Arxiv:1603.06557V4 [Math.CT] 26 Jul 2021

Home , Monoidal category, Split exact sequence

Homotopy in Exact Categories

Jack Kelly

Author address:

Jack Kelly, The Hamilton Mathematics Institute, School of Mathematics, Trinity Col- lege Dublin, Dublin 2, Ireland Email address: [email protected] arXiv:1603.06557v4 [math.CT] 26 Jul 2021

Contents

Chapter 1. Introduction 1 1.1. Background and Motivation 1 1.2. Goals and Layout 2 1.3. Notation and Conventions 4

Chapter 2. Exact Category Generalities 7 2.1. Recollections on Exact Categories 7 2.2. Homological Properties of Exact Categories 11 2.3. Resolutions in Exact Categories 21 2.4. Monoidal Exact Categories and Monads in Exact Categories 24 2.5. Quasi-Abelian Categories 28 2.6. Generators in Exact Categories 30

Chapter 3. Examples 39 3.1. Categories of Topological Vector Spaces 39 3.2. The Exact Category with Respect to a Generator 49 3.3. Exact Categories of Grothendieck Type 51 3.4. The Split Exact Structure 52

Chapter 4. Model Structures on Exact Categories 54 4.1. Cotorsion Pairs 54 4.2. Model Structures on Chain Complexes 58 4.3. The Projective Model Structure 68 4.4. The Dold-Kan Correspondence 72 4.5. The Injective Model Structure, the K-Projective Model Structure, and Examples 79

Chapter 5. Filtered and Graded Objects in Exact Categories 82 5.1. Preliminaries 82 5.2. Graded Objects in General Categories 82 5.3. Filtered Objects in General Categories 83 5.4. Filtered and Graded Objects in Exact Categories 90

Chapter 6. Homotopical Algebra in Exact Categories 102 6.1. Algebra in Monoidal (Model) Categories 102 6.2. Operads 105 6.3. Model Categories of Monoids and Algebras 107 6.4. Homotopical Algebra in Exact Categories 114

Appendix A. Model Categories 118 A.1. Weak factorisation Systems and Model Structures 118 A.2. Coﬁbrant Generation 119 A.3. Homotopy Colimits 119 A.4. Monoidal Model Categories 120

iii iv CONTENTS

A.5. Transferred Model Structures 121

Bibliography 123 Abstract

In this monograph we develop various aspects of the homotopy theory of exact categories. We introduce diﬀerent notions of compactness and generation in exact categories, and use these to study model structures on categories of chain complexes Ch∗(E) which are induced by cotorsion pairs on E. As a special case we show that under very general conditions the categories Ch+(E), Ch≥0(E), and Ch(E) are equipped with the projective model structure, and that a generalisation of the Dold-Kan correspondence holds. We also establish conditions under which categories of ﬁltered objects in exact categories are equipped with natural model structures. When E is monoidal we also examine when these model structures are monoidal and conclude by studying some homotopical algebra in such categories. In particular we provide conditions under which Ch(E) and Ch≥0(E) are homotopical algebra contexts, thus making them suitable settings for derived geometry.

2010 Mathematics Subject Classiﬁcation. Primary: 18G35; Secondary: 18G55. The majority of this research was conducted during the author’s graduate studies, which was supported by the EPSRC studentship BK/13/007, while employed at the University of Oxford on the EPSRC grant ‘Symmetries and Correspondences’, and while supported by the Simons Foundation at Trinity College Dublin under the program “Targeted Grants to Institutes”.

CHAPTER 1

Introduction

1.1. Background and Motivation Derived Geometry. Derived geometry has proved crucial for understanding intersection theory, de- formation theory and moduli theory in algebraic, smooth, and, recently, complex analytic geometry. There are two dominating abstract models for derived geometry. Lurie’s approach [Lur09a] uses a higher-categorical generalization of ringed spaces, namely structured (∞, 1)-topoi. This is an (∞, 1)-topos X together with a limit-preserving functor O : G → X where G is a geometry - an (∞, 1)-category satisfying certain properties. For example taking (G)op to be the (∞, 1)-category of simplicial rings gives a reasonable notion of derived algebraic stacks. David Spivak [S+10] considers derived smooth manifolds by taking as (G)op the category of simplicial C∞-rings. Mauro Porta and Tony Yue Yu [PY16],[Por15a][Por15b], [Por17],[PY18b], [PY17], [PY18a] are developing derived analytic geometry by taking Gop to be the category of simplicial rings equipped with a holomorphic functional calculus. In particular they have proven GAGA, base-change, and Riemann Hilbert type theorems. Thanks to their work the field of derived analytic geometry has developed rapidly in recent years, leading to GAGA, base-change, and Riemann-Hilbert type theorems. They have also announced a Hochschild-Kostant- Rosenberg theorem. Despite these crucial results there are some drawbacks to their setup. For example, there is no obvious definition of the category of quasi-coherent sheaves on a derived analytic space. Base-change theorems for coherent sheaves are difficult to formulate and prove. As in the algebraic case the push-forward functor between categories of coherent sheaves is only defined for proper maps. Moreover, constructing the pullback functor is very technical since the algebraic tensor product of sheaves is not the correct notion in the analytic setting. We believe that a model of derived analytic geometry based on Toënand Vezzosi’s formalism [TTV08] of derived geometry relative to a monoidal model category M would resolve many of these issues. Toënand Vezzosi’s model for derived geometry is inspired by the theory of (non-derived) geometry relative to a symmetric monoidal category (developed for instance in [Del07] and [Ban17]). This is a category-theoretic framework which views geometry as the unification of algebra and topology. The algebra describes local pieces and a Grothendieck topology allows one to glue these local pieces and obtain global objects. In [TTV08] they introduce the notion of a homotopical algebraic geometry context. Up to some technical details, a homotopical algebraic geometry context consists of a monoidal model category M such that the category Alg Comm(M ) of unital commutative monoids in M is a model category with the transferred op model structure, and (Alg Comm(M )) is equipped with a homotopy Grothendieck topology τ. We regard .= ( ( ))op as a category of affine spaces. The category of derived stacks on is then the Aff M Alg Comm M M category of functors X : → satisfying descent for τ-hypercovers. For derived algebraic geometry Aff M sSet one considers either the category M = Ch(R) of chain complexes of modules over a ring R (in characteristic zero), or the category M = sRMod of simplicial R-modules.

Monoidal Categories and Analytic Geometry. For the purposes of motivation we will give a brief overview of one approach to a formulation of derived analytic geometry. Details will appear in a forthcoming paper [BBKK]. Let (X, OX ) be a complex manifold. For each open set U the set OX (U) has a canonical structure of a Fr´echet space. Moreover, the restriction maps OX (V ) → OX (U) are continuous. Let F be a coherent sheaf on X. Cartan’s Theorem B implies that on a coordinate neighbourhood (or more generally a

1 2 1. INTRODUCTION

Stein neighbourhood), there is an exact sequence m n OX (V ) → OX (V ) → F (V ) → 0 The quotient topology on F (V ) makes it a Fréchet space. Thus sheaves on complex spaces have natural topological structures. It is therefore tempting to view (non-derived) complex analytic geometry as geometry relative to the symmetric monoidal category of Fréchet spaces. Unfortunately this does not seem possible. However in fhZ op [BBBK18] the authors construct a Grothendieck topology τ on a subcategory St of (Alg Comm(Fr)) . St is equivalent to the category of (dagger) Stein spaces and when k = C the coverings in their topology correspond to coverings of Stein spaces by Stein spaces. In particular the category of complex analytic spaces embeds in the category of schemes on this site. This construction is somewhat ad hoc but as usual passing to the derived world proves enlightening. The Grothendieck topology τ fhZ of [BBBK18] makes use of the homological structure on Fr which is a quasi- abelian, and therefore exact, category. It is an additive category with classes of admissible monomorphisms and admissible epimorphisms which provide a well-defined notion of homology. There are also notions of projective objects, exact functors, derived categories, and derived functors. If E is a monoidal exact category with a left-derivable tensor product ⊗, then we say a map A → B of commutative monoids in E L is a homotopy epimorphism if the map B ⊗A B → B is a quasi-isomorphism. The opposites of these maps make up the covers in τ fhZ . The obstacle to such covers defining a topology on the entire category op (Alg Comm(Fr)) is that they are not stable under base-change (because of the derived tensor product). If Ch(Fr) were a good enough monoidal model category then we could easily extend the definition of a homotopy epimorphism. Moreover as a homotopy cover in such a model category the issue of base change op would disappear and would give a genuine model topology on (Alg Comm(Ch(Fr))) . Tragically Fr is not good enough. It is neither complete nor cocomplete and does not have enough projectives. Fortunately it does nicely embed in a complete and cocomplete exact category with enough projectives, namely the category CBornC of complete bornological spaces over C. It is sometimes convenient to pass to the even bigger category Ind(BanC), the formal completion of the category of Banach spaces by filtered colimits.

1.2. Goals and Layout

The central goal of this monograph is to show that Ind(BanC) and CBornC are good categories for developing homotopical algebra, i.e. local derived geometry. More generally we show that under very general conditions on an exact category E, the category Ch(E) admits a good homotopy theory of algebras. Exact Category Generalities. Building on work of [Büh10] in Chapter 2 we begin this work by es- tablishing some technical results about exact categories in general which we will need in subsequent chapters. After recalling some basic facts we introduce various useful notions of acyclicity. We then discuss bounded and unbounded resolutions in exact categories. In particular we generalise the famous result of Spaltenstein [Spa88] to exact categories satisfying very general conditions, following a similar generalisation of this result to relative homological algebra [CNPS17]. Theorem 1.2.1 (Corollary 2.3.67). Let E be an exact category with kernels in which. Let P be a class of objects such that for each object X in E there is an object P in P together with an admissible epimorphism P X. Suppose further that P is closed under countable coproducts and satisfy axiom AB4 − k for some k. Then for any complex X• in Ch(E) there is a complex P• in Ch(P) and an admissible epimorphism P• → X• which is a quasi-isomorphism. Moreover, X• is the limit of a Ch+(P)-special direct system. We then recall a suitable idea of generators, before defining so-called elementary and weakly elementary exact categories. These technical notions will be crucial for controlling the homotopy theory of an exact category and avoiding set-theoretic smallness concerns. Next we define monoidal exact categories and establish some basic properties of them. In particular we prove the existence of an induced exact structure on modules for commutative monoids internal to such categories. More generally we study monads on exact categories and their categories of algebras. 1.2. GOALS AND LAYOUT 3

In Chapter 3 we study numerous examples, including the category of complete bornological spaces over a Banach ﬁeld, and the non-expanding normed and Banach categories over a non-Archimedean ﬁeld. We also relate our work to Gillespie’s work on derived categories of Grothendieck abelian categories with respect to generators in [Gil16a], and to the work of Gillespie [Gil16a], Estrada-Gillespie-Odaba¸si[EGO17], and Krause [Kra12] on pure exact structures.

Model Structures on Exact Categories. In Chapter 4 we discuss model structures on exact categories. There is a general theory of model structures on weakly idempotent complete exact categories due to [Hov02], [Gil11] and [St’12ˇ ] using cotorsion pairs. A pair of classes of objects (L, R) in an exact category E is said to be a cotorsion pair if L ∈ L if and only if Ext1(L, R) = 0 for all R ∈ R, and R ∈ R if and only if Ext1(L, R) = 0 for all L ∈ L. In [Gil04] Gillespie suggests a strategy for producing a model structure on Ch(E), given a cotorsion pair on an abelian category E, which can easily be adapted to exact categories more generally. When E is monoidal we give conditions on (L, R) such that the induced model structure is monoidal and satisﬁes the monoid axiom. However we give very general conditions on an exact category E such that it does work for the projective cotorsion pair (Proj(E), Ob(E)), where Proj(E) is the class of projective objects in E. In particular we prove the following

Theorem 1.2.2 (Theorem 4.3.58). Let E be an exact category satisfying the following conditions (1) E has enough projectives. (2) E has kernels. (3) We suppose that countable coproducts of projectives exist and satisfy axiom AB4−k for some k ∈ Z. Then, applied to the cotorsion pair (Proj(E), Ob(E)), Gillespie’s strategy produces a model structure on Ch(E).

We call this the projective model structure on Ch(E). If E = RMod is the category of R-modules over a ring R then this is the usual projective model structure. Under some stronger assumptions, namely that the category E has generators which are presented relative to the class of admissible monics, the result of the above theorem can be deduced from results of [St’12ˇ ]. Moreover, Gillespie proved Theorem 4.3.58 in [Gil16a] in the case that E is the G-exact structure on a Grothendieck abelian category. An advantage of our result is that it avoids many set-theoretic concerns and we suggest examples where this is useful. In particular at the end of the chapter we give natural examples of exact categories with very different set-theoretic properties which satisfy the conditions of Theorem 4.3.58. This result should also be compared with the model structures of [CH02]. Indeed, if an exact category has enough projectives then the class of exact sequences on E are the P-exact sequences for P the class of projectives, in the sense of [CH02] Section 1.1. When the underlying category of E is abelian Christensen and Hovey show that the projective model structure exists on Ch(E) (and Ch≥0(E)). We suspect that the fact that E is abelian is not necessary for their proof, and that one only needs an additive category with kernels. Although our proof of the existence of resolutions is similar to [CH02] Theorem 2.2, we actually give a more general version of their Case A. Moreover our proof strategy for the existence of the projective model structure is somewhat different - we work with cotorsion pairs. This allows us to derive useful properties of the projective model structure. We call a monoidal exact category monoidal elementary if it is elementary and its projectives are flat and closed under the tensor product. We then prove the following result.

Theorem 1.2.3 (Theorem 4.3.68). Let E be a monoidal elementary exact category. Then the projective model structure on Ch(E) is monoidal and satisﬁes the monoid axiom. We then prove a generalisation of the Dold-Kan correspondence.

Theorem 1.2.4 (Theorem 4.4.76). Let E be a weakly idempotent complete exact category with enough projectives. Endow Ch≥0(E) and sE with their projective model structures. Then the functors

Γ: Ch≥0(E) → sE 4 1. INTRODUCTION and

N : sE → Ch≥0(E) form a Quillen equivalence.

In Chapter 5 we discuss model structures for graded and ﬁltered objects in exact categories. This will prove crucial for generalising Koszul duality results in [Kel19]. Finally in Chapter 6 we discuss model structures on categories of dg-modules and dg-algebras in monoidal exact categories. By modying results of [WY19] and [Whi17] we show the following

Theorem 1.2.5 (Theorem 6.4.32). Let E be an elementary exact category which is also a closed symmetric monoidal category and let P be any non-symmetric operad in Ch∗(E) for ∗ ∈ {≥ 0, ∅}. Then the transferred model structure exists on the category Alg P(Ch∗(E)) of P-aglebras in Ch∗(E). If E is enriched over Q and P is a symmetric operad, then the transferred model structure also exists on Alg P(Ch∗(E)). We also prove a Dold-Kan correspondence for algebras over operads (Theorem 6.4.38) , again following [WY19], as well as a generalisation of the cosimplicial Dold-Kan equivalence of [CC04] (Theorem 6.4.39). Finally in Theorem 6.4.41 we show that whenever E is monoidal elementary, the categories Ch(E) and Ch≥0(E) are homotopical algebra contexts in the sense of [TTV08]. In particular they are suitable settings for theories of derived geometry. In future joint work with Oren Ben-Bassat and Kobi Kremnitzer we will develop a model of derived analytic geometry by applying this to the quasi-abelian category CBornk. In appendix A we recall some basic facts and set out our conventions regarding model categories.

1.3. Notation and Conventions Throughout this work we will use the following notation. • 1-categories will be denoted using the mathpzc font C, D, E, etc. In particular we denote by Ab the category of abelian groups and QVect the category of Q-vector spaces. If M is a model category, or a category with weak equivalences, its associated (∞, 1)-category will be denoted M. • Operads will be denoted using capital fractal letters C, P, etc. Algebras over an operad will generally be denoted using capital letters X,Y , etc. The category of algebras over an operad will be denoted

Alg P • We denote the operads for unital associative algebras, unital commutative algebras, non-unital commutative algebras, and Lie algebras by Ass, Comm, Commnu, and Lie respectively. • For the operad Ass, Comm, Lie we will denote the corresponding free algebras by T (V ),S(V ), and L(V ) respectively. We also denote by U(L) the universal enveloping algebra of a Lie algebra L. • Unless stated otherwise, the unit in a monoidal category will be denoted by k, the tensor functor by ⊗, and for a closed monoidal category the internal hom functor will be denoted by Hom. Monoidal categories will always be assumed to be symmetric, with symmetric braiding σ. • Filtered colimits will be denoted by lim→. Projective limits will be denoted lim←. • The ﬁrst inﬁnite ordinal will be denoted ℵ0. Let us now introduce some conventions for chain complexes.

Definition 1.3.6. A chain complex in a pre-additive category E is a sequence

dn dn−1 K• = ... / Kn / Kn−1 / Kn−2 / ...

where the Ki are objects and the di are morphisms such that dn−1 ◦ dn = 0. The morphisms are called diﬀerentials.A morphism of chain complexes f• : K• → L• is a collection of morphisms fn : Kn → Ln such that the following diagram commutes for each n: 1.3. NOTATION AND CONVENTIONS 5

K K dn+1 dn ... / Kn+1 / Kn / Kn−1 / ...

n n−1 fn+1 f f dL L n+1 dn ... / Ln+1 / Ln / Ln−1 / ... The category whose objects are chain complexes and whose morphisms are as described above is called the category of chain complexes in E, denoted Ch(E). We also define Ch≥0(E) to be the full subcategory of Ch(E) on complexes A• such that An = 0 for n < 0, Ch≤0(E) to be the full subcategory of Ch(E) on complexes A• such that An = 0 for n > 0, Ch+(E), the full subcategory of chain complexes A• such that An = 0 for n << 0, Ch−(E), the full subcategory of chain complexes A• such that An = 0 for n >> 0 and Chb(E) to be the full subcategory of Ch(E) on complexes A• such that An 6= 0 for only finitely many n.A lot of the statements in the rest of this document apply to several of these categories at once. In such cases we will write Ch∗(E), and specify that ∗ can be any element of some subset of {≥ 0, ≤ 0, +, −, b, ∅}, where by definition Ch∅(E) = Ch(E).

We will frequently use the following special chain complexes. Definition 1.3.7. If E is an object of a pointed category E we let Sn(E) ∈ Ch(E) be the complex whose nth entry is E, with all other entries being 0. We also denote by Dn(E) ∈ Ch(E) the complex whose nth and (n − 1)st entries are E, with all other entries being 0, and the diﬀerential dn being the identity. Let us also introduce some notation for truncation functors.

Definition 1.3.8. Let E be an additive category which has kernels. For a complex X• we denote by τ≥nX the complex such that (τ≥nX)m = 0 if m < n, (τ≥nX)m = Xm if m > n and (τ≥nX)n = Ker(dn). The differentials are the obvious ones. The construction is clearly functorial. Dually we define the truncation functor τ≤k. All of the above categories are naturally enriched over Ch(Ab). We denote the enriched hom by Hom(−, −). For notational clarity we recall its definition here.

Definition 1.3.9. Let X•,Y• ∈ Ch(E). We deﬁne Hom(X•,Y•) ∈ Ch(Ab) to be the complex with Y Hom(X•,Y•)n = HomE (Xi,Yi+n) i∈Z

and diﬀerential dn deﬁned on HomE (Xi,Yi+n) by Y n X df = di+n ◦ f − (−1) f ◦ di Let (E, ⊗, k) be a monoidal additive category, i.e. ⊗ is an additive bifunctor. There is an induced 0 monoidal structure on Ch∗(E) for ∗ ∈ {≥ 0, ≤ 0, +, −, b, ∅}. The unit is S (k). If X• and Y• are chain complexes then we set M (X• ⊗ Y•)n = Xi ⊗ Yj i+j=n

If i + j = n, then we deﬁne the diﬀerential on the summand Xi ⊗ Yj of (X• ⊗ Y•)n by

X•⊗Y• X• i Y• dn |Xi⊗Yj = di ⊗ idY• + (−1) idX• ⊗ dj 0 If ∗ ∈ {≥ 0, ≤ 0, +, −, b, ∅} then (Ch∗(E), ⊗,S (k)) is a monoidal additive category. If (E, ⊗, k, Hom) is a closed monoidal additive category then we deﬁne a functor Hom(−, −): Ch(E)op × Ch(E) → Ch(E)

Y Hom(X•,Y•)n = HomE (Xi,Yi+n) i∈Z 6 1. INTRODUCTION

and diﬀerential dn deﬁned on HomE (Xi,Yi+n) by

X• i Y• d = Hom(di , id) + (−1) Hom(id, di+n) This does deﬁne an internal hom on the monoidal category

(Ch(E), ⊗,S0(k)) The internal hom on chain complexes also restricts to a bifunctor

op Hom(−, −): Chb(E) × Chb(E) → Chb(E) Then

0 (Chb(E), ⊗,S (k), Hom) is a closed monoidal additive category. In fact, in both of these categories there are natural isomorphisms of chain complexes of abelian groups. ∼ Hom(X•, Hom(Y•,Z•)) = Hom(X• ⊗ Y•,Z•)

The categories Ch∗(E) for ∗ ∈ {+, −, b, ∅} also come equipped with a shift functor. It is given on objects A[1] A • by (A•[1])i = Ai+1 with diﬀerential di = −di+1. The shift of a morphism f is given by (f•[1])i = fi+1. [1] is an auto-equivalence with inverse [−1]. We set [0] = Id and [n] = [1]n for any integer n.

Finally, we deﬁne the mapping cone as follows.

Definition 1.3.10. Let X• and Y• be chain complexes in an additive category E and f• : X• → Y•. The mapping cone of f•, denoted cone(f•) is the complex whose components are

cone(f•)n = Xn−1 ⊕ Yn and whose diﬀerential is Å X ã cone(f) −dn−1 0 dn = Y −fn−1 dn

There are natural morphisms τ : Y• → cone(f) induced by the injections Yi → Xi−1 ⊕ Yi, and π : cone(f) → X•[−1] induced by the projections Xi−1 ⊕ Yi → Xi−1. The sequence

Y• → cone(f) → X•[−1] is split exact in each degree. Acknowledgements. The author would like to thank Kobi Kremnizer, Kevin McGerty, Theo B¨uhler, andTimoth´eeMoreau for useful comments. The author is also very grateful to the referee for many helpful comments and suggestions which have greatly improved this paper. CHAPTER 2

Exact Category Generalities

In this chapter we will establish some technicalities about exact categories which will be used throughout this work. In particular we will discuss acyclicity of complexes and the existence of unbounded resolutions. We will also discuss various notions of generation and compactness in such categories, and introduce the notion of a monoidal exact category. Finally we will see when exact structures can be lifted to categories of algebras for some monad acting on an exact category. The results in this chapter will prove crucial for studying the homotopy theory of exact categories in Chapter 4.

2.1. Recollections on Exact Categories In this section we review the rudiments of exact categories, following [B¨uh10]. In the following E will be an additive category. A kernel-cokernel pair in E is a pair of composable maps (i, p), i : A → B, p : B → C such that i = Ker(p) and p = Coker(i). If Q is a class of kernel-cokernel pairs and (i, p) ∈ Q, then we say that i is an admissible monic and p is an admissible epic with respect to Q.

Definition 2.1.1. A Quillen exact structure on an additive category E is a collection Q of kernel- cokernel pairs such that (1) Isomorphisms are both admissible monics and admissible epics. (2) Both the collection of admissible monics and the collection of admissible epics are closed under composition. (3) If

f A / B

f 0 X / Y is a pushout diagram, and f is an admissible monic, then f 0 is as well. (4) If

f 0 A / B

f X / Y is a pullback diagram, and f is an admissible epic, then f 0 is as well.

Let (E, Q) be an exact category. We call a null sequence

i p 0 / A / B / C / 0 short exact if (i, p) is a kernel-cokernel pair in Q. We will use interchangeably the notion of kernel-cokernel pair and short exact sequence. When it is not likely to cause confusion, we will suppress the notation (E, Q) to E. When studying exact categories it is natural to consider so-called exact functors:

7 8 2. EXACT CATEGORY GENERALITIES

Definition 2.1.2. Let (E, P), (F , Q) be exact categories. A functor F : E → F is said to be exact (with respect to P and Q) if for any short exact sequence 0 → X → Y → Z → 0 in P, 0 → F (X) → F (Y ) → F (Z) → 0 is a short exact sequence in Q. Definition 2.1.3. Let (E, P) be an exact category. An exact subcategory of (E, P) is an exact category (F , Q) where F is a subcategory of E and the inclusion functor is exact. On any additive category one can deﬁne the split exact structure for which the kernel-cokernel pairs are the split exact sequences. Any exact category contains this is an exact subcategory. The split exact structure is therefore the minimal exact structure on an additive category. If an additive category has all kernels and cokernels, then it also has a maximal exact structure. This is due to Sieg and Wegner [SW11]. One particularly nice class of additive categories has the class of all kernel-cokernel pairs as its maximal exact structure. Definition 2.1.4. An additive category E with all kernels and cokernels is said to be quasi-abelian if the class qac of all kernel-cokernel pairs forms an exact structure on E. The following is then tautological. Proposition 2.1.5. Let E be a quasi-abelian category, and let Q be a class of kernel-cokernel pairs on E such that (E, Q) is an exact category. Then the identity functor idE is an exact functor (E, Q) → (E, qac). We will study quasi-abelian structures in more detail later. For now let us note that abelian categories are quasi-abelian. In an abelian category all monics are kernels of their cokernels, and all epics are cokernels of their kernels. It therefore trivially follows that both classes are closed under composition. It is also clear that both classes contain all isomorphisms. It is a standard exercise that in an abelian category, monomorphisms are pushout-stable and epimorphisms are pullback-stable. See for example [Fre64] Theorem 2.54. Let us now record some basic results about exact categories which will prove useful. Proposition 2.1.6. Let i A / / B

f f 0 i0 A0 / / B0 be a commutative diagram in which the horizontal morphisms are admissible monics. Then the following are equivalent (1) The square above is a pushout. (2) The sequence i −f ( f 0 i0 ) 0 / A / B ⊕ A0 / B0 / 0 is short exact. (3) The square above is bicartesian. (4) The square is part of a commutative diagram

i p A / / B / / C

f f 0 0 i0 p A0 / / B0 / / C with short exact rows.

Proof. See [B¨uh10] Proposition 2.12. 2.1. RECOLLECTIONS ON EXACT CATEGORIES 9

Proposition 2.1.7. Let E be an exact category and A ⊂ E a full additive subcategory. Suppose that for every morphism f : A → B which is admissible in E, a kernel and cokernel of f in E exist in A. Then the collection of all kernel-cokernel pairs (i : A → B, p : B → C) which are exact in E where A, B, C ∈ A defines an exact structure on A which makes it an exact subcategory of E. Proof. It is clearly sufficient to show that this collection of kernel-cokernel pairs endows A with an exact structure. The first and second conditions are clearly satisfied. Let

i A / B

f f 0 i0 A0 / B0 be a pushout diagram in E with f an admissible monic, and i and i0 in A. We need to show that Y is (isomorphic to) an object of A. But there is an exact sequence

i −f ( f 0 i0 ) 0 / A / B ⊕ A0 / B0 / 0 in E. Now a cokernel of the map A → B ⊕ A0 in E exists in A, so B0 is isomorphic to an object of A. The last condition is dual to this one. For technical reasons, unless stated otherwise we will assume from now on that all exact categories are weakly idempotent complete. This means that every retraction has a kernel, or equivalently, that every coretraction has a cokernel. Note that the condition is self-dual. Quasi-abelian categories are in particular weakly idempotent complete. In weakly idempotent complete exact categories, we then have the following useful result, often called the Obscure Axiom. Proposition 2.1.8 (The Obscure Axiom). Let E be a weakly idempotent complete exact category. (1) Suppose that i : A → B is a morphism. If there exists a morphism j : B → C such that the composite ji : A → C is an admissible monic, then i is an admissible monic. (2) Suppose that i : A → B is a morphism. If there exists a morphism j : C → A such that i ◦ j is an admissible epic, then i is an an admissible epic.

Proof. See [B¨uh10] Proposition 2.16. Another useful result which translates from abelian categories is the 3 × 3 Lemma, [B¨uh10] Corollary 3.6. Proposition 2.1.9 (3 × 3 Lemma). Consider a commutative diagram in which all columns are exact sequences and all rows are null sequences.

f 0 g0 A0 / B0 / C0

a b c f g A / B / C

a0 b0 c0 f 00 g00 A00 / B00 / C If two of the rows are exact sequences then the remaining row is also an exact sequence. This has an immediate corollary. Corollary 2.1.10 ([B¨uh10] Proposition 2.9). Let f : X → Y and a : A → B be admissible monomorphisms (resp. admissible epimorphisms). Then f ⊕ a : X ⊕ A → Y ⊕ B is an admissible monomorphism (resp. admissible epimorphism). 10 2. EXACT CATEGORY GENERALITIES

subsectionAbelianisations Let (E, Q) be an exact category. Let F be a full subcategory of E. Suppose that F is closed under extensions, that is if 0 → A → B → C → 0 is a short exact sequence in (E,Q) with A and C objects of F , then B is an object of F as well. Let QF consist of those kernel-cokernel pairs (i : A → B, q : B → C) in F which when regarded as pairs of morphisms + in E are kernel-cokernel pairs in Q. It is then straightforward to show ([BMSG 10]) that (F , QF ) is an exact subcategory of (E, Q). It turns out that any small exact category can be obtained as a full subcategory of an abelian category which is closed under extensions. This is the main content of the Gabriel-Quillen Embedding Theorem which provides an invaluable tool for studying exact categories. Theorem 2.1.11 (The Gabriel-Quillen Embedding Theorem). Let E be a small exact category. Then there is an abelian category A(E) and a fully faithful additive functor I : E → A(E) which is exact, reflects exactness, and preserves all kernels. Moreover the essential image of I is closed under extensions. A(E) may be chosen to be the category of left-exact functors E → Ab. If in addition E is weakly idempotent complete then a morphism f : E → F in E is an admissible epic if and only if I(f) is an epic in A(E). Proof. See Appendix A in [Büh10]. Definition 2.1.12. We call an embedding I : E → A of an exact category into an abelian category a left abelianisation of E if (1) I is fully faithful. (2) I is exact. (3) I reflects exactness. (4) The essential image of I is closed under extensions. (5) I preserves all kernels which exist. (6) If f is a morphism in E, then f is an admissible epic if and only if I(f) is an epic. In particular, Theorem 2.1.11 says that any weakly idempotent complete small exact category admits a left abelianisation. In fact, if the final assumption is removed, then all small exact categories have abelianisations. There is an obvious dual notion of a right abelianisation. It is clear that right abelianisations of small weakly idempotent complete exact categories also exist. Indeed, if Eop → A is a left-abelianisation of Eop, then E → Aop is a right-abelianisation of E. 2.1.1. Generation of Exact Subcategories. Let E be a locally small additive category and A a small full subcategory. By an argument similar to [hc] we can find a small full exact subcategory of E containing A. In the rest of the section we assume that given a small subcategory E of A one can choose direct sums for finite collections of objects in E, and kernels and cokernels of morphisms in E. For example one might assume that such limits and colimits can be made functorial in the ambient category (e.g. if E is locally presentable). Proposition 2.1.13. There is a small full additive subcategory Σ(A; E) of E containing A. Proof. We let Σ(A; E) be the full subcategory whose objects are the zero object and a choice of a direct sum in E for each finite collection of objects of A. This is clearly additive, contains A, and is small. Now let E be an exact category and A a small full subcategory. Proposition 2.1.14. There is a small full exact subcategory Ex(A, E) of E containing A. Proof. By Proposition 2.1.13 we may assume that A is additive. Let Ex1(A, E) denote the full subcategory of E consisting of a choice of kernels and cokernels of morphisms f : A → A0 which are admissible in E. We set Exn+1(A; E) .= Ex1(Exn(A; E); E) We claim that ∞ [ Ex(A; E) .= Exn(A; E) n=1 2.2. HOMOLOGICAL PROPERTIES OF EXACT CATEGORIES 11

1 S∞ n works. Since Ex (A; E) is small for A small this would prove the claim. n=1 Ex (A; E) is clearly closed under taking kernels and cokernels of those morphisms which are admissible in E. By Proposition 2.1.7 it is an exact subcategory. The point of this is that even if a category E is not small, when working with small diagrams in E we can pass to an abelianisation.

2.2. Homological Properties of Exact Categories 2.2.1. Notions of Acyclicity. In a general exact category, arbitrary kernels and cokernels may not exist. Therefore it is not in general possible even to write down candidates for the homology objects of a chain complex. Even if all kernels and cokernels do exist, then there are multiple candidates for the homology which are not isomorphic in general. For example, given a null sequence

f g Γ = E / F / G i.e. g ◦ f = 0, one could consider both Coker(Im(f) → Ker(g)) and Im(Ker(g) → Coker(f)). In an abelian category these are isomorphic, but for general additive categories this is not the case. Despite these ambiguities, there are still various useful notions of acyclicity in exact categories, which we discuss below. First let us deﬁne several classes of morphisms. Definition 2.2.15. A morphism f : E → F in an exact category is said to be (1) weakly left admissible if it has a kernel and the map Ker(f) → E is admissible. (2) weakly right admissible if it has a cokernel, and the map F → Coker(f) is admissible. (3) weakly admissible if it is both weakly left admissible and weakly right admissible. The following characterisation of weakly admissible morphisms is immediate. Proposition 2.2.16. A morphism f : E → F in an exact category E is weakly admissible if and only if it admits a decomposition

f E / F < <

< # # fˆ < $ $ Ker(f) Coim(f) / Im(f) Coker(f) where the sequences Ker(f) E Coim(f) and Im(f) F Coker(f) are short exact. Definition 2.2.17. Let f be a morphism in exact category. Then f is said to be admissible if it is weakly admissible and the map Coim(f) → Im(f) is an isomorphism. Remark 2.2.18. Admissible epimorphisms and admissible monomorphisms are admisssible morphisms in the sense above. This is not how admissible morphisms are usually deﬁned (see e.g. [B¨uh10]). However the notions are equivalent. 12 2. EXACT CATEGORY GENERALITIES

Proposition 2.2.19. Let f : E → F be a morphism in an exact category E. Then the following are equivalent. (1) f is admissible. (2) f admits a decomposition E I F (3) There is a commutative diagram

f E / F < @

< # # Kerf I @ Cokerf where the sequences Kerf E I and I F Coker(f) are short exact. Proof. 1 and 3 are clearly equivalent thanks to Proposition 2.2.16. Also 3 ⇒ 2 trivially. Let us show that 2 ⇒ 1. Since I F is an admissible monic, the kernel of f exists, and coincides with the kernel of E I. Hence Ker(f) → E is an admissible monic and in particular E → I is a coimage of f. Dually, the cokernel of f exists, it coincides with the cokernel of G F , and I F is an image of f. Corollary 2.2.20. A morphism f : E → F in an exact category is an isomorphism if and only if it is both an admissible epic and an admissible monic. Proof. Axiomatically an isomorphism is both an admissible monic and an admissible epic. Conversely, suppose f is both an admissible monic and an admissible epic. Since it is an admissible monic the map E → Coim(f) is an isomorphism. Since it is an admissible epic the map Im(f) → E is an isomorphism. Since f is admissible the map Coim(f) → Im(f) is an isomorphism. The claim now follows from the commutative diagram f E / F O ∼ ∼

∼ Coim(f) / Im(f) We are now ready to introduce our various notions of acyclic sequences. Definition 2.2.21. A null sequence

f g X / Y / Z is said to be (1) weakly acyclic if f is weakly right admissible, g has a kernel, and the natural map Im(f) → Ker(g) is an isomorphism. (2) weakly coacyclic if g is weakly left admissible, f has a cokernel, and the natural map Coker(f) → Coim(g) is an isomorphism. (3) admissibly acyclic if it is weakly acyclic and f is admissible, (4) admissibly coacyclic if it is weakly coacyclic and g is admissible (5) admissible if both f and g are admissible. (6) acyclic if it is both admissibly acyclic and admissibly coacyclic. 2.2. HOMOLOGICAL PROPERTIES OF EXACT CATEGORIES 13

Remark 2.2.22. If a null sequence

f g X / Y / Z is weakly acyclic then g is automatically weakly left admissible. Definition 2.2.23. A complex

fn fn−1 Xn / Xn−1 / ... / X0 is said to be weakly acyclic/ weakly coacyclic/ admissibly acyclic/ admissibly coacyclic/ admissible/ acyclic if for each 1 ≤ i ≤ n − 1 each sequence

fi+1 fi Xi+1 / Xi / Xi−1 is weakly acyclic/ weakly coacyclic/ admissibly acyclic/ admissibly coacyclic/ admissible/ acyclic. Let us now set up some tools for determining whether a complex is acyclic. We can partially test acyclicity by passing to a left abelianisation.

Proposition 2.2.24. Let I : E → A be a left abelianisation of E. (1) If

fn fn−1 Xn / Xn−1 / ... / X0 is admissibly acyclic in E then

I(fn) I(fn−1) I(Xn) / I(Xn−1) / ... / I(X0) is exact in A. (2) If fi is weakly admissible for 2 ≤ i ≤ n and

I(fn) I(fn−1) I(Xn) / I(Xn−1) / ... / I(X0) is exact, then

fn fn−1 Xn / Xn−1 / ... / X0 is admissibly acyclic. (3) If fi is weakly left admissible for 1 ≤ i ≤ n − 1 and

I(fn) I(fn−1) I(Xn) / I(Xn−1) / ... / I(X0) is exact in A, then

fn fn−1 Xn / Xn−1 / ... / X0 is admissibly acyclic. Proof. Clearly it is suﬃcient to prove the claims for sequences

f g X / Y / Z (1) Suppose the above sequence is admissibly acyclic. Since f is admissible I preserves Im(f). By assumption I preserves all kernels. Hence

I(f) I(g) I(X) / I(Y ) / I(Z) is exact. 14 2. EXACT CATEGORY GENERALITIES

(2) Suppose now that I(f) I(g) I(X) / I(Y ) / I(Z) is exact and that f is weakly admissible. Since I preserves all kernels, and cokernels of admissible morphisms, we have I(Coim(f)) =∼ CoimI(f). Now CoimI(f) =∼ ImI(f) =∼ KerI(g) Since I is fully faithful, Coim(f) is a kernel of g. Finally, note that we have a factorisation of Coim(f) → Ker(g) Coim(f) → Im(f) Ker(g) By Proposition 2.1.8 Im(f) Ker(g) is also an (admissible) epic. By Corollary 2.2.20 it is an isomorphism. Therefore Coim(f) → Im(f) is as well. By Proposition 2.2.19 we are done. (3) We can factor f as f 0 X / Ker(g) / Y with Ker(g) → Y an admissible monic. We need to show f 0 is an admissible epic. Since I preserves kernels, it sends the diagram above to

I(f 0) I(X) / KerI(g) / I(Y ) Since I(f) I(g) I(X) / I(Y ) / I(Z) is exact, I(f 0) is an epic. Thus f 0 is an admissible epic, and we are done. Let us give an immediate application. Proposition 2.2.25. Let f : X → Y and g : Y → Z be admissible morphisms in a weakly idempotent complete exact category E. Then the sequence 0 → Ker(f) → Ker(g ◦ f) → Ker(g) is admissibly acyclic. If f is an admissible epimorphism then 0 → Ker(f) → Ker(g ◦ f) → Ker(g) → 0 is exact. Proof. For the first claim it suffices to show that Ker(f) → Ker(g ◦ f) is weakly left admissible. Then we may pass to a left abelianisation. The composition Ker(f) → Ker(g ◦ f) → X is an admissible monomorphism since f is admissible. Thus Ker(f) → Ker(g ◦ f) is an admissible monomorphism. The second claim follows since a left abelianisation preserves admissible epimorphisms. Although the functor I reflects short exact sequences, it need not in general reflect acyclicity of unbounded complexes. However it does for a certain nice class of complexes.

Definition 2.2.26. A complex X• in an exact category is said to be good if for each n there is m < n such that dm has a kernel. X• is said to be cogood if for each n there is m > n such that dm has a cokernel. Example 2.2.27. Bounded below complexes are good. We will frequently use the following trick for good complexes.

Proposition 2.2.28. Let X• be a good complex in an exact category. Suppose that for any n such that X dn has a kernel, the induced map 0 dn+1 : Xn+1 → ZnX is an admissible epic. Then X• is acyclic. 2.2. HOMOLOGICAL PROPERTIES OF EXACT CATEGORIES 15

Proof. Suppose dm has a kernel. By assumption dm+1 factors as

Xm+1 ZmX → Xm A priori ZmX → Xm is not admissible. However it is a monomorphism. Therefore, since Xm+1 ZmX is admissible its kernel exists and it coincides with the kernel Zm+1X of dm+1. Since Xm+1 ZmX is admissible it is in particular weakly left admissible. Therefore dm+1 is also weakly left admissible. Now consider dm+2. By assumption it factors as

dm+2 : Xm+2 Zm+1X Xm+1

Thus dm+2 is an admissible morphism whose image is Zm+1X. An easy induction then shows that X• is acyclic. Since I preserves kernels and reﬂects admissible epimorphisms, Proposition 2.2.28 gives the following.

Corollary 2.2.29. Let (X•, d•) be a complex in E. Let I : E → A be a left abelianisation of E. Suppose X• is good. Then X• is acyclic if and only if I(X•) is. X 0 Proof. Suppose I(X•) is acyclic, and dn has a kernel ZnX. By assumption I(dn+1): I(Xn+1) → 0 ZnI(X) = I(ZnX) is an epimorphism. Thus dn+1 : Xn+1 → ZnX is an admissible epimorphism. This can be used to prove the following useful result about truncations functors Y X X Proposition 2.2.30. Let f : X → Y be a weak equivalence in Ch(E). If dn , dn , and dn−1 have kernels, then τ≥nf : τ≥nX → τ≥nY is a weak equivalence. It is a nice exercise to prove this result without appealing to abelianisations. 2.2.2. Exactness of Functors. Before continuing, let us introduce some further useful notions of exactness for functors, following [Sch99] Section 1.1.5. Part 1) of Proposition 2.2.24 above says that the functor I is admissibly exact. This is a stronger notion than exactness. It will be useful in later contexts, so we make a deﬁnition. Definition 2.2.31. A functor F : E → F between exact categories is said to be admissibly (co)exact if for any admissibly (co)acyclic sequence X → Y → Z in E, the sequence F (X) → F (Y ) → F (Z) is admissibly (co)acyclic. A functor which is both admissibly exact and admissibly coexact is said to be strongly exact. Proposition 2.2.32. Let F : E → F be an additive functor between exact categories with kernels. Suppose that F commutes with cokernels and preserves admissible epimorphisms. Then F is admissibly coexact. Dually if F : E → D is an additive functor between categories with cokernels, which commutes with kernels and preserves admissible monomorphisms then F is admissibly exact. Proof. The fact that F preserves cokernels implies that if p : Y → Z is a cokernel of i : X → Y , then F (p): F (Y ) → F (Z) is a cokernel of F (i): F (X) → F (Y ). Therefore Im(F (i)) = Ker(F (p)). Now F (Coim(p)) =∼ F (Coker(i)) =∼ Coker(F (i)). Therefore it remains to show that F (Coim(p)) =∼ Coim(F (p)). ∼ ∼ But both p and F (p) are admissible epimorphisms, so F (Coim(p)) = F (Z) = Coim(F (p)) Example 2.2.33. It is easy to show that taking ﬁnite direct sums is a strongly exact functor. Indeed being both a limit and a colimit, this functor commutes with all limits and colimits. To see that the functor is exact, just use Corollary 2.1.10 We also have the following notion. 16 2. EXACT CATEGORY GENERALITIES

Definition 2.2.34. A covariant functor F : E → F between exact categories is said to be right exact if for any short exact sequence 0 → X → Y → Z → 0 in E, the sequence F (X) → F (Y ) → F (Z) → 0 is admissibly coacyclic in F .

A contravariant functor F : E → F between exact categories is said to be right exact if for any short exact sequence 0 → X → Y → Z → 0 in E, the sequence F (Z) → F (Y ) → F (X) → 0 is admissibly coacyclic in F . Dually one deﬁnes left exactness. In particular Proposition 2.2.32 says that a right exact functor between exact categories with kernels which commutes with cokernels is admissibly coexact.

2.2.3. Homotopies and Quasi-Isomorphisms. 2.2.3.1. Homotopies. Let us now discuss homological properties of maps between complexes.

Definition 2.2.35. A homotopy between morphisms of chain complexes f•, g• : K• → L• is a collection of morphisms Di : Ai → Bi+1 such that K L fi − gi = Di−1 ◦ di + di+1 ◦ Di

We then say f• ∼ g•.

Definition 2.2.36. Two complexes K• and L• are said to be homotopy equivalent if there are maps

g : K• → L• and f : L• → K• such that f ◦ g ∼ idK• and g ◦ f ∼ idL• . If p q A / B / C

α f g X / Y / Z is a diagram with the top and bottom row being null sequences, we will also say that it is homotopic to zero if there are two maps D : B → X and D0 : C → Y such α = f ◦ D − D0 ◦ q. We can use homotopies in an exact category to test for acyclicity.

Proposition 2.2.37. Let E be a weakly idempotent complete exact category, and let

f g X / Y / Z be a null sequence. Suppose that g has a kernel. Then the induced map f 0 : X → Ker(g) is an admissible epimorphism if and only if there is a diagram

p q A / B / C

α f g X / Y / Z

which is homotopic to zero, and such that the induced map αe : Ker(q) → Ker(g) is an admissible epic. 2.2. HOMOLOGICAL PROPERTIES OF EXACT CATEGORIES 17

Proof. Suppose that g has a kernel and that the induced map f 0 : X → Ker(g) is an admissible epimorphism. Consider the diagram 0 / X / 0

f f g X / Y / Z By assumption the induced map fe : X → Ker(g) is an admissible epic. Moreover the diagram is clearly homotopic to 0 via the maps D = id : X → X and D0 = 0 : 0 → Y . Conversely suppose we have a diagram

p q A / B / C D D0 α ~ f g X / Y / Z 0 such that g has a kernel, α = f ◦ D − D ◦ q, and αe is an admissible epic. We have the factorisation of f

fe X / Ker(g) / Y

Moreover, αe = fe◦ D|Ker(q). By Proposition 2.1.8 fe is an admissible epic. Corollary 2.2.38. Let E be a weakly idempotent complete exact category, and let

f g X / Y / Z be a null sequence. The sequence is admissibly acyclic if and only if g is weakly left admissible and there is a diagram p q A / B / C

α f g X / Y / Z which is homotopic to zero, and such that the induced map αe : Ker(q) → Ker(g) is an admissible epic. Proof. Suppose the sequence is admissibly acyclic. By Remark 2.2.22 g is weakly left admissible. For the converse, note that by Proposition 2.2.37 and the fact that Ker(g) → Y is admissible, we have a decomposition of f X Ker(g) Y By Proposition 2.2.19 f is an admissible morphism whose image is Ker(g). We can also test split exactness by looking at homotopy.

Proposition 2.2.39. Let E be a weakly idempotent complete exact category, and let

f g Γ := X / Y / Z be a null sequence. The sequence is admissibly acyclic in the split exact structure if and only if g is weakly left admissible and the diagram f g X / Y / Z

idX idY idZ f g X / Y / Z is homotopic to zero. 18 2. EXACT CATEGORY GENERALITIES

Proof. Suppose the diagram is homotopic to the zero. If we can show that g is also weakly left admissible in the split exact structure, then the claim follows from Corollary 2.2.38. By Corollary 2.2.38 we already know that the sequence is admissibly acyclic, so Im(f) =∼ Ker(g). Let D : Y → X and D0 : Z → Y 0 be maps such that idY = f ◦ D − D ◦ g. The map f ◦ D : Y → Y factors as

(·f◦D) i Y / Im(f) / Y where i is the inclusion. But f ◦ D ◦ i = f ◦ D ◦ i − D ◦ g ◦ i = i ∼ since g ◦ i = 0. It follows that (·f ◦ D) ◦ i = IdIm(f). This implies that the map Ker(g) = Im(f) → Y is split, and so is an admissible monic in the split exact structure.

Corollary 2.2.40. Let X• be a good complex in a weakly idempotent complete exact category E .

