Arxiv:1209.3423V2 [Math.CT] 29 Jun 2019 Xml 3.6

Home , Cokernel, Commutative diagram, Exact sequence

arXiv:1209.3423v2 [math.CT] 29 Jun 2019 xml 3.6. Example sequence. exact short stable category, plete eetfils uha algebraic appl as important such have fields, and ferent categories, abelian beyond algebra ocp featctgr nteadtv aehscrystaliz has case algebraic additive on the m Quillen in The category category. of abelian exact an of than relevanconcept structure many algebraic where poorer etc., a analysis functional and algebraic nadtv aeoyedwdwt itnuse ls fk of class exac distinguished Quillen a A called with ones. pairs, endowed closest of category the those additive only as an mention such to notions [23], previous Yoneda different extends This [10]. htpsot rsrekresadplbcspeev coke preserve kern pullbacks kernels and all kernels with preserve of categories categories pushouts additive that abelian that for namely (i.e. choice right category, categories p the abelian exact maximal is the this an While be of pairs. a obviously definition should pair, the conflations kernel-cokernel develoin of a to class is which the conflation to for tri every respect the Since with from h sequences, algebra. apart exact to category, split possibility additive the the ther any by is if on it wonder structure maximality a to arbitrary exact its an natural than on is structure important exact It More greatest the is, sequences. which exact then split the are 21], [5, [19]. analysis theory visi functional approximation power, in or their those [6] attention main as e the such Quillen into applications, on rebring accounts and 5], exhaustive [2, present papers recent eral HNSAL HR XC EUNE EIEAN DEFINE SEQUENCES EXACT SHORT STABLE WHEN h uhrwudas iet hn iePetfrilluminati for Prest Mike thank to like also would author The phrases. and words Key xc aeoispoieasial etn o developing for setting suitable a provide categories Exact 2000 Date vr diiectgr a mleteatsrcue who structure, exact smallest a has category additive Every XC TUTR NA DIIECATEGORY ADDITIVE AN ON STRUCTURE EXACT etme 1 2013. 21, September : ahmtc ujc Classification. Subject Mathematics hsi ie ytesal hr xc eune,ie kerne c i.e. semi-stable a sequences, and exact kernel short semi-stable a stable st of the We consisting by category. pairs additive given any is on this structure exact Quillen imal Abstract. conflations uphsrcnl hwdteeitneo nqemax- unique a of existence the showed recently has Rump aifigcranaim seDfiiin31.Sev- 3.1). Definition (see axioms certain satisfying , diiectgr,eatctgr,(eky dmoetco idempotent (weakly) category, exact category, Additive K ter 1] n a ensmlfidb Keller by simplified been has and [15], -theory 1. ETMUCRIVEI SEPTIMIU K Introduction ter,agbacgoer n topology, and geometry algebraic -theory, 1 80,1E0 18G50. 18E10, 18E05, gdsusoslaigto leading discussions ng n oenl such cokernels and ree o quasi- for even or , okernel. oe structures model aeoishave categories t nl)[,1,20], 17, [7, rnels) dtv category. dditive l nsm new some in ble d when udy l-cokernel di h work the in ed atcategories xact ctosi dif- in ications rtcandidate first iloegiven one vial homological p ernel-cokernel s important ost elr[]and [8] Heller v natural a ave econflations se xss and exists, e aeoyis category t sil class ossible homological el-cokernel m- 2 SEPTIMIU CRIVEI it is no longer suitable for the more general setting of preabelian categories (i.e. additive categories with kernels and cokernels). It has already been observed by Richman and Walker [16, p. 522] that the class of all kernels (cokernels) in a preabelian category is not closed under pushouts (pullbacks), and so the class of all kernel-cokernel pairs cannot define an exact structure in this setting. In order to study homological algebra in preabelian categories they introduced semi-stable kernels (cokernels) as those kernels (cokernels) which are preserved by pushouts (pullbacks), and the stable short exact sequences as those kernel-cokernel pairs consisiting of a semi-stable kernel and a semi-stable cokernel. Recently, Sieg and Wegner have made use of these concepts, and showed that the class of stable short exact sequences defines the unique maximal exact structure on any preabelian category [21, Theorem 3.3]. A natural extension of the definition of semi- stable kernels and semi-stable cokernels from a preabelian category to an arbitrary additive category allowed the generalization of the above result in [3, Theorem 3.5], which shows that the class of stable short exact sequences defines the unique maximal exact structure on any weakly idempotent complete additive category. Note that many classes of additive categories, such as the accessible categories (which have a natural exact structure consisting of the pure exact sequences) [14] and the triangulated categories (which only have the trivial exact structure), are weakly idempotent complete. But there are additive categories (even exact) which are not weakly idempotent complete, for instance any category of free modules in which there exist projective modules which are not free (e.g., see [6, p. 2894]). The remaining challenge was to determine a greatest exact structure on any additive category. A step towards that direction has recently been made by Rump, which shows that there exists a greatest exact structure on any additive category [18]. His interesting approach uses a new concept of one-sided exact category (also, see [1]), by constructing the maximal left exact structure and the maximal right exact structure, and then deducing the existence of the greatest exact structure. Nevertheless, the question whether this is defined by the class of stable kernel-cokernel pairs has still remained open. The main obstacle is to prove that the semi-stable kernels and the semi-stable cokernels satisfy Quillen’s “obscure axiom” (see [3, 18]). In the present paper we establish an equivalent condition for the stable short exact sequences to define an exact structure on an additive category, and we shall show that this is not always checked. Our approach uses essen- tially the property that every additive category has an additive idempotent (or Karoubian) completion [9, p. 75], which in turn is weakly idempotent complete [2]. The technique is to employ the canonical fully faithful functor H : C → C between an additive category C and its idempotent completion C b b in order to transfer properties. One of the main ingredients is to prove that H preserves stable short exact sequences. Then we show that the stable short exact sequences define an exact structure on an additive category C if and only if C is stable under pushouts and pullbacks for (C,H) in the sense b of the forthcoming Definition 3.2. The latter condition is usually easier to check, and we illustrate this in two relevant situations. We also give some applications to categories of chain complexes and projective spectra. STABLESHORTEXACTSEQUENCES 3

