with y $ 0, choose z € y. Define f:x—>y to be f(t) =tiftey, fj=zift¢y.1.1 Logical foundations of the theory 3 Obviously f is surjective and therefore y € U. It should be noticed that — assuming the axiom of choice in our set the- ory — condition (4) in definition 1.1.2 could have been replaced precisely by zé€U and yCrayedl. Now the axiom on the existence of universes is just Axiom 1.1.4 Every set belongs to some universe. Not much is known about this axiom from the point of view of set theory. Because of the property eeu and yCrsyel, it sounds reasonable to think of the elements of a universe as being “sufficiently small sets”. If you choose to use the theory of universes as a foundation for category theory, the following convention has to remain valid throughout this book. Convention 1.1.5 We fix a universe U and call “small sets” the ele- ments of U. Obviously we now have the following Proposition 1.1.6 There exists a set S with the propertyr Ee S@ux is a small set. Proof Just choose S =U. An analogous statement is valid for small abelian groups, small topo- logical spaces, and so on .... For example a small group is a pair (G, +) where G is a small set (and there is just a set of them) and + is a suitable mapping G x G——>G (and there is just a set of them); so we can draw the conclusion by proposition 1.1.3. An alternative way to handle these problems of size is to use the Gédel-Bernays theory of sets and classes. In the Zermelo-Frankel theory, the primitive notions are “set” and “membership relation”. In the Gédel- Bernays theory, there is one more primitive notion called “class” (think of it as “a big set”); that primitive notion is related to the other two by the property that every set is a class and, more precisely: Axiom 1.1.7 |B| ee between the classes of objects of J and B; the image of A € & is written F(A) or just FA; (2) for every pair of objects A, A’ of x, a mapping } f(A, A!) ——> &(FA, FA’); the image of f € (A, A’) is written F(f) or just Ff. These data are subject to the following axioms:6 The language of categories (1) for every pair of morphisms f € f(A, A’), 9 € S(A', A”) F(g0 f) = F(g9) 0 F(f); (2) for every object Ac F(la) = Ira. Given two functors F: —>@ and G: 8—->@, a pointwise compo- sition immediately produces a new functor GoF: .« —>@. This compo- sition law is obviously associative. The identity functor on the category of (i.e. choose every mapping in definition 1.2.2 to be the identity) is clearly an identity for that composition law. A careless argument could thus lead to the conclusion that categories and functors constitute a new category ... but this can easily be reduced to a contradiction us- ing proposition 1.1.1! The point is that, in the axioms for a category, it is required to have a set of morphisms between any two objects. And when the categories sf and @ merely have a class of objects, there is no way to force the functors from # to & to constitute a set. All along in this book we shall realize how crucial it is, in category theory, to distin- guish all the time between sets and classes. To facilitate the language, we particularize definition 1.2.1. Definition 1.2.3 A category @ is called a small category when its class |@| of objects is a set. The next result is then obvious (see 1.1.8). Proposition 1.2.4 Small categories and functors between them consti- tute a category. Examples 1.2.5 Let us first list some obvious examples of categories and the correspond- ing notation, when it is classical. 1.2.5.a Sets and mappings: Set. 1.2.5.b Topological spaces and continuous mappings: Top. 1.2.5.c Groups and group homomorphisms: Gr. 1.2.5.d Commutative rings with unit and ring homomorphisms: Rng. 1.2.5.e Real vector spaces and linear mappings: Vectr. 1.2.5.f Real Banach spaces and bounded linear mappings: Bango. 1.2.5.g Sets and injective mappings. 1.2.5.h Real Banach spaces and linear contractions: Ban,. And so on.1.2 Categories and functors 7 Examples 1.2.6 Here is a list of some mathematical devices which can also be seen as categories. 1.2.6.a Choose as objects the natural numbers and as arrows from n to m the matrices with n rows and m columns; the composition is the usual product of matrices. 1.2.6.b A poset (X,<) can be viewed as a category 2 whose objects are the elements of X; the set 2 (x,y) of morphisms is a singleton when xz < y and is empty otherwise. The possibility of defining a (unique) composition law is just the transitivity axiom of the partial order; the existence of identities is just the reflexivity axiom. 1.2.6.c Every set X can be viewed as a category 2 whose objects are the elements of X and the only morphisms are identities. (2(z,y) is a singleton when x = y and is empty otherwise). A category whose only morphisms are the identities is called a discrete category. 1.2.6.d A monoid (M,-) can be seen as a category .@ with a single object * and M = .M@(x,*) as a set of morphisms; the composition law is just the multiplication of the monoid. As a special case, we can view any group as a category. When a ring with unit is considered as a special case of a category, the composition law of that category is generally that induced by the multiplication of the ring. Examples 1.2.7 From a given category @, one very often constructs new categories of “diagrams in @”. Here are some basic contructions. 1.2.7.a Let us fix an object I € @. The category @/I of “arrows over I” is defined by the following. © Objects: the arrows of @ with codomain I. © Morphisms from the object (f: A——>1) to the object (g: B——> 1): the morphisms h: A——> B in @ such that goh = f (the “commu- tative triangles over I”); see diagram 1.3. The composition law is that induced by the composition of @. Notice that when @ is the category of sets and mappings, a mapping f: A—>I can be identified with the I-indexed family of sets (f~'(i)),_, so that the previous category is just that of J-indexed families of sets and J-indexed families of mappings. 1.2.7.b Again fixing an object I € @, we define the category I/@ of “arrows under I”.8 The language of categories A——4—— B NA Diagram 1.3 I YN A——j 8 Diagram 1.4 e Objects: the arrows of @ with domain I. e Morphisms from the object f: 1——> A to the object g: 1—— B: the morphisms h: A——> B in @ such that ho f = g (the “commutative triangles under J”); see diagram 1.4. The composition law is induced by that of @. 1.2.7.c The category Ar(@) of arrows of @ has for objects all the arrows of ¢; a morphism from the object (f:A——>B) to the object (g: C—— D) is a pair (hk: A—>C, k: B—— D) of morphisms of @, with the property ko f = goh (“a commutative square”); see diagram 1.5. Again, the composition law is that induced pointwise by the composition in @. In examples 1.2.7.a,b,c, it is easy to check that when @ is small, so are the three categories @/I, I/@, Ar(@). Examples 1.2.8 Let us finally mention some first examples of functors. 