Lectures on Homological Algebra. Weizhe Zheng

Transcription

1 Lectures on Homological Algebra Weizhe Zheng

2 Morningside Center of Mathematics Academy of Mathematics and Systems Science, Chinese Academy of Sciences Beijing , China University of the Chinese Academy of Sciences, Beijing , China

3 Contents 1 Categories and functors Categories Functors Universal constructions Adjunction Additive categories Abelian categories Projective and injective objects Projective and injective modules Derived categories and derived functors Complexes Homotopy category, triangulated categories Localization of categories Derived categories Extensions Derived functors Double complexes, derived Hom Flat modules, derived tensor product Homology and cohomology of groups Spectral objects and spectral sequences Summary of properties of rings and modules 105 iii

4 iv CONTENTS

5 Chapter 1 Categories and functors Very rough historical sketch Homological algebra studies derived functors between categories of modules (since the 1940s, culminating in the 1956 book by Cartan and Eilenberg [CE]); abelian categories (Grothendieck s 1957 Tōhoku article [G]); and derived categories (Verdier s 1963 notes [V1] and 1967 thesis of doctorat d État [V2] following ideas of Grothendieck). 1.1 Categories Definition A category C consists of a set of objects Ob(C), a set of morphisms Hom(X, Y ) for every pair of objects (X, Y ) of C, and a composition law, namely a map Hom(X, Y ) Hom(Y, Z) Hom(X, Z), denoted by (f, g) gf (or g f), for every triple of objects (X, Y, Z) of C. These data are subject to the following axioms: (associativity) Given morphisms X f Y g Z h W, we have h(gf) = (hg)f. (unit law) For every object X of C, there exists an identity morphism id X End(X) := Hom(X, X) such that fid X = f, id X g = g for all f Hom(X, Y ), g Hom(Y, X). The morphism id X is clearly unique. Remark For convenience we usually assume that the Hom sets are disjoint. In other words, every morphism f Hom(X, Y ) has a unique source X and a unique target Y. Remark Russell s paradox shows that not every collection is a set. Indeed, the collection R of sets S such that S S cannot be a set, for otherwise R R if and only if R R. To avoid the paradox, the conventional ZFC (Zermelo Fraenkel + axiom of choice) set theory does not allow the existence of a set containing all sets or unrestricted comprehension. In category theory, however, it is convenient to introduce a collection of all sets in some sense. In NBG (von Neumann Bernays Gödel) 1

6 2 CHAPTER 1. CATEGORIES AND FUNCTORS set theory, which is an extension of ZFC set theory, one distinguishes between sets and proper classes. Another approach, which we adopt, is to assume the existence of an uncountable Grothendieck universe U. 1 Elements of U are called small sets. The following table loosely summarizes the basic terminological differences of the NBG class set proper class two approaches. ZFC + U set small set large set We will mostly be interested in categories whose Hom sets are small, which are sometimes called locally small categories. A category C is called small if it is locally small and Ob(C) is small. Example (1) Let R be a ring. category objects morphisms Set small sets maps Top small topological spaces continuous maps Grp small groups homomorphisms of groups R-Mod (e.g. Ab) small (left) R-modules homomorphisms of R-modules In all of the above examples, composition is given by composition of maps. (2) Any partially ordered set (S, ) can be regarded as a category by { } x y, Hom(x, y) = otherwise. (3) Any monoid M can be regarded as a category with one object and End( ) = M. Conversely, given any object X of a category C, End(X) is a monoid. Definition A morphism f : X Y in C is called an isomorphism if there exists a morphism g : Y X such that gf = id X and fg = id Y. The morphism g is unique and is called the inverse of f, denoted by f 1. Remark The identity map id X is an isomorphism. Isomorphisms are stable under composition. In particular, Aut(X) is a group. Definition Let f : X Y be a morphism in a category C. We say that f is a monomorphism if for every pair of morphisms (g 1, g 2 ): W X satisfying fg 1 = fg 2, we have g 1 = g 2. We say that f is an epimorphism if for every pair of morphisms (h 1, h 2 ): Y Z satisfying h 1 f = h 2 f, we have h 1 = h 2. In other words, f is a monomorphism if and only if the map Hom(W, X) Hom(W, Y ) carrying g to fg is an injection; f is an epimorphism if and only if the map Hom(Y, Z) Hom(X, Z) carrying h to hf is an injection. Remark One can show that a morphism in Set, Top, Grp, or R-Mod is a monomorphism (resp. epimorphism) if and only if it is an injection (resp. surjection). See Problems for Grp. There are epimorphisms that are not surjective in the category Ring of small rings and in the category HausTop of small Hausdorff topological spaces. 1 A Grothendieck universe U is a set satisfying the following conditions: y x U implies y U; x, y U implies {x, y} U; x U implies P (x) U where P (x) is the power set of x; x i U, i I U implies i I x i U. TG (Tarski Grothendieck) set theory is obtained from ZFC by adding Tarski s axiom, which states that for every set x, there exists a Grothendieck universe U x.

7 1.2. FUNCTORS 3 Remark An isomorphism is necessarily a monomorphism and an epimorphism. The converse does not hold in general. For example, in Top, the continuous map R disc R carrying x to x, where R disc denotes the set R equipped with the discrete topology, is a monomorphism and an epimorphism, but not an isomorphism. We leave the proof of the following lemma as an exercise. Lemma Consider morphisms X f Y g Z. Then (1) If f and g are monomorphisms, then gf is a monomorphism. (2) If f and g are epimorphisms, then gf is an epimorphism. (3) If gf is a monomorphism, then f is a monomorphism. (4) If gf if an epimorphism, then g is an epimorphism. Definition The opposite category C op of a category C is defined by Ob(C op ) = Ob(C) and Hom C op(x, Y ) = Hom C (Y, X). A morphism f of a category C is a monomorphism in C if and only if it is an epimorphism in C op. 1.2 Functors Functors Definition Let C and D be categories. A functor F : C D consists of a map Ob(C) Ob(D) and, for every pair of objects (X, Y ) of C, a map Hom C (X, Y ) Hom D (F X, F Y ), compatible with composition and identity: F (id X ) = id F X for all X Ob(C) and F (gf) = F (g)f (f) for all morphisms X f Y g Z. Remark Given functors F : C D and G: D E, we have the composite functor GF : C E. For any category C, we have the identity functor id C. We can thus organize small categories and functors into a category Cat. Example (1) We have forgetful functors Top Set and R-Mod Ab Grp Set. (2) We have a functor Set R-Mod carrying a set S to the free R-module R (S) = s S Rs. (3) We have a functor H n : Top Ab carrying a topological space X to its n-th singular homology group Hn sing (X). (4) For any object X in a category C with small Hom sets, we have functors Hom(X, ): C Set and h C (X) = Hom(, X): C op Set. Definition A contravariant functor from C to D is a functor C op D. Definition Let (C i ) i I be a family of categories. The product category C = i I C i is defined by Ob(C) = i I Ob(C i ) and Hom C ((X i ), (Y i )) = i I Hom Ci (X i, Y i ). A functor C D E is sometimes called a bifunctor.