Our characterization is inspired by that of an exact full subcategory C′ of an additive category C (in the sense that the restriction of an exact structure E from C to the full subcategory C′ is still an exact structure) [4, The- orem 2.6]. As already noted by Dierolf and Sieg [4], any extension-closed subcategory of an exact category is an exact subcategory in the above sense. But there are exact subcategories C′ of some exact category C which are not extension-closed in C (e.g., see [4, Remark 2.3], which argues that the full subcategory LCS of locally convex spaces of the category TVS of topolog- ical vector spaces with the exact structure given by the short topologically exact sequences is an exact subcategory which is not extension-closed in TVS), even if the Gabriel-Quillen embedding theorem allows one to realize them as extension-closed subcategories of some abelian categories.

2. Idempotent completion An additive category is called idempotent complete if every idempotent morphism has a kernel [9, Définition 1.2.1]. A remarkable result states that every additive category has an idempotent completion (also called Karoubian completion) [9, Lemme 1.2.2]. More precisely, for every additive category C, there exists an idempotent complete additive category C and a fully faithful b additive functor H : C → C. b The category C has as objects the pairs (A, p), where A is an object of C b and p : A → A is an idempotent morphism in C, and as morphisms between two objects (A, p) and (B,q) of C the morphisms f : A → B in C such that b f = qfp. The biproduct in C is given by (A, p) ⊕ (B,q) = (A ⊕ B,p ⊕ q). b The functor H : C → C is defined by H(A) = (A, 1 ) on objects A of C, and b A by H(f) = f on morphisms f in C (also see [2, Section 6]). We denote by Im(H) the essential image of H. The next remark will be important in what follows. Remark 2.1. (i) Every idempotent complete additive category is weakly idempotent complete, in the sense that every section has a cokernel, or equivalently, every retraction has a kernel [2, Lemma 7.1]. (ii) Every object of the idempotent completion (C,H) of an additive cate- b gory C is a direct summand of an object in Im(H). Indeed, for every object (A, p) of C, the morphism p : (A, p) → (A, 1 ) splits, whence (A, p) is a di- b A rect summand of H(A) = (A, 1 ), because C is weakly idempotent complete A b (e.g., see [2, Remark 7.4]). The identity morphism idA of (A, p) is p. Lemma 2.2. Let C be an additive category and let (C,H) be its idempotent b completion. Then H preserves and reflects pullbacks and pushouts. Proof. Since H is fully faithful, it reflects pullbacks and pushouts [13, Chap- ter II, Theorem 7.1]. We show that H preserves pullbacks, the preservation ′ ′ ′ of pushouts being dual. Let (B = B ×C C ,g,d ) be a pullback of some morphisms d : B → C and h : C′ → C in C. Let (D,p) be an object of C for some object D and idempotent morphism p : D → D in C, and b let α : (D,p) → H(B), β : (D,p) → H(C′) be morphisms in C such that ′ b H(d)α = H(h)β. Then α : D → B, β : D → C , α = αp and β = βp. By the