introduction to the theory of derivators · introduction to the theory of derivators moritz groth...

140
INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely categorical approach to homological algebra and homotopical algebra, which was proposed independently by Grothendieck, Heller, Franke, and pos- sibly others. While stable derivators provide an enhancement of triangulated categories, the theory of derivators is more general in that it also applies to unstable homotopy theories (like the homotopy theory of topological spaces). As a slogan, derivators are minimal extensions of the more classical derived categories of abelian categories or homotopy categories of model categories to a framework with a well-behaved calculus of homotopy limits and homotopy Kan extensions. As we shall indicate here, this calculus encodes interesting constructions arising in various areas of pure mathematics. Starting with a short review of homological algebra including a discussion of derived categories, we spend some time giving a thorough motivation for the definition of a derivator. We use a proof of the result that (strong) sta- ble derivators take canonically values in triangulated categories as a pretext to study various basic aspects of derivators. As we shall see the calculus of homo- topy Kan extensions specializes for example to functorial cone constructions and functorial (higher) octahedron diagrams. One advantage of derivators over triangulated categories is that derivators admit exponentials, i.e., given a derivator and a small category, there is a corresponding derivator of diagrams of that fixed shape. As a consequence, we can apply derivators to abstract representation theory, and we conclude this book by an outlook on such applications. While classically in representation theory of quivers the focus is on representations over fields, some basic results extend to representations over rings, schemes, differential-graded algebras, ring spectra or in arbitrary abstract stable homotopy theories arising in algebra, geometry, and topology. Contents 1. Introduction and overview 3 1.1. From classical derived functors to derived categories 3 1.2. From derived categories to derivators 5 1.3. Derivators arising in algebra, geometry, and topology 7 1.4. Stable derivators as an enhancement of triangulated categories 8 1.5. Towards abstract representation theory 9 2. Abelian categories and classical derived functors 10 2.1. Review of abelian categories 10 2.2. Classical derived functors 13 2.3. Group cohomology as a derived limit 16 3. Derived categories of abelian categories 19 3.1. Towards derived cokernels 20 Date : April 25, 2015. 1

Upload: others

Post on 28-Jul-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS

MORITZ GROTH

Abstract. In this book we give an introduction to the theory of derivators, a

purely categorical approach to homological algebra and homotopical algebra,

which was proposed independently by Grothendieck, Heller, Franke, and pos-sibly others. While stable derivators provide an enhancement of triangulated

categories, the theory of derivators is more general in that it also applies to

unstable homotopy theories (like the homotopy theory of topological spaces).As a slogan, derivators are minimal extensions of the more classical derived

categories of abelian categories or homotopy categories of model categories to

a framework with a well-behaved calculus of homotopy limits and homotopyKan extensions. As we shall indicate here, this calculus encodes interesting

constructions arising in various areas of pure mathematics.Starting with a short review of homological algebra including a discussion

of derived categories, we spend some time giving a thorough motivation for

the definition of a derivator. We use a proof of the result that (strong) sta-ble derivators take canonically values in triangulated categories as a pretext to

study various basic aspects of derivators. As we shall see the calculus of homo-

topy Kan extensions specializes for example to functorial cone constructionsand functorial (higher) octahedron diagrams.

One advantage of derivators over triangulated categories is that derivators

admit exponentials, i.e., given a derivator and a small category, there is acorresponding derivator of diagrams of that fixed shape. As a consequence, we

can apply derivators to abstract representation theory, and we conclude this

book by an outlook on such applications. While classically in representationtheory of quivers the focus is on representations over fields, some basic results

extend to representations over rings, schemes, differential-graded algebras, ringspectra or in arbitrary abstract stable homotopy theories arising in algebra,

geometry, and topology.

Contents

1. Introduction and overview 31.1. From classical derived functors to derived categories 31.2. From derived categories to derivators 51.3. Derivators arising in algebra, geometry, and topology 71.4. Stable derivators as an enhancement of triangulated categories 81.5. Towards abstract representation theory 92. Abelian categories and classical derived functors 102.1. Review of abelian categories 102.2. Classical derived functors 132.3. Group cohomology as a derived limit 163. Derived categories of abelian categories 193.1. Towards derived cokernels 20

Date: April 25, 2015.

1

Page 2: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

2 MORITZ GROTH

3.2. Cones and cofibers 233.3. Derived categories and derived functors 263.4. Derived categories as 2-localizations 303.5. Cones as derived cokernels 323.6. Coherent versus incoherent diagrams 343.7. The case of category algebras 374. Derived categories as triangulated categories 414.1. The homotopy category of an abelian category 414.2. Triangulated categories 444.3. Exact morphisms and classical triangulations 474.4. Beyond triangulated categories 475. Kan extensions 495.1. Motivation and definition 495.2. Final functors 525.3. Pointwise Kan extensions 545.4. Basic properties and first examples 596. Basics on derivators 636.1. Prederivators 636.2. Derivators 676.3. Examples of derivators 706.4. Limits versus homotopy limits 737. Homotopy exact squares and Kan extensions 747.1. The calculus of mates 757.2. Homotopy exact squares 797.3. First applications to Kan extensions 808. Pointed derivators 858.1. Basics on pointed derivators 858.2. Suspensions, loops, cofibers, and fibers 868.3. Parametrized Kan extensions 908.4. Cartesian and cocartesian squares 938.5. Iterated cofiber constructions 969. Stable derivators 1019.1. Basics on stable derivators 1019.2. The preadditivity of stable derivators 1049.3. The additivity of stable derivators 1069.4. Morphisms and natural transformations 1089.5. Strongness of derivators 1129.6. Canonical triangulations in stable derivators 1129.7. Exact morphisms of stable derivators 11910. Towards abstract representation theory 12010.1. Morita equivalences 12010.2. Derived equivalences 12010.3. Strong stable equivalences 12110.4. Abstract tilting theory of An-quivers 12410.5. Universal tilting modules 128Appendix A. Some category theory 128A.1. Adjunctions 128A.2. Limits and colimits 131

Page 3: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 3

A.3. Basic 2-categorical terminology 134Appendix B. Examples of derivators 135B.1. Represented derivators 135B.2. Homotopy derivators of model categories 135References 137

1. Introduction and overview

This book offers an introduction to the theory of derivators, providing one ofthe many different approaches to homological algebra and homotopical algebra.Starting with the context of homological algebra, one goal of derivators is to addressthe following problem: the passage from the category Ch(A) of chain complexesin an abelian category A to the corresponding derived category D(A) results ina loss of information. In the theory of derivators, this problem is addressed byalso keeping track of derived categories of diagram categories D(AA), A ∈ Cat ,and various constructions between them, thereby enhancing the classical derivedcategories to a more flexible notion.

A very similar situation arises in homotopy theory where the passage from thecategory Top of topological spaces to the homotopy category Ho(Top) should besuitably refined. In this case, this is achieved by also considering homotopy cat-egories of diagram categories Ho(TopA), A ∈ Cat , together with suitable functorsbetween them. Among these functors are the homotopy limit and homotopy col-imit functors holimA,hocolimA : Ho(TopA)→ Ho(Top) as well as the more generalhomotopy Kan extension functors. These functors encode various interesting con-structions including reduced supensions, loop spaces, Borel constructions, homo-topy fixed points, and spectrifications of prespectrum objects.

It turns out that, besides these two key examples, there is a plethora of addi-tional situations leading to such ‘systems of diagram categories’ A 7→ D(A). Bydefinition a derivator simply axiomatizes key formal properties of such systems,leading to a purely categorical framework to study typical situations arising in al-gebra, geometry, and topology. In particular, homotopy (co)limits and homotopyKan extensions are characterized by ordinary universal properties, making themaccessible to elementary categorical techniques.

Derivators were introduced independently by Grothendieck [Gro], Heller [Hel88],Franke [Fra96], and possibly others. We refer the reader to the webpage mentionedin [Gro] for a comprehensive bibliography on the subject. While this book offer anintroductory account of basic aspects of the theory of derivators, we intend to comeback to a more systematic treatment somewhere else.

In the remainder of this introduction we try to fill the above paragraphs withmore life. This section is a rather informal account, while the precise mathematicsonly begin in §2. We include some references in the remainder of this introduction;further references can be found, together with the corresponding details, in theappropriate sections.

1.1. From classical derived functors to derived categories. To begin withone typical input leading to derivators, we stick to the framework of homological al-gebra [Wei94, GM03] and, say, consider categories of modules over rings, categoriesof abelian sheaves on topological spaces [?, ?], or more general abelian categories.

Page 4: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

4 MORITZ GROTH

Classical homological algebra is the study of such abelian categories and additivefunctors between them. Typical examples of such functors are tensor product andhom functors associated to bimodules over rings and direct and inverse image func-tors associated to continuous maps between topological spaces.

Let us recall that abelian categories essentially axiomatize key properties of thecalculus of kernels, cokernels, and short exact sequences of module homomorphisms.Correspondingly, there is the class of exact functors between abelian categories,i.e., functors preserving short exact sequences (these functors are, in particular,additive). However, typical examples of additive functors, like the ones mentionedabove, fail to be exact. In many cases the functors under consideration preserveexactness on one side, say such a functor F : A → B sends short exact sequences0 → X ′ → X → X ′′ → 0 in A to exact sequences FX ′ → FX → FX ′′ → 0in B. Since the kernel of the morphism FX ′ → FX is in general non-trivial, weare interested in systematic tools to study such kernels, and dually for functorspreserving exactness on the other side.

Classical homological algebra offers various techniques, based on projective, in-jective, and more general F -adapted resolutions, to define (classical) left and rightderived functors. In the case of a right exact functor F : A → B as above this leadsto additive left derived functors LkF : A → B, k ≥ 0, such that L0F ∼= F . Togetherwith suitable natural connecting homomorphisms δ = δk these functors associate toa short exact sequence 0→ X ′ → X → X ′′ → 0 in A a natural long exact sequence

. . .→ (L1F )(X)→ (L1F )(X ′′)δ→ FX ′ → FX → FX ′′ → 0

in B. These long exact sequences are convenient for both calculational and the-oretical purposes. Moreover, they can be characterized by universal properties interms of universal δ-functors [Gro57] as we recall in §2.

Implicitly in the above construction we use the trivial observation that everyadditive functor F : A → B yields by a levelwise application an additive functorF : Ch(A)→ Ch(B) between the corresponding categories of chain complexes. Oneeasily checks that the original functor F : A → B is exact if and only if the inducedfunctor F : Ch(A) → Ch(B) preserves quasi-isomorphisms. (Let us recall that achain map C → D is a quasi-isomorphism if the induced maps HkC → HkD,k ∈ Z, between homology objects are isomorphisms.) The point of this differentcharacterization is that it suggests the following perspective on derived functors:try to approximate the functor F : Ch(A)→ Ch(B) in a universal way by a functorwhich preserves quasi-isomorphisms.

The natural domain for such functors are the classical derived categories ofabelian categories [Ver96, KS06]. Ignoring set-theoretic issues for the time be-ing, let us recall that the derived category D(A) of A is obtained from Ch(A)by inverting the class of quasi-isomorphisms. The resulting localization functorγ : Ch(A) → D(A) has the defining universal property that every other functorCh(A) → C which sends quasi-isomorphisms to isomorphisms factors uniquelythrough γ.

To relate this to classical derived functors, we again consider our right exactfunctor F : A → B together with the induced functor F : Ch(A) → Ch(B). In

Page 5: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 5

gereral, it is not possible to find a dashed arrow in

Ch(A)F //

γ

=

Ch(B)

γ

D(A)@// D(B)

making the above diagram commute. As we recall in §3, a left derived functor orleft hyperderived functor is a functor LF : D(A)→ D(B) together with a universalnatural transformation LF γ → γ F . These desired ‘exact approximations’(and their duals) exist in typical situations and can be constructed by means ofsuitable resolutions. The classical derived functors are recovered by passing tohomology objects and using stalk complexes, i.e., there are natural isomorphisms(LkF )(X) ∼= (Hk LF )(X), k ∈ Z, where for X ∈ A we also denote the associatedcomplex concentrated in degree zero by the same symbol.

As an illustration of such left derived functors we establish the following examplein §3: given an arbitrary abelian category A, the classical construction of the coneof a chain map (together with a suitable universal natural transformation) simplyamounts to constructing the left derived cokernel functor. Thus, denoting theabelian category of morphisms in A by A[1], we consider the right exact cokernelfunctor F = cok: A[1] → A and obtain the cone functor as the corresponding leftderived functor, Lcok = C : D(A[1])→ D(A).

1.2. From derived categories to derivators. As we just recalled, the naturaldomain for derived functors are derived categories, and these derived categoriescome with localization functors γ : Ch(A)→ D(A). It is important to be aware ofthe fact that the categories Ch(A) and D(A) have rather different formal properties.While the category Ch(A) is again abelian and hence is, in particular, finitelycomplete and finitely cocomplete, the category D(A) is typically rather ill-behaved.Often the only limits and colimits which exist in D(A) are (finite) products andcoproducts. More importantly, even derived limits and derived colimits can neitherbe constructed nor characterized using the derived category D(A) only.

A classical way of adressing these bad properties of D(A) is by endowing themwith more structure. Derived categories are often considered as triangulated cate-gories [Pup67, Ver96, Nee01, HJR10], hence together with a triangulation consistingof an equivalence Σ: D(A)→ D(A) and a class of so-called distinguished triangles.These triangles are diagrams of the form X → Y → Z → ΣX, and this structure isthen subject to a certain list of axioms; see §4. A typical slogan is that the distin-guished triangles in D(A) are certain shadows of short exact sequences in Ch(A).In fact, they essentially arise from iterations of the cone construction.

The theory of triangulated categories is very sucessful and has been used for along time in various areas of pure mathematics, including algebra, algebraic ge-ometry, and topology, and this continues until today. Nevertheless, from the verybeginning on (see already the introduction to [Hel68]) it was obvious that the ax-ioms of a triangulated category have certain defects — a reminiscent of the fact thatthe passage from Ch(A) to D(A) results in a loss of information. We conclude §4by a short discussion of some of these defects. One goal of this book is to showthat these problems are avoided by the theory of stable derivators.

Page 6: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

6 MORITZ GROTH

As mentioned above, triangulations consist of additional structure imposed oncertain additive categories, like derived categories of abelian categories. The thirdobject in a distinguished triangle X → Y → Z → ΣX is often referred to as ‘thecone’ of the morphism X → Y . A crucial observation however is that in generalthese cones do not depend functorially on the morphism in D(A), i.e., in general,there is no cone functor D(A)[1] → D(A). But, as we already saw, there is afunctorial cone construction C = Lcok: D(A[1]) → D(A). To put it in words:there is a functorial cone construction defined on the derived category of the arrowcategory but not on the arrow category of the derived category. As an upshot, thissuggests that one should refine the passage A 7→ D(A) by also keeping track of thecategory D(A[1]).

To continue building towards derivators we make the following seemingly pickyobservation. Let us recall that the cokernel of a chain map f : X → Y is some‘colimit type construction’. In fact, while the cokernel is not the colimit of thediagram X → Y (which would be isomorphic to Y ), the cokernel cok(f) can beconstructed as the pushout on the right in

Xf// Y

Xf//

Y

Xf

//

Y

0 0 // cok(f).

(One easily checks that this reduces to the usual universal property of the cokernel.)The first above step extends a morphism of chain complexes to a span of chaincomplexes by adding the morphism X → 0, while the second step forms the pushoutsquare. From a more category theoretic perspective, this passage to the span andthe passage to the pushout square (as opposed to the pushout corner only) areexamples of Kan extensions [ML98]. In §5 we recall some basics concerning Kanextensions, including a few examples and first key properties.

As an upshot, this discussion suggests that in order to also encode a constructionof the cone functor C : D(A[1])→ D(A) at the level of derived categories, we should,using suggestive notation, also keep track of the derived categories D(Ap) andD(A) together with suitable functors between them. And for more sophisticatedpurposes it will be useful to also remember more general derived categories ofdiagram categories, say D(AA) for arbitrary small categories A. Pursuing thismore systematically, one is led to consider a 2-functor

DA : A 7→ D(AA),

the derivator DA of the abelian category A, thereby also keeping track of derivedrestriction functors. The derived category D(A) is recovered by considering dia-grams of trivial shape, and one goal of this book is to show that DA enhances D(A)to a more flexible notion.

As we discuss in §6, the notion of a derivator axiomatizes key properties of 2-functors like DA. For example, derived restriction functors have right adjoints andleft adjoints, given by derived Kan extension functors. As special cases, this yieldsderived limits and derived colimits. It turns out that the more general derived Kanextensions can be calculated pointwise in terms of certain derived limit and derivedcolimit expressions, and this property is turned into one of the key axioms of aderivator.

Page 7: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 7

1.3. Derivators arising in algebra, geometry, and topology. Before we ex-pand on the content of the remaining sections §§7-10, let us step back a little and seewhat we discussed so far. We essentially started with the pair (Ch(A),WA) consist-ing of the category of chain complexes together with the class of quasi-isomorphismsand refined the localization process

(Ch(A),WA) 7→ D(A) = Ch(A)[W−1A ].

But pairs of the form (Ch(A),WA) are only one class of examples of categoriestogether with a chosen class of morphisms ‘which we would like to treat as isomor-phisms’.

For example, in homotopy theory [?] one does not want to distinguish topologicalspaces as soon as they are connected by a zig-zag of weak homotopy equivalences.(Let us recall that a continuous map f : X → Y is a weak homotopy equivalence ifthe induced maps πk(X,x0)→ πk(Y, f(x0)), k ≥ 0, x0 ∈ X, are bijections.) Denot-ing by WTop the class of all weak homotopy equivalences, the homotopy category ofspaces Ho(Top) is defined as the localization

(Top,WTop) 7→ Ho(Top) = Top[W−1Top].

Again, the category Ho(Top) lacks the existence of most limits and colimits, and,moreover, the calculus of homotopy limits, homotopy colimits, and homotopy Kanextensions [BK72] is not visible to that category alone. Similarly to the previoussituation, this can be fixed by refining the passage to Ho(Top) by considering the(homotopy) derivator of spaces, a suitable 2-functor

HoTop : A 7→ Ho(TopA) = TopA[(WATop)−1].

There is a plethora of additional examples of such 2-functors arising in variousareas of algebra, geometry, and topology. For example in stable homotopy theoryone is interested in the stable homotopy category of spectra [Vog70, Ada74], whichcan be enhanced by the (homotopy) derivator of spectra. To mention an additionalalgebraic example, stable categories arising in stable module theory or group rep-resentation theory can be similarly refined by suitable derivator [?]. And as anexplicit example of the derivator of an abelian category, associated to a scheme Xthere is the abelian category of quasi-coherent OX -modules. The correspondingderivator DX enhances D(X), the derived category of the scheme [?].

More abstractly, it can be shown that every Quillen model category has anunderlying homotopy derivator [Cis03] and a similar result is true for complete andcocomplete ∞-categories [GPS14]. In this book, we do not assume the reader tohave background knowledge on Quillen model categories [Qui67, Hov99, DS95] or∞-categories [Joy08, Lur09, Gro10]. Consequently, those readers are asked to takefor granted the existence of derivators of abelian categories, spaces, spectra, andother examples, and instead to focus on how to work with such a structure.

At the same time we also want to satisfy those readers who know about Quillenmodel categories. Since a careful, reasonably self-contained proof that arbitraryQuillen model categories have homotopy derivators is rather involved we refrainfrom including such a proof here. As a compromise, in Appendix B we discussthe construction of homotopy derivators associated to combinatorial model cate-gories [Gro13]. This class of examples is more easily established and it includesmany interesting examples. In particular, all the above-mentioned explicit exam-ples of derivators can be realized that way.

Page 8: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

8 MORITZ GROTH

1.4. Stable derivators as an enhancement of triangulated categories. Asalready mentioned, in §6 we define derivators, which can be roughly thought ofas systems of diagram categories coming with restriction functors. One key axiomfor a derivator is that these restriction functors have adjoints on both sides, andthese adjoints are referred to as Kan extension functors. As a special case, everyderivator admits abstract limit and colimit functors. These Kan extension functorsare hence a common generalization of

(i) limits, colimits, and Kan extensions in ordinary complete and cocompletecategories,

(ii) derived limits, derived colimits, and derived Kan extensions between derivedcategories associated to abelian categories, and

(iii) homotopy limits, homotopy colimits, and homotopy Kan extensions betweenhomotopy categories associated to Quillen model categories or ∞-categories.

It turns out that a good deal of basic manipulation rules for limits and Kanextensions extends from ordinary category theory to the context of an arbitraryderivator, and hence applies in the derived and the homotopical situation. Theformalism behind these rules is governed by the notion of a homotopy exact square;see §7.

A derivator is pointed if it admits a zero object, i.e., an object which is simul-taneously initial and final. Typical examples of pointed derivators are derivatorsassociated to abelian categories and the homotopy derivator of pointed topologi-cal spaces. In every pointed derivator, one can define suspensions, loops, cofibers,and fibers, generalizing the classical constructions from homological algebra andhomotopy theory. In §8 we study some aspects of these functors, including iteratedcofiber constructions. For example, we will see that in every pointed derivator thereare suspension-loops and cofiber-fiber adjunctions, and that a threefold iterationof the cofiber construction amounts to passing to the suspension. The calculus ofsuch functors relies on basic facts about the derivator version of pullback squaresand pushout squares, which will also be established in §8.

A pointed derivator is stable if a square is a pullback square if and only if itis a pushout square. It can be shown that this is the case if and only if thesuspension-loops adjunction is an equivalence. Typical examples of stable deriva-tors are derivators associated to abelian categories and the homotopy derivator ofspectra. In §9 we define stable derivators and establish some basic aspects of theircalculus. With the construction of canonical triangulations in mind, we show thatstable derivators are additive. It turns out that suitable shadows of cofiber se-quences induce canonical triangulations on the values of (strong) stable derivators.Moreover, restriction functors and Kan extension functors can be turned into exactfunctors with respect to these triangulations. We also include a short discussionof morphisms of derivators, and show that, more generally, exact morphisms ofderivators induce exact functors with respect to canonical triangulations.

Stable derivators provide an enhancement of triangulated categories and theyfix some of the typical defects of triangulated categories. For example, at the levelof derivators, cone constructions enjoy universal properties and they are functorial.Similarly, there are functorial octahedron diagrams and even functorial higher oc-tahedron diagrams. As a further instance, let us recall that in general it is not true

Page 9: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 9

that diagram categories with values in triangulated categories are again triangu-lated. The corresponding result is true for stable derivators and, as a consequence,the theory of stable derivators can be applied to representation theory.