(1) X• is acyclic whenever there is a complex Y•, a morphism of complexes f• : Y• → X• which is Y X homotopic to 0, and such that the induced maps fn : Ker(dn ) → Ker(dn ) are admissible epimorphisms.

(2) X• is split exact whenever idX• is homotopic to 0. Proof. The ﬁrst assertion follows from Proposition 2.2.28 and Proposition 2.2.37. For the second assertion note that X• is acyclic by the ﬁrst assertion. In particular each

Xn+1 → Xn → Xn−1

is acyclic, and Xn → Xn−1 is (weakly left) admissible. Thus we may use Proposition 2.2.39. 2.2.3.2. Quasi-isomorphisms. Recall that in an abelian category a map of complexes induces a map on homology. The map is said to be a quasi-isomorphism if the induced map on homology is an isomorphism. Quasi-isomorphisms can also be characterised in terms of their mapping cone. A map of chain complexes in an abelian category is a quasi-isomorphism if and only if its mapping cone is acyclic. As remarked previously, in an exact category we cannot in general deﬁne the homology of a complex. However the construction of the mapping cone makes sense in any additive category. By the previous remarks, the following deﬁnition is sensible.

Definition 2.2.41. Let E be an exact category. A map f• : X• → Y• of complexes of E is said to be a quasi-isomorphism if cone(f•) is acyclic. Proposition 2.2.42. If E is idempotent complete then homotopy equivalences are quasi-isomorphisms.

Proof. See [B¨uh10] Proposition 10.9. The next proposition is an immediate consequence of Corollary 2.2.29.

Proposition 2.2.43. Let I : E → A be a left abelianisation of an exact category E. Let f• : X• → Y• be a morphism of complexes. Suppose cone(f) is good. Then f is a quasi-isomorphim if and only if I(f) is.

Remark 2.2.44. As for abelian categories, one can define the derived category D∗(E) of an exact category E by localizing Ch∗(E) at the quasi-isomorphisms. For details see for example [Büh10] Section 10. 2.2.4. Ext Groups and Projective Objects. In order to study cotorsion pairs in exact categories in Section 4, we will need the notion of Ext groups in exact categories. Recall for an abelian category A one can define the groups Extn(A, B) for any pair of objects A, B ∈ A regardless of whether A has enough projectives by the Yoneda construction. This construction goes through mutatis-mutandis for exact categories. The elements are Yoneda equivalences classes of n-extensions and the binary operation is the Baer sum. All the proofs for the above facts work as the abelian case. The interested reader can adapt the relevant proofs in [Buc59] for example. The first Ext group Ext1(A, B) can also be computed by passing to a left abelianisation. More generally we have the following straightforward result. 2.2. HOMOLOGICAL PROPERTIES OF EXACT CATEGORIES 19

Proposition 2.2.45. Let E and F be exact categories. Let F : E → F be a fully faithful exact functor which reflects exactness. Suppose that the essential image of E is closed under extensions. Then F induces a natural isomorphism of abelian groups 1 ∼ 1 ExtE (−, −) = ExtF (F (−),F (−)) Remark 2.2.46. In the above we make the implicit assumption that each Extn(A, B) is a set. This always holds for exact categories with enough projectives, which can be seen from the discussion in the following section. At this point we recall the notion of a projective object in an exact category, and mention how they relate to the Ext functor. Definition 2.2.47. An object P in an exact category E is said to be projective if the functor Hom(P, −): E → Ab is exact. Remark 2.2.48. By Proposition 2.2.32, for any projective object P the functor Hom(P, −) is admissibly exact. Example 2.2.49. In the split exact structure every object is projective. As in the abelian case one has the following result. The equivalence of the first three conditions can be found in [Büh10] Proposition 11.3. Proposition 2.2.50. The following are equivalent. (1) P is projective. (2) Given a map f : P → C and an admissible epic e : B → C, there is a morphism g : P → B such that the following diagram commutes B ? g e f P / C (3) Any admissible epic with codomain P splits. (4) Ext1(P,A) vanishes for any object A. (5) Extn(P,A) vanishes for any object A and any n ≥ 1. 2.2.5. Exact Structures on Chain Complexes. Let E be an exact category and consider the category Ch∗(E) for ∗ ∈ {∅, b, ≥, ≤, +, −}. Say that a sequence 0 → A• → B• → C• → 0 is exact precisely if for each i ∈ Z the sequence 0 → Ai → Bi → Ci → 0 is exact. Since limits and colimits in Ch∗(E) are computed degree-wise this is an exact structure on Ch(E). Proposition 2.2.51. Let F : A → B be a fully faithful exact functor which reflects exactness and whose essential image is closed under extensions. Then for ∗ ∈ {≥ 0, ≤ 0, +, −, b, ∅} the induced functor

Ch∗(F ): Ch∗(A) → Ch∗(B) is a fully faithful exact functor which reﬂects exactness and whose essential image is closed under extensions.

Proof. Since exactness of chain complexes is deﬁned level wise, Ch∗(F ) is clearly exact and reﬂects exactness. It is clearly faithful. Let us check that it is full. Let (X•, d•) and (Y•, δ•) be chain complexes in A. Let f• : F (X•) → F (Y•) be a chain map. For each n there is some gn : Xn → Yn with fn = F (gn). Moreover

F (gn ◦ dn+1) = F (gn) ◦ F (dn+1) = fn ◦ F (dn+1) = F (δn+1) ◦ fn+1 = F (δn+1 ◦ gn+1)

Since F is faithful, gn ◦ dn+1 = δn+1 ◦ gn+1. It remains to show that the essential image of Ch∗(F ) is closed under extensions. So suppose we have an exact sequence of chain complexes.

f• g• 0 / F (X•, d•) / (Q•, γ•) / F (Y•, δ•) / 0 20 2. EXACT CATEGORY GENERALITIES

0 −1 For each n pick an object Pn ∈ A and an isomorphism pn : Qnf→F (Pn). Let γn = pn−1 ◦ γn ◦ pn : 0 F (Pn) → F (Pn−1). Then (P•, γ•) is a chain complex. Moreover by construction we have an isomorphism p• : Q• → F (P•) whose nth component is pn.

Corollary 2.2.52. Let I : E → A(E) is a left abelianisation of E. Then Ch∗(I): Ch∗(E) → Ch∗(A(E)) is a left abelianisation of Ch∗(E).

Proof. By the previous proposition, it remains to check that Ch∗(I) preserves kernels, and Ch∗(I)(f•) is an admissible epimorphism if and only if f• is. However this is clear since everything is computed degree- wise.

2.2.6. A Useful Example: The Degree-Wise Exact Structure. Let E be an additive category, and endow it with the split exact structure. The induced exact structure on Ch(E) is called the degree-wise n split exact structure, and we denote the Ext functors in this structure by Extdw. We conclude this section with a brief discussion of the relation between extensions in the degree-wise split exact structure and the Ch(Ab)-enriched structure on Ch(E). This is also done in a model theoretic context for modules over a ring in [Gil11] Section 5.2 and the proofs are formally the same.

p• q• Proposition 2.2.53. A sequence of chain complexes 0 / X• / Z• / Y• / 0 is split exact in each degree if and only if it is isomorphic to a complex of the form

0 → X• → cone(f•) → Y• → 0

for some morphism of complexes f• : Y•[1] → X•. Proof. The sequence

0 → X• → cone(f•) → Y• → 0 is clearly split exact in each degree, so any complex isomorphic to it is split exact in each degree as well. Suppose

p• q• 0 / X• / Z• / Y• / 0

is split exact in each degree. Let αn : Zn → Xn be such that αn ◦pn = idXn and βn : Yn → Zn be a map such Z that qn ◦ βn = idYn . We may assume also that αn ◦ βn = 0. Deﬁne f• : Y•[1] → X• by fn = αn ◦ dn+1 ◦ βn+1. This is easily seen to be a map of chain complexes. Let αn : Zn → Xn ⊕ Yn denote the isomorphism induced by the degree-wise splitting. A straightforward computation shows that this gives a map of chain complexes α• : Z• → cone(f•). Thus we get an isomorphism of exact sequences.

p• q• 0 / X• / Z• / Y• / 0

α• 0 / X• / cone(f•) / Y• / 0

Definition 2.2.54. (1) A complex X• is said to be contractible if the map X• → 0 is a homotopy equivalence. (2) A complex X• is said to be split acyclic if it is acyclic in the split exact structure.

Proposition 2.2.55. A map of chain complexes f• : X• → Y• is homotopic to 0 if and only if the sequence

0 → Y• → cone(f•) → X•[−1] → 0 is split exact. 2.3. RESOLUTIONS IN EXACT CATEGORIES 21

Proof. Suppose that f• is homotopic to 0. Let {Dn : Xn → Yn+1} be a homotopy. We then get a map αn = (idXn−1 ,Dn−1): Xn−1 → cone(f)n. It is straightforward to check that this gives a chain map α• : X•[−1] → cone(f•). Moreover it obviously gives a splitting of cone(f•) → X•[−1]. Conversely suppose the sequence is split exact. Let α• : X•[−1] → cone(f•) be a splitting of the map cone(f•) → X•[−1]. It is an easy computation to check that the collection of compositions {Dn−1 : Xn−1 → cone(f•)n → Yn} is a homotopy between f and 0. We recover the following standard result (see e.g. [Gil16a] Lemma 2.8).

Corollary 2.2.56. For chain complexes X•,Y• in an additive category E. we have 1 ∼ Extdw(X,Y [n − 1]) = HnHom(X•,Y•) = HomCh(E)(X,Y [n]) ∼ where ∼ is chain homotopy. Q Proof. By direct computation, one ﬁnds that f ∈ i Hom(Xi,Yi+n) deﬁnes a chain map f• : X• → Y•[n] if and only if f ∈ Ker(dn). Similarly, f• is then null-homotopic if and only if it is in Im(dn+1). This gives the isomorphism HnHom(X•,Y•) = HomCh(E)(X,Y [n]) ∼ 1 ∼ The isomorphism Extdw(X,Y [n − 1]) = HomCh(E)(X,Y [n]) ∼ follows from Proposition 2.2.53 and Propo- sition 2.2.55.

2.3. Resolutions in Exact Categories We will need some results about resolutions in exact categories later.

2.3.1. Bounded Resolutions. We begin by discussing the easier case of bounded resolutions.

Definition 2.3.57. An exact category E is said to have enough projectives if for any object X of E,there is a projective object P and an admissible epimorphism P X. Lemma 2.3.58. Let P be a subclass of Ob(E), the objects of E. Assume that for any object E of E there is an object P ∈ P and an admissible epimorphism P E. Then, for any bounded below complex E of Ch+(E), there is a bounded below complex P whose entries are objects of P, and a quasi-isomorphism u : P → E

where each uk : Pk → Ek is an admissible epimorphism. Moreover, this construction can be made functorial if the choice of admissible epimorphism P E can be made functorial. Proof. This is proved in [B¨uh10] Theorem 12.7 for the case that P is the class of projectives in an exact category with enough projectives. However the proof goes through the same.

Lemma 2.3.59. Let A, B be objects in an exact category E. Let f : A → B be a morphism. Let P• be a complex with P−1 = A, Pn = 0 for n < −1 and Pn projective for n > 0. Also let Q• be an acyclic complex with Q−1 = B and Qn = 0 for n < −1. Then there is a chain map f• : P• → Q• with f−1 = f. Moreover, f• is unique up to homotopy.

Proof. See [B¨uh10] Theorem 12.4. As in the abelian case one can deﬁne derived functors between derived categories of exact categories. There are also notions of adapted classes for functors. Proposition 2.2.50 and Lemma 2.3.58 essentially say that as in the abelian case, if a category E has enough projectives, then the class of projective objects is op n . ∼ adapted to the functor Hom(−,A): E → Ab. It can be shown that R Hom(−,A) .= Hn(RHom(−,A)) = Extn(−,A). 22 2. EXACT CATEGORY GENERALITIES

2.3.2. Unbounded Resolutions. When dealing with the model structures on unbounded chain complexes., we will also need to have unbounded resolutions. For this we will modify the famous Theorem 3.4 in [Spa88] and its proof to work for more general exact categories. In the following we shall let B be a class of complexes in E which is stable under shifts, and we shall assume that for any bounded below complex X• there is a bounded below complex B• in B and a quasi-isomorphism B• → X• which is an admissible epimorphism in each degree. We will call such a class a bounded resolving class. Let us now recall some notions from Spaltenstein’s paper.

n Definition 2.3.60. Let B be a class of complexes. A direct system (P• )n∈E in Ch(E) is a B-special direct system if it satisﬁes the following conditions. (1) E is well-ordered. n m (2) If n ∈ E has no predecessor then P• = lim→m

We denote by lim→B the class of complexes which are limits of B-special direct systems. Proposition 2.3.61. Let E be an exact category which has kernels. Suppose that B is a bounded resolving n class. Then for any complex X• there exists a B-special direct system (P• )n≥−1 and a direct system of chain n n maps f : P• → τ≥nX• such that (1) f n is a quasi-isomorphism for every n ≥ 0. (2) f n is an admissible epimorphism in each degree.

n n −1 Proof. We construct the data (P• )n≥−1 and (f )n≥−1 by induction. For n = −1 we take P• = 0 −1 −1 n−1 −1 n−1 and so f = 0. Let now n ≥ 1, and suppose that P• ,...,P• and f , . . . , f have been constructed. n−1 n−1 Let P• = P• and Y• = τ≥nX•. Denote by f the composite P• → τ≥n−1X• → Y•. By assumption we can find a quasi-isomorphism g : Q• → cone(f)[1] which is an admissible epimorphism in each degree, and Q•[−1] ∈ B. Now we have a degree-wise splitting, cone(f)[1] = P• ⊕ Y•[1]. We therefore get two maps 0 00 0 g : Q• → P• and g : Q• → Y•[1] which are admissible epimorphisms in each degree, and such that g is a chain map. Define P n .= cone(−g0) and let f n : cone(−g0) = Q[1] ⊕ P → Y be defined by f n = g00[1] + f. As in [Spa88] Lemma 3.3, by direct calculation f n is a chain map and cone(f n) = cone(g)[1]. Since g is a quasi-isomorphism f n is as well. Moreover the sequence n−1 n 0 → P• → P• → Q•[1] → 0 is split exact in each degree. Following the work of [CNPS17] in the relative homological algebra setting, we will give a general condition under which unbounded resolutions exist.

Definition 2.3.62. Let E be an exact category with kernels and let k ∈ Z. A complex X is said to be homologically concentrated in degrees ≤ k if τ≥kX is acyclic. The following deﬁnition is an exact category version [CNPS17] Deﬁnition 6.1.

Definition 2.3.63. Let E be an exact category with kernels and k ∈ Z. A bounded resolving class B is said to satisfy condition AB4-k if whenever {Bn}n∈ is a collection of objects of B concentrated in degree L N ≤ 0, then Bn exists and is homologically concentrated in degrees ≤ −k + 1. n∈N Lemma 2.3.64. Let E be an exact category with kernels and countable ﬁltered colimits, and let B be a bounded resolving class satisfying condition AB4 − k. Let

K0 → K1 → ... → Km → Km+1 → ... 2.3. RESOLUTIONS IN EXACT CATEGORIES 23 be a sequence with each Km → Km+1 an admissible monomorphism, and each Km is a complex in B which is homologically concentrated in degrees ≤ n. Then lim→m Km is homologically concentrated in degrees ≤ n − k + 1. Proof. The proof is an adaptation of [CNPS17] Proposition 6.3. The sequence a a 0 → Kn → Kn → colimnKn → 0 n n is degree-wise split. Indeed this can be proven by observing that the sequence Y Y 0 → limnHom(Kn,E) → Hom( Kn,E) → Hom( Kn,E) → 0 n n for any E ∈ E is a degree-wise split exact sequence of complexes of abelian groups. By passing to an ` abelianisation and using the long exact sequence, on homology, it suﬃces to prove that n Kn is concentrated in degrees ≤ n − k + 1. By shifting we may assume that n = 0, and then we are done. Corollary 2.3.65. Let E be an exact category with kernels. Let B be a bounded resolving class satisfying condition AB4 − k for some k. Then any chain complex X• in E admits a lim→B resolution which is an admissible epimorphism in each degree.

n n n Proof. Fix a B-special direct system (P• )n≥−1 and a direct system of chain maps f : P• → τ≥nX• such that (1) f n is a quasi-isomorphism for every n ≥ 0. (2) f n is an admissible epimorphism in each degree. n Let P• be the direct limit of the special direct system. For each n the composition P• → P• → X• is an admissible epimorphism in degrees > n. Thus P• → X• is an admissible epimorphism in all degrees. It remains to prove that P• → X• is a weak equivalence. This can be proven as in [CNPS17] Theorem 6.4, mutatis mutandis. Now let P be any class of objects in E. Suppose that for each object X in E there is an object P in P together with an admissible epimorphism P X. By Lemma 2.3.58 the class Ch+(P) of chain complexes with entries in P is a bounded resolving class. Let us introduce the following notion.

Definition 2.3.66. Let E be an exact category. We say that a class of objects P in E satisﬁes condition AB4 − k for some k if Ch+(P) satisﬁes condition AB4 − k. From the proof of Corollary 2.3.65 we then immediately have the following.

Corollary 2.3.67. Let E be a weakly idempotent complete exact category with kernels. Let P be a class of objects such that for each object X in E there is an object P in P together with an admissible epimorphism P X. Suppose further that P satisﬁes condition AB − k for some k ∈ Z. Then for any complex X• in Ch(E) there is a complex P• in Ch(P) and an admissible epimorphism P• → X• which is a quasi-isomorphism. Moreover, X• is the limit of a Ch+(P)-special direct system.

For a class of objects P in an exact category E, let AdMonP denote the class of admissible monomorphisms whose cokernel is in P. We will also need the following acyclicity result, also proved in [Spa88] for abelian categories.

Proposition 2.3.68. Let T be a class of complexes in Ch(E). The class of all complexes A• ∈ Ch(E) such that Hom(A•,T•) is acyclic for every T• in T is closed under special direct limits.

Proof. It is clear from the deﬁnition of the contravariant functor Hom(−,T•) that it transforms n n colimits into limits. If (P• )n∈E is a B-special direct system then (Hom(P• ,T•))n∈E is a B-special inverse system of acyclic complexes of abelian groups, where we use the terminology of [Spa88] Section 2. Lemma 2.3 in [Spa88] says that the inverse limit of such a system is again acyclic. 24 2. EXACT CATEGORY GENERALITIES

2.4. Monoidal Exact Categories and Monads in Exact Categories We conclude this section with a brief note on monoidal exact categories, and exact structures on categories of modules over monoids. More generally we put an exact structure on the category of algebras for an additive monad which is compatible in a precise sense with the exact structure on the underlying category. First we need a general definition. Definition 2.4.69. (1)A monoidal exact category is a symmetric monoidal category (E, ⊗, k) where E is an exact category, and the monoidal functor ⊗ : E × E → E is cocontinous in each variable. E is said to be compatibly monoidal if in addition X ⊗ (−) is right exact for any X ∈ E. (2)A closed monoidal exact category is a closed symmetric monoidal category (E, ⊗, k, Hom) with E an exact category. E is said to be compatibly closed monoidal if in addition X ⊗ (−) is right exact, and Hom(X, −), Hom(−,X) are left exact for any X ∈ E Note that if (E, ⊗, Hom, k) is a closed monoidal exact category, then (E, ⊗, k) is automatically a monoidal exact category. Indeed for each object X, X ⊗ (−) is a left adjoint so it preserves colimits. Definition 2.4.70. Let (E, ⊗, k) be an exact category equipped with a symmetric monoidal structure where the tensor functor is additive. An object F of E is said to be (strongly) flat if the functor F ⊗ (−) is (strongly) exact. In the familiar category of R-modules over some ring R with the usual monoidal structure, projectives are always flat. Moreover the tensor product of two projective R-modules is again projective. This is not always guaranteed for an arbitrary monoidal exact category. However it is a useful property to have, in particular when dealing with the projective model structure later. We therefore make a definition. Definition 2.4.71. A monoidal exact category in which projective objects are flat and P ⊗P 0 is projective whenever both P and P 0 are is said to be projectively monoidal . A projectively monoidal exact category is said to be strongly projectively monoidal if projectives are strongly flat. In closed exact categories we have the following observation. Observation 2.4.72. Let (E, ⊗, k, Hom) be a closed monoidal exact category with enough projectives such that the underlying monoidal category is projectively monoidal. Then for any projective P , the functor Hom(P, −): E → E is exact. The proof follows immediately from the adjunction between ⊗ and Hom. It is shown in the quasi-abelian case in [Sch99], for example, and the proof works identically in the exact case. Now let R be a unital associative monoid internal to a monoidal exact category (E, ⊗, k). It turns out that there is an exact structure on the additive category RMod where a null sequence 0 → X → Y → Z → 0

in RMod is exact if and only if it is a short exact sequence when regarded as a null sequence in E. This follows from a more general result about cocontinuous monads in exact categories. Proposition 2.4.73. Let E be an exact category and let T : E → E be a cocontinuous monad. There is an exact structure on ET where a null sequence 0 → X → Y → Z → 0 in ET is exact if and only if it is a short exact sequence when regarded as a null sequence in E. We call this exact structure the induced exact structure. Proof. This follows from the general fact that if T is a cocontinuous monad on any category E then the forgetful functor ET → E creates limits and colimits and reﬂects isomorphisms. For a proof of this see [Bor94] Proposition 4.3.1 and Proposition 4.3.2. This exact structure inherits a lot from the exact structure on E. In fact we have the following lemma. 2.4. MONOIDAL EXACT CATEGORIES AND MONADS IN EXACT CATEGORIES 25

Lemma 2.4.74. Let | − | : D → E be a functor between exact categories which reflects exactness and creates both kernels and cokernels. Then | − | reflects admissible monomorphisms, admissible epimorphisms, weakly admissible morphisms, admissible morphisms, and admissibly acyclic sequences. Proof. Let f : X → Y be a morphism in D. Supposethat |f| is an admissible monomorphism. Then there is an exact sequence 0 → |X| → |Y | → Coker(|f|) → 0 Since | − | creates cokernels and reflects exactness 0 → X → Y → Coker(f) → 0 is an exact sequence in D. Thus f is an admissible monomorphism. That |−| reflects admissible epimorphisms is proved similarly. Note in particular that this means | − | reflects isomorphisms. Suppose now that |f| : |X| → |Y | is weakly admissible. Then there is a decomposition

|f| |X| / |Y | ; ;

; $ $ |fˆ| ; $ $ Ker(|f|) Coim(|f|) / Im(|f|) Coker(|f|) Since | − | reﬂects exactness and creates both kernels and cokernels there is a decomposition in D

f X / Y ; <

; # # fˆ < $ $ Ker(f) Coim(f) / Im(f) Coker(f) Thus f is weakly admissible. If in addition |f| is admissible then |fˆ| is an isomorphism. Since | − | reflects isomorphisms fˆ is an isomorphism, so f is admissible. Finally suppose |f| |g| |X| / |Y | / |Z| is admissibly acyclic. Then |f| is admissible, |g| has a kernel and the map Im(|f|) → Ker(|g|) is an isomorphism. By the above f is admissible. Since | − | creates kernels and cokernels and also reflects isomorphisms Im(f) → Ker(g) is also an isomorphism. As a consequence of this and Remark 2.5.86 later, if E is (quasi)-abelian, then so is ET . In particular categories of modules for monoid objects in monoidal (quasi)-abelian categories are themselves (quasi)- abelian. Before concluding this discussion of monoidal exact categories, let us briefly mention induced monoidal structures on chain complexes. So, let (E, ⊗, k) be a monoidal exact category. Recall from Section 1.3 0 there is an induced monoidal exact structure (Ch∗(E), ⊗,S (k)) on Ch∗(E) for ∗ ∈ {≥ 0, ≤ 0, +, −, b, ∅}. Since colimits of chain complexes are computed degree-wise, finite direct sums are strongly exact, and a null sequence of chain complexes is admissibly coacyclic if and only if it is so in each degree, it is clear that this 0 monoidal structure is compatible if the one on E is, so that (Ch∗(E), ⊗,S (k)) is a compatible monoidal exact category for ∗ ∈ {≥ 0, ≤ 0, +, −, b}. Now suppose (E, ⊗,S0(k), Hom) is a closed monoidal exact category. Then

0 (Chb(E), ⊗,S (k), Hom) is a closed monoidal exact category which is compatible if the closed monoidal exact structure on E is. Note that the closed symmetric monoidal exact category

(Ch(E), ⊗,S0(k), Hom) 26 2. EXACT CATEGORY GENERALITIES

need not be a compatible closed monoidal exact category since inﬁnite direct sums/ products need not be admissibly coexact/ admissibly exact. When we deal with unbounded complexes later we shall assume this to be the case. However it will still be a monoidal exact category, and we shall see shortly that compatibility is guaranteed for a closed monoidal structure on a quasi-abelian category.

2.4.1. The ⊗-Pure Exact Structure. In this section we discuss pure exact structures on exact categories relative to tensor products. This gives a method for modifying an exact structure in order to make a certain class of objects ﬂat. This is done in [EGO17] Section 3 for the case of the G-exact structure on a Grothendieck abelian category.

Definition 2.4.75. Let E be a monoidal exact category, and S a class of objects in E. A short exact sequence in E is said to be S-pure if it remains exact after tensoring with any object of E. An Ob(E)-pure exact sequence is said to be ⊗-pure

Lemma 2.4.76. Suppose E is a symmetric monoidal weakly idempotent complete exact category with enough ﬂat objects. If C ∈ E is ﬂat then every short exact sequence 0 → A → B → C → 0

is pure.

Proof. Suppose Z is arbitrary and let

0 → X → Y → Z → 0

be a short exact sequence with Y ﬂat. Let J : E → A be a right abelianisation. Since the functor J(− ⊗ −) commutes with cokernels in each variable, we have a diagram

J(A ⊗ X) / J(A ⊗ Y ) / J(A ⊗ Z) / 0

J(B ⊗ X) / J(B ⊗ Y ) / J(B ⊗ Z) / 0

J(C ⊗ X) / J(C ⊗ Y ) / J(C ⊗ Z) / 0

0 0 0 with coacyclic rows and columns. The bottom row is short exact since C is ﬂat. Since Y is ﬂat the middle column is short exact. Then the argument becomes a simple diagram chase.

Proposition 2.4.77. The class of S-pure exact sequences deﬁnes an exact structure on E for E weakly idempotent complete.

Proof. Clearly split exact sequences are S-pure, and isomorphisms are both S-pure monomorphisms and S-pure epimorphisms. Next we show show that S-pure monomorphisms are stable under pushout. Let i : A → B be a S-pure monomorphism. Consider a pushout diagram

A / B

X / Y 2.4. MONOIDAL EXACT CATEGORIES AND MONADS IN EXACT CATEGORIES 27

Since tensoring with Z preserves colimits,

A ⊗ Z / B ⊗ Z

X ⊗ Z / Y ⊗ Z is a pushout. But by assumption A ⊗ Z → B ⊗ Z is an admissible monic. Hence X ⊗ Z → Y ⊗ Z is also an admissible monic. Now let 0 / X / P / Z0 / 0

g 0 / X / Y / Z / 0 be a commutative diagram where, g is a pure epimorphism, both rows are exact, and the right-hand square is a pullback. We need to show that the top row is pure-exact. Let C be an object of S. The composite map X ⊗ C → P ⊗ C → Y ⊗ C is an admissible monomorphism, so X ⊗ C → P ⊗ C is an admissible monomorphism. The cokernel is P ⊗ C → Z0 ⊗ C. In particular the sequence 0 → X ⊗ C → P ⊗ C → Z0 ⊗ C → 0 is exact, and P → Z0 is a pure epimorphism. It is clear that the composition of two pure monomorphisms is a pure monomorphism. Let us show that the composition of two pure epimorphisms is a pure epimorphism. Let f : X → Y and g : Y → Z be pure epimorphisms, and C an object of S. We have that Ker(f) ⊗ C is the kernel of the admissible epimorphism X ⊗ C → Y ⊗ C. Therefore Ker(f) ⊗ C → X ⊗ C is an admissible monomorphism. This map factors through Ker(f) ⊗ C → Ker(g ◦ f) ⊗ C and is therefore an admissible monomorphism. It follows that 0 → Ker(f) ⊗ C → Ker(g ◦ f) ⊗ C → Ker(g) ⊗ C → 0 is a short exact sequence. Consider the diagram 0 0 0

0 / Ker(f) ⊗ C / Ker(f) ⊗ C / 0 / 0

0 / Ker(g ◦ f) ⊗ C / X ⊗ C / Z ⊗ C / 0

0 / Ker(g) ⊗ C / Y ⊗ C / Z ⊗ C / 0

0 0 0 The top row is obviously exact, and the bottom is exact since g is a pure epimorphism. We have shown that the left-hand column is exact, and the right is obviously exact. The middle is exact since f is a pure epimorphism. Therefore the middle row is exact.

Notation 2.4.78. If (E, Q) is an exact category equipped with a closed symmetric monoidal structure, we will denote the ⊗-pure exact structure by (E, Q⊗). Remark 2.4.79. In the ⊗-pure exact structure all objects are ﬂat. 28 2. EXACT CATEGORY GENERALITIES

2.5. Quasi-Abelian Categories Let us apply what we have seen so far to the particular case of quasi-abelian categories. The theory of quasi-abelian categories is developed signiﬁcantly in [Sch99] which is our main reference here. Applications to categories of topological vector spaces can be found in [Pro00b].

2.5.1. Strict Morphisms. First we explain as in [Sch99] Remark 1.1.11 that Deﬁnition 2.1.4 is equivalent to the one given in [Sch99]. Recall that in a ﬁnitely complete and cocomplete additive category, any morphism f : E → F gives rise to a commutative diagram

f E / F O

Coim(f) / Imf In any abelian category the map Coim(f) → Im(f) is an isomorphism. However this is not true in general. −1 For example, consider the standard example of the category Fr of Fr´echet spaces. Then Coim(f) = E f (0), Im(f) = f(E) and the natural map Ef −1(0) → f(E) is the obvious one. By the Open Mapping Theorem Coim(f) → Im(f) is an isomorphism if and only if f has closed range, which is not always the case. Definition 2.5.80. Let E be an additive category with all kernels and cokernels. A morphism f : E → F in E is said to be strict if Coim(f) → Im(f) is an isomorphism. Proposition 2.5.81 ([Sch99] Remark 1.1.2). Let E be a ﬁnitely complete and cocomplete additive category. (1) A monic is strict if and only if it is the kernel of some morphism. In this case it is the kernel of its cokernel. (2) An epic is strict if and only if it is the cokernel some morphism. In this case it is the cokernel of its kernel.

Proof. (1) Let f : E → F and write if : Ker(f) → E for the canonical map. Let us show that if is strict. First note that for any monic A → B, the coimage is id : A → A. Let us compute the image of if . It is given by Ker(Coker(Ker(f) → E) → E By some abstract nonsense this is just Ker(f) → E. Conversely suppose m : X → E is a strict monic. Then the maps E → Coim(m) → Im(m) are all isomorphisms, i.e. we get a commutative diagram m X / E O ∼

∼ Coim(m) / Im(m) Since Im(m) → E is a kernel of Coker(m), so is m : X → E. (2) This is dual to the ﬁrst part. Proposition 2.5.82. The class of strict epics (resp. monics) in a quasi-abelian category E is stable by composition.

Proof. See [Sch99] Proposition 1.1.7. Corollary 2.5.83. A ﬁnitely complete and cocomplete additive category E is quasi-abelian if and only if the following two conditions hold. 2.5. QUASI-ABELIAN CATEGORIES 29

(1) If f A / B

f 0 X / Y is a pushout diagram, and f is a strict monic, then f 0 is as well. (2) If f 0 A / B

f X / Y is a pullback diagram, and f is a strict epic, then f 0 is as well. Let us now describe the admissible morphisms in the quasi-abelian exact structure. Proposition 2.5.84 ([Sch99] Remark 1.1.2). Let E be a finitely complete and cocomplete additive category. A morphism f : E → F in E is strict if and only if it can be written as f = i ◦ p where p : E → I is a strict epic and i : I → F is a strict monic. Proof. Suppose f admits a decomposition f = i ◦ p as in the statement. Then Ker(f) = Ker(p). So Coim(f) = Coim(p). Since p is strict Coim(p) =∼ Im(p). Since p is an epic, Im(p) = I. Similarly Im(f) = Im(i) = I. Conversely suppose f is a strict morphism. Now E → Coim(f) is a strict epic, and Im(f) → F is a strict monic. But since f is strict, Coim(f) =∼ Im(f), so this gives the decomposition of f. Corollary 2.5.85. A morphism in a quasi-abelian category is admissible in the quasi-abelian exact structure if and only if it is strict. Remark 2.5.86. An exact structure on a finitely complete and cocomplete additive category coincides with the quasi-abelian structure if and only if every morphism is weakly admissible. Then as a consequence of Proposition 2.5.81, a finitely complete and cocomplete additive category is abelian if and only if every morphism is admissible. 2.5.2. The Left Heart. Homology in quasi-abelian categories is significantly easier than in more general exact categories. For example, there is an even stronger abelian embedding. Theorem 2.5.87. Let E be a quasi-abelian category. There exists a left abelianisation I : E → LH(E) of ∼ E such that I has a left adjoint C : LH(E) → E with C ◦ I = idE , i.e. E is a reflective subcategory of LH(E). Moreover the induced functor on derived categories D(I): D(E) → D(LH(E)) is an equivalence.

Proof. See [Sch99] Proposition 1.1.26, Corollary 1.2.27, Proposition 1.2.28, and Proposition 1.2.31. LH(E) is called the left heart of E. The embedding of E into its left heart also behaves extremely well with respect to projectives. Proposition 2.5.88. (1) An object P of E is projective if and only if I(P ) is projective in LH(E). (2) E has enough projectives if and only if LH(E) has enough projectives. In this case an object of LH(E) is projective if and only if it is isomorphic to I(P ) where P is projective in E. Proof. See [Sch99] Proposition 1.3.24. Moreover left abelianisations of quasi-abelian categories allow us to test acyclicity of any unbounded complex. Indeed as a consequence of Remark 2.5.86 and Corollary 2.2.29 we get. 30 2. EXACT CATEGORY GENERALITIES

Corollary 2.5.89. Let I : E → A be a left abelianisation of E where E is a quasi-abelian category. Then a complex X• in E is acyclic if and only if I(X•) is acyclic. In particular a map of complexes f : X → Y is a quasi-isomorphism if and only if I(f) is. 2.5.3. Monoidal Quasi-Abelian Categories. Let us briefly discuss (strongly) projectively monoidal quasi-abelian categories, i.e. a (strongly) projectively monoidal exact category in which the underlying exact category is quasi-abelian. We first make the following observation. Observation 2.5.90. An additive functor F : E → F between quasi-abelian categories is right exact if and only if it preserves cokernels of strict morphisms. See for example Section 1.1 in [Sch99]. This implies that if (E, ⊗, k) is a monoidal category with E quasi-abelian and ⊗ additive, then it is a monoidal quasi-abelian category if and only if X ⊗(−) preserves colimits for each object X of E. In particular if (E, ⊗, Hom, k) is a closed monoidal category with E quasi-abelian and ⊗, Hom additive functors, then (E, ⊗, Hom, k) is in fact a closed monoidal quasi-abelian category. Proposition 2.5.91. Let (E, ⊗, Hom, k) be a complete and cocomplete closed monoidal quasi-abelian category which is also projectively monoidal. Then there is a monoidal structure ‹⊗, Homfi on LH(E) such that (LH(E), ‹⊗, Homfi ,I(k)) is a closed monoidal abelian category. Moreover I : E → LH(E) is a lax monoidal functor. If (E, ⊗, Hom, k) is strongly projectively monoidal then (LH(E), ‹⊗, Homfi ,I(k)) is projectively monoidal. Proof. See [Sch99] Proposition 1.5.3 and Corollary 1.5.4.

2.6. Generators in Exact Categories In this recall the definition of an admissible generating collection from [SS11ˇ ], which follows Schneiders’ deinition of strict generators in quasi-abelian categories [Sch99] Definition 2.1.5/ Proposition 2.1.7. This will come into play later when we discuss cofibrant generation of model structures, where some compactness assumptions are required. If G is a collection of objects in an exact category we denote by L G the collection of all small coproducts of objects in G. We will use the word ‘collection’ because we will also be interested in proper classes of generators. Definition 2.6.92. A collection of objects G in an exact category E is said to be an admissible generating collection if for each object E of E there is an object Q of L G and an admissible epimorphism Q E. An admissible generating collection G is said to be a projective generating collection if all objects in G are projective. The next two results are adaptations of the proof of [Sch99] Proposition 1.3.23 to the exact case. Proposition 2.6.93. Let G be an admissible generating collection in a weakly idempotent complete exact category E. Suppose f : E → F is a morphism such that for each G in G then map Hom(G, E) → Hom(G, F ) is an epimorphism. Then f is an admissible epimorphism. Proof. Pick an admissible epimorphism : P → F where P ∈ L G. By assumption there is a morphism 0 0 : P → E such that = f ◦ . By Proposition 2.1.8 f is then an admissible epimorphism. Proposition 2.6.94. Let G be a generating collection in a weakly idempotent complete exact category E. A complex e0 e00 0 / E / E / E00 with e00 weakly left admissible is admissibly acyclic if and only if for each G ∈ G the sequence 0 / Hom(G, E0) / Hom(G, E) / Hom(G, E00) is acyclic in Ab. If in addition the objects of G are projective, then a sequence

e0 e00 E / E / E00 2.6. GENERATORS IN EXACT CATEGORIES 31

with e00 weakly left admissible is admissibly acyclic if and only if for each P ∈ G the sequence

Hom(P,E0) / Hom(P,E) / Hom(P,E00) is acyclic in Ab. Proof. Suppose that for each G ∈ G the sequence 0 / Hom(G, E0) / Hom(G, E) / Hom(G, E00)

is acyclic in Ab. Since e00 is weakly left admissible it is suﬃcient to show that e0 is a kernel of e00. Then e0 is automatically an admissible monic. To show this one can follow the proof in [Sch99]. At one point in that proof the existence of a resolution of X by objects of ⊕G is used. Here instead we may use Lemma 2.3.58 Finally let us consider the assertion about projective generators. Proposition 2.2.32 implies that

Hom(P,E0) / Hom(P,E) / Hom(P,E00) is acyclic. For the converse ﬁrst consider the sequence

e00 0 / Ker(e00) / E / E00 Since Hom(P, −) preserves kernels, Proposition 2.6.93 implies that

e0 e00 E / E / E00 is admissibly acyclic. In particular if E is quasi-abelian, then every morphism is weakly admissible, so in this case one has that a sequence e0 e00 E / E / E00 is admissibly acyclic if and only if for each P ∈ P the sequence

Hom(P,E0) / Hom(P,E) / Hom(P,E00) is acyclic in Ab. For general exact categories we still have the following result. Corollary 2.6.95. Let G be a projective generating collection in a weakly idempotent complete exact category E. Let X• be a complex. Suppose that X• is good. Then X• is acyclic if and only if Hom(G, X•) is acyclic for each G ∈ G. Proof. Since each G ∈ G is projective the functors Hom(G, −) preserve acyclic complexes. Conversely X 0 suppose Hom(G, X•) is acyclic for each G ∈ G, and dn has a kernel ZnX. By assumption Hom(G, dn+1): 0 Hom(G, Xn+1) → ZnHom(G, X) = Hom(G, ZnX) is an epimorphism for each n. Thus dn+1 : Xn+1 → ZnX is an admissible epimorphism. Now apply Proposition 2.2.28. Given a set of generators G of an exact category E it is often possible, as in [Gil16a] to construct an exact structure on E such that G is a set of projective generators. We will discuss this in the next chapter.

2.6.1. Elementary Exact Categories. It is convenient to have generators satisfying some compactness conditions. Recall that a poset J is said to be λ-filtered for a cardinal λ if any subset S of J with |S| < λ has an upper bound. Recall ([CH02] Definition 4.1) that the cofinality of a limit ordinal γ is the smallest cardinal κ such that there exists a subset T of γ of cardinality κ with sup(T ) = γ. If γ is a successor ordinal its cofinality is defined to be 1.

Definition 2.6.96. Let E be a category, S a class of morphisms in E, and κ a cardinal. An object E of E is said to be 32 2. EXACT CATEGORY GENERALITIES

(1) (κ, S)-small if the canonical map lim Hom(E,F ) → Hom(E, lim F ) −→ β −→ β β∈λ β∈λ is an isomorphism for any cardinal λ with cofin(λ) ≥ κ and any λ-indexed transfinite sequence where Fi → Fi+1 is in S. (2) S-small if it is (κ, S)-small for some cardinal κ (3) (κ, S)-compact if the canonical map lim Hom(E,F ) → Hom(E, lim F ) −→ β −→ β β∈λ β∈λ is an isomorphism for any regular cardinal λ ≥ κ and any λ-indexed transfinite sequence where Fi → Fi+1 is in S. (4) S-compact if it is (κ, S)-small for some cardinal κ (5) (κ, S)-presented if the natural map lim Hom(E,F ) → Hom(E, lim F ) −→ i −→ i i∈I i∈I is an isomorphism for any λ-filtered inductive system F : I → E whose colimit exists where λ ≥ κ is regular, and such that F (α) ∈ S for any morphism α in I. (6) S-presented if it is (κ, S)-presented for some cardinal κ. (7) S-tiny if it is (0, S)-presented, where 0 is the first ordinal. (8) tiny if it is S-tiny for S = Mor(E). The terminology ‘tiny’, and both ‘elementary’ and ‘quasi-elementary’ below are following the corresponding definitions for quasi-abelian categories in [Sch99]. The definition of ‘small’ is following Definition 4.1 in [CH02], and the definition of ‘compact’ is [St’12ˇ ] Definition 3.3 (where it is called small). The notion of ‘presented’ is following the discussion in [Lur09b] Section A.1.1. Definition 2.6.97. Let E be an exact category, S a collection of morphisms in E, and ? ∈ {small, presented, compact tiny}. E is said to be (1) projectively generated if it has a projective generating set. (2) (κ, S)?-elementary if it is complete, cocomplete and has a projective generating set consisting of (κ, S) − ? objects. (3) S?-elementary if it is (κ, S)?-elementary for some κ. (4) S-elementary if it is complete, cocomplete, and has a projective generating set consisting of S-tiny objects. (5) quasi-elementary if it is complete, cocomplete and has a projective generating set consisting of S-tiny objects, where S is the class of split monomorphisms. (6) AdMon-elementary if it is elementary for the class of admissible monomorphisms. (7) elementary if it is S-elementary for S = Mor(E). Proposition 2.6.98. A cocomplete quasi-abelian category is (quasi)-elementary if and only if its left heart is (quasi)-elementary.

Proof. See [Sch99] Proposition 2.1.12.

Let I be a category, E an exact category and S a class of morphisms in E. Denote by FunS (I; E) the category of functors D : I → E such that D(i → j) is in S for any morphism i → j in I. Denote also cont cocont by FunS (I; E) and FunS (I; E) the full subcategories of FunS (I; E) consisting of functors which are cocont continuous, and cocontinuous respectively. Note that if I = ℵ0, then FunS (ℵ0; E) = FunS (ℵ0; E).

Definition 2.6.99. We say that E has (I; S)-(co)limits if for any functor D ∈ FunS (I; E), a (co)limit cont cont of D exists. We say that E has (I; S) -limits if for any functor D ∈ FunS (I; E), a limit of D exists. cocont cocont Finally we say that E has (I; S) -colimits if for any functor D ∈ FunS (I; E), a colimit of D exists. 2.6. GENERATORS IN EXACT CATEGORIES 33

Let Ch(S) denote the class of morphisms in Ch(E) consisting of those morphisms f• : A• → B• such cont cocont that fn ∈ S for each n. Clearly if E has (I; S)-limits, (I; S) -limits, (I; S)-colimits, or (I; S) -colimits then Ch(E) also has corresponding (co)limits for the class Ch(S). Definition 2.6.100. Suppose that E has (I; S)-colimits. We say that (I; S)-colimits are exact in E if

for any functor F ∈ FunS (I; Ch(E)) such that F (i) acyclic for any object i in I, the colimit lim→I F (i) is acyclic. Similarly one defines exactness of (I; S)cocont-colimits, (I; S)-limits, and (I; S)cont-limits. We will be particularly interested in the cases S = AdMon is the class of admissible monomorphism, or S = SplitMon is the class of split monomorphisms. The following proposition is immediate from Proposition 2.6.94 and Corollary 2.6.95 but it has a useful consequence. Proposition 2.6.101. Let E be a complete and cocomplete elementary exact category, I a filtered category, and 0 → F → G → H → 0 a null sequence of functors I → E such that for each i ∈ I, 0 → F (i) → G(i) → H(i) → 0 is exact. Suppose there is a class P of projective generators of E such that the maps lim Hom(P,F (i)) → Hom(P, lim F (i)) −→ −→ i∈I i∈I lim Hom(P,G(i)) → Hom(P, lim G(i)) −→ −→ i∈I i∈I lim Hom(P,H(i)) → Hom(P, lim H(i)) −→ −→ i∈I i∈I are isomorphisms for any P ∈ P. Then the sequence 0 → lim F (i) → lim G(i) → lim H(i) → 0 −→ −→ −→ i∈I i∈I i∈I is exact. In particular if, for example, E is a (κ, S)small-elementary, then for any regular λ ≥ κ,(λ, S)cocont- colimits in E exist and are exact. If E is elementary then all filtered colimits are exact. This also motivates a more general definition 2.6.102 below. Definition 2.6.102. Let E be an exact category and S a collection of morphisms in E. E is said to be (1) weakly (λ; S)-elementary for an ordinal λ if E has (λ; S)cocont-colimits and (λ; S)cocont-colimits are exact. (2) weakly S-elementary if for any ordinal λ E is weakly (λ; S)-elementary. (3) weakly AdMon-elementary if it is weakly S-elementary for S = AdMon of admissible monomorphisms. (4) weakly elementary if it is weakly S-elementary for S = Mor(E). In particular S-elementary exact categories are weakly S-elementary. Another useful fact is that if E has enough injectives then it is weakly (ℵ0; AdMon)-elementary. Proposition 2.6.103. If a weakly idempotent complete exact category E has enough injectives then it is weakly (ℵ0; AdMon)-elementary. Proof. In fact we shall prove the following. Let 0 → A → B → C → 0 be an exact sequence in Fun(ℵ0; E). Suppose that C is in FunAdMon(ℵ0; E). We claim that

0 → lim→n An → lim→n Bn → lim→n Cn → 0 is exact. Noting that Hom(−,I) sends colimits to limits, by the dual of Proposition 2.6.94 it suﬃces to prove that for each injective I, the sequence

0 → lim←n Hom(Cn,I) → lim←n Hom(Bn,I) → lim←n Hom(An,I) → 0 34 2. EXACT CATEGORY GENERALITIES

is exact. Since I is injective lim←n Hom(Cn,I) is the limit of a Mittag-Leffler system. Thus the sequence of abelian groups is exact ([Sta21] Section 10.86). Proposition 2.6.104. Let λ be an ordinal, and E be a weakly (λ0; AdMon)-elementary exact category for all λ0 ≤ λ. Then λ-transfinite compositions of admissible monics are admissible monics. Proof. The proof is by transfinite induction. Since finite compositions of admissible monics are admissible, the successor case is clear. For the limit case let Λ be a limit ordinal, and consider the commutative diagram

E0 / E0 / E0 / ...

cλ E0 / Eλ / Eλ0 / ...