pullback property there is a unique morphism w : D → B′ in C such that 4 SEPTIMIU CRIVEI gw = α and d′w = β. Then the morphism wp : (D,p) → H(B′) in C sat- ′ ′ b ′ isfies the equalities H(g)wp = α and H(d )wp = β. If w : (D,p) → H(B ) is another morphism in C such that H(g)w′ = α and H(d′)w′ = β, then ′ ′ ′ b′ w : D → B and w = w p in C. By the pullback property of d and h, it follows that w′ = w, and so w′ = w′p = wp. Hence (H(B′),H(g),H(d′)) is a pullback of H(d) and H(h) in C. b The notion of stable short exact sequence was introduced in [16] for preabelian categories, and generalized to arbitrary categories in [3] as follows. Definition 2.3. Let C be a category. A cokernel d : B → C in C is called ′ ′ ′ a semi-stable cokernel if there exists a pullback (B = B ×C C ,g,d ) of d along any morphism h : C′ → C:

d′ B′ / C′

g h d B / C such that d′ : B′ → C′ is a cokernel. The notion of semi-stable kernel is i deﬁned dually. A short exact sequence, i.e. a kernel-cokernel pair, A → d B → C is called stable if i is a semi-stable kernel and d is a semi-stable cokernel. We need two well-known result on pullbacks, whose duals for pushouts hold as well. Lemma 2.4. [11, Lemma 5.1] Consider the following diagram in a category C such that the squares are commutative and the right square is a pullback:

i′ d′ A′ / B′ / C′

f g h i d A / B / C Then the left square is a pullback if and only if so is the rectangle. Lemma 2.5. [16, Theorem 5] Let d : B → C and h : C′ → C be morphisms in a category C such that d has a kernel i : A → B, and the pullback of d and h exists. Then there exists a commutative diagram in C:

i′ d′ A / B′ / C′

g h i d A / B / C in which the right square is a pullback and i′ : A → B′ is the kernel of d′. Let us note some useful remarks, which will be freely used together with their dual versions. Remark 2.6. (i) Every semi-stable cokernel is the cokernel of its kernel by [3, Remark 2.5]. (ii) The pullback of a semi-stable cokernel along any morphism exists and is again a semi-stable cokernel by Lemma 2.4. STABLESHORTEXACTSEQUENCES 5

(iii) Every projection onto a direct summand is a semi-stable cokernel by [3, Corollary 3.3]. In particular, every isomorphism is a semi-stable cokernel. The following result is one of the keys for establishing our main theorem. Proposition 2.7. Let C be an additive category and let (C,H) be its idem- b potent completion. Then H preserves stable short exact sequences. i d Proof. Let A → B → C be a stable short exact sequence in C. Then H(i) H(d) H(A) → H(B) → H(C) is a kernel-cokernel pair in C by Lemma 2.2. We b show that H(d) is a semi-stable cokernel, the proof that H(i) is a semi-stable kernel following in a dual manner. Let γ : Z → H(C) be a morphism in C. Then Z is a direct summand of b an object in Im(H), say Z ⊕ Z′ = H(C′) for some objects Z′ of C and C′ of ′ ′ b C. The composition [ γ 0 ] = γ [ 1 0 ] : H(C ) = Z ⊕ Z → H(C) must be of the form H(h) for some morphism h : C′ → C in C. Since d is a semi-stable cokernel, we have the following commutative diagram in C: ′ ′ A i / B′ d / C′