1.2.8.a The “forgetful functor” U: Ab—— Set from the category Ab of abelian groups to the category Set of sets maps a group (G,+) to the underlying set G and a group homomorphism f to the corresponding mapping f. 1.2.8.b If R is a commutative ring, let us write Mod for the cate- gory of R-modules and R-linear mappings. Tensoring with R produces a functor from the category Ab of abelian groups to Modr: — ® R: Ab————> Mod.1.2 Categories and functors 9 A—tL— Bs Cl aoa Diagram 1.5 An abelian group A is mapped to the group A ®z R provided with the scalar multiplication induced by the formula (a@r)r’ =a@(rr’). A group homomorphism f: A——> B is mapped to the R-linear mapping f @idr. 1.2.8.c We obtain a functor P: Set—— Set from the category of sets to itself by mapping a set A to its power set P(A) and a mapping f: A—— B to the “direct image mapping” from P(A) to P(B). 1.2.8.d Given a category @ and a fixed object C € @, we define a functor @(C, —):€ ———> Set from @ to the category of sets by first putting @(C, —)(A) = @(C, A). Now if f: A——>B is a morphism of @, the corresponding mapping @(C,-)(f) = 6(C, f): €(C, A) ——> @(C, B) is defined by the formula €(C, f)(9) = fog for an arrow g € ¢(C,A). Such a functor is called a “representable functor” (the functor is “represented” by the object C). 1.2.8.e Given two categories «/, @ and a fixed object B € &, we define the “constant functor to B” Ap: 8 ——>@ by Ap(A)=B, Az(f) =1210 The language of categories FA —°4— GA Ff Gf FA! —qq 7 GA Diagram 1.6 for every object A € & and every morphism f of x. 1.3 Natural transformations General topology studies, in particular, topological spaces and continu- ous functions between them. But given two continuous functions from a space to another one, there exists also the notion of a “homotopy” be- tween those two continuous functions, which allows you to “pass” from one function to the other one. A similar situation exists for categories and functors. Definition 1.3.1 Consider two functors F,G:.0 __3@ from a category & toacategory &. A natural transformation a: F = G from F to G is a class of morphisms (a4: FA——>GA) acs of @ indexed by the objects of & and such that for every morphism f: A——> A’ in %, ayo F(f) = G(f)o aa. (see diagram 1.6) It is an obvious matter to notice that, when F, G, H are functors from to and a: F => G, 8:G => H are natural tranformations, the formula (Boa)a = Baca, defines a new natural transformation Boa: F = H. That composition law is clearly associative and possesses a unit at each functor F: this is just the natural transformation 17 whose A-component is 1r4. Again a careless argument would deduce the existence of a category whose objects are the functors from « to @ and whose morphisms are the natural transformations between them. But since and @ have merely classes of objects, there is in general no way to prove the existence of a set of natural transformations between two functors! But when is small, that problem disappears and we get the following result.1.3 Natural transformations 11 Proposition 1.3.2 Let & be a small category and @ an arbitrary category. The functors from & to @ and the natural transformations between them constitute a category; that category is small as long as B is small. We prove now the first important theorem of this book. We refer to example 1.2.8.d for the description of the representable functors. Theorem 1.3.3 (The Yoneda lemma) Consider a functor F': x6 —~+Set from an arbitrary category & to the category of sets and mappings, an object A € & and the correspond- ing representable functor f(A, —):.o—— Set. There exists a bijective correspondence 6,4: Nat(sf(A, -), F) —=—> FA between the natural transformations from #(A,—) to F and the ele- ments of the set F'A; in particular those natural transformations con- stitute a set. The bijections 07,4 constitute a natural transformation in the variable A; when x is a small category, the bijections 07,4 also constitute a natural transformation in the variable F. Proof For a given natural transformation a: (A, —) => F, we define 9p, 4(@) = aa(14). With a given element a € FA we associate, for every object B € o/, a mapping T(a)p: #(A, B) ———> FB defined by 7(a)a(f) = F(f)(a). This class of mappings defines a natural transformation t(a):0(A,-) > F since, for every morphism g: B——>C in w, the relation Fgor(a)p =rT(a)oo #(A,g) (see diagram 1.7) reduces to the equality Vf € f(A, B) F(go f)(a) = Fg((Ff)(a)), which follows from the functoriality of F. 9r,a and 7 are inverse to each other. Indeed, starting from a € FA we have 9r,a(T(a)) = 7(a)a(La) = F(1a)(a) = 1ra(a) =a.12 The language of categories (A,B) 78, pp &(A,g) Fo (A, C)——> FC T(a)e Diagram 1.7 On the other hand, starting from a: (A, —) => F and choosing a mor- phism f: A—>B in #, T(6r,4(2)) (Ff) = 7(@a(1a)) p(f) = F(f)(ea(1a)) = ap (#(A, f)(1a)) =ap(f ola) = ap(f), where the third equality follows from the naturality of a. This proves the first part of the theorem. To prove the naturality of the bijections, let us consider the functor N: x6 —>Set defined by N(A) = Nat(s#(A,—), F). and for every morphism f: A——> B in @ N(f):Nat(./(A, —), F) —>Nat((B, —), F) N(f)(a) = a0 #(f,—) (see example 1.3.6.c for the definition of (f,—)). We are claiming the existence of a natural transformation 7: N => F defined by na = OF,4. Indeed, with the previous notation, (67,8 © N(f)) (a) = Or,B(a0 (f,-)) = (ac A(f,—)) ,(1B) an(f), F(f)(aa(1a)) = (apo A (A, f))(1a) = az(f). (F(f) ° Or.) (a)1.3 Natural transformations 13 Moreover, when . is a small category, it makes sense to consider the category Fun(.%, Set) of functors from wo to Set and natural transforma- tions between them. For a fixed object A € & we consider this time the functor M: Fun(.?, Set) —> Set defined by M(F) = Nat(s(A, —-), F); for a functor G: «/ ——> Set and a natural transformation 7: F > G, M(v7): Nat(¢(A, —), F) ———>Nat(#(A, —), G) is defined by M(y)(a) = yoa. On the other hand we consider the functor “evaluation in A” ev,: Fun(., Set) —> Set defined by eva(F) = FA, eva(y) =a. We claim to have a natural transformation u:M => eva defined by Lr = Op. Indeed, with the previous notation, (9c,4 © M(7)) (a) = 6c,a(7°.