1.5. Towards abstract representation theory. The aim of the final section §10is to give a short outlook on applications of derivators to abstract representationtheory [GS14c, GS14b, GS14a, GS15]. Here the word ‘abstract’ is meant to indicatethe following. While classically in representation theory of quivers the focus is onrepresentations over fields, it turns out that some basic results extend to the contextof an abstract stable derivator.

To expand on this, we begin by recalling that a quiver Q is simply an orientedgraph and that a representation of Q is simply a diagram Q → Mod(k). Here,k is an arbitrary field and we abuse notation by also denoting the category freelygenerated by Q by the same symbol. For quivers with finitely many vertices thereis an equivalence of categories

Mod(k)Q ' Mod(kQ),

where kQ denotes the (k-linear) path algebra of the quiver Q.Recall that tilting theory [Hap88, AHHK07] is a derived version of Morita the-

ory [Mor58], studying derived equivalences between algebras. We say that twoquivers Q and Q′ are derived equivalent (over k) if there is an exact equivalence ofderived categories of the corresponding path algebras,

D(kQ)∆' D(kQ′).

Obviously, in principle, this might depend on the ground field under consideration.Here we take a different perspective on tilting theory and show that some basic

results are formal consequences of stability. To be more specific, we say that twoquivers Q and Q′ are strongly stably equivalent if for every stable derivator D thereis a natural equivalence

DQ ' DQ′

between the corresponding derivators of representations. Specializing to the stablederivator of a field, it follows that strongly stably equivalent quivers are derivedequivalent over arbitrary fields.

However, a priori, it is much more restrictive to be strongly stably equivalent,since we ask for such equivalences for arbitrary stable derivators. In particular, wealso obtain exact equivalences of derived categories in the context of representa-tions over rings, schemes, differential-graded algebras, ring spectra, or many otherstable homotopy theories arising in algebra, geometry, or topology. To keep thisbook at a reasonable length, here we restrict ourselves to only indicating that allDynkin quivers of type A of a fixed length are strongly stably equivalent. This isclosely related to the existence of canonical higher triangulations in the sense ofMaltsiniotis [Mal05]; see [GS14a]. We conclude §10 with an outlook on additionalresults from abstract representation theory.

Prerequisites. Mostly for motivational purposes, some background knowledgeReformulate!

from homological algebra and/or homotopy theory is useful. While not strictlynecessary, it would be helpful if the reader is familiar with the elementary yoga oftriangulated categories. We also assume some basic acquaintance with the languageof category theory. For the convenience of the reader, in Appendix A we collect

Page 10: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

10 MORITZ GROTH

some basics from category theory which a reader with less background might wantto refer to as needed.

Acknowledgements. This book is based on various introductory talks on deriva-tors as well as on a course on derivators given by the author at the Universityof Bonn during the winter term 2014-2015. I thank the students for their inter-est and their remarks, comments, and questions during the course. I also thankThorge Jensen, Alexander Korschgen, Greg Stevenson, and Jan Stovıcek for helpfulcomments on earlier versions of this book.

While teaching this course on derivators, the author was supported by the Max-Planck-Gesellschaft. A good deal of the final writing of the book was done duringa visit at the CRM in Bellaterra. The author is very grateful for the hospitality atthe CRM as well as a research visiting grant from the CRM and the University ofBielefeld making this visit possible.

2. Abelian categories and classical derived functors

We assume that the reader is familiar with basic homological algebra, includingthe construction of classical derived functors and interesting examples. Never-theless, mostly to set the stage and to smoothen the transition to the theory ofderived categories, here we quickly recall some basics. For more details and ex-amples we refer the reader to the literature; see for example [Rot79], the moreadvanced [Wei94, GM03, KS06], and the original [CE99, Gro57].

In §2.1 we recall basic definitions, emphasizing the distinction between propertiesand structures. In §2.2 we sketch the construction of classical derived functors andshow that they yield universal δ-functors. In §2.3 we illustrate the notion of derivedlimit and derived colimit functors by observing that group cohomology and grouphomology, respectively, are special cases of these more general notions.

2.1. Review of abelian categories. To put it as a slogan, homological algebrais classically the study of abelian categories and derived functors. We begin byrecalling basic definitions concerning abelian categories.

Definition 2.1. A preadditive category is a pointed category with finite biprod-ucts.

Thus, denoting the singleton by ∗, a preadditive categoryA satisfies the followingthree defining axioms.

(i) The category A has a zero object, i.e., an object 0 ∈ A which is both finaland initial,

homA(X, 0) ∼= ∗ ∼= homA(0, Y ), X, Y ∈ A.

Such a category is also called a pointed category.(ii) The category A has finite coproducts and finite products.(iii) For any X,Y ∈ A the canonical map X t Y → X × Y from the coproduct to

the product is an isomorphism.

Just to be completely specific, in matrix notation the canonical map in axiom (iii)is given by (

1 0

0 1

): X t Y → X × Y.

Page 11: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 11

Following standard notation, we write X ⊕ Y for any of X t Y ∼= X × Y and referto it as the biproduct of X and Y .

Remark 2.2. (i) All axioms of a preadditive category ask for properties. Theonly structure is the category itself.

(ii) Any preadditive category A can be canonically endowed with an enrich-ment in the category AbMon of abelian monoids. In fact, given morphismsf, g : X → Y , then the sum f + g : X → Y is the composition

f + g : X∆→ X ⊕X f⊕g→ Y ⊕ Y ∇→ Y,

where ∆ is the diagonal X → X×X and∇ is the fold map ∇ : Y tY → Y . Weleave it to the reader to check that this defines an abelian monoid structure onhomA(X,Y ) with neutral element 0: X → 0→ Y and that the compositionis bilinear.

(iii) A closely related perspective on this enrichment is as follows (we will comeback to in §9). In a preadditive category, the fold map ∇ : Y ⊕Y → Y endows

Promise.any Y ∈ A with the structure of an abelian monoid object. We leave it asan exercise to verify that such an abelian monoid structure is equivalent tospecifying a lift of the represented functor homA(−, Y ) : Aop → Set againstthe forgetful functor AbMon→ Set,

AbMon

Aop

homA(−,Y )//

::

Set.

Similarly, the diagonal map ∆: X → X ⊕ X endows any X ∈ A withthe structure of an abelian comonoid object. And one checks that such anabelian comonoid structure is equivalent to a lift of the corepresented functorhomA(X,−) : A → Set against the forgetful functor AbMon→ Set.

Definition 2.3. A preadditive category A is additive if for every X ∈ A the shearmap (

1 1

0 1

): X ⊕X → X ⊕X

is an isomorphism.

Lemma 2.4. The following are equivalent for a preadditive category A.

(i) The category A is additive.(ii) Identity morphisms id : X → X have additive inverses in homA(X,X).

(iii) The abelian monoids homA(X,Y ) are abelian groups.(iv) The fold map ∇ : Y ⊕ Y → Y endows Y ∈ A with the structure of an abelian

group object.(v) The diagonal map ∆: X → X ⊕X endows X ∈ A with the structure of an

abelian cogroup object.

Proof. This proof is left as an exercise.

Thus, to emphasize, from the defining exactness properties of an additive cate-gory one can construct the structure of an enrichment in the category Ab of abeliangroups. The good notion of functors between additive categories are additive func-tors, i.e., functors which preserve zero objects and finite direct sums. Given an

Page 12: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

12 MORITZ GROTH

additive functor F : A → B it follows that the functor is automatically compatiblewith this enrichment, i.e., that the map homA(X,Y )→ homB(FX,FY ) is a grouphomomorphism for every X,Y ∈ A.

Finally, abelian categories are additive categories admitting kernels and cokernelsand such that the first Noether isomorphism theorem is true. In more detail, let usconsider a morphism f : X → Y in an additive category with kernels and cokernels.

(i) The image im(f) of f is the kernel of the canonical map Y → cok(f) to thecokernel, yielding the diagram

im(f)→ Y → cok(f).

(ii) The coimage coim(f) of f is the cokernel of the canonical map ker(f)→ Xfrom the kernel, hence there is the diagram

ker(f)→ X → coim(f).

One easily checks that f factors through a canonical map coim(f)→ im(f),

f : X → coim(f)→ im(f)→ Y.

In the case of the category of abelian groups A = Ab the map coim(f) → im(f)is the usual isomorphism X/ ker(f) → im(f) as guaranteed by the first Noetherisomorphism theorem.

Definition 2.5. An abelian category is an additive category satisfying the fol-lowing two properties.

(i) Every morphism has a kernel and a cokernel.(ii) For every morphism f : X → Y the canonical map coim(f) → im(f) is an

isomorphism.

Thus, as a slogan, abelian categories axiomatize additive categories allowing forthe ‘usual calculus of short exact sequences’.

Examples 2.6. (i) The category Ab and the category Mod(R) of left R-modulesover a ring R is abelian.

(ii) Given a topological space X, the categories of presheaves or sheaves on Xwith values in an abelian category is again abelian.

(iii) Given an abelian category A, the corresponding category Ch(A) of chaincomplexes is again abelian.

(iv) A category is abelian if and only if its opposite is abelian.(v) If Ai, i ∈ I, are abelian categories, then so is the product category

∏i∈I Ai.

In order to state a generalization of Examples 2.6(v) we recall the followingdefinition.

Definition 2.7. Let C be a category and let A be a small category. The func-tor category or diagram category CA = Fun(A, C) has as objects the functorsX : A→ C and as morphisms the natural transformations α : X → Y .

Lemma 2.8. Let A be an abelian category and let A be a small category. Thefunctor category AA = Fun(A,A) is again abelian.

Proof. This proof is left as an exercise, the only hint being that limits and colimitsare constructed pointwisely.

Page 13: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 13

2.2. Classical derived functors. We now focus on additive functors betweenabelian categories. Let us recall that a functor F : A → B between additive cate-gories is additive if it preserves zero objects and finite biproducts. As a consequencethere is the following result.

Lemma 2.9. Any additive functor between abelian categories preserves split shortexact sequences.

Proof. This proof is left as an exercise.

While, in general, short exact sequences are not preserved by additive functors,many functors showing up in nature preserve exactness ‘on one side’, motivatingthe following definitions.

Definition 2.10. Let F : A → B be an additive functor between abelian categories.

(i) The functor F is left exact if for every exact sequence 0→ X ′ → X → X ′′

in A also the sequence 0→ FX ′ → FX → FX ′′ is exact.(ii) The functor F is right exact if for every exact sequence X ′ → X → X ′′ → 0

in A also the sequence FX ′ → FX → FX ′′ → 0 is exact.(iii) The functor F is exact if it is left exact and right exact.

Examples 2.11. Let R,S be rings and let M be an R-S-bimodule.

(i) The tensor product M ⊗S − : Mod(S) → Mod(R) is right exact, but, ingeneral, not left exact.

(ii) The hom functor homR(M,−) : Mod(R) → Mod(S) is left exact, but, ingeneral, not right exact.

Remark 2.12. (i) By means of Proposition A.15 one observes that any abeliancategory is finitely complete and that it is complete if and only if it admitsproducts. There is a dual statement concerning colimits and coproducts.

(ii) Proposition A.22 implies that an additive functor between abelian categoriesis left exact in the sense of homological algebra (Definition 2.10) if and onlyif it is left exact in the sense of category theory (Definition A.21). Togetherwith Lemma A.23 this yields a formal proof of the positive statements in Ex-amples 2.11.

We next recall the classical construction of left derived functors. Thus, letA,B beabelian categories, let F : A → B be a right exact functor, and let us for simplicityassume that A has enough projective objects. Hence, for every object X ∈ A wecan find a projective resolution, i.e., an exact sequence

. . .→ P2 → P1 → P0ε→ X

such that all Pi are projective. Since additive functors preserve zero objects, weobtain an induced chain complex FP ∈ Ch(B) which is given by

. . .→ FP2 → FP1 → FP0.

A final application of the homology functors Hn : Ch(B)→ B, n ≥ 0, concludes thedefinition of the (classical) left derived functors LnF : A → B, i.e., we set

(LnF )(X) = Hn(FP ), n ≥ 0.

Since we assume the reader to be familiar with basic homological algebra, wecontent ourselves by claiming that the above construction is well-defined, that it can

Page 14: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

14 MORITZ GROTH

be extended to additive functors LnF : A → B, and that, using the right exactnessof F , there is a natural isomorphism L0F ∼= F .

As a first justification of the definition of classical left derived functor, let usrecall further that they measure systematically the deviation of F from being leftexact. More precisely, as a consequence of the horseshoe lemma, every short exactsequence 0→ X ′ → X → X ′′ → 0 in A gives rise to a long exact sequence

. . .→ (L1F )(X)→ (L1F )(X ′′)δ→ FX ′ → FX → FX ′′ → 0

in B. In particular, the image im(δ) : (L1F )(X ′′) → FX ′ of the connecting homo-morphism agrees with the kernel of FX ′ → FX.

An even better justification is given by the observation that classical left derivedfunctors yield the universal way to measure the deviation from being left exact(see Theorem 2.15). To make this precise we first recall the following definitions.

Continue here.

Definition 2.13. Let A and B be abelian categories. A homological δ-functorT from A to B consists of

(i) additive functors Tn : A → B, n ≥ 0, and(ii) for every short exact sequence 0 → X ′ → X → X ′′ → 0 in A of morphisms

δn : Tn(X ′′)→ Tn−1(X ′), n ≥ 1,

such that for every such short exact sequence we obtain a natural long exact se-quence

. . .→ T1(X)→ T1(X ′′)δ1→ T0(X ′)→ T0(X)→ T0(X ′′)→ 0.

The naturality means that every morphism of short exact sequences gives riseto a commutative ladder in B. Thus, the connecting homomorphisms δn of (ii)assemble to suitable natural transformations.

Definition 2.14. (i) A morphism of homological δ-functors α : S → Tconsists of natural transformations αn : Sn → Tn, n ≥ 0, such that for everyshort exact sequence 0→ X ′ → X → X ′′ → 0 in A the diagram

Sn(X ′′)δ //

αn

Sn−1(X ′)

αn−1

Tn(X ′′)δ// Tn−1(X ′)

commutes for all n ≥ 1.(ii) A homological δ-functor T is universal if for every homological δ-functor S

and every natural transformation α0 : S0 → T0 there is a unique morphismof homological δ-functors α : S → T extending the given α0.

Here is a justification of the seemingly adhoc construction of classical left derivedfunctors.

Theorem 2.15. Let F : A → B be a right exact functor between abelian categoriesand let A have enough projective objects. The functors LnF, n ≥ 0, together withthe connecting homomorphisms define a universal homological δ-functor.

Independence? Proof. We assume that the reader knows that the LnF together with the connectinghomomorphisms assemble to a homological δ-functor. To prove its universality,

Page 15: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 15

let T be a homological δ-functor from A to B and let α0 : T0 → F be a naturaltransformation.

For n ≥ 1 we assume by induction that a partial morphism of δ-functors consist-ing of αj , 0 ≤ j < n, has already been constructed and that it is unique. We wantto show that there is a unique way of extending it up to degree n. Given X ∈ Awe choose a short exact sequence

(2.16) 0→ K → P → X → 0

such that P is a projective object. Since the left derived functors LnF, n ≥ 1, vanishon projective objects, by induction assumption we have a commutative (solid arrow)diagram with exact rows

TnXδ //

∃!

Tn−1K //

αn−1

Tn−1P

αn−1

0 // (LnF )Xδ// (Ln−1F )K // (Ln−1F )P.

An easy diagram chase shows that there is a unique dashed morphism αn : TnX →(LnF )X such that the square on the left commutes. Thus, the compatibility withshort exact sequences of the form (2.16) implies that there is at most one way ofdefining the desired αn : Tn → LnF . In particular, the morphism is independent ofthe choice of (2.16).

As for the existence it remains to show that these unique morphisms assembleinto natural transformations and that they are compatible with arbitrary shortexact sequences (and not only the ones of the form (2.16)). To verify the naturalitylet us consider a morphism f : X → Y in A. Such a morphism can be extended toa morphism of short exact sequences

0 // K //

P //

X

// 0

0 // L // Q // Y // 0

in which P,Q are projective objects. Associated to this we obtain the followingdiagram in which all squares commute with possibly the exception of the naturalitysquare on the left,

TnXδ //

''

Tn−1K

((

TnY //

Tn−1L

(LnF )X

''

// (Ln−1F )K

((

(LnF )Yδ

// (Ln−1F )L.

A diagram chase shows that in that square the two possibly different compositionsTnX → (LnF )Y agree when composed with δ : (LnF )Y → (Ln−1F )L. Since againthis morphism δ is a monomorphism, also the square on the left commutes whichis to say that we constructed a natural transformation αn : Tn → LnF .

Page 16: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

16 MORITZ GROTH

It remains to show that αn is compatible with all short exact sequences. Forthis purpose, we note that for any short exact sequence 0 → X1 → X2 → X3 → 0and any short exact sequence 0 → K → P → X3 → 0 such that P is a projectiveobject we can extend the identity morphism id: X3 → X3 to a morphism of shortexact sequences, yielding a commutative diagram

0 // K //

P //

X3

=

// 0

0 // X1// X2

// X3// 0.

Similarly to the previous step, associated to this morphism we obtain a diagram

TnX3δ //

=

((

Tn−1K

))

TnX3//

Tn−1X1

(LnF )X3

= ''

// (Ln−1F )K

((

(LnF )X3δ

// (Ln−1F )X1

in which all squares with possibly the exception of the front square commute (byinduction assumption and by the previous steps). Precomposing the two morphismsTnX3 → (Ln−1F )X1 with id: TnX3 → TnX3, a short diagram chase implies thatalso the front face commutes, concluding the proof.

Dualizing these constructions we are led to right derived functors of left exactfunctors F : A → B. In this case, assuming the existence of enough injective objectsin A, every object X ∈ A has an injective resolution 0 → X → I0 → I1 → . . ..If we denote the truncated cochain complex by I, then the n-th (classical) rightderived functor RnF is defined by

(RnF )(X) = Hn(FI), n ≥ 0.

These classical right derived functors assemble to universal cohomological δ-functorsin the obvious sense.

For variants of these results and a discussion of examples and applications werefer the reader to the literature. Here we only mention that in the case of Exam-ples 2.11 we obtain the torsion products TorRn (M,N) and Ext-groups ExtnR(M,N)as typical examples of classical left derived and right derived functors, respectively.

2.3. Group cohomology as a derived limit. In this subsection we observe aclose relation between group (co)homology and certain derived limit and derivedcolimit functors. This is meant to be a first illustration that derived (co)limitfunctors encode interesting constructions. The reader who is not familiar withgroup (co)homology can safely skip most of this subsection.

Recall from Lemma 2.8 that for every abelian category A and every small cat-egory A the functor category AA is again abelian. Assuming A to have productsin case A is not finite, the limit functor limA : AA → A exists and is part of anadjunction

(∆A, limA) : A AA,

Page 17: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 17

showing that limA is a left exact functor (see Lemma A.23 and Remark 2.12). If wemoreover assume that AA has enough injective objects (which follows for examplefrom A having enough injective objects), then we can apply the techniques of §2.2in order to obtain (classical right) derived limit functors

RnlimA : AA → A, n ≥ 0.

Making dual assumptions on A, we obtain an adjunction (colimA,∆A) : AA Aand associated (classical left) derived colimit functors

LncolimA : AA → A, n ≥ 0.

Here we are only interested in the category A = Ab of abelian groups. Thecategory of abelian groups is complete and cocomplete. Moreover, it can be shownthat all functor categories AbA have enough injectives and projectives so that thederived (co)limit functors exist.

In our special situation we do not need this general statement since we can arguemore directly. Given a discrete group G, there is an associated category, againdenoted by G, which has one object ∗ only and such that G is the correspondingendomorphism monoid,

homG(∗, ∗) = G.

We recall that the integral group ring ZG on G is defined as follows. Theunderlying abelian group is the free abelian group on G. Denoting by eg thegenerator associated to g ∈ G, the ring multiplication is the bilinear extensionof the assignment

eg · eg′ = egg′ , g, g′ ∈ G.The reader easily checks that there is an equivalence of categories

(2.17) AbG ' Mod(ZG),

so that AbG clearly has enough injectives and projectives. We will not distinguishnotationally objects corresponding to each other under this equivalence.

The following two constructions apply to modules M ∈ Mod(ZG) and are centralto the theory of (co)homology of groups. Here we simplify notation and writegm = egm, g ∈ G,m ∈M .

(i) The abelian group of invariants is MG = m ∈ M | gm = m, g ∈ G andthis defines a functor (−)G : Mod(ZG)→ Ab.

(ii) The abelian group of coinvariants is MG = M/gm−m | g ∈ G,m ∈ M,defining a functor (−)G : Mod(ZG)→ Ab.

One can check directly that the invariants functor is left exact and that thecoinvariant functor is right exact. A different way of seeing this is as follows. Weobserve that the integral group ring ZG comes with an augmentation map

ε : ZG→ Z :∑

riegi 7→∑

ri,

which is easily seen to be a ring homomorphism. Restriction of scalars on the leftand on the right applied to the regular bimodule ZZZ yields

ZGZZ and ZZZG,

respectively. We refer to these G-actions as trivial G-actions.

Lemma 2.18. Let G be a discrete group.

Page 18: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

18 MORITZ GROTH

(i) There is a natural isomorphism (−)G ∼= homZG(Z,−) : Mod(ZG) → Ab,where Z is endowed with the trivial left G-action.

(ii) There is a natural isomorphism (−)G ∼= Z⊗ZG − : Mod(ZG)→ Ab, where Zis endowed with the trivial right G-action.

Proof. This proof is left as an exercise.

Together with Examples 2.11 this also yields the claimed exactness properties.As a special case of Ext- and Tor-functors we can make the following definitions(but refer the reader to almost any book on homological algebra or, for example,[Bro94] for more details).

Definition 2.19. Let G be a discrete group, let M ∈ Mod(ZG), and let n ≥ 0.

(i) The n-th group cohomology of G with coefficients in M is the abeliangroup

Hn(G;M) =(Rn(−)G

)(M) ∼= ExtnZG(Z,M).

(ii) The n-th group homology of G with coefficients in M is the abelian group

Hn(G;M) =(Ln(−)G

)(M) ∼= TorZGn (Z,M).