g h A i / B d / C in which the right square is a pullback, and the rows are kernel-cokernel pairs (see Lemma 2.5). The morphism H(d′) : H(B′) → Z ⊕ Z′ is of the form [ u ] for some morphisms u : H(B′) → Z and v : H(B′) → Z′ in C. v b Then we have the following commutative diagram in C: b ′ u H(i ) [ v ] H(A) / H(B′) / Z ⊕ Z′

H(g) [ γ 0 ] H(A) / H(B) / H(C) H(i) H(d) in which the right square is a pullback and the rows are kernel-cokernel γ 0 0 pairs by Lemmas 2.2 and 2.5. Since H(d)0 = [ ] [ 1 ], by the pullback property there is a unique morphism ε : Z′ → H(B′) such that H(g)ε = 0 u 0 ′ and [ v ] ε = [ 1 ]. In particular, vε = idZ , and so v is a retraction. Since C ′ b is weakly idempotent complete, v has a kernel, say j : K → H(B ). Then H(d)H(g)j = γuj. We claim that the following commutative square: uj K / Z

H(g)j γ H(B) / H(C) H(d) is a pullback of H(d) and γ in C. To this end, let a : D → H(B) and b γ 0 b b : D → Z be morphisms such that H(d)a = γb. Then H(d)a = [ ] 0 , hence the pullback square of H(d) and [ γ 0 ] yields the existence of a unique ′ u b morphism δ : D → H(B ) such that H(g)δ = a and [ v ] δ = 0 . Since 6 SEPTIMIU CRIVEI vδ = 0 and j = ker(v), there is a unique morphism w : D → K such that δ = jw. Then we have H(g)jw = a and ujw = b. For uniqueness, suppose that there is another morphism w′ : D → K such that H(g)jw′ = a and ′ ′ u ′ 0 ujw = b. It follows that H(g)(jw − jw ) = 0 and [ v ] (jw − jw ) = [ 0 ], ′ whence jw − jw = 0 by the pullback property of H(d) and [ γ 0 ]. Then w = w′, because j is a monomorphism. Thus H(d) and γ have the claimed pullback in C. b u Since H(d)H(g)= γ [ 1 0 ] [ v ], the pullback square of H(d) and γ yields the existence of a unique morphism r : H(B′) → K such that H(g)jr = H(g) u ′ and ujr = [ 1 0 ] [ v ]. Denote k = rH(i ) : H(A) → K. Then we have the following commutative diagram in C: b ′ u H(i ) [ v ] H(A) / H(B′) / Z ⊕ Z′

r [ 1 0 ] k uj H(A) / K / Z

H(g)j γ H(B) / H(C) H(d) in which the upper row is a kernel-cokernel pair. The vertical rectangle and the lower square are pullbacks, hence so is the right-upper square by uj Lemma 2.4. Since ujidK = [ 1 0 ] , the pullback property of uj and 0 ′ ′ [ 1 0 ] yields the existence of a unique morphism j : K → H(B ) such that ′ u ′ uj rj = idK and [ v ] j = 0 . Then the equality H(g)jr = H(g) implies that ′ ′ u ′ 0 H(g)j = H(g)j , and so H(g)(j − j ) = 0. We also have [ v ] (j − j ) = [ 0 ]. ′ Now the pullback property of H(d) and [ γ 0 ] implies that j − j = 0, and so ′ rj = rj = idK. The commutative right-upper square in the previous diagram implies that ujr is an epimorphism, hence so is uj. Also, we have ujk = 0. We claim that uj = coker(k). To this end, let t : K → T be a morphism such that ′ u ′ tk = 0. Then the equality trH(i ) = 0 and the fact that [ v ] = coker(H(i )) ′ imply the existence of a unique morphism [ z z′ ] : Z ⊕ Z → T such that ′ u ′ ′ [ z z ] [ v ]= tr, that is, zu + z v = tr. It follows that zuj + z vj = trj, that is, zuj = t. The uniqueness of the morphism z : Z → T such that zuj = t follows because uj is an epimorphism. Hence uj is a cokernel. Consequently, H(d) is a semi-stable cokernel in C. b 3. Maximal exact structure We consider the following concept of exact category given by Quillen [15], as simplified by Keller [10]. Definition 3.1. By an exact category we mean an additive category C endowed with a distinguished class E of short exact sequences satisfying the axioms [E0], [E1], [E2] and [E2op] below. The short exact sequences in E are called conflations, while the kernels and cokernels appearing in such exact sequences are called inflations (denoted by ֌) and deflations (denoted by ։) respectively. STABLESHORTEXACTSEQUENCES 7