@) = (yoa)a(la), (eva(7) © 9F,4) (a) = ya(@a(La))- In proposition 1.3.2 we have used a first composition law for natural transformations. In fact, there exists another possible type of composi- tion for natural transformations. Proposition 1.3.4 Consider the following situation: seta Ta aku B¢ where ~, B, @ are categories, F, G, H, K are functors and a, B are natural transformations. The formula, for every A € of, (B* @)4 = Boao H(a4) = K(a4) 0 Bra defines a natural transformation Bea:HoF>KoG. called the “Godement product” of the two natural transformations a and B. Proof (8 * a), is thus defined considering diagram 1.8 which is in- deed commutative by naturality of 8. The proposition asserts, for every morphism f: A—— A’ in &, the commutativity of the outer rectangle14 The language of categories Hratle), ca Bra Boa KFA—=—~KGA K(aa) Diagram 1.8 Hratlea), no a—804 5 KGA HFf HGf KGf , , , HFA Hoa) HGA “Boar KGA Diagram 1.9 in diagram 1.9. It holds since the first square commutes by naturality of a and functoriality of H and the second square commutes by naturality of 8. The proof of the next proposition is a straightforward exercise left to the reader. Proposition 1.3.5 Consider the situation F G =O > glia ig KIB o¢ Ly M6 where Y, B, © are categories, F, G, H, K, L, M are functors and a, B, y, 6 are natural transformations. The following equality holds: (6 * 7) 0 (8 * a) = (608) *(yoa). For the sake of brevity and with the notations of the previous propo- sitions, we shall often write G@ * F instead of @ * 1p or G * a instead of lg *a. Examples 1.3.6 1.3.6.a Consider the power set functor P: Set—— Set defined in 1.2.8.c and the identity functor id: Set —-> Set. The mappings “singleton” orn: E———>P(E)1.4 Contravariant functors 15 which map an element z € E to the singleton {x} constitute a natural transformation o:id > P. 1.3.6.b Consider the category Vectg of real vector spaces and the bidual functor ()**: Vect, ———— Vectr. The canonical morphisms ov: V——>V™, vru™ for every vector space V, define a natural transformation from the iden- tity functor to the bidual functor. 1.3.6.c Consider a category and a morphism f: A——> B of o#. We obtain a natural transformation &(f,—): f(B,—) => f(A,—) between the functors represented by A and B (see 1.2.8.d) by putting, for every object C € and every morphism g € (B,C), A(f,-)o(g) = 9° f. Generally we shall write «(f,C) for the mapping #(f,—)c. 1.3.6.d Given two categories «, @ and a fixed morphism b: B—— B’, we define the “constant natural transformation on b” Ay: Ag > Ag by (Ay)a = b for every object A € wo (see 1.2.8.e for the definition of Az, Ap). 1.4 Contravariant functors If of is a small category, we know it makes sense to consider the category of functors from « to Set and natural transformations between them (see 1.3.2). In examples 1.2.8.d and 1.3.6.c we have defined a mapping Y*: J—> Fun(wa, Set), Y*(A) = #(A,-), Y*(f) = A(F,-), where A € |.| is an object of » and f is a morphism of #. It is rather obvious that, given morphisms f g A——> B—2—>C in &, we obtain the following equalities: Y*(go f) =¥*(f)o¥*(g), Y*(1z) = lye.16 The language of categories So Y* is a mapping which “reverses the direction of every morphism” , f:A——> B, _Y*(f): ¥*(B) ——"(A) , and — up to this reversing process — preserves the composition law and identities. This is what we shall call a “contravariant functor”. Definition 1.4.1 A contravariant functor F from a category # to a category & consists of the following: (1) a mapping |#| —— || between the classes of objects; the image of A € of is written F(A) or just FA; (2) for every pair of objects A, A’ € , a mapping f(A, A’) ———> BFA’, FA); the image of f € (A, A’) is written F(f) or just Ff. These data are subject to the following axioms: (1) for every pair of morphisms f € (A, A’), g € &(A’, A”), F(go f) = F(f) 0 F(9); (2) for every object A € &, F(1a) = 1a. When confusion could be possible, we shall emphasize the fact that we are definitely working with a functor in the sense of definition 1.2.2 by calling it a covariant functor. The notion of a natural transformation can easily be carried over to the contravariant case. Definition 1.4.2 Consider two contravariant functors F,G: 3B from a category & to a category B. A natural transformation a: F > G from F to G is a class of morphisms (a4: FA——>GA) ac of B indexed by the objects of and such that for every morphism f: A——> A’ in &, G(f)oa4 =a,ao F(f) (see diagram 1.10). All the results of sections 1.2 and 1.3 can be transposed to the con- travariant case; this is a straightforward exercise left to the reader. More- over, we should mention at this point that the validity of this transpo- sition can also be obtained as an application of the duality principle of section 1.10.1.4 Contravariant functors 17 FA—°4— GA Ff Gf FA! —qqy 7 GA’ Diagram 1.10 Examples 1.4.3 1.4.3.a We started this section with the example of the “contravariant Yoneda embedding” Y*: f———> Fun(&#, Set) for a small category x. 1.4.3.b Example 1.2.8.d can be “dualized”; given a category w and an object A € & we define a contravariant functor Af (—, A): x ———> Set by the formulas f(—, A)(B) = @(B, A) for every object B € x, and f(—,A)(f): d(C, A)—>.f(B, A), A(-,A)(f)(9) = 90°F for all morphisms f: B——>C and g: C——>A in #. 1.4.3.c Example 1.3.6.c can be “dualized” as well. With the previous notation we obtain a natural transformation &(-, f): f(—, B) ——> 0(-,C) for f: B——>C, by putting 8(—,f)p(h) =foh ‘a every object D and every morphism h: D—— B. Generally, we shall ‘ite w(D, f) for the mapping (-, f)p. /1.4.3.d Again using the previous notation, example 1.4.3.a itself can be “dualized”. Let us write Fun*(./, Set) for the category of contravari- ant functors from a small category 2 to Set. The “covariant Yoneda18 The language of categories embedding” is the covariant functor Y,: ———> Fun* (of, Set) defined by the formulas Y,(A) = o(-, 4), ¥.(f) = &(-, f) for every object A € & and every morphism f of x. 1.4.3.e Consider the category Rng of commutative rings with unit and the category Top of topological spaces and continuous mappings. The construction of the Zariski spectrum of a ring gives rise to a contravariant functor Sp: Rng————> Top. For a given ring A, Sp(A) is the Zariski spectrum of A, that is the set of prime ideals of A provided with the topology generated by the fundamental open subsets Oa = {P € Sp(A) la ¢P} for every element a € A. For a given ring homomorphism f: A—— B, the inverse image process maps a prime ideal of B to a prime ideal of A; therefore we get a mapping Sp(f):Sp(B) —>Sp(A), Sp(f)(P) = f-(P), which is easily proved to be continuous. 1.4.3.f£ The last example in this section is that of a contravariant functor P*: Set—— Set which coincides on the objects with the covariant functor P: Set—— Set defined in 1.2.8.c. Thus P*(X) is the power set of X and for a given mapping f:X —>Y, P*(f):P*(Y¥) ——>P*(X), P*(f)U) =f") is the inverse image mapping. 