The equivalences (2.17) offer a different perspective on group (co)homology. Tomake it precise, we give the following definition, which is a rather obvious weakeningof the assumption that two functors are naturally isomorphic.

Definition 2.20. Let F1 : C1 → D1 and F2 : C2 → D2 be functors. The functorsF1, F2 are equivalent if there are equivalences φ : C1 ' C2, ψ : D1 ' D2 and anatural isomorphism ψ F1

∼= F2 φ,

C1F1 //

φ '

∼=

D1

ψ'

C2F2

// D2.

The point is that equivalent functors are often equally good for many practicalpurposes. For example, many properties of functors (like being fully faithful, essen-tially surjective, preserving certain (co)limits) are common to equivalent functors.

Proposition 2.21. Let G be a discrete group.

(i) The invariants functor (−)G : Mod(ZG) → Ab and limG : AbG → Ab areequivalent.

(ii) The coinvariants functor (−)G : Mod(ZG) → Ab and colimG : AbG → Abare equivalent.

Proof. More precisely, we show that for every discrete group G the following dia-gram commutes up to natural isomorphisms

Mod(ZG)

(−)G

zz

(−)G

%%'

∼= ∼=Ab Ab.

AbGcolimG

dd

limG

99

Page 19: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 19

In this diagram the vertical functor is the equivalence (2.17). The claim concerningthe triangle on the right follows from the general construction of limits in termsof equalizers and products; see (A.16). In fact, in this particular case, the generalconstruction specializes to the equalizer of abelian groups

limGM ∼= eq(M ⇒

∏g∈G

M),

where one map is the diagonal map while the other map is m 7→ (gm)g∈G. Clearly,this equalizer is naturally isomorphic to the invariants MG ∈ Ab.

Similarly, using the explicit construction of colimits in (A.17), one observes thatalso the triangle on the left commutes up to a natural isomorphism.

Corollary 2.22. Let G be a discrete group and let n ≥ 0.

(i) The functor Hn(G;−) : Mod(ZG)→ Ab and RnlimG : AbG → Ab are equiv-alent.

(ii) The functor Hn(G;−) : Mod(ZG) → Ab and LncolimG : AbG → Ab areequivalent.

Proof. This is purely formal consequence of Proposition 2.21 and the details areleft to the reader.

Thus, there is a very close relation between group cohomology functors Hn(G;−)and derived limit functors RnlimG and, similarly, between group homology functorsHn(G;−) and derived colimit functors LncolimG. Of course, we do not claim thatthis different perspective simplifies the study of group (co)homology. Instead, thepoint of these observations was to illustrate that the seemingly very abstract derived(co)limit functors specialize to well-known constructions which are of independentinterest.

In the remainder of this book we indicate that suitable combinations of derivedlimit functors (and, more generally, derived Kan extension functors) encode in-teresting constructions, and that such constructions are conveniently organized bymeans of derivators.

3. Derived categories of abelian categories

In this section we recall the construction of the derived category of an abeliancategory [?]. These derived categories can be thought of as rather refined invariantsof the given abelian categories [?]. Important special cases are derived categoriesof rings [Hap88] and derived categories of schemes [?, ?], which respectively showup prominently in representation theory and algebraic geometry. While in thissection we consider derived categories as plain categories only (without additionalstructure), in §4 we briefly discuss the classical triangulations on derived categories.

In §3.1 we observe that cokernel functors in general fail to be exact, but that theyare exact on monomorphisms. Following the typical reasoning from homologicalalgebra, this suggest to ‘redefine’ the cokernel, thereby obtaining cone and cofiberfunctors (see §3.2). In §3.3 we recall the definition of derived categories and thecorresponding notion of derived functors. In §3.4 we show that derived categories,although defined as localizations, enjoy the stronger universal property of being2-localizations. This allows us to conclude in §3.5 that for every abelian categorythe cone functor is the left derived cokernel functor. In general, these cone functorsdo not factor through the morphism category of the derived category. To put

Page 20: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

20 MORITZ GROTH

this into context, in §3.6 we emphasize that one has to distinguish between derivedcategories of diagram categories and diagram categories of derived categories. Usingagain the property of a 2-localization, we observe that these categories are relatedby underlying diagram functors. In §3.7 we illustrate these underlying diagramfunctors in the context of category algebras (like group algebras, path algebras, andincidence algebras) and observe that for semisimple coefficients they are equivalentto homology functors. This shows that underlying diagram functors in generaldiscard relevant information.

3.1. Towards derived cokernels. Let A be an abelian category and let A be asmall category. In §2.3 we mentioned first examples of such categories such thatthe limit and colimit functors

limA, colimA : AA → A

exist but fail to be exact. The existence of the adjunction (colimA,∆A) implies thatcolimA is right exact, but, in general colimA is not left exact, and dually for limA.In this subsection we discuss exactness properties of (co)products and (co)kernels.

We already observed that abelian categories are finitely cocomplete and thatarbitrary finite colimits can be constructed from finite coproducts and cokernels,and dually. Hence we consider these two types of constructions independently. Letus recall that a category is discrete if all morphisms are identity morphisms. Givena discrete category D and an arbitrary category C, there is a canonical isomorphismCD ∼=

∏d∈D C.

Lemma 3.1. Let A be an abelian category and let D be a finite discrete category.The finite biproduct functor ⊕

d∈D

: AD → A

is exact.

Proof. Given an abelian category A, finite coproduct and finite product functorsare part of adjunctions

(∐d∈D

,∆D) : AD A and (∆D,∏d∈D

) : A AD,

showing that∐d∈D is right exact and

∏d∈D is left exact. The preadditivity of

abelian categories yields natural isomorphisms∐d∈D∼=⊕

d∈D∼=∏d∈D, implying

that finite biproducts are exact.

Hence, finite coproducts and finite products in abelian categories are exact. Ofcourse this lemma already applies to preadditive categories but here we focus onabelian categories.

Remark 3.2. Infinite (co)products in abelian categories are, in general, not exact.However, one can impose additional axioms on abelian categories guaranteeingfirst of all the existence of infinite coproducts or infinite products and, second, thatinfinite coproducts or infinite products are exact. In this order, attributing creditto [Gro57], these assumptions are referred to as axioms (AB3), (AB3*), (AB4), and(AB4*). We refer the reader to the literature for more on these axioms, includinga discussion of examples of abelian categories satisfying these stronger axioms; see[Gro57] or [Fai73, §14].

Page 21: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 21

We now turn to the case of (co)kernels which, in general, happen to fail to beexact. To begin with, as a special case of Lemma 2.8, associated to an abeliancategory A there is the abelian category A[1] of arrows in A. Here, [1] is thepartially ordered set [1] = (0 < 1), considered as a category. Thus, objects in A[1]

are morphisms X0 → X1 in A while a morphism X → Y is a commutative square

X0//

Y0

X1// Y1.

Since limits and colimits in A[1] are constructed levelwise (see again the proofof Lemma 2.8 which was left as an exercise), a short exact sequence in A[1] corre-sponds precisely to a morphism of short exact sequences in A, i.e., a commutativediagram

(3.3)

0 // X0

f

// Y0//

g

Z0//

h

0

0 // X1// Y1

// Z1// 0,

such that the rows are short exact sequences.

Lemma 3.4. Let A be an abelian category.

(i) The kernel functor ker : A[1] → A is left exact.(ii) The cokernel functor cok: A[1] → A is right exact.

(iii) In general, the functors ker, cok: A[1] → A are not exact.

Proof. Since both functors are additive, the first two statements are simply a refor-mulation of the snake lemma. In fact, given a short exact sequence of morphismsas in (3.3), the snake lemma yields an exact sequence

0→ ker(f)→ ker(g)→ ker(h)→ cok(f)→ cok(g)→ cok(h)→ 0,

establishing the first two statements.To establish the third statement it suffices to consider any non-zero object X ∈ A

and the corresponding short exact sequence of morphisms

0 // 0

// Xid //

id

X //

0

0 // Xid// X // 0 // 0.

In fact, by the snake lemma we obtain an induced exact sequence

0→ 0→ 0→ X∼=→ X → 0→ 0→ 0,

showing that the kernel is not right exact and that the cokernel is not left exact.

We now reformulate the failure of exactness using a slightly different perspective.

Definition 3.5. Let A be an abelian category and let f : X → Y be a morphismin Ch(A). The chain map f is a quasi-isomorphism, in notation f : X

∼→ Y, if itinduces isomorphisms in homology,

Hk(f) : Hk(X)∼=→ Hk(Y ), k ∈ Z.

Page 22: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

22 MORITZ GROTH

The class of all quasi-isomorphisms is denoted by W = WA.

We recall that any additive functor F : A → B induces an additive functorF : Ch(A) → Ch(B) between the corresponding categories of chain complexes, ob-tained by applying F in all degrees.

Lemma 3.6. An additive functor F : A → B between abelian categories is exact ifand only if the induced functor F : Ch(A)→ Ch(B) preserves quasi-isomorphisms.

Proof. This proof is left as an exercise.

This motivates the following definition.

Definition 3.7. Let A,B be abelian categories. A functor F : Ch(A) → Ch(B) isexact or homotopy-invariant if it preserves quasi-isomorphisms.

Lemma 3.6 is to be seen in contrast with the following result.

Lemma 3.8. Let F : A → B be an additive functor between abelian categories. Theinduced functor F : Ch(A)→ Ch(B) preserves chain homotopies and, in particular,chain homotopy equivalences.

Proof. This proof is left as an exercise.

For later applications of Lemma 3.6 to limit and colimit functors, it is conve-nient to have a refined version of Lemma 2.8. Therein we consider the followingisomorphism of categories. Given a diagram of chain complexes X : A → Ch(A),one easily checks that

Yk(a) = X(a)k, k ∈ Z, a ∈ A,

defines a chain complex Y ∈ Ch(AA), and conversely. Moreover, this obviouslyyields an isomorphism of categories

(3.9) Ch(A)A ∼= Ch(AA) : X ↔ Y.

Given two functors X ′, X ′′ : A → Ch(A) and a morphism X ′ → X ′′ in Ch(A)A,i.e., a natural transformation α : X ′ → X ′′, we say that α is a levelwise quasi-isomorphism if the chain maps αa : X ′(a)→ X ′′(a), a ∈ A, are quasi-isomorphisms.The class of all levelwise quasi-isomorphims is denoted by WA

A .

Lemma 3.10. Let A be an abelian category and let A be a small category.

(i) The functor category AA = Fun(A,A) is abelian.(ii) The isomorphism of categories (3.9) identifies the class WA

A of levelwise quasi-isomorphisms with the class WAA of quasi-isomorphisms.

Proof. This proof is left as an exercise.

Let us recall that kernels and cokernels in Ch(A) are calculated degreewise. Thus,Lemma 3.4, Lemma 3.6, and Lemma 3.10 together imply the following warning.

Warning 3.11. Let A be an abelian category. In general, the functors

cok, ker : Ch(A)[1] → Ch(A)

are not homotopy-invariant, i.e., they do not send levelwise quasi-isomorphisms toquasi-isomorphisms.

Page 23: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 23

Despites this warning, there are the following positive statements concerninglevelwise quasi-isomorphisms X → Y in Ch(A)[1], i.e., commutative squares ofchain complexes as in

(3.12)

X0f//

X1

Y0 g// Y1.

Lemma 3.13. Let A be an abelian category and let (3.12) be a levelwise quasi-isomorphism in Ch(A)[1].

(i) If f and g are monomorphisms, then the induced map cok(f) → cok(g) is aquasi-isomorphism.

(ii) If f and g are epimorphisms, then the induced map ker(f) → ker(g) is aquasi-isomorphism.

Proof. We give a proof of (i) and observe that there is a morphism of short exactsequences of chain complexes

0 // X0

f// X1

//

cok(f) //

0

0 // Y0 g// Y1

// cok(g) // 0.

The induced long exact sequences in homology and the 5-lemma allow us to concludethe proof of (i). The case of (ii) is dual.

Thus, the cokernel and the kernel functors are exact on certain morphisms ofchain complexes. The idea behind the construction of derived (co)kernel functorsis now simply to quasi-isomorphically approximate arbitrary morphisms of chaincomplexes by these good ones.

3.2. Cones and cofibers. In this subsection we recall some classical constructionsfrom homological algebra. This includes the seemingly adhoc construction of thecone functor, a certain ‘corrected version’ of the cokernel functor. While a firstjustification of this construction is provided by Proposition 3.23, a more conceptualone will be given in §3.5. Everything can be dualized to yield a similar ‘correction’of the kernel functor.

To motivate these constructions we include the following different description ofthe cokernel. Let f : X0 → X1 be a morphism in an abelian category A. We notethat the cokernel of f can also be constructed by considering pushout squares

(3.14)

X0f//

X1

0 // cok(f).

In fact, the universal property of this pushout square is easily checked to reduce tothe usual universal property of the cokernel. Thus, although the cokernel cok(f) isnot the colimit of the diagram f : X0 → X1 (which would be isomorphic to X1) it

Page 24: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

24 MORITZ GROTH

is the colimit of the span (0← X0 → X1). We will come back to this intermediatestep later.

Now, as recalled in Warning 3.11, in general the cokernel functor is not exact.Using this different description of the cokernel, a classical way of fixing this problemis by replacing the morphism X0 → 0 in (3.14) quasi-isomorphically by a niceinclusion, namely by the inclusion i : X0 → CX0 of X0 in its cone CX0. Wequickly recall this construction and, since we also have a use for further variants ofsuch constructions, we allow us to include a minor digression.

Digression 3.15. Let FZ denote the category of finitely generated, free abeliangroups and let Chb(FZ) be the category of bounded chain complexes in FZ. Werecall that for every additive category A there is a two-variable functor

(3.16) ⊗ : Chb(FZ)× Ch(A)→ Ch(A)

which is additive in both variables separately.To begin with there is a similar biadditive action ⊗ : FZ×A → A which is already

determined by asking that there is a natural isomorphism Z⊗X ∼= X,X ∈ A. Thus,for a free abelian group F of rank r, there is a natural isomorphism F ⊗X ∼= r ·X,where r ·X denotes the r-fold direct sum X ⊕ . . .⊕X.

This action can be extended to chain complexes by the following ‘convolutiontype construction’. Given chain complexes F ∈ Chb(FZ) and X ∈ Ch(A) we definethe underlying graded object of F ⊗X ∈ Ch(A) by

(F ⊗X)k =⊕p+q=k

Fp ⊗Xq, k ∈ Z.

(This direct sum is finite by the boundedness assumption on F .) The differentials(F ⊗ X)k → (F ⊗ X)k−1 are obtained from those on F and X using the usualKoszul sign convention. (Recall that this convention says that, in differential gradedalgebra, ‘whenever the order of two homogeneous elements of degree p and q isswapped, a sign (−1)pq has to appear’.)

We illustrate the action (3.16) by the following examples.

Examples 3.17. (i) Let I = Ccell([0, 1]) ∈ Chb(FZ) be the reduced cellular chaincomplex of the interval endowed with the CW structure consisting of two0-cells and one 1-cell only. More explicitly, I0 = I1 = Z, the differentialI1 → I0 is the identity, and all Ik, k 6= 0, 1, vanish. We call I the intervaland it is easily verified that I is a contractible chain complex. For X ∈ Ch(A)there is a natural isomorphism I ⊗ X ∼= CX where C : Ch(A) → Ch(A) isthe cone functor. (If you did not see a definition of C before, then you cansimply take this construction as a definition and we suggest as an exercise tounravel it.) The notation is of course motivated from similiar constructionsfor pointed topological spaces, where smashing with the interval yields the(reduced) cone construction.

(ii) Let S0 = Ccell(0, 1) ∈ Chb(FZ) be the reduced cellular chain complex of the0-sphere endowed with the CW structure conisting of two 0-cells only. Thus,S0 is the stalk complex Z. For X ∈ Ch(A) there is a natural isomorphismS0 ⊗X ∼= X.

(iii) The continuous map 0, 1 → [0, 1] yields a map S0 → I in Chb(FZ) andhence a natural transformation S0⊗− → I⊗− of functors Ch(A)→ Ch(A).

Page 25: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 25

Under the above natural isomorphisms, for X ∈ Ch(A) this yields the usualnatural inclusion i : X → CX.

(iv) Let S1 ∈ Chb(FZ) be the reduced cellular chain complex of the 1-sphereconsidered as a quotient CW complex [0, 1]/0, 1. Thus, S1 is the abeliangroup Z considered as a complex concentrated in degree 1. For X ∈ Ch(A)there is a natural isomorphism S1 ⊗X ∼= ΣX, where Σ: Ch(A)→ Ch(A) isthe usual ‘shift against the differential’.

Having recalled the cone construction for chain complexes, we now replace thevertical morphism on the left in (3.14) by X0 → CX0 and hence consider thepushout diagram

(3.18)

X0f//

i

X1

cof(f)

CX0// Cf.

Definition 3.19. Let A be an abelian category and let f : X0 → X1 be in Ch(A).

(i) The chain complex Cf in (3.18) is the cone of f .(ii) The chain map cof(f) : X1 → Cf in (3.18) is the cofiber of f .

It is immediate that both the cone and the cofiber are functorial in f , i.e., wehave functors

C : Ch(A)[1] → Ch(A) and cof : Ch(A)[1] → Ch(A)[1].

Lemma 3.20. Let A be an abelian category and let X ∈ Ch(A).

(i) The cone CX is contractible.(ii) The cone of an isomorphism is contractible.

Proof. Since the interval I is contractible and the construction (3.16) is additive inthe first variable it follows from Lemma 3.8 that I ⊗ X ∼= CX is contractible. Iff : X0 → X1 is an isomorphism, then the construction (3.18) yields an isomorphismCX0

∼= Cf . Thus, by the first part, Cf is contractible.

To prove the homotopy invariance of the cone construction we collect the follow-ing lemma.

Lemma 3.21. Let A be an abelian category and let us consider a square

X0i //

j

X1

p

X2 q// X

in A. The square is a pushout square if and only if the sequence

X0(i,j)t→ X1 ⊕X2

(p,−q)→ X → 0

is exact.

Proof. This proof is left as an exercise.

Page 26: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

26 MORITZ GROTH

In the special case of the defining pushout square (3.18) of the cone, we henceobtain a short exact sequence

(3.22) 0→ X0 → X1 ⊕ CX0 → Cf → 0,

since the first map is clearly also a monomorphism (see Examples 3.17(iii)). Thereader easily checks that this short exact sequence is functorial in the morphismf : X0 → X1. While this construction allows us to give a partial justification of thecone construction (see again Lemma 3.6), a better one will be given in §3.5 (seeTheorem 3.42).

Proposition 3.23. Let A be an abelian category. The cone C : Ch(A)[1] → Ch(A)sends levelwise quasi-isomorphisms to quasi-isomorphisms.

Proof. We again consider a diagram of chain complexes (3.12) such that the verticalmaps are quasi-isomorphisms, i.e., a levelwise quasi-isomorphism f → g. Associatedto such a levelwise quasi-isomorphism we obtain (by the functoriality of (3.22)) amorphism of short exact sequences

0 // X0//

X1 ⊕ CX0//

Cf //

0

0 // Y0// Y1 ⊕ CY0

// Cg // 0.

Since the objects CX0, CY0 have trivial homology objects (Lemma 3.20), anychain map between them is a quasi-isomorphism. As finite direct sums of quasi-isomorphisms are again quasi-isomorphisms, in the above diagram also the secondvertical morphism is a quasi-isomorphism. The long exact sequence in homologyand the 5-lemma imply that the morphism Cf → Cg is a quasi-isomorphism, con-cluding the proof.

3.3. Derived categories and derived functors. As a preparation for the defi-nition of derived categories, let us reconsider the construction of classical derivedfunctors between abelian categories. Thus, let F : A → B be a right exact functorbetween abelian categories and let A have enough projective objects. The key stepsin the construction of the LkF, k ≥ 0, consisted of choosing projective resolutions

(3.24) . . .→ P2 → P1 → P0ε→ X

for X ∈ A, applying F to the truncated chain complex

P = (. . .→ P2 → P1 → P0),

Page 27: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 27

and then passing to homology. A different way of saying that (3.24) is a resolutionis to say that the commutative diagram

(3.25)

...

...

P2

// 0

P1//

0

P0 ε// X

defines a quasi-isomorphism ε : P → X where X also denotes the associated com-plex concentrated in degree 0, i.e., the complex (. . .→ 0→ X → 0→ . . .) ∈ Ch(A).

This simple trick of rewriting the horizontal diagram (3.24) in a vertical fashion asin (3.25) is an important step. In fact, it suggests that, secretely in the constructionof classical derived functors, we already passed from objects in the abelian category(like modules over a ring) to chain complexes. From a more abstract perspectivethis amounts to passing from a good old category to a ‘homotopy theory’, namelyto the ‘homotopy theory’ (Ch(A),WA) consisting of the category Ch(A) togetherwith the class W = WA of quasi-isomorphisms. We say a bit more about ‘abstracthomotopy theories’ in §6.3 and Appendix B.

Promise.The passage to stalk complexes defines a fully faithful functor A → Ch(A),

suggesting that if we understand Ch(A) we also understand A. For various reasonswe would like to invert the quasi-isomorphisms WA in Ch(A). For example thiswould allow us to identify all projective resolutions of a fixed object X ∈ A.

Add more motivationalcomments?Definition 3.26. Let A be an abelian category. The derived category D(A) of

A is the localization of Ch(A) at the class WA of quasi-isomorphisms,

D(A) = Ch(A)[W−1A ].

By the very definition, the derived category is hence a pair (D(A), γ) consistingof a category D(A) and a localization functor γ : Ch(A)→ D(A) which sends quasi-isomorphisms to isomorphisms and which is initial with this property. Thus, everyfunctor F : Ch(A) → C which sends quasi-isomorphisms to isomorphisms factorsuniquely through D(A),

Ch(A)∀F //

γ

C, F (WA) ⊆ IsoC ,

D(A).

∃!

<<

Warning 3.27. There is a typical warning in the context of such localizations,namely, that, in general, such localizations do not exist (at least not in a fixeduniverse). A different way of saying this is that such localizations can always beconstructed (see for example [GZ67]), but, in general, the resulting categories are

Page 28: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

28 MORITZ GROTH

not locally small, i.e., there might be proper classes of morphisms between certainobjects. There are various ways of dealing with this issue.