8 4 CHAPTER 1. CATEGORIES AND FUNCTORS Example (1) For any category C with small Hom sets, we have a functor Hom(, ): C op C Set. (2) Let R, S, and T be rings. We have functors S : (R, S)-Mod (S, T )-Mod (R, T )-Mod, Hom R-Mod (, ): ((R, S)-Mod) op (R, T )-Mod (S, T )-Mod, Hom Mod-T (, ): ((S, T )-Mod) op (R, T )-Mod (R, S)-Mod. Here (R, S)-Mod denotes the category of small (R, S)-bimodules, which can be identified with (R Z S op )-Mod. Natural transformations Definition Let F, G: C D be functors. A natural transformation α: F G consists of morphisms α X : F X GX in D for all objects X of C, such that for every morphism f : X Y of C, the following diagram commutes F X F f F Y α X α Y GX Gf GY. Example Let U : R-Mod Set be the forgetful functor carrying an R- module to its underlying set. Let F : Set R-Mod be the free module functor. Then the injection S R (S) carrying s to 1 s defines a natural transformation α: id Set UF. Remark Given functors F, G, H : C D and natural transformations α: F G and β : G H, we have the (vertically) composite natural transformation βα: F H. Functors C D and natural transformations form a category Fun(C, D). Isomorphisms in this category are called natural isomorphisms 2. A natural transformation α is a natural isomorphism if and only if α X is an isomorphism for every object X of C. There is also a horizontal composition of natural transformations: Given a natural transformation α: F G between functors C D and a natural transformation α : F G between functors D E, we have α α: F F G G between functors C E. This composition satisfies various compatibilities. Small categories, functors, and natural transformations, together with horizontal and vertical compositions, form a 2-category. There is an obvious notion of isomorphism of categories. A more useful notion is the following. Definition An equivalence of categories is a functor F : C D such that there exists a functor G: D C and natural isomorphisms id C GF and F G id D. 3 The functors F and G are then called quasi-inverses of each other. 2 Some authors call them natural equivalences. 3 Some authors write for equivalences and = for isomorphisms. We will write for isomorphisms and state equivalences verbally.

9 1.2. FUNCTORS 5 Quasi-inverses of a functor F are unique up to natural isomorphisms. Example Let X be a (path-connected) simply-connected space. The fundamental groupoid Π 1 (X) is equivalent to { }, but not isomorphic to { } unless X is a singleton. Remark Given a composable pair of functors C F D G E, if two of F, G, and GF are equivalences of categories, then so is the third one. If F F is a natural isomorphism of functors, then F is an equivalence of categories if and only if F is. Faithful functors, full functors Definition A functor F : C D is faithful (resp. full, resp. fully faithful) if the map Hom C (X, Y ) Hom D (F X, F Y ) is an injection (resp. surjection, resp. bijection) for all X, Y Ob(C). Lemma Let F : C D is fully faithful functor. (1) Let f : X Y be a morphism of C such that F f is an isomorphism. Then f is an isomorphism. (2) Let X and Y be objects of C such that F X F Y. Then X Y. Proof. (1) Let g be an inverse of F f and let g : Y X be such that F g = g. Then g is an inverse of f. (2) Let f : F X F Y be an isomorphism and let f : X Y be such that F f = f. By (1), f is an isomorphism. Remark There is an obvious notion of subcategory. For a subcategory of a category, the inclusion functor is faithful. A full subcategory is a subcategory such that the inclusion functor is fully faithful. Example The category Ab is a full subcategory of Grp. The forgetful functor Grp Set is faithful, but not fully faithful. Definition A functor F : C D is essentially surjective if for every object Y of D, there exists an object X of C and an isomorphism F X Y. Proposition A functor F : C D is an equivalence of categories if and only if it is fully faithful and essentially surjective. Corollary Let F : C D be a fully faithful functor. Then F induces an equivalence of categories C D 0, where D 0 is the full subcategory of D spanned by the image of F. Lemma For any category C, there exists a full subcategory C 0 such that the inclusion functor C 0 C is an equivalence of categories and isomorphic objects in C 0 are equal.