1.5 Full and faithful functors An abelian group is a set provided with some additional structure; a group homomorphism is a mapping which satisfies some additional prop- erty. So, in some vague sense, the category of abelian groups is “included”1.5 Full and faithful functors 19 in the category of sets... the expected “inclusion” being the functor de- scribed in example 1.2.8.a. But this functor is by no means injective since on the same set G, there exist in general many different abelian group structures. In fact this functor is what we shall call a “faithful functor”. Definition 1.5.1 Consider a functor F: &<—— & and for every pair of objects A, A’ € &, the mapping (A, A’) ——>@(FA,FA'), fr Ff. (1) The functor F is faithful when the abovementioned mappings are injective for all A, A’. (2) The functor F is full when the abovementioned mappings are sur- jective for all A, A’. (3) The functor F is full and faithful when the abovementioned map- pings are bijective for all A, A’. (4) The functor F is an isomorphism of categories when it is full and faithful and induces a bijection |s&|—> |@| on the classes of ob- jects. The reader will easily adapt definition 1.5.1 to the case of contravari- ant functors. Definiton 1.5.1.4 is a special instance, in the category of small categories and functors, of the general notion of isomorphism in a category. Proposition 1.5.2 The Yoneda embedding functors described in ex- amples 1.4.3.a,d are full and faithful functors. Proof In the case of the contravariant Yoneda embedding, we have to prove that given two objects A, B in a small category A, the canonical mapping f(A, B) ——— Nat(#(B,-), (A,-)), f > (f,-) is bijective. This is a special case of the Yoneda lemma (see 1.3.3) applied to the functor ./(A,—) and the object B. The case of the covariant embedding is proved in a “dual” way. Let us conclude with some terminology concerning subcategories. Definition 1.5.3 A subcategory & of a category & consists of: (1) a subclass |B| C |.f| of the class of objects, (2) for every pair of objects A, A! € xf, a subset B(A, A') C f(A, A’), in such a way that (1) f € BA, A’) and g € B(A', A”) > go f € BA, A"),20 The language of categories ra—t— op Fa Gb FA ne GB’ Diagram 1.11 (2) VA EB, 14 € BA, A). A subcategory # of # thus gives rise to an injective (and therefore faithful) inclusion functor 8—> #. Definition 1.5.4 A subcategory B of a category # is called a full subcategory when the inclusion functor 8——>.# is also a full functor. @ is thus full in «& when A, A'€ B= BA, A’) = A(A, A’). The category of sets and injections between them is a (non-full) subcat- egory of the category of sets and mappings. The category of finite sets and mappings between them is a full subcategory of the category of sets and mappings. A full subcategory can clearly be defined by just giving its class of objects. 1.6 Comma categories We indicate now a quite general process for constructing new categories from given ones. This type of construction will be used very often in this book. Definition 1.6.1 Consider two functors F: —>€ and G:8—>@. The “comma category” (F,G) is defined in the following way. (1) The objects of (F,G) are the triples (A, f,B) where AE #,BEB are objects and f: FA——>GB is a morphism of @. (2) A morphism of (F,G) from (A, f,B) to (A’, f’, B’) is a pair (a,b), where a: A——> A’ is a morphism of #, b: B——>B is a morphism " of @, and f' o F(a) = G(b) o f (see diagram 1.11). (3) The composition law in (F,G) is that induced by the composition laws of and @, thus (a’,b’) 0 (a,b) = (a’ 0.a,b' 0b).1.6 Comma categories 21 (FRa—“. a U Y G A ae @ Diagram 1.12 Va 1 gQ—— &B ul) of |G aap oo F Diagram 1.13 Proposition 1.6.2 Consider functors F: ~—>@, G:8—+>@ and their corresponding comma-category (F,G). There are two functors U: (F,G)—> &, V: (F,G) —>@ (see diagram 1.12); moreover there exists a canonical natural transformation a: FoU >GovV. Proof With the notation of 1.6.1 it suffices to define U(A, f,B) =A, V(A, f,B) = B, U(a,b) =a, V(a,b) =b. The equality F oU = GoV has no reason at all to hold in general. The natural transformation a is easily defined by a(4,7,3) = f; the fact that it is a natural transformation is just condition 1.6.1.(2). Proposition 1.6.3 In the situation and with the notations of 1.6.2, consider a category 2, two functors U': J—>&, V': J—>B (see diagram 1.13) and a natural transformation a’: FoU'>GoV’. In that case there exists a unique functor W: 9——>(F,G) such that UoW=U', VoW=V’, atW =a’.22 The language of categories Proof The conditions imposed on W indicate immediately what it should be: W(D) = (U'D, op, V'D) for an object D € 9 and W(d) = (U'd,V'd) for a morphism d of J, which already proves the uniqueness of such a W. To prove the existence, it suffices to observe that the previous formulas indeed define a functor W: J—>(F,G). We shall refer to proposition 1.6.3 as the “universal property” of the comma, category. A special but very important case of a comma category is the “cate- gory of elements” of a functor F': of — Set. Definition 1.6.4 Consider a functor F: J —— Set from a category J to the category of sets. The category Elts(F’) of “elements of F” is defined in the following way. (1) The objects of Elts(F’) are the pairs (A,a) where A € |.¢| is an object anda é FA. (2) A morphism f:(A,a)——>(B,b) of Elts(F’) is an arrow f: A—>B of A such that Ff (a) =. (3) The composition of Elts(F’) is that induced by the composition of A. Let us write 1 for the discrete category with a single object +; 1: 1—— Set, x + {*} is the functor which maps the unique object * of 1 to the singleton {+}. In other words, we view 1 as the full subcategory of Set generated by a singleton set. Since an element a € F'A can be seen as a morphism from a singleton to F'A, thus as a morphism of the type 1(A)—> F(A) in Set, the category Elts(F) is exactly the comma category (1, F). Notice that the forgetful functor ¢r: Elts(F’) —>. is defined by ¢r(A,a) = A on the objects and by ¢r(f) = f on the morphisms. Another interesting example of a comma category is the “product” of two categories. Definition 1.6.5 The product of two categories ¥ and & is the category x @ defined in the following way. (1) The objects of xf x @ are the pairs (A, B) with A € ||, BE |B| objects of #, B.1.7 Monomorphisms 23 (2) The morphisms (A, B) —>(A’, B’) of of x @ are the pairs (a,b) where a: A——> A’ is a morphism of J and b: B——>B’ is a mor- phism of 2. (3) The composition in x @ is that induced by the compositions of of and &, namely (a’,b’) 0 (a,b) = (a’ a,b! ob). With the product x B are associated the two “projection” functors pa: 8 xX B———> B, pp: A x B———>B@ defined by the formulas pa(A,B) = A, pa(A, B) = B, Pet(a,b) =a, pa(a,b) = db. These data satisfy the following “universal property” . Proposition 1.6.6 Consider two categories and &. For every cate- gory 2 and every pair of functors F: 9 —> #, G: J—> &, there exists a unique functor H: 9 —> 0 x @ such that pyoH =F, pgoH =G. Proof H is the functor defined by H(D) =(FD,GD) for an object D of J, H(d) = (Fd,Gd) for a morphism d of 2. Let us now observe the existence of a unique functor Ay: —>1: this is the “constant functor” to the unique object of 1 (see 1.2.8.e). Since 1 has just one single mapping, the comma category (A,y, Ag) is isomorphic to the product category x @. Proposition 1.6.6 is then a particularization of proposition 1.6.3. A point of terminology: a functor F:./ x @——>@ defined on the Product of two categories is generally called a “bifunctor” (a functor of two “variables” ). 