[E0] The identity morphism 10 : 0 → 0 is a deflation. [E1] The composition of two deflations is again a deflation. [E2] The pullback of a deflation along an arbitrary morphism exists and is again a deflation. [E2op] The pushout of an inflation along an arbitrary morphism exists and is again an inflation. We should note that the duals of the axioms [E0] and [E1] on inflations as well as both sides of Quillen’s “obscure axiom” hold in any such exact category [10]. Following [4, Definition 2.4], we introduce the following terminology.

Deﬁnition 3.2. Let C be an additive category and let (C,H) be its idempo- b tent completion. Consider an exact structure E on C. Then C is called stable b H(i) H(d) for pushouts for (C, E) if for every conﬂation H(A) ֌ H(B) ։ H(C) b and every pushout diagram

H(i) H(A) / / H(B)

H(f) w u H(A′) / / S in C, there exists an object B′ of C such that S =∼ H(B′). b Dually, one deﬁnes C to be stable for pullbacks for (C, E). b The following proposition extends [4, Theorem 2.6].

Proposition 3.3. Let C be an additive category and let (C,H) be its idempo- b tent completion. Consider an exact structure E on C, and the distinguished b class E0 of kernel-cokernel pairs A ֌ B ։ C in C such that the induced kernel-cokernel pair H(A) ֌ H(B) ։ H(C) is a conflation in C. Then E b 0 defines an exact structure on C if and only if C is stable under pushouts and pullbacks for (C, E). b Proof. Suppose first that E0 defines an exact structure on C. We only show that C is stable under pushouts for (C, E), the part for pullbacks being dual. H(i) H(d) b Let H(A) ֌ H(B) ։ H(C) be a conflation and H(f) : H(A) → H(A′) a morphism in C, and consider a pushout of H(i) and H(f). Then we have b an induced commutative diagram in C: b H(i) H(d) H(A) / / H(B) / / H(C)

H(f) w u v H(A′) / / S / / H(C)

i d in which the lower row is also a conﬂation. Since A ֌ B ։ C is a kernel- cokernel pair in C by Lemma 2.2, it follows that it belongs to E0, hence it is a conﬂation in C. Then there exists the pushout of i and f in C and we 8 SEPTIMIU CRIVEI have a commutative diagram in C: i d A / / B / / C

f g ′ ′ i d A′ / / B′ / / C in which the lower row is also a conﬂation. It follows that H(i) H(A) / / H(B)

H(f) H(g) H(i′) H(A′) / / H(B′) is a commutative square in C. By the universal property of the pushout b of H(i) and H(f) there exists a unique morphism λ : S → H(B′) in C ′ ′ b such that λu = H(i ) and λw = H(g). Then H(d )λu =0= vu and H(d′)λw = H(d)= vw, and again the universal property of the pushout of H(i) and H(f) implies that H(d′)λ = v. Hence the following diagram in C b is commutative: u v H(A′) / / S / / H(C)

λ H(i′) H(d′) H(A′) / / H(B′) / / H(C) i′ d′ Since A′ ֌ B′ ։ C is a conflation in C, the lower row of the above diagram is a conflation in C by the choice of E0. Now the Short Five Lemma in C [2, ′ b Corollary 3.2] implies that λ is an isomorphism, and so S =∼ H(B ). This shows that C is stable under pushouts for (C, E). b Conversely, suppose that C is stable under pushouts and pullbacks for (C, E). It is clear that the axiom [E0] holds for E . In order to show [E2], b 0 i d ′ let A ֌ B ։ C be a kernel-cokernel pair in E0, and let f : A → A be a H(i) H(d) morphism in C. Then H(A) ֌ H(B) ։ H(C) is a conflation in C, and b we may consider the pushout of H(i) and H(f) in C. Since C is stable under b pushouts for (C, E), we may choose a pushout diagram of the following form: b H(i) H(d) H(A) / / H(B) / H(C)