0 / Coker(cλ) / Coker(cλ0 ) / ... with short exact columns. Taking the direct limit over Λ, we get a short exact sequence

0 → E0 → E → C → 0

In particular E0 → E is admissible. Note that Proposition 2.6.101 and Proposition 2.6.104 are proven for the G-exact structure (see the next chapter) in [Gil16a] Corollary 5.3. Before concluding this section, let us say something brief about projective limits.

op Definition 2.6.105. Let E be an exact category. A projective diagram A ∈ Fun(ℵ0; E ) in E is said to be lim←-acyclic if for any exact sequence 0 → A → B → C → 0 op in Fun(ℵ0; E ) the sequence

0 → lim←n An → lim←n Bn → lim←n Cn → 0 is exact in E. The following is dual to Proposition 2.6.103. Proposition 2.6.106. If E be a weakly idempotent complete exact category. If E has enough projectives op then any sequence A ∈ Fun(ℵ0; E ) such that An → An−1 is an admissible epimorphism is lim←-acyclic. 2.6.2. Generators in Categories of Chain Complexes. Our goal now is to show that if E is an elementary exact category then so is Ch∗(E), for ∗ ∈ {+, ≤ 0, −, b, ≥ 0, ∅}. Much of this is based on the following technical result. Lemma 2.6.107. Let E be a weakly idempotent complete exact category. For any object C ∈ E and X,Y ∈ Ch(E) we have natural isomorphisms ∼ n (1) HomE (C,Yn) = HomCh(E)(D (C),Y ) ∼ n (2) HomE (Xn−1,C) = HomCh(E)(X,D (C)) Y ∼ n Y (3) Ker(HomE (C, dn )) = HomCh(E)(S (C),Y ). In particular if Ker(dn ) exists Y ∼ n then HomE (C, Ker(dn )) = HomCh(E)(S (C),Y ) X ∼ n X (4) Ker(HomE (dn+1,C)) = HomCh(E)(X,S (C)) In particular if Coker(dn+1) exists then X ∼ n HomE (Coker(dn+1),C) = HomCh(E)(X,S (C)) 1 ∼ 1 n (5) ExtE (C,Yn) = ExtCh(E)(D (C),Y ) 1 ∼ 1 n+1 (6) ExtE (Xn,C) = ExtCh(E)(X,D (C)) 2.6. GENERATORS IN EXACT CATEGORIES 35

X (7) Let X be a complex such that Ker(dn ) exists. Then there is a monomorphism 1 X 1 n Ext (C, Ker(dn )) ,→ Ext (S (C),X) If X is acyclic then this is an isomorphism. X (8) Let X be a complex such that Coker(dn+1) exists. Then there is a monic 1 X 1 n Ext (Coker(dn+1),C) ,→ Ext (X,S (C)) If X is acyclic then this is an isomorphism.

Proof. By Proposition 2.2.45 and Corollary 2.2.52 it is suﬃcient to prove statements 1−3, 5, 6, 7 under the assumption that E is abelian. In this context the result is Lemma 3.1 in [Gil04] and Lemma 4.2 in [Gil08]. Statement 4 is dual to to 3, and statement 8 is dual to 7.

Remark 2.6.108. It is possible to prove most of this lemma internally in an exact category without passing to an abelianisation. At this point we can prove the following lemma. It provides one of our main applications of generating sets, namely a convenient method for testing acyclicity. It is a modiﬁcation of Lemma 3.7 in [Gil07].

Lemma 2.6.109. Let E be an exact category with a collection of generators G. Let X be a chain complex. n n+1 Suppose that X• is good. If for every G ∈ G each map f : S (G) → X extends to D (G), then X is acyclic.

Proof. By Proposition 2.2.28 it is enough to show that whenever dm has a kernel, the induced map 0 d : Xm+1 → ZmX is an admissible epic. For this it is enough to show that for each G ∈ G,

0 Hom(G, d ) : Hom(G, Xm+1) → Hom(G, ZmX)

is surjective, i.e. that any map f : G → ZmX lifts to a diagram

Xm+1 < d0 f G / ZnX But this is equivalent to showing that the chain map Sn(G) → X induced by f extends to a morphism Dn+1(G) → X. Since there is a bijective correspondence between diagrams of the form

Sn−1(G) / X

f Dn(G) / Y

and maps of the form Sn(G) → cone(f), which induces a bijection between lifts in the above diagram and extensions of the map Sn(G) → cone(f) to a map Dn+1(G) → cone(f), we immediately get the following.

Corollary 2.6.110. Let E be a weakly idempotent complete exact category with a collection of generators G. Let g : X → Y be a morphism of complexes. Then g is acyclic if and only if f has the right lifting property with respect to all maps of the form Sn(G) → Dn+1(G) for n ∈ Z,G ∈ G. 36 2. EXACT CATEGORY GENERALITIES

Next we characterise projective objects in categories of chain complexes. It is well known that projective objects in the category of chain complexes in an abelian category are precisely the split exact complexes with projective entries. See for example [Hov99] Proposition 2.3.10. We generalise the result to exact categories. The fact that in Ch(E) the projective complexes are contractible complexes with projective components, and that if E has enough projectives then so does Ch(E) is proven in [Gil16b] Proposition 2.6 and Corollary 2.7. Proposition 2.6.111. Let E be an exact category, and let ∗ ∈ {≥ 0, ≤ 0, +, −, b, ∅}. Then contractible 0 complexes of projectives are projective objects in Ch∗(E). In addition, if P is projective in E then S (P ) is projective in Ch≥0(E). Conversely, if a complex X• is a projective in Ch∗(E) for ∗ ∈ {+, −, b, ≥ 0, ≤ 0, ∅} then every Xn is projective. Moreover, if ∗ ∈ {+, −, b, ∅} then X• is contractible. In particular if E is weakly idempotent complete then X• is projective if and only if it is a split exact complex of projective objects of E. Proof. The claims for ∗ ∈ {+, −, b, ∅} are established in [Gil16b] Proposition 2.6 and Corollary 2.7. In fact the proof there is restricted to ∗ ∈ {∅}but the proof is the same for the other cases. Let us show that 0 S (P ) is a projective object in Ch≥0(E) whenever P is projective in E. Indeed in this case, Lemma 2.6.107 0 ∼ 0 implies that HomCh(E)(S (P ),Y•) = HomE (P,Y0). Since P is projective, S (P ) is as well.

We can now show that Ch∗(E) has enough projectives. (This is well known for Ch(A) with A abelian. See for example [Wei95] Exercise 2.2.2).

Corollary 2.6.112. Let E be an exact category with enough projectives. Then Ch∗(E) has enough projectives for ∗ ∈ {+, −, b, ≤ 0, ≥ 0, ∅}

Proof. The case for ∗ ∈ {+, −, b, ∅} is established in [Gil16b] Corollary 2.7. Now let X• ∈ Ch≥0(E). n 0 For n > 0 the object D (P ) is projective in Ch≥0(E). S (P ) is also projective in Ch≥0(E). For n > 0, as n before there is a projective object Pn and a morphism D (Pn) → X• which is an admissible epimorphism in degree n. For n = 0 pick a projective object P0 and an admissible epimorphism P0 → X0. Since 0 X−1 = 0, this induces a map S (P0) → X• which is an admissible epimorphism in degree 0. Let P• = L n 0 n>0 D (Pn) ⊕ S (P0). Then we have an admissible epimorphism P• X•.

In particular we have shown that Ch∗(E) has a set of projective generators whenever E does. Corollary 2.6.113. Suppose P is a collection of admissible generators for an exact category E. Then ∗ n D (P) = {D (P ): P ∈ P, n ∈ Z}∩Ch∗(E) is a collection of generators for Ch∗(E) and ∗ ∈ {+, −, b, ≤ 0, ∅}. ∗ . ∗ 0 For ∗ ∈ {≥ 0}, D‹ (P) .= D (P) ∪ {S (P ): P ∈ P} is a collection of generators for Ch∗(E). They are projective generating collections if P is. Proof. The proof of Corollary 2.6.112 shows that the collection in the statement of the proposition are admissible generating collection. Proposition 2.6.111 establishes the second assertion.

We are nearly ready to show that Ch∗(E) is elementary for ∗ ∈ {+, ≥ 0, ≤ 0, −, b, ∅}. It remains to identify some suitably compact objects in complexes. However by Lemma 2.6.107 we have the following. Proposition 2.6.114. Let E be an object satisfying one of the smallness conditions of Deﬁnition 2.6.96. Then Dn(E) and Sn(E) satisfy the same smallness condition in Ch(E). As a consequence we have Corollary 2.6.115. For ? ∈ {small, compact, presented, teeny} let E be a (κ, S)?-elementary exact ? category. Then Ch∗(E) is (κ, Ch(S)) -elementary for ∗ ∈ {+, ≤ 0, ≥ 0, −, b, ∅}. In particular if E is S- elementary then Ch(E) is Ch(S)-elementary

Proof. Let P be a projective generating set consisting of tiny objects. The sets D∗(P) (resp. D‹∗(P)) n are projective generating sets in Ch∗(E) for ∗ ∈ {≤ 0, +, −, b, ∅} (resp. ∗ ∈ {≥ 0}). For each n ∈ Z D (P ) n is tiny, as is S (P ), by Proposition 2.6.114. 2.6. GENERATORS IN EXACT CATEGORIES 37

2.6.3. Generators in Monoidal Exact Categories. Let us briefly mention a useful compatibility condition between generators and monoidal structure. Definition 2.6.116. A monoidal exact category which has a collection of flat admissible generators is said to be flatly generated. Definition 2.6.117. A projectively monoidal exact category which is also (λ; S)-elementary is said to be monoidal (λ; S)elementary

Proposition 2.6.118. Suppose that (E, ⊗, k) is a ﬂatly generated monoidal exact category in which direct sums are exact. Then every projective object is ﬂat.

Proof. In this case every projective will be a summand of a flat object, and therefore flat. In particular to check that a category is projectively monoidal, it suffices to find a collection of flat generators.

2.6.4. Generators and Adjunctions. We conclude this section with a note about passing generating collections through adjunctions. The speciﬁc application we have in mind is to categories of algebras over compatible monads. We have the following general setup.

F : E D :| − | is an adjunction where E and D are exact categories. We have the following result which is standard for abelian categories. Proposition 2.6.119. Let F a | − | be an adjunction as above. Suppose that | − | is an exact functor. If P is a projective object of E then F (P ) is a projective object of D. Proof. Let 0 → X → Y → Z → 0 be a short exact sequence in D, and let P be projective in E. Then we have a diagram 0 / Hom(F (P ),X) / Hom(F (P ),Y ) / Hom(F (P ),Z) / 0

0 / Hom(P, |X|) / Hom(P, |Y |) / Hom(P, |Z|) / 0 The vertical arrows are isomorphisms and the bottom row is exact since | − | is exact and P is projective. Hence the top row is short exact as well. We know how adjunctions act on projectives. Let us now see what happens on generating collections. Proposition 2.6.120. Let F a | − | be an adjunction as above. Suppose that | − | reﬂects admissible epimorphisms, and that E has an admissible generating collection G. Let F (G) denote the collection {F (G): G ∈ G} of objects of D. Then F (G) is an admissible generating collection in D. Proof. Let X be an object of D. Suppose there is some object Q of E and an admissible epimorphism p : Q → |X|. There is an induced morphism pe : F (Q) → X. Then p coincides with the composition Q → |F (Q)| → |X|. By Proposition 2.1.8, the map |pe| is an admissible epimorphism. Since | − | reﬂects admissible epimorphisms, pe is an admissible epimorphism in D. Now let G be an admissible generating collection in E, and let X be an object of D. Since G is an L admissible generating collection, there is an object G of G and an admissible epimorphism G |X|. The induced morphism F (G) → X is an admissible epimorphism by the above remarks. Since F is a left adjoint L it preserves colimits, so F (G) is an element of F (G). The following is clear. 38 2. EXACT CATEGORY GENERALITIES

Proposition 2.6.121. Let F : E D :| − | be an adjunction between any categories, κ an ordinal, and S a class of morphisms in D. Let |S| denote the class of morphisms in E which are images of morphisms in S under | − |. (1) Suppose that | − | commutes with (λ, S)cocont-colimits whenever λ ≥ κ is regular. If G ∈ E is (κ, S)-small then F (G) is (κ, |S|)-small. (2) Suppose that | − | commutes with (λ, S)cocont-colimits whenever λ ≥ κ. If G ∈ E is (κ, S)-compact then F (G) is (κ, |S|)-compact. (3) Suppose that | − | commutes with (I, S)cocont-colimits whenever I is a λ-filtered category for λ ≥ κ regular. If G ∈ E is (κ, |S|)-presented then F (G) is (κ, S)-presented. Using this proposition, Proposition 2.6.119 and Proposition 2.6.120, we then get the following. Proposition 2.6.122. Let F a |−| be an adjunction as above, κ an ordinal, and S a class of morphisms in D. Let |S| denote the class of morphisms in E which are images of morphisms in S under | − |. (1) Suppose that |−| is exact, reflects exactness and commutes with (κ, S)cocont-colimits. If E is weakly (κ, |S|)- elementary then D is weakly (κ, S)-elementary. (2) Suppose that | − | is exact and reflects admissible epimorphisms. If G is a projective generating collection in E then F (G) is a projective generating collection in D. (3) Suppose that |−| is exact, reflects admissible epimorphisms, and commutes with (λ, S)cocont-colimits for any ordinal λ. If E is |S|-elementary then D is S-elementary. Proof. The first claim is obvious. The second and third assertions follow the previous Proposition, Proposition 2.6.119, and Proposition 2.6.120. Note that by Proposition 2.6.121 the third claim in Proposition 2.6.122 can be generalised to statements about D being (κ, S)?-elementary for more general ?, by only requiring that | − | commutes with certain colimits. We have omitted this level of generality so as not to have to write out all the cases. However the proof is identical. Example 2.6.123. Let T be a compatible monad on an exact category E. Then the forgetful functor | − | : ET → E has a right adjoint F : E → ET assigning to an object the free T -algebra on it. By construction of the exact structure on ET in Proposition 2.4.73, the functor |−| is admissibly exact and reflects exactness. Moreover it creates limits and colimits. By Lemma 2.4.74, Proposition 2.6.122 is applicable in such categories. CHAPTER 3

Examples

In this chapter we give examples of interesting exact categories which satisfy very diﬀerent set-theoretic properties but which are all weakly AdMon-elementary. In the next chapter we shall see that E being weakly AdMon-elementary and having kernels is enough for the category Ch(E) to be equipped with the projective model structure. The moral of the story is that often diﬃcult to check set-theoretic assumptions can be ignored to some extent when discussing such model structures.

3.1. Categories of Topological Vector Spaces In this section we let k be a Banach ring, that is, a unital commutative ring k together with a map | − | : k → R>0 such that for all x, y ∈ k we have (1) |x| = 0 ⇔ x = 0 (2) |x + y| ≤ |x| + |y| (3) |xy| ≤ |x||y| (4) k is complete with respect to the topology deﬁned by | − |. k is said to be non-Archimedean if |x + y| ≤ max{|x|, |y|} and Archimedean otherwise. Over such rings we can consider categories of topological k-modules. For details of claims made in this section consult [Sch99], [BBB16], [BBK13], and [BBBK18].

3.1.1. Categories of Normed and Banach Modules.

Definition 3.1.1. A normed k-module is a k-module V together with a map || − || : V → R>0 such that for all λ ∈ k and for all x, y ∈ V we have (1) ||x|| = 0 ⇔ x = 0 (2) ||x + y|| ≤ ||x|| + ||y|| (3) ||λx|| ≤ |λ|||x|| If V is complete with respect to the metric deﬁned by || − || then V is said to be a Banach k-module.

If k is non-Archimedean then V is said to be non-Archimedean if ||x + y|| ≤ max{||x||, ||y||}. We denote by Normk the category whose objects are normed k-modules and whose morphisms are bounded k-linear maps. Bank is the full subcategory of Normk on Banach k-modules. For k non-Archimedean we nA nA also consider the full subcategories Normk and Bank of non-Archimedean normed and Banach spaces respectively. All of these categories are additive, ﬁnitely complete, and ﬁnitely cocomplete. The inclusions

nA nA Bank → Normk, Bank → Normk have left-adjoint functors given by completion. They are also symmetric monoidal. If E and F are objects in Normk then we deﬁne E ⊗π F to be their usual module tensor product endowed with the cross-norm

n n n X X o ||u|| = inf ||ei||||fi|| : u = ei ⊗ fi i=1 i=1

39 40 3. EXAMPLES

nA nA If E and F are objects in Normk we deﬁne E ⊗π F to be their usual module tensor product endowed with the norm. n n n X o ||x||π = inf max{||ai||||bi||}i=1 : x = ai ⊗ bi i=1 We refer to both of these constructions as the projective tensor product. If E and F are Banach spaces then E⊗ˆ πF is the completion of their projective tensor product as normed spaces. These constructions are functorial in each of the categories deﬁned above and form part of symmetric monoidal structures on them with unit the ground ring k. These monoidal structures are in fact closed. The module Homk(E,F ) of bounded maps between E and F can be given the structure of a normed space. The norm of T : E → F is

||T (e)||F ||T || = supe∈E\{0} ||e||E

This gives an internal Hom functor, which we denote by Hom. Thus (Bank, ⊗ˆ π, Hom) is a monoidal quasi- abelian category. Details can be found in [BBB16] Proposition 3.17 and Proposition 3.19. Finally, the projective objects l1(I) are flat by [BBB16] Lemma 3.26. By Proposition 2.6.118 this category is projectively monoidal. There are unfortunately some problems with this category. Although it is finitely complete and cocomplete it does not even have countable colimits in general. The larger category Tˆc of complete locally convex topological spaces is complete and cocomplete, but tragically it is not quasi-abelian ([Sch99] Proposition 3.1.14). Instead we pass to the formal completion Ind(Bank) of Bank by filtered colimits.

3.1.2. Ind and Pro Categories. Recall that if C is a U-small category for some universe U, and V is a universe, then the V-ind-completion of C is a category constructed as follows. Objects are diagrams E : I → C where I is a V-small ﬁltered category. If E : I → C and F : J → C are objects in Ind(C) (where we suppress universes in the notation) then we write

HomInd(C)(E,F ) = lim←I lim→J HomC (Ei,Fj) Details of this can be found in [KS05] Chapter 6. Proposition 3.1.2. Let E be a quasi-abelian category with enough projectives. Then Ind(E) is a cocomplete elementary quasi-abelian category. Moreover, if E is a closed monoidal exact category, then its ind-completion has a canonical exact closed monoidal structure extending the one on E. Finally if E is projectively monoidal then so is Ind(E).

Proof. See [Sch99] Proposition 2.1.16 and Proposition 2.1.19.

Corollary 3.1.3. The category Ind(Bank) is a locally presentable, closed monoidal elementary quasi- abelian category.

The category Ind(Bank) is not concrete. However it does have a natural concrete full subcategory m m Ind (Bank). An object of Ind (Bank) is a formal colimit “lim→”Ei such that any map Ei → Ej is a monomorphism (not necessarily admissible!). It is shown in [Mey07] Theorem 1.139 and Section 1.5.3 that this category is equivalent to the concrete category CBornk of complete bornological k-modules, via the m disection functor diss : CBornk → Ind (Bank). These are spaces equipped with an appropriate notion of ‘bounded subsets’. To a (complete) locally convex space E one can functorially assign both the von Neumann bornology vN(E) and the compact bornology Cpt(E). The von Neumann bornology is composed of the subsets of E absorbed by all zero neighbourhoods. The compact bornology is composed of subsets with compact closure. There is a natural transformation of functors Cpt → vN. For details see [Mey07] Section 1.1.4. If V is a nuclear locally convex space then the map Cpt(V ) → vN(V ) is an isomorphism by [BBBK18] Lemma 3.67. There is also the dual notion of the V-pro-completion of C, which is deﬁned to be P ro(C) = Ind(C op)op It is the formal completion of C by projective limits. 3.1. CATEGORIES OF TOPOLOGICAL VECTOR SPACES 41

For k a Banach ring P ro(Bank) contains Tˆc,k as a full subcategory. Indeed if E is an object of Tˆc,k ˆ ˆ defined by a family of seminorms P then define PB(E) = “lim←p∈P ”Ep where Ep is the completion of E with respect to the metric defined by the semi-norm p. This construction is functorial, lax monoidal, and PB : Tˆc,k → P ro(Bank) is fully faithful. If E is a quasi-abelian category enough projectives and injectives then by Proposition 2.1.15 in [Sch99] both Ind(E) and P ro(E) are complete and cocomplete. In particular by an obvious Kan extension there is a canonical functor PI : P ro(E) → Ind(E) Again this is lax monoidal. Returning to the case E = Ban, there is a natural isomorphism of functors PI ◦ PB =∼ diss ◦ vN (see [BBBK18]), and therefore a natural transformation diss ◦ Cpt → PI ◦ PB. Let E be a Banach space and F a complete locally convex space. Then ∼ HomInd(Ban)(E,PI ◦ PB(F )) = HomP ro(Ban)(E,PB(F )) =∼ Hom (E,F ) Tˆc,k ∼ = HomCBorn(E, vN(F )) ∼ = HomInd(Ban)(E, diss ◦ vN(F )) Where the isomorphisms just arises from the fact that Hom(E, −) commutes with projective limits in both Tˆc and CBorn When restricted to the category of nuclear Fréchet spaces the functor diss ◦ vN is fully faithful by Example 3.22 [BBBK18]. Moreover since map diss ◦ Cpt → PI ◦ PB is a natural isomorphism this restriction of PI ◦ PB is also strong monoidal [Mey07]) Theorem 1.87. In particular the category of nuclear Fréchet algebras over C embeds fully faithfully in the category of commutative complete bornological algebras. Since the category CBornC has good categorical properties, in particular it is closed monoidal AdMon-elementary, this is evidence that it provides a convenient setting in which to study analytic algebra. 3.1.3. The Non-Expanding Normed and Banach Categories. Each of the normed and Banach categories considered in the previous section has a corresponding ‘non-expanding’ subcategory. If E and ≤s F are normed spaces and s ∈ R≥0 then we denote by Homk (E,F ) ⊂ Homk(E,F ) the set of maps of k-modules of norm at most s. Composition gives a map ≤r ≤s ≤rs Homk (E,F ) ⊗ Homk (F,G) → Homk (F,G) nA nA In particular there are wide subcategories of Normk, Bank, Normk , Bank consisting of maps of norm ≤1 ≤1 nA,≤1 nA,≤1 at most 1 which we denote by Normk , Bank , Normk , Bank . They are equipped with closed symmetric monoidal structures by restricting the ones on the larger categories. If k is non-Archimedean then these categories are also additive and in fact quasi-abelian. In both the Archimedean and non-Archimedean case these categories are complete and co-complete. Details of this can be found in [BBK13] Appendix A. For convenience we recall how to construct arbitrary ≤1 nA,≤1 `≤1 coproducts in Bank and, for k non-Archimedean, Bank . For k Archimedean the coproduct i∈I Ai of ≤1 a collection {Ai}i∈I of Banach spaces in Bank is

kMod Y X {(ai)i∈I ⊂ Ai : ||ai|| < ∞} i∈I i∈I

P QkMod with the norm ||(ai)|| = i∈I ||ai||. Here denotes the product in the category of k-modules. For k ≤1,nA ≤1 `≤1 non-Archimedean the coproduct in both Bank and Bank the coproduct i∈I Ai of a collection {Ai}i∈I of Banach spaces is the subspace

kMod Y {(ai)i∈I ⊂ Ai : limi∈I ||vi|| = 0} i∈I endowed with the norm ||(ai)i∈I || = supi∈I ||ai||. 42 3. EXAMPLES

3.1.3.1. Rescaling Functors. For r ∈ R>0 we denote by (−)r : Normk → Normk the endofunctor which sends a normed space E to Er which has the same underlying k-vector space as E but with norm rescaled by r. On morphisms it does nothing. It is evidently an autoequivalence, and in fact an automorphism, with inverse given by (−) 1 . Moreover it restricts to an auto-equivalence on all the normed and Banach categories r deﬁned above. These functors satisfy the following useful property.

sr ≤r ≤ t Proposition 3.1.4. Let E and F be Banach k-modules. Then Homk (Es,Ft) = Homk (E,F )

Proof. Let f : Es → Ft have norm at most r, so that for any e ∈ E, 1 r sr ||f(e)|| = ||f(e)|| ≤ ||e|| = ||e|| F t Ft t Es t E sr Conversely suppose f : E → F has norm at most t . Then we get the same inequality as above. 3.1.3.2. The Quasi-Abelian Exact Structure. Although we will not go into the details here, it is not hard ≤1,nA ≤1,nA to see that the categories Normk and Bank are quasi-abelian. However let us note the following. ≤1,nA ≤1,nA Proposition 3.1.5. In both Normk and Bank we have the following. (1) A monomorphism f : A → B is admissible in the quasi-abelian exact structure if and only if it is an isometry with closed image. (2) An epimorphism g : B → C is admissible in the quasi-abelian exact structure if and only if it is a set-theoretic epimorphism and ||g(b)|| = infa∈Ker(g)||b − a||. Proof. (1) Suppose that f : A → B is admissible in the quasi-abelian exact structure. Then it is the kernel of its cokernel g : B → C. Therefore f induces an isometric isomorphism with the normed subspace K = {b ∈ B : g(b) = 0}. In particular f is an isometry. Conversely ∼ ≤1,nA suppose that f is an isometry with closed image. Then A = f(A) in Normk . The cokernel of f is isometrically isomorphic to the quotient space Bf(A), and the kernel of B → Bf(A) is {b ∈ B : g(b) = 0} = f(A) =∼ A. (2) Suppose that g : B → C is an admissible epimorphism in the quasi-abelian exact structure. Then it is the cokernel of its kernel, which is the subspace A = {b ∈ B : g(b) = 0}. In particular g induces an isometric isomorphism g : BA =∼ C. So

||g(b)|| = ||[b]|| = infa∈A||b − a|| Moreover B → BA is a set-theoretic epimorphism, so g is as well. Conversely suppose that g is a set-theoretic epimorphism, and that ||g(b)|| = infa∈Ker(g)||b − a||. Then g clearly induces an isometric isomorphism. ≤1,nA Remark 3.1.6. In the case of Bank we may remove the assumption in Proposition 3.1.5 1) that f has closed image, since an isometry of Banach spaces always has closed image. ≤1,nA 3.1.3.3. The Strong Exact Structure. We introduce a diﬀerent exact structure on Normk (resp. ≤1,nA Bank ). ≤1,nA Definition 3.1.7. (1) We say that a morphism f : A → B in Normk is a strong monomorphism if it is an isometry with closed image, and any b ∈ B has a closest point ab ∈ f(A). ≤1,nA (2) We say that a morphism g : B → C in Normk is a strong epimorphism if for any c ∈ C there is a bc ∈ B with g(bc) = c and ||bc|| = ||c||. ≤1,nA (3) We say that a morphism f : A → B in Bank is a strong monomorphism (resp. strong ≤1,nA epimorphism) if it is a strong monomorphism (resp. strong epimorphism) in Normk . Corollary 3.1.8. A strong monomorphism is an admissible monomorphism in the quasi-abelian exact structure. A strong epimorphism is an admissible epimorphism in the quasi-abelian exact structure. 3.1. CATEGORIES OF TOPOLOGICAL VECTOR SPACES 43

Proposition 3.1.9. A map f : A → B is a strong monomorphism if and only if it is the kernel of a strong epimorphism. A map g : B → C is a strong epimorphism if and only if it is the cokernel of a strong monomorphism. Proof. Suppose that f : A → B is a strong monomorphism. Then in particular it is an admissible monomorphism in the quasi-abelian exact structure so it is the kernel of its cokernel g : B → C. Let us show that g is a strong epimorphism. Let c = [b] ∈ C = B f(A). Now ||[b]|| = infa∈A||b − f(a)||. By assumption there is some ab such that ||[b]|| = ||b − f(ab)||. Moreover g(b − f(ab)) = [b]. So g is a strong epimorphism. Conversely suppose that f : A → B is the kernel of a strong epimorphism g : B → C. Let b ∈ B. There is b0 ∈ B such that g(b) = g(b0) and ||g(b)|| = ||b0||. Now b − b0 ∈ A. We claim that b − b0 is a closest point to b in A. Indeed for any a ∈ A ||b − (b − b0)|| = ||b0|| = ||g(b)|| = ||g(b − a)|| ≤ ||b − a|| So we get a class of kernel-cokernel pairs

f g 0 / A / B / C / 0 where f is a strong monomorphism and g is a strong epimorphism. We denote this class by strong. We are going to prove the following. Theorem 3.1.10. The collection strong of strong kernel-cokernel pairs is an exact structure on both ≤1,nA ≤1,nA Normk and Bank . We do this in several steps. It is clear that isomorphisms are strong epimorphisms and strong monomorphisms. It is also clear that the projection A⊕B → B is a strong epimorphism and the inclusion A → A⊕B is a strong monomorphism. Proposition 3.1.11. Let f A / B

g g0 f 0 X / Y ≤1,nA ≤1,nA 0 be a pushout diagram in Normk or Bank . If f is a strong monomorphism then so is f . ≤1,nA ≤1,nA Proof. We shall prove it for Normk . The case of Bank is similar. The space Y is isometrically isomorphic to the quotient normed space X ⊕ B{−g(a), f(a)} Let [(x, b)] be an element of X ⊕ B {−g(a), f(a)}, and let ab ∈ A be such that f(ab) is a closest point to 0 b in f(A). Consider the element [(x − g(ab), 0)] which is in the image of f . We claim that this is a closest point to [(x, b)] in X ⊕ B{−g(a), f(a)}. Let [(x0, 0)] be an element in the image of f 0. Then 0 0 ||[(x, b)] − [(x , 0)]|| = infa∈Amax{||x − x + g(a)||, ||b − f(a)||} 0 Now ||b − f(ab)|| ≤ ||b − f(a)|| and 0 ≤ ||x − x + g(a)||. 0 ||[(0, b − f(ab)]|| = ||[(x, b)] − [(x − g(ab), 0)]|| ≤ ||(0, b − f(ab))|| ≤ ||[(x, b)] − [(x , 0)]|| 0 0 for any x ∈ X. Therefore [(x − g(ab), 0)] is a closest point to [(x, b)] in the image of f . Proposition 3.1.12. Let f 0 A / B

g0 g f X / Y 44 3. EXAMPLES

≤1,nA ≤1,nA 0 be a pullback diagram in Normk or Bank . If f is a strong epimorphism then so is f . Proof. A is (isometrically isomorphic to) the subspace {(x, b): f(a) = g(b)} of X ⊕ B, with f 0 being (x, b) 7→ b. Let b ∈ B and let x ∈ X be such that f(x) = g(b) and ||g(b)|| = ||x||. Then (x, b) ∈ A and f 0(x, b) = b. Moreover ||(x, b)|| = max{||x||, ||b||}. If ||x|| ≤ ||b|| then we are done. Suppose ||x|| ≥ ||b||. Then ||b|| ≤ ||x|| = ||g(b)|| ≤ ||b|| so ||x|| = ||b||. In either case ||(x, b)|| = ||b||. It is clear that the composition of strong epimorphisms is a strong epimorphism. To conclude the proof of Theorem 3.1.10 let us next show that compositions of strong monomorphisms are strong. More generally we have the following. Proposition 3.1.13. Let (C, d) be an ultrametric space and A ⊂ B ⊂ C be subspaces. Let c ∈ C. Suppose that c has a nearest point bc in B and that bc has a nearest point ac ∈ A. Then c has a nearest point in A.

Proof. Let a ∈ A. If d(a, c) = d(bc, c) then a is a nearest point to C in B and hence therefore in A. Hence we may assume that d(a, c) > d(bc, c) for all a ∈ A. In particular d(bc, a) = d(a, c). So d(ac, c) = d(bc, ac) < d(bc, a) = d(a, c) for all a ∈ c and ac is a closest point to c in A. Corollary 3.1.14. The composition of strong monomorphisms f : A → B and g : B → C in ≤1,nA ≤1,nA Normk , and hence in Bank is a strong monomorphism. Now let us establish some properties of this exact structure. The following is clear. Proposition 3.1.15. Let f : A → B be a strong monomorphism. Then for [b] ∈ Bf(A) we have ||[b]|| = ||b − f(a)|| where f(a) is a closest point to b in f(A).

≤1,nA ≤1,nA Proposition 3.1.16. In both Normk and Bank products and coproducts preserve strong monomorphisms, strong epimorphisms and kernels. Coproducts preserve cokernels and products preserve cokernels of admissible monomorphisms. In particular they are exact for the strong exact structures. ≤1,nA Proof. Let us ﬁrst prove the claims about products. It suﬃces to show this for Normk . First note that products always commute with kernels. Now let

fi gi 0 / Ai / Bi / Ci / 0 be a strong exact sequence. We write the product sequence

f g 0 / A / B / C / 0 We need to show that this sequence is exact. Let us show that the map g is a strong epimorphism. Indeed by Proposition 3.1.15 ||([bi])|| = supi∈I ||bi − fi(ai)|| where ai is such that fi(ai) is a closest point to bi in fi(Ai). Now

||ai|| = ||fi(ai)|| = ||(fi(ai) − bi) + bi|| ≤ max{||fi(ai) − bi||, ||bi||} ≤ ||bi||

So (ai) ∈ A. Moreover ||(bi − fi(ai))|| = ||([bi])|| and π(bi − fi(ai)) = ([bi]). Now let us show that f is Q a strong monomorphism. It is clearly an isometry. Let c = (ci) ∈ Ci. For each i pick bi ∈ Bi with i∈I Q gi(bi) = ci and ||ci|| = ||bi||. Then clearly supi∈I ||bi|| = supi∈I ||ci||. Set b = (bi) ∈ i∈I Bi. Then g(b) = c n Q n and ||c|| = ||b||. A sequence (bi ) converges to (bi) in i∈I Bi if and only if bi converges to bi uniformly Bi. n In particular each bi converges to bi in Bi. It follows that the image of f is closed in B. Finally let (bi) ∈ B and for each i pick a closest point fi(ai) to bi in fi(Ai). Now

||([bi])|| = supi∈I infai∈Ai ||bi − fi(ai)||

Pick ai such that fi(ai) is a closest point to bi in fi(Ai). Then ||(bi)|| = supi∈I ||bi −fi(ai)||. By a computation similar to the previous part of the proof supi||ai|| ≤ supi||bi|| < ∞ and (ai) ∈ A. Moreover for any (eai) ∈ A we have ||(bi) − f((ai))|| = supi∈I ||bi − fi(ai)|| ≤ supi∈I ||bi − fi(eai)|| = ||(bi) − (ai)|| 3.1. CATEGORIES OF TOPOLOGICAL VECTOR SPACES 45

So f((ai)) is a closest point to (bi) in f(A).

Finally it is clear that f is a kernel of g and therefore the sequence is exact.

Coproducts always preserve cokernels. It is obvious that coproducts preserve strong epimorphisms in ≤1,nA ≤1,nA Normk and for Bank the proof is similar to the proof that products preserve strong epimorphisms. It is clear that coproducts preserve kernels. nA,≤1 nA,≤1 Corollary 3.1.17. In Normk and Bank products are admissibly coexact and coproducts are strongly exact for the strong exact structure. nA nA 3.1.3.4. Completion Functors. There is a completion functor Cpl : Normk → Bank which sends a ˆ nA nA normed space A to its separated completion A. It is left adjoint to the inclusion functor ι : Bank → Normk . ≤1 ≤1,nA ≤1,nA It restricts to a functor Cpl : Normk → Bank . Again it is left adjoint to the inclusion functor ≤1 ≤1,nA ≤1,nA ι : Bank → Normk . Proposition 3.1.18. The functor Cpl is exact for the quasi-abelian exact structure.

Proof. This is in [Pro00b] 3.1.13 for k = C, but the proof works for any Banach ring. We are going to show the following. Proposition 3.1.19. The functor Cpl≤1 is exact for the strong exact structure. First we need two basic facts about Cauchy sequences in non-Archimedean ﬁelds.

Proposition 3.1.20. Let (an) be a sequence in k such that ||an+1 − an|| → 0. Then (an) is a Cauchy sequence.

Proof. For any pair m > n we have ||am − an|| ≤ supn≤i≤m−1{||ai+1 − ai||}. Let δ > 0 and let N be such that ||aj+1 − aj|| < δ for j > N. Then for m > n > N we have ||am − an|| < δ. By Lemma 2.19 in [Bak11] we have

Proposition 3.1.21. Let (an) be a Cauchy sequence in k. If (an) does not converge to zero then the sequence (|an|) is eventually constant. Combining these two propositions we get the following. Proposition 3.1.22. Let f g 0 / A / B / C / 0 ≤1,nA be an strong exact sequence in Normk . Then

fˆ gˆ 0 / Aˆ / Bˆ / Cˆ / 0 ≤1,nA is a strong exact sequence in Bank . Proof. By Proposition 3.1.18 the complex is a kernel-cokernel pair. Thus it remains to show thatg ˆ is a strong epimorphism. Let [(cn)] be a non-zero equivalence class of Cauchy sequences in C. By Proposition 3.1.21 we may assume that ||cn|| is a constant r. Pick eb0 such that g(eb0) = c0 and ||b0|| = ||c0||. For each Pn n + 1 pick ebn+1 such that g(ebn+1) = cn+1 − cn and ||ebn+1|| = ||cn+1 − cn||. Write bn = k=0 ebn. Then g(bn) = cn. Moreover

r = ||cn|| ≤ ||bn|| ≤ maxk≤n||ebn|| = max{||c0||, max1≤k≤n||ck − ck−1||} ≤ r

Hence ||bn|| = r. Moreover ||bn+1 − bn|| = ||ebn+1|| = ||cn+1 − cn|| → 0, so by Proposition 3.1.20, (bn) is a Cauchy sequence. 46 3. EXAMPLES

≤1,nA ≤1,nA Proposition 3.1.23. For each r ∈ R>0 the object kr is projective in both Normk and Bank . In ≤1,nA ≤1,nA particular the strong exact structures on both Normk and Bank have enough functorial projectives. ≤1,nA Proof. Let us first prove the proposition for Normk . It suffices to show that the functor Hom(kr, −): ≤1,nA Normk → Ab preserves cokernels. Let f : A → B be a strong monomorphism with cokernel g : B → C. We need to show that the map 1 1 B 0, → B 0, B r C r on open balls is an epimorphism. This follows immediately from the definition of strong epimorphism. ≤1,nA . L For the second assertion, let E ∈ Normk . Write P(E) = e∈E k||e||. There is a map P(E) → E induced by the isometry

k||e|| → E, λ 7→ λe This is clearly a strong epimorphism. ≤1,nA ≤1 For Bank we use the fact that Cpl is exact and preserves projectives since it is left adjoint to an exact functor. ≤1,nA 3.1.3.5. Smallness Conditions. Let D : I → Normk be a diagram with I a directed category. Write . Ai = D(i) and fji = D(i ≤ j). Suppose that the fji are isometries. The direct limit A .= lim→Ai is constructed as follows. The underlying vector space lim→Ai is the direct limit of the underlying vector spaces of the Ai. Namely it is the disjoint union of the Ai quotient by the relation ai ∼ fji(ai) for any j > i. . If ai ∈ Ai and aj ∈ Aj then [ai] + [aj] .= [fKi(ai) + fKj(aj)] where K is any upper bound of i and j. If . . λ ∈ k then λ[ai] .= [λai]. We deﬁne a norm on this vector space by ||[ai]|| .= ||ai||. This is well-deﬁned because if ai ∼ aj then aj = fji(ai), so ||aj|| = ||ai||. Clearly ||[ai]|| = 0 if and only if ai = 0 and if λ ∈ k then ||λ[ai]|| = |λ|||ai||. Finally

||[ai] + [aj]|| = ||[fKi(ai) + fKj(aj)]||

= ||fKi(ai) + fKj(aj)||

≤ max{||fKi(ai)||, ||fKj(aj)||}

= max{||ai||, ||aj||}

= max{||[ai]||, ||[aj]||}

So this is a non-Archimedean norm. The map fi : Ai → A sends ai to [ai].

≤1,nA Proposition 3.1.24. The normed space described above is the direct limit in Normk .

Proof. Let gi : Ai → C be a cocone from D. There is a unique map of vector spaces g : A → C such that g ◦ fi = gi. It remains to show that g is bounded with ||g|| ≤ 1. Let [ai] ∈ A with [ai] = fi(ai). Then ||g([ai])|| = ||gi(ai)|| ≤ ||ai|| = ||[ai]||.

Corollary 3.1.25. (1) Suppose that for each j < k, fkj : Aj → Ak is an admissible monomor- ≤1,nA phism in the quasi-abelian exact structure on Normk . Then for each i the map Ai → A is an admissible monomorphism in the quasi-abelian exact structure. (2) Suppose that for each j < k, fkj : Aj → Ak is an admissible monomorphism in the strong exact ≤1,nA structure on Normk . Then for each i the map Ai → A is an admissible monomorphism in the strong exact structure.

Proof. (1) It is clear from the deﬁnition of the norm on A that fi is an isometry. Suppose that n n (fi(ai )) converges to [aj] with aj ∈ Aj. Let K be an upper bound of i and j. Then ([fKi(ai )]) n converges to [fKj(aj)]. But by the deﬁnition of the norm on A this clearly means that fKi(ai ) converges to fKj(aj) in AK . Since fKi has closed image, fKj(aj) = fKi(ai). Since fKi is an n n isometry (ai ) converges to ai, so (fi(ai )) converges to fi(ai), and fi has closed image. 3.1. CATEGORIES OF TOPOLOGICAL VECTOR SPACES 47

(2) Let [aj] ∈ A with aj ∈ Aj. Let K be an upper bound of i and j. The map fKi : Ai → AK is a strong monomorphism. Therefore fKj(aj) has a closest point fKi(ai) in fKi(Ai). We claim that 0 0 [ai] is a closest point to [aj] in fi(Ai). Indeed let [ai] ∈ fi(Ai) with ai ∈ Ai. Then 0 0 0 ||[aj] − [ai]|| = ||[fKj(aj) − fKi(ai)]|| = ||fKj(aj) − fKi(ai)||

≥ ||fKj(aj) − fKi(ai)|| = ||[aj] − [ai]|| 3.1.26 Let I be a filtered category. The functors lim : (I, Norm≤1,nA) → Corollary . −→ FunAdMon k Norm≤1,nA and lim : (I, Ban≤1,nA) → Ban≤1,nA are exact. k −→ FunAdMon k k ≤1,nA Proof. By Proposition 3.1.22 it suffices to prove the claim for Normk . Let i p 0 / X / Y / Z / 0 ≤1,nA be a short exact sequence in Fun(I, Normk ) where I is filtered. It suffices to show that the map lim→Y → lim→Z is a strong epimorphism. Fix a class [zi] ∈ lim→Z where zi ∈ Zi. There is some yi ∈ Yi with pi(yi) = zi and ||zi|| = ||yi||. But then p([yi]) = [zi] and ||[zi]|| = ||zi|| = ||yi|| = ||[yi]||. This immediately gives the following. ≤1,nA ≤1,nA Corollary 3.1.27. The strong exact structures on both Normk and Bank are weakly AdMon- elementary.

Proposition 3.1.28. For r ∈ R>0 the objects kr are tiny with respect to the class of admissible monomor- ≤1,nA phisms in the strong exact structure on Normk .

Proof. We need to show that for any r ∈ R>0 the map

lim→BAi (0, r) → Blim→Ai (0, r)

is an isomorphism. It suﬃces to prove that it is an epimorphism. Let [ai] ∈ lim→Ai be such that ||[ai]|| =

||ai|| ≤ r. Then ai ∈ BAi (0, r) ,→ lim→BAi (0, r) maps to [ai].

Proposition 3.1.29. For any r ∈ R>0 the k-Banach space kr is not tiny with respect to the class of admissible monomorphisms in the strong exact structure (or even the split exact structure). However every object is ℵ1-presented. i Proof. Consider the sequence with Xi = kr and the map Xi Xi+1 being the inclusion of the ﬁrst i copies of k. The group lim→Hom(kr,Xi) is the ascending union of the closed balls in Xi of radius r, while Hom(kr, lim→Xi) is the closed ball of radius r in lim→Xi. The map

lim→Hom(kr,Xi) → Hom(kr, lim→Xi) ⊕i is the obvious inclusion. Consider the example with Xi = k with Xi → Xi+1 being the split injection i i+1 k → k which is the inclusion of the first i copies of k. Then lim→Xi is the space of sequences in k converging to 0 with the supremum norm. The group lim→Hom(kr,Xi) is the group of finite sequences of 1 1 norm at most r , while Hom(kr, lim→Xi) is the group of sequences converging to 0 with norm at most r . It is clear that for a non-discrete field the map

lim→Hom(kr,Xi) → Hom(kr, lim→Xi) is not an epimorphism. The last claim is [AR94] 1.48. Recall that a Banach space E is said to have the Hahn-Banach extension property if for every subspace D of E, every bounded functional f : D → k there is an extension g : E → k of f with ||g|| = ||f||. Theorem 3.1.30 ([Pro00a] Theorem 4.12). If k is spherically complete then every Banach space over k has the Hahn-Banach extension property. 48 3. EXAMPLES

Proposition 3.1.31. Let E be a non-zero Banach space with the Hahn-Banach extension property and 0 ∼ 0 let e ∈ E. Then there is a Banach space E and an isometric isomorphism E = E ⊕ k||e||. In particular if ≤1,nA k is spherically complete then there are no non-nonzero tiny objects in Bank . D E D E Proof. Let e be the span of e in E. The map f : e → k||e|| sending e to 1 is an isometric

isomorphism with inverse g sending 1 to e. Therefore f extends to a map f : E → k||e|| with ||f|| = 1. Moreover f ◦ g = Id . Since Ban≤1 is quasi-abelian and in particular weakly-idempotent complete this k||e|| k||e|| gives a splitting. 3.1.3.6. The Monoidal Structure. The following is straightforward using that both functors are left adjoints. Proposition 3.1.32. Consider the functors nA nA nA Cpl ◦ ⊗π : Normk ⊗ Normk → Bank and nA nA nA ⊗ˆ π ◦ Cpl × Cpl : Normk ⊗ Normk → Bank There is a natural isometric isomorphism

φ : Cpl ◦ ⊗π → ⊗ˆ π ◦ Cpl × Cpl In particular we get a natural isomorphism ≤1 ≤1 ≤1 ≤1 φ : Cpl ◦ ⊗π → ⊗ˆ π ◦ Cpl × Cpl

Proposition 3.1.33. Let r ∈ R>0. Consider the functors nA nA (−)r ◦ Cpl : Normk → Bank and nA nA Cpl ◦ (−)r : Normk → Bank Then there is a natural isometric isomorphism

ζ :(−)r ◦ Cpl → Cpl ◦ (−)r In particular this induces a natural isomorphism of functors ≤1 ≤1 ∼ ≤1 ζ :(−)r ◦ Cpl = Cpl ◦ (−)r

Proposition 3.1.34. Let s, r ∈ R>0 and consider the functors nA nA nA (−)rs ◦ ⊗π : Normk × Normk → Normk and nA nA nA ⊗π ◦ (−)r × (−)s : Normk × Normk → Normk Then there is an natural isometric isomorphism

η : ⊗π ◦ (−)r × (−)s → (−)rs ◦ ⊗π At this point let us make the following remark Remark 3.1.35. The rescaling functors are exact for both the quasi-abelian and strong exact structures. By Proposition 3.1.33 and Proposition 3.1.32 we get

Corollary 3.1.36. Let s, r ∈ R>0 and consider the functors nA nA nA (−)rs ◦ ⊗π : Bank × Bank → Bank and nA nA nA ⊗π ◦ (−)r × (−)s : Bank × Bank → Bank Then there is an natural isometric isomorphism

η : ⊗π ◦ (−)r × (−)s → (−)rs ◦ ⊗π 3.2. THE EXACT CATEGORY WITH RESPECT TO A GENERATOR 49

≤1,nA ≤1,nA Corollary 3.1.37. Projective objects in Normk an Bank are ﬂat in both the quasi-abelian and strong exact structures. Proof. By Proposition 3.1.34 and Corollary 3.1.36, we only need to note that tensoring with k is the identity functor and hence is exact. ≤1,nA ≤1,nA Corollary 3.1.38. The tensor product of projective objects in Normk an Bank is a projective object. ≤1,nA ∼ Proof. It suﬃces to prove this in Normk for objects of the form kr with r ∈ R>0. But kr ⊗ks = krs which is projective. We summarise this section with the following result. ≤1 Theorem 3.1.39. Bank is a projectively monoidal weakly AdMon-elementary exact category which is ≤1 ℵ1-presentable but not ℵ0-presentable. Normk is a monoidal AdMon-elementary exact category. 3.2. The Exact Category with Respect to a Generator

3.2.1. The Gabriel-Popescu Theorem. Let G be a Grothendieck abelian category with a set {Gi}i∈I of generators. By Proposition 3.3 in [Gil16a] there is the structure of an exact category on G, called the G-exact structure, whereby a null sequence 0 → X → Y → Z → 0 is exact if 0 → Hom(Gi,X) → Hom(Gi,Y ) → Hom(Gi,Z) → 0 is exact for all i ∈ I. By Theorem A in [Gil16a] projective resolutions always exist in the G-exact structure, and as a consequence the projective model structure exists on Ch(G), where G is equipped with the G-exact structure. Gillespie further studies the case in which Gi are tiny. In this case the G-exact structure has exact coproducts. Moreover since the G-exact category has enough projectives, it has exact products, and Gillespie prove the injective model also exists Let us prove a generalisation of this for what we shall call strongly efficient exact categories. The terminology here is following the notion of an efficient exact category in [St’12ˇ ], which we will discuss in the next section. An exact category E is said to be strongly efficient if Definition 3.2.40. (1) It is locally presentable. (2) It has an admissible generator G. (3) It has kernels. (4) Kernels commute with filtered colimits. The proof of the following is a minor generalisation of [Sta21] Lemma 19.14.1. Proposition 3.2.41. Let G be an admissible generator of a weakly idempotent complete exact category E, and let X be an object of E. The natural map M π : G → X α∈Hom(G,X) is an admissible epimorphism. L Proof. There is some set I and an admissible epimorphism γ : i∈I G → X. For each i let γi be the restriction of γ to the copy of G in the coproduct indexed by i. We may regard L G as a summand of L L i∈I L α∈Hom(G,X) G where the copy of G in i∈I G indexed by i corresponds to the copy of G in α∈Hom(G,X) G indexed by γi. Moreover the composition M M G → G → X i∈I α∈Hom(G,X) is γ. This is an admissible epimorphism, so by the obscure axiom π is an admissible epimorphism. 50 3. EXAMPLES

There is also an easy adaptation of [Sta21] Lemma 19.14.1. Lemma 3.2.42. Let E be a strongly efficient exact category, and let G be an admissible generator. Then the functor R : Hom(G, −): E → End(G)Mod has a left adjoint such that the counit LR(G) → G is an isomorphism. Proof. The fact that the functor has a left adjoint L follows immediately from the adjoint functor theorem. However it will be helpful to give an explicit construction. Fix a functorial projective resolution functor Q : Mod End(G,G) → Ch≥0(Mod End(G,G)) such that for each module M and each n ≥ 0, Q(M)n . is a free End(G, G)-module. For M an End(G, G)-module define L(M) .= coker(Q(M)1 → Q(M)0). Let L L L L Q(M)0 = i∈I End(G, G) and Q(M)1 = j∈J End(G, G). This determines a map i∈I G → j∈J G. Define L(M) to be the cokernel of this map. For X ∈ E we have an exact sequence Y Y 0 → Hom(L(M),X) → R(X) → R(X) i∈I j∈J which is isomorphic to the exact sequence M M 0 → K → Hom( End(G, G),R(X)) → Hom( End(G, G),R(X)) i∈I j∈J But K is isomorphic to Hom(M,R(X)) due to the exact sequence M M End(G, G) → End(G, G) → M → 0 j∈J i∈I Hence L is left adjoint to R. Moreover this construction in fact shows that the counit LR(G) → G is an isomorphism. Using the previous two results, the following Lemma can be proven exactly as in [Sta21] Lemma 19.14.2.