1.7 Monomorphisms When a composition law appears in some mathematical structure, spe- cial attention is always paid to those elements which are “cancellable” or “invertible” for that composition. This section is devoted to the study of left cancellable morphisms in a category.24 The language of categories Definition 1.7.1 A morphism f: A——B in a category @ is called a monomorphism when, for every object C € @ and every pair of mor- phisms g, h: C_3A, the following property holds: (fog = foh) = (g=h). We shall generally use the symbol f: A-—>B to emphasize the fact that f is a monomorphism. Proposition 1.7.2 In a category @, (1) every identity morphism is a monomorphism, (2) the composite of two monomorphisms is a monomorphism, (3) if the composite ko f of two morphisms is a monomorphism, then f is a monomorphism. Proof We use the notation of 1.7.1 and consider another morphism k: B—> D. (1) is obvious. (2) If f and k are monomorphisms, kofoh=kofogs>foh=fog>h=g. (3) If ko f is a monomorphism, fog=fohskofog=kofoh>g=h. The following terminology is rather classical. Definition 1.7.3 Consider two morphisms f: A——> B and g: B—> A in a category. When go f = 14, f is called a section of g, g is called a retraction of f and A is called a retract of B. Proposition 1.7.4 In a category, every section is a monomorphism. Proof By 1.7.2.(1,3). Let us now say a word about the effect of a functor on a monomor- phism. Definition 1.7.5 Consider a functor F:. 7 —>&. (1) F preserves monomorphisms when, for every morphism f of &, f monomorphism => Ff monomorphism. (2) F reflects monomorphisms when, for every morphism f of , Ff monomorphism => f monomorphism.1.7 Monomorphisms 25 Proposition 1.7.6 A faithful functor reflects monomorphisms. Proof Consider a faithful functor F': x —~+ @, a morphism f: A——> A’ in &, and suppose Ff is a monomorphism in #. Choose another object A” € of and two morphisms g, h: A” —>A in of. fog=foh>FfoFg=FfoFh => Fg=Fh =>g=h where the second implication holds since Ff is a monomorphism and the last one follows from the faithfulness of F. Examples 1.7.7 1.7.7.a In the category Set of sets and mappings, the monomorphisms are exactly the injections. Indeed, an element a € A can be viewed as a mapping @: {*}—> A from the singleton to A; therefore, given a monomorphism f: A——> B and elements a, a’ € A, f(a) = f(a’) > foa=fod >a=a >a=a'. Conversely, if f: A—— B is injective and g,h:C ___; A are mappings such that f og = f oh, then for every element ce C fog=foh= f(g(c)) = f(h(O) = g(c) = h(c) and therefore g = h. 1.7.7.b In the category Top of topological spaces and continuous map- Pings or its full subcategory Comp of compact Hausdorff spaces, the Monomorphisms are exactly the continuous injections. Indeed, an el- ement of a space A corresponds to a continuous mapping {*x}—>A from the singleton to A; therefore the argument of 1.7.7.a can be carried Over. 1.7.7.c In the categories Gr of groups and Ab of abelian groups, the monomorphisms are exactly the injective group homomorphisms. The argument is again analogous, using now the bijective correspondence between the elements a € G of a group and the group homomorphisms @: Z—+4G from the group of integers to G; we recall the correspondence: G(z) =z-a, a=a(1).26 The language of categories 1.7.7.4 In the category Rng of commutative rings with a unit, the monomorphisms are exactly the injective ring homomorphisms. Repeat the argument using now the ring homomorphisms with domain the ring Z[X] of polynomials with integral coefficients: an element r € R of a ring R corresponds to the ring homomorphism 7: Z[X] > R mapping the polynomial p(X) to p(r); conversely r = 7(X). 1.7.7.e In the category Modp of right modules on a ring R with unit, the monomorphisms are exactly the injective R-linear mappings. Use again the same argument using the R-linear mapping with domain the ring R itself: an element m € M of a R-module M corresponds to the linear mapping ™: R——> M mapping r to mr; conversely m = 7(1). 1.7.7.£ In the category Ban, of real Banach spaces and linear contrac- tions, the monomorphisms are exactly the injective linear contractions. The elements of the unit ball of a Banach space B are in bijective cor- respondence with the linear contractions a: R—— B; just put ar) =ra, a=(l1). Therefore a monomorphism f: B——> B’ is such that the implication f(a) = f(a’) = a=a' holds for elements a, a’ in the unit ball of B; by linearity of f, this fact extends to arbitrary elements a, a’ € B. The converse is once more obvious. 1.7.7.g The previous examples could give the wrong impression that, in “concrete” examples, a monomorphism is always exactly an injective morphism. This is false as shown by the following counterexamples. We give first an “algebraic” counterexample. Consider the category Div of divisible abelian groups and group homo- morphisms between them. The quotient morphism q:Q——>-Q/Z of the additive group of rational numbers by the group of integers is definitely not an injection, but it is a monomorphism in Div. Indeed, choose G a divisible group and f, g:G——3Q two group homomorphisms such that qo f =qog. Putting h = f —g we have goh = 0 and the thesis becomes h = 0. Given an element z € G, h(z) is an integer since goh = 0. If h(x) £0 note that1.8 Epimorphisms 27 and therefore (qoh) (ais) #° which is a contradiction. 1.7.7.h Let us give now a “topological” counterexample. We consider the category whose objects are the pairs (X,x) where X is a connected topological space and x € X is a base point; in this category, a morphism f:(X,x2)—>(Y, y) is a continuous mapping f: X —>Y which preserves the base points, i.e. such that f(r) = y. Let us consider the projection x of the circular helix on the circle S?, nm: (H, hk) ———>(S1, 8), with h € H and s = n(h). If f:(X,z)——(S", s) is a morphism in our category which admits a “lifting” g: (X, x) ———> (Hh) through the projection 7, that lifting is necessarily unique (see Spanier, page 67). But this expresses exactly the fact that 7 is a monomorphism. 1.8 Epimorphisms We now turn our attention to right cancellable morphisms in a category. Definition 1.8.1 A morphism f: B—— A in a category @ is called an epimorphism when, for every object C € @ and every pair of morphisms g,h: A__3C, the following property holds: (go f =ho f) > (g=h). We shall generally use the notation {: B—» A to emphasize the fact that f is an epimorphism. Proposition 1.8.2 In a category @, (1) every identity morphism is an epimorphism, (2) the composite of two epimorphisms is an epimorphism, (3) if the composite f ok of two morphisms is an epimorphism, then f is an epimorphism. Proof We use the notation of 1.8.1 and consider another morphism k: D—> B. (1) is obvious.28 The language of categories (2) If f and k are epimorphisms, hofok=gofok>hof=gof>h=g. (3) If f ok is an epimorphism, g°f=hofsgofok=hofoksg=h. Proposition 1.8.3 In a category, every retraction is an epimorphism. Proof By 1.8.2.(1,3). Transposing definition 1.7.5 to the case of epimorphisms, we obtain Proposition 1.8.4 A faithful functor reflects epimorphisms. Proof Consider a faithful functor F: .