(i) One can ignore these issues (as it was done in the more classical literature),reassuring the reader every now and then by claiming that all steps can bejustified.

(ii) One can restrict attention to the large class of Grothendieck abelian categoriessince in that case all these set-theoretic problems happen to disappear. Seefor example [Fai73, §14] for a discussion of Grothendieck abelian categories.

(iii) The language of Quillen model categories [Qui67, Hov99, DS95] is an abstractframework which, among many other things, allows us to deal with suchproblems. As soon as the pair (Ch(A),WA) can be extended to a Quillenmodel category, these set-theoretic problems automatically disappear. And,in fact, in the case of Grothendieck abelian categories such model structuresalways exist; see [Hov01].

(iv) If one imposes certain boundedness conditions on the chain complexes andassumes the existence of sufficently many injective or projective objects, thenone can show that corresponding derived categories are equivalent to quo-tient categories of suitable categories of chain complexes. In particular, thesecategories are again locally small categories.

For an actual example of such set-theoretic problems we refer to an old example ofFreyd (see the reprint [Fre03]) as discussed in [Kra10, Example 4.15].

For the remainder of this section we assume that the derived categories underconsideration exist. (Thus, we essentially follow approach (i).)

Examples 3.28. (i) Let R be a ring and let Mod(R) be the Grothendieck abeliancategory of R-modules. The derived category of the ring R is the derivedcategory

D(R) = D(Mod(R)) = Ch(R)[W−1R ].

In particular in the case of finite-dimensional algebras over fields, such derivedcategories have been studied a lot in representation theory; see for example [?,?, ?].

(ii) Let X be a scheme, let Mod(X) be the category of quasi-coherent OX -modules, and let Ch(X) = Ch(Mod(X)) be the corresponding category ofunbounded cochain complexes. The derived category of the scheme Xis the derived category

D(X) = D(Mod(X)) = Ch(X)[W−1X ].

References in this case include [?, ?, ?, ?].

Corollary 3.29. Let A,B be abelian categories and let F : Ch(A)→ Ch(B) be anexact functor. There is a unique functor F : D(A)→ D(B) such that

Ch(A)F //

γ

Ch(B)

γ

D(A)F// D(B)

commutes. This applies, in particular, if F comes from an exact functor F : A → B.

Page 29: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 29

Proof. Since the functor γ F : Ch(A) → D(B) sends quasi-isomorphisms to iso-morphisms this is immediate from the defining universal property of the localizationfunctor γ : Ch(A) → D(A) (see Definition 3.26). The special case coming from anexact functor A → B is simply Lemma 3.6.

A first trivial example comes from the homology functors. Given an abeliancategory A, we denote by AZ the category of Z-graded objects in A. (Thus, weconsider Z as a discrete category and AZ is the corresponding diagram category.)Taking all homology functors Hk : Ch(A)→ A, k ∈ Z, at once we obtain a functorH∗ : Ch(A)→ AZ.

Example 3.30. Let A be an abelian category. The functor H∗ : Ch(A)→ AZ sendsquasi-isomorphisms to isomorphisms and hence induces a functor H∗ : D(A)→ AZ.

The following example is important.

Example 3.31. Let A be an abelian category. The cone C : Ch(A)[1] → Ch(A)induces by Proposition 3.23 and Lemma 3.10 a functor

C : D(A[1])→ D(A).

Thus, to emphasize, there is a functorial cone construction defined on the derivedcategory of the arrow category of an abelian category. We will get back to this later.

As recalled in §2, many additive functors showing up in nature are only left exactor right exact. In such situations one has to work harder in order to obtain certainuniversal induced functors at the level of derived categories.

Definition 3.32. Let A,B be abelian categories and let F : Ch(A) → Ch(B) bea functor. A left derived functor of F is a pair (LF, ε) consisting of a functorLF : D(A) → D(B) and a natural transformation ε : (LF )γ → γF such that forevery other such pair (G : D(A) → D(B), α : Gγ → γF ) there is a unique naturaltransformation α′ : G→ LF such that

α = ε (α′γ) : Gγ → (LF )γ → γF.

Thus, a left derived functor can be depicted by a square populated by a naturaltransformation as in

Ch(A)F //

γ

Ch(B)

γ

D(A)LF// D(B)

@Hε

and the universal property of a left derived functor can be illustrated as follows

Ch(A)F //

γ

Ch(B)

γ

=

Ch(A)F //

γ

Ch(B)

γ

D(A)G// D(B)

@Hα

D(A) ⇑α′LF))

G

55 D(B).

@Hε

Remark 3.33. (i) The idea behind the definition is that the left derived functor(LF, ε) is a universal approximation of F by an exact functor.

Page 30: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

30 MORITZ GROTH

(ii) Since derived functors are defined by a universal property, once one knowsthat derived functors exist, then they are unique up to a canonical naturalisomorphism. The existence of derived functors is often settled using suitableresolutions.

(iii) There is the dual notion of a right derived functor consisting of a functorRF : D(A)→ D(B) and a universal natural transformation η : γF → (RF )γsuch that (RF, η) is suitably initial.

(iv) We warn the reader that Definition 3.32 differs slightly from the one found intypical books on homological algebra; see [Wei94, GM03]. While in those ref-erences in the definition of a left derived functor both F and LF are assumedto be exact functors of triangulated categories (see §4), here we define derivedfunctors as it is typically done in the theory of Quillen model categories.

In §3.5 we show that the functor C : D(A[1]) → D(A) constructed in Exam-ple 3.31 together with a suitable natural transformation is a left derived functor ofthe cokernel cok: Ch(A[1]) → Ch(A) (see Theorem 3.42). As a preparation we es-tablish in §3.4 that derived categories enjoy a seemingly stronger universal propertythan that one of a localization.

3.4. Derived categories as 2-localizations. Derived categories of abelian cate-gories are defined as suitable localizations (Definition 3.26) and as such they satisfya universal property. In this subsection we show that they actually enjoy a strongeruniversal property, namely that derived categories are 2-localizations. This seem-ingly technical result has interesting consequences as we see here and in §§3.5-3.6.

We begin collecting the following lemma.

Lemma 3.34. Let C be a category, let A be a small category, and let X,Y : A→ Cbe functors. There is a bijection between

(i) the set of natural transformations α : X → Y ,(ii) the set of functors H : A× [1]→ C such that H(−, 0) = X and H(−, 1) = Y ,

and(iii) the set of functors K : A→ C[1] such that 0∗K = X and 1∗K = Y .

Proof. This proof is left to the reader.

Let us consider an abelian category A together with the localization functorγ : Ch(A)→ D(A). For every category C, precomposition of functors with γ yieldsa map

γ∗ : hom(D(A), C)→ hom(Ch(A), C).If we denote by homW(Ch(A), C) ⊆ hom(Ch(A), C) those functors which send quasi-isomorphisms to isomorphisms, then the defining universal property of γ is that forall categories C precomposition with γ induces a bijection

γ∗ : hom(D(A), C)∼=→ homW(Ch(A), C).

It turns out that this can be refined to an isomorphism of categories. For thispurpose, we denote by

FunW(Ch(A), C) ⊆ Fun(Ch(A), C)

the full subcategory of the functor category spanned by all functors which sendquasi-isomorphisms to isomorphisms.

Page 31: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 31

Lemma on path objects Proposition 3.35. Let A be an abelian category. The localization γ : Ch(A) →D(A) is a 2-localization, i.e., for every category C precomposition with γ inducesan isomorphism of categories

γ∗ : Fun(D(A), C)∼=→ FunW(Ch(A), C).

In particular, for every pair of functors F ′, G′ : D(A) → C there is a bijectionγ∗ : hom(F ′, G′)→ hom(F ′γ,G′γ).

Proof. By definition of the localization, γ∗ is bijective on objects and it hencesuffices to show that γ∗ is also fully faithful. Let F,G ∈ FunW(Ch(A), C) withcorresponding factorizations F = F ′γ, G = G′γ and let α : F → G be a natu-ral transformation. Note that α : F → G is equivalently specified by the functorα : Ch(A)→ C[1] defined by

α(X) = (αX : FX → GX).

If w : X → Y is a quasi-isomorphism, then, since F,G send quasi-isomorphisms toisomorphisms, the vertical morphisms in

FXα //

∼= Fw

GX

∼=Gw

FYα// GY,

are isomorphisms, showing that α ∈ FunW(Ch(A), C[1]). The universal property of

γ implies that there is a unique functor α′ : D(A)→ C[1] such that

Ch(A)α //

γ

C[1]

D(A)∃!α′

II

commutes. But α′ now corresponds to a natural transformation α′ : F ′ → G′

defined by

α′(X) = (α′X : F ′X → G′X), X ∈ D(A).

The reader easily checks that this is the unique natural transformation F ′ → G′

satisfying α′γ = α.

As a first illustration of this 2-categorical universal property we collect the follow-ing result. This result makes precise the slogan that ‘exact functors (Definition 3.7)do not have to be derived’; see again Corollary 3.29.

Corollary 3.36. Let A,B be abelian categories, let F : Ch(A) → Ch(B) be exact,and let F : D(A)→ D(B) be the unique induced functor such that

Ch(A)F //

γ

Ch(B)

γ

D(A)F// D(B)

commutes.

Page 32: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

32 MORITZ GROTH

(i) The pair (F : D(A)→ D(B), id : F γ → γ F ) is a left derived functor of F .(ii) The pair (F : D(A)→ D(B), id : γF → F γ) is a right derived functor of F .

Proof. We take care of statement (i), the case of (ii) is dual. To this end, let

(G : D(A)→ D(B), α : G γ → γ F )

be an instance of a pair among which a left derived functor is supposed to be ter-minal (see Definition 3.32). Since γ F = F γ, the natural transformation α readsas α : G γ → F γ. By Proposition 3.35 there is a unique natural transformationα′ : G→ F such that

α = α′γ = id (α′γ),

i.e., we have

Ch(A)F //

γ

Ch(B)

γ

=

Ch(A)F //

γ

Ch(B)

γ

D(A)G// D(B)

@Hα

D(A) ⇑α′F))

G

55 D(B).

@Hid

Thus, (F, id) enjoys the universal property of a left derived functor.

Using that γ : Ch(A) → D(A) is a 2-localization, in the above proof we couldavoid the explicit construction of a natural transformation between functors definedon derived categories. This is convenient since, in general, morphisms in derivedcategories are a bit tricky to understand. We will see a further instance of such asituation in the next subsection.

3.5. Cones as derived cokernels. We now take a closer look at the cone con-struction C : Ch(A[1])→ Ch(A) as defined via the pushout square (3.18). Clearly,CX0 → 0 induces a morphism of spans (CX0 ← X0 → X1) → (0 ← X0 → X1),and hence an induced morphism of pushout diagrams

X0f

//

= ##

X1=

%%

X0//

X1

CX0

""

// Cf

$$

0 // cok(f).

The exact sequence of Lemma 3.21 is easily seen to be natural in the pushoutdiagram, hence we obtain a natural morphism of exact sequences

(3.37)

0 // X0//

= ∼

X1 ⊕ CX0

// Cf //

0

X0f

// X1// cok(f) // 0,

and, in particular, a natural transformation

(3.38) φ : Cf → cok(f).

Page 33: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 33

Remark 3.39. We observe that the first two components of the natural morphismof exact sequences (3.37) define a natural levelwise quasi-isomorphism

(3.40)

X0//

= ∼

X1 ⊕ CX0

X0f

// X1.

Moreover, the domain of this quasi-isomorphism is a monomorphism and we knowby Lemma 3.13 that the cokernel preserves quasi-isomorphisms between monomor-phisms. We refer to this by saying that (3.40) is a functorial resolution onCh(A[1]) which is adapted to the functor cok: Ch(A[1])→ Ch(A). It follows thatthe morphism φ : Cf → cok(f) is a quasi-isomorphism for all monomorphisms f .

As a preparation for the theorem we recall that, by Example 3.31, the coneinduces a functor C : D(A[1])→ D(A). The natural transformation (3.38) yields anatural transformation ε : C γ → γ cok given by

(3.41) ε : C γ = γ C γφ→ γ cok.

We also collect the corresponding dual statement concerning the fiber functorF : D(A[1])→ D(A). Since its construction is dual to that of C : D(A[1])→ D(A),we leave the details to the reader.

Theorem 3.42. Let A be an abelian category.

(i) The cone functor C : D(A[1]) → D(A) together with ε defined in (3.41) is aleft derived functor of the cokernel cok: Ch(A[1])→ Ch(A).

(ii) The fiber functor F : D(A[1]) → D(A) together with a dually defined η is aright derived functor of the kernel ker : Ch(A[1])→ Ch(A).

Proof. We have to show that (C, ε) satisfies the universal property of Definition 3.32.For this purpose let G : D(A[1]) → D(A) be a functor and let α : G γ → γ cokbe a natural transformation,

Ch(A[1])cok //

γ

Ch(A)

γ

D(A[1])G// D(A).

AI

We claim that it is enough to show that there is a unique natural transformationψ : G γ → C γ such that

α = ε ψ : G γ → C γ → γ cok.

In fact, by Proposition 3.35 we then deduce that there is a unique natural trans-formation α′ : G → C such that ψ = α′γ, which immediately yields the desireduniversal property of (C, ε).

Page 34: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

34 MORITZ GROTH

In order to construct such a transformation ψ we consider (X0f→ X1) ∈ Ch(A[1])

and associate to it the following commutative diagram

Gγ(X0 → X1 ⊕ CX0)α //

∼=

γcok(X0 → X1 ⊕ CX0)

Cγ(X0 → X1 ⊕ CX0)ε∼=oo

∼=

Gγ(X0 → X1)α

// γcok(X0 → X1) Cγ(X0 → X1).ε

oo

In this diagram, the vertical morphisms are induced by the resolutions adaptedto the cokernel (see Remark 3.39) and it hence follows that the two outer verticalmorphisms are isomorphisms. Moreover, both squares commute by naturality of αand ε, respectively. Finally, the top horizontal morphism

ε = γφ : Cγ(X0 → X1 ⊕ CX0)→ γcok(X0 → X1 ⊕ CX0)

is an isomorphism since X0 → X1 ⊕ CX0 is a monomorphism; see again Re-mark 3.39. Thus, the above commutative diagram shows us that there is at mostone such natural transformation ψ. These morphisms Gγ(f)→ Cγ(f) actually as-semble to a natural transformation Gγ → Cγ as one can check from the naturalityof the above diagram in (f : X0 → X1).

Note that Proposition 3.35 again allowed us to avoid the explicit construction ofa natural transformation between functors defined on derived categories.

Remark 3.43. We include a short philosophical remark concerning the constructionof the cone as a derived version of the cokernel. The cokernel is a categoricalconstruction which amounts to ‘collapsing’ the image of a morphism. As observedin Warning 3.11 the cokernel is not an exact or homotopy invariant construction inthat levelwise quasi-isomorphisms are not always sent to quasi-isomorphisms.

One obtains the cone from the cokernel by, instead of collapsing the image,simply ‘adding the potential of collapsing the image’. This is made precise by thedefining pushout square (3.18). In fact, thinking geometrically, by glueing a coneon the image of the morphism we add the potential of collapsing it since we cannow push it to the apex of the cone. Similar comments also apply to other derived(co)limit constructions and this picture also extends to the context of topologicalspaces and homotopy (co)limits.

3.6. Coherent versus incoherent diagrams. In this subsection we briefly dis-cuss differences between coherent and incoherent diagrams. The main point of thetheory of derivators is that many interesting constructions are available at the levelof coherent diagrams while, in general, this is not the case for incoherent diagrams.

To already get used to the terminology from the theory of derivators, we make thefollowing definition. Let us recall that given an abelian category B the localizationfunctor γ : Ch(B) → D(B) can be chosen to be the identity on objects. Hence,objects in the derived category are again simply chain complexes.

Definition 3.44. Let A be an abelian category and let A be a small category.

(i) An object in D(AA) is a coherent diagram of shape A.(ii) An object in D(A)A is an incoherent diagram of shape A.

Thus, a coherent diagram is simply a chain complex X ∈ Ch(AA), i.e., a chaincomplex of functors A → A. Under the isomorphism Ch(AA) ∼= Ch(A)A from

Page 35: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 35

Lemma 3.10 this corresponds to a diagram X : A → Ch(A). The adjective ‘coher-ent’ is meant to indicate that this is a strictly commutative diagram in the sensethat it preserves compositions and identies on the nose.

In contrast to this, an incoherent diagram is a functor X : A → D(A). Let usthink of the derived category as being realized as a quotient category, i.e., objectsare suitably nice chain complexes and morphisms are chain homotopy classes. Anincoherent diagram then is merely a homotopy-commutative diagram. For example,for every pair of composable morphisms f : a0 → a1 and g : a1 → a2 in A it followsthat the triangle on the right in the diagram

a1

g

'

X(a1)

X(g)

$$

a0

f

??

gf// a2, X(a0)

X(f)::

X(gf)// X(a2),

commutes up to an unspecified chain homotopy. (In that triangle we of courseabused notation and wrote X(f), X(g), and X(gf) respectively for representativesof the corresponding chain homotopy classes.) Now, in general, it is not possibleto replace a homotopy-commutative diagram up to quasi-isomorphism by a strictlycommutative one.

Put differently, the two categories D(AA) and D(A)A are honestly differentcategories. As an additional application of Proposition 3.35 we observe next thatthese categories are related by a functor D(AA)→ D(A)A. Since the constructionof this functor is a bit abstract, we discuss some specific examples in §3.7.

Let us denote by 1 the terminal category consisting of one object ∗ and itsidentity morphism only. For any category A the evaluation at this unique objectyields an isomorphism of categories

A1 ∼= A.

Given an object a ∈ A we also write a : 1 → A for the corresponding functor.Similarly, any morphism f : a→ b in A yields a natural transformation

1

a''

b

77 A,

which we again denote by f .

Lemma 3.45. Let A be an abelian category, let A be a small category, and leta ∈ A. The evaluation functor a∗ : Ch(AA)→ Ch(A) sends quasi-isomorphisms toquasi-isomorphisms. Hence, there is a unique evaluation functor a∗ : D(AA)→D(A) such that the following diagram commutes,

Ch(AA)a∗ //

γ

Ch(A)

γ

D(AA)a∗// D(A).

Proof. This is immediate from Lemma 3.10 and Definition 3.26.

Page 36: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

36 MORITZ GROTH

Given a morphisms of coherent diagrams g : X → Y in D(AA) we simplifynotation by writing

ga : Xa → Ya, a ∈ A,for the image of g under a∗ : D(AA)→ D(A).

Now, every morphism f : a → b in A induces a transformation f∗ : a∗ → b∗

between the corresponding evaluation functors a∗, b∗ : AA → A. A levelwise ap-plication gives rise to a natural transformation f∗ : a∗ → b∗ between chain-levelevaluation functors a∗, b∗ : Ch(AA)→ Ch(A).

Lemma 3.46. Let A be an abelian category, let A be a small category, and letf : a → b be a morphism in A. There is a unique induced natural transformationf∗ : a∗ → b∗ between a∗, b∗ : D(AA)→ D(A) such that γf∗ = f∗γ,

Ch(AA) ⇓f∗a∗))

b∗55

γ

Ch(A)

γ

D(AA) ⇓f∗a∗))

b∗55 D(A).

Proof. This is immediate from Proposition 3.35 and Lemma 3.45.

Given a coherent diagram X ∈ D(AA), we denote the corresponding componentof the natural transformation f∗ : a∗ → b∗ by

Xf : Xa → Xb,

which is a morphism in D(A). As always we see that uniqueness implies functori-ality. Given morphisms f : a→ b and g : b→ c, the uniqeness implies the equations

Xg Xf = Xgf and Xida = idXa .

Thus, associated to every coherent diagram X ∈ D(AA) we obtain an incoherentdiagram diaA(X) : A→ D(A) defined by

diaA(X) : A→ D(A), a 7→ Xa, f 7→ Xf .

We refer to diaA(X) as the underlying diagram of X.Finally, let us again consider a morphism g : X → Y in D(AA). Then there is

a natural transformation diaA(g) : diaA(X)→ diaA(Y ) between the correspondingincoherent diagrams. In fact, the component of diaA(g) at an object a ∈ A is simplythe morphism

diaA(g)a = ga : Xa → Ya.

We leave it to the reader to check that diaA(g) is a natural transformation. Thefunctoriality of this construction is summarized in the following proposition.

Proposition 3.47. Let A be an abelian category and let A be a small category.The assignments X 7→ diaA(X) and g 7→ diaA(g) define a functor

diaA : D(AA)→ D(A)A

Proof. This proof is left as an exercise.

We refer to this functor as the underlying diagram functor.

Page 37: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 37

Warning 3.48. (i) Just to re-emphasize, the domain and the target categoriesof diaA are very different. While the domain is the derived category of adiagram category the target is a diagram category of the derived category.In general, diaA is not an equivalence and it is important to distinguishbetween these two categories. Intuitively speaking, the functor diaA takesa strict diagram to the underlying homotopy-commutative diagram, therebydiscarding important information.

(ii) As a special case let us consider A = [1] and the associated underlying dia-gram functor dia[1] : D(A[1])→ D(A)[1] from the derived category of the arrowcategory to the arrow category of the derived category. One slogan concerningdefects of triangulated categories (see §4) is that cone constructions are notfunctorial at that level. In the case of derived categories, this statement takesthe more precise form that the cone functor C : D(A[1])→ D(A) (see Exam-ple 3.31) does not factor through dia[1] : D(A[1]) → D(A)[1], i.e., in general,there is no dashed arrow

D(A[1])C //

dia[1]

D(A)

D(A)[1]

@

::

making the diagram commute. In general, the functorial cone constructiononly exists at the level of coherent diagrams.

While we come back to the above warning in §9.5, in the following subsection wePromise!

illustrate the construction of underlying diagram functors. In particular, this yieldsexamples of underlying diagram functors which are far from being equivalences.

3.7. The case of category algebras. Since the construction of underlying dia-gram functors (Proposition 3.47) was fairly abstract, in this subsection we want toidentify them in more specific situations. These special cases illustrate a few fairlytypical features of underlying diagram functors.

To begin with we collect a lemma about the homology functor (Example 3.30) inthe context of semisimple abelian categories, i.e., abelian categories in whichevery short exact sequence splits.

Lemma 3.49. For every semisimple abelian category A the homology functorH∗ : D(A)→ AZ is an equivalence of categories.