0 Lemma 3.2.43. Let f : M → R(X) be injective in Mod End(G,G). Then the adjoint map f : L(M) → X is injective. This allows us to prove a version of the Gabriel-Popescu Theorem, mimicking the proof of [Sta21] Theorem 19.14.3. Theorem 3.2.44 (Gabriel-Popescu). Let E be a strongly efficient exact category, and let G be an admissible generator. Then the functor R : Hom(G, −): E → End(G)Mod is fully faithful with a left adjoint. Proof. We need to show that the counit v : L◦R(X) → X is an ismorphism. It is adjoint to the identity map R(X) → R(X) so it is an monomorphism. It remains to show that it is an admissible epimorphism L for any X. Let π : α∈Hom(G,X) G → X be the canonical map, which is an admissible epimorphism. Each α ∈ Hom(G, X) is an element of R(X) and therefore determines a map α0 : G =∼ LR(G) → LR(X) such that 0 0 L 0 v ◦ α = α. This gives a map π : α∈Hom(G,X) G → LR(X) such that v ◦ π = π. Since π is an admissible epimorphism, v must be as well. Using this Gabriel-Popescu theorem, the proofs of Proposition 3.3 and Corollary 3.4 in [Gil16a] also work for this more general setting. Corollary 3.2.45. Let E be a strongly efficient exact category with generator G. Then the collection of G-exact sequences defines an exact structure on E. Moreover G is a projective generator for the G-exact structure on E. nA,≤1 nA,≤1 Example 3.2.46. This is almost how the strong exact structures on Normk and Bank arise. nA,≤1 Indeed the category sNormk of semi-normed non-Archimedean k-modules with non-expanding bounded morphisms is strongly efficient as an exact category with the quasi-abelian exact structure. The strong exact L≤1 nA,≤1 nA,≤1 structure is the one determined by the generator r>0 kr. Both Normk and Bank are extension- nA,≤1 nA,≤1 closed subcategories of sNormk , and therefore inherit the strong exact structure from sNormk . 3.3. EXACT CATEGORIES OF GROTHENDIECK TYPE 51

3.3. Exact Categories of Grothendieck Type In [St’12ˇ ] St’ov´ıˇcekintroducesˇ efficient exact categories, and exact categories of Grothendieck type. As suggested by the name, the latter are supposed to generalise Grothendieck abelian categories. In particular, St’ov´ıˇcekshowsˇ that such categories have enough injectives, and in fact that the injective model structure (explained in the next chapter) exists on unbounded complexes in such categories under a mild extra assumption. Let us recall the definitions. Definition 3.3.47 ( [St’12ˇ ], Definition 3.11). A exact category E is said to be efficient if (1) E is weakly idempotent complete. (2) Transfinite compositions of admissible monics exist and are admissible monics. (3) Every object of E is compact relative to the class of all admissible monics. (4) E has an admissible generator. Definition 3.3.48 ([St’12ˇ ] 3.2). Let E be an exact category. E is said to be deconstructible in itself if there is a set S of objects of E, such that every object X of E has a presentation ∼ X = colimα<λSα

where λ is a regular ordinal, Sα ∈ S, and for α + 1 < λ, Sα → Sα+1 is an admissible monomorphism. Definition 3.3.49 ([St’12ˇ ] Definition 3.11). An exact category E is said to be of Grothendieck type if it is efficient and deconstructible in itself. Lemma 3.3.50 ([St’12ˇ ] Corollary 5.9.). Let E be an exact category of Grothendieck type. Then E has enough functorial injectives. 3.3.1. The λ-Pure Exact Structure. Following an argument of [EGO17] in the case of the ⊗-pure exact structure, we show that exact structures on locally presentable additive categories in which filtered colimits of split exact sequences are exact are of Grothendieck type. This is closely related to work of [Kra12] where the additive category is assumed to be locally finitely presentable, and of [Gil16a], for the the case E is a Grothendieck abelian category. Let E be a locally λ-presentable additive category for some regular cardinal λ. Following [AR94], say that a null sequence 0 → X → Y → Z → 0 in E is λ-pure exact if for any λ-presentable object H, the sequence of abelian groups 0 → Hom(H,X) → Hom(H,Y ) → Hom(H,Y ) → 0 is exact. Lemma 3.3.51 ([AR94], 2.30/ [Gil16a] Proposition A.1). A map f : X → Y in E is a λ-pure monomorphism if and only if it is a λ-directed colimit of split monomorphisms. Proposition 3.3.52 ([EGO17], Proposition 2.9). Let P : I → E be a λ-directed system. Then the map L i∈I Pi → colimI Pi is a λ-pure epimorphism. Proof. Proposition 2.9 in [EGO17] is stated and proven in the context that E is a Grothendieck abelian category, however the proof also works for any additive category which is cocomplete and finitely L complete. Let us give a short independent proof. The direct sum i∈I Pi is also a λ-directed system, L ∼ L ∼ and it follows that for H λ-presentable, Hom(H,Pi) = Hom(H, Pi), and colimI Hom(H,Pi) = L i∈I i∈I Hom(H, colimI Pi). The map Hom(H,Pi) → colimI Hom(H,Pi) is an epimorphism of abelian groups, L i∈I so i∈I Pi → colimI Pi is a λ-pure epimorphism. The following is proven in [Kra12] when the additive category is assumed to be locally finitely presentable, and of [Gil16a], for the the case E is a Grothendieck abelian category. Proposition 3.3.53. The collection of all λ-pure exact sequences defines an exact structure on E with enough projectives. 52 3. EXAMPLES

Proof. Clearly split exact sequences are λ-pure exact, and λ-pure epimorphisms are stable by composition and pullback. Let f : X → Y be a λ-pure monomorphism. Then f is λ-directed colimit of split monomorphisms. Since pushouts commute with colimits, and pushouts of split monomorphisms are split monomorphisms, any pushout of f is a λ-directed colimit of split monomorphisms, and therefore a λ-pure monomorphism. Moreover, a composition of λ-directed colimits of pure monomorphisms, is again a λ-directed colimit of split monomorphisms. To show that the λ-pure exact structure has enough projectives, ∼ let X be an object of E. We may write X as a λ-directed colimit X = colimI Hi where Hi is λ-presentable. L L But then by Proposition 3.3.52 the map i∈I Hi → X is a λ-pure epimorphism. Moreover i∈I Hi is projective. Lemma 3.3.54. Let E be a locally λ-presentable additive category. Suppose that (E, Q) is an exact category and that for any ﬁltered category I, the functor lim→ : Fun(I, E) is exact (in particular if E is elementary). Then (E, Q) is an exact category of Grothendieck type.

Proof. Let (E, Qλ) denote the λ-pure exact structure on E. Since E is weakly elementary, and split monomorphisms are exact in any exact category, the identity functor (E, Qλ) → (E, Q) is exact. Thus a generator for (E, Qλ) is still a generator for (E, Q). Similarly, the fact that (E, Qλ) is deconstructible in intself implies that (E, Q) is. The only missing axiom is that transﬁnite compositions of admissible monomorphisms are admissible monomorphisms, but we have in fact assumed something stronger. As we will explain later, the next corollary is essentially a generalisation of [Gil16a] Corollary 5.12 (and the part of Proposition 5.6 which says that the G-exact structure has enough injectives). Corollary 3.3.55. Let E be a strongly eﬃcient exact category, with E locally λ-presentable. Suppose that E has a set of generators {Gi}i∈I with each Gi tiny. Then the G-exact structure on E is of Grothendieck L type, where G = i∈I Gi.

Proof. Since the Gi are tiny, they are in particular λ-presentable. It follows that λ-pure exact sequences are exact in the G-exact structure. Also since the Gi are tiny, the G-exact structure is elementary. This next result then generalises [EGO17] Lemma 3.6. Corollary 3.3.56. Let E be a locally λ-presentable additive category equipped with a closed monoidal structure (⊗, Hom). Suppose that (E, Q) is an exact category and that for any ﬁltered category I, the functor lim→ : Fun(I, E) is exact. Then the ⊗-pure exact structure on (E, Q⊗) is of Grothendieck type. Proof. The functor X ⊗ (−) commutes with colimits, and sends split exact sequences to split exact sequences. Thus it sends λ-pure exact sequences to λ-pure exact sequences. 3.4. The Split Exact Structure We conclude with an example of a category which has no small generating set whatsoever but is still weakly AdMon-elementary. Let E be an additive category and endow it with the split exact structure. Let us prove the following useful lemma. Its proof, as well as the proof of Corollary 4.1.1 later, is a minor generalisation of [St’12ˇ ] Proposition 5.7 (the proof is essentially the same, and is also similar to the corresponding statement for abelian categories - Lemma 6.2 in [Hov02]). Lemma 3.4.57. Let E be an exact category. Let λ be an ordinal and f p 0 / Y / Z / X / 0 cocont be a short exact sequence in F un(λ, E) with X in F unadm (λ, E). Suppose that 1 Ext (Coker(Xα → Xα+1),Yα+1) = 0 for any α ≤ λ such that α + 1 ≤ λ. Then

0 → lim→Y → lim→Z → lim→X → 0 is a split exact sequence. 3.4. THE SPLIT EXACT STRUCTURE 53

Proof. For α ≤ β in λ, denote the corresponding maps in the diagrams by xα,β : Xα → Xβ, zα,β : Zα → Zβ, and yα,β : Yα → Yβ. The proof is by transﬁnite induction. If λ = 0 or λ is a successor ordinal

then the claim is clear. Now let λ = lim→α<λ α be a limit ordinal. Suppose the claim has been proven for 1 all α < λ. Let α < λ. Since Ext (Xα,Yα) = 0 there is some splitting tα : Xα → Zα of pα. We are going to modify the tα to sα so that they are compatible, i.e. zα,γ sα = sγ xα,γ for all α ≤ γ. We will do this by transﬁnite induction. Set s0 = t0. If γ is a limit ordinal let sγ : colimα<γ Xα → Zγ be the map whose restriction to Zα is zα,γ sα, where zα,γ : Zα → Zγ is the transﬁnite composition of the continuous functor γ → E, β 7→ Zβ. Then by construction zα,γ sα = sγ xα,γ . Now for the successor case γ = α + 1. Suppose we have constructed sα. Let us construct sα+1. We have

pα+1(zα,α+1sα − tα+1xα,α+1) = xα,α+1 ◦ pα ◦ sα − pα+1 ◦ tα+1 ◦ xα,α+1

= xα,α+1 − xα,α+1 = 0

Therefore there is a map h : Xα → Yα+1 such that fα+1h = zα,α+1sα − tα+1xα,α+1. Since xα,α+1 : Xα → 1 Xα+1 is an admissible monic and Ext (Coker(xα,α+1),Yα+1) = 0, the long exact Ext sequence implies that there is a map g : Xα+1 → Yα+1 such that gxα,α+1 = h. Let sα+1 = tα+1 + fα+1g. Then clearly sα+1 is a section of pα+1. Moreover

sα+1xα,α+1 = tα+1 ◦ xα,α+1 + fα+1g ◦ xα,α+1

= zα,α+1sα − fα+1 ◦ h + fα+1 ◦ h

= zα,α+1sα as required. Proposition 3.4.58. If E is an additive category with kernels and countable coproducts then the split exact structure is weakly (ℵ0; AdMon)-elementary. Proof. It has kernels by assumption. It trivially has enough projectives since every object is projective. The fact (ℵ0; AdMon)-colimits exist and are exact follows from Lemma 3.4.57. For E = Ab this category has no small generating set by [CH02] Section 5.4. This can be generalised to other exact structures deﬁned by projective classes as discussed in the same paper. CHAPTER 4

Model Structures on Exact Categories

In this chapter we discuss model structures on categories of chain complexes in exact categories. We give very general conditions under which unbounded complexes are equipped with the projective model structure. We also investigate when such a model structure is monoidal and satisﬁes the monoid axiom, which will be crucial for studying homotopical algebra in exact categories in the next section. Finally we generalise the Dold-Kan correspondence.

4.1. Cotorsion Pairs In [Hov02], Hovey introduced the notion of a compatible model structure on an abelian category. He showed that there is a 1-1 correspondence between such model structures and purely homological data now known as Hovey triples. Gillespie noticed that this correspondence generalises to weakly idempotent complete exact categories, and explains in [Gil11] how to adapt Hovey’s proofs. Moreover in [St’12ˇ ], St’ov´ıˇcekexplainedˇ the relationship between cotorsion pairs and compatible weak factorisation systems. In the next two subsections we will recall some of Hovey’s/ Gillespie’s/St’ov´ıˇcek’sresultsˇ both for the reader’s convenience and because we will need many of the individual propositions later anyway. For basic facts about weak factorisation systems and model structures in general see Appendix A. Let S be a class of objects in an exact category E. We shall denote by ⊥S the class of all objects X such that Ext1(X,S) = 0 for all S ∈ S, and by S⊥ the class of all objects X such that Ext1(S, X) = 0 for all S ∈ S. The class S⊥ is called the class of S-injectives, and the class ⊥S is called the class of S-projectives. For a class of objects P in an exact category E, let AdMonP denote the class of admissible monomorphisms whose cokernels is in P.

Corollary 4.1.1. Let E be an exact category. Let S be a class of objects in E, and let L = ⊥S. Then L is closed under retracts and ﬁnite extensions. If E is cocomplete and weakly (λ, AdMonL)-elementary then L is closed under λ-transﬁnite extensions.

Proof. First we show that L is closed under retracts. Note that it is sufficient to show that for a given Y ∈ E, the collection of objects X such that Ext1(X,Y ) = 0 is closed under retracts. Let X be such that Ext1(X,Y ) = 0 and let X0 be a retract of X. Then X0 is a summand of X, and so Ext1(X0,Y ) = 0. Let us show that L is closed under transfinite extensions. Again it is sufficient to show that for any object Y ∈ E the collection of all X with Ext1(X,Y ) = 0 is closed under transfinite extensions and retracts. cocont Let λ be an ordinal X : λ → E an object of F unadm (λ, E). Let

f p 0 / Y / N / lim→X / 0

54 4.1. COTORSION PAIRS 55

1 represent an element of Ext (lim→X,Y ). For each β ∈ λ, pull this short exact sequence back through the map xβ : Xβ → lim→X. For α ≤ γ in λ we get a commutative diagram.

f p 0 / Y / N / Xβ / 0 O O kγ jγ

fγ pγ 0 / Y / Nγ / Xγ / 0 O O kα,γ jα,γ

fα pα 0 / Y / Nα / Xα / 0 Each of the sequences

0 → Y → Nγ → Xγ → 0

is exact. Moreover for each α + 1 ≤ λ the map kα,α+1 has the same cokernel as jα,α+1. In particular it is in AdMonL. It suﬃces to show that the map lim→Nα → N is an isomorphism. Then we may apply Lemma 3.4.57. Indeed by assumption we have a commutative diagram of exact sequences

0 / Y / lim→Nα / lim→Xα / 0

0 / Y / N / X / 0 The ﬁrst and last vertical maps are isomorphisms, so the middle one is as well. Let us now deﬁne cotorsion pairs, and discuss their relation with weak factorisation systems. We shall largely follow the notation of [St’12ˇ ].

Definition 4.1.2. Let E be an exact category. A cotorsion pair on E is a pair of families of objects (L, R) of E such that L = ⊥R and R = L⊥. Definition 4.1.3. A cotorsion pair (L, R) is said to have enough (functorial) projectives if for every X ∈ E there is an admissible epic p : Y → X, (functorial in X), such that Y ∈ L and Ker(p) ∈ R. It is said to have enough (functorial) injectives if, for every X, there is an admissible monic i : X → Z, (functorial in X), such that Z ∈ R and Coker(i) ∈ L. A cotorsion pair is said to be (functorially) complete if it has enough (functorial) projectives and enough (functorial) injectives.

Example 4.1.4. Our main example is the projective cotorsion pair. Let E be an exact category. Let Proj(E) denote the collection of projective objects of E. Then (Proj(E), Ob(E)) is clearly a cotorsion pair. Suppose that E has enough (functorial) projectives. Then the cotorsion pair (Proj(E), Ob(E)) is trivially (functorially) complete.

Notation 4.1.5. Let E be an exact category and (L, R) a weak factorisation system on E. Denote by CokerL the collection of objects L such L is a cokernel of some map in, L and by KerR the collection of objects R such that R is the kernel of some map in R. Given classes of objects A, B in E, we denote by Inﬂ(A) the class of admissible monics with cokernel in A and by Deﬂ(B) the class of admissible epics with kernel in B.

Definition 4.1.6. Let E be an exact category. A weak factorisation system (L, R) on E is said to be compatible if (1) f ∈ L if and only if f is an admissible monic and 0 → Coker(f) belongs to L. (2) f ∈ R if and only if f is an admissible epic and Ker(f) → 0 belongs to R. The following result is Theorem 5.13 in [St’12ˇ ]. 56 4. MODEL STRUCTURES ON EXACT CATEGORIES

Theorem 4.1.7. Let E be an exact category. Then

(L, R) 7→ (CokerL, KerR ) and (A, B) 7→ (Infl(A), Defl(B)) define mutually inverse bijective mappings between compatible weak factorisation systems and complete cotorsion pairs. The bijections restrict to mutually inverse mappings between compatible functorial weak factorisation systems and functorially complete cotorsion pairs.

4.1.1. Compatible Model Structures. Having described the bijection between cotorsion pairs and compatible weak factorisation systems, we now introduce compatible model structures, and explain how they too correspond to purely homological data. Remember that we do not assume our model categories are complete or cocomplete. Let (C, F, W) be a model structure on an additive category E.

Definition 4.1.8. Let E be an exact category. Let (C, F, W) be a model structure on E. The model structure is said to be compatible if both (C ∩ W, F) and (C, F ∩ W) are compatible weak factorisation systems.

Let us now deﬁne the corresponding homological data. As for abelian categories, we will call a subcategory D of an exact category E thick if whenever 0 → A → B → C → 0 is a short exact sequence and two of the objects are in D, then so is the third.

Definition 4.1.9. A Hovey triple on an exact category E is a triple (C, W, F) of collections of objects of E such that the full subcategory on W is closed under retracts and thick, and that both (C, F ∩ W) and (C ∩ W, F) are complete cotorsion pairs.

We then have the following theorem (Theorem 6.9 in [St’12ˇ ]). It is originally due to [Hov02] in the abelian case and [Gil11] Theorem 3.3 in the more general exact case.

Theorem 4.1.10. Let E be a weakly idempotent complete exact category. Then there is a bijection between Hovey triples and compatible model structures. The correspondence assigns to a Hovey triple (C, W, F) the model structure (C, W, F) such that (1) C = Infl(C) (2) F = Defl(F) (3) W consists of morphisms of the form p ◦ i where i ∈ Infl(W) and p ∈ Defl(W).

Before we move on let us mention a more general notion than compatible model structures. We will need it when we consider the projective model structure on Ch≥0(E).

Definition 4.1.11. Let E be an exact category. A model structure (C, F, W) on E is said to be left pseudo-compatible if there are classes of objects C and W such that (1) The full subcategory on W is thick. (2) A map f is in C (resp. C ∩ W) if and only if it is an admissible monic with cokernel in C (resp. C ∩ W). (3) An admissible monic is in W if and only if its cokernel is in W. As before C/ W /C ∩ W are called the cofibrant /trivial/ trivially cofibrant objects. The pair (C, W) will be called the left homological Waldhausen pair of the model structure. Dually one defines right pesudo-compatible model structures and right homological Waldhausen pairs

The terminology comes from the notion of a Waldhausen category, in which classes of weak equivalences and coﬁbrations are speciﬁed. Clearly any compatible model structure is left pseudo-compatible. 4.1. COTORSION PAIRS 57

4.1.2. Small Cotorsion Pairs and Cofibrant Generation. When working with model categories, it is computationally convenient that they be generated by suitably compact objects (see Appendix A for exactly what we mean here). In this section, we study what conditions on the cotorsion pairs defining a compatible model structure guarantee that the model structure is cofibrantly small. The material here is adapted from [Hov02] §6 to exact categories. Gillespie has also done this in [Gil16a] Section 5.2 for the G-exact structure on a Grothendieck abelian category. Definition 4.1.12. Let E be an exact category. A cotorsion pair (L, R) on E is said to be cogenerated by a set if there is a set of objects G in L such that X ∈ R if and only if Ext1(G, X) = 0 for all G ∈ G. Definition 4.1.13. Suppose E is an exact category. A cotorsion pair (L, R) is said to be small if the following conditions hold (1) L contains a set of admissible generators of E. (2) (L, R) is cogenerated by a set G. (3) For each G ∈ G there is an admissible monic iG with cokernel G such that, if HomE (iG,X) is surjective for all G ∈ G, then X ∈ R.

The set of iG together with the maps 0 → Ui for some generating set {Ui} contained in L is called a set of generating morphisms of (L, R). There is an easy example. Example 4.1.14. Recall the projective cotorsion pair (Proj(E), Ob(E)). Suppose that the category E is projectively generated, with P a generating set of projectives. We claim that in this case the projective cotorsion pair is small. Indeed by assumption Proj(E) contains a set of generators P. This set trivially cogenerates the cotorsion pair as well. The third condition is also trivial. We now come to the connection between cofibrantly small model structures and cotorsion pairs. The proof of the following is a straightforward modification of [Hov02] Lemma 6.7 and Proposition 5.4 in [Gil16a]. Lemma 4.1.15. Let E be a weakly idempotent complete exact category together with a compatible weak factorisation system (L, R) with corresponding cotorsion pair (L, R). If the cotorsion pair is small with generating morphisms I = {0 → Ui} ∪ {iG}, then this weak factorisation system is cofibrantly small. If in addition the generating morphisms have cell(I)-presented domain, the weak factorisation system is cofibrantly generated. 4.1.3. Cotorsion Pairs on Monoidal Exact Categories. In this section (E, ⊗, k) is a monoidal exact category. We will now study sufficient conditions on cotorsion pairs defining a model category structure so that the resulting structure is monoidal. We generalise the work of [Hov02] §7 to exact categories. In fact this has mostly been done in [St’12ˇ ], apart from the monoid axiom. The theorem below is proven for compatible model structures in [St’12ˇ ]. However the proof for left pseudo-compatible model structures is formally identical. Note that [St’12ˇ ] assumes that the monoidal unit k is in fact already cofibrant, which makes the final assumption below automatic. This does not affect the proof. Theorem 4.1.16. Let E be a closed symmetric monoidal exact category. Suppose that E has a left pseudo- compatible model structure with Waldhausen pair (C, W). Suppose the following conditions are satisfied. (1) Every cofibration is pure. (2) If X,Y ∈ C then X ⊗ Y ∈ C. (3) If X,Y ∈ C and one of them is in W, then X ⊗ Y ∈ C ∩ W. (4) If C → k is an acyclic fibration with C in C, then for any object X of E, C ⊗ X → X is a weak equivalence Then E is a monoidal model category. Note that Lemma 2.4.76 says that if cofibrant objects are flat then condition 1 in Theorem 4.1.16 is automatically satisfied. 58 4. MODEL STRUCTURES ON EXACT CATEGORIES

Theorem 4.1.17. Let E be a complete and cocomplete, monoidal exact category. Suppose that E has a left pseudo-compatible model structure satisfying the hypotheses of Theorem 4.1.16. In addition, suppose that the following conditions hold (1) If X ∈ C ∩ W and Y is arbitrary, then X ⊗ Y is in W. (2) Transfinite compositions of weak equivalences which are also pure monics with cokernel in C ∩ W ⊗ Ob(E) are still weak equivalences. Then the model structure satisfies the monoid axiom. Proof. The first condition implies that if i is an acyclic cofibration, then i ⊗ Y is a weak equivalence. By Propositions 2.4.77 and the fact that pushouts commute with cokernels any pushout of i ⊗ Y is a weak equivalence as well as a pure monic. By the second condition, any transfinite composition of such maps is a weak equivalence. If in E transfinite compositions of admissible monics are admissible monics (e.g. if E is weakly AdMon- elementary) then one can replace the second condition by requiring that the class W is closed under transfinite compositions of pure monomorphisms. By this we mean that if λ is some ordinal, and X : λ → E a continuous functor such that 0 → X0 is a weak equivalence, and for each i < j in λ the map Xi → Xj is a pure monic which is also a weak equivalence, then Xλ is in W. (This is the condition used in [Hov02] Theorem 7.4). Since W forms a thick subcategory and X0 → Xλ is an admissible monic, this is equivalent to the cokernel of the map X0 → Xλ being in W which in turn is equivalent to X0 → Xλ being a weak equivalence.

4.2. Model Structures on Chain Complexes Generalising results of [Gil04] and and [Gil16a], in this section we describe a method for constructing compatible model structures on categories of chain complexes Ch∗(E) from cotorsion pairs on E. Note that what we describe below will not always produce a model structure. However we will show in the next chapter that it does in the case that E has enough projectives, and the cotorsion pair is the projective one (Example 4.1.4). First we define the collections of objects which will be candidates for the (trivially) fibrant and (trivially) cofibrant objects. Definition 4.2.18. Let (L, R) be a cotorsion pair on an exact category E. Let X ∈ Ch(E) be a chain complex.

(1) X is called an L complex if it is acyclic and ZnX ∈ L for all n. The collection of all L complexes is denoted Le. (2) X is called an R complex if it is acyclic and ZnX ∈ R for all n. The collection of all R complexes is denoted R‹. (3) X is called a K-L complex if Hom(X,B) is exact whenever B is an R complex. The class of all K-L complexes is denoted KL. (4) X is called a K-R complex if Hom(A, X) is exact whenever A is an L complex. The class of all K-L complexes is denoted KR. (5) X is called a dgL complex if X is a K-L complex and Xn ∈ L for each n ∈ Z. The collection of all dgL complexes is denoted dgﬁL. (6) X is called a dgR complex if X is a K-R complex and Xn ∈ R for each n ∈ Z. The collection of all dgR complexes is denoted dgﬁR.

Notation 4.2.19. We define the collections Le, R‹, dgfiL, dgfiR,KL,KR similarly in the categories Ch∗(E) for ∗ ∈ {≥ 0, ≤ 0, +, −, b}. We will use the same notation for these collections irrespective of which category of chain complexes we are working in.

Remark 4.2.20. In Ch∗(E) for ∗ ∈ {+, −, ≥ 0, b, ∅} all of the above classes are closed under shifts [n] for n ≤ 0. For ∗ ∈ {+, −, ≤ 0, b, ∅} they are closed under shifts [n] for n ≥ 0. Let us start to populate these collections. We ﬁrst make the following easy observation. 4.2. MODEL STRUCTURES ON CHAIN COMPLEXES 59

Proposition 4.2.21. Let X be an R-complex. Then Xn ∈ R for each n. Proof. For each n we have a short exact sequence

0 → ZnX → Xn → Zn−1X → 0 and ZnX,Zn−1X ∈ R. By Corollary 4.1.1 R is closed under extensions. With this in hand the result belows generalises immediately from [Gil04] Lemma 3.4. Lemma 4.2.22. (1) Bounded below complexes with entries in L are dgL complexes. (2) Bounded above complex with entries in R are dgR complexes. Gillespie’s crucial Proposition 3.6 in [Gil04] does not hold in arbitrary exact categories. However some of it can be salvaged to give the following two results.

Proposition 4.2.23. Let (L, R) be a cotorsion pair in an exact category E. Then in Ch∗(E) for ∗ ∈ {+, −, b, ∅} we have (1) dgfiL = ⊥R‹. (2) dgfiR = Le⊥ (3) R‹ ⊆ (dgfiL)⊥ (4) Le ⊆ ⊥(dgfiR) (5) Suppose E has enough L-objects. Let X ∈ (dgfiL)⊥ be good. Then X is an R-complex. (6) Suppose E has enough R-objects. Let X ∈ ⊥dg(R‹) be cogood. Then X is an L-complex. Proof. Parts 1) and 3) are easily seen to generalise to the exact case from the Gillespie’s proof. ⊥ 1 1 (1) Let X ∈ R‹. Then Ext (X,B) = 0 whenever B is an R complex. In particular Extdw(X,B) = 0. Hence Hom(X,B) is exact whenever B is an R complex by Corollary 2.2.56. It remains to show Xn ∈ L. Let B ∈ R. By Lemma 2.6.107 we have 1 1 n+1 Ext (Xn,B) = Ext (X,D (B)) = 0 n+1 ⊥ since D (B) ∈ R‹. So Xn ∈ L, and R‹ ⊂ dgfiL. Now let X ∈ dgfiL. Since the entries of X are in L, for any Y ∈ R‹, any short exact sequence 0 / Y / Z / X / 0 1 is split exact in each degree. But also Extdw(X,Y ) = 0. Hence, any sequence as above must be split exact, i.e. Ext1(X,Y ) = 0. (2) This is dual to the previous part. 1 1 (3) Let X ∈ R‹ and A ∈ dgfiL. Note that since Xn ∈ R, Ext (X,A) = Extdw(X,A). Now since 1 Hom(A, X) is exact, Extdw(X,A) = 0. (4) This is dual to the previous part. (5) Let us show that X is acyclic. We will again use Proposition 2.2.28. Let n be such that dn has a 0 0 kernel. Since we have enough L-objects, we may choose an admissible epic fn : A → ZnX for some 0 n 0 1 n 0 A ∈ L. By Lemma 2.6.107 this induces a map f : S (A ) → X. Now Extdw(S (A )[−1],X) ⊂ Ext1(Sn(A0)[−1],X) = 0 by assumption. Hence f is homotopic to 0. Applying Proposition 2.2.37 0 the map dn+1 : Xn+1 → ZnX is an admissible epic. By Proposition 2.2.28 X is acyclic. To see that ZnX ∈ R, we note that since X is acyclic, we have for any A ∈ L, 1 ∼ 1 n ExtE (A, ZnX) = Ext (S (A),X) = 0 ⊥ Since (L, R) is a cotorsion pair, ZnX ∈ R. Hence X ∈ R‹ and so (dgfiL) ⊆ R‹. (6) The proof for the second part is dual. We also have the following. 60 4. MODEL STRUCTURES ON EXACT CATEGORIES

Proposition 4.2.24. Let ∗ ∈ {≥ 0}, and let (L, R) be a cotorsion pair in E with enough L-objects. Then dgfiL = ⊥R‹ and R‹ = (dgfiL)⊥. Dually, if the cotorsion pair has enough R-objects, then for ∗ ∈ {≤ 0} dgfiR = Le⊥ and Le = ⊥dgfiR. Proof. The proofs of parts (3) and (5) in the previous proposition go through here, as does the proof that dgfiL ⊂ ⊥R‹. Now let X ∈ ⊥R‹. The same proof as in part (1) of the previous proposition shows that each Xn must be an object in L. Thus X is a bounded below complex of objects in L and hence a dgfiL complex. We get an immediate corollary.

Corollary 4.2.25. Let (L, R) be a cotorsion pair on an exact category E with enough L-objects and enough R-objects.

(1) (dgfiL, R‹) is a cotorsion pair on Ch≥0(E) and Ch+(E). If E has all kernels then it is a cotorsion pair on Ch(E). (2) (Le, dgfiR) is a cotorsion pair on Ch≤0(E) and Ch−(E). If E has all cokernels then it is a cotorsion pair in Ch(E). (3) (Le, dgfiR) and (dgfiL, R‹) are cotorsion pairs in Chb(E). (4) If E has all kernels and cokernels, in particular if E is quasi-abelian, then (Le, dgfiR) and (dgfiL, R‹) are cotorsion pairs in Ch(E). 4.2.1. Existence of dg-Model Structures. The hope now is that the class W of acyclic complexes satisfies Le = dgfiL ∩ W, R‹ = dgfiR ∩ W and that the cotorsion pairs (dgfiL, R‹) and (Le, dgfiR) are functorially complete. It is not at all clear that this will be the case. In [YD14] it is shown that for a complete and cocomplete abelian category it is always the case. We suspect this result can be easily adapted for complete or cocomplete exact categories satisfying a similar condition. In fact in private communication [Mor20] TimothéeMoreau has shown that it holds whenever E has exact products and exact coproducts. For the projective model structure we will only require that complexes of projectives satisfy the AB4-k axiom for some k (though we suspect Moreau’s work can be easily generalised to arbitrary complete cotorsion pairs on E satisfying some version of the AB4−k axiom.) In this section we will give some partial results about the induced cotorsion pairs on Ch(E) without any exactness assumptions on products or coproducts. First we need acyclic complexes to form a thick subcategory.

Proposition 4.2.26. Let E be a weakly idempotent complete exact category. Then for ∗ ∈ {≥ 0, ≤ 0, +, −, b} the full subcategory on W is a thick subcategory of Ch∗(E). If E has all kernels then this is also true for ∗ = {∅}.

Proof. One may assume that E is abelian by passing to a left abelianisation for ∗ ∈ {≥ 0, +, b}, (or a right abelianisation for ∗ ∈ {≤ 0, −}). The result in this case follows from the long exact sequence on homology. Remark 4.2.27. Timoth´eeMoreau has pointed out a nice proof to us which does not use abelianisations.

It turns out that we always have the inclusions Le ⊂ dgfiL ∩ W, and R‹ ⊂ dgfiR ∩ W. This follows from the next result, which is an easy modification of the proof of [Gil04] Lemma 3.9. Lemma 4.2.28. Every chain map from an L complex to an R complex is homotopic to 0.

Corollary 4.2.29. Let (L, R) be a cotorsion pair in an exact category. Then Le ⊂ dgﬁL ∩ W, and R‹ ⊂ dgﬁR ∩ W. 4.2. MODEL STRUCTURES ON CHAIN COMPLEXES 61

In order to have any chance of getting the reverse inclusion, we’ll need the cotorsion pair on E to be hereditary. The following deﬁnition and the subsequent proposition are immediate generalisations of [Roz99] §1.2.3 from abelian categories to exact categories. Definition 4.2.30. A cotorsion pair (L, R) is said to be hereditary if Exti(A, B) = 0 for any A ∈ L,B ∈ R and i ≥ 1. Example 4.2.31. Clearly the projective cotorsion pair is hereditary. Proposition 4.2.32. Let (L, R) be a hereditary cotorsion pair on a weakly idempotent complete exact category E. Then (1) L is resolving. That is L is closed under taking kernels of admissible epis. (2) R is coresolving. That is R is closed under taking cokernels of admissible monics. If E has enough R-projectives then (L, R) is hereditary if and only if L is resolving. Dually if E has enough L-injectives then (L, R) is hereditary if and only if R is coresolving. With this result in hand [Gil04] Theorem 3.12 generalises immediately to the exact setting.

Theorem 4.2.33. Let (L, R) be a hereditary cotorsion pair in an exact category E. If E has enough projectives then in Ch∗(E) for ∗ ∈ {≥ 0, +, ∅}, dgfiR ∩ W = R‹. If E has enough injectives then in Ch∗(E) for ∗ ∈ {≤ 0, −, ∅} dgfiL ∩ W = Le. In particular, if E has enough projectives and injectives, then the induced cotorsion pairs on E are compatible. Lemma 3.14 in [Gil04], which partially handles the case in which we may not have enough injectives or projectives also passes essentially unaffected to exact categories.

Lemma 4.2.34. Let E be an exact category and (L, R) a cotorsion pair on E. Consider the categories Ch∗(E) for any ∗ ∈ {≥ 0, ≤ 0, +, −, b, ∅}. (1) If (Le, dgfiR) is a cotorsion pair with enough projectives and dgfiR ∩ W = R‹ then dgfiL ∩ W = Le. (2) If (dgfiL, R‹) is a cotorsion pair with enough injectives and dgfiL ∩ W = Le then dgfiR ∩ W = R‹. These next two results partially deal with the issue of completeness.

Lemma 4.2.35. Let E be an exact category. Suppose

f 0 / B / A / X / 0 is a short exact sequence of complexes in the degree-wise exact structure with both B and cone(f) either good or cogood. Then B is acyclic if and only if f is a quasi-isomorphism.

Proof. Let I : E → A a suitable abelianisation . Then by [Wei95] Exercise 1.59 there is a long exact sequence

... / Hn+1(Ker(I(f•))) / Hn(cone(I(f•))) / Hn(Coker(I(f•))) / ...

/ Hn−1(Ker(I(f))) / ...

If f• is a quasi-isomorphism, then cone(I(f•)) is acyclic. It is also an admissible epimorphism, so Coker(I(f•)) = 0. Hence Ker(I(f•)) = I(B) is acyclic. If B is acyclic then again since Coker(I(f•)) = 0, Hn(cone(I(f•))) = 0 as well. Thus I(f) is a quasi- isomorphism, so f is as well. 62 4. MODEL STRUCTURES ON EXACT CATEGORIES

Proposition 4.2.36. Let (L, R) be a functorially complete cotorsion pair on a weakly idempotent complete exact category E. Then any complex X• of Ch∗(E) where ∗ ∈ {≥ 0, +} admits a resolution by an object L• of dgﬁL whose kernel is an acyclic complex R• with Rn ∈ R. In particular the cotorsion pair (dgProjﬂ, Ob›) on both Ch≥0(E) and Ch+(E) has enough functorial projectives.

Proof. Let X• be an object of Ch∗(E) where ∗ ∈ {≥ 0, +}. By an easy adaptation of the proof of Lemma 2.3.58, one can ﬁnd a (functorial) quasi-isomorphism f• : L• → X• with each Ln an object of L, which is an admissible epimorphism, and whose kernel is a complex R• with Rn ∈ R. Now L• is a dgL complex by Lemma 4.2.22. By Lemma 4.2.35 R• is acyclic. 4.2.2. Properties of dg-Model Structures.

Definition 4.2.37. Let E be a weakly idempotent complete exact category and (L, R) a cotorsion pair on E.

(1) We say that (L, R) is dg≥0-compatible if (dgfiL, R‹) is a functorially complete cotorsion pair on Ch≥0(E), W ∩ dgfiL = Le and the model structure whose cofibrations are Infl(dgfiL), and whose acyclic cofibrations are Infl(Le) exists on Ch≥0(E). (2) We say that (L, R) is dg≤0-compatible if (Le, dgfiR) is a functorially complete cotorsion pair on Ch≤0(E), W∩dgfiR = R‹ and the model structure whose fibrations are Defl(dgfiR), and whose acyclic fibrations are Defl(R‹) exists on Ch≤0(E). (3) For ∗ ∈ {b, +, −∅} we say that (L, R) is dg∗-compatible if (Le, dgfiR) and (dgfiL, R‹) are (functorially) complete cotorsion pairs on Ch∗(E), dgL ∩ W = Le, and dgR ∩ W = R‹

Let us establish some properties of model structures arising from dg∗-compatible cotorsion pairs. First note the following easy observation.

Proposition 4.2.38. Let (L, R) be a cotorsion pair on E which has kernels, and which is both dg≥0- compatible and dg-compatible. Then with the induced model structures on Ch≥0(E) and Ch(E) the adjunction

i: Ch≥0(E) Ch(E):τ≥0 is a Quillen adjunction, where i is the natural inclusion functor.

Proposition 4.2.39. Let E be weakly idempotent complete.

(1) If (L, R) is dg≥0-compatible and f : X → Y is a fibration, then f is an admissible epimorphism in each strictly positive degree. If f is an acyclic fibration then f is a quasi-isomorphism and an admissible epimorphism in each degree. (2) If (L, R) is dg≤0-compatible and f : X → Y is a cofibration, then f is an admissible monomorphism in each strictly negative degree. If f is an acyclic cofibration then f is a quasi-isomorphism and an admissible monomorphism in each degree. Proof. We prove the first claim, the second being dual. First assume that f is a fibration. Then f has the right lifting property with respect to acyclic cofibrations. In particular it has the right lifting property with respect to maps of the form 0 → Dn(L) where n ≥ 1 and L ∈ L. Since (L, R) is complete, this implies that fn is an admissible epimorphism in each strictly positive degree. Now suppose that f is an acyclic fibration. By the first part fn is an admissible epimorphism in each strictly positive degree. But f also 0 has the right lifting property with respect to the map 0 → S (L) for any L in L, implying that f0 is an admissible epimorphism. Now for any n and any L ∈ L one can always find a lift in the diagram

Sn(L) / X

f Dn+1(L) / Y 4.2. MODEL STRUCTURES ON CHAIN COMPLEXES 63

Indeed for n < −1 this is trivial, and for n ≥ 0 this follows from the fact that f is an acyclic fibration. For n = −1 the finding a lift amounts to finding a lift in the diagram

0 / X0

f0 L / Y0 i.e. a lift in the diagram 0 / X

S0(L) / Y Now the claim follows from Corollary 2.6.110.

Proposition 4.2.40. Let (L, R) be a dg∗-compatible cotorsion pair for ∗ ∈ {∅, ≥ 0, b, +}. If f is a coﬁbration (resp. trivially coﬁbrant) in Ch∗(E), then so is f[−1].

Proof. It suﬃces to observe that the classes dgﬁL and Le are closed under applying [−1].

Proposition 4.2.41. Let (L, R) be a dg∗-compatible cotorsion pair for ∗ ∈ {∅, ≥ 0, b, +}. If f : X → Y is a map between cofibrant objects then cone(f) is cofibrant. It is trivially cofibrant whenever f is a quasi- isomorphism. Proof. There is an exact sequence 0 → Y → cone(f) → X[−1] → 0

Thus the result follows from Proposition 4.2.40 and Corollary 4.1.1.

If (L, R) is dg∗-compatible, then there is an induced model structure on Ch∗(E). The resulting model structure will have quasi-isomorphisms as its weak equivalences.

Proposition 4.2.42. Suppose that ∗ ∈ {≥ 0, ≤ 0, +, −, b} and that (L, R) is a dg∗-compatible cotorsion pair on an exact category E. The weak equivalences in the induced model structure are precisely the quasi- isomorphisms. If E has all kernels then this is also true for ∗ ∈ {∅}. Proof. First we show that admissible monics and admissible epics which are weak equivalences are quasi-ismorphisms. By duality it suﬃces to show it for epics and ∗ ∈ {≥ 0, b, +, −, ∅, ≤ 0}. Let f : B → C be an an admissible epic which is a weak equivalence. It is suﬃcient to show that I(f) is a quasi-isomorphism, where I : E → A(E) is a suitable abelianisation. Now we have an exact sequence

g f 0 / A / B / C / 0 Let us argue that A is acyclic. We can factor f as f = p ◦ i where p is an acyclic fibration and i is an acyclic cofibration. By Proposition 4.2.39 for ∗ ∈ {≥ 0, ≤ 0} and Theorem 4.1.10 for ∗ ∈ {b, +, −, ∅}, both i is an admissible monomorphism with acyclic cokernel, and p is an admissible epimorphism with acyclic kernel. By the snake lemma there is an exact sequence 0 → ker(f) → ker(p) → coker(i) → 0. In particular ker(f) is acyclic by Lemma 4.2.35. Again by Lemma 4.2.35, f is acyclic. Now let f be a morphism of Ch∗(E). Factor it as p ◦ i where i is a fibration, p is a cofibration and either p or i is trivial, and therefore a quasi-isomorphism. By the exact triangle (after passing to an abelianisation) cone(i) → cone(f) → cone(p) →+1 and the fact that acyclic complexes form a thick subcategory, we find that f is a quasi-isomorphism if and only if the other factor is trivial. 64 4. MODEL STRUCTURES ON EXACT CATEGORIES

Proposition 4.2.43. Let E be an exact category and S a class of morphisms in E closed under finite direct sums. Suppose that E is weakly S-elementary. Then transfinite compositions of quasi-isomorphisms in Ch(E) which are also maps in S are quasi-isomorphisms. Proof. The proof is by transfinite induction. Since a finite composition of quasi-isomorphisms is a quasi- isomorphism, the successor part of the induction is finished. Now let λ be a limit ordinal and F : λ → Ch(E) a continuous functor with F (α ≤ β)n ∈ S for any morphism α ≤ β in λ and n ∈ Z. For α ≤ β ≤ λ denote by fα,β the map Fα → Fβ. For β ≤ λ write fβ = f0,β. It is clear that ∼ cone(fλ) = lim→β<λ cone(fβ)

Since each fβ is a quasi-isomorphism, cone(fβ) is acyclic. Since E is weakly S-elementary, this implies

lim→β<λ cone(fβ) is acyclic, which means that cone(fλ) is acyclic and hence that fλ is a quasi-isomorphism. Corollary 4.2.44. Let E be a weakly idempotent complete exact category and S a class of morphisms in E such that E is weakly S-elementary as an exact category. Suppose that there is a model category structure on Ch(E) in which the weak equivalences are the quasi-isomorphisms, and the class of coﬁbrations is contained in Ch(S). Then Ch(E) is weakly Ch(S)-elementary as a model category (Deﬁnition A.3.17).