~ —>@, a morphism f: A’ —> A and suppose F'f is an epimorphism in %. Choose another object A” € & and two morphisms g,h: A__; A” in . Then gof=hog>FgoFf=FhoFf =>Fg=Fh =g=h, where the second implication holds since F’f is an epimorphism and the last one follows from the faithfulness of F’. The similarity of the previous proofs with those of section 1.7 is strik- ing: this is a special instance of the “duality principle” described in section 1.10. Examples 1.8.5 1.8.5.a In the category Set of sets and mappings, the epimorphisms are exactly the surjective mappings. Choose f: A——>B a surjective map- ping and g, h: B_3.C two mappings such that go f = ho f. For every element 6 € B, we can find an element a € A such that f(a) = b; therefore 9(b) = g(f(a)) = A(F(a)) = h(d), which proves the equality g = h. Conversely, if f: A—— B is an epimorphism, consider the two-element set {0,1} and the following mappings g, h: B——> {0, 1}: g(b) =1if be f(A), g(b) = 0 if b ¢ F(A), h(b) = 1 for every be B.1.8 Epimorphisms 29 Clearly go f =ho f is the constant mapping on 1; therefore g = h and f(A) =B. 1.8.5.b In the category Top of topological spaces and continuous map- pings, the epimorphisms are exactly the surjective continuous mappings. The previous proof applies when {0,1} is provided with the indiscrete topology. 1.8.5.c In the category Haus of Hausdorff topological spaces and con- tinuous mappings between them, the epimorphisms are exactly the con- tinuous mappings with a dense image. We recall that a continuous map- ping f: A——B has a dense image precisely when every element b € B is the limit of a net of elements of f(A), i.e. a set of elements indexed by a filtered poset (see 2.13.1); when B is a Hausdorff space, the limit of a converging net is unique. Suppose f: A——>B has a dense image and choose g,h: Bc such that go f = ho f. Given an element b € B, choose a net (a;)icr of elements in A such that b = lim f(a;). By continuity of g,h we have g(b) = lim(g 0 f)(a;), h(b) = lim(ho f)(a;). Since go f = ho f and the limit is unique, we conclude that g(b) = h(b) and thus g = h. Conversely if f: 4A——B is an epimorphism, and B is not empty, A cannot be empty. Indeed if BIIB is the space constituted by two disjoint copies of B, BIB is a Hausdorff space and the two canonical inclusions 4,12: B =3B Il B are continuous and distinct. A empty would yield 4,0 f =ig0f and thus 7; = ig, since f is an epimorphism. Now consider the quotient of B which identifies with a single point the closure f(A) of the image of A; this is a Hausdorff space as a quotient of a Hausdorff space by a closed subspace; write p: B——> B/ f(A) for the corresponding continuous projection. Since f(A) is not empty, we can consider as well the constant mapping g: B——> B/'f(A) on the equivalence class of the elements of f(A). Clearly po f = qo f and therefore p = g, which proves the equality f(A) = B. 1.8.5.d In the category Gr of groups and their homomorphisms, the epimorphisms are exactly the surjective homomorphisms. Indeed, a sur- jective homomorphism is clearly an epimorphism. Conversely suppose f: A—> B is an epimorphism. We can factor f through its image A———> f(A) ———> B, thus through a surjection followed by an injection. By 1.8.2.(3), the30 The language of categories injective part is an epimorphism and so the problem reduces to proving that an epimorphic inclusion is an identity. Given two groups G, H with a common subgroup K, it is possible to construct the amalgamation of G and H over K: this is the group G «x H of words constructed with the “letters” of G and H, the two copies of a “letter” of K being identified in G*x% H. The amalgamation property for groups tells us that the two canonical morphisms G——>G «x H, H———>G kx H, are injective and that two “letters” of G and H are identified in Gxx H just when they are the two copies of a “letter” in K (see Kurod). If we apply that amalgamation property choosing the inclusion f(A) @ B twice, we first deduce the equality of the two canonical inclusions i: B——> B kya) B, to: B———> B kg(a) B since they coincide on f(A) and f(A)——>B is an epimorphism. But then each element of B is already in f(A) by the amalgamation property. 1.8.5.e Consider a ring R with unit. In the category Mod, of right R- modules, the epimorphisms are exactly the surjective linear mappings. In particular, choosing R = Z, the epimorphisms of the category of abelian groups are exactly the surjective homomorphisms. Again a surjective linear mapping is clearly an epimorphism. Conversely if f: A——> B is an epimorphism, consider both the quotient mapping and the zero mapping p: B———> B/ f(A), 0: B————> B/ f (A). From the equality pof=0=0of we deduce p = 0 and thus B = f(A). 1.8.5.f The form of epimorphisms in the category of commutative rings with unit is known (see exercise 1.11.13); let us just emphasize the fact that epimorphisms of rings are not necessarily surjective. Consider the inclusion of the ring Z of integers in the ring Q of rational numbers, i:Z—>Q. This is clearly not a surjection but it is an epimorphism of rings. Indeed given another ring A and two ring homomorphisms f,g:Q—_3 A which agree on the integers, we deduce first that for every integer 0 4 z € Z, z is invertible in Q and therefore f(z) and g(z) are invertible in A; clearly way ~# (=) aay 722):1.9 Isomorphisms 31 Since f and g agree on the integers, f (+) = g (+) and finally, By a 2 1 £(2) =1(2-2) =10-4() =9(2')-9 (:) =a(#' :) =9(=). 1.8.5.g In the category Ban, of Banach spaces and linear contractions, the epimorphisms are the linear contractions with dense image. Choose f: A—— B with a dense image and g, h: BC such that gof =hof. Since g and h agree on f(A), by continuity g, h agree on on f(A) = Bas well; therefore g = h. Conversely if f: A——>B is an epimorphism, the quotient of B by the closed subspace f(A) is a Banach space and both the quotient mapping p and the zero mapping are linear contractions: p: B——— B/ f(A), 0: B———> B/f (A). From the equalities po f = 0 = 0of, we deduce p = 0 and thus B = f(A). 1.9 Isomorphisms We consider finally the case of those morphisms of a category which are invertible. Definition 1.9.1 A morphism f:A——B in a category @ is called an isomorphism when there exists a morphism g: B——>A of @ which satisfies the relations fog=1p, gof=la. Clearly such a morphism g is necessarily unique; indeed if h: B——>A is another morphism with the same properties foh=1p, hof=la, we conclude that g=g°lp=gofoh=l1,4oh=h. Therefore we shall call such a morphism g “the” inverse of f and we shall denote it by f-!. Proposition 1.9.2 In a category, (1) every identity is an isomorphism, (2) the composite of two isomorphisms is an isomorphism, (3) an isomorphism is both a monomorphism and an epimorphism.32 The language of categories Proof (1) is obvious. (2) If f: A—>B and g: B—->C are isomorphisms, so is go f and (go f)* = frog”. (3) is just the conjunction of 1.7.4 and 1.8.3 Proposition 1.9.3 In a category, if a section is an epimorphism, it is an isomorphism. Proof If gof =1, and f: A——B is an epimorphism, from fogof = f we deduce fog =1g. Proposition 1.9.4 Every functor preserves isomorphisms. Proof Obvious. Transposing definition 1.7.5 to the case of isomorphisms, we obtain Proposition 1.9.5 A full and faithful functor reflects isomorphisms. Proof Obvious. Examples 1.9.6 1.9.6.a In the category Set of sets, the isomorphisms are exactly the bijections. 