Proof. We refer the reader to [GM03, §] for a discussion of this lemma. Check ref!

As a special case, given a semisimple ring R (for example, R could be a field k),the homology functor is an equivalence of categories

H∗ : D(R) ' Mod(R)Z.

This observation will allow us to identify particular instances of underlying diagramfunctors with suitable homology functors.

Let A be a small category with finitely many objects only and let R be a ring.The category algebra RA is the following R-algebra. The underlying R-moduleof RA is free with basis given by the morphisms in A,

RA =⊕

f : a0→a1

R.

Page 38: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

38 MORITZ GROTH

Let us denote by ef ∈ RA the generator corresponding to the morphism f : a0 → a1.The multiplication is essentially the R-bilinear extension of the composition in A,i.e., it is the bilinear extension of the assignment

(eg, ef ) 7→egf , s(g)=t(f),0 , otherwise.

It is easy to see that Σa∈Aeida ∈ RA is a neutral element for this multiplication(and this is the reason why we assumed A to have finitely many objects only).

Having a module over a category algebra RA is as good as having a diagramX : A → Mod(R). More precisely, given such a diagram X, the reader checksthat the direct sum

⊕a∈AX(a) ∈ Mod(R) actually can be turned into an RA-

module. Moreover, this construction extends to a functor and there is the followingstatement about this functor.

Proposition 3.50. Let A be a category with finitely many objects only and let Rbe a ring. There is an equivalence of categories

Mod(R)A ' Mod(RA) : X 7→⊕a∈A

X(a).

Proof. This proof is left as an exercise. As a hint we suggest to take a look at theidempotent elements eida ∈ RA for all a ∈ A in order to establish the essentialsurjectivity.

As general references for representations of category algebras and small categorieswe mention [?, ?]. Here we content ourselves by illustrating the notion of a categoryalgebra by the following examples.

Examples 3.51. Let R be a ring.

(i) Any discrete group G can be considered as a category with one object only(as in §2.3). In this case the category algebra is simply the group algebra RGand the proposition reproduces the equivalence of categories

Mod(R)G ' Mod(RG).

(ii) Let us recall that a quiver Q is simply an oriented graph, hence a quadrupelQ = (Q0, Q1, s, t : Q1 → Q0) given by a set of vertices, a set of edges, andsource and target maps. Abusing notation, we also denote by Q the corre-sponding path category, i.e., the category freely generated by the quiver Q(this can be thought of as a many object version of the construction of the freemonoid on a set). Assuming Q0 to be finite, the category algebra RQ is thepath algebra of the quiver and the proposition specializes to the equivalence

Mod(R)Q ' Mod(RQ).

(iii) Let P = (P0,≤) be a partially ordered set. We abuse notation and denoteby P the category with P0 as set of objects and morphisms given by

homP (x, y) =

∗ , x ≤ y,∅ , otherwise.

Under the assumption that P0 is finite, in this case the category algebra RPis the incidence algebra of P and the proposition yields an equivalence ofcategories

Mod(R)P ' Mod(RP ).

Page 39: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 39

Proposition 3.50 describes an equivalence of abelian categories. Every suchequivalence induces a derived equivalence in the following sense.

Lemma 3.52. Let F : A ' B be an equivalence of abelian categories A,B. Thenapplying F levelwisely induces equivalences

F : AZ ' BZ, F : Ch(A) ' Ch(B), and F : D(A) ' D(B).

Proof. This proof is left as an exercise.

We want to apply this lemma to the situation of Proposition 3.50. Thus, let Abe a small category with finitely many objects only and let R be a ring. Theequivalence Mod(R)A ' Mod(RA) induces the two vertical equivalences in thediagram

(3.53)

D(Mod(R)A)diaA //

'

D(R)AH∗ // (Mod(R)Z)A

∼= // (Mod(R)A)Z

'

D(RA)H∗

// Mod(RA)Z.

Lemma 3.54. For every small category A with finitely many objects only and forevery ring R the diagram (3.53) commutes up to a canonical natural isomorphism.

Proof. Unraveling definitions, this is an immediate consequence of the fact thathomology is additive. In more detail, using the explicit description of the equiv-alence in Proposition 3.50 we see that if we trace X ∈ D(Mod(R)A) clockwiselythrough (3.53) then we obtain

⊕a∈AH∗(Xa). Tracing the object X counterclock-

wisely through (3.53) we get H∗(⊕

a∈AXa). Since homology commutes with finitedirect sums this shows that the two compositions are canonically naturally isomor-phic.

In the semisimple case there is the following corollary (see again Definition 2.20).

Corollary 3.55. Let A be a small category with finitely many objects only and let Rbe a semisimple ring. The functor diaA : D(Mod(R)A) → D(R)A is equivalentto the homology functor H∗ : D(RA) → Mod(RA)Z. In particular, diaA is anequivalence if and only if H∗ is an equivalence.

Proof. Since R is semisimple, the homology functor H∗ : D(R) → Mod(R)Z is anequivalence of categories by Lemma 3.49. Hence the same is true for the inducedfunctor H∗ : D(R)A → (Mod(R)Z)A and the first claim follows since (3.53) com-mutes up to natural isomorphism (Lemma 3.54). Finally, it is easy to see thatif two functors F1, F2 are equivalent in the sense of Definition 2.20 then F1 is anequivalence if and only if F2 is an equivalence.

If we specialize to the case of a field and take up again Examples 3.56 then weobtain the following examples.

Examples 3.56. Let k be a field.

(i) For every discrete group G the functor diaG : D(Mod(k)G)→ D(k)G is equiv-alent to the homology functor

H∗ : D(kG)→ Mod(kG)Z.

Page 40: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

40 MORITZ GROTH

(ii) For every quiver Q with only finitely many objects the underlying diagramfunctor diaQ : D(Mod(k)Q)→ D(k)Q is equivalent to the homology functor

H∗ : D(kQ)→ Mod(kQ)Z.

(iii) For every finite, partially ordered set P the underlying diagram functordiaP : D(Mod(k)P )→ D(k)P is equivalent to the homology functor

H∗ : D(kP )→ Mod(kP )Z.

These examples illustrate the following two points.

Remark 3.57. (i) In general, underlying diagram functors are far from beingequivalences. In fact, an abstract way to see this is as follows. The derivedcategories in Examples 3.56 are at the same time triangulated categories(see §4) and abelian. But any such category is necessarily semisimple — aproperty which in general is not enjoyed by group algebras, path algebras,and incidence algebras. (We will later take a closer look at specific exam-ples of underlying diagram functors, showing more explicitly that one has todistinguish between coherent and incoherent diagrams.)

(ii) Even if we are in one of the few cases in which derived categories admit(co)limits different from (co)products, then derived (co)limits and categorical(co)limits of course do not match up to an application of underlying diagramfunctors.

As a specific example, let k be a field and let G be a discrete group. Sincethe derived category D(k) is equivalent to Mod(k)Z it admits colimits ofshape G. However, the diagram

D(Mod(k)G)diaG //

LcolimG

D(k)G' // (Mod(k)Z)G

colimG

D(Mod(k))H∗

// Mod(k)Z

does not commute up to a natural isomorphism. Since LcolimG is equivalentto the group homology functor H∗(G;−) (see again §2.3), by Examples 3.56the previous diagram commutes up to a natural isomorphism if and only ifthis is the case for

D(Mod(kG))H∗ //

H∗(G;−)

Mod(kG)Z

(−)G

D(Mod(k))H∗

// Mod(k)Z.

Now, consider any G and M ∈ Mod(kG) such that the group homologyHn(G;M) is non-trivial for some n > 0. Then (H∗M)G is the vector spaceMG concentrated in degree zero, while this is not the case for the imageof M under the remaining composition. In particular, neither of the abovediagrams commutes up to a natural isomorphism.

Page 41: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 41

4. Derived categories as triangulated categoriesTODO: proof-read and

expand this section. Derived categories are obtained from categories of chain complexes by invertingthe class of quasi-isomorphisms. It is important to note that these two categories(the category of chain complexes and the derived category) behave quite differently.

(i) Categories of chain complexes are again abelian and as such they are finitelycomplete and finitely cocomplete. In contrast to this, in general, the onlylimits and colimits which exist in derived categories are finite biproducts.

(ii) More importantly, at the level of chain complexes one can construct functorialresolutions of diagrams leading to derived limits and derived colimits (see §2.3and §3.5 for special cases). However, this calculus of derived limits andderived colimits is not visible at the level of the derived category.

While (i) is a matter of fact one has to live with, there are various ways of tryingto deal with (ii). One way is given by endowing derived categories with additionalstructure, thereby turning them into triangulated categories. These triangulationsessentially encode certain shadows of iterated cofiber constructions, hence by §3.5of iterated derived cokernel constructions.

In this section we review some basics concerning triangulated categories in gen-eral and the classical triangulations on derived categories in particular. While herewe only sketch the construction of these classical triangulations most details areomitted. The point being that these triangulations are special cases of canonicaltriangulations in stable derivators which we establish in detail in §9.

Classically, the derived category is obtained by first passing to the homotopyAdapt the remainder ofintroduction.category of an abelian category and observing that it can be turned into a trian-

gulated category. The derived category is then obtained by a further localizationand, using a theorem of Verdier, this allows one to conclude that derived categoriesare triangulated categories. Here we content ourselves by collecting a few of thekey steps.

Despites being a very sucessful theory, triangulated categories suffer certain de-fects. We conclude this section by a short discussion of such defects. Later in thisbook we will see that these defects can be fixed by working with stable derivatorsinstead.

4.1. The homotopy category of an abelian category. In this subsection weconsider the homotopy category K(A) of an abelian category A. This categoryoccurs as an intermediate step in the classical construction of the derived cate-gory D(A) = Ch(A)[W−1

A ]. Since the chain homotopy relation ' is a congruencerelation (i.e., we have equivalence relations on all homCh(A)(X,Y ) which are com-patible with compositions), we can make the following definition.

Definition 4.1. Let A be an abelian category. The homotopy category K(A)of A is the quotient category

K(A) = Ch(A)/ ' .

Thus, objects in K(A) are simply chain complexes and morphisms are chainhomotopy classes of morphisms. The composition is defined by choosing represen-tatives and passing to the chain homotopy class of the corresponding composition.

By definition there is a quotient functor γ′ : Ch(A)→ K(A) which is the univer-sal example of a functor identifying chain homotopic maps. Note that a chain map

Page 42: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

42 MORITZ GROTH

f : X → Y is a chain homotopy equivalence if and only if γ′(f) is an isomorphism.Similarly, a chain complex X is sent to zero in K(A) if and only if X is contractible.

As a preparation for a different characterization of γ′, we recall the cylinderconstruction at the level of chain complexes. This construction is a further instanceof the biadditive pairing described in Digression 3.15. It might be convenient tobriefly recall Examples 3.17.

Examples 4.2. (i) Let I+ = Ccell([0, 1]+) ∈ Chb(FZ) be the reduced cellularchain complex of [0, 1]+ = [0, 1]t∗ endowed with the CW structure consistingof three 0-cells and one 1-cell only. More explicitly, we have I0 = Z⊕Z, I1 = Z,and Ik = 0, k 6= 0, 1 with non-trivial differential (−id, id) : Z ⊕ Z → Z. Thecontinuous map [0, 1]→ ∗ induces a chain map I+ → S0 which is easily seento be a chain homotopy equivalence. The cylinder cyl(X) of X ∈ Ch(A) isdefined as

cyl(X) = I+ ⊗X.

It follows from Digression 3.15, Examples 3.17, and Lemma 3.8 that the chainmap I+ → S0 induces a natural chain homotopy equivalence cyl(X)→ X.

(ii) Let S0+ = Ccell(0, 1+) ∈ Chb(FZ) be the reduced cellular chain complex

of 0, 1+ endowed with the CW structure consisting of three 0-cells only.Thus, S0

+∼= S0 ⊕ S0 and for X ∈ Ch(A) there is a natural isomorphism

S0+ ⊗X ∼= X ⊕X.

(Recall that the space 0, 1+ is the wedge of two copies of 0, 1.)(iii) The continuous map 0, 1 → [0, 1] yields a map S0

+ → I+ in Chb(FZ) andhence a natural transformation S0

+⊗− → I+⊗− of functors Ch(A)→ Ch(A).Under the above isomorphisms, for X ∈ Ch(A) this yields the usual naturalinclusion i : X ⊕ X → cyl(X). We denote the two resulting inclusions byi0, i1 : X → cyl(X).

The point of this cylinder construction is the following.

Lemma 4.3. Let f, g : X → Y be chain maps. There is a bijection between chainhomotopies s : f → g and chain maps H : cyl(X)→ Y making the following diagramcommute

X

i0 $$

f

cyl(X)H // Y.

X

i1::

g

@@

Proof. This proof is left as an exercise.

Thus, as in topology, the cylinder is the natural domain for homotopies. In thespecial case of the constant chain homotopy of the identity id : X → X one checks

Page 43: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 43

that the resulting homotopy

(4.4)

X

i0 $$

id

cyl(X)HX // X

X

i1::

id

@@

is the chain homotopy equivalence cyl(X) → X from Examples 4.2(i). These ob-servations can be reformulated as follows.

Lemma 4.5. Let A be an abelian category and let C be a category. The followingare equivalent for a functor F : Ch(A)→ C.

(i) For every pair of chain homotopic maps f, g we have Ff = Fg.(ii) The functor F sends chain homotopy equivalences to isomorphisms.

Proof. Clearly, (i) implies (ii). Conversely, (ii) implies that F sends the chainhomotopy equivalences HX in (4.4) to isomorphisms in C. Consequently, the inclu-sions i0, i1 : X → cyl(X) have the same images in C and statement (i) is hence animmediate consequence of Lemma 4.3.

Corollary 4.6. Let A be an abelian category and let HEA be the class of chainhomotopy equivalences in Ch(A). The functor γ′ : Ch(A) → K(A) exhibits K(A)as a localization of Ch(A) at the class of chain homotopy equivalences,

K(A) ∼= Ch(A)[HE−1A ].

Proof. In view of Lemma 4.5 this is simply a reformulation of the universal propertyof the quotient functor γ′ : Ch(A)→ K(A).

This corollary shows us that the homotopy category is indeed an intermediatestep in the construction of the derived category. Since K(A) is a quotient categoryit is simpler to put ones hands on it.

Lemma 4.7. The homotopy category K(A) of an abelian category is additive.

Proof. This is a special case of the more general fact that quotients of additivecategories by additive congruence relations are again additive. We leave the detailsto the reader.

Warning 4.8. It is not true, in general, that the homotopy category of an abeliancategory is abelian.

The additive category K(A) can be endowed with some additional structurewhich can be thought of as shadows of the existence of short exact sequences onthe category Ch(A). First, the suspension functor Σ: Ch(A) → Ch(A) clearlyinduces a suspension functor Σ: K(A) → K(A), which is again an equivalence ofcategories.

A triangle in K(A) is a diagram of the form X → Y → Z → ΣX and amorphism of triangles is a commutative diagram

X //

f

Y //

g

Z //

h

ΣX

Σf

X ′ // Y ′ // Z ′ // ΣX ′.

Page 44: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

44 MORITZ GROTH

Note that we insist that the vertical morphism to the very right is the suspension ofthe morphism to the very left. These notions of course make sense in any categorywith an endofunctor.

We now single out a particular class of triangles on K(A). Recall from (3.18)and Definition 3.19 that associated to any chain map f : X → Y there is the coneCf which is endowed with a chain map cof(f) : Y → Cf . There is also a canonicalmap Cf → ΣX. To construct it, we note that by the definition of the cone CX ofa chain complex there is a natural short exact sequence

0→ Xi→ CX

q→ ΣX → 0.

(In fact, by means of Examples 3.17 this is induced by S0⊗− → I ⊗− → S1⊗−.)The universal property of the defining pushout diagram (3.18) applied to q and thezero map 0: Y → ΣX implies that there is a unique map Cf → ΣX making thefollowing diagram

Xf//

i

Y

cof(f)

0

CX //

q //

Cf

∃!

""

ΣX

commutative. The resulting triangle

X → Y → Cf → ΣX

in K(A) is the standard triangle of f . We say that a triangle in K(A) is distin-guished if it is isomorphic to a standard triangle.

The additional structure on K(A) consisting of the suspension functor and theabove class of distinguished triangles satisfies certain properties as axiomatized bythe notion of a triangulated structure.

4.2. Triangulated categories. In this subsection we take a glimpse at the the-ory of triangulated categories. Here we only include what is strictly necessary inthis motivational section. For more details we refer the reader to the establishedliterature which includes [Nee01, HJR10].

Definition 4.9. Let T be an additive category with a self-equivalence Σ: T → Tand a class of distinguished triangles X → Y → Z → ΣX. The pair consisting of Σand the class of distinguished triangles defines a triangulated structure on T ifthe following four axioms are satisfied.

(T1) For every X ∈ T , the triangle Xid→ X → 0 → ΣX is distinguished. Every

morphism in T occurs as the first morphism in a distinguished triangle and theclass of distinguished triangles is replete, i.e., is closed under isomorphisms.

(T2) If the triangle Xf→ Y

g→ Zh→ ΣX is distinguished then also the rotated

triangle Yg→ Z

h→ ΣX−Σf→ ΣY is distinguished.

Page 45: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 45

(T3) Given two distinguished triangles and a commutative solid arrow diagram

X //

u

Y //

v

Z //

∃w

ΣX

Σu

X ′ // Y ′ // Z ′ // ΣX ′

there exists a dashed arrow w : Z → Z ′ as indicated such that the extendeddiagram commutes.

(T4) For every pair of composable arrows f3 : Xf1→ Y

f2→ Z there is a commutativediagram in which the rows and columns are distinguished triangles:

Xf1 // Y

g1 //

f2

C1h1 //

ΣX

Xf3

// Z

g2

g3// C3

h3

//

ΣX

Σf1

C2

h2

C2

Σg1h2

h2

// ΣY

ΣYΣg1

// ΣC1

A triangulated category is an additive category together with a triangulatedstructure.

A few comments about these axioms are in order.

Remark 4.10. (i) Axiom (T2) is the rotation axiom, axiom (T3) the (map-ping) cone axiom, and axiom (T4) the octahedron axiom. The nameoctahedron axiom is motivated by a different way of drawing the diagram inaxiom (T4); see for example [Wei94, p. 375].

(ii) Triangulated categories were invented independently by Puppe [Pup67] (mo-tivated by Algebraic Topology) and Verdier [Ver96] (motivated by AlgebraicGeometry) in his 1967 thesis, the difference being that Puppe triangulationsare only asked to satisfy axioms (T1)-(T3).

(iii) One way to think of distinguished triangles is that they are some shadows ofcertain derived cokernel constructions on ‘a model in the background’ (thiswill be justified in §4.3). Motivated by this we refer to the third object in adistinguished triangle as a cone (as already reflected in the name of axiom(T3)). The axioms (T1)-(T4) capture some compatibility properties satisfiedby such derived cokernel constructions.

(iv) Using this interpretation of the third objects in distinguished triangles, wecan think of the octahedron axiom as a triangulated category version of thethird Noether isomorphism theorem. In fact, while the Noether isomorphismtheorem says that ‘the quotient of two quotients is again a quotient’ theoctahedron axiom asks that ‘a cone of two cones is again a cone’.

(v) Note that axiom (T3) asks for a weak functoriality of the cone construction:any ‘partial morphism’ of triangles can be extended to an actual morphism.However, we do not ask for the uniqueness of such an extension which is to

Page 46: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

46 MORITZ GROTH

say that the cone is not necessarily a functorial construction T [1] → T . Thislack of functoriality has important implications and we come back to thisin §4.4.

(vi) There is a stronger version of the rotation axiom, asking that a triangle isdistinguished if and only if the rotated diagram is distinguished. It can beshown under the assumption of (T1) and (T3) that this stronger version ofthe rotation axiom is a consequence of the weaker one.

(vii) Also the octahedron axiom is given in a seemingly weaker form than inother references. The observation that the stronger version follows from thisweaker one goes back to [KV87] and this form was recently used by Holmand Jørgenson in their survey article in [HJR10]. For a short discussion ofthis fact and related references see [HJR10, Remark 3.2]. The motivation forus to choose these slightly simpler axioms is that this will later simplify theproof that stable derivators give rise to canonical triangulations.

We suggest the reader who just saw the definition of a triangulated category forthe first time to provide the proofs of the following statements.

Proposition 4.11. Let T be a triangulated category and let W ∈ T . For everydistinguished triangle X → Y → Z → ΣX the represented functor homT (W,−)yields an exact sequence of abelian groups

homT (W,X)→ homT (W,Y )→ homT (W,Z).

Proof. This proof is left as an exercise.

Using the rotation axiom (TR2) in its symmetric form we see that representedfunctors send distinguished triangles to long exact sequences. This is often referredto by saying that represented functors are homological.

Also the proof of the following 5-lemma is suggested as an exercise.

Proposition 4.12. Let T be a triangulated category and let

X //

f

Y //

g

Z //

h

ΣX

Σf

X ′ // Y ′ // Z ′ // ΣX ′

be a morphism of distinguished triangles. If two of the maps f, g, h are isomorphismsthen so is the third one.

Proof. This proof is left as an exercise.

Corollary 4.13. Let X → Y be a morphism in a triangulated categories. Any twodistinguished triangles extending the morphism are isomorphic.

Proof. This is immediate from axiom (T3) and the 5-lemma.

Warning 4.14. In general these isomorphisms are not canonical. We will get backto this in §4.4.

Lemma 4.15. A morphism in a triangulated category is an isomorphism if andonly if any of its cones is zero.

Proof. This proof is left as an exercise.

Solutions to all these exercises and much more can be found in [Nee01, §1].

Page 47: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 47

4.3. Exact morphisms and classical triangulations. We again consider anabelian category A. Recall from the end of §4.1 that we defined a suspensionfunctor Σ: K(A)→ K(A) and a class of distinguished triangles on K(A).

Theorem 4.16. The homotopy category K(A) of an abelian category A togetherwith the equivalence Σ: K(A)→ K(A) and the above class of distinguished trianglesis a triangulated category.

Proof. The category K(A) is additive by Lemma 4.7 and Σ: K(A) → K(A) isan equivalence. Axiom (T1) is immediate while the remaining axioms are moreinvolved. We refer the reader to [HJR10] or [Wei94, Proposition 10.2.4] for moredetails.