Proof. Let λ be a limit ordinal and F : λ → Ch(E) a continuous functor with F (α ≤ β)n ∈ S. Let F‹ → F be a coﬁbrant replacement for F . Then F‹(α ≤ β) is a coﬁbration so is in Ch(S). Thus the map

lim→α<λ F‹ → lim→α<λ F is an equivalence, as required. Such model structures are also both left and right proper. More generally, we have the following. Proposition 4.2.45. Let E be a weakly idempotent complete exact category. Let ∗ ∈ {≥ 0, ≤ 0, +, −, b}. Suppose there is a model structure on Ch∗(E) whose weak equivalences are the quasi-isomorphisms and such that any coﬁbration is an admissible monomorphism in each degree. Then the model structure is left proper. If E has all kernels then this is also true for Ch(E). Dually, if any ﬁbration is an admissible epimorphism in each degree then the model structure is right proper. Proof. The dual case is slightly easier to write down, so we will prove that. We need to check that, given a pull-back diagram p0 A• / B•

q0 q p X• / Y• where p is an admissible epic, and q is a quasi-isomorphism, then q0 is a quasi-isomorphism. By Lemma 2.1.6 without loss of generality, we may assume that the category E is actually abelian. We argue by elements. A• is isomorphic to {(x, b) ∈ X• × B• : p(x) = q(b)} 0 0 A 0 with q and p being the restrictions of the projections. Suppose (x, b) ∈ Ker(dn ) is such that q (x, b) = x = 0. B A X But then q(b) = p(x) = 0. So b = dn+1(eb) for some b, and (x, b) = dn+1((0,eb)). Now suppose x ∈ Ker(dn ). Y B Y Then p(x) ∈ Ker(dn ). Thus there is a b ∈ Ker(dn ) and a ye ∈ Yn+1 such that q(b) = p(x) + dn+1(ye). Now, X p is an epic, so there is xe ∈ Xn+1 such that ye = p(xe). Write a = (x + dn+1(xe), b). Then a ∈ A• and 0 X 0 q (a) = x + dn+1(xe). This shows that q is a quasi-isomorphism. Corollary 4.2.46. Let E be a weakly idempotent complete exact category. Let ∗ ∈ {≥ 0, ≤ 0, +, −, b}. Suppose there is a model structure on Ch∗(E) whose weak equivalences are the quasi-isomorphisms and such that any cofibration is an admissible monomorphism in each degree. Then degree-wise admissible monomorphisms are h-cofibrations (Definition A.3.11). If E has all kernels then this is also true for Ch(E). Dually, if any fibration is an admissible epimorphism in each degree then degree-wise admissible epimorphisms are h-fibrations. 4.2. MODEL STRUCTURES ON CHAIN COMPLEXES 65

Proof. We prove the left properness claim, the right properness claim being dual. The model structure is left proper by Proposition 4.2.45. By the proof of Proposition 4.2.45 pushouts of weak equivalences along degree-wise admissible monomorphisms are weak equivalences. Moreover degree-wise admissible monomorphisms are pushout-stable. The result now follows from Poposition A.3.14.

We also have the following useful result.

Proposition 4.2.47. Suppose there is a model structure on Ch(E), where E is weakly idempotent complete, whose weak equivalences are the quasi-isomorphisms. Then the model category is stable (A.3.16).

`L Proof. We need to compute the suspension functor 0 X 0. Now X → cone(IdX ) is a h-coﬁbration. `L ` ∼ Thus 0 X 0 can be computed as the pushout cone(IdX ) X 0 = X[−1]. Thus as one should expect, the suspension functor is the shift functor, which is an auto-equivalence.

4.2.3. Small dg-Cotorsion Pairs. Let us now examine when the cotorsion pair (dgﬁL, R‹) is small. Again this is done in [Gil16a] Section 4.4 for the G-exact structure on a Grothendieck abelian category.

Proposition 4.2.48. Let (L, R) be a cotorsion pair in an exact category E which has a set of admissible generators G. Suppose that (L, R) is cogenerated by a set {Ai}i∈I . Then (dgﬁL, R‹) is cogenerated by the set

n n S = {S (G): G ∈ G, n ∈ Z} ∪ {S (Ai): n ∈ Z, i ∈ I} for ∗ ∈ {+} (and. ∗ ∈ {∅} if E has kernels) and

n n S = {S (G): G ∈ G, n ≥ 0} ∪ {S (Ai): n ≥ 0, i ∈ I} for ∗ ∈ {≥ 0}. Furthermore, suppose (L, R) is small with generating morphisms the maps {0 → G : G ∈ G} together with monics ki as below (one for each i ∈ I).

ki 0 / Yi / Zi / Ai / 0

Then (dgﬁL, R‹) is small with generating morphisms the set

n n−1 n n n n Ie = {0 → D (G)} ∪ {S (G) → D (G)} ∪ {S (ki): S (Yi) → S (Zi)}

for ∗ ∈ {+} (and. ∗ ∈ {∅} if E has kernels) and

Ie = {0 → S0(G)} ∪ {0 → Dn(G): n > 0} ∪ {Sn−1(G) → Dn(G): n > 0}

n n n ∪{S (ki): S (Yi) → S (Zi): n ≥ 0} for ∗ ∈ {≥ 0}.

Proof. For ∗ ∈ {+, ∅} the proof of [Gil07] Proposition 3.8 generalises immediately to exact categories. Now consider the case ∗ ∈ {≥ 0}. The only diﬀerence in the proof is that now the generating set for Ch≥0(E) n 0 is {D (G): G ∈ G : n > 0} ∪ {S (G): G ∈ G}. This is also a subset of dgﬁL.

n Remark 4.2.49. The proof of the above proposition in fact shows that if Hom(S (F ),X•) is acyclic for any generating collection consisting of objects in L (i.e. not necessarily a set), then X• is a R‹-complex. Remark 4.2.50. In the situation of the previous proposition, if the domains of the generating morphisms for the cotorsion pair (L, R) are (κ, S)−? for ? ∈ {small, compact, presented, tiny} then the domains of the maps in I are Ch(S) − ? by Proposition 2.6.114. 66 4. MODEL STRUCTURES ON EXACT CATEGORIES

4.2.4. Monoidal Model Structures on Chain Complexes. In this section we investigate when cotorsion pairs on monoidal exact categories induce monoidal model structures on the category of chain complexes. First we have the following easy results, which says when a complex is flat. Proposition 4.2.51. Let (E, ⊗, k) be an additive symmetric monoidal category with E an exact category. 0 For ∗ ∈ {≥ 0, ≤ 0, b, +, −} the flat objects in (Ch∗(E), ⊗,S (k)) are precisely the complexes F• in Ch∗(E) such that for each n ∈ Z, Fn is flat. If in addition countable direct sums exist and are exact, then the flat 0 objects in (Ch(E), ⊗,S (k)) are also the complexes F• that for each n ∈ Z, Fn is flat Proof. Let 0 / X• / Y• / Z• / 0

be a short exact sequence in Ch∗(E). Let F• be a complex. Then the nth row of

0 / X• ⊗ F• / Y• ⊗ F• / Z• ⊗ F• / 0 is L L L 0 / i+j=n Xi ⊗ Fj / i+j=n Yi ⊗ Fj / i+j=n Zi ⊗ Fj / 0 Since the direct sums involved are exact, this sequence is short exact if for each i, j,

0 / Xi ⊗ Fj / Yi ⊗ Fj / Zi ⊗ Fj / 0 is short exact. It follows immediately that a complex whose entries are flat in E is itself a flat object in Ch∗(E). To see that a flat complex must have flat entries, simply take a short exact sequence in E, and regard it as a short exact sequence in Ch∗(E) concentrated in degree 0. Definition 4.2.52. Let (E, ⊗) be a monoidal exact category. A cotorsion pair (L, R) on E is said to be monoidally dg∗-compatible for ∗ ∈ {≥ 0, +, ∅} if

(1) (L, R) is dg∗ compatible. (2) For ∗ ∈ ∅ countable product functors are admissibly exact and and countable coproduct functors are admissibly coexact (3) L contains k, and is closed under ⊗. (4) short exact sequences in L are L-pure. Now let ∗ ∈ {≥ 0, ∅}. If in addition objects of L are flat, E is weakly (λ, PureMon)-elementary for any ordinal λ, where PureMon is the class of pure monomorphisms, and every (trivially) cofibrant complex is an (ℵ0; PureMon)-extension of bounded below (trivially) cofibrant complexes, then the cotorsion pair is said to be strongly monoidally dg∗-compatible.

Proposition 4.2.53. Let (L, R) be a hereditary monoidally dg≥0-compatible cotorsion pair on E, with E weakly idempotent complete..

(1) If L• is a dgLe-complex and R• is a R‹-complex then Hom(L•,R•) is a R‹-complex. (2) If L• is a Le-complex and R• is a dgfiR complex, then Hom(L•,R•) is a R‹-complex. n Proof. By Remark 4.2.49 for both statements it suffices to show that Hom(S (F ), Hom(L•,R•)) is n ∼ n acyclic for any F ∈ L, and all appropriate n. But Hom(S (F ), Hom(L•,R•)) = Hom(S (F ) ⊗ L•,R•). n Thus for the first statement we need to show that S (F ) ⊗ L• is a dgfiL-complex when L• is, and for the second we need to show that it is a Le-complex when L• is. Since objects of F are flat and L is closed under ⊗, if we can show that first statement then the second follows immediately. Let X• be a R‹-complex. Then n ∼ n Hom(S (F ) ⊗ L•,X•) = Hom(L•, Hom(S (F ),X•)) n To show that this is acyclic it now suffices to show that Hom(S (F ),X•) is a R‹-complex. By shifting we may assume that n = 0, and then this complex is just the internal hom functor in E taken degree-wise, Hom(F,X•). Since X• is an acyclic complex with Xn ∈ R and ZnX ∈ R for each n, this ∼ complex is clearly exact, and ZnHom(F,X•) = Hom(F,ZnX). Finally we reduce to showing that for 4.2. MODEL STRUCTURES ON CHAIN COMPLEXES 67

F ∈ L and G ∈ R, Hom(F,G) ∈ R. As the cotorsion pair (L, R) is complete, it suffices to show that 0 Hom(Z•,S (Hom(F,G))) is acyclic whenever Z• is a bounded below Le complex. Let Z• be such a com- 0 ∼ 0 0 0 plex. Now Hom(Z•,S (Hom(F,G))) = Hom(Z• ⊗ S (F ),S (G)). Z• ⊗ S (F ) is the complex obtained from Z• by tensoring with F degree-wise. By the assumptions on L this is clearly a Le-complex. More- 0 0 0 0 0 over S (G) is a dgfiR-complex. Thus Hom(Z•, Hom(S (F ),S (G))) is acyclic, and Hom(S (F ),S (G)) is a dgfiR-complex.

Proposition 4.2.54. Let (L, R) be a monoidally dg∗-compatible cotorsion pair for ∗ ∈ {≥, ∅} on weakly idempotent complete E. The model category structure induced by (L, R) on Ch∗(E) is monoidal. Proof. For the first part we use Theorem 4.1.16. First suppose that the cotorsion pair is monoidally 0 0 dg∗-compatible. Clearly S (k) is cofibrant. Let L and L be dgfiL complexes and let R be a R‹ complex. Then Hom(L ⊗ L0, R) =∼ Hom(L, Hom(L0, R)) By Proposition 4.2.53 Hom(L0, R) is a R‹-complex. Therefore Hom(L, Hom(L0, R)) is acyclic. Hence L⊗L0 is a dgfiL-complex. In particular the class of cofibrant objects is closed under ⊗. If one of them is acyclic 0 then again using Proposition 4.2.53 L ⊗ L is also acyclic. Proposition 4.2.55. Let L be a class of objects in a monoidal weakly idempotent complete exact category E, and suppose that E is weakly PureMon-elementary (1) Suppose that any admissible monomorphism with cokernel in L is pure. If X is any complex then for any complex L in Le, L ⊗ X is acyclic. (2) Suppose that objects in L are flat. If L is an (ℵ0; PureMon)-extension of bounded below complexes of objects in L then for any acyclic complex X, X ⊗ L is acyclic.

In particular if there is a strongly monoidally dg∗-compatible cotorsion pair (L, R) for ∗ ∈ {≥ 0, ∅}, then the induced model structure satisfies the monoid axiom. Moreover in this case if C is cofibrant and X is acyclic then C ⊗ X is acyclic. n n n+1 Proof. In either case we can write it as L = lim→L where the maps L → L are pure monomorphisms, and each Ln is bounded below. If L is in Le we may assume that each Ln is acyclic. Let X be ∼ n n n+1 any complex. Then X ⊗ L = lim→n (X ⊗ L ). Once again the maps X ⊗ L → X ⊗ L are pure monomorphisms. Suppose that either X or L is acyclic. We want to show that X ⊗ L is acyclic. By Proposition 4.2.43 it suffices to show that Ln ⊗ X is acyclic. Without loss of generality let us assume that Ln is concentrated in degrees ≥ 0. First suppose condition 1) is satisfied. We need to show that Ln ⊗ X is acyclic whenever Ln is a bounded below complex in Le. Such a complex can be written as an (ℵ0; PureMon)-extension of bounded complexes in Le. So we may in fact assume that Ln is bounded. We induct on the length of Ln. Suppose L is a complex in Le of length k + 1. There is an exact sequence k+1 0 → D (Lk+1) → L → τ≤kL → 0 n This is pure exact since τ≤kL is a complex of objects in L. Therefore tensoring with X gives an exact sequence k+1 ≥k 0 → X ⊗ D (Lk+1) → X ⊗ L → X ⊗ τ L → 0 k+1 By assumption X ⊗ τ≤kL is acyclic. Up to a shift, X ⊗ D (Lk+1) is X ⊗ cone(IdLk+1 ) and is therefore acyclic. Again by thickness, X ⊗ L is acyclic. n 0 n Suppose now that conidition 2) is satisfied. Now L may be obtained from S (L0 ) by a transfinite composition of pushouts of the form Sk(F ) / A

Dk+1(F ) / B 68 4. MODEL STRUCTURES ON EXACT CATEGORIES

Tensoring with X gives a pushout diagram

Sk(F ) ⊗ X / A ⊗ X

Dk+1(F ) ⊗ X / B ⊗ X Sk(F ) → Dk+1(F ) is a pure monomorphism. Therefore there is an exact sequence 0 → Sk(F ) ⊗ X → Dk+1(F ) ⊗ X → Sk+1(F ) ⊗ X → 0 By pushout there is an exact sequence 0 → A ⊗ X → B ⊗ X → Sk+1(F ) ⊗ X → 0 Sk+1(F ) ⊗ X is acyclic since, up to shift, it is tensoring with a ﬂat object. By induction on the length of the complex we may assume that A ⊗ X is acyclic. Since acyclic objects are thick, B ⊗ X is acyclic. Moreover n the map A ⊗ X → B ⊗ X is a pure monomorphism. Therefore L ⊗ X is an (ℵ0; PureMon)-extension of acyclic objects, and so is acylic. 4.3. The Projective Model Structure In this section we specialise to the cotorsion pair (Proj(E), Ob(E)). E will be an exact category with enough functorial projectives. We denote the collection of all projective objects in E by Proj(E)

Definition 4.3.56. Let E be an exact category. If it exists, the projective model structure on Ch∗(E), for ∗ ∈ {+, ∅, ≤ 0} is the model structure in which • Weak equivalences are quasi-isomorphisms. • Fibrations are degree-wise admissible epics. • Coﬁbrations are maps which have the left lifting property with respect to acyclic ﬁbrations.

Proposition 4.3.57. Let E be an exact category. Suppose that the cotorsion pair (dgProj‚(E), Obﬂ(E)) on Ch∗(E) for ∗ ∈ {+, ≥ 0, ∅} has enough functorial projectives. Then it has enough functorial injectives.

Proof. Let X• be an object of Ch∗(E), and let f• : L• → X• be a quasi-isomorphism and admissible epimorphism with acyclic kernel, and L• ∈ dgProj‚(E). We have a short exact sequence

0 → X• → cone(f•) → L•[−1] → 0 cone(f•) is an acyclic complex, so it is in Ob‡(E). Clearly L•[−1] ∈ dgProj‚(E). We are now ready to prove the following theorem. Theorem 4.3.58. Let E be a weakly idempotent complete exact category with enough projectives. Then the projective model structure exists on Ch+(E) and is compatible. It is functorial if E has enough functorial projectives. It is coﬁbrantly generated if E is elementary, and combinatorial if E is locally presentable. If DG-projective resolutions exist, in particular if E has all kernels and projectives satisfy condition AB4 − k for some k, then this is all true for Ch(E) as well. Proof. Consider the projective cotorsion pair (Proj(E), Ob(E)) on E. By Corollary 4.2.25,

(dgProj‚(E), Ob‡(E)) is a cotorsion pair on Ch+(E). It is functorially complete by Proposition 4.2.36 and Proposition 4.3.57.

We claim that (Proj‚(E), dgOb‡(E)) is also a cotorsion pair on Ch+(Ob(E)). First note that Proj‚(E) consists of split exact complexes of projectives. By Proposition 2.6.111 this is precisely the class of projective objects in Ch+(E). Then by Proposition 4.2.53 dgOb‡(E) = Ch+(Ob(E)). Hence (Proj‚(E), dgOb‡(E)) is just the projective cotorsion pair. Now Ob‡(E) is the class of all acyclic complexes, W. Thus dgOb‡(E)∩W = 4.3. THE PROJECTIVE MODEL STRUCTURE 69

Ch+(E) ∩ W = W = Ob‡(E). Moreover Ch+(E) has enough projectives by Corollary 2.6.112. By Lemma 4.2.34 it remains to prove that (Proj‚(E), dgOb‡(E)) is (functorially) complete. But in a category with enough (functorial) projectives the projective cotorsion pair is always (functorially) complete by Example 4.1.4. Assume further that E is elementary. Then by Example 4.1.14, the cotorsion pair (Proj‚(E), dgOb‡(E)) is small and by Proposition 4.2.48 the cotorsion pair (dgProj‚(E), Ob‡(E)) is small. By Lemma 4.1.15, the model structure is coﬁbrantly generated. The fact about combinatoriality is clear. The proof for unbounded complexes works in almost exactly the same way. All that needs to be veriﬁed in this case is that (dgProj‚(E), Ob‡(E)) is complete. Now the class of projectives is closed under (ℵ0; AdMon)- extensions by Corollary 4.1.1. Completeness therefore follows from from Corollary 2.3.67, Proposition 2.3.68 and Proposition 4.3.57. There is another common situation in which we have DG-projective resolutions, which was established in the relative homological algebra setting in [CH02] Proposition 4.2. This result can be considered as a version of ‘Case B’ of [CH02] Theorem 2.2, while the AB4 − k critestion is a generalisation of ‘Case A’. Proposition 4.3.59. Let E be an exact category with kernels and coproducts. Suppose that E has a collection P of (functorially) generating projectives such that there exists a cardinal κ for which each P ∈ P is (κ; AdMonProj)-small. Then Ch(E) has (functorial) DG-projective resolutions. Proof. The proof for exact categories works essentially identically to [CH02] Proposition 4.2, mutatis- mutandis. The only subtlety is that since we have assumed E has kernels, it holds that a map f : X → Y in Ch(E) is a weak equivalence if and only if for any P ∈ P, Hom(P, f) is a weak equivalence in Ch(Ab).

Remark 4.3.60. The projective model structure on Ch(E) (but not on Ch≥0(E)) is a projective model structure on an exact category in the sense of [Gil11] Definition 4.4. Remark 4.3.61. The existence of the projective model structure on bounded below chain complexes on a quasi-abelian category with enough projectives was already known to Bühler [Büh11] (see Appendix C). The proof there is more direct. In fact the proof works for any idempotent complete exact category in which the class of all kernel-cokernel pairs forms the exact structure (all kernels and cokernels need not exist). Recall that if E is (quasi)-elementary quasi-abelian, then Proposition 2.6.98 says that LH(E) is as well. Thus the projective model structure exists on Ch(LH(E)). Moreover the induced functor I : Ch(E) → Ch(LH(E)) is then right Quillen. Indeed it is left adjoint to the induced functor C : Ch(LH)(E) → Ch(E). It preserves fibrations since I : E → LH(E) is a left abelianisation, and it preserves quasi-isomorphisms by Corollary 2.5.89. Moreover by Theorem 2.5.87, Proposition 2.5.88 and Proposition 4.2.42 it induces an equivalence between the homotopy categories. We therefore have Proposition 4.3.62. Let E be an elementary quasi-abelian category. Then the adjunction

C ( Ch(LH(E)) Ch(E) h I is a Quillen equivalence between the projective model structures.

We claim that the projective model structure exists also on Ch≥0(E) for E an exact category with kernels. It will be strongly left pseudo-compatible, but not compatible. Definition 4.3.63. Let E be an exact category. If it exists, the projective model structure on Ch≥0(E), is the model structure in which • Weak equivalences are quasi-isomorphisms. • Fibrations are degree-wise admissible epics in each strictly positive degree. • Coﬁbrations are maps which have the left lifting property with respect to acyclic ﬁbrations. 70 4. MODEL STRUCTURES ON EXACT CATEGORIES

Theorem 4.3.64. Let E be an exact category with enough projectives and which has all kernels. Then the projective model structure exists on Ch≥0(E). Moreover it is a strong left pseudo-compatible model structure with Waldhausen pair (dg·Proj(E), W). In particular the acyclic coﬁbrations are the degree-wise admissible monics whose cokernels are split exact complexes of projectives. If E is SplitMon-elementary then it is cofbrantly generated. In particular if E is locally presentable and SplitMon-elementary then the projective model structure is combinatorial.

Proof. The class of weak equivalences satisfies the 2-out-of-6 property since it does so in Ch+(E). Denote the class of fibrations by F and of weak equivalences by W. Also denote the class of admissible monomorphisms with degree-wise projective cokernel by C. By Proposition 2.6.110 F ∩ W consists of quasi- isomorphisms which are admissible epimorphisms in each degree. By Proposition 4.2.36 and Proposition 4.3.57, it follows (C, F ∩ W) is a (compatible) weak factorisation system with corresponding cotorsion pair (dg‰Proj(E), W). In particular the cofibrations in the sense of Definition 4.3.63 coincide with the class C. It therefore remains to check that (C∩W, F) is a weak factorisation system. Let us first check the lifting conditions. First suppose a map A• → B• in Ch≥0(E) has the left lifting property with respect to maps X• → Y• in Ch≥0(E) which are admissible epimorphisms in each strictly positive degree. Let E• → F• be a map between any complexes in Ch(E) which is an admissible epimorphism in all degrees. Consider a diagram

A• / E•

B• / F•

Since A• and B• are in Ch≥0 we can factor the above diagram as

A• / τ≥0E• / E•

B• / τ≥0F• / F•

Now the map τ≥0E• → τ≥0F• is an epimorphism in each strictly positive degree. By assumption we can ﬁnd a lift as follows.

A• / τ≥0E• / E• <

B• / τ≥0F• / F•

Thus the map A• → B• has the left lifting property with respect to all degree-wise epimorphisms in Ch+(E). By Theorem 4.3.58 A• → B• is an admissible monic whose cokernel is a split exact complex of projectives. n Now, any acyclic cofibration is of the form A• → A•⊕ ⊕n>0 D (Pn) where each Pn is a projective object in E, and the map is the inclusion into the first factor of the direct sum. Clearly then it is enough to show that the collection of maps {0 → Dn(P ): n > 0,P is projective } has the left lifting property with respect to F, and that a map is in F if and only if it has the right lifting property with respect to these maps. However this follows from Lemma 2.6.107 and Proposition 2.6.93. It remains to find a (functorial) factorisation. Let f• : X• → Y• be a map in Ch≥0(E). We can factor it in Ch+(E) as

n X• → X•⊕ ⊕n≥0 D (Pn) → Y• 4.3. THE PROJECTIVE MODEL STRUCTURE 71

n n where X• → X•⊕ ⊕n≥0 D (Pn) is the inclusion into the ﬁrst factor, and X•⊕ ⊕n≥0 D (Pn) → Y• is an admissible epimorphism in each degree. Then

n X• → X•⊕ ⊕n>0 D (Pn) → Y•

n is also a factorisation of f•, X• → X•⊕ ⊕n>0 D (Pn) is an acyclic cofibration in Ch≥0(E), and X•⊕ ⊕n>0 n D (Pn) → Y• is an admissible epimorphism in each strictly positive degree. We prove the statement about cofibrant generation. Suppose that P is a projective generating set consisting of SplitMon-tiny objects. It follows from Proposition 4.2.48 that the weak factorsiation system (C, F ∩ W) is cofibrantly generated. From our proof above that (C ∩ W, F) is a weak factorisation system, it follows that {0 → Dn(P ): n > 0,P ∈ P} is a set of generating morphisms for (C ∩ W, F), so it is also a cofibrantly generated weak factorisation system. The claim about combinatoriality is clear.

Remark 4.3.65. The existence of the projective model structure on Ch≥0(E) in the case that E is quasi- abelian was also known. This is mentioned in a math.stackexchange.com exchange, [uh], as an adaptation of the proof for Ch+(E) in [Büh11]. We finish with the non-positively graded case, which appears for the category of abelian groups in [CC04]. Theorem 4.3.66. Let E be an exact category which has cokernels and DG-projective resolutions. Then the projective model structure exists on Ch≤0(E). Moreover, it is the right transferred model structure along the adjunction τ≤0 : Ch(E) Ch≤0(E):i≤0 where i≤0 is the inclusion functor. In particular it is cofibrantly generated if E is elementary, and combinatorial if E is locally presentable.

Proof. Let g : X → Y be a map of complexes in Ch≤0(E). We factor it as c : X → Y‹ , f : Y‹ → Y , where c is a cofibration in Ch(E), f is a fibration in Ch(E), and one of them is acyclic. We therefore get a factorisation of f as τ≤0c : X → τ≤0Y‹, τ≤0f : τ≤0Y‹ → Y . If c was an equivalence then so is τ≤0c, and likewise for f and τ≤0f. It remains to show that τ≤0c is a cofibration, and τ≤0f a fibration. Let X / A

τ≤0c g τ≤0Y‹ / B be a commutative diagram with g an acyclic ﬁbration. Then g is an acyclic ﬁbration in Ch(E). Thus there is a lift in the diagram X / A

c g Y‹ / B

Since A is concentrated in non-positive degrees the map Y‹ → A factors through τ≤0Y‹, and gives a lift in Ch≤0(E). Thus τ≤0c is a coﬁbration. Now the composition Y‹n → (τ≤0Y‹)n → Yn is an admissible epimorphism for each n ≤ 0, so (τ≤0Y‹)n → Yn is an admssible epimorphism for each n ≤ 0. Hence τ≤0f is a ﬁbration. The fact that it is the right-transferred model structure is clear.

Remark 4.3.67. Note that since the model category structure on Ch≤0(E) is transferred from the one on Ch(E), if P is a projective generating set for E, then {0 → Dn(P ): P ∈ P, n < 0} is a generating set of acyclic coﬁbrations for Ch≤0(E), and {0 → Dn(P ): P ∈ P, n < 0} ∪ {S0(P ) → 0 : P ∈ P} ∪ {Sn(P ) → Dn+1(P ): n < 0,P ∈ P} 72 4. MODEL STRUCTURES ON EXACT CATEGORIES

is a set of generating cofibrations for Ch≤0(E). 4.3.1. The Projective Model Structure on Monoidal Exact Categories. We now turn our attention to monoidal model structures on categories of chain complexes. Using the technology developed earlier, it is reasonably easy to prove the following. Theorem 4.3.68. Let E be a projectively monoidal weakly idempotent complete exact category with enough projectives. Then the projective model structure on Ch+(E) is monoidal. If E also has kernels, then the projective model structure on Ch∗(E) is monoidal. If in addition E is weakly elementary then Ch∗(E) satisfies the monoid axiom, and moreover all of this is also true for Ch(E) Proof. By Theorem 4.3.58, Theorem 4.3.64, and Proposition 4.2.54 all that remains to notice is that k is projective, projective objects are flat, and the class of projective objects is closed under ⊗.

The model category Ch≤0(E) is not a monoidal model category in general. Indeed consider the generating cofibration S0(P ) → 0, and let C be cofibrant. In particular 0 → C is a cofibration. The pushout-product axiom would imply that S0(P ) ⊗ C → 0 is a cofibration. Consider C = S1(Q) for some projective Q, and let P be the monoidal unit. Then S0(P )⊗C =∼ S1(P ⊗Q). The map S1(P ⊗Q) → 0 is very rarely a cofibration. Duality. In any closed monoidal category (E, ⊗, k, Hom) one can consider the functor (−)∨ : E → Eop,E 7→ Hom(E, k) This functor is contravariantly self-adjoint. ∨ Proposition 4.3.69. Let E be a monoidal elementary exact category. The functor (−) : Ch∗(E) → op Ch∗(E) is left Quillen for the projective model structure on the left and its opposite model structure on the right. op Proof. Since any object of Ch∗(E) is cofibrant and Hom(−, k) clearly preserves degree-wise split exact sequences all that remains to prove is that it sends trivially cofibrant objects to acyclic objects. Indeed ∨ if P• is trivially cofibrant then P• → 0 is a homotopy equivalence. Hence 0 → (P•) is a homotopy equivalence and we are done. 4.4. The Dold-Kan Correspondence In this section we generalise the Dold-Kan correspondence for abelian groups to exact categories exact categories with enough projectives. If C is a category, we denote by sC the functor category [∆op, C], where ∆ is the usual simplicial category. We use this to show that when E is elementary the projective model structure on Ch(E) and Ch≥0(E) are Kan complex-enriched. Let us recall the Dold-Kan correspondence for abelian categories. The exposition here follows [Wei95] 8.4. For an abelian category A, there are functors

Γ: Ch≥0(A) → sA,N : sA → Ch≥0(A) constructed as follows. To construct N, we first define an ‘unnormalised’, simpler functor C : sA → Ch≥0(A). Given an object A ∈ sA let CA be the complex with (CA)n = An, and differential ∂n :(CA)n → (CA)n−1 given by n X i (−1) di i=0

where the di are the degeneracy maps. One checks straightforwardly that this is well-defined as a chain complex. Now let n−1 \ (NA)n = Ker(di) i=0 n NA ⊂ CA is in fact a subcomplex, and the restricted differential is δn = (−1) dn : NAn → NAn−1. Since by definition a map of simplicial objects commutes with the face maps, the constructions CA and NA are functorial. Moreover we have the following 4.4. THE DOLD-KAN CORRESPONDENCE 73

Proposition 4.4.70 ([GJ09], Theorem III 2.1, 2.4). The natural inclusion N → C is a natural homotopy equivalence, and a split monomorphism.

The construction of Γ is more involved. For a chain complex C ∈ Ch≥0(A), one sets M Γ(C)n = Cη η:[n][p],p≤n where for η :[n] [p], Cη = Cp. Given a morphism α :[n] → [m] in ∆, deﬁne a morphism Γ(C)(α): Γm(C) → Γn(C) by its restriction Γ(α, η): Cη → Γ(C) to each summand Cη as follows. For each surjection η :[n] → [p] we consider its epi-mono factorisation η0 of ηα. α [m] / [n]

η0 η

[q] / [p] 0 If p = q so that ηα = η then we take Γ(α, η) to be the natural identiﬁcation of Cη with the summand Cη0 of Γm. If p = q + 1 and = p, so that the image of ηα is {0, . . . , p − 1}, then we take Γ(α, η) to be the composition d Cη = Cp / Cp−1 = Cη0 / Γm(C) Otherwise we take Γ(α, η) to be 0. The Dold-Kan Correspondence says the following Theorem 4.4.71 (Dold-Kan for Abelian Categories). Let A be an abelian category. Then the functors

Γ: Ch≥0(A) → sA,N : sA → Ch≥0(A) form an equivalence of categories.

Proof. See [Wei95] §8.4. The constructions of Γ and C make sense in any additive category, and N makes sense in any additive category which has kernels. Thus for an additive category E with kernels we get functors

Γ: Ch≥0(E) → sE,N : sE → Ch≥0(E)

constructed mutatis mutandis as above. We also get a functor C : sE → Ch≥0(E), and a natural split inclusion and homotopy equivalence i : N → C. Corollary 4.4.72 (Dold-Kan for Exact Categories). Let E be an elementary exact category. The functors Γ: Ch≥0(E) → sE,N : sE → Ch≥0(E) deﬁned above are weakly inverse to each other. In particular they give equivalences of categories.

Proof. Pick a left abelianisation I : E → A. Then I extends to functors sE → sA and Ch≥0(E) → Ch≥0(A), which we will also denote by I. Since I preserves kernels we get a commutative diagram.

N sA / Ch≥0(A) O O I I

N sE / Ch≥0(E) It is also clear from the construction of Γ that the following diagram commutes

sA o Ch≥0(A) O Γ O I I

sE o Ch≥0(E) Γ 74 4. MODEL STRUCTURES ON EXACT CATEGORIES

Since the functor I is fully faithful, Theorem 4.4.71 implies the result. Remark 4.4.73. This result is actually overkill. It has been pointed out to us by Theo Bühlerthat the Dold-Kan equivalence is valid for any weakly idempotent complete additive category. A proof (which in fact works on the level of quasi-categories) can be found in [Joy08] Section 35. If A = Ab is just the category of abelian groups, then there is a well-known model structure on the category sAb. The weak equivalences (resp. fibrations) are those maps of simplicial abelian groups which are weak equivalences (resp. fibrations) on the underlying simplicial set. As usual, the cofibrations are maps of simplicial abelian groups which have the left lifting property with respect to the trivial fibrations. Moreover, the category Ab is an elementary abelian category. As a set of compact projective generators we can take P = {Z}. Thus there is a projective model structure on Ch≥0(Ab). In this case the functors N and Γ also form a Quillen equivalence between these model categories. For a proof see [GJ09] Chapter 3 Section 2. The model structure on sAb is a special case of a much more general model structure. Notation 4.4.74. (1) Let Z be an object in a category C. We denote by sZ the constant simplicial object in sC which is Z in each degree, and such that the face and degeneracy maps are all idZ . (2) If C is additive, then the category sC is enriched over sAb in an obvious way. We denote the enriched hom functor by HomsC The result below is a special case of Theorem 6.3 in [CH02]. Theorem 4.4.75. Let E be a complete and cocomplete exact category with enough projectives. There is a model structure on sE in which a map f : X → Y is a weak equivalence (respectively fibration) if and only if the induced map HomsE (sP,A) → HomsE (sP,B) is a weak equivalence (respectively fibration) for all projectives P . Theorem 4.4.76 (Model Dold-Kan for Elementary Exact Categories). Let E be a complete and cocomplete exact category with enough projectives. Endow Ch≥0(E) and sE with their projective model structures. Then the functors Γ: Ch≥0(E) → sE,N : sE → Ch≥0(E) form a Quillen equivalence. We use the following notion. Definition 4.4.77. Let M , N be model categories. M is said to be generated by a collection of functors {Fi : M → N }i∈I if a map f : X → Y in M is a fibration (resp. weak equivalence) if and only if Fi(f) is a fibration (resp. weak equivalence) for each i ∈ I. By construction the model structure on sE is generated by the functors

{HomsE (sP, −): sE → sAb}P ∈P where we endow sAb with its projective model structure. The model structure on Ch≥0(E) is generated by a similar set of functors. Proposition 4.4.78. Let E be a weakly idempotent complete exact category with class of projectives P. The projective model structure on Ch≥0(E) is generated by the functors 0 {Hom(S (P ), −): Ch≥0(E) → Ch≥0(Ab): P ∈ P}

where we endow Ch≥0(Ab) with its projective model structure.

Proof. The fibrations in Ch≥0(E) are the degree-wise admissible epics in positive degree, and the fibrations in Ch≥0(Ab) are the degree-wise epics in positive degree. Let f• : X• → Y• be a morphism in 0 Ch≥0(E). Then the components of Hom(S (P ), f•) are HomE (P, fn). Now f• is a fibration if and only if each fn is an admissible epimorphism for n > 0. This is true if and only if HomE (P, fn) is an epic for each 0 n > 0 and each P ∈ P, i.e. if and only if Hom(S (P ), f•) is a fibration for each P ∈ P. 4.4. THE DOLD-KAN CORRESPONDENCE 75

0 ∼ 0 It is clear that Hom(S (P ), cone(f•)) = cone(Hom(S (P ), f•)). Now by Corollary 2.6.95, cone(f•) is 0 acyclic if and only if Hom(S (P ), cone(f•)) is acyclic for all P ∈ P. Equivalently, f• is a weak equivalence 0 if and only if Hom(S (P ), f•) is a weak equivalence for each P ∈ P. With these structures in hand, we will use the following result in order to prove the theorem.

0 0 Proposition 4.4.79. Let M , N , M , N be model categories. Suppose M is generated by functors {Fi : 0 0 0 0 0 0 M → N }i∈I , and M is generated by functors {Fi : M → N }i∈I . Let G : M → M and H : M → M be adjoint functors G a H Suppose also that there is a Quillen adjunction P a Q, with P : N → N 0 and Q : N 0 → N such that for each i ∈ I the diagram M o M 0 H 0 Fi Fi NNo 0 Q commutes. Then G a H is a Quillen adjunction. Proof. We need to show that H preserves (acyclic) fibrations. Let f be an (acyclic) fibration in M 0. 0 0 0 By assumption, for each i, Fi (f) is an (acyclic) fibration in N . Since Q is right Quillen, Q ◦ Fi (f) is an (acyclic) fibration. By commutativity of the diagram M o M 0 H 0 Fi Fi NNo 0 Q

Fi ◦ H(f) is an (acyclic) fibration for each i ∈ I. Again by assumption, H(f) is an (acyclic) fibration. Before proving the theorem, we shall make the following easy observation. Proposition 4.4.80. Let M and M 0 be model categories, and G : M → M 0 and H : M 0 → M be Quillen adjoint functors G a H Suppose further that (1) The unit and counit maps of the adjunction are weak equivalences. (2) G preserves weak equivalences of the form X → HY where X is cofibrant and Y is fibrant. (3) H preserves weak equivalences of the form GX → Y where X is cofibrant and Y is fibrant. Then G a H is a Quillen equivalence. Proof. Let X be a cofibrant object of M and Y a fibrant object of M 0. Suppose that f : GX → Y is a weak equivalence. Then by assumption HGX → HY is a weak equivalence. Also by assumption X → HGX is a weak equivalence. Hence X → HY is a weak equivalence. Conversely suppose that X → HY is a weak equivalence. Then GX → GHY is a weak equivalence by assumption. Also by assumption GHY → Y is a weak equivalence. Thus GX → Y is a weak equivalence. Proof of Theorem 4.4.76. We first note that the following diagrams commute (up to natural isomorphism).

sE o Ch≥0(E) Γ

0 HomsE (sP,−) Hom(S (P ),−) sAb o Ch≥0(Ab) Γ 76 4. MODEL STRUCTURES ON EXACT CATEGORIES

N sE / Ch≥0(E)

0 HomsE (sP,−) Hom(S (P ),−)

N sAb / Ch≥0(Ab) The second diagram follows from the fact that Hom(P, −): E → Ab preserves kernels (and therefore inter- sections). The ﬁrst diagram follows from the fact that Hom(P, −): E → Ab preserves ﬁnite direct sums. By Proposition 4.4.79 the adjunction is a Quillen adjunction. Let us now check the hypotheses of Proposition 4.4.80. The unit and counit maps are isomorphisms. In particular they are weak equivalences. In the Dold-Kan correspondence for abelian groups, it can be shown that the functors N : sAb → Ch≥0(Ab) and Γ: Ch≥0(Ab) → sAb both preserve all weak equivalences. By the commutativity of the above diagrams, this also implies that the functors N : sE → Ch≥0(E) and Γ : Ch≥0(E) → sE also preserve all weak equivalences. Of course if E is a weakly idempotent complete additive category with kernels, and there is a model structure on Ch≥0(E), then via the categorical equivalence Ch≥0(E) → sE there is a transferred model category structure on sE which is Quillen equivalent to the one on Ch≥0(E). The point of the above is that for the projective model structure on Ch≥0(E), the transferred model structure coincides with the one established in Theorem 6.3 of [CH02].

Remark 4.4.81. Let E be a small complete and cocomplete exact category equipped with a small dg≥0- compatible cotorsion pair (L, R). Equip Ch≥0(E) with the resulting model category structure, and sE with the model category structure transferred along the Dold-Kan equivalence Suppose that the cotorsion pair is also dg-compatible. By Proposition 4.2.38 we also have an adjunction

i ◦ N : sE Ch(E):Γ ◦ τ≥0 4.4.1. Exactness of the Dold-Kan Functors and Left Properness. As a diagram category, sE can be equipped with the structure of an exact category, where exact sequences are determined term-wise. With this exact structure we have the following.

Proposition 4.4.82. Both of the functors N : sE → Ch≥0(E) and Γ: Ch≥0(E) → sE are exact. Proof. Γ is clearly exact, and N is a retract of the functor C which is also clearly exact. The following is tautological. Proposition 4.4.83. Let C be a model category, and F : C → D an equivalence. Equip D with the transferred model structure. Let f : X → Y be a left proper map in C. Then F (f) is left-proper in sE. Corollary 4.4.84. Admissible monomorphisms in sE are left-proper. 4.4.2. The Cosimplicial Dold-Kan Correspondence. In this section we discuss a generalisation of the cosimplicial Dold-Kan correspondence of [CC04] to exact categories. If E is a weakly idempotent complete exact category then so is Eop. Thus there is an equivalence

Γc : Ch≤0(E) csE :Nc

between the categories Ch≤0(E) and the category csE of cosimplicial objects. Therefore a model structure on Ch≤0(E) induces a Quillen equivalent model structure on csE. In this case, it is not immediately clear that a map f : X → Y in csE is a weak equivalence if and only if the map Hom(P, f) of cosimplicial abelian groups is an equivalence for each projective P , where Hom(P, −) is applied level-wise. This is because cokernels appear in the construction of Nc, and Hom(P, −) does not necessarily commute with these. Fortunately we still have the following. Proposition 4.4.85. A map f : X → Y in cs(E) is a weak equivalence if and only if for any projective P , Hom(csP, f) is a weak equivalence in csAb. 4.4. THE DOLD-KAN CORRESPONDENCE 77

Proof. Let Cc : csE → Ch≤0(E) be cosimplicial unnormalised Moore complex functor (i.e. the unnormalised Moore complex functor computed in the opposite category). There is a natural homotopy equivalence and split epimorphism Cc → Nc. Thus a map f : X → Y in csE is an equivalence if and only if Cc(f) is an equivalence. Now there is a commutative diagram

Ch≤0(E) o csE Cc Hom(S0(P ),−) Hom(csP,−) Ch≤0(Ab) o csAb Cc and the result follows. 4.4.2.1. An Alternative Cosimplicial Dold-Kan Correspondence. In [CC04] Castiglioni and Corti˜nacon- struct an alternative pair of functors between cosimplicial and non-positively graded complexes. Their construction also works for more general additive categories. Let E be a cocomplete additive category. Then E is tensored over the category Ab of abelian groups. For n ≥ 0, write V n for the abelian group given by the kernel of the canonical map n M Z → Z i=0 Ln n • Let {ei : 0 ≤ i ≤ n} be the standard basis of i=0 Z. Then {vi = ei − e0 : 0 ≤ i ≤ n} is a basis of V . V may be regarded as an object of csAb as follows. For α :[n] → [m] a map, set

αvi = vα(i) − vα(0) and extend to V n by linearity. Let T (V ) denote the free associative monoid in csAb on V . We shall denote the multiplication on this algebra by µ. Write

Q : Ch≤0(E) → csE

for the following functor. Let (A•, d) ∈ Ch≤0(E). Write ∞ M i n (QA)n = A−n ⊗ T (Z ) i=0 For any map α :[n] → [m] of ﬁnite sets, there is a map

(QA)n → (QA)m deﬁned by

IdA ⊗ α + d ⊗ µ(vα0 ⊗ α) It is immediately veriﬁed as in [CC04] Section 3 that this is a well-deﬁned object of csE. Proposition 4.4.86 (Proposition 7.4/ Remark 7.5 [CC04]). The functor

Q : Ch≤0(E) → csE has a right adjoint H. Proof. Of course with some mild assumptions one can prove this using the adjoint functor theorem. As in [CC04] we give an explicit construction. For B ∈ csE write M M M DB = Dn(P ) n≤0 P ∈P Hom(Q(Dn(P )),B) n n For s ∈ Hom(Q(D (P )),B) let js : D (P ) → B be the inclusion. Denote by α : QDB → B the map deﬁned n on the summand Q(js): QD (P ) → B by s. Let K = colimf:I→DB:α◦Q(f)=0I, and set HB = coker(K → I). The functor Q commutes with colimits, so Q(HB) =∼ coker(Q(K) → QDB). In particular α : QDB → B factors through a mapα ˆ : DHB → B. Thus (HB, αˆ) is an object of the arrow category Q ↑ B. To prove 78 4. MODEL STRUCTURES ON EXACT CATEGORIES

that H is a right adjoint is equivalent to proving that it is a ﬁnal object of this category. Now the proof proceeds formally the same way as [CC04] Proposition 7.4. A formally identical proof to [CC04] Theorem 4.2 i) shows that there is a natural homotopy equivalence pˆ : Q → Γc Now let E be a weakly idempotent complete exact category such that the projective model structure exists on Ch≤0(E), and equip Ch≤0(E) and csE with the projective model structures. Lemma 4.4.87. The adjunction Q: Ch≤0(E) csE :H is a Quillen equivalence. Proof. It suﬃces to prove that the adjunction is a Quillen adjunction. Indeed if this is the case then we get an adjunction of homotopy categories

LQ: Ho(Ch≤0(E)) Ho(csE):RH

The functors Q and Γc preserve all equivalences (since Γc does, and Q is homotopy equivalent to it), so we get equivalences of functors of homotopy categories ∼ ∼ ∼ LQ = Q = Γc = LΓc Since Γc : Ch≤0(E) csE :Nc ∼ is a Quillen equivalence, LΓc = LQ is an equivalence of categories. It remains to show that adjunction between Q and H is Quillen. Now we have shown that the functor Q preserves all equivalences. It therefore suffices to prove that it preserves cofibrations. We will show that in fact NcQ(f) is a cofibration for any generating cofibration f : X → Y . By the proof of Theorem 4.2 in [CC04], for X an object of Ch≤0(E), ∼ L∞ (NcQX)−n = −r=n Xr ⊗ Z[Surr,n], where Surr,n is the set of surjections from the set with r elements to the set with n elements. In particular, if X is bounded below then so is NcQA. First consider a cofibration 0 0 0 ∼ 0 of the form S (P ) → 0 with P projective. (NcQS (P ))m = 0 for m < 0, so NcQS (P ) = S (P ), and 0 ∼ 0 −n NcQ(S (P ) → 0) = S (P ) → 0 is a cofibration. Next consider a cofibration of the form S (P ) → −n+1 −n −n+1 D (P ) for n ≥ 1 and P projective. NcQ(S (P ) → D (P )) is degree-wise split, and the cokernel ∼ −n+1 C satisfies C−m = (NcQS (P ))−m. This is a bounded below complex of projectives, and is therefore DG-projective. 4.4.3. The Simplicial Model Structure. In this section we determine conditions under which the model structures on Ch≥0(E) and Ch(E) are simplicial (refer to Section A.4.1 for the terminology in this section). First note that we have the following result, using Proposition A.4.25, Theorem 4.4.76, and Remark 4.4.81.

Proposition 4.4.88. Let E be a small complete and cocomplete exact category equipped with a small dg≥0-compatible cotorsion pair (L, R). Endow Ch≥0(E) and sE with their induced model structures. Let M be a Ch≥0(E)-model category. Then M is a sE-model category. If (L, R) is also dg-compatible, then any Ch(E)-model category is also a sE-model category.

We claim that if E is any complete and cocomplete exact category equipped with a dg∗-compatible cotorsion pair for ∗ ∈ {≥ 0, ∅}, then with the induced exact structure Ch∗(E) is a Ch∗(Ab)-model category. In particular by the above proposition it is a sAb-model category. Since there is a Quillen adjunction Z ⊗ (−): sSet sAb :| − |

this in turn implies by Proposition A.4.25 that Ch∗(E) is a simplicial model category. Now by Prop. 3.46 in [KK82] we have the following.

Proposition 4.4.89. Let E be a complete and cocomplete additive category. Then Ch∗(E) is tensored, cotensored, and enriched over Ch∗(Ab) for ∗ ∈ {≥ 0, ∅} 4.5. THE INJECTIVE MODEL STRUCTURE, THE K-PROJECTIVE MODEL STRUCTURE, AND EXAMPLES 79

Let us denote the tensoring, enrichment, and cotensoring by

⊗ : Ch∗(Ab) × Ch∗(E) → Ch∗(E) op Hom : Ch∗(E) × Ch∗(E) → Ch∗(Ab) (−) op (−) : Ch∗(E) × Ch∗(E) → Ch∗(E)

Proposition 4.4.90. Let E be a complete and cocomplete exact category equipped with a dg∗-compatible cotorsion pair (L, R) for ∗ ∈ {≥ 0, ∅}. Then the two-variable adjunction above is a two-variable Quillen adjunction.