1.9.6.b In the category Top of topological spaces, the isomorphisms are exactly the homeomorphisms. Since a continuous bijection is in general not a homeomorphism, this provides an example where the converse of statement 1.9.2.(3) does not hold (see 1.7.7.b and 1.8.5.b). 1.9.6.c In the categories Gr of groups, Ab of abelian groups and Rng of commutative rings with unit, the isomorphisms are the bijective ho- momorphisms. 1.9.6.d In the category Modp of right modules over a ring R, the isomorphisms are the bijective R-linear mappings. 1.9.6.e In the category Ban, of real Banach spaces and bounded linear mappings, the isomorphisms are the bounded linear bijections. An iso- morphism is obviously bijective. Conversely if f: A——> B is a bounded linear bijection, the inverse mapping f~!: B—— A is certainly linear. By the open mapping theorem, f is open because it is surjective; but “f open” means precisely “f—! continuous” and thus f~1 is bounded.1.10 The duality principle 33 1.9.6.f In the category Ban; of real Banach spaces and linear contrac- tions, the isomorphisms are exactly the isometric bijections. An isometric bijection is obviously an isomorphism. Conversely if the linear contrac- tion f: A——B has an inverse mapping f~!: B—— A which is also a linear contraction, then for every element a € A llall = IF F(@)I < FOI and thus |la|| = || f(@)|| since f is contracting. 1.9.6.g In the category Cat of small categories and functors, the iso- morphisms are those defined in 1.5.1. 1.9.6.h Going back to example 1.2.6.d, a group can be seen as a cat- egory with a single object all of whose morphisms are isomorphisms. 1.10 The duality principle At this point the reader will have noticed that every result proved for co- variant functors has its counterpart for contravariant functors and every result proved for monomorphisms has its counterpart for epimorphisms. These facts are just special instances of a very general principle. Definition 1.10.1 Given a category , the dual category #* is defined in the following way: (1) |sf*| = |s2| (both categories have the same objects); (2) for all objects A, B of xf*, o6*(A,B) = (B, A) (the morphisms of «f* are those of “written in the reverse direc- tion”; to avoid confusion, we shall write f*: A——>B for the mor- phism of s* corresponding to the morphism f: B——>A of sf); (3) the composition law of s/* is given by frog" =(gof)*. Metatheorem 1.10.2 (Duality principle) Suppose the validity, in ev- ery category, of a statement expressing the existence of some objects or morphisms or the equality of some composites. Then the “dual state- ment” is also valid in every category; this dual statement is obtained by reversing the direction of every arrow and replacing every composite f og by the composite g o f. Proof If S denotes the given statement and S* denotes its dual state- ment, proving the statement S* in a category is equivalent to proving34 The language of categories the statement S in the category *, and this is supposed to be valid. For example, the notion of f: A——> B being a monomorphism in means VC € oA Vg,hEe S(C,A) fog=foh>g=h. The dual notion is thus that of a morphism f: B——> A which satisfies VCE oS VghEe A(A,C) gof=hofs>g=h ... which is exactly the notion of an epimorphism. With that remark in mind, it is obvious that all the results of section 1.8 are just the dual statements of the results of section 1.7: so, formally, the validity of the latter follows at once from the validity of the former via the duality principle. The case of contravariant functors can also be reduced to the case of covariant functors via the consideration of the dual category: a con- travariant functor from # to & is just a covariant functor from * to B (or, equivalently, a covariant functor from # to B*). It is interesting to notice that, in category theory, some notions are their own dual. For example f: A—— B is an isomorphism when dg: B——>A gof=la, fog=lp. The dual notion is that of a morphism f: B—— A with the property dg: A——— B fog=l1y,4, gof=lsB ... but this is again the definition of f being an isomorphism. Examples 1.10.3 1.10.3.a With every category we can associate a bifunctor, still written /, A: A* x of ——— Set, defined by the following formulas: e (A, B) is the set of morphisms from A to B; e if f: A/—-A and g: B——> B’ are morphisms of /, A (f,9): f(A, B)——>. f(A’, B’), A(f,g)(h) =gohof. Fixing the first variable A we obtain the covariant functor defined in 1.2.8.d and fixing the second variable B we obtain the contravariant functor defined in 1.4.3.b. The bifunctor .¥ is called the “Hom-functor”1.10 The duality principle 35 of the category x (from “homomorphism”); it is “contravariant in the first variable and covariant in the second variable” . 1.10.3.b The dual of the category of sets and mappings is equiva- lent to the category of complete atomic boolean algebras and (V — A)- preserving homomorphisms. Indeed, writing CBA for the second cate- gory, the contravariant power set functor can be seen as a contravariant functor P*: Set——> CBA. It is well-known that every complete atomic boolean algebra B is isomorphic to the power-set PX of its set X of atoms. Let us prove now that P* is a full and faithful functor. Given two sets X and Y, the mapping Set(X, Y) > CBA(P"Y, P*X), fro fo? is obviously injective. To prove it is surjective, let us consider a morphism g: P*Y —>P*X in CBA and an element x € X = g(Y) (g preserves the top element). Now Y is the union of its singletons and g preserves unions, so there exists some y € Y such that z € g({y})- Such an element y is necessarily unique since x € g({y’}) with y’ # y would imply x € o({y} 9 {y'}) = 9(0) = 0, because g preserves intersections and the bottom element. Writing f(z) for that element y, it follows easily that g is just f—!. 1.10.3.c The dual of the category of abelian groups and their homo- morphisms is equivalent to the category of compact abelian groups and continuous homomorphisms. This is just the Pontryagin duality theo- rem: with every abelian group A is associated its group of characters A= Hom (A,U) where U is the circle group and the topology of A is that induced by the product topology U4; with every homomorphism f: A—+B is associated the morphism f: B—— A of composition with f. 1.10.3.d The category of finite abelian groups and their homomor- Phisms is equivalent to its own dual category. Indeed, it suffices to par- ticularize the Pontryagin duality to the case of finite groups: when A is finite, A is isomorphic to A as a group and therefore is finite. But the finite compact groups are just the finite discrete groups, thus finally just the finite groups.36 The language of categories 1.11 Exercises 1.11.1 If two ordered sets A, B are viewed as categories (see 1.2.6.b), prove that a functor from A to B is just an order preserving mapping. If f,g: A__3B are two such functors, prove that there exists a (single) natural transformation from f to g if and only if for every element a € A, f(a) < g(a). 