In the notation of Corollary 4.6 there clearly is an inclusion HEA ⊆WA and, bythat corollary, the localization functor γ : Ch(A) → D(A) hence factors uniquelyover γ′ : Ch(A)→ K(A), yielding the canonical functor

γ′′ : K(A)→ D(A).

Passing through the homotopy category K(A) as an intermediate step, the derivedcategory is obtained as the localization D(A) = K(A)[W−1]. There are generaltechniques which allow us to conclude that such localizations are again triangulatedcategories and that the localization functor is exact in the following sense; see forexample [Wei94, §10] and [Kra10].

Definition 4.17. Let T , T ′ be triangulated categories. An exact functor T → T ′is a pair (F, σ) consisting of

(i) an additive functor F : T → T ′ and(ii) a natural transformation σ : FΣ→ ΣF

such that for every distinguished triangle X → Y → Zh→ ΣX in T the image

triangle FX → FY → FZσFh→ ΣFX is distinguished in T ′.

Remark 4.18. Note that exactness of a functor of triangulated categories is not aproperty but that it amounts to specifying an additional structure. If we want toemphasize this then we refer to σ : ΣF → FΣ as an exact structure on F .

The following theorem is a consequence of the more general Verdier localizationtheorem; again we refer the reader to the literature.

Theorem 4.19. Let A be an abelian category. The derived category D(A) can beturned into a triangulated category and the localization functor γ′′ : K(A)→ D(A)into an exact functor.

derived functorsbetween derivedcategories are often

exact

4.4. Beyond triangulated categories. The theory of triangulated categories isa large theory and has been successful in many applications in various areas ofpure mathematics for roughly fifty years by now. For a few sample applications invarious areas we refer the reader to the book [HJR10].

Despites all these successes, from the very beginning on (see already the intro-duction to [Hel68]) it was also apparent that the axioms of a triangulated categorycome with certain defects. We include a few comments along these lines.

Page 48: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

48 MORITZ GROTH

(i) One crucial observation is that the cone construction at the level of trian-gulated categories is not functorial. For every morphism in T there is adistinguished triangle (axiom (T1)), hence we can associate a cone object toit. There are the following (closely related) drawbacks to this definition of acone.(a) Cone objects are not characterized by a universal property.(b) We already observed that any two cones of a fixed object are isomorphic,

but only by non-canonical isomorphisms.(c) Similarly, every morphism of morphisms in T yields an induced mor-

phism on cones (axiom (T3)), but, again, this morphism is not uniqueand hence it lacks functoriality.

And in fact, it turns out that if there is a cone functor T [1] → T then thetriangulated category is semisimple ([Ver96, Proposition 1.2.13]). This is tobe seen in contrast to Example 3.31.

(ii) The axioms of a triangulated category ask for the additional structure ofa suspension functor and of distinguished triangles. This datum does notsatisfy any universal property and is hence non-canonical. This is partiallyreflected in the fact that exact morphisms of triangulated categories haveto be defined by means of additional structure as well. But in the typicalexamples the fact that we have an exact morphism of triangulated categoriesreflects the idea that ‘in the background on the models’ there is a morphismof ‘stable homotopy theories’ which has the property of preserving certainfinite homotopy (co)limits.

(iii) As we saw in Theorem 3.42 the cone functor C : D(A[1]) → D(A) is a leftderived cokernel functor, hence satisfies a universal property. There are ofcourse many further derived (co)limit functors which are important in ap-plications. Given a nice abelian category A and a small category A, oftenone can define derived (co)limit functors LcolimA, RlimA : D(AA) → D(A).A crucial observation however is that, say, LcolimA does not factor throughthe underlying diagram functors diaA : D(AA)→ D(A)A of Proposition 3.47,i.e., in general, there is not a dashed arrow

D(AA)Lcolim//

diaA

D(A)

D(A)A@

::

making the diagram commutative.A typical slogan referring to this fact is that ‘diagrams in triangulated

categories do not carry enough information (to canonically determine theirhomotopy (co)limits)’. A different but related slogan is that the underly-ing diagram functor amounts to passing from strict diagrams to homotopycommutative diagrams, and that such homotopy commutative diagrams donot carry enough information. (In fact, one would need homotopy coherentdiagrams to construct associated homotopy (co)limits.)

(iv) Still a related observation is as follows. Given a triangulated category Tand a small category A there is no canonical triangulation on the diagramcategory T A.

Page 49: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 49

(v) A further related observation is that the axioms of a triangulated category,while satisfied in many examples, might miss interesting aspects of homolog-ical algebra or homotopy theory. As mentioned above, the axioms encodesome structure which is a shadow of iterated cofiber seqences and certaincompatibilities. The octahedron axiom (T4) goes one step further and con-siders two composable morphisms together with the various associated dis-tinguished triangles. However, the axioms have nothing to say if one wantsto consider longer strings of composable arrows and all the related distin-guished triangles. Pursuing this further leads to higher triangulations; seefor example [BBD82, Remark 1.1.14] and [Mal05].

The aim of the remainder of this course is to show that stable derivators providean alternative theory which successfully adresses the above issues. In fact, one canthink of a stable derivator as a minimal, purely categorical extension of a derivedcategory to a framework which comes with a well-behaved calculus of derived limitsand derived colimits.

5. Kan extensions

In this section we briefly review some basics concerning the theory of Kan exten-sions from ordinary category theory. Kan extensions are certain universal construc-tions which can be thought of as ‘relative versions of limits and colimits’. The basicidea behind a derivator is to axiomatize key formal properties which are commonto

(i) the calculus of limits, colimits, and Kan extensions in complete and cocom-plete categories,

(ii) the calculus of derived limits, derived colimits, and derived Kan extensionsin derived categories, and

(iii) the calculus of homotopy limits, homotopy colimits, and homotopy Kan ex-tensions in Quillen model categories or complete and cocomplete∞-categories.

While here we establish these formal properties for ordinary categories, in §6 theyare essentially turned into the definition of a derivator and then one establishesthe non-trivial results that abelian categories, Quillen model categories, and ∞-categories yield derivators (see §6 and Appendix B).

In §5.1 we define Kan extensions and observe that they generalize (co)limits.In §5.2 we recall a short discussion of final functors. In §5.3 we note that in completeand cocomplete categories Kan extensions can be constructed pointwisely using themore well-known (co)limits. In §5.4 we collect a few basic results about ordinaryKan extensions which we later extend to arbitrary derivators. We also mentionsome first examples to illustrate Kan extensions.

5.1. Motivation and definition. To motivate the notion of a Kan extension, werecall from §3 (see, in particular, Theorem 3.42) that for every abelian category Athe cone functor C : D(A[1]) → D(A) is the derived cokernel. Given a morphismf : X0 → X1 in Ch(A), the cone Cf can be constructed as the pushout object in

X0f//

i

X1

cof(f)

CX0// Cf.

Page 50: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

50 MORITZ GROTH

Thus, in all detail, the cone of a chain map is obtained by associating a particularspan to it, passing to the pushout square of the span, and then evaluating thepushout square on the final vertex.

(5.1)

X0f// X1

X0f//

i

X1

X0f//

i

X1

cof(f)

CX0 CX0// Cf Cf

Similarly to the case of cokernels, also for cones it is often important to keep trackof the morphism cof(f) : X1 → Cf instead of the chain complex Cf only. And toobtain a detailed description of the construction of cof(f) it suffices to replace thefinal step in (5.1) by the restriction to the vertical morphism on the right.

Note that the fact that this construction of the cone is seemingly complicated isnot a consequence of working in the derived setting. In fact, this is already true forcokernels in ordinary category theory: given a morphism f in a finitely cocompletecategory C admitting a zero object, the cokernel is not the colimit of the morphism(which would be isomorphic to its target) but it is an instance of the more generalnotion of a weighted colimit. Similarly to the derived setting, the cokernel of amorphism f : X0 → X1 in such a category C can be obtained in the following steps.

(5.2)

X0f// X1

X0f//

X1

X0f//

X1

0 0 // cok(k) cok(k)

The only difference between (5.1) and (5.2) is that in the derived setting we replacethe map X0 → 0 by the inclusion X0 → CX0 in order to fix the problem that, ingeneral, pushouts are not exact or homotopy-invariant.

As an upshot, a detailed construction of the cokernel of a morphism passesthrough intermediate steps, the first two being the extension of the morphism toa span and then the extension to a pushout square. The purely categorical notionwhich is in the background of these two intermediate steps is the notion of a Kanextension as we define it next.

To abstract from our specific situation, let us consider the following extensionproblem. Let u : A → B be a functor between small categories and let X : A → Cbe a functor taking values in a not necessarily small category C. The aim is to find‘extensions of X along u’,

(5.3)

A

u

X // C

B.

∃?

>>

The notion of ‘an extension’ can be made precise in different ways, namely by askingthat the diagram (5.3)

(i) commutes on the nose,(ii) commutes up to a specified natural isomorphism, or(iii) commutes up to a specified universal natural transformation.

Page 51: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 51

The first condition is ‘against the philosophical principle from category theory thatone should not ask that functors are equal but only naturally isomorphic’. Whilesatisfied in many specific situations, in general, the second condition turns out tobe too strict. The third condition is the one we would like to axiomatize and, ofcourse, there are two versions of such universal natural transformations, dependingon whether we want the datum to be initial or final.

Definition 5.4. Let u : A → B be a functor between small categories, let C be acategory, and let X : A→ C.

Draw pasting.(i) A left Kan extension of X along u is a functor LKanu(X) : B → C together

with a natural transformation η : X → LKanu(X) u satisfying the followinguniversal property. For every pair (Y : B → C, α : X → Y u) there is aunique β : LKanu(X)→ Y such that α = (βu) · η : X → Y u.

(ii) A right Kan extension of X along u is a functor RKanu(X) : B → Ctogether with a natural transformation ε : RKanu(X) u→ X satisfying thefollowing universal property. For every pair (Y : B → C, α : Y u→ X) thereis a unique β : Y → RKanu(X) such that α = ε · (βu) : Y u→ X.

These notions are dual to each other and in this section we will mostly focuson left Kan extensions. For convenience, we included the following diagramaticdescription of this definition. A left Kan extension (LKanu(X), η) can be depictedby a triangle populated by a natural transformation as in

AX //

u

C

B

LKanu(X)

??

and the universal property of the left Kan extension can be illustrated as followsDo it!

To develop some intuition for this notion let us consider the following specialcase. Recall that we denote the terminal category by 1.

Example 5.5. Let A be a small category and let X : A → C be given. A leftKan extension of X along πA : A → 1 consists of a functor LKanπA(X) : 1 → Cand a universal natural transformation η : X → LKanπA(X) πA. The functorLKanπA(X) : 1 → C amounts to picking an object c ∈ C and the natural transfor-mation is of the form η : X → ∆A(c),

a

∀f

X(a)

X(f)

ηa // c

a′, X(a′),

ηa′

==

thereby specifying a cocone on X. Together with the initiality of this datum, wesee that the left Kan extension (c, η) is an initial cocone on X, which is to say thatc = colimAX and that η is the colimiting cocone.

Thus, we just saw that the notion of a left Kan extension along πA : A → 1reduces to the notion of a colimit. In particular, such Kan extensions exist as soonas the target category is cocomplete. The goal of §5.3 is to show that this is truemore generally and that those more general Kan extensions can be constructed

Page 52: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

52 MORITZ GROTH

pointwisely by certain colimits. Before we get to that, in §5.2 we include a shortdigression on final functors.

5.2. Final functors. Let C be a cocomplete category and let u : A → B. Forevery diagram X : B → C there is a canonical comparison map between colimB Xand colimA(X u), and we say that u is final if this comparison map is always anisomorphism.

We begin by defining the canonical comparison map and for this purpose weconsider colimB X ∈ C together with its colimiting cocone η which has componentsηb : Xb → colimB X, b ∈ B. Precomposition along u yields the functor Xu : A→ C.Applying the same precomposition to the colimiting cocone η we obtain a cocone αon X u given by αa = ηua : (X u)(a) → colimB X, a ∈ A. In fact, for everymorphism f : a→ a′ in A the triangle

X(ua)

X(uf)

ηua // colimB X

X(ua′)

ηua′

99

commutes since η is a cocone. But also the diagram X u : A → C has a colimitcolimA(X u) and α hence factors uniquely through the corresponding colimitingcocone. Thus, there is a canonical morphism

(5.6) colimA(X u)→ colimB X,

which is compatible with the cocones, i.e., such that the diagram

X(ua)

αa=ηua // colimB X

colimA(X u)

∃!

77

commutes for all a ∈ A. (The unlabelled morphisms belong to the colimiting coconeof X u.)

Dually, if C is a complete category then for every X : B → C there is a canonicalmorphism

(5.7) limBX → lim

A(X u)

which is induced from the respective limiting cones.

Definition 5.8. Let u : A→ B be a functor between small categories.

(i) The functor u is final if the canonical map (5.6) is an isomorphism for everycocomplete category C and every diagram X : B → C.

(ii) The functor u is cofinal if the canonical map (5.7) is an isomorphism forevery complete category C and every diagram X : B → C.

Since these notions are dual to each other we mostly focus on final functors.

Proposition 5.9. Right adjoint functors between small categories are final.

Page 53: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 53

Proof. Let C be a cocomplete category and let (u, v) : A B be an adjunctionbetween small categories. It is easy to check that the associated precompositionfunctors define an adjunction

(v∗, u∗) : CA CB .Recall that adjunctions can be composed. In our situation, if we combine the aboveadjunction with the adjunction (colimB ,∆B) : CB C, then we see that colimB v∗is left adjoint to u∗∆B . But since the functor u∗∆B is equal to ∆A, the uniquenessof left adjoints implies that there is a canonical isomorphism colimA

∼= colimB v∗.With a bit more care one checks that this isomorphism agrees with the desiredcanonical map (5.6).

Corollary 5.10. Let A be a small category admitting a final object ∗. Every cate-gory C has colimits of shape A and for X : A→ C there is a natural isomorphism

X(∗) ∼= colimAX.

Proof. This is immediate from Proposition 5.9 since (πA, ∗) : A 1 is an adjunc-tion.

There is a combinatorial criterion which allows us to characterize final functors.To state the result we need the following definition. For later reference, we alsomake the dual definition.

Definition 5.11. Let u : A → B be a functor between small categories and letb ∈ B be an object.

(i) The slice category (b/u) has as objects pairs (a ∈ A, f : b → u(a)) and asmorphisms (a, f)→ (a′, f ′) morphisms a→ a′ in A such that

bf//

f ′

u(a)

u(a′)

commutes.(ii) The slice category (u/b) has as objects pairs (a ∈ A, f : u(a) → b) and as

morphisms (a, f)→ (a′, f ′) morphisms a→ a′ in A such that

u(a)f//

b

u(a′)

f ′

>>

commutes.

Proposition 5.12. Let u : A→ B be a functor between small categories.

(i) The functor u is final if and only if the categories (b/u), b ∈ B, are non-emptyand connected.

(ii) The functor u is cofinal if and only if the categories (u/b), b ∈ B, are non-empty and connected.

To be completely specific, u is final if and only if for each b ∈ B the followingtwo properties are satisfied.

Page 54: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

54 MORITZ GROTH

(i) There is an object in (b/u), i.e., an object a ∈ A and a morphism b → u(a)in B.

(ii) Any two objects in (b/u) can be connected by a finite zigzag of morphisms.

For a proof of Proposition 5.12 we refer the reader, for example, to [KS06, §2.5].Here we content ourselves by showing how this proposition can be used to re-establish Corollary 5.10 and Proposition 5.9 and by relating the proposition to amore classical statement about colimits of diagrams defined on partially orderedsets (aka. posets).

Examples 5.13. (i) Let B be a poset, let A ⊆ B be a subposet, and let i : A→ Bbe the corresponding inclusion functor. The inclusion is final if and only if forevery b ∈ B the category (b/i) is non-empty and connected. Thus, we ask forthe existence of a ∈ A such that b ≤ a. Moreover, any a1, a2 ∈ A such thatb ≤ a1 and b ≤ a2 have to be connected by a finite zigzag of morphisms in(b/i). A typical sufficient condition is that A is directed, i.e., that any twoelements in A have a common successor. Thus, a directed subposet which isunbounded in the larger poset yields a final functor.

(ii) Let B be a small category, let ∗ ∈ B be a final object, and let us considerthe functor ∗ : 1 → B classifying the final object. For every b ∈ B the slicecategory (b/∗) is isomorphic to 1 and is hence non-empty and connected.This reproduces Corollary 5.10.

(iii) Let A,B be small categories and let v : B → A be a right adjoint functor.We want to show that for every a ∈ A the slice category (a/v) is non-emptyand connected. The unit of the adjunction (u, v) : A B yields a mor-phism ηa : a → vu(a), i.e., an object (u(a), ηa) ∈ (a/v), hence showing that(a/v) is non-empty. The explicit description of the adjunction isomorphismin terms of the unit (see (A.2)) actually shows that (u(a), ηa) is an initialobject in (a/v), so that the slice category is, in particular, connected. Thisrecovers Proposition 5.9.

5.3. Pointwise Kan extensions. In this subsection we show that in cocompletecategories left Kan extensions along functors between small categories always exist.Moreover, there is an explicit construction of such left Kan extensions using colimitsover slice categories. Given the importance of these formulas to theory of derivators,we include complete and fairly detailed proofs. And again, duality allows us tomostly focus on left Kan extensions.

For the remainder of this subsection we assume that C is a cocomplete categoryand that u : A → B is a functor between small categories. The defining universalproperty of left Kan extensions (Definition 5.4) suggests the following constructionof them.

Let us consider a diagram X : A → C and let (Y, α) be a datum among whichthe left Kan extension is supposed to be initial, i.e., we have a functor Y : B → Cand a natural transformation α : X → Y u. For every morphism g : a → a′ in Awe obtain the naturality square

X(a)X(g)

//

αa

X(a′)

αa′

Y u(a)Y u(g)

// Y u(a′).

Page 55: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 55

Now, for every object b ∈ B, the value Y (b) has to be compatible with thesenaturality squares in the following sense. Let us assume we can find morphismsf : u(a)→ b and f ′ : u(a′)→ b in B which are compatible with g in the sense thatf = f ′ u(g). For each such datum the above naturality square extends to thefollowing commutative diagram

(5.14)

X(a)X(g)

//

αa

X(a′)

αa′

Y u(a)Y u(g)

//

Y (f)

Y u(a′)

Y (f ′)zz

Y (b)

in C. If we ignore the line in the middle, then we see that Y (b) is part of a coconeon a certain diagram. And the universal property of a left Kan extension suggeststhat we should form universal such cocones by passing to colimits. It turns outthat this indeed works and we now carry out the details.

Note first that in the above heuristics (5.14) the slice category (u/b) from Defi-nition 5.11 came up.

Lemma 5.15. Let u : A→ B be a functor between small categories.

(i) For every b ∈ B there is a projection functor p = pb : (u/b) → A defined onobjects by (a, f : u(a)→ b) 7→ a and on morphisms by g 7→ g.

(ii) For every morphism h : b→ b′ in B there is a functor (u/h) : (u/b)→ (u/b′)defined on objects by (a, f : u(a) → b) 7→ (a, h f : u(a) → b → b′) and onmorphisms by g 7→ g.

(iii) For every morphism h : b→ b′ the diagram

(u/b)(u/h)

//

pb

(u/b′)

pb′zz

A

commutes.

Proof. This is obvious and is left to the reader.

Thus, for every object b ∈ B we can consider the diagram

X pb : (u/b)→ A→ C,

and let us note that in the above heuristics (5.14) we observed that Y (b) is part ofa cocone on precisely this diagram. This suggests that we make the definition

(5.16) L(X)(b) = colim(u/b)X pb, b ∈ B.

Moreover, as a special instance of the canonical morphisms (5.6), for every mor-phism h : b → b′ in B we use the relation pb′ (u/h) = pb from Lemma 5.15 toobtain a canonical morphism

(5.17) L(X)(h) : colim(u/b)X pb → colim(u/b′)X pb′ .

Page 56: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

56 MORITZ GROTH

The uniqueness of these canonical morphisms show that they are functorial, therebyconcluding the definition of the functor L(X) : B → C.

Lemma 5.18. Let C be a cocomplete category, let u : A→ B be a functor betweensmall categories, and let X : A → C. The assignments (5.16) and (5.17) define afunctor L(X) : B → C.

Proof. This is immediate from the above discussion.

In order to conclude the construction of a left Kan extension, we need a naturaltransformation η : X → L(X) u. For a ∈ A the pair (a, id : u(a) → u(a)) definesan object in (u/u(a)) and we let ηa be the corresponding canonical map belongingto the colimiting cocone of L(X)(u(a)) = colim(u/u(a))X pu(a),

(5.19) ηa = ι(a,idu(a)) : Xa → colim(u/u(a))X pu(a).

Given a morphism g : a→ a′ in A, let us consider the following diagram

Xa

ι(a,idu(a))//

X(g)

ι′(a,u(g))

**

colim(u/u(a))X pu(a)

Xa′ι′(a′,id

u(a′))

// colim(u/u(a′))X pu(a′),

in which the respective colimiting cocones are denoted by ι and ι′. The uppertriangle commutes by definition of (5.17) as a particular instance of the canonicalmorphism (5.6), while the lower triangle commutes since ι′ is a cocone. Thus, themorphisms (5.19) define a natural transformation η : X → L(X) u.

Proposition 5.20. Let C be a cocomplete category, let u : A → B be a functorbetween small categories, and let X : A → C. The pair (L(X), η) consisting ofthe functor L(X) : B → C defined by (5.16),(5.17) and the natural transformationη : X → L(X) u defined by (5.19) is a left Kan extension of X along u.

Proof. By Definition 5.4 we have to show that for every Y : B → C the assignment

φ : homCB (L(X), Y )→ homCA(X,Y u) : β 7→ (βu) ηis a bijection. It follows from the explicit construcion (5.16) and (5.19) that thecomponent of φ(β) at an object a ∈ A is given by

φ(β)a = βua ι(a,id) : X(a)→ colimX pua → (Y u)(a).

We next construct the following map in the converse direction

ψ : homCA(X,Y u)→ homCB (L(X), Y ).