Proof. It suffices to prove that ⊗ : Ch∗(Ab) × Ch∗(E) → Ch∗(E) is a left Quillen bifunctor. Let f : A → B be a cofibration in Ch∗(Ab) and g : X → Y be a cofibration in Ch∗(Ab). We may assume that f is a generating cofibration. First suppose it is a generating acyclic cofibration of the form 0 → Dn(Z). By shifting 1 1 ∼ we may assume that it is of the form 0 → D (Z). Observe that for any complex Z, D (Z) ⊗ Z = cone(IdZ ). The pushout of the diagram

0 / cone(IdX )

is of course just cone(IdX ). So we need to show that the map cone(IdX ) → cone(IdY ) is an acyclic coﬁbration. Let C be the cokernel of g : X → Y . Then the cokernel of cone(IdX ) → cone(IdY ) is cone(IdC ). However this follows from Proposition 4.2.41. Now suppose f is a generating coﬁbration of the form Sn(Z) → Dn+1(Z). Again by shifting we may assume that it is of the form S0(Z) → D1(Z). Note that S0(Z) ⊗ Z = Z for any complex Z. The pushout of the diagram

X / cone(IdX )

g Y

is cone(g). Thus we need to show that the map cone(g) → cone(IdY ) is a cofibration, and an acyclic cofibration when g is acyclic. The map is clearly an admissible monomorphism. Moreover its cokernel is isomorphic to C, which is cofibrant, and trivially cofibrant when g is acyclic. Finally, for the ≥ 0 case, we need cofibrations of the form 0 → Sn(Z). Again we may assume that n = 0, in which case S0(Z) ⊗ (−) is isomorphic to the identity functor. Therefore in this case the pushout-product axiom is obvious.

Corollary 4.4.91. Let E be a complete and cocomplete exact category equipped with a dg≥0-compatible cotorsion pair (L, R). Then Ch≥0(E) is a simplicial model category. In fact it is Kan-complex enriched. If (L, R) is also dg-compatible then this is also true of Ch(E). 4.5. The Injective Model Structure, the K-Projective Model Structure, and Examples 4.5.1. The Injective Model Structure.

Definition 4.5.92. Let E be an exact category. If it exists, the injective model structure on Ch∗(E), for ∗ ∈ {−, b, ∅, ≤ 0, ≥ 0} is the model structure in which • Weak equivalences are quasi-isomorphisms. • Coﬁbrations are degree-wise admissible monomorphisms for ∗ ∈ {−, b, ∅, ≥ 0}, and degree-wise admissible monomorphisms in strictly negative degree for ∗ ∈ {≤ 0}. Proposition 4.5.93. Let E be an exact category. Suppose that both the projective and injective model structures exist on Ch∗(E) for ∗ ∈ {∅, ≥ 0, ≤ 0}. Then the adjunction

Id: Ch∗(E) Ch∗(E):Id is a Quillen equivalence, where the left hand side is equipped with the projective model structure, and the right-hand side with the injective model structure. 80 4. MODEL STRUCTURES ON EXACT CATEGORIES

Proof. Equivalences in both model structures are the same. Moreover for ∗ ∈ {∅, ≥ 0}, cofibrations on the left-hand side are in particular degree-wise admissible monomorphisms. For ∗ ∈ {≤ 0}, cofibrations on the left-hand side are degree-wise admissible in strictly negative degree. Thus the adjunction is Quillen. The induced adjunction on homotopy categories is just the identity adjunction, so is clearly an equivalence. By duality we have the following. Proposition 4.5.94. Let E be a weakly idempotent complete exact category with enough injectives. Then the injective model structure exists on Ch≤0(E). If E has cokernels, countable products of injectives, and injectives satisfy axiom AB4 ∗ −k (the dual of axiom AB4 − k) for some k ∈ Z, then this is also true for Ch(E). Thanks to Lemma 3.3.54, we also have the following. Corollary 4.5.95. If a countably complete and countably cocomplete exact category E has both enough injectives and enough projectives, then both the injective and projective model structures exist on Ch(E). In particular if a quasi-abelian category E has enough injectives then Ch(P ro(E)) is equipped with the injective model structure. This was proven by Pridham for E = BanC in [Pri17]. 4.5.1.1. Exact categories of Grothendieck type. In [St’12ˇ ], St’ov´ıˇcekprovesˇ the following Theorem 4.5.96 ([St’12ˇ ] Theorem 7.11). Let E be an exact category of Grothendieck type such that the class W of acyclic complexes is deconstructible in Ch(E). Then the injective model structure exists on Ch(E). When E has countable products, and such products are exact, we also get the following existence result. Theorem 4.5.97. Let E be an exact category of Grothendieck type. If E has cokernels, countable products of injectives, and injectives satisfy axiom AB4∗−k for some k ∈ Z. Then the injective model structure exists on Ch(E). Proof. The category Eop has enough projectives by Lemma 3.3.50, and countable coproducts of projectives satisfy AB4−k for some k. Thus the projective model structure exists on Ch(Eop), i.e. the injective model structure exists on Ch(E). Corollary 4.5.98. Let E be a complete and cocomplete elementary exact category which is locally presentable. Then both the projective and injective model structures exist on Ch(E), and they are Quillen equivalent. 4.5.2. The K-Projective Model Structure. Throughout this section E is a weakly idempotent complete exact category. In [Gil16b] Gillespie proves a general theorem concerning the existence of exact model structures in which the cofibrant objects are of the form KL (rather than dgfiL). Theorem 4.5.99 ([Gil16b] Theorem 6.3). Let E be a weakly idempotent complete exact category with enough projectives. Let (U, F) be a cotorsion pair in Ch(E) such that (1) (U, F) is complete. (2) U is thick. (3) U ∩ F is the class of projectives in Ch(E). Then there is a model structure on Ch(E) in which (1) the cofibrations are the degree-wise split monomorphisms with cokernel in KU (2) the trivial cofibrations are the degree-wise split monomorphisms with contractible cokernel. (3) the fibrations are the degree-wise split epimorphisms. (4) the trivial fibrations are the degree-wise split epimorphisms with kernel in F. By definition, the K-projective model structure is the model structure of the theorem for the cotorsion pair (dgProjfl(E), W). The following is proven for Grothendieck abelian categories in [Gil16b] Section 6 but the proof is the same. 4.5. THE INJECTIVE MODEL STRUCTURE, THE K-PROJECTIVE MODEL STRUCTURE, AND EXAMPLES 81

Proposition 4.5.100. If X ∈ KProj is acyclic then it is homotopy equivalent to 0. ∼ Proof. Hom(X,X) is acyclic. Thus [IdX ] ∈ Hom(X,X) ∼= H0Hom(X,X) = 0 is homotopy equivalent to the zero map. Suppose the projective model structure exists on Ch(E). In particular the cotorsion pair (dgProjfl(E), W) is complete. Thus, as mentioned in [Gil16b] Corollary 6.6 in the case that E is a Grothendieck abelian category with a projective generator, the K-projective model structure also exists on Ch(E). Moreover, the identity functor is clearly left Quillen from the projective model structure to the K-projective model structure. We claim it is a Quillen equivalence. Let f : X → Y be an equivalence in the K-projective model structure. Then f factors as p ◦ i where i is a degree-wise split monomorphism with contractible cokernel, and p is a degree-wise split epimorphism with acyclic kernel. Thus both p and i, and therefore f, are equivalences in the projective model structure. On the other hand, let f : X → Y be an equivalence in the projective model structure. We factor f as p ◦ i where i is a cofibration in the K-projective model structure, p is a fibration in the K-projective model structure, and p is trivial in the K-projective model structure. In particular it is also trivial in the projective model structure. Thus i is also trivial in the projective model structure. i is a degree-wise split monomorphism with K-projective cokernel C. Since it is trivial in the projective model structure C is acyclic and hence contractible, as required. 4.5.3. Examples. All the examples of Section 3 satisfy the assumptions of Theorem 4.3.58 such that their categories of unbounded complexes have projective model structures. The model structures for Ch(Ind(Bank)), Ch(CBornk), and for unbounded complexes in the contracting normed and Banach categories are monoidal and satisfy the monoid axiom. In fact if any quasi-abelian category E has enough projectives then the projective model structure exists on Ch(Ind(E)). 4.5.3.1. The Derived Category With Respect to a Generator. Let E be a strongly efficient exact category L with a generator G = i∈I Gi. Consider the G-exact structure on E. Then G is a projective generator of E. Moreover since E is locally presentable, G is κ-presented for some ordinal κ. By Proposition 4.3.59 DG-projective resolutions exist. Thus we get the following, which for the case that E is a Grothendieck abelian category is Corollary 4.7 in [Gil16a]. Corollary 4.5.101. Equip E with the G-exact structure. Then the projective model structure exists on Ch(E). Moreover it is combinatorial.

Suppose now that each Gi is tiny. Consider the G-exact structure on E. By construction this category has enough projectives. By Corollary 3.3.55 it is of Grothendieck type, and therefore by Lemma 3.3.54 it has enough injectives. Therefore we get the following, which generalises [Gil16a] Corollary 5.12. Corollary 4.5.102. Both the projective and injective model structures exist on Ch(E) where E is equipped with the G-exact structure. CHAPTER 5

Filtered and Graded Objects in Exact Categories

In [Sch99] Schneiders shows that the category of exhaustively Z-filtered abelian groups is an elementary quasi-abelian category. In particular the category of chain complexes of filtered abelian groups or, equivalently, filtered chain complexes of abelian groups, is equipped with a projective model structure. We generalise this to quasi-abelian and exact categories. ( In [SS16] Schapira and Schneiders study derived categories of Λ-filtered objects in abelian categories, where Λ is a more general filtered category. Similar results to theirs should hold for exact categories, though we shall not deal with that here). We will also relate our results to the work of [GP18] on filtered objects in more general model categories and (∞, 1)-categories, and to the work of [CCN21].We will also study monoidal structures on categories of filtered objects. This is crucial for the Koszul duality result Theorem 4.23[Kel19] where one needs to understand associated graded objects of filtered cooperads. Our considerations of homotopically complete filtered objects will be important for joint work with Kobi Kremnizer and Devarshi Mukherjee on a bornological version of the HKR theorem, following [Rak20].

5.1. Preliminaries Before continuing let us ﬁx some notation.

Notation 5.1.1. (1) If I is a category then we denote by Itop the category obtained by freely adjoining a terminal object top to I, and by I• the category obtained by freely adjoining a discrete element •. (2) Throughout this section Z will denote the set of integers regarded as a discrete category, and Z< the poset of integers with increasing ordering. 5.1.1. Restricted Diagram Categories. Let E be a category, I be a small category, and S a class of morphisms in E. We denote by FunS (I, E) the full subcategory of Fun(I, E) consisting of functors F such that for any α : i → j in I, F (α) is in S. It will be convenient to have a description of limits and colimits in such categories. Let D : J → FunS (I, E) be a diagram. For each object j ∈ J and i ∈ I write Di(j) for the 0 functor D(j) evaluated at i, and for α : i → i in I write Dα(j) for the corresponding map. For each i ∈ I this gives a functor Di : J → E.

Proposition 5.1.2. Suppose that for each morphism α in I the map (co)limJ Dα : (co)limJ Di → (co)limJ Di0 is in S. Then the functor i 7→ (co)limJ Di(j) is a (co)limit of the diagram D in FunS (I, E). Proof. The universal property is checked directly.

5.2. Graded Objects in General Categories Before specialising to exact categories, we will discuss graded and ﬁltered objects in arbitrary categories. Let us ﬁx a complete and cocomplete category E. 5.2.1. Graded Objects.

Definition 5.2.3. The category of Z-graded objects in E, denoted Gr(E), is the diagram category Fun(Z, E). ` We will write an object of ( ) as An. The following is clear. Gr E n∈Z 82 5.3. FILTERED OBJECTS IN GENERAL CATEGORIES 83

Proposition 5.2.4. The category Gr(E) is complete and cocomplete. For n ∈ Z we denote by Sn : E → Gr(E) the functor sending an object X to the graded object with n n (S (X))n = X and (S (X))m = 0 for m 6= n. 5.2.2. Pointed Graded Objects. It will also be convenient to consider the category of pointed graded objects.

Definition 5.2.5. The category of pointed graded objects in E, denoted Gr •(E), is the category of functors Fun(Z•, E). ` We will write an object of Gr •(E) as ( n∈ An,A•). Again Gr •(E) is complete and cocomplete. Functors n Z S : E → Gr •(E) for n ∈ Z or n = • are deﬁned in the obvious way. 5.2.3. Monoidal Structures on Graded Objects. Suppose that (E, ⊗, k) is a (symmetric) monoidal ` ` category. The induced (symmetric) monoidal structure on Gr(E) is easy to deﬁne. If i Ai and j Bj are graded objects then their monoidal product is

a a . a a ( Ai) ⊗ ( Bj) .= Ai ⊗ Bj i j n i+j=n

It is functorial in the obvious way. Gr •(E) can also naturally be given the structure of a (symmetric) monoidal category by deﬁning

a a . a a ( An,A•) ⊗ ( Bn,B•) .= ( Ai ⊗ Bj,A• ⊗ B•) n∈Z n∈Z n i+j=n 5.2.4. Model Structures on Graded Objects. Let E be a combinatorial model category. Say that a map in f = (fn) in Gr(E) (resp. Gr •(E)) is a fibration/ cofibration/ weak equivalence if for each n ∈ Z (resp. n ∈ Z•), fn is a fibration/ cofibration/ weak equivalence in E. Since everything is computed component-wise in categories of graded objects it is clear that this determines a combinatorial model structure on Gr(E) (resp. Gr •(E)). Note that if I (resp. J) is a set of generating cofibrations (resp. generating acyclic cofibrations) for E then {Sn(f): f ∈ I, n ∈ Z} (resp. {Sn(f): f ∈ J, n ∈ Z}) is a set of generating cofibrations (resp. generating acyclic cofibrations) for Gr(E), and similarly for Gr •(E) by also taking n = •. If E is a monoidal model category then, again since everything is computed component-wise, both Gr(E) and Gr •(E) are monoidal model categories.

5.3. Filtered Objects in General Categories In this section we again fix a complete and cocomplete pointed category E. The initial/ terminal object will be denoted 0. Definition 5.3.6. Let S be a class of maps in E. < (1) The category of S-filtered objects, denoted Filt S (E), is the functor category FunS (Ztop, E). < (2) The category of exhaustively S-filtered objects, denoted Filt S (E), is the functor category FunS (Z , E).

We shall write a filtered object as a tuple (Atop, αi, ai), where Atop is the value of the diagram at the terminal object, and αi : Ai → Atop, ai : Ai → Ai+1 are maps in S. A map (Atop, αi, ai) → (Btop, βi, bi) will be written as (gtop, gi). We shall write the data of an exhaustively filtered object as (ai : Ai → Ai+1), or (ai) when the Ai are understood, and a map of exhaustively filtered objects as (gi : Ai → Bi).

Remark 5.3.7. The category Filt S (E) is equivalent to the full subcategory of Filt S (E) consisting of objects (Atop, αi, ai) such that Atop together with the maps αi : Ai → Atop is a direct limit of the diagram

a−1 a0 A1 ... / A−1 / A0 / A1 / A2 / ...

We will freely identify Filt S (E) with this subcategory. We will typically only be interested in classes S satisfying the properties below. 84 5. FILTERED AND GRADED OBJECTS IN EXACT CATEGORIES

Definition 5.3.8. Let E be a pointed category. A subclass S ⊂ Mor(E) is said to be a filtering class if (1) S contains all identity morphisms. (2) S is closed under coproducts. (3) S contains all monomorphisms of the form X → X ` Y for X,Y ∈ E. If λ is an ordinal then S is said to be λ-closed if λ-transfinite compositions of maps in S are in S, and is said to be left-cancellable if whenever g ◦ f is in S and g ∈ S, then f ∈ S. We will mainly be interested in the cases S = Mor(E), or S is the class of regular monomorphisms when E is additive, which we denote by RegMon. Recall that in an additive category E a morphism is said to be a regular monomorphism if it is the kernel of a morphism. When E is exact, we will be interested in S = AdMon. It is useful to define the S-image of a map. Definition 5.3.9. Let S be a class of maps, and let f : X → Y be a map in E. The S-image of f, if it exists, is a factorisation of f

i fe X / X‹ / Y with fe ∈ S, such that whenever 0 i0 fe X / Xf0 / Y is a factorisation of f with fe0 in S, there is a unique map g : X‹ → X‹0 such that the diagram below commutes

i fe X / X‹ / Y

IdX g IdY 0 i0 fe X / Xf0 / Y

E is said to have functorial S-images if there is a functor ImS : Mor(E) → E and natural transformations domain → ImS → codomain such that for any map f : X → Y , X → ImS → Y is an S-image of f. Definition 5.3.10. A map f : X → Y is said to be S-epic if for any g : Y → Z, all maps in the diagram ImS (A → X → Y ) → ImS (ImS (A → X) → Y ) → ImS (X → Y ) .are isomorphisms Example 5.3.11. (1) If S = Mor(E) and f : X → Y is a map, then the S-image is the factorisation

Id f X X / X / Y S-epic maps are isomorphisms. (2) If E is additive, S = RegMon and f : X → Y is a map, then the S-image is the factorisation

i fe X / Im(f) / Y where here Im(f) → Y is the usual image, namely Ker(Coker(f)) → Y . S-epic maps are categorical epimorphisms. We have a shift functor for ﬁltered objects.

Notation 5.3.12 ([CCN21], Section 2.1). For r ∈ Z denote by < r >: Filt S (E) → Filt S (E) the functor which sends (Atop, αi, ai) to the object A < r >= (Atop, αi+r, ai+r). Note that this restricts to a functor < r >: Filt S (E) → Filt S (E).

Notation 5.3.13. Following [GP18] we denote by Seq(E) the category Filt Mor(E)(E), and by Seqtop(E) the category Filt Mor(E)(E). 5.3. FILTERED OBJECTS IN GENERAL CATEGORIES 85

There is an obvious forgetful functor R : Filt S (E) → Filt S (E) induced by the restriction functor < < Fun(Ztop, E) → Fun(Z , E). In nice circumstances this functor has a left adjoint which realises Filt S (E) as a coreflective subcategory of Filt S (E). Let A be an exhaustively filtered object, and consider the filtered object (Atop, αi, ai) defined as follows. Atop = lim→n An. ai = Ai → Ai+1 and αi is the canonical map

Ai → lim→n An. This construction is naturally functorial, and we easily get the following result.

Proposition 5.3.14. Suppose that S is ℵ0-closed. Then the functor C : Filt S (E) → Filt S (E) is right adjoint to the forgetful functor R : Filt S (E) → Filt S (E) . Moreover the unit Id → R ◦ C is a natural isomorphism. Let S and R be filtering classes with S ⊂ R, such that S is left-cancellable and functorial S-images exist. Denote by IS;R : Filt R(E) → Filt S (E) the functor defined as follows. For (Atop, αi, ai) an object in Filt R(E), let αi denote the map ImS (αi) → Atop, and let ai denote the induced map ImS (αi) → ImS (αi+1). This is a well-defined object of Filt S (E). Moreover if RS;R : Filt S (E) → Filt R(E) is the obvious inclusion functor, then there is a natural isomorphism IS;R ◦ RS;R → Id. Moreover if R is ℵ0-closed, this restricts to a well defined functor IS;R : Filt R(E) → Filt S (E). In fact these functors often form an adjunction. Proposition 5.3.15. Let S and R be filtering classes with S ⊂ R. Suppose that functorial S-images exist, and S is left-cancellable. Then there is an adjunction

IS;R : Filt R(E) Filt S (E):RS;R where RS;R is the obvious inclusion functor. Moreover if R and S are ℵ0-closed this restricts to an adjunction

IS;R : Filt R(E) Filt S (E):RS;R

5.3.1. Complete and Bounded Below Objects. Let A = (Atop, αi, ai) be a ﬁltered object. We deﬁne Ae to be the projective limit of the diagram. ... / Atop A−1 / Atop A0 / Atop A1 / Atop A2 / ...

. Define Aen .= Ker(Aetop → Atop An). This construction is functorial, and there is a natural transformation A → Ae. As in, e.g. [CCN21] 2.1, we define complete filtered objects.

Definition 5.3.16. A filtered object (Atop, αi, ai) is said to be complete if A → Ae is an isomorphism of filtered objects. The full subcategory of Filt S (E) on objects equipped with a complete filtration will be denoted V V by Filt S (E). The full subcategory of exhaustive and complete objects will be denoted Filt S (E) We generally will not be interested in complete objects which are not exhaustively filtered. The inclusion V functor J : Filt S (E) → Filt S (E) does not in general have a left adjoint (problems arise because of the appearance of both limits and colimits in the definition of complete objects). When dealing with model categories, this is one of the reasons why it is better to work with (homotopically) complete filtered objects, as studied in detail in [GP18]. For certain exact categories we shall show that one can work with complete objects.

Definition 5.3.17. A ﬁltered object (Atop, αi, ai) is said to be bounded below if for suﬃciently large i > 0 A−i = 0. The full subcategory of Filt S (E) (resp. Filt S (E)) consisting of bounded below objects is + + + + denoted Filt S (E) (resp. Filt S (E)). The full subcategory of Filt S (E) (resp. Filt S (E)) consisting of objects N0 N0 (Atop, αi, ai) such that Ai = 0 for i < 0 is denoted Filt S (E) (resp. Filt S (E)).

V + + V Note that Filt S (E) (resp. Filt S (E)) is a full subcategory of Filt S (E) (resp. Filt S (E)). Moreover the N0 N0 category Filt S (E) (resp. Filt S (E)) is in fact a coreﬂective subcategory of Filt S (E) (resp. Filt S (E)). The N0 N0 right adjoint to the inclusion sends an object A = (Atop, αi, ai) to the object A with Ai = 0 for i < 0, N0 and Ai = Ai for i ≥ 0 and i = top. 86 5. FILTERED AND GRADED OBJECTS IN EXACT CATEGORIES

< 5.3.2. Natural Functors. For each l ∈ Ztop we denote by (−)l the functor

Filt S (E) → E which sends a ﬁltered object

(Atop, αi, ai)

to Al. It sends a morphism (ftop, fi):(Atop, αi, ai) → (Btop, βi, bi) to fl. For i ∈ Z we denote by Qi : Filt S (E) → E the functor deﬁned on objects by

Qi(Atop, αi, ai) = coker(αi : Ai → Atop)

It is defined on morphisms in the obvious way. Finally we denote by Fi : E → Filt S (E) the functor which sends an object A of E to the following filtered object. (Fi(A))j is 0 for j < i,(Fi(A))top = A, and (Fi(A))j = A for i ≤ j < ∞, with the structure maps being the obvious ones. Again it is defined on morphisms in the obvious way. Proposition 5.3.18. There are adjunctions

Qi a Fi+1 a (−)i+1.

Proof. Let us first prove the second adjunction. Fix an object A of E, and a filtered object (Btop, βi, bi). Let f : A → (B)i+1 be a map in E. There is an induced map fe : Fi+1A → B defined as follows. fej = 0 for j < i + 1 and fej is the composition A → Bi+1 → Bj for i + 1 ≤ j < ∞. fetop is given by the composition βi+1 ◦ f. This gives a map

HomE (A, (B)i+1) → HomFilt(E)(Fi+1A, B) It is straightforward to verify that it is natural in both A and B. It is clearly an isomorphism of abelian groups. Let us now show the first adjunction. Let (Btop, βi, bi) be a filtered object, and let f : coker(βi : Bi → Btop) → A be a morphism in E. There is an induced map fe :(Btop, βi, bi) → Fi+1A defined as follows. fej is 0 for j < i + 1, and for i + 1 ≤ j < ∞ or j = top, fej is given by the composition Bj → Btop → Btop Bi+1 → A This gives a homomorphism of abelian groups

HomE (Qi(Btop, βi, bi),A) → HomFilt(E)((Btop, βi, bi),Fi+1A)

which is clearly natural in (Btop, βi, bi) and A. It is also clearly an isomorphism. Note that these functors are all also well-defined at the level of exhaustively filtered objects, at least for i 6= top. ` There is a functor filttop : Gr •(E) → Filt S (E). It sends a pointed graded object ( j∈ Ej,Etop) to the ` ` ` ` ` Z filtered object ( Ej, Ek → Ej, Ek → El). It acts on morphisms in the obvious j∈Z• k≤i j∈Z• k≤i l≤i+1 way. There is also a functor grtop : Filt S (E) → Gr •(E), called the associated graded functor defined as follows. To a filtered object A = (Atop, αi, ai) it assigns the pointed object grtop(A) with grtop(A)i = coker(ai−1 : Ai−1 → Ai) for i ∈ Z and grtop(A)• = Atop. Again it acts on morphisms in the obvious way. The functor grtop is neither left nor right adjoint adjoint to filttop. However we have the following. P Proposition 5.3.19. The functor filt : Gr (E) → Filt S (E) is left-adjoint to the functor (−)n top top n∈Z• given by the composition

` (−)n ∆ ` n∈Z• ` ∼ Filt S (E) / Filt S (E) / E = Gr (E) n∈Z• n∈Z• • Here ∆ is the diagonal morphism. P Proof. The functor (−)n is a composition of right-adjoints. Computing the composition of the n∈Z• corresponding left adjoints gives ﬁlttop. 5.3. FILTERED OBJECTS IN GENERAL CATEGORIES 87

` We denote by filt : Gr(E) → Filt S (E) the functor which sends a graded object j∈ Ej to the exhaustively ` ` P Z filtered object ( Ek → El). This functor also has a right adjoint (−)n defined by the k≤i l≤i+1 n∈Z composition ` ∆ n∈ (−)n ( ) / ` ( ) Z / ` =∼ ( ) Filt S E n∈Z Filt S E n∈Z E Gr E

Finally we define gr : Filt S (E) → Gr(E) to be the functor sending A = (Atop, αi, ai) to the graded object . gr(A)i .= coker(ai−1 : Ai−1 → Ai). Note that gr ◦ filt is equivalent to the identity functor. 5.3.3. Limits and Colimits of Filtered Objects. We can use Proposition 5.1.2 to analyse limits and colimits of filtered objects. The first easy result to note is the following.

Proposition 5.3.20. Let E be a complete and cocomplete category. Then Seqtop(E) and Seq(E) are complete and cocomplete. We also have the following important result.

Proposition 5.3.21. Let A = (A∞, αi, ai) be an object of Filt S (E). Suppose that each Ai and Atop satisfy one of the smallness conditions of Definition 2.6.96, S is closed under the corresponding colimits, and for sufficiently large |i|, ai is an isomorphism. Then A satisfies the same smallness condition in Filt S (E). This also holds for the exhaustively filtered category.

Proof. We prove the claim for non-exhaustively filtered objects. Let D : I → Filt S (E) be a relevant filtered diagram. By Proposition 5.1.2 the colimit is computed by taking the colimit in each degree of the filtration. For each k ∈ Z, there is an ik ∈ I such that Ak → colim(−)k ◦D factors through (−)k(ik). Let n ∈

N0 be such that An → An+i, and A−n−i → A−n is an isomorphism for any i ∈ N0. Let i = max−n≤k≤niik . Then the map A → colimD factors through D(i).

Note that if S is ℵ0-closed then, since Filt S (E) is a coreﬂective subcategory of Filt S (E), the former will be complete and cocomplete as long as the latter is. Moreover in this case, colimits computed in Filt S (E) coincide with colimits computed in Filt S (E). For certain limits the computations in the exhaustive and non-exhaustive categories also coincide. Indeed the following is essentially tautological.

Proposition 5.3.22. Let D : J → Filt(E)S . If each D(j) is exhaustively ﬁltered, and (co)limits of diagrams of shape J commute with ℵ0-transﬁnite compositions of morphisms in S, then the formula in Proposition 5.3.21 is also a (co)limit in Filt S (E).

In the situation that we get and adjunction IS;Mor(E) : Seqtop(E) Filt S (E):RS;Mor(E), then the right hand side is a coreflective subcategory of Seqtop. In particular we get the following. Corollary 5.3.23. Let S be a filtering class. Suppose that functorial S-images exist , and S is left- cancellable. Then Filt S (E) is complete and complete. If S is ℵ0-closed this is also true for Filt S (E) 5.3.4. Monoidal Structures on Filtered Objects. For this subsection let (E, ⊗, k) be a monoidal category such that ⊗ commutes with countable colimits in each variable, and let S be left-cancellable. Sup- pose further that functorial S-images exist. For S-filtered objects A = (Atop, αi, ai) and B = (Btop, βi, bi), define a S-filtered object A ⊗ B. Define (A ⊗ B)top := Atop ⊗ Btop. For each n ∈ Z consider the cat- < egory {(i, j) ∈ Z : i + j ≤ n}, and the diagram on this category sending (i, j) to Ai ⊗ Bj. Write . (A⊗˜ B)n .= colim(i,j):i+j≤nAi ⊗ Bj. This gives a well defined object of Seq(E), (Atop ⊗ Btop, (A⊗˜ B)n → (A⊗˜ B)n+1, (A⊗˜ B)n → Atop ⊗ Btop). Notice that this is just Day convolution, as in [GP18] Section 2.23. . ˜ Define A ⊗ B = IS;Mor(E)(A⊗B).

Proposition 5.3.24. If S is left-cancellable then (Atop ⊗ Btop, (α ⊗ β)n, (a ⊗ b)n) is a S-ﬁltered object. Suppose that (1) For any sequence A0 → A1 → ... 88 5. FILTERED AND GRADED OBJECTS IN EXACT CATEGORIES

together with a map lim→n An → B, the map

lim→n ImS (An → B) → ImS (lim→n An → B) is an isomorphism. (2) ⊗ preserves colimits in each variable. If A and B are exhaustive, then A ⊗ B is exhaustive.

Proof. The ﬁrst claim is clear. The maps (A ⊗ B)n → ImS (An ⊗ Bn → Atop ⊗ Btop) and ImS (Ai ⊗ Bn−i → Atop ⊗ Btop) → (A ⊗ B)n furnish an isomorphism (and its inverse) ∼ lim→i lim→j ImS (Ai ⊗ Bj → Atop ⊗ Btop) = (A ⊗ B)n We then have. ∼ lim→i lim→j ImS (Ai ⊗ Bj → Atop ⊗ Btop) = ImS (lim→i lim→j Ai ⊗ Bj → Atop ⊗ Btop) ∼ = ImS (Atop ⊗ Btop → Atop ⊗ Btop)

= Atop ⊗ Btop

The monoidal functor ⊗ : Filt S (E) ⊗ Filt S (E) → Filt S (E) has a unit, namely F0(k). Moreover it is naturally symmetric. However it is not associative in general. It will be useful to consider monoidal subcategories of Filt S (E) wherein the restriction of the monoidal product is naturally associative. One such category is the essential image of ﬁlttop : Gr •(E) → Filt S (E). Indeed this follows from the fact that this functor is strong monoidal.

Proposition 5.3.25. Let A be a full subcategory of Filt S (E) which is closed under ⊗, contains k,and ˜ ˜ ˜ ˜ such that for any A, B, C ∈ A, the maps IS;Mor(E)((A⊗B)⊗C) → (A ⊗ B) ⊗ C and IS;Mor(E)(A⊗(B⊗C)) → A ⊗ (B ⊗ C) are isomorphisms. Then the restriction of ⊗ to A is naturally associative.

Proof. The conditions on A ensure that ((A ⊗ B) ⊗ C)n can be computed as ImS (colimi+j+k≤n(Ai ⊗ Bj) ⊗ Ck → (Atop ⊗ Btop) ⊗ Ctop). This is clearly associative. Proposition 5.3.26. Let S be a ﬁltering class such that

(1) for any map A → B, A → ImS (A → B) is a S-epimorphism (2) ⊗ preserves S-epimorphisms and countable colimits in each variable

Then Filt S (E) is associative.

Proof. Again the assumptions imply that ((A⊗B)⊗C)n can be computed as ImS (colimi+j+k≤n(Ai ⊗ Bj) ⊗ Ck → (Atop ⊗ Btop) ⊗ Ctop). This is clearly associative. 5.3.4.1. Closed Monoidal Structures. Suppose that E is closed monoidal, with internal hom Hom, and let S be a filtering class. Then Filt S (E) is enriched over E, with Homtot(A, B) defined to be the equaliser of the two obvious maps Q Q , n∈ Hom(An,Bn) n∈ Hom(An,Bn+1) Z• 1 Z• As in [CCN21] Section 2 (which considers filtered vector spaces ), it is also enriched over Seqtop(E) by defining Homfilt(A, B) to be the filtered object

(Hom(Atop,Btop), Homtot(A, B < i >) → Hom(Atop,Btop), Homtot(A, B < i >) → Homtot(A, B < i+1 >))

If S = Mor(E), i.e. Filt S (E) = Seqtop(E) then

(Seqtop(E), ⊗, Homfilt) is a closed monoidal category, as in [GP18] 2.23. Moreover

(Seq(E),R(C(−) ⊗ C(−)), RHomfilt(C(−),C(−))) 5.3. FILTERED OBJECTS IN GENERAL CATEGORIES 89

is also a closed monoidal category.

Let S be left-cancellable, and suppose that functorial S-images exist. The inclusion RS;Mor(E) : Filt S (E) → Seqtop(E) has a left adjoint IS;Mor(E). If it happens that Homfilt(A, B) is an object of Filt S (E) whenever A ∈ Seqtop(E) and B ∈ Filt S (E), then by abstract nonsense

(Filt S (E), ⊗,F0(k), Homfilt)

is a closed monoidal category. Finally if A ⊗ B is exhaustive whenever A and B are, and S is ℵ0-closed, then (Filt S (E),R(− ⊗ −),F0(k),R(Homfilt)) is also a closed monoidal category.

5.3.5. Model Structures on Filtered Objects. Let E be a pointed combinatorial model category.

Definition 5.3.27. The ﬁltered model structure on Filt S (E) (resp. Filt S (E)), if it exists, is the model structure transferred along the adjunction X X filttop : Gr •(E) Filt S (E): (−)n (resp. filt: Gr(E) Filt S (E): (−)n) n∈Z• n∈Z

Note that if I (resp. J) is a generating set of cofibrations (resp. trivial cofibrations) then {Fig : g ∈ I, i ∈ Z} (resp. {Fig : g ∈ J, i ∈ Z}) is a generating set of cofibrations (resp. trivial cofibrations) for the filtered model structure on Filt S (E). For Filt S (E) we have to add Ftopg for g ∈ I (resp. g ∈ J). As in [GP18] Section 3, we have the following.

Lemma 5.3.28. The ﬁltered model structure exists on Seqtop(E), and Seq(E)

Proof. The ﬁltered model structure on Seqtop(E) (resp. Seq(E)) is just the projective model structure for functors ( [Rie14] Theorem 12.3.2).

Remark 5.3.29. Let S be a filtering class such that Filt S (E) is a (functorial) model category with the filtered model structure. Suppose that all cofibrations in E are contained in S. Because S contains cofibrations, the cofibrant objects in Filt S (E) coincide with the cofibrant objects in Seqtop(E) (as a consequence of the description of cofibrant objects in the projective model structure for functors). By [MG15] Lemma 2.8

the (∞, 1)-categories presented by Seqtop(E) and Filt S (E) are equivalent. This is also true for the exhaustively filtered categories. Note that in fact since an object A of Filt S (E) is fibrant if and only if each Ai is fibrant top for each i ∈ Z , Seqtop(E) and Filt S (E) in fact have the same fibrant-cofibrant objects. Often we get a Quillen equivalence. Proposition 5.3.30. Let S and R be filtering monomorphic classes with S ⊂ R such that the filtering model structure exists on both Filt R(E) and Filt S (E). Suppose that functorial S-images exist, that S is left- cancellable, that all cofibrations are maps in S, and that both S and R are ℵ0-closed.. Then the adjunctions

IS;R : Filt R(E) Filt S (E):RS;R

IS;R : Filt R(E) Filt S (E):RS;R are Quillen equivalences. Proof. We prove this for the exhaustive case, the case for non-exhaustive filtered objects being similar. Denote by ΣS (−) (resp. ΣR (−) ) the functor Σ (−) : ( ) → ( ) resp. (Σ (−) : n∈Z n n∈Z n n∈Z n Filt S E Gr E n∈Z n ( ) → ( )). The fact that the adjunction is Quillen follows from the fact ΣR (−) ◦ R = Filt R E Gr E n∈Z n S;R ΣS (−) . Clearly a map f in Ch( ( )) is a weak equivalence if and only if R (f) is a weak equivalence. n∈Z n Filt S E S;R Moreover any cofibrant X in Ch(Filt R(E) is in the image of RS;R(f). Let X → RS;R(Y ) be a map with X ∼ cofibrant and Y fibrant. Since X is cofibrant, X = RS;RIS;R(X). Hence X → RS;R(Y ) is an equivalence if and only if IS;R(X) → Y is. 90 5. FILTERED AND GRADED OBJECTS IN EXACT CATEGORIES

5.3.5.1. Homotopically Complete Objects. Let S be a ﬁltering monomorphic class and E a combinatorial left proper model category.

Definition 5.3.31. A coﬁbrant object (Atop, αi, ai) in either Filt S (E) or Filt S (E) is said to be homotopically complete if the map Atop → holim←n (Atop An) is an equivalence, where here holim←n denotes the homotopy limit. Note that in general homotopically complete objects need not be complete, and complete objects need not be homotopically complete. Any bounded below ﬁltered object is both complete and homotopically complete.

Definition 5.3.32 ([GP18] Definition 3.6). Suppose the filtering model structure exists on Filt S (E).A map f : X → Y is said to be a derived graded equivalence if gr(Q(f)) is a graded equivalence, where Q is the cofibrant replacement functo Definition 5.3.33 ([GP18] Definition 3.5). The homotopically complete filtered model structure on Filt S (E), if it exists, is the left Bousfield localisation of the filtered model structure at the derived graded equivalences. Remark 5.3.34. Let S be a filtering class containing cofibrations such that the filtering model structure exists on Filt S (E). Suppose the homotopically complete model structure exists on both Filt S (E) and Seq(E). As left Bousfield localisation does not change the cofibrant objects, with the homotopically complete model structures, Filt S (E) and Seq(E) still have the same cofibrant objects. Moreover a map f : X → Y between such cofibrant objects is an equivalence in either Filt S (E) or Seq(E) precisely if gr(f) is an equivalence. Hence, once again, they present the same (∞, 1)-categories. Again we often have a Quillen equivalence. Proposition 5.3.35. Let S and R be filtering monomorphic classes with S ⊂ R such that the filtering and homotopically complete model structures exists on both Filt R(E) and Filt S (E). Suppose further that cofibrations are contained in S, that S is left-cancellable, and that functorial S-images exist.. Then the adjunction IS;R : Filt R(E) Filt S (E):RS;R is a Quillen equivalence where both sides are equipped with the homotopically complete model structure.

Proof. The fact that cofibrations are contained in S means that Filt R(E) and Filt S (E) have the same ∼ cofibrant objects. In particular for X cofibrant in Filt R(E), we have gr(X) = gr(IS;R(f)). Thus a map f : X → Y between cofibrant objects in Filt R(E) is a graded equivalence if and only if IS;R(f) is a graded equivalence. Thus the adjunction on the Bousfield localisations at derived graded equivalences is a Quillen equivalence. Theorem 5.3.36 ([GP18], Proposition 3.31). Let E be a stable, combinatorial, left proper model category. The homotopically complete filtered model structure exists on Seq(E). Moreover in this case the model structure is the left Bousfield localisation at maps of the form const(0) → const(K), where K is a cofibrant generator of E, and const(0) is the constant filtered object with const(K)top = const(K)i = K for all i ∈ Z, with the maps ai and αi being the identity. 5.3.5.2. Monoidal Model Structures. Suppose that (E, ⊗, k) is a monoidal model category. Let S be a filtering class such that the filtering model structure exists on Filt S (E) (resp. Filt S (E)) and is combinatorial. Suppose further that the model structure on E is monoidal, and that the filtered monoidal structure on Filt S (E) (resp. Filt S (E)) is associative. It follows from the fact that the functor filttop (resp. filt) is

strong monoidal, and Gr •(E) (resp. Gr(E)) is a monoidal model category, that Filt S (E) (resp. Filt S (E)) is a monoidal model category.

5.4. Filtered and Graded Objects in Exact Categories In this section we ﬁx a complete and cocomplete exact category E. 5.4. FILTERED AND GRADED OBJECTS IN EXACT CATEGORIES 91

5.4.1. Graded and Pointed Graded Objects. As diagram categories, the categories Gr(E) and ` ` Gr •(E) have natural exact structures, in which a map f : n X → n Y is an admissible epimorphism/ admissible monomorphism if for each n ∈ Z (resp. n ∈ Z• is an admissible epimorphism/ admissible monomorphism.

5.4.1.1. Compatible Model Structures on Graded Objects. For a class of objects O in E we deﬁne by ` Gr(O) the class of objects in Gr(E) of the form i∈I Ai where each Ai ∈ O. Since exactness is degree-wise the following is clear. Proposition 5.4.37. Let E be an exact category and let (L, R) be a cotorsion pair on E. Then (Gr(L), Gr(R)) is a cotorsion pair on Gr(E). Moreover if (L, R) is dg∗-compatible then so is (Gr(L), Gr(R)). Finally if (L, R) is monoidally dg∗-compatible then so is (Gr(L), Gr(R)).

The model structure induced on Ch∗(Gr(E)) is the component-wise one of Section 5.2.4. One can of course repeat all this for Gr •(E). 5.4.2. Filtered Objects in Exact Categories. In this section we consider categories of the form Filt S (E) where E is an exact category and S a ﬁltering monomorphic class. We will typically be interested in the cases S = RegMon and S = AdMon. Since regular monomorphisms in a quasi-abelian category are left-cancellable, we have the following useful result.

Proposition 5.4.38. Let E be a complete and cocomplete weakly (ℵ0; RegMon)-elementary exact category whose underlying additive category is quasi-abelian, and such that RegMon is ℵ0-closed. The pairs of functors below are adjunction

IMor(E);RegMon : Seqtop(E) Filt RegMon(E):RMor(E);RegMon

IMor(E);RegMon : Seq(E) Filt RegMon(E):RMor(E);RegMon

is an adjunction. In particular, in this case Filt RegMon(E) and Filt RegMon(E) are complete and cocomplete. 5.4.2.1. Exactness Properties of Categories of Filtered Objects. We shall see that in general categories of filtered objects do not have exact structures. However we can still define the notion of an exact sequence of filtered objects. Proposition 5.4.39. Let f g 0 / (Atop, αi, ai) / (Btop, βi, bi) / (Ctop, γi, ci) / 0

be a null sequence in Filt S (E). It is a kernel-cokernel pair if for each i ∈ Z and i = top.

fi gi 0 / Ai / Bi / Ci / 0 is a kernel-cokernel pair. If E is a abelian then the converse is true. Proof. By Proposition 5.1.2 if for each i ∈ Z and i = top.

fi gi 0 / Ai / Bi / Ci / 0 is a kernel-cokernel pair then

f g 0 / (Atop, αi, ai) / (Btop, βi, bi) / (Ctop, γi, ci) / 0 is a kernel-cokernel pair. Now suppose that E is abelian, and that f g 0 / (Atop, αi, ai) / (Btop, βi, bi) / (Ctop, γi, ci) / 0 is a kernel-cokernel pair. Then Ai → Bi is a kernel of Bi → Ci. Moreover Ci = Im(Bi → Btop Atop). Therefore Bi → Ci is an epimorphism, and we are done. 92 5. FILTERED AND GRADED OBJECTS IN EXACT CATEGORIES

Definition 5.4.40. We say that a null sequence

0 / (Atop, αi, ai) / (Btop, βi, bi) / (Ctop, γi, ci) / 0

in Filt S (E) is exact if for each i ∈ Z and i = top the null sequence

0 / Ai / Bi / Ci / 0

is an exact sequence in E. A null-sequence as above in Filt S (E) is said to be exact if each sequence 0 → Ai → Bi → Ci → 0 is exact for all i ∈ Z.

Definition 5.4.41. (1) A map f : A → B in Filt S (E) or is said to be an admissible monomorphism if there is an exact sequence

f 0 / A / B / C / 0

in Filt S (E). (2) A map g : B → C in Filt S (E) is said to be an admissible epimorphism if there is an exact sequence g 0 / A / B / C / 0

in Filt S (E). Likewise one deﬁnes admissible monomorphisms and admissible epimorphisms in Filt S (E).

Proposition 5.4.42. Let E be a complete and cocomplete exact category. Then Seqtop(E) and Seq(E) are complete and cocomplete exact categories with the classes of admissible monomorphisms and admissible epimorphisms deﬁned above. The classes of admissible monomorphisms and epimorphisms are clearly stable under composition, and contain isomorphisms. We claim that admissible epimorphisms are stable under pullback. Proposition 5.4.43. Let Y / X

f g Y‹ / X‹ be a commutative diagram where the vertical maps are admissible monomorphisms and the horizontal maps are admissible epimorphisms. Let A / X

h g Ae / X‹ be a diagram where h is an admissible monomorphism. Then the map A × Y → A × Y is an admissible X e Xf ‹ monomorphism. Proof. There is a diagram of exact sequences

0 / A ×X Y / A ⊕ Y / X / 0

0 A × Y A ⊕ Y X 0 / e Xf ‹ / e ‹ / ‹ / The middle vertical map is an admissible monomorphism, so the ﬁrst is as well.

Corollary 5.4.44. Let g : B → C be an admissible epimorphism in Filt AdMon(E) and let f : A → C be any morphism. Then the pullback A ×C B exists and the map A ×C B → A is an admissible epimorphism. 5.4. FILTERED AND GRADED OBJECTS IN EXACT CATEGORIES 93

Proof. By Proposition 5.1.2 and Proposition 5.4.43 the pullback is given by (A ×C B)i = Ai ×Ci Bi

for i ∈ Z and i = top. with the maps Ai ×Ci Bi → Aj ×Cj Bj for j > i being the obvious ones. The situation for pushouts of admissible monomorphisms is less satisfying. In general it only holds when E is quasi-abelian.

Proposition 5.4.45. Suppose that E is a quasi-abelian category. Then Filt AdMon(E) is an exact category. If E is abelian then Filt AdMon(E) is a quasi-abelian category. Finally, if E is weakly (ℵ0; Admon)-elementary, then all of this is true for Filt AdMon(E). Proof. It remains to show that pushouts of admissible monomorphisms are admissible monomorphisms. Let

(ftop,fi) (Atop, αi, ai) / (Btop, βi, bi)

0 (ftop,fi) (Xtop, χi, xi) / (Ytop, ξi, yi)

be a pushout diagram with (ftop, fi) an admissible monic. By Proposition 5.3.18, the diagram

ftop Atop / Btop

0 ftop Xtop / Ytop is a pushout diagram. The ﬁltration on Y is given by Yi = Im(Bi ⊕ Xi → Ytop). It remains to see that Xi → Im(Bi ⊕ Xi) is an admissible monomorphism. But Xi → Im(Bi ⊕ Xi → Ytop) coincides with the composition Xi → Xtop → Ytop. Xtop → Ytop is an admissible monomorphism as the pushout of an admissible monomorphism. The claim when E is abelian follows immediately from Proposition 5.4.39. The claim for exhaustively ﬁltered objects also follows from the fact that in weakly (ℵ0; AdMon)-elementary quasi-abelian categories we get isomorphisms ∼ ∼ ∼ lim→Im(Bi ⊕ Xi → Ytop) = Im(lim→Bi ⊕ Xi → Ytop) = Im(Btop ⊕ Xtop → Ytop) = Ytop

V The categories Filt S (E) and Filt S (E) may considered as subcategories of Filt S (E) in the following sense. Proposition 5.4.46. Let

(ftop,fi) (gtop,gi) 0 / (Atop, αi, ai) / (Btop, βi, bi) / (Ctop, γi, ci) / 0

be a short exact sequence of in Filt S (E).

(1) Suppose that E is weakly (ℵ0; S)-elementary. If (Atop, αi, ai) and (Ctop, γi, ci) are exhaustive then so is (Btop, βi, bi) (2) If S ⊂ AdMon (Atop, αi, ai) and (Ctop, γi, ci) are complete then so is (Btop, βi, bi). Proof. (1) Consider the diagram of short exact sequences

0 / lim→Ai / lim→Bi / lim→Ci / 0

0 / Atop / Btop / Ctop / 0 The two outer vertical maps are isomorphisms so the middle one is as well. 94 5. FILTERED AND GRADED OBJECTS IN EXACT CATEGORIES

(2) By passing to a right abelianisation we may assume that E is abelian. By the 3 × 3 lemma we get for each n ∈ Z an exact sequence 0 → Atop An → Btop Bn → Ctop Cn → 0 Taking projective limits commutes with kernels, so we get a commutative diagram

0 / Atop / Btop / Ctop / 0

0 / lim←n Atop An / lim←n Btop Bn / lim←n Ctop Cn in which the top and bottom rows are exact, and the ﬁrst and last maps are isomorphisms. The Snake Lemma implies that the middle map is also an isomorphism, as required.