1.11.2 If two monoids M and N are viewed as categories (see 1.2.6.d), prove that a functor from M to N is just an homomorphism of monoids. What is a natural transformation between two such functors? 1.11.3 In exercise 1.11.2, if M and N are groups, show the existence of a natural transformation between two functors f,g: M—_;N if and only if f and g are conjugate: Ine N Wm eM f(m)=n7!-g(m)on. 1.11.4 If G is a group considered as a category (see 1.9.6.h), prove that a natural transformation on the identity functor of G is just an element of the centre of G. 1.11.5 Prove that a covariant representable functor preserves monomor- phisms. 1.11.6 Prove that a contravariant representable functor maps an epi- morphism to a monomorphism. 1.11.7 Prove that the forgetful functor Rng—— Set which maps a ring to its underlying set is faithful and representable by the ring Z[X], but does not preserve epimorphisms. (Hint: see 1.8.5.f.] 1.11.8 If , B, @ are small categories, prove the isomorphism of cate- gories Fun(f x B,@) = Fun(./, Fun(@,@)), where Fun denotes the category of functors and natural transformations. 1.11.9 Prove that a retraction which is a monomorphism is necessarily an isomorphism. 1.11.10 Determine the nature of the monomorphisms, epimorphisms and isomorphisms in examples 1.2.7. 1.11.11 Consider a small category and the corresponding functor category Fun(.¥, Set). Prove that a morphism a of Fun(.¢, Set) (a natural transformation) is a monomorphism if and only if each component aa, A € &, is a monomorphism in Set. [Hint: use the Yoneda lemma].1.11 Exercises 37 ol : x —t + y A ——_5 B Diagram 1.14 1.11.12 The statement in 1.11.11 is no longer valid when Set is replaced by an arbitrary category @. Consider the categories of diagram 1.14 (as a convention, identity arrows are not shown) where, in #, the two compos- ites fog and f oh are equal to k. The category Fun(.#/, Z) is the category of arrows of B (see 1.2.7.c). The pair (1g, f): (B, 18, B) —>(B, f,C) is a monomorphism in Fun(./, &) while f is not a monomorphism in &. 1.11.13 Consider the category Rng of commutative rings with unit. A morphism f: A—— B is an epimorphism precisely when given any element b € B, the equality 1@b=6@ 1 holds in B @, B. This is also equivalent to saying that the morphism B—— B @, B is surjective, or again equivalently is an epimorphism.Limits We have seen in chapter 1 that the models of a mathematical theory and the corresponding homomorphisms very often constitute an interesting example of category: the category of sets and mappings, the category of vector spaces and linear mappings, the category of topological spaces and continuous mappings, and so on. With a given mathematical structure are very often associated “op- erations on models or homomorphisms”: cartesian product, quotient, kernel, union, intersection, and so on... It is the aim of this chapter to develop a general theory containing most of those constructions as particular cases. 2.1 Products Everybody knows how to construct the cartesian product of two sets A and B; this is just Ax B= {(a,b)|ace A; be B}. This “cartesian product” is provided with two “canonical” projections pa: Ax B—>A, pa(a,b) =a, pp: Ax B—>B, pp(a,b) =b. Moreover, if C is a set and f:C——~ A, g:C——>B are arbitrary map- Pings, there exists a unique mapping h: C——> A x B which makes dia- gram 2.1 commutative. Indeed, h(c) = (f(c), 9(c))- Replacing “set” by “category” and “mapping” by “functor”, the sit- uation of diagram 2.1 recaptures precisely the fundamental property of the “product of two categories” as studied in 1.6.5 and 1.6.6. This fact is much more general and makes sense in every category. 382.1 Products 39 PA PB Diagram 2.1 Definition 2.1.1 Let @ be a category and A, B € @ two objects of @. A (cartesian) product of A and B is, by definition, a triple (P,pa,pp) where (1) P € @ is an object, (2) pa: P——A and pg: P—>B are morphisms, and this triple is such that for every other triple (Q,qa,qB) where (1) Q € @ is an object, (2) ga: Q——A and gg: Q——> B are morphisms, there exists a unique morphism r: Q——> P such that qa = paor and 9B =PpBor. It is a fundamental observation that: Proposition 2.1.2 In a category, the cartesian product of two objects (when it exists) is unique up to isomorphism. Proof In the category @ , consider two products (P,pa,pp) and (Q, 44,98) of the same objects A, B. Since (P,pa,ps) is a product, there exists a morphism r:Q——>P such that ga = paor and gg = pp or. Since (Q,4qa, qe) is a product, there exists a morphism s: P——>Q such that PA = 4a08 and pp = QB Os. Applying definition 2.1.1 to the triple (P,p,, pg) seen both as a prod- uct and as “another triple”, we deduce the existence of a unique mor- phism t: P——> P such that pa = pag ot and pg = ppot. Clearly t = 1p40 Limits such a morphism. But the relations PA=QA°S=Pa°Tos, PB= qB°S=pporos indicate that t = ros is another morphism of this kind. The uniqueness of t therefore implies ro s = 1p. In an analogous way sor = 1g andr, s are inverse isomorphisms. In view of the previous theorem, it makes sense to write “the” product of A and B as (A x B,pa,pp) or just A x B. It is a common mistake to think that the projections p,, pg of a product are epimorphisms. This is not true, not even in the category of sets and mappings! The projection pa: A x B——>A in the category of sets is not surjective when B is empty and A is non-emtpy; thus it is not an epimorphism (see example 1.8.5.a). Another common mistake is to think that once the object A x B in a product has been fixed, the corresponding projections pa,pg are necessarily unique. This is not true at all (see example 2.1.7.i): in the definition of a product both the object A x B and the projections p4, pp are defined up to isomorphism. Proposition 2.1.3 In a category, when the corresponding products exist, the following isomorphisms hold: AxB2=BxA; Ax(BxC)=(Ax B)xC. Proof If (Ax B,pa,pp) is a product of A and B, it suffices to notice that (A x B,pg,pa) is a product of B and A and apply proposition 2.1.2. An analogous argument holds in the second case. Writing (A x B,p4,pp) and (B x A,p',,p',) for the products of A, B and B, A, the unique morphism 7: Ax B——> Bx A such that pa = plot and pg = pz oT is generally called the “twisting isomorphism”. It is indeed an isomorphism as proved in 2.1.3. In the case of the category Set of sets and mappings, it is the usual twisting bijection 7(a, b) = (b, a). Proposition 2.1.3 indicates in particular that the existence of all binary products allows the definition (up to isomorphism) of the product of n objects of the category (n € N,n > 2). In fact, it makes sense to define the product of an arbitrary family of objects in a given category. Definition 2.1.4 Let I be a set and (C;)ier a family of objects in a given category @. A product of that family is a pair (P,(pi)ier) where