10 6 CHAPTER 1. CATEGORIES AND FUNCTORS Proof. By the axiom of choice, we can pick a representative in each isomorphism class of objects C. For every object X, pick an isomorphism α X : X R X, where R X is the representative of the class of X. Let C 0 be the full subcategory of C spanned by the representatives. Consider the functor F : C C 0 carrying X to R X and f : X Y to α Y fαx 1 : R X R Y. Let i: C 0 C be the inclusion functor. Then α X defines natural isomorphisms id C0 F i and id C if. Proof of Proposition Let F be an equivalence of categories and let G be a quasi-inverse. Since F G id, F is essentially surjective. Since GF id, G is essentially surjective. For objects X, X of C, since the composition (GF ) X,X : Hom C (X, X ) F X,X Hom D (F X, F X ) G F X,F X Hom C (GF X, GF X ) is an isomorphism, F X,X is an injection. Similarly, for objects Y, Y of D, F GY,GY is a surjection. Since G is essentially surjective, it follows that F X,X is a surjection. Now let F be a fully faithful and essentially surjective functor. We apply the lemma to C and to D. Let i: C 0 C be the inclusion functor and choose a quasiinverse j of the inclusion functor i : D 0 D. Then j F i: C 0 D 0 is fully faithful and essentially surjective, thus a surjection on objects. By Lemma (2), j F i is also an injection an objects. Thus j F i is an isomorphism of categories. Since j and i are equivalences of categories, it follows that F is an equivalence of categories. Yoneda s lemma and representable functors Let C be a category with small Hom sets. For every object X of C, consider the functor h C (X) = Hom C (, X): C op Set. Lemma (Yoneda). For every functor F : C op Set, the map given by φ(α) = α X (id X ) is a bijection. φ: Nat(h C (X), F ) F (X) We leave it as an exercise to state a dual of Yoneda s lemma. Proof. We construct the inverse ψ : F (X) Nat(h C (X), F ) by ψ(x) Y (f) = F (f)(x) for f : Y X. We have (φψ)(x) = ψ(x) X (id X ) = F (id X )(x) = x. Moreover, (ψφ)(α) Y (f) = F (f)(φ(α)) = F (f)(α X (id X )) = α Y (h C (f)(id X )) = α Y (f). Note that h C (X) is functorial in X, in the sense that we have a functor h C : C Fun(C op, Set). Corollary The functor h C is fully faithful. This functor is called the Yoneda embedding. Proof. Indeed, the map Hom C (X, Y ) Nat(h C (X), h C (Y )) given by h C coincides with the bijection ψ constructed in the proof of the lemma for F = h C (Y ). Applying Lemma , we get the following.

11 1.3. UNIVERSAL CONSTRUCTIONS 7 Corollary (1) Let f : X Y be a morphism such that h C (f): h C (X) h C (Y ) is a natural isomorphism. Then f is an isomorphism. (2) Let X, Y be objects such that h C (X) h C (Y ). Then X Y. Definition We say that a functor F : C op Set is represented by an object X of C if there exists a natural isomorphism F h C (X). 1.3 Universal constructions Initial objects, final objects, zero objects Definition Let C be a category. An object X of C is called an initial object if, for every object Y of C, there exists precisely one morphism X Y. An object Y of C is called a final (or terminal) object if, for every object X of C, there exists precisely one morphism X Y. Remark An initial object of C is a final object of C op and a final object of C is an initial object of C op. Proposition If X 1 and X 2 are initial objects of C, then there exists a unique isomorphism between them. If Y 1 and Y 2 are final objects of C, then there exists a unique isomorphism between them. Proof. By definition there exists a unique morphism f : X 1 X 2 and a unique morphism f : X 2 X 1. By the uniqueness of morphisms X 1 X 1, we have f f = id X1, and similarly ff = id X2. Thus f is an isomorphism. The case of final objects follows by duality. Example Let R be a ring. category initial object final object Set Top { } Grp {1} {1} R-Mod (e.g. Ab) {0} {0} Ring Z {0} Field none none (S, ) least element (if any) greatest element (if any) Definition If an object is both initial and final, it is called a zero object. Remark If C admits a zero object, then for every pair of objects X, Y, there exists a unique morphism X Y, called the zero morphism, that factors through a zero object. Zero objects and zero morphisms are often denoted by 0. Products, coproducts Definition Let (X i ) i I be a family of objects in C. A product of (X i ) i I is an object P of C equipped with morphisms p i : P X i, i I, satisfying the following

12 8 CHAPTER 1. CATEGORIES AND FUNCTORS universal property: for each object Q of C equipped with morphisms q i : Q X i, i I, there exists a unique morphism q : Q P such that q i = p i q. A coproduct of (X i ) i I is an object U of C equipped with morphisms u i : X i C, i I, satisfying the following universal property: for each object V of C equipped with morphisms v i : X i V, i I, there exists a unique morphism v : U V such that v i = vu i. Remark A product in C is a coproduct in C op and a coproduct in C is a product in C op. Remark For I =, a product is a final object and a coproduct is an initial object. Proposition The product of (X i ) i I, if it exists, is unique up to unique isomorphism. More precisely, if (P, (p i )) and (P, (p i)) are products of (X i ), then there exists a unique morphism f : P P such that p i = fp i for all i I. Proof. Indeed, by the universal property of product, we have unique morphisms f : P P and f : P P such that p i = fp i, p i = f p i for all i I. It follows that p i = f fp i for all i I, so that f f = id P by the universal property (applied to Q = P ). Similarly, ff = id P. Therefore, f is an isomorphism. Notation We let i I X i denote the underlying object of a product of (X i ) i I if it exists. We let i I X i denote the underlying object of a coproduct of (X i ) i I if it exists. We say that C admits finite (resp. small, etc.) products if C admits limits indexed by all finite (resp. small, etc.) set I. Example Let R be a ring. category small coproduct small product Set disjoint union Top usual product R-Mod (e.g. Ab) direct sum In the case of Top, the product P = i I X i is equipped with coarsest topology (sometimes called the Tychonoff topology) for which the projections P X i are continuous. The disjoint union U = i I X i is equipped with finest topology for which the inclusions X i U are continuous. In the categories Grp, Ring, CRing, small products are usual products. In the category CRing of small commutative rings, the coproduct of a pair of rings is tensor product. Example In the category associated to a partially ordered set (S, ), product means infimum and coproduct means supremum. In particular, if is a total order, then (S, ) admits products of pairs of objects and coproducts of pairs of objects. Remark Let (X i ) i I be a family of objects of C and let I = j J I j be a partition. If P j = i I j X i exists for each j, and P = j J P j exists, then P is a project of (X i ) i I. In particular, a category admitting products of pairs of objects admits finite nonempty products.