Corollary 5.4.47. If E is weakly (ℵ0; S)-elementary and S is ℵ0-closed, then Filt S (E) is an extension- closed, coreﬂective, exact subcategory of Filt S (E).

In general Filt S (E) and Filt S (E) are not exact. However when D is an exact category we can and will say that a functor F : Filt S (E) → D or F : Filt S (E) → D is exact if it sends an exact sequence as defined in Definition 5.4.40 to an exact sequence in D. Likewise one defines exact functors F : D → Filt S (E) or F : D → Filt S (E).

Remark 5.4.48. Although the categories Filt S (E) and Filt S (E) may not be exact, they are strongly left exact in the sense of [BC13] Deﬁnition 3.2.

Example 5.4.49. It is clear that the functors (−)l,Fi, and ﬁlt are exact functors. 5.4.3. Exactness of AdMon-Filtered Objects. In this subsection we consider the special case that S ⊆ AdMon. As we shall see, in many instances this is better behaved than the general case. 5.4.3.1. Complete Objects. We begin by analysing complete objects.

Lemma 5.4.50. Let E be an exact category with enough projectives. Then the inclusion functors Filt AdMon(E) → V V Filt AdMon(E) and Filt AdMon(E) → Filt AdMon(E) have left adjoints. Proof. For each n the map Atop An → Atop An+1 is an admissible epimorphism. Let P be projective. Then Hom(P,Atop An) is a Mittag-Leffler system of abelian groups. The maps Atop → Atop An are also admissible epimorphisms. Thus Hom(P,Atop) → Hom(P, lim←n Atop An) is a epimorphism, so Atop → lim←n Atop An is an admissible epimorphism. In particular for each n the map Aen → Aetop is an admissible monomorphism, and by the Obscure Lemma each Aen → Aen+1 is an admissible monomorphism. Thus Ae is ∼ a well-defined object of Filt AdMon(E). Moreover for any projective P we have Hom(P, Ae) = HomÂ(P,A). ˜ Thus A is complete, since completion is well-defined for abelian groups.

V V Proposition 5.4.51. Any object of Filt AdMon(E)) (resp. Filt AdMon(E)) is a projective limit of objects + + in Filt AdMon(E) (resp. Filt AdMon(E)). ≥n ≥n Proof. Let A = (Atop, αi, ai) be a filtered object. Denote by A the filtered object with Atop = ≥n ≥n ≥n ≥n Atop An−1, Am = 0 for m < n, and Am = Am An−1 for m ≥ n. The maps αi and ai are defined in the obvious way. There is a natural map A≥n−1 → A≥n. Moreover there is a natural isomorphism ∼ ≥n ∼ ≥n Ae = lim←n A . In particular if A is complete then A = lim←n A . We introduce a concept which will be useful for dealing with graded equivalences of complete filtered objects later.

Definition 5.4.52. A ﬁltered object (Atop, αi, ai) in Seqtop(E) is said to be completion acyclic if for any m ∈ Z, the projective sequence Am Am−n is lim←-acyclic. 5.4. FILTERED AND GRADED OBJECTS IN EXACT CATEGORIES 95

In particular any bounded-below ﬁltered object is completion acyclic.

Proposition 5.4.53. Suppose that E has enough projectives. Then any ﬁltered object (Atop, αi, ai) in Filt AdMon(E) is completion acyclic. Proof. This follows immeditaley from Proposition 2.6.103, and the fact that Am A−+n → Am Am−n−1 is an admssible epimorphism. 5.4.3.2. Exactness and the Associated Graded Functor. In this section we show that often one can determine when a map is an admissible epimorphism/ admissible monomorphism by looking at the associated graded map.

Proposition 5.4.54. Let S be a class contained in AdMon, and let f : A → B be a map in Filt S (E). Then

(1) If f is an admissible monomorphism then grtop(f) is an admissible monomorphism. The converse is true if the ﬁltrations on A and B are complete, C = Coker(f) exists in Filt S (E), and A is completion acyclic. (2) If f is an admissible epimorphism then grtop(f) is an admissible epimorphism. The converse is true if the ﬁltrations on A, and B are complete, and for each m ∈ Z and m = top the sequence K given by the kernel of Am Am−n → Bm Bm−n is lim←-acyclic. (3) If a null sequence 0 → A → B → C → 0

in Filt S (E) is exact then

0 → grtop(A) → grtop(B) → grtop(B) → 0 is exact. The converse is true if the filtration on A is completion acyclic and all of A, B, C are complete. Proof. The argument (particularly for the second claim) is a generalisation of [CCN21] Lemma 2.16 (2). Let us prove the first claim, the others are similar. The statement that f being an admissible monomorphism implies that grtop(f) is follows by definition and by the 3 × 3 lemma. Let us prove the (partial) converse statement. Suppose that gr(f)top is an admissible monomorphism. Let C be the cokernel of f. For each n, m ∈ Z (or m = top) we have an exact sequence 0 / Am Am−n / Bm Bm−n / Cm Cm−n / 0 Taking the projective limit over n, and noting that the filtrations are complete, with A completion acyclic gives that 0 / Am / Bm / Cm / 0 is exact for each m ∈ Z, and m = top. Remark 5.4.55. If A and B are bounded below then the statements in the previous Proposition all become if and only if statements. For the second claim, if we let Km,n denote the kernel of Am Am−n → Bm Bm−n. The Snake Lemma implies that Km,n → Km,n+1 is an admissible epimorphism. Thus if projective limits of admissible epimorphic systems are exact, e.g. if E has enough projectives, then the second statement of the previous proposition is also if and only if whenever A and B are complete.

Corollary 5.4.56. Let S be a class of morphisms contained in AdMon. A map f in Filt S (E) between objects with bounded below ﬁltrations is an admissible monomorphism (resp. epimorphism) if and only if gr(f) is. Moreover a null sequence 0 → A → B → C → 0

in Filt S (E) in which all objects have bounded below ﬁltrations is exact if and only if 0 → gr(A) → gr(B) → gr(B) → 0 is exact. 96 5. FILTERED AND GRADED OBJECTS IN EXACT CATEGORIES

5.4.3.3. Generators and Projectives. In this subsection we discuss what it means for a collection of objects in Filt S (E) to be a (projective) generating collection.

Proposition 5.4.57. Let G be an admissible generating set in E. Then for any object A of Filt S (E) L L (resp. Filt S (E)) there is an object X of < Fi(G) (resp. Fi(G)) and an admissible epimorphism i∈Ztop i∈Z X → A.

Proof. Let (Atop, αi, ai) be a ﬁltered object. For each i ∈ Z or i = top pick some Gi ∈ G and an L admissible epimorphism Gi Ai. Then ( Fi)Gi ⊕ FtopGtop → A is an admissible epimorphism. The i∈Z claim for exhaustively ﬁltered objects is similar.

Although Filt S (E) is not an exact category, we still have the notion of a projective object.

Definition 5.4.58. An object P of Filt S (E) (resp. Filt S (E)) is said to be projective if the functor Hom(P, −) is exact. The following is clear.

Proposition 5.4.59. Let P be a projective object in E, and i ∈ Z. Then FiP is projective in both Filt S (E) and Filt S (E) for any ﬁltering class of morphisms S. FtopP is projective in Filt S (E). Using the fact that Proposition 2.6.119 clearly works in this more general context, we can quite easily classify projective objects when S ⊆ AdMon.

Proposition 5.4.60. Let S ⊆ AdMon and assume that E is a weakly (ℵ0; S)-elementary exact category. If a ﬁltered object (Atop, αi, ai) is projective in Filt S (E) then Atop Ai is projective in E for all i ∈ Z. In + Filt S (E) A is projective if and only if Atop Ai is projective for each i ∈ Z.

Proof. The ﬁrst assertion is a consequence of the fact that the functor Qi : Filt S (E) → E is left adjoint to the exact functor Fi+1 : E → Filt S (E). Thus Qi(Atop, αi, ai) = Atop Ai is projective. The second assertion is a consequence of Corollary 4.1.1, and the fact that the functor (−) is left adjoint to an exact functor, and so preserves projectives. Using Propositions 5.3.21, 5.4.57, and 5.4.59 we get the following corollary.

Corollary 5.4.61. Let E be an elementary quasi-abelian category. Then Filt AdMon(E) is an elementary exact category. If E is abelian then it is an elementary quasi-abelian category. In particular for the case E = Ab is the category of abelian groups, this recovers Proposition 3.1.5 of [Sch99].

5.4.3.4. Filtered Objects in Monoidal Exact Categories. Recall that the functor ﬁlttop : Gr •(E) → Filt S (E) is strong monoidal. On the other hand the associated graded functor grtop : Filt S (E) → Gr(E) is unfortunately only lax monoidal. However we have the following.

j j j j Proposition 5.4.62. For 1 ≤ j ≤ k let A = (Atop, α , a ) be a ﬁltered object. (1) Suppose that for each n the map M A1 ⊗ ... ⊗ Ak → A1 ⊗ ... ⊗ Ak i1 ik top top i1+...+ik=n

Nk j Nk j is admissible. Then the map j=1 gr(A ) → gr j=1 A is an admissible epimorphism.

j j (2) If for each 1 ≤ j ≤ k and each i ∈ Z the map each map Ai → Ai+1 is a pure monomorphism, then Nk j Nk j the map j=1 gr(A ) → gr j=1 A is an isomorphism.

Proof. (1) Let I denote the image of the map L A1 ⊗ ... ⊗ Ak → A1 ⊗ ... ⊗ n i1+...+ik=n i1 ik top k Atop. By the obscure lemma the map In → In+1 is an admissible monomorphism. Moreover 5.4. FILTERED AND GRADED OBJECTS IN EXACT CATEGORIES 97

the map L A1 ⊗ ... ⊗ Ak → I is an admissible epimorphism. Hence the map i1+...+ik=n i1 ik n L A1 ⊗ ... ⊗ Ak → I I is an admissible epimorphism. The obscure lemma i1+...+ik=n+1 i1 ik n+1 n then implies the result. k k (2) Suppose now that for each 1 ≤ k ≤ n and each i ∈ Z the map Ai → Ai+1 is a pure monomorphism. k k k Equivalently 0 → Ai → Ai+1 → gri+1(A ) is a pure exact sequence. By tensoring there is an induced n-dimensional chain complex which is exact along each axis. There is an acyclic sequence

k M A ⊗ ... ⊗ A ⊗ ... ⊗ A → A ⊗ ... ⊗ A → gr (A) ⊗ ... ⊗ gr (A) → 0 i1+1 il ik+1 i1+1 ik+1 i1+1 ik+1 l=1

Moreover this is a pure exact sequence. Hence there is a pure exact sequence.

k X 0 → A ⊗ ... ⊗ A ⊗ ... ⊗ A → A ⊗ ... ⊗ A → gr (A) ⊗ ... ⊗ gr (A) → 0 i1+1 il ik+1 i1+1 ik+1 i1+1 ik+1 l=1

This completes the proof.

Definition 5.4.63. A ﬁltered object H in Filt S (E) is said to be ﬂat if the functor H ⊗ (−): Filt S (E) → Filt S (E) is exact.

Proposition 5.4.64. Let X be a ﬂat object of E such that X ⊗ f ∈ S whenever f ∈ S. Then for any i ≥ 0, FiX is a ﬂat object of Filt S (E).

Proof. If X is ﬂat and A = (Atop, αi, ai) is an object of Filt S (E), then (FiX ⊗ A)j is 0 for j < i and X ⊗ Aj for j ≥ i. From the deﬁnition of exact sequences in Filt S E it is clear that FiX ⊗ A is exact.

We cannot precisely classify ﬂat objects. However we have the following.

Proposition 5.4.65. Let S ⊂ AdMon. Suppose that for any object A of E, and any exact sequence 0 → X → Y → Z → 0 with X → Y in S, the sequence

0 → ImS (X ⊗ A → Y ⊗ A) → Y ⊗ A → Z ⊗ A → 0 is exact. If a filtered object (Htop, ti, hi) is flat then Htop, and Htop Hi are flat.

Proof. Suppose that (Htop, ti, hi) is ﬂat. Let

0 / A / B / C / 0 be an exact sequence in E. Then

0 / F0A / F0B / F0C / 0 is exact in Filt S (E). Therefore

0 / H ⊗ F0A / H ⊗ F0B / H ⊗ F0C / 0 is exact. In particular

0 / Htop ⊗ A / Htop ⊗ B / Htop ⊗ C / 0 98 5. FILTERED AND GRADED OBJECTS IN EXACT CATEGORIES

is exact. Hence Htop is ﬂat. Moreover, by assumption we have the following diagram. 0 0 0

0 / ImS (Hi ⊗ A → Htop ⊗ A) / ImS (Hi ⊗ B → Htop ⊗ B) / ImS (Hi ⊗ C → Htop ⊗ C) / 0

0 / Htop ⊗ A / Htop ⊗ B / Htop ⊗ C / 0

0 / Htop Hi ⊗ A / Htop Hi ⊗ B / Htop Hi ⊗ C / 0

0 0 0 The columns are exact, and the top two rows are exact by assumption. Therefore the third row is exact, so Htop Hi is flat. The assumptions of the propositions are satisfied, for example, if E is quasi-abelian and S = AdMon. The assumptions of Proposition 5.4.64 are satisfied for S = Mor(E) We then get the following.

Corollary 5.4.66. (1) If E is a monoidal elementary exact category then Seqtop(E) and Seq(E) are monoidal elementary exact categories. (2) If E is a monoidal elementary quasi-abelian category then Filt AdMon(E) is a monoidal elementary exact category.

5.4.3.5. Model Structures on Categories of Filtered Objects. As we have seen, the categories Filt S (E) and Filt S (E) in general only have exact structures when E is quasi-abelian and S = AdMon. However one is often still able to equip these categories with natural model structures, even when E is a more general exact category. We shall assume that E is weakly (ℵ0; S)-elementary and that S is ℵ0-closed throughout this section. Denote by Ch∗(S) the class of morphisms in Ch∗(E) consisting of morphisms f• : X• → Y• ∼ such that fn is in S for each n ∈ Z. We consider the categories Filt Ch∗(S)(Ch∗(E)) = Ch(Filt S (E)) and ∼ Filt Ch∗(S)(Ch∗(E)) = Ch(mathpzcF iltS (E)). Let O be a class of objects in E. Denote by Filt S (O) the class of objects in Filt S (E) of the form (Atop, αi, ai) such that Atop ∈ O, Ai ∈ O for each i ∈ Z and gr(Atop, αi, ai) is in Gr(O). + Proposition 5.4.67. Suppose that O is extension closed, S ⊂ AdMon, (Atop, αi, ai) ∈ Filt S (E)and gr(A)top ∈ O. Then for each i ∈ Z we have Ai ∈ O. In particular if (Atop, αi, ai) ∈ Filt S (E), and O is closed under (ℵ0; S)-extensions, then (Atop, αi, ai) ∈ Filt S (O) if and only if gr(A) ∈ Gr(O). If further O is + closed under taking cokernels of maps in S, then (Atop, αi, ai) ∈ Filt S (O) if and only if Ai ∈ O for each 0 ≤ i < ∞.

Proof. Without loss of generality we may assume that Ai = 0 for i < 0. Suppose that gr(A) ∈ Gr(O). In particular A0 ∈ O. Moreover for each i ≥ 0 there is an exact sequence

0 → Ai → Ai+1 → gri(A) → 0

Since gri+1(A) ∈ O, an easy induction gives Aj ∈ O for each 0 ≤ j < ∞. If A is exhaustively filtered then Atop is an (ℵ0; S)-extension of objects in O, so the second claim follows. The final claim follows again from the exact sequence above. Let (L, R) be a cotorsion pair on E. In particular L and R are extension closed. Since L is closed under transfinite extensions by Proposition 5.4.67 we get the following. 5.4. FILTERED AND GRADED OBJECTS IN EXACT CATEGORIES 99

+ Corollary 5.4.68. Suppose that E is weakly (ℵ0; S)-elementary. Let A ∈ Filt S (E). Then gr(A) ∈ Gr(L) if and only if A ∈ Filt S (L). Moreover Lemma 3.4.57 implies the following. + + Proposition 5.4.69. Let A ∈ Filt S (L) and B = (Btop, βi, bi) ∈ Filt S (E) be such that Bi ∈ R for each i ∈ Z. Then any ﬁltered exact sequence 0 → B → C → A → 0 splits.

Definition 5.4.70. We say that a morphism f in Filt S (E) (resp. Filt S (E)) is a ﬁltered quasi- isomorphism if fi is a weak equivalence for each i ∈ Z (resp. for each i ∈ Z and i = top). Definition 5.4.71. A complex A ∈ Filt S (E) is said to be quotient acyclic if Atop An is acyclic for each n. This will allow us to detect acyclic complete objects using the associated graded functors under mild assumptions. Note that any acyclic complex is completion acyclic.

Example 5.4.72. If A ∈ Filt AdMon(E) is bounded below then it is acyclic if and only if it is quotient acyclic. Proposition 5.4.73. (1) Let E be an exact category with enough projectives. Then a complex A V in Filt AdMon(Ch(E)) is quotient acyclic if and only if it is acyclic. (2) If E is weakly (ℵ0; AdMon)-elementary then a complex A in Filt AdMon(Ch(E)) is quotient acyclic if and only if gr(A) is acyclic. Proof. (1) Suppose A is quotient acyclic. Then each A An is acyclic. Moreover each Atop An → ∼ A An+1 is an admissible epimorphism. Thus for each projective P , Hom(P, lim←n Atop An) = ∼ lim←n Hom(P,Atop An) is acyclic. Hence Atop = lim←n Atop An is acyclic. This also implies that each An is also acyclic. (2) This proof is a generalisation is a modiﬁcation of [CCN21] Lemma 2.16 (1). Suppose E is weakly (ℵ0; AdMon)-elementary, and let A be such that gr(A) is acyclic. Each An → An+1 is an admissible monomorphism and an equivalence. Since E is weakly (ℵ0; AdMon)-elementary the transﬁnite composition An → Atop is an admissible monomorphism which is an equivalence. Thus Atop An is acyclic.

V Corollary 5.4.74. Let f : X → Y be a map in Filt AdMon(Ch(E)). Suppose that E has enough projectives and is weakly (ℵ0; AdMon)-elementary. Then f is an equivalence if and only if gr(f) is an equivalence.

Corollary 5.4.75. Suppose that E has enough projectives and is weakly (ℵ0; AdMon)-elementary. V V V Let f : X → Y be a graded equivalence Filt AdMon(Ch(E)). Then f : X → Y is an equivalence in V Filt AdMon(Ch(E)).

V Proof. This follows immediately from the fact that the natural map gr(A) → gr(A) is an isomorphism.

Let (L, R) be a dg∗-compatible cotorsion pair on E and consider the induced model structure on Ch∗(E).

Theorem 5.4.76. Suppose that E is a weakly S-elementary exact category and that (L, R) is a dg∗- compatible cotorsion pair for ∗ ∈ {∅, ≥ 0} and that acyclic coﬁbrations are degree-wise split. Then the

transferred model structure exists on Filt Ch∗(S)(Ch∗(E)) and Filt Ch∗(S)(Ch∗(E)). If g is a filtered (acyclic) cofibration then gr(g) is a graded (acyclic) cofibration. 100 5. FILTERED AND GRADED OBJECTS IN EXACT CATEGORIES

Proof. We prove the claim for exhaustive filtered objects, the non-exhaustive case being similar. Let us first show that transfinite compositions of pushouts of coproducts of maps of the form filt(f), where f is a generating cofibration in Ch∗(Gr(E)), exist in Filt Ch∗(S)(Ch∗(E)). Indeed such a map f is in each homological degree a split monomorphism. Since by assumption S is closed under direct sums, by Proposition P 5.1.2 these colimits exist and the functor (−)n commutes with them. It therefore suffices to show that n∈Z transfinite compositions of pushouts of coproducts of maps of the form filt(f) where f is a generating acyclic P cofibration is a weak eauivalence. But since such colimits are computed degree-wise and (−)n commutes n∈Z with such colimits this is clear. The claim about the associated graded functor follows from the fact that gr sends cell(filt(I)) and cell(filt(J)) to cell(I) and cell(J) respectively, where I is a collection of generating cofibrations in Ch∗(Gr(E)), and J is a collection of generating acyclic cofibrations in Ch∗(Gr(E)). Proposition 5.4.77. The class of cofibrant/ trivially cofibrant) objects in the transferred model structure of Theorem 5.4.76 on Filt S (Ch∗(E)) is Filt S (C) (resp. Filt S (C ∩ W)) where C (resp W ∩ C) is the class of cofibrant (resp. trivially cofibrant) objects in Ch∗(E). In particular a bounded below exhaustively filtered object A is cofibrant/ trivially cofibrant if and only gr(A) is cofibrant/ trivially cofibrant as a graded object. Proof. The first claim follows from the fact that the objects described are the cofibrations (trivial

coﬁbrations) in the projective model structure (for functors) on Seq(Ch∗(E)) (resp. Seqtop(Ch∗(E))), and

the computation of a transfinite composition lim→α<λ Xα where X0 ∈ Filt S (E) (resp. Filt S (E)) and each Xα → Xα+1 is a pushout of a generating cofibration/ acyclic cofibration is the same in both Seq(Ch∗(E))

(resp. Seqtop(Ch∗(E))) and X0 ∈ Filt S (E) (resp. Filt S (E)).

Theorem 5.4.78. Suppose that E is a weakly S-elementary exact category, with S ℵ0-closed, left can- celling, and such that functorial S-images exist. Suppose further that (L, R) is a monoidally dg∗-compatible cotorsion pair for ∗ ∈ {∅, ≥ 0}. Suppose further that Filt S (E) is a monoidal category with the induced monoidal structure, and that the induced model structure on Ch∗(E) has a collection of generating cofibrations which are split exact in each degree. The transferred model structure on Filt Ch∗(S)(Ch∗(E)) is n monoidal. If Ch∗(E) has maps of the form 0 → D (F ), with F ∈ L, as generating acyclic cofibrations, then FiltCh∗(S)(Ch∗(E)) satisfies the monoid axiom.

Proof. We have already explained why Filt Ch∗(S)(Ch∗(E)) is a monoidal model category. Now let us prove the monoid axiom. A transfinite composition of pushouts of tensor products of objects with generating acyclic cofibrations will be of the form A → A ⊕ Y , where Y is a direct sum of objects of the form n X ⊗ Fi(D (L)) for some L ∈ L. Y is clearly trivially cofibrant, so A → A ⊕ Y is a trivial cofibration. 5.4.3.6. Model Structures for Complete Objects. Let E be a weakly idempotent complete exact category with enough projectives. Since projective limits of admissibly epimorphic projective sequences in E are exact by Proposition 2.6.106, we immediately get the following. Proposition 5.4.79. Let E be a weakly idempotent complete exact category with enough projectives. Let A = (Atop, αi, ai) be an object of Ch(Filt AdMon(E)). Then the map lim←n Atop An → holim←n Atop An is an equivalence. In particular if A is complete then it is homotopically complete, and A is homotopically complete if and only if lim←n Atop An is acyclic.

Proposition 5.4.80. Let E be an elementary exact category, and let A = (Atop, αi, ai) be an object of

Ch∗(Filt AdMon(E)) for ∗ ∈ {≥ 0, ∅}. Then A is homotopically complete if and only if holim←n An is trivial.

Proof. For the stable model category Ch(E) this follows from [GP18] Section 2.7. For Ch≥0(E), it follows because all the necessary computations happen in Ch(E).

Let E be an elementary quasi-abelian category. Then Ch(Filt AdMon(E)) is a combinatorial, left proper model category when equipped with the projective model structure. The left Bousﬁeld localisation of Ch(Filt AdMon(E)) at maps of the form 0 → const(P ) where P is a projective generator, and const(P ) is the constant ﬁltered object with const(P )i = const(P )top = P for all ∈ Z exists. The adjunction

IAdMon;Mor(E) : Seq(Ch(E)) Filt AdMon(Ch(E)) :RAdMon;Mor(E) 5.4. FILTERED AND GRADED OBJECTS IN EXACT CATEGORIES 101

is a Quillen equivalence. IAdMon;Mor(E) is the identity on the maps 0 → const(P ), so there is an induced Quillen equivalence of the localised model structures by [Hir09]. The equivalences for the localised model structure on Seq(Ch(E)) are the derived graded equivalences by Theorem 5.3.36. Since the adjunction is a Quillen equivalence, the equivalences for the localisation of right-hand side are also the derived graded equivalences. Since in Filt AdMon(Ch(E)) the associated graded functor preserves equivalences, the weak equivalences are in fact all graded equivalences. We have proven the following.

Theorem 5.4.81. The homotopically complete ﬁltered model structure exists on Filt AdMon(Ch(E)).A map f is an equivalence if and only if gr(f) is an equivalence.

V On the other hand, we know that Filt AdMon(E) is an extension closed subcategory of Filt AdMon(E). V Thus it is an exact category. Moreover FiP is projective in Filt AdMon(E) for P projective and i ∈ Z or i = top. Using that P is projective and so Hom(P, −) commutes with completion, it is easy to see that if P V V V is tiny, then Fi is tiny in Filt AdMon(E). Thus Filt AdMon(E) and Filt AdMon(E) are elementary quasi-abelian categories. Theorem 5.4.82. Let E be an elementary quasi-abelian category. Then the complete model structure V exists on Ch(Filt AdMon(E)), and it is combinatorial. Moreover the adjunction

V V (−): Ch(Filt AdMon(E)) Ch(Filt AdMon(E)) :J is a Qullen equivalence when the category on the left-hand side is equipped with the homotopically complete monoidal structure.

V Proof. J clearly preserves ﬁbrations. By Corollary 5.4.74 and Proposition 5.4.75,( −) and J both preserve and reﬂect all equivalences. It follows that the Quillen adjunction is in fact a Quillen equivalence.

Now suppose that E is closed symmetric monoidal elementary quasi-abelian category. Then Filt AdMon(E) is closed symmetric monoidal elementary. R(Homfilt(−, −)) deﬁnes an internal hom on Filt AdMon(E). In- deed the functor Hom(X, −) is kernel preserving, so for A and B objects of Filt AdMon(E), R(Homfilt(A, B)) is in Filt AdMon(E). As for vector spaces over a ﬁeld in [CCN21] Section 2.1, we then have the following. Proposition 5.4.83. Let E be a closed symmetric monoidal elementary quasi-abelian category. If Y is V V

an object in Filt AdMon(E) and X is any object in Filt AdMon(E) then Homfilt(X,Y ) is in Filt AdMon(E). In V V particular (Filt AdMon(E), − ⊗ −,F0(k), Homfilt) is a closed symmetric monoidal elementary quasi-abelian category. ∼ ≥n Proof. Since Y is complete, we may write it as Y = lim←n Y . ∼ ≥n ≥n Then Homfilt(X,Y ) = lim←n Homfilt(X,Y ). Homfilt(X,Y ) is also a bounded-below ﬁltered object, V and hence complete. As a reﬂective subcategory Filt AdMon(E) is closed under colimits in Filt AdMon(E). The result follows by abstract nonsense. CHAPTER 6

Homotopical Algebra in Exact Categories

In this ﬁnal chapter we show that monoidal elementary exact categories are good settings in which to do homotopical algebra. In particular we show that they are naturally homotopical algebra contexts in the sense of [TTV08]. e also establish Dold-Kan equivalences for algebras over operads. As will be explained in future work, over Q this essentially implies that for derived geometry relative to E one can work with either simplicial objects are non-negatively graded complexes, and over Z justiﬁes working with simplicial objects.

6.1. Algebra in Monoidal (Model) Categories Before specialising to complexes in exact categories, let us establish some general results concerning the existence of model structures on categories of algebras over operads. We ﬁrst recall basic facts concerning algebra in arbitrary monoidal categories. Throughout this section (C, ⊗, k) is a monoidal category, with monoidal functor ⊗ and (M , •M ) is a unital left C-module. Because we want to deal with operads as associative monoids in a certain non-symmetric monoidal category, we will not assume that the monoidal structure is symmetric. We shall assume that C is ﬁnitely complete and cocomplete, and that the tensor product commutes with coproducts. What follows is largely standard. Much of it can be found in [BBK13] for example.

6.1.1. Associative Monoids. We denote the category of (unital) associative monoids internal to C by Alg Ass(C). There is a faithful forgetful functor | − |Ass : Alg Ass(C) → C. If C has countable products then | − | has a left adjoint T which can be constructed explicitly. Namely for V ∈ C, set

⊗n Tn(V ) = V

∞ M T (V ) = Tn(V ) n=0 ⊗0 where by deﬁnition T0(V ) = V = k. Now ⊗ preserves colimits in each variable, so

∞ ∼ M T (V ) ⊗ T (V ) = Tm(V ) ⊗ Tn(V ) m,n=0 The multiplication m : T (V ) ⊗ T (V ) → T (V )

is deﬁned on the summand Tm(V ) ⊗ Tn(V ) by the composition ∼ Tm(V ) ⊗ Tn(V ) = Tm+n(V ) → T (V )

⊗m ⊗n ∼ ⊗(m+n) where the isomorphism Tm(V ) ⊗ Tn(V ) = V ⊗ V = V = Tm+n(V ) is the natural isomorphism. The identity is given by the inclusion e : k = T0(V ) → T (V ). m and e endow T (V ) with the structure of a unital associative monoid. It is clear that V → T (V ) is functorial in V , and it is straightforward to check that T is left adjoint to | − |.

102 6.1. ALGEBRA IN MONOIDAL (MODEL) CATEGORIES 103

6.1.1.1. Commutative Monoids. If C is symmetric monoidal then we denote the category of (unital) commutative monoids by Alg Comm(C). If C has ﬁnite coequalizers and countable coproducts then the forgetful functor | − |Comm : Alg Comm(C) → C has a left-adjoint, which can be constructed explicitly as follows. The ⊗n symmetric group on n letters Σn acts on Tn(V ) = V . Let Sn(V ) = Tn(V )Σn be the coinvariants for this action. We then set ∞ M S(V ) = Sn(V ) n=0 The associative monoid structure on T (V ) descends to an associative monoid structure on S(V ). One checks easily that it is commutative and that it is a left adjoint.

6.1.2. Modules. Fix objects A and B of Alg Ass(C). We denote by AMod (M ) the category of left

modules for A in M . There is a forgetful functor | − |A : AMod (M ) → M . This functor has a left adjoint. It sends an object E to the object A • E with the obvious left action of A. The adjunction

A • (−): M AMod (M ):| − |A is called the free-forgetful adjunction. If M is a right C-module we denote by Mod A(M ) the category of right modules for A in M . If M is a C-bimodule, we denote by AMod B(M ) the category of A − B bimodules. Regard C as a bimodule over itself, and let M be a left C-module. Let E be a right A-module in C with action morphism

aE : E ⊗C A → E and F a left A-module in M with action morphism

aF : A •M F → F If the category C has ﬁnite coequalisers, then we deﬁne

E •A F to be the coequaliser of the maps

aE * E ⊗C A •M F E •M F aF 4 This deﬁnes a bifunctor

•A : Mod A(C) × AMod (M ) → M

If E is a B − A bimodule in C, F an A bimodule in M , then E ⊗A F is naturally a B-module, i.e. •A gives a bifunctor

BMod A(C) × AMod (M ) → BMod (M ) In particular if C is symmetric monoidal, M a left C-module, and A is a commutative monoid then this gives a bifunctor

AMod (C) × AMod (M ) → AMod (M )

If we regard C as a bimodule over itself, then this endows AMod (C) with a symmetric monoidal structure, and AMod (M ) with a left AMod (C)-module structure. Suppose that α : A → B is a map of commutative monoids in C. Then B is an object of Alg Comm(AMod (C)). Thus we get an extension of scalars adjunction.

B •A (−): AMod (M ) BMod (M ):| − |α Suppose further that the monoidal structure on C is closed, and let Hom(−, −) denote the internal hom functor. Then one can also construct an internal hom, HomA(−, −) functor on AMod (C) by a similar method as used to construct ⊗A. This makes (AMod (C), ⊗A, HomA(−, −),A) a closed monoidal category. See for example [BBK13] for details. 104 6. HOMOTOPICAL ALGEBRA IN EXACT CATEGORIES

6.1.2.1. Adjunctions of Categories of Modules. In this subsection we let (C, ⊗C , kC ) and (D, ⊗D , kD ) be monoidal categories, (M , •M ) a left C-module, and (N , •N ) a left D-module. Definition 6.1.1. A pair of functors R : D → C, R : N → M is said to be lax monoidal if there are natural transformations

: kC → R(kD )

µ : R(−) ⊗C R(−) → R(− ⊗N −)

ν : R(−) •M R(−) → R(− •N −) satisfying obvious associativity and unitality axioms (see e.g. [Lur21] Section 2.1.5). If and µ, and ν are isomorphisms then the pair of functors is said to be strong monoidal. One deﬁnes an oplax monoidal pair of functors dually. Note that in particular R is a lax monoidal functor between monoidal categories. Definition 6.1.2. A pair of adjunctions L: C D :R L: M N :R is said to be lax monoidal if the pair (R, R) is lax monoidal. Let L: C D :R L: M N :R be a lax monoidal pair of adjunctions. By doctrinal adjunction ([Kel74]) (L, L) is an oplax monoidal pair of functors - we spell out the argument here. Let , µ, and ν be the maps realising (R, R) as lax monoidal. ∨ ∨ Deﬁne : L(kC ) → kD to be the adjoint of , µ to be the composition

L(µL(−),L(−)) L(− ⊗C −) / L(RL(−) ⊗C RL(−)) / LR(L(−) ⊗D L(−)) / L(−) ⊗D L(−) and ν∨ to be the composition

L(νL(−),L(−)) L(− •M −) / L(RL(−) •M RL(−)) / LR(L(−) •N L(−)) / L(−) •N L(−) where the ﬁrst map comes from the unit transformations and the last map from the counit transformation.

Now let L: C D :R L: M N :R be a lax monoidal pair of adjunctions, and T an associative monoid in D. By lax monoidality, R(T ) is an associative monoid in C. Moreover R induces a well-deﬁned functor

RT : T Mod (N ) → R(T )Mod (M ) Exactly as in [SS00] Page 305, this functor has a left adjoint constructed as follows. For M an R(T )-module in M let αM : T •N (L(R(T ) •M M)) → T •N L(M)

be induced by the multiplication map R(T ) •M M → M, and let βM denote the composition

∨ νR(T ),M L(R(T ) •M M) / LR(T ) •N L(M) / T •N L(M)

Denote by LT (M) the coequalizer of the maps αM and βM . LT is a functorial construction, and is left adjoint to RT . 6.2. OPERADS 105

∼ Remark 6.1.3. If R(T ) •M M is a free S-module, then LT (R(T ) •M M) = T •N M. If α : S → R(T ) is a map of associative monoids, then by composition with the extension of scalars adjunction we get a composite adjunction

Lα : SMod (M ) T Mod (N ):Rα

6.2. Operads Here we will recall some facts about operads and their algebras. We shall ﬁx a symmetric monoidal category (C, ⊗, k), and assume now that the tensor product commutes with all colimits in each variable. Most of the deﬁnitions and claims in this section can be found in [LV12].

. 6.2.1. Non-Symmetric Sequences. Consider the category Gr (C) .= Fun( 0, C) where 0 is the N0 N N discrete category of non-negative integers.

Definition 6.2.4. Let M and N be objects of Gr (C). The tensor product of M and N, denoted N0 M ⊗ N, is the graded object deﬁned by M (M ⊗ N)(n) = (M(i) ⊗ N(j)) i+j=n

Definition 6.2.5. (1) The graded object I0 is deﬁned by I0(i) = 0 for i 6= 0 and I0(0) = k. (2) The graded object I is deﬁned by I(i) = 0 for i 6= 1 and I(1) = k.

Definition 6.2.6. Let M and N be objects of Gr (C). The composite product of M and N is the N0 graded object deﬁned by . M ⊗k M ◦ns N .= (M(k) ⊗ N (n)) k≥0

Proposition 6.2.7. (Gr (C), ⊗,I0) and (Gr (C), ◦,I) are monoidal categories. N0 N0 6.2.2. Symmetric Sequences. Let (C, ⊗, k) be a symmetric monoidal category with all small coproducts. Note that in this case (Gr (C), ⊗,I0) is a symmetric monoidal category. N0 6.2.2.1. Discrete Groups in Monoidal Categories. We denote by k[−]: Set → C the functor which sends L a set S to the object k[S] = S k. If f : S → T is a map of sets, then k[f]: k[S] → k[T ] is the morphism which sends the copy of k indexed by s ∈ S to the copy indexed by f(s) ∈ T . Objects and morphisms in the essential image of the functor k[−]: Set → C will be called discrete. Proposition 6.2.8. Let (C, ⊗, k) be a monoidal category. Suppose that ⊗ preserves all coproducts. Endow Set with its Cartesian monoidal structure. Then the functor k[−]: Set → C is strong monoidal. Proof. Let S and T be sets. Then a a a a a a a k ⊗ k =∼ k ⊗ k =∼ k =∼ k S T S T S T S×T In particular Set → C sends groups to Hopf monoids. If G is a group we call k[G] the group monoid of G in C.

6.2.3. The Category of Σ-Modules. We denote by k[Σ] the monoid in (Gr (C), ⊗,I0) deﬁned as N0 follows. In degree n it is given by the monoid k[Σn], the free monoid on the symmetric group in n letters. It is an associative monoid in the monoidal category (Gr (C), ⊗,I). N0

Definition 6.2.9. The category of Σ-modules in C, denoted Mod Σ(C) is the category of right k[Σ]- modules. 106 6. HOMOTOPICAL ALGEBRA IN EXACT CATEGORIES

Definition 6.2.10. Let M and N be two Σ-modules. The composite product of M and N, denoted M ◦ N is deﬁned by M ⊗k M ◦ N(n) = (M(k) ⊗Σk N (n) k≥0

Proposition 6.2.11. (Mod Σ, ◦,I) is a monoidal category. 6.2.4. Operads and Algebras. Definition 6.2.12. (1) The category of non-symmetric operads is the category of associative monoids Alg (Gr (C), ◦ns,I). Ass N0 (2) The category of symmetric operads is the category of associative monoids Alg Ass(Mod Σ(C), ◦,I).

(1) Alg Ass(C) is the category of algebras over the non-symmetric operad with Ass(n) = k for all n. (2) Alg Comm(C) is the category of algebras over the symmetric operad with Comm(n) = k for all n, regarded as a trivial Σn-module.

(3) In the additive setting there is an operad Lie such that Alg Lie(C) is the category of Lie algebras. By thinking of objects of C as graded objects concentrated in degree 0, we may regard C as a full subcategory of both Gr (C) and Mod Σ(C). This makes C into a left (Gr (C), ◦ns,I)-module and a left (Mod Σ(C), ◦,I)- N0 N0 module Definition 6.2.13. Let P be either a symmetric or non-symmetric operad in C. The category of P- algebras, denoted Alg P(C) is the category PMod (C) (where in the non-symmetric case C is a left (Gr(C), ◦ns,I)- module, and in the symmetric case C is a left (Mod Σ(C), ◦,I)-module ). 6.2.4.1. Colimits of Algebras. In this section we recall from [Har10] how to compute certain colimits in the category of aglebras over an operad. Proposition 6.2.14 ( [Har10] Proposition 7.28). Let P be a non-symmetric operad and X a P-algebra. There exists an object PX in Gr (C) together with, for any Y ∈ Gr (C), an isomorphism, natural in X N0 N0 and Y , a ∼ X (P ◦ns Y ) = PX ◦ns (Y )

As in [Har10] Definition 7.31, for s : A → B a map in Gr (C), we define Qt (s) for t ≥ 1 and 0 ≤ q ≤ t N0 q t . ⊗t t . ⊗t t as follows. Q0(s) = A , Qt(s) = B and for 0 < q < t, Qq(s) is defined by the pushout.

t ⊗(t−q) q ⊕(q) t (X ⊗ Qq−1(s)) / Qq−1(s)

t ⊗(t−q) ⊗q ⊕(q) t (X ⊗ B ) / Qq(s)

q where the top map is the obvious projection, and the left-hand map is induced by natural map Qq−1(s) → B⊗q. Proposition 6.2.15 ([Har10] Proposition 7.32). Let s : A → B be a map in C, and let X be a P-algebra. Consider a pushout diagram P(A) / X

P(s) P(B) / P Then P is naturally isomorphic to a ﬁltered colimit ∼ P = lim→n Xn 6.3. MODEL CATEGORIES OF MONOIDS AND ALGEBRAS 107 where X0 = X, and for n ≥ 1 the map Xn−1 → Xn is given by the pushout diagram in C

n PX (n) ⊗ Qn−1(s) / Xn−1

⊗n PX (n) ⊗ B / Xn ∼ The ﬁltration P = lim→n Xn will be called the standard ﬁltration. There is also a version of this when P is a symmetric operad in C. As in [Har10] Proposition 7.6 for X a P-algebra, there is a Σ-module PX in C such that for any Σ-module Y in C, there is an isomorphism, natural in X and Y , a ∼ X (P ◦ Y ) = PX ◦ Y

The pushout P of a map P ◦ s : P ◦ A → P ◦ B along a map P ◦ A → X can also be computed using a standard ﬁltration ∼ P = lim→n Xn n ⊗n where the map Xn−1 → Xn is given by pushout along PX (n) ⊗Σn Qn−1(s) → PX (n) ⊗Σn B 6.2.4.2. Adjunctions of Operads and Algebras. Let C and D be symmetric monoidal categories, and let

L: C D :R be a lax monoidal adjunction. There are two obvious induced lax monoidal pairs of adjunctions.

Lns :(Gr (C), ⊗,I0) (Gr (D), ⊗,I0):Rns,L: C D :R N0 N0

Lns :(Gr (C), ◦ns,I) (Gr (D), ◦ns,I):Rns,L: C D :R N0 N0 Thus if T is a non-symmetric operad in D, P a non-symmetric operad in C, and α : P → R(T) a map of operads, then we get an adjunction

Lα : Alg P(C) Alg T(D):Rα

Consider the monoid k [Σ] in (Gr (D), ⊗,I0). Using the map k [Σ] → R(k [Σ]) we get an adjunction. D N0 C D

LΣ : Mod Σ(C) Mod Σ(D):RΣ

If R is a symmetric monoidal functor, then RΣ is in fact a lax monoidal functor (Mod Σ(D), ◦,I) → (Mod Σ(C), ◦,I), and we get the following lax monoidal pair of adjunctions.

LΣ :(Mod Σ(C), ◦,I) (Mod Σ(D), ◦,I):RΣ,L: C D :R So again if T is a symmetric operad in D, P a symmetric operad in C, and α : P → R(T) a map of operads, then we get an adjunction

Lα : Alg P(C) Alg T(D):Rα Remark 6.2.16. It is important to keep in mind that L is in general not lax symmetric monoidal. However it is oplax symmetric monoidal. In particular if X is a right Σn-module and Y a left Σn-module in

C, we get a map L(X ⊗Σn Y ) → L(X) ⊗Σn L(Y ). This is the map in Deﬁnition 2.3.2 of [WY19].

6.3. Model Categories of Monoids and Algebras

Let (C, ⊗C , k) be a monoidal category, and (M , •M ) be a left C-module which is also a combinatorial model category. Again, we are not assuming that ⊗ is symmetric, nor that • commutes with any colimits a priori. 108 6. HOMOTOPICAL ALGEBRA IN EXACT CATEGORIES

6.3.1. Existence of Model Structures. Definition 6.3.17. An associative monoid R in C is said to be admissible in M if the transferred model structure along the free-forgetful adjunction

R ⊗ (−): M RMod (M ):| − |R

exists on RMod (M ). We will use the following slightly generalised version of [SS00] Deﬁnition 3.3. Definition 6.3.18. Let R be an associative monoid in C. A collection S of weak equivalences in M is said to satisfy the R-monoid axiom if

(1) every map of the form R •M s where s ∈ S is a h-cofibration. R (2) M is weakly S -elementary, where S is the class of pushouts of maps of the form R •M s for s ∈ S. In particular, a monoidal model category satisfies the monoid axiom of [SS00] precisely if the class of acyclic cofibrations satisfies the R-monoid axiom for any associative monoid R, where we regard C as a left-module over itself. Proposition 6.3.19. Suppose that the class of acyclic cofibrations satisfies the R-monoid axiom, and that the forgetful functor | − |R : RMod (M ) → M commutes with transfinite compositions. Then R is admissible. Suppose further that | − |R commutes with all colimits and the class of cofibrations satisfies the R-monoid axiom. Then RMod (M ) is left proper. Proof. The first claim is essentially [SS00] Theorem 4.1/ Remark 4.2, and follows immediately from Theorem A.5.28. The second follows from Proposition A.3.19. 6.3.1.1. Model Structures on Algebras Over Operads. We now specialise to the situation that C is a monoidal category which is also a monoidal category, and we consider C as a left (Gr , ◦ns,I)-module or a N0 left (Mod Σ, ◦,I)-module. Notation 6.3.20. Let P be an operad, X a P-algebra, and S be a class of maps in C. (1) Write SullP(S; X) for the class of algebras which can be obtained as a transfinite composition

X0 → X1 → ... → Xα → ...

where X0 = X, and each Xα → Xα+1 is a pushout of a map of the form P ◦ s, where s ∈ S. P;X (2) For a functor L : C → D, denote by L(S) the class of maps of the form L(Yn−1) → L(Yn) where Yn−1 → Yn is a map appearing in the standard ﬁltration of the pushout of an algebra Y ∈ SullP(S; X) along a map of the form P(s), for s ∈ S. The following deﬁnition is based on a comment after Remark 6.1.3 on Page 36 [WY18], where we are also relaxing the condition that C be a monoidal model category. Definition 6.3.21. Let P be a non-symmetric (symmetric) operad, and S a class of maps in C and L : C → D a functor where D is also a combinatorial model category. A collection of maps S in C is said to satisfy the weak P-algebra axiom relative to (L; X) if for any s ∈ S (1) the square below is a homotopy pushout for any n ≥ 1. n á n ë L(PX (n) ⊗ Qn−1(s)) / L(Xn−1) L(PX (n) ⊗Σn Qn−1(s)) / L(Xn−1)

⊗n ⊗n L(PX (n) ⊗ B ) / L(Xn) L(PX (n) ⊗Σn B ) / L(Xn) (2) D is weakly L(S)P;X -elementary. S is said to satisfy the weak P-algebra axiom relative to X if it satisﬁes the weak P-algebra axiom relative to (IdC ; X) and it is said to satisfy the weak P-algebra axiom if it satisﬁes the weak P-algebra axiom relative to X for all X ∈ Alg P(C). 6.3. MODEL CATEGORIES OF MONOIDS AND ALGEBRAS 109

White and Yau prove in [WY18] Theorem 6.1.1 that when a monoidal model category satisfies a slightly stronger condition that the weak P-algebra axiom relative to the class of cofibrant P-algebras (called the P- algebra axiom in [Whi17] Definition C.1), then the operad P is semi-admissible, that is, there is a transferred

semi-model structure on Alg P(C) (for a semi-model structure, one only requires the left lifting property of trivial cofibrations with cofibrant domain against fibrations, and only maps with cofibrant domain can be factored as a trivial cofibration followed by a fibration). White and Yau actually prove this for coloured operads, and mention that their proof holds under the more general hypotheses of Definition 6.3.21 (though we also relax the requirement that M be a monoidal model category). For completeness we give the proof here.