To this end, let α : X → Y u and let us fix an object b ∈ B. For every object(a, f : u(a) → b) ∈ (u/b) we form the map Y f αa : Xa → (Y u)(a) → Y (b).By (5.14) these maps define a cocone on X pb : (u/b) → C and there is hence aunique map ψ(α)b : L(X)(b) = colimX pb → Y (b) such that

(5.21)

colimX pb∃! ψ(α)b

// Y (b)

X(a)

ι(a,f)

OO

αa// Y (ua)

Y f

OO

Page 57: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 57

commutes. Here, the morphism ι(a,f) belongs to the colimiting cocone. We leave itto the reader to check that for every morphism h : b→ b′ in B the diagram

colimX pb

L(X)(h)

ψ(α)b// Y (b)

h

colimX pb′ψ(α)b′

// Y (b′)

commutes, i.e., that ψ(α) : L(X)→ Y is a natural transformation.It remains to check that φ and ψ are inverse to each other. Let β : L(X) → Y

and let α = φ(β) = (βu) η. For every b ∈ B and (a, f) ∈ (u/b) we consider thediagram

colimX pbβb // Y (b)

X(a)

ι(a,f)88

ηa// colimX pua

βua

//

L(X)(f)

OO

Y (ua),

Y f

OO

in which the square is a naturality square of β. Since ηa = ιa,idua (see (5.19))and ιa,f belong to the respective colimiting cocones, the definition of L(X)(f)(see (5.17)) as an instance of the canonical map (5.6) implies that also the trianglecommutes. But the uniqueness of the map ψ(α)b in the definition of ψ (see (5.21))implies that βb = ψ(φ(β))b.

Conversely, let α : X → Y u be given and we want to show that (φψ)(α) = α,i.e., that (ψ(α)u) η = α. If we consider a ∈ A and the corresponding object(a, idu(a)) ∈ (u/a), then as a special case of (5.21) we see that the diagram

colimX puaψ(α)ua

// Y (ua)

X(a)

ι(a,idua)

OO

αa// Y (ua)

=

OO

commutes. Using once more the precise form of η (see (5.19)) we conclude thatφψ(α) = α, concluding the proof.

Definition 5.22. Let C be a category and let u : A → B be a functor betweensmall categories. The assignments

u∗ : CB → CA : X 7→ X u, β 7→ βu

define a restriction functor or precomposition functor.

For later reference we summarize our findings concerning the existence and theprecise form of left Kan extensions in the following theorem.

Theorem 5.23. Let C be a cocomplete category and let u : A → B be a functorbetween small categories.

(i) For every diagram X : A→ C a left Kan extension LKanu(X) exists and thisdefines a functor LKanu : CA → CB which is left adjoint to u∗ : CB → CA,

(LKanu, u∗) : CA CB .

Page 58: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

58 MORITZ GROTH

(ii) For every diagram X : A → C and every object b ∈ B there is a canonicalisomorphism

colim(u/b)X pb ∼= LKanu(X)b.

Proof. Associated to u there is the restriction functor u∗ : CB → CA and a leftKan extension of X along u is precisely an initial object in the category (X/u∗).By Proposition 5.20 we know that such an initial object always exists, and fromthis it follows immediately that any choices of such left Kan extensions define afunctor LKanu : CA → CB which is left adjoint to u∗ : CB → CA.

Moreover, any two initial objects are canonically isomorphic, implying that anyleft Kan extension LKanu(X) is canonically isomorphic to the one constructedin (5.16).

The corresponding result for right Kan extensions is dual. Given a functoru : A→ B between small categories and b ∈ B there is a functor

(5.24) q = qb : (b/u)→ A

defined on objects by (a, f : b→ u(a)) 7→ a and on morphisms by g 7→ g.

Theorem 5.25. Let C be a complete category and let u : A → B be a functorbetween small categories.

(i) For every diagram X : A → C a right Kan extension RKanu(X) : B → Cexists and this defines a functor RKanu : CA → CB which is right adjoint tou∗ : CB → CA,

(u∗,RKanu) : CB CA.(ii) For every diagram X : A → C and every object b ∈ B there is a canonical

isomorphismRKanu(X)b ∼= lim

(b/u)X qb.

Proof. This is the dual of Theorem 5.23.

Thus, in sufficiently (co)complete categories, Kan extensions always exist andcan be constructed using pointwise formulas, thereby sort of reducing the moregeneral notion of a Kan extension to the more well-known notion of a limit or acolimit.

Warning 5.26. There are examples of Kan extensions (necessarily in contexts lack-ing certain limits or colimits) which cannot be calculated using pointwise formulas.For a particular example we refer the reader to [?]. Such Kan extensions will notplay a role in this book.

Ref!

Before we take up more interesting examples in §5.4, we revisit Example 5.5.

Example 5.27. Let C be a complete and cocomplete category, let A be a smallcategory, and let πA : A→ 1 be the unique functor.

(i) Under the isomorphism C1 ∼= C the adjunction (LKanπA , π∗A) : CA C1 gets

identified with (colimA,∆A) : CA C.(ii) Under the isomorphism C1 ∼= C the adjunction (π∗A,RKanπA) : C1 CA gets

identified with (∆A, limA) : C CA.

This leads us to think of Kan extensions as ‘relative versions of limits and colim-its’, which are obtained by replacing πA : A→ 1 by more general functors u : A→ Bbetween small categories.

Page 59: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 59

5.4. Basic properties and first examples. In this subsection we illustrate thenotion of Kan extensions by collecting a few examples and also establish the usefulproperty that Kan extensions along fully faithful functors are again fully faithful,thereby justifying the terminology Kan extensions.

Since Kan extensions will be used a lot in the remainder of this book, let usintroduce the following notation.

Notation 5.28. Let C be a complete and cocomplete category and let u : A→ Bbe a functor between small categories. We also denote Kan extension functors by

u! = LKanu : CA → CB and u∗ = RKanu : CA → CB .In the special case that u = b : 1→ B classifies the object b ∈ B we thus write

b! = LKanb : C → CB and b∗ = RKanb : C → CB .

As first examples we already saw that Kan extensions along functors πA : A→ 1reduce to limits and colimits (Example 5.27). The other extreme case is given byKan extensions along functors 1 → B defined on the terminal category, and thisleads to free and cofree diagrams.

Example 5.29. Let B be a small category, let b ∈ B, and let C be a complete andcocomplete category (actually, the existence of products and coproducts suffices).We describe the Kan extension functors b!, b∗ : C → CB along b : 1→ B.

(i) The left Kan extension functor b! : C → CB sends an object X ∈ C to the freediagram b!(X) : B → C generated by X. Given an object b′ ∈ B, the slicecategory (b/b′) is simply the discrete category on homB(b, b′). The pointwiseformula yields the first isomorphism in

b!(X)b′ ∼= colim(b/b′)X pb′ ∼=∐

homB(b,b′)

X.

The second isomorphism is a consequence of (b/b′) being a discrete categoryso that the colimit reduces to a coproduct. The defining universal propertyof the free diagram b!(X) is that there are natural isomorphisms

homCB (b!(X), Y ) ∼= homC(X,Y (b)), Y ∈ CB ,i.e., b! is left adjoint to the evaluation functor b∗ : CB → C.

(ii) The right Kan extension functor b∗ : C → CB sends an object X ∈ C to thecofree diagram b∗(X) : B → C generated by X. In this case there arenatural isomorphisms

b∗(X)b′ ∼= lim(b′/b)

X qb′ ∼=∏

homB(b′,b)

X.

Remark 5.30. Note that this example shows that Kan extensions are, in general,not fully faithful. In other words, assuming that the respective Kan extensionsexist, for a diagram X : A→ C the diagrams

LKanu(X) u,RKanu(X) u : A→ Cobtained by first extending and then restricting back are, in general, not naturallyisomorphic to X. In fact, for free and cofree diagrams we have isomorphisms

b!(X)b ∼=∐

homB(b,b)

X and b∗(X)b ∼=∏

homB(b,b)

X.

Page 60: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

60 MORITZ GROTH

We illustrate the concept of free diagrams by two more specific examples.

Examples 5.31. Let C be a complete and cocomplete category.

(i) Given a quiver Q and a vertex q0 ∈ Q we obtain the free diagram functor(q0)! : C → CQ. Specializing to Mod(R) for a ring R we obtain a functor(q0)! : Mod(R) → Mod(R)Q. The universal property of free diagrams yieldsnatural isomorphisms

homMod(R)Q((q0)!(R), X) ∼= homR(R,X(q0)) ∼= Xq0 ,

i.e., (q0)!(R) corepresents the evaluation functor (q0)∗ : Mod(R)Q → Mod(R).It follows that (q0)!(R) is a projective object in the abelian category Mod(R)Q.Moreover, (q0)!(R) is an indecomposable object since any non-trivial decom-position would by the Yoneda lemma be induced by a non-trivial idempotentendomorphism of q0 ∈ Q which is impossible since Q is a free category.

To specialize this a bit, let us consider a field R = k so that (q0)!(k) is theusual indecomposable projective representation P (q0) corresponding to thevertex q0 ∈ Q. In the case of the linearly oriented A3-quiver (1 → 2 → 3)this yields the indecomposable projective representations

P (1) = (kid→ k

id→ k), P (2) = (0→ kid→ k), and P (3) = (0→ 0→ k).

(ii) Let G be a discrete group, considered as a groupoid with one object, and leti : ∗ → G classify the unique object. The free and the cofree diagram functorsi!, i∗ : C → CG, respectively, send an object X ∈ C to the G-objects

i!(X) ∼=∐g∈G

X and i∗(X) ∼=∏g∈G

X.

The G-actions simply permute summands and factors.Specializing this to C = Mod(R) for a ring R and implicitly using the

equivalence Mod(R)G ' Mod(RG), this recovers the induction and coin-duction functors Mod(R) → Mod(RG), i.e., the respective adjoints of therestriction of scalar functor i∗ : Mod(RG)→ Mod(R).

We mention one more example of Kan extension functors.

Example 5.32. Let C be a complete and cocomplete category, let G be a discretegroup, and let H < G be a subgroup. The Kan extensions along the correspondingfunctor i : H → G give rise to adjunctions

(i!, i∗) : CH CG and (i∗, i∗) : CG → CH .

The functors i! and i∗ are, respectively, referred to as induction and coinduction,and they make perfectly well sense for not necessarily injective group homomor-phisms. The reader is invited to use the pointwise formulas to obtain a more precisedescription of these functors and to specialize to the case of modules over a ring.

The following property of Kan extensions is important, and this result justifiesthe terminology Kan extensions.

Proposition 5.33. Let C be a complete and a cocomplete category and let u : A→B be a functor between small categories.

(i) The functor u! : CA → CB is fully faithful and induces an equivalence onto thefull subcategory of CB spanned by all Y such that the counit ε : (u!u∗)Y → Yis an isomorphism.

Page 61: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 61

(ii) The functor u∗ : CA → CB is fully faithful and induces an equivalence onto thefull subcategory of CB spanned by all Y such that the unit η : Y → (u∗ u∗)Yis an isomorphism.

Proof. By duality it is enough to take care of the first statement, and it followsfrom Lemma A.4 that u! is fully faithful if and only if the unit η : 1 → u∗u! isa natural isomorphism. For X : A → C and a ∈ A it follows from the pointwiseformula (Theorem 5.23) that u!(X)u(a)

∼= colim(u/u(a))X pu(a). Note that thefully faithfulness of u implies that (a, idu(a) : u(a) → u(a)) is a terminal object in(u/u(a)). Hence by Corollary 5.10 the functor (a, idu(a)) : 1 → (u/u(a)) is final,and, as a particular instance of (5.6), there is a canonical isomorphism

X(a) = X pu(a)(a, 1)→ colim(u/u(a))X pu(a).

Note that this canonical morphism is simply the map ι(a,1) belonging to the colim-iting cocone of colim(u/u(a))X pu(a). But this map agrees with the component ηaof the adjunction unit η : 1 → u∗u! by the above explicit construction of left Kanextensions (see (5.19)). Since this is the case for all a ∈ A, we conclude that ηindeed is a natural isomorphism, thereby establishing that u! is fully faithful. Thedescription of the essential image of u! is a special case of Lemma A.4.

Given a diagramX : A→ C in a cocomplete category, let us recall that the colimiton X is an initial cocone on X, i.e., it consists of the colimit object colimAX ∈ Ctogether with the colimiting cocone ιa : Xa → colimAX, a ∈ A. The colimitingcocone is a natural transformation X → ∆A colimAX, and as noted in Lemma 3.34the natural domain for natural transformations is the cylinder A× [1]. If the targetof the transformation is a constant diagram, then the tranformation factors overthe cocone in the sense of the following definition, and dually.

Definition 5.34. Let A be a small category.

(i) The cocone AB of A is the small category obtained by freely adjoining anew terminal object ∞ ∈ AB.

(ii) The cone AC on A is the small category obtained by freely adjoining a newinitial object −∞ ∈ AB.

Let us be more specific about the construction of the cocone AB. The objectsin A are the objects in A and a unique new object ∞ ∈ AB. The set of morphismsis defined by

homAB(x, y) =

homA(x, y) , x, y ∈ A,∗ , y =∞,∅ , x =∞, y ∈ A.

And the composition law in AB is unique determined by the fact that there is afully faithful inclusion functor A → AB : x 7→ x. The construction of AC is dualand there is also a fully faithful functor A→ AC.

Examples 5.35. Let C be a complete and cocomplete category and let A be a smallcategory.

(i) Since the inclusion functor i : A → AB is fully faithful, we obtain by Propo-

sition 5.33 the left Kan extension functor i! : CA → CAB

is fully faithful. Onecan check that the essential image consists precisely of the colimiting cocones.In particular, for X : A → C, the diagram i!(X) : AB → C encodes colimAXtogether with the colimiting cocone.

Page 62: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

62 MORITZ GROTH

(ii) Dually, since the inclusion functor i : A → AC is fully faithful, the right

Kan extension functor i∗ : CA → CACis fully faithful. The essential image

consists precisely of the limiting cones. Starting with X : A→ C, the diagrami∗(X) : AC → C encodes limAX together with the limiting cone.

(iii) We denote by = [1]× [1] the commutative square, i.e., the category

(0, 0) //

=

(1, 0)

(0, 1) // (1, 1)

and by ip : p→ and iy : y→ the full subcategories obtained by removingthe final object (1, 1) and the initial object (0, 0), respectively. Note that is the cocone on p and the cone on y. Hence, specializing the previous twopoints, the Kan extension functors

(ip)! : Cp → C and (iy)∗ : Cy → C

are fully faithful and the essential images consist precisely of the pushoutsquares and the pullback squares, respectively.

To conclude this section, let us briefly come back to the construction of kernelsand cokernels in terms of Kan extensions; see again the discussion of (5.2) and itsdual. With Examples 5.35 in mind it only remains to take care of the passage froma morphism to the span in (5.2). To this end we consider the functor

j : [1]→ p and k : [1]→yclassifying the horizontal morphism (0, 0)→ (1, 0) and (0, 1)→ (1, 1), respectively.

Lemma 5.36. Let C be a category admitting a zero object and let X0 → X1 be amorphism in C, thereby determing a functor X : [1]→ C.

(i) The right Kan extension (i0)∗(X) : p→ C is naturally isomorphic to the dia-gram on the left in the diagram

X0//

X1 0

0, X0// X1.

(ii) The left Kan extension (i1)!(X) : y → C is naturally isomorphic to the dia-gram on the right in the above diagram.

Proof. By duality it is enough to take care of the first statement. The functori0 : [1] → p is fully faithful. Hence also (i0)∗ : C[1] → Cp is fully faithful. In partic-ular, for X : [1] → C, the counit ε : (i0)∗(i0)∗X → X is an isomorphism. Thus, itremains to show that the value of (i0)∗(X) at the lower left corner (0, 1) ∈ p is azero object. By the pointwise formula for right Kan extensions (Theorem 5.25) wecalculate

(i0)∗(X)(0,1)∼= lim

((0,1)/i0)X q(0,1)

= lim∅X q(0,1)

∼= 0.

Page 63: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 63

In fact, it is obvious that the slice category ((0, 1)/i0) is empty, so that the limitreduces to a zero object.

Thus, the intended passage from a morphism to the span in (5.2) amounts toforming the right Kan extension along i0 : [1] → p. Combining this with Exam-ples 5.35 we see that for a category C with a zero object the Kan extension functors

C[1] (i0)∗→ Cp i!→ C

amount to passing from a morphism f : X0 → X1 to the pushout square

X0f//

X1

0 // cok(f)

containing the cokernel.While it is arguable how much is gained by this description of the cokernel if

one only cares about ordinary cokernels, the point is that the same descriptionyields a construction of functorial cones at the level of derived categories of abeliancategories and homotopy categories of Quillen model categories or ∞-categories.This description only relies on having a suitable framework for a ‘calculus of abstractKan extensions’, and precisely such a framework is axiomatized by the notion ofa derivator. In §6 we define derivators and in §§7-10 we illustrate the notion byindicating a few constructions available in such a derivator.

6. Basics on derivatorsTODO: proof-read and

expand from here toend of file.

In §§6.1-6.2 we define derivators and introduce some basic terminology. To put itas a slogan, derivators can be thought of as a minimal, purely categorical extensionof a derived category to a framework which comes with a powerful calculus ofderived limits and, more generally, derived Kan extensions. Similarly, the derivatorof spaces is such an extension of the homotopy category of spaces to a frameworkwith a calculus of homotopy limits and homotopy Kan extensions.

In §6.3 we mention quite a few examples of derivators, but postpone the proofsthat these actually are examples to the appendix. Finally, in §6.4 we emphasizethe difference between categorical limits on the one hand side and derived limits orhomotopy limits on the other side.

6.1. Prederivators. Let A be an abelian category and let D(A) be the derivedcategory. By Theorem 3.42 there is a functorial cone C : D(A[1]) → D(A) which,in general, does not factor through D(A)[1]. More generally, as already emphasizedin §4.4, it is important to distinguish between the following two types of categories.

(i) Derived categories of diagram categories have as objects strict diagrams andthese diagrams allow for many important constructions.

(ii) Diagram categories in derived categories are less well behaved in that theirobjects do not carry enough information and many constructions can not beperformed at that level anymore.

These categories are related by functors diaA : D(AA)→ D(A)A, which, in general,are far from being equivalences; see Proposition 3.47 and Warning 3.48. Sincethe categories D(AA) can not be reconstructed from D(A), this suggests that the

Page 64: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

64 MORITZ GROTH

passage to derived categories should be refined by keeping track of the categoriesD(AA).

The formalization is as follows. Let Cat denote the 2-category of small categories,functors, and natural transformations. Similarly, let CAT be the 2-category of (notnecessarily small) categories, functors, and natural transformations.

Definition 6.1. A prederivator D is a strict 2-functor D : Catop → CAT .

The 2-category Catop is obtained from Cat by reversing the direction of the functorswhile the direction of the natural transformations is unchanged. Thus, given aprederivator D , for every small category A we obtain a category D(A). Associatedto a functor u : A → B in Cat , there is an induced precomposition functor orrestriction functor

u∗ : D(B)→ D(A).

To emphasize the dependence on D , we sometimes also write D(u) or u∗D . Finally,given functors u, v : A → B in Cat and a natural transformation α : u → v, weobtain an induced natural transformation α∗ : u∗ → v∗, as depicted in the diagram

A

u''

v

77 B, D(B)

u∗++

v∗33 D(A).

This datum is compatible with compositions and identities in a strict sense, i.e., wehave equalities of the respective expressions and not only coherent natural isomor-phisms between them.

Recall that 1 denotes the terminal category. Given a prederivator D , we refer toD(1) as the underlying category of D . As we will see in this book, the structureof a derivator D enhances the underlying category D(1) to a more flexible notion.

Examples 6.2. (i) Every category C gives rise to a represented prederivatory(C) defined by

y(C)(A) := CA, A ∈ Cat .

Given a functor u : A→ B in Cat , we define y(C)(u) to be the precompositionfunctor u∗ : CB → CA : X 7→ X u. Finally, given two functors u, v : A → Band a natural transformation α : u→ v the reader easily checks that

α∗X = X α : X u→ X v, X ∈ CB ,

defines a natural transformation α∗ : u∗ → v∗, concluding the definition ofthe 2-functor y(C). The underlying category of y(C) is canonically isomorphicto C itself,

y(C)(1) ∼= C.This example is interesting when it comes to generalizing notions from

ordinary category theory to derivator theory; in fact, the notions should bedefined for derivators such that in the represented case they reduce to theclassical ones.

(ii) Let A be an abelian category, let Ch(A) be the category of unboundedchain complexes, and let W be the class of quasi-isomorphisms in Ch(A).Given a small category A, we again denote by WA the levelwise quasi-isomorphisms, i.e., those natural transformations X → Y : A → Ch(A) such

Page 65: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 65

that all components are quasi-isomorphisms. The homotopy prederivatorDA : Catop → CAT is defined by the localizations

DA(A) := Ch(A)A[(WA)−1], A ∈ Cat .

Since the restriction functors u∗ : Ch(A)B → Ch(A)A clearly preserve level-wise quasi-isomorphisms, there are canonical functors u∗ : DA(B) → DA(A)compatible with the localization functors (compare to Lemma 3.45). Usingthat these localizations are 2-localizations (see Proposition 3.35), the readereasily concludes the definition of the 2-functor DA : Catop → CAT . The un-derlying category of this prederivator is canonically isomorphic to the derivedcategory D(A). More generally, it follows from Lemma 3.10 that there areisomorphisms

DA(A) ∼= D(AA), A ∈ Cat .

(iii) Let Top denote the category of topological spaces and continuous maps. Werecall that a continuous map f : X → Y is a weak homotopy equivalence iffor all x0 ∈ X and all n ≥ 0 the induced map f∗ : πn(X,x0) → πn(Y, f(x0))is a bijection. For every small category A we denote by WA the levelwiseweak homotopy equivalences in TopA. The (homotopy) prederivator oftopological spaces is defined by

HoTop(A) := TopA[(WA)−1], A ∈ Cat .

As in the previous example, the reader easily checks (using a variant of Propo-sition 3.35) that this extends to a 2-functor HoTop : Catop → CAT . The un-derlying category of HoTop is canonically isomorphic to the usual homotopycategory of spaces,

HoTop(1) ∼= Ho(Top).