13 1.3. UNIVERSAL CONSTRUCTIONS 9 Remark The universal property for product can be summarized as a bijection Hom C (Q, X i ) Hom C (Q, X i ). i I i I We defined i I X i by spelling out the functor Hom(, i I X i ). Comma categories Let F : C D be a functor. For an object Y of D, we let (F Y ) denote the category defined as follows. An object of (F Y ) is a pair (X, f), where X is an object of C and f : F X Y is a morphism of D. A morphism (X, f) (X, f ) is a morphism h: X X such that f = f (F h), i.e. the following diagram commutes F X F h F X f f Y. Dually, we let (Y F ) denote the category of pairs (X, f), where X is an object of C and f : Y F X is a morphism of D. A morphism (X, f) (X, f ) is a morphism h: X X such that f = (F h)f. We have an isomorphism of categories (F Y ) op (Y F op ), where F op : C op D op. We may regard a family of objects (X i ) i I of C as one object X of the category C I Fun(I, C). Here I denotes both the index set and the associated discrete category. Let : C C I be the diagonal functor sending A to the constant functor of value A. Then a product of (X i ) i I is the same as a final object of ( X) and a coproduct is the same as an initial object of (X ). Limits, colimits The above construction generalizes to the case where I is a category. Let C and I be categories and let : C C I := Fun(I, C) be the diagonal functor carrying X to the constant functor of value X. Definition Let F : I C be a functor. A limit (also called projective limit) of F is a final object of ( F ) and a colimit (also called inductive limit) of F is an initial object of (F ). We sometimes refer to ( F ) as the category of objects of C over F (or cones to the base F ) and denote it by C /F. Dually, we sometimes refer to (F ) as the category of objects of C under F (or cones from the base F ) and denote it by C F/. Limits and colimits are unique up to unique isomorphisms. Remark Let us spell out the definition of limit. A limit of F is an object L of C equipped with morphisms p i : L F (i), i Ob(I) such that p j = F (f)p i for all morphisms f : i j and satisfying the following universal property: for every object M of C equipped with morphisms q i : M F (i), i Ob(I) satisfying q j = F (f)q i for all morphisms f : i j, there exists a unique morphism a: M L such that q i = p i a for all i. We leave it as an exercise to spell out the definition of colimit.

14 10 CHAPTER 1. CATEGORIES AND FUNCTORS Notation We let lim F or lim i I F (i) denote the underlying object of a limit of F if it exists. We let colim F or colim i I F (i) denote the underlying object of a colimit of F if it exists. (Other notation: lim for limit and lim for colimit.) Example For I = ( ), F : I C is represented by a pair of morphisms f, g : X Y with the same source and target. The limit of F is called the equalizer of the pair and denoted by Eq(f, g). The colimit of F is called the coequalizer and denoted by Coeq(f, g). If C admits a zero object and g is the zero morphism, these are called respectively the kernel and the cokernel of f: Ker(f) = Eq(f, 0), Coker(f) = Coeq(f, 0). Example For I = ( ), F : I C is given by a pair of morphisms f : A B, g : C B with the same target in C. A limit of F is called a pullback of (f, g). Let us spell out the definition. A pullback of (f, g) is a pair of morphisms a: D A, c: D C such that fa = gc and satisfying the following universal property: for every pair of morphisms a : E A, c : E C, there exists a unique morphism e: E D such that a = ae and c = ce. The universal property can be visualized as follows E a c e D c A f B. We write A B C for the underlying object of the limit. Dually, for I = ( ), a colimit is called a pushout. We say that C admits finite (resp. small, etc.) limits if C admits limits indexed by all finite (resp. small, etc.) category I. Example The category Set admits small limits. For a functor F : I Set, lim F is represented by the subset L i Ob(I) F (i) consisting of elements (x i ) such that F (f)(x i ) = x j for every morphism f : i j is small, whenever L. The same holds for limits in Grp, R-Mod, and Ring. Example The category Set admits small colimits. For a functor F : I Set with I a small category, colim F is represented by the quotient Q = F (i)/ a i Ob(I) by the equivalence relation generated by x F (f)(x) for f : i j and x F (i), whenever Q is small. Similarly, the category R-Mod admits small colimits. For a functor F : I Set with I a small category, colim F is represented by the quotient Q = F (i)/m i Ob(I) by the R-submodule M generated by x F (f)(x) for f : i j and x F (i), whenever Q is small. C g

15 1.4. ADJUNCTION 11 Definition We say that a functor F : C D preserves limits if for every functor S : I C, and every limit a: X S of S, (F Y ) = F Y F a F S is a limit of GS. Proposition Let C be a category with small Hom sets. For every object X of C, the functor Hom C (X, ): C Set preserves limits, and, in particular, small limits. Proof. Let F : I C be a functor. Then Hom C (X, lim F ) Hom C I( X, F ) lim Hom C (X, F ). (We leave it to the reader to give a more rigorous proof.) Remark None of the forgetful functors preserve finite coproducts. Universal constructions Grp Set, Ab Set, Ab Grp Let F : C D be a functor and let Y be an object of D. A universal arrow from U to Y is a final object of (F Y ). A universal arrow from Y to U is an initial object of (Y F ). Remark Note that (X 0, ɛ: F X 0 Y ) is a final object of (F Y ) if and only if the map Hom C (X, X 0 ) Hom D (F X, Y ) carrying f to the composite F X F f ɛ F X 0 Y is a bijection. Example Let U : Grp Set be the forgetful functor and let S be a small set. The free group F S with basis S, equipped with the map i: S UF S satisfies the following property: for every small group G equipped with a map f : S UG, there exists a unique homomorphism h: F S G such that f = (U h)i. Thus (F S, i) is an initial object of (S U). 1.4 Adjunction Adjunction Definition (Kan). Let C and D be categories. An adjunction is a triple (F, G, φ), where F : C D and G: D C are functors, and φ is a natural isomorphism φ XY : Hom D (F X, Y ) Hom C (X, GY ). We then say that F is left adjoint to G and G is right adjoint to F and we sometimes write φ: F G. If C and D have small Hom sets, then φ is a natural isomorphism of functors C op D Set. Example The free group functor F : Set Grp is left adjoint to the forgetful functor U : Grp Set.