H(f) H(g) H(i′) H(d′) H(A′) / / H(B′) / H(C) in which the lower row is a conflation in C. Using Lemma 2.2 we obtain b a pushout of i and f in C reflected from C in which the lower row is a b op kernel-cokernel pair in E0, as required. Dually, [E2 ] holds for E0. f g f ′ Finally we need to prove [E1]. To this end, let A ֌ B ։ C and A′ ֌ ′ g ′ H(f) H(g) C ։ C be kernel-cokernel pairs in E0. Then H(A) ֌ H(B) ։ H(C) H(f ′) H(g′) and H(A′) ֌ H(C) ։ H(C′) are conflations in C by the choice of b STABLESHORTEXACTSEQUENCES 9

′ E0. By the axiom [E1] for the exact structure E, it follows that H(g g) = H(g′)H(g) is a deﬂation in C. Since C is stable under pullbacks for (C, E), b b we may choose a pullback of H(g) and H(f ′) in C of the form: b

H(α) H(K) / / H(A′) H(k) H(f ′) H(g) H(B) / / H(C)

Since H(f ′) is a monomorphism in C, so is H(k). We claim that H(k) : b H(K) → H(B) is the kernel of H(g′g) in C. To this end, let t : T → H(B) b be a morphism in C such that H(g′g)t = 0. Since H(f ′) is the kernel of H(g′), b ′ ′ there exists a unique morphism µ : T → H(A ) such that H(f )µ = H(g)t. Now by the universal property of the pullback of H(g) and H(f ′) there exists a unique morphism λ : T → H(K) such that H(k)λ = t and H(α)λ = µ. If there is also a morphism λ′ : T → H(k) in C such that H(k)λ′ = t and ′ ′ b H(α)λ = µ, then it follows that λ = λ, because H(k) is a monomorphism. Hence H(k) is the kernel of the deflation H(g′g) in C, and so H(k) is an b inflation in C. Then by the choice of E0 it follows that the kernel-cokernel b ′ k g g ′ pair K ֌ B ։ C belongs to E0. This shows that [E1] holds for E0, and so E0 defines an exact structure on C.

Now we are in a position to give our main result.

Theorem 3.4. Let C be an additive category, and let (C,H) be its idempo- b tent completion. Consider the maximal exact structure E in C given by max b the stable short exact sequences in C. Then the stable short exact sequences b deﬁne an exact structure on C if and only if C is closed under pushouts and pullbacks for (C, E ). In this case, it is the maximal exact structure on C. b max

Proof. Stable short exact sequences define the maximal exact structure Emax in C by [3, Theorem 3.5]. b If the stable short exact sequences define an exact structure on C, then C is closed under pushouts and pullbacks for (C, E ) by Proposition 3.3. b max Conversely, assume that C is closed under pushouts and pullbacks for (C, E ). By [3, Theorem 3.5], the stable exact sequences define an exact b max structure on the (weakly) idempotent complete category C. Consider the b distinguished class E0 of kernel-cokernel pairs A ֌ B ։ C in C such that the induced kernel-cokernel pair H(A) ֌ H(B) ։ H(C) is a stable exact sequence in C. Then E defines an exact structure on C by Proposition 3.3, b 0 and its conflations must be stable short exact sequences. On the other hand, if A → B → C is a stable short exact sequence in C, then the short exact sequence H(A) → H(B) → H(C) is stable in C by Proposition 2.7, hence b A → B → C is a conflation in C with respect to E0. Therefore, E0 coincides with the class of stable short exact sequences. Clearly, this is the maximal exact structure on C. 10 SEPTIMIU CRIVEI