Theorem 6.3.22. Suppose that C is a combinatorial model category, which is also a symmetric monoidal category. Let P be either a symmetric or non-symmetric operad in C such that the class of acyclic cofibrations in C satisfies the weak P-algebra axiom. Then P is admissible. Proof. Let S denote the class of acyclic cofibrations. We need to show any map of P-algebras of the form X → Y which is obtained as a transfinite composition of pushouts of maps of the form P(f), for f ∈ S, is an equivalence. Such a map can be written, as a transfinite composition, computed in C, of maps in SP. Since homotopy pushouts of equivalences are equivalences, any map in SP is an equivalence. Moreover homotopy colimits of equivalences are equivalences, and the result is proven. One can also show that if the class of acyclic cofibrations satisfies the weak P-algebra axiom relative to cofibrant P-algebras then one gets a semi-model structure, as in [WY18]. One just needs to repeat the proof of the theorem assuming that X is a cofibrant P-algebra.

We have the following useful trick for categories enriched over Q. It follows immediately from the fact

that for any right Σn module X and any map f of left Σn-modules, X ⊗Σn f is a retract of X ⊗ f. Proposition 6.3.23. Let C be a combinatorial model category which is enriched over Q, and is also a symmetric monoidal category. Let P be a symmetric operad in C. Suppose when regarded as a non-symmetric operad the class of acyclic cofibrations in C satisfies the weak P-algebra axiom. Then the class of acyclic cofibrations satisfies the weak P-algebra axiom when it is regarded as a symmetric operad. 6.3.2. Equivalences of Model Categories of Algebras. In this section we establish when Quillen equivalences of model categories with monoidal structures lift to Quillen equivalences modules. Our proofs are modifications of the proofs of [SS03] Theorem 3.12, Proposition 3.16, and particularly for the case of algebras over operads, [WY19] Theorem A.

Theorem 6.3.24. Let (C, ⊗C ) and (D, ⊗D ) be monoidal categories, (M , •M ) a C-module and (M , •N ) a D-module where M and N are combinatorial model categories. Let L: C D :R

L: M N :R be a lax monoidal pair of adjunctions such that the bottom adjunction is a Quillen equivalence of model categories. Let S be an associative monoid in C, T an associative monoid in D, and α : S → R(T ) a map of monoids such that (1) S is admissible in M (2) T is admissible in N (3) For any coﬁbrant B in SMod (M ) the map χB : LL|B|S → |LαB|T which is adjoint to the underlying map in M of the unit |B|S → R(|LαB|T ) is an equivalence. Then the adjunction Lα : SMod (M ) T Mod (M ):Rα is a Quillen equivalence. 110 6. HOMOTOPICAL ALGEBRA IN EXACT CATEGORIES

Proof. It is clearly a Quillen adjunction. Let B be a cofibrant S-module, and Y a fibrant T -module. We need to show that a map LαB → Y is an equivalence if and only if B → RαY is an equivalence. Let C → |B|S be a cofibrant replacement. Then LC → |LαB|T is an equivalence by assumption. Note also that |Y |T is fibrant in N . Now since L: M N :R ∼ is a Quillen equivalence, we have C → R|Y |T = |RαY |S is an equivalence if and only if LC → |Y |T is an equivalence. By the two-out-of-three property this implies that |LαB → Y |T is an equivalence if and only if |B → RαY |S is an equivalence, which completes the proof. We immediately get the following.

Corollary 6.3.25. Let (C, ⊗C ) and (D, ⊗D ) be monoidal categories, (M , •M ) a left C-module and (N , •N ) a left D-module which are both combinatorial model categories. Let

L: M N :R

L: M N :R be a lax monoidal pair of adjunctions such that the bottom adjunction is a Quillen equivalence of model categories. Suppose that L and L are strictly monoidal. Let α : S → R(T ) be an isomorphism of associative monoids in C, where S and T are admissible. Then the adjunction

Lα : SMod (M ) T Mod (N ):Rα is a Quillen equivalence. 6.3.2.1. Equivalences of Categories of Alegbras over Operads. Here we once again specialise to algebras over operads. Let (C, ⊗, k) be a closed monoidal category which is also a combinatorial model category. Our results here are adaptations of [WY19] to slightly more general conditions. Essentially, where in the commutative cubes below White/ Yau’s axioms implies that the front and back faces are homotopy pushouts, we shall impose these conditions.

Lemma 6.3.26. Let P be a non-symmetric (resp. symmetric) operad Y be an object of SullP(S; X) where

(1) LX → Lα(X) is an equivalence. (2) The collection of maps S satisﬁes the weak P-algebra axiom relative to X, and relative to (L; X). (3) The collection of maps {L(s): s ∈ S}satisﬁes the weak T-algebra axiom relative to L(X). 0 (4) For any s : A → B ∈ S any X ∈ SullP(S; X), and any n ≥ 0 the maps

n+1 n+1 n+1 n+1 0 0 0 0 L(PX (n+1)⊗Q(s)n ) → TX (n+1)⊗Qn (L(s)) (resp. L(PX (n+1)⊗Σn+1 Q(s)n ) → TX (n+1)⊗Qn (L(s)))

⊗(n+1) ⊗(n+1) ⊗(n+1) ⊗(n+1) 0 0 0 0 L(PX (n+1)⊗B ) → TX (n+1)⊗L(B) (resp. L(PX (n+1)⊗Σn+1 B ) → TX (n+1)⊗B ) are equivalences. Then the map

LY → LαY is an equivalence.

Proof. We ﬁrst prove that for any map s : A → B in S, and any object X such that LX → Lα(X) is an equivalence, in any pushout diagram P(A) / X

P(s) s0 P(B) / X0 6.3. MODEL CATEGORIES OF MONOIDS AND ALGEBRAS 111

0 0 0 we have that LX → LαX is an equivalence. In this case LαX is given by the pushout diagram

T(A) / Lα(X)

P(s) s0

0 T(B) / Lα(X )

0 0 0 Consider the standard ﬁltrations lim→n Xn and lim→n X‹n of X and LαX respectively. L(X ) is given by a

transﬁnite composition lim→n L(Xn) where L(Xn) → L(Xn+1) is given by the pushout

n+1 L(PX (n + 1) ⊗ Q(s)n / L(Xn)

n+1 L(PX (n + 1) ⊗ B ) / L(Xn+1) The map 0 0 LX → LαX is given by the colimit

L(X0) / L(X1) / ... / L(Xn) / ...

X‹0 / X‹1 / ... / X‹n / ...

where X0 = X and X‹0 = LαX. We prove by induction that each map L(Xn) → X‹n is an equivalence. Then since N is both weakly {L(s): s ∈ S}T-elementary and weakly L(S)P-elementary, this would give 0 0 that LX → LαX is an equivalence. Note that L(X0) → X‹0 is an equivalence by assumption. Also by assumption, in the commutative cube below

n+1 L(PX (n + 1) ⊗ Q(s)n ) / L(Xn)

t n+1 z TX (n + 1) ⊗ Q(L(s))n / X‹n

⊗(n+1) L(PX (n + 1) ⊗ B ) / L(Xn+1)

t z ⊗(n+1) TX (n + 1) ⊗ L(B) / X‹n+1 the front and back faces are homotopy pushouts. By assumption the back-to-front maps on the left-hand face are equivalences. By the induction hypothesis the top right back-to-front map is an equivalence. Thus the bottom right back-to-front map is also an equivalence. Now since L and Lα both commute with ﬁltered

(in fact all) colimits, the forgetful functor from Alg T(N ) to N commutes with filtered colimits, and N is T P both weakly {L(s): s ∈ S} -elementary and weakly L(S) -elementary, LαY is a transfinite composition of maps in the first class, and LY a transfinite composition of maps in the second, the map LY → LαY is an equivalence.

Corollary 6.3.27. Let (C, ⊗C ) and (D, ⊗D ) be (symmetric) monoidal categories, which are both combinatorial model categories. Let L: C D :R 112 6. HOMOTOPICAL ALGEBRA IN EXACT CATEGORIES

be a lax monoidal adjunction which is a Quillen equivalence of model categories. Let P be an admissible (symmetric) operad in C, T an admissible (symmetric) operad in D, and α : P → R(T) a map of operads. Finally, suppose that there is a generating set I of cofibrations in C such that (1) L preserves equivalence between the underlying objects of cofibrant P-algebras. (2) The map LL(P(0)) → T(0) is an equivalence. (3) I satisfies the weak P-algebra axiom relative to P(0) and relative to (L; P(0)). (4) {L(f): f ∈ I} satisfies the weak T-algebra axiom relative to T(L(0)). (5) Whenever X0 is a cofibrant P-algebra, s ∈ I, and n ≥ 0, the maps

n+1 n+1 n+1 n+1 0 0 0 0 L(PX (n+1)⊗Q(s)n ) → TX (n+1)⊗Qn (L(s))(resp. L(PX (n+1)⊗Σn+1 Q(s)n ) → TX (n+1)⊗Qn (L(s)))

⊗(n+1) ⊗(n+1) ⊗(n+1) ⊗(n+1) 0 0 0 0 L(PX (n+1)⊗B ) → TX (n+1)⊗L(B) (resp. L(PX (n+1)⊗Σn+1 B ) → TX (n+1)⊗B ) are equivalences. Then the adjunction

Lα : Alg P(M ) Alg T(N ):Rα is a Quillen equivalence.

n n Remark 6.3.28. We are implicitly using here the natural transformation L(Qt (s)) → Qt (L(s)) coming from the oplax monoidal structure of L.

Proof. The assumptions and Lemma 6.3.26 imply that for any coﬁbrant algebra X, the map

LL|X|P → |LαX|T is an equivalence. Then the result follows from Theorem 6.3.24. 6.3.2.2. Commutative Monoids and Properness. Consider the commutative operad Comm. For X a commutative monoid, the Σ-module CommA is given by CommX (n) = X. In particular for Z a left Σn- ∼ module, CommX (n) ⊗Σn Z = (Z)Σn . Thus a class of maps S satisﬁes the weak Comm-axiom relative to X if the square below is a homotopy pushout for any n ≥ 1 and any 0 < q < n.

n X ⊗ (Qn−1(s))Σn / Xn−1

⊗n X ⊗ (B )Σn / Xn

and C is SComm;X -elementary. For S the class of acyclic cofibrations, this is implied by White’s commutative monoid axiom, and for S the class of cofibrations, this is implied by White’s strong commutative monoid axiom [Whi17] Definition 3.1. In [Whi17] Theorem 4.17, White gives conditions under which the model category Alg Comm(C) is left proper. Following the strategy of White’s proof, let us prove a generalisation. First we will need a definition.

Definition 6.3.29. Let C be a symmetric monoidal category equipped which is also a model category. An object X of C is said to be K-flat if for any equivalence f, X ⊗ f is also an equivalence. Theorem 6.3.30. Suppose (1) Comm is admissible. (2) E is left proper. n+1 ⊗n+1 (3) If f : X → Y is an acyclic cofibration, then (Qn (f))Σn+1 and (Y )Σn+1 are K-flat for all n ≥ 0. (4) The class S of cofibrations in C satisfies the weak Comm-algebra axiom.

Then the category Alg Comm(E) is left proper. 6.3. MODEL CATEGORIES OF MONOIDS AND ALGEBRAS 113

Proof. It suffices to prove that for g : X → Y a cofibration in E, the map S(g): S(X) → S(Y ) is a h-cofibration. Cosider a commutative diagram

f S(X) / A / B

S(g) g0 f 0 S(Y ) / A0 / B0 in which both squares are pushouts and f is an equivalence. We need to show that f 0 is an equivalence. The map f 0 is given by the colimit of the maps on standard ﬁltrations

A0 / A1 / ... / An / ...

f f1 fn B0 / B1 / ... / Bn / ... where A0 = A, B0 = B, and An+1, Bn+1 ﬁt into commutative cubes.

n+1 A ⊗ (Qn (g))Σn+1 / An

n+1 u | B ⊗ (Qn (g))Σn+1 / Bn

⊗n +1 A ⊗ (Y )Σn+1 / An+1

⊗n +1 u | B ⊗ (Y )Σn+1 / Bn+1

n+1 ⊗n+1 Since (Qn (g))Σn+1 and (Y )Σn+1 are K-ﬂat, the horizontal maps on the left face are equivalences. The map An → Bn is an equivalence by the inductive step. The back-left and front-left maps are left proper by assumption, and the front and back faces are pushouts, and therefore homotopy pushouts. Hence An+1 → Bn+1 is an equivalence. Suppose for a moment that C be a pointed model category. The following result will be important for us.

Lemma 6.3.31. Let C is a pointed model category. Suppose that the category Alg Comm(C) is equipped with nu the tranferred model structure (resp. is left proper). Then the category Alg Comm(C) of non-unital commutative monoids is equipped with the trasferred model structure (resp. is left proper). nu Proof. Let X0 → lim→α<λ Xα be a transfinite composition of pushouts in Alg Comm(C) of maps of nu ` ` the form S (f), with f a trivial cofibration. Then k X0 → lim→ k Xα is a transfinite composi- α<λ ` tion of pushouts in Alg (C) of maps of the form S(f), with f a trivial cofibration. Thus k X0 → ` Comm lim→α<λ k Xα is an equivalence. As a retract of this map, X0 → lim→α<λ Xα is an equivalence.

Suppose now Alg Comm(C) is left proper. Consider a pushout.

f Snu(X) / A / B

Snu(g) g0 f 0 Snu(Y ) / A0 / B0 114 6. HOMOTOPICAL ALGEBRA IN EXACT CATEGORIES

in which both squares are pushouts and f is an equivalence. Then

k ` f S(X) / k ` A / k ` B

S(g) k ` g0 k ` f 0 S(Y ) / k ` A0 / k ` B0 ` ` 0 is a diagram in Alg Comm(C) with both squares pushouts and k f an equivalence. Thus k f is an ` 0 0 equivalence. So, as a retract of k f , f is an equivalence. 6.4. Homotopical Algebra in Exact Categories In this section we let (E, ⊗, k) be a complete and cocomplete symmetric monoidal exact category which is quasi-elementary. We do not assume that projectives are flat, nor that the tensor product of two projectives is projective. Then the projective model structure exists on Ch≥0(E), Ch(E), Ch≥0(E), sE, and csE. These are not, in general, monoidal model categories. However the cofibrations in all of these categories are degree- wise/ level-wise split, and hence h-cofibrations (and therefore left proper). Moreover the tensor product of a trivially cofibrant object with any object is still trivial. Thus if f is an acyclic cofibration, then any pushout of a tensor product of f with any object will still be a weak equivalence. Finally, if E is weakly AdMon-elementary as an exact category, then they are all weakly AdMon-elementary as model categories. Thus we get the following.

Theorem 6.4.32. Let M be one of the additive model categories Ch≥0(E), Ch(E), Ch≥0(E), sE, csE, and let P be a non-symmetric operad in M . Then P is admissible. If M is enriched over Q then any symmetric operad is also admissible.

6.4.1. The Monoidal Dold-Kan Correspondence. Following [SS03] closely, in this section we show that the Dold-Kan equivalence is a monoidal Quillen equivalence, and deduce equivalences of categories of algebras over operads. We will also generalise the work of [CC04] on the cosimplicial Dold-Kan correspondence. Let E be an idempotent complete additive category. Suppose that it is equipped with an additive monoidal structure ⊗ : E × E → E. As explained in the introduction Ch≥0(E) has a natural monoidal structure. sE also has a monoidal structure deﬁned level-wise: if X• and Y• are objects of sE then X• ⊗ Y• is the simplicial object which in level n is Xn ⊗ Yn. In the Dold-Kan adjunction

Γ: Ch≥0(E) sE :N both the left and right adjoints are lax and oplax monoidal. Indeed it suﬃces to show that the right adjoint is both lax and oplax monoidal, then Γ inherits the same structures via doctrinal adjunction. For A and B objects of sE, deﬁne the Alexander-Whitney map

∆A,B : C(A ⊗ B) → C(A) ⊗ C(B)

on the factor An ⊗ Bn by M p q d‹ ⊗ d0 p+q=n where d‹p is induced by the map [p] → [p + q], i 7→ i q and d0 is induced by the map [q] → [p + q], i 7→ i + p Deﬁne the Eilenberg-Zilber map

∇A,B : C(A) ⊗ C(B) → C(A ⊗ B)

on the factor Ap ⊗ Bq by X sign(µ, ν)sν ⊗ sµ (µ,ν) 6.4. HOMOTOPICAL ALGEBRA IN EXACT CATEGORIES 115 where the sum is over all (p, q)-shuﬄes (µ, ν) = (µ1, . . . , µp, ν1, . . . , νq), and

. sµ = sµp−1 ◦ ... ◦ sµ2−1 ◦ sµ1−1 and similarly for sν . Both ∆A,B and ∇A,B are functorial in A and B, and restrict to maps

∆A,B : N(A ⊗ B) → N(A) ⊗ N(B)

∇A,B : N(A) ⊗ N(B) → N(A ⊗ B) Remark 6.4.33. It is shown in e.g. [Wei95] 8.5.4 that for simplicial abelian groups ∇ is lax symmetric monoidal, and the proof works for arbitrary additive categories with kernels. Theorem 6.4.34 ([Lur21]). Let A and B be simplicial objects.

(1) ∆A,B ◦ ∇A,B = IdN(A)⊗N(B) (2) The Eilenberg-Zilber map

∆A,B : N(A ⊗ B) → N(A) ⊗ N(B) is a homotopy equivalence. Proof. (1) This can be proven exactly as in [Lur21] Section 2.5.8. (2) This can be proven as in [Lur21] Section 2.5.8, with some small adjustments. It suﬃces to show that for any complex M, the map

∇A,Γ(M) N(A) ⊗ M =∼ N(A) ⊗ N(Γ(M)) / N(A ⊗ Γ(M))

is a homotopy equivalence. M can be written as an (ℵ0; SplitMon)-colimit of bounded above n n n complexes M , where Mk = Mk for k ≤ n, and Mk = 0 for k > n. Thus we may assume that M is bounded above, and prove the claim by induction on the length of M. Let M be of length 0 n. Denote by Mn[n] the complex given by Mn concentrated in degree n, and let M denote the kernel of the map M → Mn[n]. We get degree-wise split exact sequence of complexes and simplicial objects respectively 0 0 → M → M → Mn[n] → 0 0 0 → Γ(M ) → Γ(M) → Γ(Mn[n]) → 0 Thus tensoring the ﬁrst sequence with a complex and the second with a simplicial object will still result in a degree-wise split exact sequence. In particular get a commutative diagram of degree-wise split exact sequences 0 0 / N(A) ⊗ M / N(A) ⊗ M / N(A) ⊗ Mn[n] / 0

0 0 / N(A ⊗ Γ(M )) / N(A ⊗ Γ(M)) / N(A ⊗ Γ(Mn[n])) / 0 If both the ﬁrst and last vertical maps are homotopy equivalences then the middle one will be as well. It follows by induction that we may assume that M is of the form Y [n] =∼ N(Z[∆n)] ⊗ Y for some n ≥ 0 and some object Y of E. By a similar inductive argument, we may also assume that A is of the form N(Z[∆m]) ⊗ X for some m ≥ 0 and some object X of E. Thus we reduce to showing m n that the map ∇N(Z[∆ ])⊗X,N(Z[∆ ])⊗Y , which is given by the composition m n m n m n (N(Z[∆ ]) ⊗ X) ⊗ (N(Z[∆ ]) ⊗ Y ) =∼ N(Z[∆ ]) ⊗ N(Z[∆ ]) ⊗ X ⊗ Y → N(Z[∆ ⊗ ∆ ]) ⊗ X ⊗ Y is a homotopy equivalence. By the classical result for abelian groups, e.g. [Lur21] Theorem 2.5.7.14, the map m n m n N(Z[∆ ]) ⊗ N(Z[∆ ]) → N(Z[∆ ⊗ ∆ ]) is a homotopy equivalence. Thus m n m n m n (N(Z[∆ ]) ⊗ X) ⊗ (N(Z[∆ ]) ⊗ Y ) =∼ N(Z[∆ ]) ⊗ N(Z[∆ ]) ⊗ X ⊗ Y → N(Z[∆ ⊗ ∆ ]) ⊗ X ⊗ Y 116 6. HOMOTOPICAL ALGEBRA IN EXACT CATEGORIES

is also a homotopy equivalence Recall the oplax structure inherited by the functor Γ, which we denote by ∇c, is deﬁned on complexes X and Y by the composition

Γ(∇Γ(X),Γ(Y )) Γ(X ⊗ Y ) / Γ(NΓ(X) ⊗ NΓ(Y )) / ΓN(Γ(X) ⊗ Γ(Y )) / Γ(X) ⊗ Γ(Y ) Since all maps in the composite are homotopy equivalences, we get the following.

Corollary 6.4.35. Let X and Y be objects in Ch≥0(E). Then the map c ∇X,Y : Γ(X ⊗ Y ) → Γ(X) ⊗ Γ(Y ) is a homotopy equivalence. 6.4.1.1. The Cosimplicial Monoidal Dold-Kan equivalence. Let E be an cocomplete additive category equipped with a closed monoidal structure, and let A, B ∈ Ch≤0(E). Deﬁne ν : QA ⊗ QB → Q(A ⊗ B) i n j m on An ⊗ T (V ) ⊗ Bm ⊗ T (V ) by n ν = IdAn ⊗ IdBm ⊗ µ + (−1) IdAn ⊗ dB ⊗ µ ◦ (θ ⊗ Id) The following is proven formally as in [CC04], Theorem 5.2.

Lemma 6.4.36. ν is a natural isomorphism, and makes Q : Ch≤0(E) → csE a strong monoidal functor. 6.4.1.2. Dold-Kan Equivalences for Algebras. The results in this section are obtained in [WY19] for the category of k-modules where k is a unital commutative ring. The Dold-Kan equivalence for algebras was of course famously proved in [SS03] for the category of abelian groups, and the dual Dold-Kan correspondence for abelian groups in [CC04]. Let (E, ⊗, Hom, k) be a complete and cocomplete closed symmetric monoidal AdMon-elementary exact category. We suppose that Ch≥0(E) and sE are equipped with the projective model structures. Let T be an admissible non-symmetric operad in sE, P an admissible non-symmetric operad in Ch≥0(E), and α : P → N(T) a homotopy equivalence. Then we have the following. Theorem 6.4.37 (The Simplicial Dold-Kan Equivalence for Algebras). The adjunction

Γα : Alg P(Ch≥0(E)) Alg T(sE):Nα is a Quillen equivalence.

Now suppose that E is enriched over Q, let T be an admissible symmetric operad in sE, P an admissible symmetric operad in Ch≥0(E), and α : P → N(T) a homotopy equivalence of symmetric operads. The maps n+1 n+1 ⊗(n+1) ⊗(n+1) L(PX0 (n + 1) ⊗ Q(s)n ) → TX0 (n + 1) ⊗ Qn (L(s)) L(PX0 (n + 1) ⊗ B → TX0 (n + 1) ⊗ L(B) are homotopy equivalences. Since E is enriched over Q, for any n and any P-algebra X, the maps n+1 n+1 ⊗(n+1) ⊗(n+1) 0 0 0 0 L(PX (n+1)⊗Σn+1 Q(s)n ) → TX (n+1)⊗Qn (L(s)) L(PX (n+1)⊗Σn+1 B → TX (n+1)⊗L(B) are retracts of the ﬁrst pair of maps, and are therefore also homotopy equivalences. Thus we get the following Theorem 6.4.38 (The Simplicial Dold-Kan Equivalence for Symmetric Algebras). The adjunction

Γα : Alg P(Ch≥0(E)) Alg T(sE):Nα is a Quillen equivalence for P and T symmetric operads as above. Just as for Σ-coﬁbrant (coloured) operads in [WY19] Section 4.3, in general when E is not enriched over Q, one can work with Σ-K-ﬂat symmetric operads, though we will not go into this here. Ch≤0(E) is equipped with a combinatorial model structure. Again we assume that homotopy equivalences are weak equivalences. We equip csE with the model structure transferred along the dual Dold-Kan equivalence. The functor Q is strongly monoidal. Thus by Corollary 6.3.25 we get the following. 6.4. HOMOTOPICAL ALGEBRA IN EXACT CATEGORIES 117

Theorem 6.4.39. Let P be an admissible non-symmetric operad in Ch≤0(E), T be an admissible non- symmetric operad in cs(E), and α : P → H(T) a homotopy equivalence of operads. Then the adjunction

Qα : Alg P(Ch≤0)(E) Alg T(csE):Hα is a Quillen equivalence. 6.4.2. Homotopical Algebra Contexts. Before concluding let us make a connection with geometry. Recall that in [TTV08] Toënand Vezzosi introduce an abstract categorical framework in which one can ‘do’ homotopical algebra, namely a homotopical algebra context. Let us recall the truncated definition (for the category C0 in [TTV08] we always take C = C0). Definition 6.4.40. Let M be a combinatorial symmetric monoidal model category. We say that M is an homotopical algebra context (or HA context) if for any A ∈ Alg Comm(M ). (1) The model category M is proper, pointed and for any two objects X and Y in M the natural morphisms a a QX QY → X Y → RX × RY are equivalences. (2) Ho(M ) is an additive category. (3) With the transferred model structure and monoidal structure − ⊗A −, the category AMod is a combinatorial, proper, symmetric monoidal model category. (4) For any cofibrant object M ∈ AMod the functor

− ⊗A M : AMod → AMod preserves equivalences. (5) With the transferred model structures (A ) and (A ) are combinatorial Alg Comm Mod Alg Commnu Mod proper model categories.

(6) If B is cofibrant in Alg Comm(AMod ) then the functor B ⊗A − : AMod → BMod preserves equivalences. Theorem 6.4.41. Let (E, ⊗, Hom, k) be a locally presentable closed projectively monoidal exact category which is AdMon-elementary. Then Ch≥0(E) and Ch(E) are homotopical algebra context. Proof. The natural maps a a QX QY → X Y → RX × RY are clearly equivalences. All that remains to prove is the final property. Now if B is a cofibrant A-algebra then it is a retract of the free A-algebra on a cofibrant A-module. But the free A-algebra on a cofibrant A-module is cofibrant as an A-module. Hence B ⊗A (−) preserves equivalences by 4). APPENDIX A

Model Categories

A.1. Weak factorisation Systems and Model Structures Here we brieﬂy recall the deﬁnition of a model structure by means of weak factorisation systems. Details can be found in [Rie14]. Definition A.1.1. Let C be a class of morphisms in a category M . A morphism f in M is said to have the left lifting property with respect to C if in any diagram of the form A / C

f c B / D with c ∈ C, there exists a morphism h : B → C such that the following diagram commutes A / C > h f c B / D We denote the class of all morphisms which have the left lifting property with respect to C by C. Dually one defines the morphisms having the right lifting property with respect to C. The class of all such morphisms is denoted C. The following is straightforward Proposition A.1.2. Let C be a class of morphisms in a category M . Then C is closed under retracts, pushouts and transfinite composition (whenever they exist). Proof. See [Rie14] Lemma 11.1.4. Definition A.1.3. A weak factorisation system on a category C is a pair (L, R) such that (1) Any map in C can be factored as a map in L followed by a map in R. (2) L = R and R = L. A weak factorisation system is said to be functorial if the factorisation in (1) can be made functorial. We can now give a definition of the notion of a model structure in terms of weak factorisation systems. Definition A.1.4. A model structure on a category M is a collection of three wide subcategories (C, F, W) such that (1) The class W satisfies the 2-out-of-6 property (see [Rie14]). (2) Both (C ∩ W, F) and (C, W ∩ F) are weak factorisation systems. We do not assume completeness or cocompleteness of M . Definition A.1.5. A model structure on a category M is said to be functorial if the factorisation systems are functorial. Definition A.1.6. A (functorial) model category is a category together with a (functorial) model structure.

118 A.3. HOMOTOPY COLIMITS 119

A.2. Cofibrant Generation We state here our conventions regarding cofibrant generation. These are largely slightly modified definitions from [Hov99] Chapter 2. Definition A.2.7. If I is a collection of maps in category C, we denote by cell(I) the collection of transfinite compositions of pushouts of maps in I. Definition A.2.8. If I is a collection of maps in category C, we say that I satisfies the small object argument if any transfinite composition of pushouts of morphisms in I exists, and any morphism f has a factorisation f = h ◦ g where g ∈ cell(I) and h ∈ I. Definition A.2.9. Let C be a category. A weak factorisation system (L, R) on C is said to be cofibrantly small if there is a set I of maps in L such that R = I. I is called a set of generating morphisms. If in addition I admits the small object argument then the weak factorisation system is said to be cofibrantly generated. If C is locally presentable and cofibrantly generated, then the weak factorisation system is said to be combinatorial. A model category (C, W, F) is said to be cofibrantly small/ cofibrantly generated/ combinatorial if both the weak factorisation systems (C, F ∩ W) and (C ∩ W, F) are cofibrantly small/ cofibrantly generated/ combinatorial. Remark A.2.10. A cofibrantly small weak factorisation system (resp. model structure) on a locally presentable category is automatically combinatorial.

A.3. Homotopy Colimits Let C be a combinatorial model category, and I a small category. The category of functors Fun(I, C) can be equipped with the projective model structure [Rie14] Theorem 12.3.2. The functor colimI : Fun(I, C) → C is left Quillen ([Rie14] Corollary 5.1.3). For a functor F ∈ Fun(I, C), the homotopy colimit of F , denoted hocolimI (or holim→I for filtered I) is the derived functor of colimI applied to F . One defines homotopy limits dually, and denotes them by holimI , (or holim←I for cofiltered I). A.3.1. Homotopy Pushouts and Properness. Properness is a useful condition to have on model categories and on maps in model categories. Definition A.3.11 ([Whi14] Definition 8.1.). (1) A map f : X → Y in a model category C is said to be a h-cofibration if whenever

g w X / A. / B

f 0 g w0 Y / A0 / B0 is a commutative diagram in which both squares are pushouts, and w a weak equivalence, then w0 is a weak equivalence. (2) A model category is said to be left proper if cofibrations are h-cofibrations. One defines h-fibrations of maps and right properness of model categories dually. Proposition A.3.12 ([BB13] Proposition 1.6). A map f : X → Y in a left proper model category C is a h-cofibration if and only if any pushout diagram X / A

f Y / P is a homotopy pushout. This motivates the following deﬁnition. 120 A. MODEL CATEGORIES

Definition A.3.13. Let C be a model category and P ⊂ C a subcategory C. A map f : X → Y in C is said to be left proper relative to P if any pushout diagram X / A

f Y / P with A ∈ P is a homotopy pushout. If P = C, then f is said to be left proper. One defines relative right-properness dually. In particular, in a proper model category h-cofibrations are left proper. Let us make some straightforward observations. Proposition A.3.14. Let S be a pushout-stable class of maps such that any pushout of a weak equivalence along a map in S is still a weak equivalence. Then any map in S is a h-cofibration. The following is just a consequence of the 2-out-of-3 property of weak equivalences. Proposition A.3.15. Let f be a weak equivalence such that any pushout of f is a weak equivalence. Then f is left proper, and a h-cofibration. A.3.1.1. Stable Model Categories. Let C be a combinatorial pointed model category with initial-terminal object 0. The suspension functor Σ : Ho(C) → Ho(C) assigns to an object X the homotopy pushout `L 0 X 0. Definition A.3.16 ([Hov99], Definition 7.1.1). A pointed model category C is said to be stable if the functor Σ is an auti-equivalence. A.3.2. Homotopy Transfinite Compositions. Definition A.3.17. Let S be a collection of maps in a co-complete model category C. C is said to be cocont weakly S-elementary if for any ordinal λ, and any functor X ∈ FunS (λ, C) , the colimit colimα<λXα is a homotopy colimit. Example A.3.18. (1) If S is the class of cofibrations then C is weakly S-elementary. (2) If S is a class of weak equivalences such that any transfinite composition of maps in S is a weak equivalence, then it follows from the 2-out-of-3 property that C is weakly S-elementary. The following is straightforward. Proposition A.3.19. Let S be a class of maps in a model category C. Suppose that any map in S is a h-cofibration, and that C is weakly Sp-elementary, where Sp is the class of pushouts of maps in C. Then any map obtained as a transfinite composition of pushouts of maps in S is a h-cofibration.

A.4. Monoidal Model Categories Definition A.4.20. Let M , N , P be model categories. A bifunctor − ⊗ − : M × N → P is said to be left Quillen if whenever i : m → m0 and j : n → n0 are coﬁbrations then so is i⊗ˆ j, and it is an acyclic coﬁbration if either i or j is. Here i⊗ˆ j is the following map

i⊗1 m ⊗ n / m0 ⊗ n

1⊗j 1⊗j m ⊗ n0 / P i⊗ˆ j % i⊗1 m0 ⊗ n0 2 where the square is a pushout. A.5. TRANSFERRED MODEL STRUCTURES 121

Definition A.4.21. A (closed) monoidal model category is a (closed) symmetric monoidal category (V, ⊗, k) ((V, ⊗, k, Hom)) with a model structure so that the monoidal product is a left Quillen bifunctor, and the maps Q(k) ⊗ v → k ⊗ v =∼ v and v ⊗ Q(k) → v ⊗ k =∼ v are weak equivalences whenever v is cofibrant. Here Q is the cofibrant replacement functor. Another condition that is often asked of a monoidal model category is that it satisfies the so-called monoid axiom. Under certain additional technical assumptions on the model category, this guarantees the existence of a model structure on the category of algebras over any cofibrant operad. Definition A.4.22. A monoidal model category (V, ⊗, k) is said to satisfy the monoid axiom if every morphism which is obtained as a transfinite composition of pushouts of tensor products of acyclic cofibrations with any object is a weak equivalence.

A.4.1. Two-Variable Quillen Adjunctions. Let (C, D, E) be a triple of model categories. Recall (Section 10.1 [Rie14]) that a two-variable adjunction on (C, D, E) is a triple of functors op o ⊗ : C × D → E Homl : C × E → D Homr : D p × E → C together with natural isomorphisms ∼ ∼ HomE (c ⊗ d, e) = HomC (c, Homl(d, e)) = HomD (d, Homr(c, e)) The following is from Section 11.4 in [Rie14].

Definition A.4.23. Let (C, D, E) be model categories. A two-variable ajdunction on (⊗, Homl, Homr) on (C, D, E) is said to be a Quillen two-variable adjunction if ⊗ is a left Quillen bifunctor. Definition A.4.24. Let C, D be model categories. If there exists a Quillen two-variable adjunction on (C, D, D) then D is said to be a C-model category. In particular, following Deﬁnition 11.4.4 of [Rie14], a simplicial model category is just a sSet-model category. The following is clear.

Proposition A.4.25. . Let (⊗, Homl, Homr) be a two-variable Quillen adjunction on model categories (C, D, E). Let 0 L: C C :R be a Quillen adjunction. Then (L(−) ⊗ L(−), Homl(L, −),R ◦ Homr) is a Quillen two-variable adjunction on (C 0, D, E). In particular if D is a C-model category then it is also a C 0-model category.

A.5. Transferred Model Structures Definition A.5.26. Let C and D be categories with C a model category. Suppose F : C → D and G : D → C are functors with F a G. If it exists, the transferred model structure on D is the one defined as follows. (1) A map f in D is a weak equivalence precisely if G(f) is a weak equivalence in C. (2) A map f in D is a fibration precisely if G(f) is a fibration in C. (3) A map f in D is a cofibration precisely if it has the left lifting property with respect to acyclic cofibrations.

Remark A.5.27. If the transferred model structure exists on D then F a G is a Quillen adjunction. We need the following important result, which is essentially Theorem 3.3 in [Cra95]. Although the conditions are marginally diﬀerent, the proof is identical. 122 A. MODEL CATEGORIES

Theorem A.5.28. Suppose F : C → D and G : D → C are functors with F a G. Suppose that C is finitely complete and cocomplete cofibrantly generated model category with generating cofibrations I and generating acyclic cofibrations J, and that D is a category having finite limits, sequential colimits, and pushouts along maps of the form F (f) where f is a coproduct of maps in I ∪ J. Let λ (resp λ0) be an ordinal such that domains of generating cofibrations (resp. domains of generating acyclic cofibrations) are λ-presented (resp. λ0-presented) relative to pushouts of coproducts of maps in I (resp. J). Suppose that if c is a generating cofibration (resp. generating acyclic cofibration) then the domain of F (c) is λ-presented (resp. λ0-presented ) relative to pushouts of coproducts of maps of the form F (f) where f is a generating cofibration (resp. generating acyclic cofibration). Then the transferred model structure on E exists if and only if the weak equivalences in D contain any sequential colimit of pushouts of maps of the form F (g), where g is a generating trivial cofibration in C. Corollary A.5.29. Let C and D be categories, with C a cococomplete combinatorial model category and D having finite limits and all colimits. Suppose F : C → D and G : D → C are functors with F a G. If G preserves filtered colimits, then the transferred model structure on D exists if and only if the weak equivalences in D contain any map of the form F (g), where g is a generating trivial cofibration in C. Moreover the transferred model structure is cofibrantly generated. Remark A.5.30. Note that in [Cra95] it is actually proved that if an adjunction satisfying the above condition then the transferred model structure exists. The converse is clear however since as a left Quillen functor F preserves acyclic cofibrations and colimits. Bibliography

[AR94] Jiˇr´ıAdámekand Jiˇr´ıRosicky, Locally presentable and accessible categories, vol. 189, Cambridge University Press, 1994. [Bak11] Andrew Baker, An Introduction to p -adic Numbers and p -adic Analysis. [Ban17] Abhishek Banerjee, Noetherian schemes over abelian symmetric monoidal categories, International Journal of Mathematics 28 (2017), no. 07, 1750051. [BB13] Michael Batanin and Clemens Berger, Homotopy theory for algebras over polynomial monads, arXiv preprint arXiv:1305.0086 (2013). [BBB16] Federico Bambozzi and Oren Ben-Bassat, Dagger Geometry as Banach Algebraic Geometry, Journal of Number Theory 162 (2016), 391–462. [BBBK18] Federico Bambozzi, Oren Ben-Bassat, and Kobi Kremnizer, Stein Domains in Banach Algebraic Geometry, Journal of Functional Analysis 274 (2018), no. 7, 1865–1927. [BBK13] Oren Ben-Bassat and Kobi Kremnizer, Non-Archimedean Analytic Geometry as Relative Algebraic Geometry, arXiv preprint arXiv:1312.0338 (2013). [BBKK] Oren Ben-Bassat, Jack Kelly, and Kobi Kremnitzer, A Perspective on the Foundations of Derived Analytic Ge- ometry, Forthcoming. [BC13] Silvana Bazzoni and Septimiu Crivei, One-sided exact categories, Journal of Pure and Applied Algebra 217 (2013), no. 2, 377–391. [BMSG+10] Paul Frank Baum, Ralf Meyer, RubénSánchez-Garc´ıa,Marco Schlichting, and Bertrand Toën, Topics in Algebraic and Topological K-theory, Springer, 2010. [Bor94] Francis Borceux, Handbook of Categorical Algebra. 2, Volume 51 of Encyclopedia of Mathematics and its Appli- cations, 1994. [Buc59] David A Buchsbaum, A Note on Homology in Categories, Annals of Mathematics (1959), 66–74. [Büh10] Theo Bühler, Exact Categories, Expositiones Mathematicae 28 (2010), no. 1, 1–69. [Büh11] , On the Algebraic Foundations of Bounded Cohomology, American Mathematical Soc., 2011. [CC04] JoséLuis Castiglioni and Guillermo Cortiñas, Cosimplicial versus dg-rings: a version of the dold–kan correspondence, Journal of Pure and Applied Algebra 191 (2004), no. 1-2, 119–142. [CCN21] Damien Calaque, Ricardo Campos, and Joost Nuiten, Lie algebroids are curved lie algebras, arXiv preprint arXiv:2103.10728 (2021). [CH02] J Daniel Christensen and Mark Hovey, Quillen model structures for relative homological algebra, Mathematical Proceedings of the Cambridge Philosophical Society, vol. 133, Cambridge University Press, 2002, pp. 261–293. [CNPS17] Wojciech Chacholski, Amnon Neeman, Wolfgang Pitsch, and Jerome Scherer, Relative homological algebra via truncations, arXiv preprint arXiv:1702.05357 (2017). [Cra95] Sjoerd E Crans, Quillen Closed Model Structures for Sheaves, Journal of Pure and Applied Algebra 101 (1995), no. 1, 35–57. [Del07] Pierre Deligne, CatégoriesTannakiennes, The Grothendieck Festschrift, Springer, 2007, pp. 111–195. [EGO17] Sergio Estrada, James Gillespie, and Sinem Odaba¸si, Pure exact structures and the pure derived category of a scheme, Mathematical Proceedings of the Cambridge Philosophical Society, vol. 163, Cambridge University Press, 2017, pp. 251–264. [Fre64] Peter Freyd, Abelian Categories: An Introduction to the Theory of Functors, no. 1, Harper & Row, 1964. [Gil04] James Gillespie, The Flat Model Structure on Ch(R), Transactions of the American Mathematical Society 356 (2004), no. 8, 3369–3390. [Gil07] , Kaplansky Classes and Derived Categories, Mathematische Zeitschrift 257 (2007), no. 4, 811–843. [Gil08] , Cotorsion Pairs and Degreewise Homological Model Structures, Homology, Homotopy Appl 10 (2008), no. 1, 283–304. [Gil11] , Model Structures on Exact Categories, Journal of Pure and Applied Algebra 215 (2011), no. 12, 2892– 2902. [Gil16a] , The derived category with respect to a generator, Annali di Matematica Pura ed Applicata (1923-) 195 (2016), no. 2, 371–402. [Gil16b] , Exact model structures and recollements, Journal of Algebra 458 (2016), 265–306. [GJ09] Paul G Goerss and John F Jardine, Simplicial Homotopy Theory, Springer Science & Business Media, 2009.

123 124 BIBLIOGRAPHY

[GP18] Owen Gwilliam and Dmitri Pavlov, Enhancing the filtered derived category, Journal of Pure and Applied Algebra 222 (2018), no. 11, 3621–3674. [Har10] John E Harper, Homotopy theory of modules over operads and non-σ operads in monoidal model categories, Journal of Pure and Applied Algebra 214 (2010), no. 8, 1407–1434. [hc] Kevin Carlson (https://math.stackexchange.com/users/31228/kevin carlson), Applying Freyd-Mitchell’s embedding theorem on large categories, Mathematics Stack Exchange, URL:https://math.stackexchange.com/q/1006507 (version: 2014-11-04). [Hir09] Philip S Hirschhorn, Model Categories and Their Localizations, no. 99, American Mathematical Soc., 2009. [Hov99] Mark Hovey, Model Categories, no. 63, American Mathematical Soc., 1999. [Hov02] , Cotorsion Pairs, Model Category Structures, and Representation Theory, Mathematische Zeitschrift 241 (2002), no. 3, 553–592. [Joy08] AndréJoyal, Notes on quasi-categories, preprint (2008). [Kel74] G Max Kelly, Doctrinal adjunction, Category seminar, Springer, 1974, pp. 257–280. [Kel19] Jack Kelly, Koszul duality in exact categories, arXiv preprint arXiv:1905.10102 (2019). [KK82] Gregory Maxwell Kelly and Max Kelly, Basic concepts of enriched category theory, vol. 64, CUP Archive, 1982. [Kra12] Henning Krause, Approximations and adjoints in homotopy categories, Mathematische Annalen 353 (2012), no. 3, 765–781. [KS05] Masaki Kashiwara and Pierre Schapira, Categories and Sheaves, vol. 332, Springer Science & Business Media, 2005. [Lur09a] Jacob Lurie, Derived Algebraic Geometry V: Structured Spaces, arXiv preprint arXiv:0905.0459 (2009). [Lur09b] , Higher topos theory (am-170), Princeton University Press, 2009. [Lur21] Jacob Lurie, Kerodon, https://kerodon.net, 2021. [LV12] Jean-Louis Loday and Bruno Vallette, Algebraic Operads, vol. 346, Springer Science & Business Media, 2012. [Mey07] Ralf Meyer, Local and Analytic Cyclic Homology, vol. 3, European Mathematical Society, 2007. [MG15] Aaron Mazel-Gee, Quillen adjunctions induce adjunctions of quasicategories, arXiv preprint arXiv:1501.03146 (2015). [Mor20] TimothéeMoreau, Private communication, 2020. [Por15a] Mauro Porta, Derived complex analytic geometry I: GAGA theorems, arXiv preprint arXiv:1506.09042 (2015). [Por15b] , Derived complex analytic geometry II: square-zero extensions, arXiv preprint arXiv:1507.06602 (2015). [Por17] , The derived Riemann-Hilbert correspondence, arXiv preprint arXiv:1703.03907 (2017). [Pri17] JP Pridham, A K-theoretic interpretation of real Deligne cohomology, Advances in Mathematics 320 (2017), 795–826. [Pro00a] J.B. Prolla, Topics in Functional Analysis over Valued Division Rings, North-Holland Mathematics Studies, Elsevier Science, 2000. [Pro00b] Fabienne Prosmans, Derived Categories for Functional Analysis, Publications of the Research Institute for Math- ematical Sciences 36 (2000), no. 1, 19–83. [PY16] Mauro Porta and Tony Yue Yu, Higher analytic stacks and GAGA theorems, Advances in Mathematics 302 (2016), 351–409. [PY17] , Representability theorem in derived analytic geometry, arXiv preprint arXiv:1704.01683 (2017). [PY18a] , Derived Hom spaces in rigid analytic geometry, arXiv preprint arXiv:1801.07730 (2018). [PY18b] , Derived non-archimedean analytic spaces, Selecta Mathematica 24 (2018), no. 2, 609–665. [Rak20] Arpon Raksit, Hochschild homology and the derived de rham complex revisited, arXiv preprint arXiv:2007.02576 (2020). [Rie14] Emily Riehl, Categorical Homotopy Theory, no. 24, Cambridge University Press, 2014. [Roz99] JR Garc´ıaRozas, Covers and Envelopes in the Category of Complexes of Modules, vol. 407, CRC Press, 1999. [S+10] David I Spivak et al., Derived Smooth Manifolds, Duke Mathematical Journal 153 (2010), no. 1, 55–128. [Sch99] Jean-Pierre Schneiders, Quasi-abelian Categories and Sheaves, Sociétémathématiquede France, 1999. [Spa88] Nicolas Spaltenstein, Resolutions of Unbounded Complexes, Compositio Mathematica 65 (1988), no. 2, 121–154. [SS00] Stefan Schwede and Brooke E Shipley, Algebras and Modules in Monoidal Model Categories, Proceedings of the London Mathematical Society 80 (2000), no. 2, 491–511. [SS03] Stefan Schwede and Brooke Shipley, Equivalences of monoidal model categories, Algebraic & Geometric Topology 3 (2003), no. 1, 287–334. [SS11]ˇ Manuel Saor´ınand Jan St’ov´ıˇcek,ˇ On exact categories and applications to triangulated adjoints and model structures, Advances in Mathematics 228 (2011), no. 2, 968–1007. [SS16] Pierre Schapira and Jean-Pierre Schneiders, Derived categories of filtered objects, Astérisque 383 (2016), no. 103- 120, 7. [St’12]ˇ Jan St’ov´ıcek,ˇ Exact Model Categories, Approximation Theory, and Cohomology of Quasi-coherent Sheaves, Ad- vances in Representation Theory of Algebras, Bielefeld (2012), 297–367. [Sta21] The Stacks project authors, The stacks project, https://stacks.math.columbia.edu, 2021. [SW11] Dennis Sieg and Sven-Ake Wegner, Maximal exact structures on additive categories, Mathematische Nachrichten 284 (2011), no. 16, 2093–2100. BIBLIOGRAPHY 125

[TTV08] Bertrand Toën,Bertrand Toën,and Gabriele Vezzosi, Homotopical algebraic geometry ii: Geometric stacks and applications, vol. 2, American Mathematical Soc., 2008. [uh] user159931 (http://math.stackexchange.com/users/159931/user159931), The projective model structure on chain complexes, Mathematics Stack Exchange, URL:http://math.stackexchange.com/q/846202 (version: 2014-06-24). [Wei95] Charles A Weibel, An Introduction to Homological Algebra, no. 38, Cambridge university press, 1995. [Whi14] David White, Monoidal bousfield localizations and algebras over operads, arXiv preprint arXiv:1404.5197 (2014). [Whi17] , Model structures on commutative monoids in general model categories, Journal of Pure and Applied Algebra 221 (2017), no. 12, 3124–3168. [WY18] David White and Donald Yau, Bousfield localization and algebras over colored operads, Applied Categorical Struc- tures 26 (2018), no. 1, 153–203. [WY19] , Homotopical adjoint lifting theorem, Applied Categorical Structures 27 (2019), no. 4, 385–426. [YD14] Xiaoyan Yang and Nanqing Ding, On a Question of Gillespie, Forum Mathematicum, 2014.