16 12 CHAPTER 1. CATEGORIES AND FUNCTORS Example Let I be a category. If C admits limits indexed by I, then the limit functor lim: C I C is right adjoint to the diagonal functor : C C I. If C admits colimits indexed by I, then the colimit functor colim: C I C is left adjoint to the diagonal functor : C C I. Example Let X, Y, Z be small sets. Then we have a bijection Hom Set (X Y, Z) = Hom Set (X, Hom Set (Y, Z)) natural in X, Y, Z. This exhibits Y and Hom Set (Y, ) as adjoint functors. Example Let R, S, T be rings and consider small bimodules and R PT. We have an isomorphism of abelian groups Hom (R,T )-Mod (M S N, P ) = Hom (R,S)-Mod (M, Hom Mod-T (N, P )), Hom (R,T )-Mod (M S N, P ) = Hom (S,T )-Mod (N, Hom R-Mod (M, P )), R M S, S N T, natural in M, N, P. The first one exhibits S N and Hom Mod-T (N, ) as adjoint functors, and the second one exhibits M S and Hom R-Mod (M, ) as adjoint functors. Remark Let φ: F G be an adjunction. Then φ induces G op F op. Proposition Let φ: F G. isomorphism. Then G is determined by F up to natural Proof. Let φ : F G. Consider the natural isomorphism φ 1 φ: Hom D (X, GY ) Hom D (X, G Y ). By Yoneda s lemma, this is given by an isomorphism GY G Y, which is natural in Y by the naturality of φ and φ. The following proposition shows that the functoriality of an adjoint is automatic. Proposition Let F : C D be a functor. Assume that C and D have small Hom sets. Then F admits a right adjoint if and only if for every object Y of D, the functor h D (Y ) F = Hom C (F, Y ) is representable. Proof. We can construct an adjunction φ: F G as follows. For every object Y of D, choose an object GY of C and an isomorphism φ: h D (Y ) F h C (GY ). For every morphism Y Y, we get a morphism GY GY in C by Yoneda s lemma. We leave it to the reader to state duals of the preceding propositions. Proposition Let F : C D, F : D E, G: D C, G : E D be functors and let φ: F G, φ : F G be adjunctions. Then φφ : F F GG. Proof. Indeed we have Hom E (F F X, Y ) φ F X,Y Hom D (F X, G Y ) φ X,G Y Hom C (X, GG Y ).

17 1.4. ADJUNCTION 13 Unit, counit The naturality of φ means (1.4.1) φ(b f F a) = Gb φ(f) a, for all a: X X, f : F X Y, b: Y Y. Let η X = φ(id F X ): X GF X and let ɛ Y = φ 1 (id GY ): F GY Y. By (1.4.1), η : id C GF and ɛ: F G id D are natural transformations. We call η the unit and ɛ the counit. Note that (1.4.1) implies that F F η F GF ɛf F, G ηg GF G Gɛ G are identity transformations: ɛf F η = id F, Gɛ ηg = id G. Indeed, φ(ɛ F X F η X ) = φ(ɛ F X ) η X = η X = φ(id F X ) and the second relation is proved similarly. Note that (1.4.1) also implies that φ is determined by η by the rule φ(f) = Gf η X for f : F X Y. Moreover, φ 1 is determined by ɛ by the rule φ 1 (g) = ɛ Y F g for g : X GY. Example Let U : Ab Set be the forgetful functor and let F : Set Ab be the functor carrying S to Z (S) = s S Za s. The unit S UF S carries s to a s. The counit F UA A carries s A n s a s to s A n s s. Proposition Let C and D be categories. An adjunction (F, G, φ) is uniquely determined by each of the following data: (1) Functors F : C D, G: D C and natural transformations η : id C GF, ɛ: F G id D be such that ɛf F η = id F, Gɛ ηg = id G. (2) A functor F : C D, and for every object Y of D, a final object (GY, ɛ Y ) of (F Y ); (3) A functor G: D C, and for every object X of C, an initial object (F X, η X ) of (X G); Part (2) is in some sense a restatement of Proposition Proof. For (1), note we have seen that φ(f) = Gf η X is uniquely determined. Clearly φ a natural transformation. Put ψ(g) = ɛ Y F g. Then φψ(g) = φ(ɛ Y F g) = Gɛ Y GF g η X = Gɛ Y η GY g = g. Similarly, ψφ(f) = f. Thus φ is a natural isomorphism. (2) and (3) are dual to each other. We only treat (3). For any morphism a: X X in C, there exists a unique morphism F a: F X F X in D such that the diagram X η X GF X a GF a X η X GF X commutes. It is easy to check that F : C D is a functor. The above commutativity means that η : id C GF is a natural transformation. Then φ(f) = Gf η X is uniquely determined. Clearly φ is a natural transformation, and is a natural isomorphism by the universal property.

18 14 CHAPTER 1. CATEGORIES AND FUNCTORS Proposition Let φ: F G. Then (1) F is fully faithful if and only if the unit η : id C GF is a natural isomorphism. (2) G is fully faithful if and only if the counit ɛ: F G id D is a natural isomorphism. Proof. By (1.4.1), for f : X X, we have φ(f f) = φ(id F X f) = φ(id F X ) f = η X f. In other words, the composite Hom C (X, X ) F Hom D (F X, F X ) φ Hom C (X, GF X ) is induced by η X. Then (1) follows from Yoneda s lemma. We obtain (2) by duality. Corollary Let φ: F G. Then the following conditions are equivalent: (1) F is an equivalence of categories. (2) G is an equivalence of categories. (3) F and G are fully faithful. (4) The unit η : id C GF and counit ɛ: F G id D are natural isomorphisms. Under these conditions, F and G are quasi-inverse to each other. Proof. If F is an equivalence, then, by the proposition, id C GF, so that G is also an equivalence. By duality, (1) (2). It is clear that (4) (1)+(2) (3). By the proposition, (3) (4). Remark If F is an equivalence of categories and G is a quasi-inverse to F, then G is both right adjoint to F and left adjoint to F. Adjunction and (co)limits Proposition Let F : C D be left adjoint to G: D C. Then (1) F J : C J D J is left adjoint to G J : D J C J for any category J; (2) G preserves limits (that exist in D) and F preserves colimits (that exist in C). Proof. (1) follows from the determination of adjunction by unit and counit (Proposition (1)). For (2), by duality it suffices to show the first assertion. Let S : J D be a functor such that lim S exists. Consider the commutative square (1.4.2) C C J F D D J. F J We have Hom(X, lim G J S) Hom( X, G J S) Hom(F J X, S) = Hom( F X, S) Hom(F X, lim S) Hom(X, G lim S), so that G lim S is a limit of GS (again we leave it to the reader to give a more rigorous proof).