The stability condition from Theorem 3.4 is usually easier to check in specific categories. In what follows we give situations when it does or does not hold. Corollary 3.5. The stable short exact sequences define the maximal exact structure on the category of free modules. Proof. It is well known that the idempotent completion of the category F of free modules is the category P of projective modules. Consider the usual functor H : F → P and the maximal exact structure Emax on P given by the stable short exact sequences. If one starts with a stable short exact sequence H(i) H(d) in P of the form H(A) ֌ H(B) ։ H(C), there exists the pullback of H(i) and any morphism H(f) : H(A) → H(A′). Then one has an induced stable short exact sequence H(A′) → P → H(C) [2, Proposition 2.12], which splits by the projectivity of H(C). Since P is weakly idempotent complete, it follows that P =∼ H(A′) ⊕ H(C). Hence P is free, and so the stability condition for pushouts in the sense of Definition 3.2 is checked. Dually, one has the stability condition for pullbacks. Hence the stable short exact sequences define the maximal exact structure on F by Theorem 3.4. Example 3.6. Consider the ring R = k[x]/(x2) for some field k, and denote S = kR. Let C be the additive category hS, R ⊕ Ri generated by S and R ⊕ R. Then its idempotent completion C is the additive category hS, Ri b generated by S and R, and then C = Mod(R). Then we have the following b commutative diagram in C: b (i,i) S ⊕ S / / R ⊕ R / S ⊕ S

(0,1) (p,1) (0,i) 0 ⊕ S / / S ⊕ R / S ⊕ S in which i is the inclusion, p is the projection map with pi = 0, and the remaining morphisms are the canonical ones. Then the left square is a pushout, and the upper kernel-cokernel pair is stable, since C = Mod(R). b Also, the first and the last objects of the lower kernel-cokernel pair are objects in C. But the middle object S ⊕ R does not belong to C. This shows that the stability condition for pushouts in the sense of Definition 3.2 does not hold, hence the stable short exact sequences do not define an exact structure on C by Theorem 3.4. We end with two applications to categories of chain complexes and projective spectra. Corollary 3.7. Let C be an additive category, and let (C,H) be its idem- b potent completion. Consider the maximal exact structure E in C given max b by the stable short exact sequences in C, and assume that C is closed under b pushouts and pullbacks for (C, E ). Then the maximal exact structure on b max the category Ch(C) of chain complexes over C consists of the short exact sequences which are stable short exact sequences in C in each degree. Proof. If the additive category C has an exact structure E, then the additive category Ch(C) also has an exact structure Ch(E) whose conflations are the STABLE SHORT EXACT SEQUENCES 11 short exact sequences which are conflations from E in each degree (e.g., see C [2, Lemma 9.1]). In particular, if Emax is the maximal exact structure on C C, then Ch(Emax) is an exact structure on Ch(C), included in the maximal Ch(C) exact structure Emax on Ch(C). i d Ch(C) Now let A ֌ B ։ C be a conflation from Emax , that is, a stable short exact sequence in Ch(C) by Theorem 3.4. Set a degree n, let X be an object of C and let α : X → Cn be a morphism in C. Then X can be viewed as a chain complex C′ concentrated in degree n, and we may define a chain map h : C′ → C by hn = α and hm = 0 for every m 6= n. By Lemma 2.5, the pullback of d and h yields the following commutative diagram in Ch(C):

i′ d′ A / / B′ / / C′

g h i d A / / B / / C ′ ′ i ′ d ′ Ch(C) Then A ֌ B ։ C is a conflation from Emax by Theorem 3.4. Then i′n d′n in dn An ֌ B′n ։ C′n is a kernel-cokernel pair in C, and so An ֌ Bn ։ Cn is i d a stable short exact sequence in C. Thus A ֌ B ։ C is a conflation from C C Ch(C) Ch(Emax), which shows that Ch(Emax)= Emax . n Following [5, Definition 7.1] a projective spectrum X = (Xn, Xm) with values in a category C consists of a sequence (Xn)n∈N of objects of C and n n morphisms Xm : Xm → Xn in C defined for n ≤ m such that Xn = 1Xn N k n k for every n ∈ , and Xn ◦ Xm = Xm for k ≤ n ≤ m. A morphism of n projective spectra f : X → Y between two projective spectra X = (Xn, Xm) n and Y = (Yn,Ym) consists of a sequence (fn)n∈N of morphisms fn : Xn → Yn n n in C such that fn ◦ Xm = Ym ◦ fm for n ≤ m. If C is additive, then so is the category P(C) of projective spectra with values in C. Corollary 3.8. Let C be an additive category, and let (C,H) be its idem- b potent completion. Consider the maximal exact structure E in C given max b by the stable short exact sequences in C, and assume that C is closed un- b der pushouts and pullbacks for (C, Emax). Then the maximal exact structure b i d on the category P(C) consists of the short exact sequences A → B → C for in dn which An ֌ Bn ։ Cn are stable short exact sequences in C for every n ∈ N. Proof. It is similar to the proof of [5, Corollary 7.4], using Theorem 3.4.

References [1] S. Bazzoni and S. Crivei, One-sided exact categories, J. Pure Appl. Algebra 217 (2013), 377–391. [2] T. B¨uhler, Exact categories, Expo. Math. 28 (2010), 1–69. [3] S. Crivei, Maximal exact structures on additive categories revisited, Math. Nachr. 285 (2012), 440–446. [4] B. Dierolf and D. Sieg, Exact subcategories of locally convex spaces, Math. Nachr. 285 (2012), 2106–2119. [5] L. Frerick and D. Sieg, Exact categories in functional analysis, preprint 2010. http://www.math.uni-trier.de/abteilung/analysis/HomAlg.pdf. 12 SEPTIMIU CRIVEI

[6] J. Gillespie, Model structures on exact categories, J. Pure Appl. Algebra 215 (2011), 2892–2902. [7] L. Gruson, Complétion abélienne, Bull. Sci. Math. 90 (1966), 17–40. [8] A. Heller, Homological algebra in abelian categories, Ann. of Math. 68 (1958), 484– 525. [9] M. Karoubi, Algèbres de Clifford et K-théorie, Ann. Sci. Ecole´ Norm. Sup. (4) 1 (1968), 161–270. [10] B. Keller, Chain complexes and stable categories, Manuscripta Math. 67 (1990), 379– 417. [11] G.M. Kelly, Monomorphisms, epimorphisms and pull-backs, J. Aust. Math. Soc. 9 (1969), 124–142. [12] M. Künzer, Heller triangulated categories, Homology, Homotopy Appl. 9 (2007), 233– 320. [13] B. Mitchell, Theory of categories, Academic Press, New York, 1965. [14] M. Prest, Definable additive categories: purity and model theory, Mem. Amer. Math. Soc. 210 (2011), no. 987. [15] D. Quillen, Higher algebraic K-theory. I, Algebraic K-theory, I: Higher K-theories (Proc. Conf., Battelle Memorial Inst., Seattle, Wash., 1972), Springer, Berlin, 1973, pp. 85–147. Lecture Notes in Math. 341. [16] F. Richman and E.A. Walker, Ext in pre-Abelian categories, Pacific J. Math. 71 (1977), 521-535. [17] W. Rump, Almost abelian categories, Cah. Topologie Géom. Différ. Catégoriques 42 (2001), 163–225. [18] W. Rump, On the maximal exact structure of an additive category, Fund. Math. 214 (2011), 77–87. [19] M. Saor´ın and J. Sˇˇtov´ıˇcek, On exact categories and applications to triangulated ad- joints and model structures, Adv. Math. 228 (2011), 968–1007. [20] J.P. Schneiders, Quasi-abelian categories and sheaves, Mém. Soc. Math. Fr. (N.S.) 76 (1999), pp. 1–140. [21] D. Sieg and S.-A. Wegner, Maximal exact structures on additive categories, Math. Nachr. 284 (2011), 2093–2100. [22] B. Stenström, Rings of quotients, Springer, Berlin, Heidelberg, New York, 1975. [23] N. Yoneda, On Ext and exact sequences, J. Fac. Sci. Univ. Tokyo Sect. I 8 (1960), 507–576.

Faculty of Mathematics and Computer Science, “Babes¸-Bolyai” University, Str. Mihail Kogalniceanu˘ 1, 400084 Cluj-Napoca, Romania E-mail address: [email protected]