19 1.4. ADJUNCTION 15 In the case where C and D admit limits indexed by J, part (2) of the proposition can be paraphrased as follows: (1.4.2) induces by taking right adjoints a square C lim C J G D lim which commutes up to natural isomorphism. The rest of this section is not used elsewhere in these notes. In favorable cases one can give necessary and sufficient conditions for the existence of adjoints. Let us first give a criterion for the existence of an initial object. Theorem Let D be a category with small Hom sets and admitting small limits. Then D has an initial object if and only if it satisfies the following Solution Set Condition: there exists a small set S of objects of D that is weakly initial in the sense that for every object x of D, there exist s S and a morphism s x. Note that the condition is a set-theoretic one that is automatically satisfied if D is small (by taking S to be Ob(D)). Proof. (Copied from [ML2, Theorem X.2.1]) The only if part is clear: if i is an initial object, then S = {i} satisfies the Solution Set Condition. Let us show the if part. We let S also denote the full subcategory of D spanned by S. Let F : S D be the inclusion. We claim that i = lim F is an initial object of D. Choose, for every object x of D, a morphism f x : i x that factorizes through the projection p: i s to some s S. We get a cone i id D. Indeed, for every morphism x x, we have a commutative diagram i D J G J f x p s s x s p p f x x x, where x is the fiber product of s and s over x. We may assume that for all s S, f s : i s is the projection. We conclude by the following lemma. Lemma Let F : S D be a functor and let c: i id D be a cone such that cf : i F is a limiting cone. Then i is initial. Proof. Since c is a cone, the diagram i c i i c F s c F s F s

20 16 CHAPTER 1. CATEGORIES AND FUNCTORS commutes for every s Ob(S). Since cf is limiting, this implies c i = id i. Since c is a cone, for every morphism i x, the diagram c i i c x i x commutes, so that c x is the unique morphism i x. Theorem (Freyd Adjoint Functor Theorem). Let D be a category with small Hom sets and admitting small limits. Let G: D C be a functor. Then G admits a left adjoint if and only if G preserves small limits and satisfies the Solution Set Condition: for each object X of C, there exists a small set of objects S of (X G) that is weakly initial (namely, for every object x of (X G), there exist s S and a morphism s x in (X G)). Proof. Recall from Proposition that G admits a left adjoint if and only if (X G) admits an initial object for every object X of C. The only if part then follows from Proposition To show the if part, we apply Theorem It suffices to check, under the assumption that G preserves small limits, that (X G) admits small limits. Let I be a small category and let F : I (X G) be a functor. We write F i as (Y i, f i : X GY i ). Let Y be the limit of Y i in D. Since G preserves small limits, (f i ) determines a morphism f : X GY. It is easy to check that (X, f) is the limit of F. 1.5 Additive categories Additive categories Proposition Let A be a category with each Hom A (X, Y ) equipped with a structure of unital magma such that composition is bilinear. Let A 1 and A 2 be objects of A. Then the following conditions are equivalent. (1) A 1 A 2 exists. (2) A 1 A2 exists. Under these( assumptions, ) the morphism φ A1,A 2 : A 1 A2 A 1 A 2 described by ida1 0 the matrix is an isomorphism. Moreover, if Y Y and Y Y Y 0 id A2 exist, then Hom A (X, Y ) is a commutative monoid, and for f, f : X Y, f + f is given by the composition (1.5.1) X (f,f ) Y Y φ 1 Y Y Y Y (id Y,id Y ) Y. We denote the operation of Hom A (X, Y ) by + and the unit by 0 = 0 XY. The bilinearity of composition means the following: for f, f : X Y and g, g : Y Z, we have (a) g(f +f ) = gf +gf, (g +g )f = gf +g f; (b) g0 XY = 0 XZ, 0 Y Z f = 0 XZ. Condition (b) follows from (a) if Hom A (X, Z) is a group. If A admits a zero object, it follows from (b) that the zero morphism z XY : X Y (that factors through every zero object) is the unit of Hom A (X, Y ). Indeed, z XY = 0 Y Y z XY = 0 XY.

21 1.5. ADDITIVE CATEGORIES 17 Proof. By duality we may assume that (1) holds. Let i 1 = (id A1, 0): A 1 A 1 A 2 and let i 2 = (0, id A2 ): A 2 A 1 A 2. We show that (A 1 A 2, i 1, i 2 ) exhibits A 1 A 2 as a coproduct of A 1 and A 2. Let B be an object of A equipped with h 1 : A 1 B and h 2 : A 2 B. We put h = h 1, h 2 = h 1 p 1 + h 2 p 2 : A 1 A 2 B. Then hi 1 = h 1 p 1 i 1 + h 2 p 2 i 1 = h = h 1 and similarly hi 2 = h 2. If h : A 1 A 2 B is a morphism such that h i 1 = h 1, h i 2 = h 2. Then h = h (i 1 p 1 + i 2 p 2 ) = h 1 p 1 + h 2 p 2 = h by Lemma below. Therefore, (2) and the second assertion hold. Assume that Y Y exists. Then in (1.5.1), (id Y, id Y ) φ Y Y = p 1 + p 2. Thus f + f = (p 1 + p 2 )(f, f ) is given by (1.5.1). The diagram X (f,f ) Y Y φ 1 (p 2,p 1 ) (f,f) Y Y Y Y (id,id) Y (i 2,i 1 ) Y Y φ 1 Y Y Y Y (id,id) commutes, which implies f + f = f + f. Assume moreover that Y Y Y exists. For f, f, f : X Y, f + (f + f ) is given by the composition X (f,f +f ) Y Y φ 1 Y Y Y Y Y The diagram (f,(f,f )) Y (Y Y ) id Y φ 1 Y Y Y (Y Y ) (Y Y ) Y φ 1 Y (Y Y ) φ 1 id Y φ 1 Y Y Y Y idy Y,Y Y Y (Y Y ) Y (Y Y ) φ 1 Y,Y Y Y (Y Y ) (Y Y ) Y φ 1 Y Y,Y (Y Y ) Y commutes, because the inverse of the composition of the upper horizontal arrows and the inverse of the composition of the lower horizontal arrows are both given by id Y 0 0 the matrix 0 id Y 0. Therefore, f + (f + f ) = (f + f ) + f. 0 0 id Y Lemma Under the above notation, i 1 p 1 + i 2 p 2 = id A1 A 2. Proof. We have p 1 (i 1 p 1 + i 2 p 2 ) = p 1 i 1 p 1 + p 1 i 2 p 2 = p = p 1 and similarly p 2 (i 1 p 1 + i 2 p 2 ) = p 2. Therefore, i 1 p 1 + i 2 p 2 = id A1 A 2. Remark Under the assumptions (1) and (2) of Proposition 1.5.1, the composition of C (a,b) c,d A 1 A 2 B is c, d (a, b) = ca+db. In other words, composition is given by matrix multiplication, with (a, b) considered as a column vector and c, d considered as a row vector.

22 18 CHAPTER 1. CATEGORIES AND FUNCTORS Proposition Let A be a category admitting a zero object, finite products, and finite coproducts satisfying (*) The morphism φ Y Y : Y ( ) idy 0 Y Y Y described by the matrix is 0 id Y an isomorphism for every object Y of A. Then there exists a unique way to equip every Hom A (X, Y ) with the structure of a unital magma such that composition is bilinear. Moreover, Hom A (X, Y ) is a commutative monoid. Here 0 in the the description of φ Y Y through every zero object). denotes the zero morphism (that factors Proof. For morphisms f, f : X Y, we define f + f to be the composition The diagram X (f,f ) Y Y φ 1 Y Y Y Y (id Y,id Y ) Y. X (f,0) Y Y φ 1 Y Y Y Y (id Y,0) (id Y,id Y ) id Y 0 f Y (id Y,0) 1 Y 0 (id Y,0) Y commutes, so that f + 0 = f. Similarly 0 + f = f. This construction equips Hom A (X, Y ) with the structure of a unital magma. It is clear from the construction that (g + g )f = gf + gf. The diagram X (f,f ) Y Y φ 1 Y Y Y Y (id Y,id Y ) Y g g g g g (gf,gf ) Z Z φ 1 ZZ Z Z (id Z,id Z ) Z commutes, so that g(f + f ) = gf + gf. Moreover, 0f = 0 and g0 = 0. Thus the composition law on A is bilinear. The uniqueness and the fact that Hom A (X, Y ) is a commutative monoid follows from Proposition Definition An additive category is a category A admitting a zero object, finite products, finite coproducts, satisfying ( ) above, and such that the commutative monoids Hom A (X, Y ) are abelian groups. Remark By Proposition 1.5.1, if A is a category with each Hom A (X, Y ) equipped with a structure of abelian group such that composition is bilinear 4, and such that A admits a zero object, finite products (or finite coproducts), then A is an additive category. In an additive category, coproducts are also called direct sums and we often write instead of. 4 A category with each Hom A (X, Y ) equipped with a structure of abelian group such that composition is bilinear is called a preadditive category, or a category enriched over (Ab, ).

23 1.5. ADDITIVE CATEGORIES 19 Example Let R be a ring. Then R-Mod is an additive category. Indeed, R-Mod admits finite products, the Hom sets are naturally equipped with structures of abelian groups and composition is bilinear. Example Let B be an additive category. Let A B be a full subcategory such that for A and A in A, the direct sum A A in B is isomorphic to an object of A. Then A is an additive category by Lemma below. In particular, the full subcategory R-mod of R-Mod spanned by finitely generated R-modules is an additive category. Similarly, the full subcategory of R-Mod spanned by free left R-modules is also an additive category. Lemma Let B be a full subcategory of C and let F : I B be a functor. If p: X F exhibits X as the limit of F in C with X in B, then p exhibits X as the limit of F in B. Proof. This follows easily from the universal properties of limits. details to the reader. We leave the Example Let A be an additive category. Then A op is an additive category. For the next example, we need the general fact that limits can be computed pointwise in a functor category. More precisely, we have the following. Proposition Let C, I, P be categories and let F : I C P be a functor such that for each object p of P, F p = E p F : I C admits a limit τ p : L p F p. Here E p : C P C denotes the evaluation functor at p carrying G to G(p). Then there exists a unique functor L: P C such that L(p) = L p and p τ p gives a natural transformation τ : L F. Moreover, this τ exhibits L as a limit of F. Proof. This follows easily from the universal properties of limits. details to the reader. We leave the Corollary Let C, P, I be categories. If C admits limits indexed by I, then C P admits limits indexed by I and for each object p of P, the evaluation functor at p, E p : C P C, preserves limits indexed by I. Example Let A be an additive category and let P be a category. Then the functor category A P is an additive category. For X, Y : P A, Hom A P (X, Y ) is a subgroup of p Ob(P ) Hom A (X p, Y p ). Additive functors Proposition Let F : A B be a functor between additive categories. Then the following conditions are equivalent: (1) F preserves products of pairs of objects; (2) F preserves coproducts of pairs of objects; (3) for every pair of objects A, A of A, the map Hom A (A, A ) Hom B (F A, F A ) induced by F is a homomorphism.

24 20 CHAPTER 1. CATEGORIES AND FUNCTORS The proposition holds more generally for functors between categories satisfying Proposition (admitting zero objects, finite products, finite coproducts, and satisfying ( )). Proof. (1) (2). By the proof of Proposition 1.5.1, it suffices to show that F (id, 0) = (id, 0) and F (0, id) = (0, id). For this, it suffices to show that F carries zero morphisms to zero morphisms, or that F carries zero objects to zero objects. Note that any zero object 0 is the product of 0 and 0, with id 0 as the projections. By (1), F (0) is the product of F (0) and F (0), with id F (0) as the projections, so that F (0) is a zero object by Lemma below. By duality, we have (2) (1). (1) (3). Let f, g : A B. Then f + g = id, id (f, g), so that F (f + g) = id, id (F f, F g) = F f + F g. (3) (1). We must show that (F p 1, F p 2 ): F (A 1 A 2 ) F A 1 F A 2 is an isomorphism, where p j : A 1 A 2 A j is the projection, j = 1, 2. It is easy to check that F (i 1 )q 1 + F (i 2 )q 2 is an inverse to (F p 1, F p 2 ), where q j : F A 1 F A 2 F A j is the projection and i j : A j A 1 A 2 is the canonical morphism. Lemma Let A and B be objects of a category C with a zero object. If A B exists, with the projection p: A B A being an isomorphism, then B is a zero object. Proof. Let q : A B B be the projection. Consider the morphisms 0: B A and id B : B B. There exists a unique morphism f = (0, id B ): B A B such that pf = 0, qf = id B. It follows that f = 0, id B = 0, so that B is a zero object. Remark Even in an additive category, A B A does not imply B 0 in general. For example, if A = Z (S) is the free abelian group with an infinite basis S, then any bijection S S S induces an isomorphism A A A in the category Ab. Definition We say that a functor F : A B between additive categories is additive if it satisfies the conditions of the above proposition. We say that a subcategory A of an additive category B is an additive subcategory if A is additive and the inclusion functor is additive. If F is a functor between additive categories admitting a left or right adjoint, then F is additive. A composition of additive functors is additive. The term additive subcategory needs to be used with caution, as a subcategory of an additive category can be an additive category without being an additive subcategory. Remark Let B be an abelian category. A full subcategory A of B is an additive subcategory if and only if A admits a zero object 0 of B and for A and A in A, the direct sum A A in B is isomorphic to an object of A. Example Let A be an additive category and let F : P Q be a functor. Then the functor A Q A P induced by F is additive. Example Let A be an additive category. The functor : A A A is additive. It follows that the functor A A given by A A A is additive. Let B be an object of A. The functor B : A A is not additive unless B = 0.

25 1.6. ABELIAN CATEGORIES 21 Example Let R, S, T be rings. The functor S : (R, S)-Mod (S, T )-Mod (R, T )-Mod not additive in general. By contrast, it is additive in each variable. That is, given R M S and S N T, M S : (S, T )-Mod (R, T )-Mod and S N : (R, S)-Mod (R, T )-Mod are additive functors. The functor (S, S)-Mod (S, S)-Mod, A A S A is not additive unless S = 0 (assuming S small). Example Let A be an additive category with small Hom sets. The functor Hom A : A op A Ab is additive in each variable. 1.6 Abelian categories Kernels and cokernels In any category, the equalizer e: E X of a pair of morphisms (f, g): X Y, whenever it exists, is always a monomorphism. Indeed, if (a, b): A E are morphisms such that ea = eb, then a = b by the universal property of the equalizer. Dually, a cokernel, whenever it exists, is an epimorphism. Lemma (1) In a category with a zero object, every monomorphism f : X Y has zero kernel, and every epimorphism has zero cokernel. (2) In an additive category, every morphism of zero kernel f : X Y is a monomorphism and every morphism of zero cokernel is an epimorphism. Proof. We prove the assertions on monomorphisms and those on epimorphisms follow by duality. (1) We check that X X satisfies the universal property of a kernel of f. We have f0 0 X = 0. Let g : Z X be a morphism such that fg = 0 Z Y. Since g is a monomorphism, we have g = 0 Z X. (2) Let (g, h): Z X be morphisms such that fg = fh. Then f(g h) = 0. It follows that g h = 0, so that g = h. In an additive category, the equalizer of (f, g) is the kernel of f g and the coequalizer of (f, g) is the cokernel of f g. Remark Let F : I C be a functor. Assume that the products A = i Ob(i) F (i) and B = f : i j F (j) (f running through morphisms of I) exist. Then lim F can be identified with Eq(a, b), where a, b: A B are such that a f : A F (j) is the projection and b f : A F (i) F (f) F (j) is the composition of the projection with F (f). It follows that a category C admits small (resp. finite) limits if and only if C admits equalizers and small (resp. finite) products. Dually, C admits small (resp. finite) colimits if and only if C admits coequalizers and small (resp. finite) coproducts. Similar statements hold for preservation of limits and colimits. An additive category admitting kernels and cokernels admits all finite products and finite coproducts.

26 22 CHAPTER 1. CATEGORIES AND FUNCTORS Example Let R be a ring. The additive category R-Mod admits finite kernels and cokernels. Indeed, for a morphism f : A B, Ker(f) = f 1 (0) and Coker(f) = B/Im(f), where Im(f) denotes the image of f. Abelian categories Definition Let A be an additive category admitting kernels and cokernels and let f : A B be a morphism. We define the coimage and image of f to be Coim(f) = Coker(g), Im(f) = Ker(h), where g : Ker(f) A and h: B Coker(f) are the canonical morphisms. In the above situation, every morphism f : A B factors uniquely into A Coim(f) Im(f) B. Definition An abelian category is an additive category A satisfying the following axioms: (AB1) A admits kernels and cokernels. (AB2) For each morphism f : A B, the morphism Coim(f) Im(f) is an isomorphism. The axioms were introduced in Grothendieck s seminal 1957 Tōhoku paper [G]. The notion was introduced independently in Buchsbaum s 1954 thesis (under the name of exact category 5 ) (see also [B] and [CE, Appendix]). Example By the first isomorphism theorem, R-Mod is an abelian category for every ring R. The full subcategory of R-Mod consisting of Noetherian (resp. Artinian) R-modules is stable under subobjects and quotients, hence is an abelian category. The category R-mod of finitely-generated R-modules is an abelian category if and only if R is left Noetherian. Indeed, if I is a left ideal that is not finitely generated, then the morphism A A/I has no kernel in R-mod. Example Let A be an abelian category. Then A op is an abelian category. Example Let A be an abelian category and let P be a category. Then the functor category A P is an abelian category. Example The category of topological abelian groups is an additive category admitting kernels and cokernels, but does not satisfy (AB2). For example, let f be the map R disc R carrying x to x. Then Coim(f) = R disc and Im(f) = R. Remark The following properties follow from (AB2) and Lemma 1.6.1: (1) If a morphism is both a monomorphism and an epimorphism, then it is an isomorphism. (2) Every monomorphism is the kernel of its cokernel. (3) Every epimorphism is the cokernel of its kernel. 5 This terminology is no longer in use. In modern usage, exact category refers to a more general notion introduced by Quillen.