(iv) We conclude by a broad class of examples induced by Quillen model cate-gories; see the original [Qui67], the monographs [Hov99, Hir03] or the in-troductory [DS95]. Let M be a Quillen model category with weak equiv-alences W . For every small category A, the category MA comes with theclass WA of levelwise weak equivalences. Although WA does not necessarilybelong to a Quillen model structure on MA, we can define the homotopyprederivator HoM of M by

HoM(A) := (MA)[(WA)−1],

and it can be shown that these categories are locally small. The underlyingcategory is canonically isomorphic to the usual homotopy category,

HoM(1) ∼= Ho(M) =M[W−1].

These three final examples are the main motivation for the theory of derivators.

Remark 6.3. (i) Our convention for (pre)derivators (which agrees with the oneof Heller [Hel88] and Franke [Fra96]) is based on the idea that we want tomodel diagrams (covariant functors) in a fixed abstract homotopy theory.This results in the use of Catop as domain of definition for prederivators.There is an alternative but isomorphic approach (as used by Cisinski [Cis03,Cis08], Grothendieck [Gro], Maltsiniotis [Mal01, Mal07]) based on presheaves(contravariant functors) in which case Catop is replaced by Catcoop, obtainedby also reversing the orientations of the natural transformations. Althoughthe resulting theories are equivalent, many statements differ slightly since the

Page 66: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

66 MORITZ GROTH

directions of various natural transformations in each convention are reversedwith respect to the other.

(ii) Often (for example under certain finiteness conditions) it is convenient to havesome flexibility with respect to the domain of definition of (pre)derivators byrestricting the class of ‘admissible shapes’. There is the notion of a categoryof diagrams Dia, a full sub-2-category of Cat enjoying certain closure proper-ties, and the related definition of a (pre)derivator of type Dia as a 2-functorDiaop → CAT . In this introductory book we will not get into this.

We establish some basic terminology related to prederivators D . Motivated byExamples 6.2 we refer to objects of D(A) as coherent, A-shaped diagrams in D .We will often write X ∈ D if there is a small category A such that X ∈ D(A).

Warning 6.4. We want to warn the reader that this is just terminology; for anabstract prederivator D , an object X ∈ D(A) is not a diagram in whatever sense;see also Warning 6.7. However, every such object induces an ordinary diagram ofshape A as we discuss next.

We keep using implicitly the canonical isomorphisms A ∼= A1 and hence writea : 1 → A for the functor classifying a ∈ A. As a special case of a precompositionfunctor we obtain an evaluation functor a∗ : D(A) → D(1). Given a morphismg : X → Y in D(A) we will write ga : Xa → Ya for the induced map in the underlyingcategory D(1).

Similarly, every morphism f : a→ b in A yields a natural transformation of thecorresponding functors 1→ A, and for every prederivator we hence obtain

1

a''

b

77 A, D(A)

a∗**

b∗44 D(1) .

Evaluated at X ∈ D(A) this defines a map f∗ : Xa → Xb. We summarize thefunctoriality of this construction in the following lemma.

Lemma 6.5. Let D be a prederivator and let g : X → Y be a morphism in D(A).

(i) The assignment diaA(X) : A → D(1) : a 7→ Xa, f 7→ f∗, defines a functordiaA(X) : A→ D(1), the underlying (incoherent) diagram of X ∈ D(A).

(ii) There is a natural transformation diaA(g) : diaA(X) → diaA(Y ) : A → D(1)with components diaA(g)a = ga : Xa → Ya, a ∈ A.

(iii) The assignments X 7→ diaA(X) and g 7→ diaA(g) define a functor, the un-derlying diagram functor

diaA : D(A)→ D(1)A.

Proof. The proof is left as an exercise. We suggest the reader to really do thisexercise in order to get used to the concept of 2-functoriality.

This construction is an abstract version of Proposition 3.47.

Examples 6.6. We take up again Examples 6.2 and describe the associated under-lying diagram functors.

(i) For represented prederivators y(C) the functors diaA : CA → (C1)A reduce tothe canonical isomorphisms CA ∼= (C1)A.

Page 67: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 67

(ii) In the case of the homotopy prederivator DA of an abelian category, thefunctor diaA : DA(A)→ DA(1)A corresponds under DA(A) ∼= D(AA) to thefunctor diaA : D(AA)→ D(A)A of Proposition 3.47. In general, this functoris not an equivalence of categories.

(iii) For the (homotopy) prederivator of spaces we obtain the forgetful functor

diaA : Ho(TopA)→ Ho(Top)A.(iv) Similarly, if (M,W ) comes from a Quillen model category, then the functor

diaA : HoM(A)→HoM(1)A gets identified withMA[(WA)−1]→ Ho(M)A.If we use the description of Ho(M) as the category of cofibrant and fibrantobjects and homotopy classes of morphisms, then this example makes precisethat diaA sends a strict diagram to a homotopy commutative diagram.

Warning 6.7. Given a coherent diagram X ∈ D(A) we will often draw it as anordinary diagram which is to say that we draw diaA(X). However, it is important tonote that, in general, X is not determined by diaA(X), even not up to isomorphism.More specifically, the functor diaA : D(A) → D(1)A is usually not an equivalence,and one has hence to be careful and always distinguish between coherent diagramsand incoherent ones. The main point of the theory of derivators is that manyimportant constructions which are available at the level of coherent diagrams cannot be performed anymore at the level of underlying incoherent diagrams. To putit more frankly, this suggests that the functors diaA should not be applied.

Prederivators allow us to describe systems of categories of diagrams and associ-ated restriction functors and similarly for derived categories or homotopy categoriesof such diagrams. However, to be able to perform more interesting constructionswe have to impose additional axioms, leading to the notion of a derivator.

6.2. Derivators. The basic idea behind derivators is to encode collections of de-rived categories or homotopy categories together with a well-behaved calculus ofhomotopy Kan extensions. For (pre)derivators we follow the established terminol-ogy for ∞-categories and simply speak of Kan extensions as opposed to homotopyKan extensions (and similarly for related notions). This does not result in a riskof ambiguity since in the context of an abstract (pre)derivator the concept ‘Kanextension’ is meaningless.

Definition 6.8. Let D be a prederivator and let u : A→ B be a functor.

(i) The prederivator D admits left Kan extensions along u if the restrictionfunctor u∗ : D(B)→ D(A) has a left adjoint u! : D(A)→ D(B),

(u!, u∗) : D(A) D(B).

(ii) The prederivator D admits right Kan extensions along u if the restrictionfunctor u∗ : D(B)→ D(A) has a right adjoint u∗ : D(A)→ D(B),

(u∗, u∗) : D(B) D(A).

Motivated by Example 5.27, in the case of left Kan extensions along functorsπA : A → 1 to the terminal category we also speak of colimits of shape A andwrite (πA)! = colimA. Similary, right Kan extensions along πA will be referredto as limits of shape A and will occasionally be denoted by (πA)∗ = limA .Sometimes we want to emphasize notationally with respect to which prederivatorcertain functors are considered. In that case, we will also write

u∗D , uD! , colimD

A , uD∗ , and limD

A .

Page 68: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

68 MORITZ GROTH

(5.24)In classical category theory, a key formal property of Kan extensions is that in

sufficiently (co)complete categories they can be calculated pointwise; see §??. Let Cbe a complete and cocomplete category and let X : A→ C be a diagram. For everyu : A→ B in Cat the left Kan extension u!(X) : B → C and the right Kan extensionu∗(X) : B → C exist (??). Moreover, by ?? they can be calculated pointwise in thatfor every b ∈ B certain canonical natural transformations

colim(u/b)X p→ u!(X)b and u∗(X)b → lim(b/u)X q

are isomorphisms. Here, (u/b) and (b/u) are slice categories and p : (u/b)→ A andq : (b/u)→ A are the corresponding canonical projections functors.

In the represented case one can use these pointwise (co)limit expressions toconstruct a diagram of shape B which will be a Kan extension. This is impossiblefor an abstract prederivator D since objects in D(B) are just abstract objects,in particular, not ordinary diagrams. For a derivator we nevertheless ask thatKan extensions exist and can be calculated pointwise. To make this precise, let usconsider the slice squares

(6.9)

(u/b)p

//

π(u/b)

A

u

1b

// B

and

(b/u)q

//

π(b/u)

A

u

1b

// B.

AI

Let us focus on the square on the left, and consider an object (a, f : u(a) → b) in(u/b), and trace it through the diagram. Passing through the upper right cornerwe obtain u(a) while the other path yields b. Thus, in general the square does notcommute but we claim that a canonical natural transformation lives in that square,and dually.

Lemma 6.10. Let u : A→ B be a functor and let b ∈ B.

(i) The assignment (a, f : u(a)→ b) 7→ f defines a transformation u p→ b π.(ii) The assignment (a, f : b→ u(a)) 7→ f defines a transformation b π → u q.

Proof. This proof is left as an exercise.

Now, let us assume that the prederivator D admits the necessary Kan exten-sions. Associated to the slice squares (6.9) we obtain the following canonical mate-transformations

colim(u/b) p∗ = π!p

∗ η→ π!p∗u∗u! → π!π

∗b∗u!ε→ b∗u! and(6.11)

b∗u∗η→ π∗π

∗b∗u∗ → π∗q∗u∗u∗

ε→ π∗q∗ = lim(b/u)q

∗,(6.12)

where the transformations denoted η and ε are suitable adjunction units and counits,respectively. The transformation (6.11) is hence defined as the following pasting

D(1)

ε

D(u/b)π!oo

D(A)p∗oo

η

D(1)

π∗

OO

id

YY

D(B)b∗

oo

u∗

OO

D(A),u!

oo

idjj

Page 69: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 69

and similarly for (6.12). (We will say a bit more about the passage to canonicalmates in §7.1.) It turns out that the good way to express that Kan extensions in Dare pointwise is by asking these canonical mates to be isomorphisms.

Definition 6.13. A prederivator D is a derivator if it has the following properties.

(Der1) D : Catop → CAT takes coproducts to products, i.e., the canonical func-tor D(

∐i∈I Ai) →

∏i∈I D(Ai) is an equivalence. In particular, D(∅) is

equivalent to the terminal category.(Der2) For any A ∈ Cat , a morphism f : X → Y in D(A) is an isomorphism if

and only if each fa : Xa → Ya is an isomorphism in D(1).(Der3) Each restriction functor u∗ : D(B)→ D(A) has both a left adjoint u! and

a right adjoint u∗.(Der4) For any functor u : A → B and any object b ∈ B, the canonical mate-

transformations (6.11) and (6.12) associated to (6.9),

colim(u/b) p∗ ∼−→ b∗u! and b∗u∗

∼−→ lim(b/u)q∗,

are isomorphisms.

A few comments about these axioms are in order.

Remark 6.14. (i) Axiom (Der1) simply says that a coherent diagram on a dis-joint union is canonically determined by its restrictions to the summands.Axiom (Der2) encodes the idea that a transformation of two coherent dia-grams of the same shape is an isomorphism if and only if all componentsare. This is an abstraction of the fact that a natural transformation is invert-ible if and only if each component is. The remaining two axioms encode a‘(homotopical) completeness and cocompleteness property’ together with thepointwise formulas for Kan extensions.

(ii) The formalism behind axiom (Der4) (based on the calculus of canonicalmates) is arguably the key tool in the theory of derivators and will be studiedin more detail in §7. Although a bit heavy on a first view, this formalismturns out to be a very convenient tool allowing for rather mechanical proofs.Often it allows us to show that certain ‘obvious maps which clearly are iso-morphisms’ first of all exist and second turn out to be isomorphisms.

(iii) Note that axioms (Der1)-(Der4) are asking for properties while the only ac-tual structure is the underlying prederivator. This is in contrast to moreclassical approaches including triangulated categories in which case some non-canonical structure is part of the notion. We will get back to this later.

(iv) The axioms of a derivator seem to simply capture ‘rather obvious’ formalproperties of the calculus of Kan extensions. In a way, it is surprising thatthese axioms encode anything essential about homotopy theory. Nevertheless,it turns out that this is the case.

(v) Although already mentioned in Remark 6.3 we want to reemphasize thatthere is some flexibility with respect to the allowable shapes in the definitionof a derivator. By the very definition a derivator encodes the idea of having acomplete and cocomplete homotopy theory (enjoying some nice properties).Under certain finiteness assumptions (like in K-theoretic contexts) one wantsto restrict to finite shapes only and this can be done by means of categoriesof diagrams. Related to this, there are also variants of derivators satisfyingonly half of (Der3) and half of (Der4), leading to left derivators and rightderivators. We will not get into these variants in this introductory book.

Page 70: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

70 MORITZ GROTH

Related to Warning 6.7 we remark the following.

Warning 6.15. Note that we defined a derivator to be a prederivator which ishomotopically cocomplete and homotopically complete in that all restriction functorsu∗ : D(B) → D(A) have left adjoints and right adjoints. It is important to beaware of the fact that, in general, this does not imply that the categories D(A)are complete or cocomplete in the categorical sense. Thus, in the context of anabstract derivator, in general, ordinary categorical Kan extensions and categorical(co)limits do not exist.

We will expand on this warning in §6.4.

6.3. Examples of derivators. In this subsection we collect a few examples ofderivators (see again Examples 6.2). Detailed proofs that we actually obtain deriva-tors are a bit more involved and we suggest the reader to treat them as black boxeson a first reading. For convenience, in the appendix we nevertheless include detailedproofs at least in the case of represented derivators and in the case of homotopyderivators of combinatorial model categories.

We begin with the examples originating from ordinary category theory.

Proposition 6.16. Let C be a category. The represented prederivator y(C) is aderivator if and only if C is complete and cocomplete.

As already mentioned, this example is convenient when it comes to generalizingnotions from ordinary category theory to derivator theory.

We next turn to examples from homological algebra and homotopical algebra.It turns out that various different approaches to homotopical algebra forget toderivators, i.e., that the typical passages to homotopy categories factor throughderivators. Here we will focus on Quillen model categories. To begin with, let usrecall that a Quillen model category is combinatorial if the model structure is cofi-brantly generated and if the underlying category is locally presentable. We do notinclude a discussion of locally presentable categories here and instead only mentionthat chain complexes over a ring (more generally, chain complexes in Grothendieckabelian categories) and simplicial sets yield examples of locally presentable cate-gories. For an introduction to locally presentable categories we refer to [Gro10, §3.1]while detailed accounts can be found in [GU71, MP89, AR94] and [Bor94a, Bor94b].

Theorem 6.17. Let (M,W ) be a combinatorial model category. The homotopyprederivator

HoM : Catop → CAT : A 7→ MA[(WA)−1]

is a derivator, the homotopy derivator of M.

A proof of this result [Gro13] is included in §B.2.

Remark 6.18. Renaudin

Before we collect more specific examples, let us recall the following definition.

Definition 6.19. Let A be an abelian category.

(i) A generator is an object G ∈ A such that f : X → Y in A is an isomorphismif and only if f∗ : homA(G,X)→ homA(G, Y ) is a bijection.

(ii) The category A is Grothendieck abelian if it is cocomplete, if filteredcolimits in A are exact, and if it has a generator.

Page 71: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 71

It can be shown that Grothendieck abelian categories are also complete, see forexample [KS06, §8.3]. To illustrate this notion we include the following examples.

Examples 6.20. (i) The category of R-modules, R a ring, is a Grothendieckabelian category.

(ii) The category of quasi-coherent OX -modules on an arbitrary scheme X is aGrothendieck abelian category.

(iii) For every Grothendieck abelian category A the category Ch(A) of chain com-plexes in A is again a Grothendieck abelian category.

A proof of the following result can be found in [Hov01].

Theorem 6.21. Let A be a Grothendieck abelian category. The category Ch(A)admits a combinatorial model structure with the class of quasi-isomorphisms asclass of weak equivalences.

The results stated so far already allow us to obtain interesting examples of deriva-tors arising in homological algebra, algebraic geometry, and topology (see Exam-ples 6.24). To indicate that there are many additional examples of derivators, westate the following two theorems. The first result is due to Cisinski [Cis03] and itis a rather deep result.

Theorem 6.22. Let (M,W ) be a model category. The homotopy prederivator

HoM : Catop → CAT : A 7→ MA[(WA)−1]

is a derivator, the homotopy derivator of M.

Similarly, ∞-categories give rise to derivators. A sketch proof of the followingresult can be found in [GPS14]. Recall that the nerve NA of a small category Ais the simplicial set which in simplicial degree n is given by the set of strings of ncomposable arrows a0 → a1 → . . .→ an.

Theorem 6.23. Let C be a complete and cocomplete∞-category. There is a deriva-tor

HoC : Catop → CAT : A 7→ Ho(CNA),

the homotopy derivator of the ∞-category C.

Thus, derivators encode aspects of the calculus of derived Kan extensions in niceabelian categories and, more generally, of the calculus of homotopy Kan extensionsin model categories and∞-categories. We now include a few more specific examples.

Examples 6.24. (i) For every ring R the category of R-modules is Grothendieckabelian and

DR : Catop → CAT : A 7→ Ch(R)A[(WA)−1]

hence defines a derivator, the derivator of the ring R. In particular, wehave the derivator DZ of the integers and the derivator Dk of a field k. Thesederivators enhance the more classical derived categories.

Note that DR encodes a lot of information concerning the ring R as weillustrate next.

finitely many objects(a) Given a discrete group G there is an isomorphism DR(G) ∼= D(RG),

i.e., DR knows about derived categories of all group algebras RG. More-over, as special cases of (derived) Kan extension functors, DR encodesgroup cohomology and group homology as well as their versions for chaincomplexes (see §2.3).

Page 72: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

72 MORITZ GROTH

(b) Given a (finite) quiver Q there is an isomorphism DR(Q) ∼= D(RQ),which is to say that the derived categories of path algebras RQ of anyquiver Q are encoded by DR.

(c) As a further variant, DR also encodes derived categories of incidencealgebras RP of (finite) posets P as well as derived categories of categoryalgebras RA (of finite categories A).

(ii) Given a scheme X, let Ch(X) be the category of chain complexes of quasi-coherent OX -modules on X. For every small category A we denote by WA

the levelwise quasi-isomorphisms in Ch(X)A. The prederivator

DX : Catop → CAT : A 7→ Ch(X)A[(WA)−1]

is a derivator, the derivator of the scheme X. The underlying categoryDX(1) is isomorphic to the more classical derived category D(X).

(iii) More generally, for every Grothendieck abelian category A there is the deriva-tor

DA : Catop → CAT : A 7→ Ch(A)A[(WA)−1],

the homotopy derivator of the Grothendieck abelian category A.(iv) The category Top of topological spaces can be endowed with the Serre model

structure such that the weak equivalences are the weak homotopy equiva-lences. Denoting by WA the levelwise weak homotopy equivalences in TopA,there is a derivator

HoTop : Catop → CAT : A 7→ TopA[(WA)−1],

the (homotopy) derivator of topological spaces.(v) Let us recall the notion of a spectrum in the sense of topology. A spec-

trum X consists of pointed topological spaces Xn, n ≥ 0, and pointed mapsΣXn → Xn+1. A morphism f : X → Y of spectra is a collection of pointedmaps fn : Xn → Yn compatible with the structure maps. The category Spof spectra can be endowed with the stable model structure, having the prop-erty that the homotopy category Ho(Sp) is the classical stable homotopycategory SHC (see [Vog70] and [Ada74, Part III]). We denote the associatedhomotopy derivator HoSp by

(6.25) Sp : Catop → CAT : A 7→ SpA[(WA)−1]

and refer to it as the derivator of spectra.

There are many additional examples and we will mention some of them in ourshort outlook on abstract representation theory.

Recall the duality principle from ordinary category theory which allows one often,say, to formally deduce results concerning colimits from similar results concerninglimits. This duality principle is based upon the passage to opposite categories. Inorder to obtain a similar principle we include the following example.

Example 6.26. Given a prederivator D , the opposite prederivator Dop is definedby setting

Dop(A) = D(Aop)op.

The following result formalizes the self-duality of the axioms of a derivator. Itleads to the important duality principle. In more detail, many statements in thetheory of derivators have dual versions and by the proposition it will be sufficientto state and prove only one of them.

Page 73: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 73

Proposition 6.27. A prederivator D is a derivator if and only if the opposite Dop

is a derivator.

Proof. The proof is left as an exercise. As a hint we mention that for u : A → Bthe left Kan extension functor uDop

! with respect to Dop is closely related to theright Kan extension functor (uop)D

∗ with respect to D .

6.4. Limits versus homotopy limits. We expand a bit on Warning 6.15 andconsider a derivator D and a small category A, asking ourselves whether the cat-egory D(A) admits categorical (co)limits of shape B. In this subsection only, weuse the expressions categorical (co)limits and categorical Kan extensions in orderto refer to the usual constructions from category theory and the terms homotopy(co)limits and homotopy Kan extensions to refer to adjoints to restriction functorsu∗ : D(D)→ D(C) in derivators D .

By definition of categorical (co)limits these would be adjoint functors to thediagonal functor ∆B : D(A) → D(A)B which sends X ∈ D(A) to the constant B-shaped diagram with value X. The point of this subsection is that, in general, theaxioms of a derivator do not imply the existence of such adjoints.

However, there are adjoints to a closely related functor. Let πB : B ×A→ A bethe projection away from B with restriction functor π∗B : D(A) → D(B × A). Bydefinition of a derivator, π∗B admits a left adjoint (πB)! : D(B ×A)→ D(A) and aright adjoint (πB)∗ : D(B ×A)→ D(A). Note that the two functors

(6.28) ∆B : D(A)→ D(A)B and π∗B : D(A)→ D(B ×A)

have the same domain but that the targets, in general, are essentially differentcategories. In fact, objects in D(B ×A) are diagrams which are coherent in the A-and the B-direction, while objects in D(A)B are coherent in the A-direction only.

The two functors in (6.28) are related by the following partial underlying diagramfunctors. Given categories A and B, evaluating on objects and morphisms in Bonly we obtain a partial underlying diagram functor

diaB,A : D(B ×A)→ D(A)B .

In formulas, associated to b ∈ B there is b × 1A : A ∼= 1 × A → B × A and givenX ∈ D(B ×A) we set

(6.29) diaB,A(X)(b) = (b× 1A)∗X ∈ D(A).

Lemma 6.30. Let D be a prederivator and let A,B ∈ Cat.

(i) There is a partial underlying diagram functor diaB,A : D(B × A) → D(A)B

defined by (6.29).(ii) The functor diaB,A satisfies ∆B = diaB,A π∗B : D(A)→ D(A)B.

Proof. This proof is left as an exercise.

The underlying diagram functor diaB : D(B) → D(1)B of Lemma 6.5 is canon-ically isomorphic to diaB,1 : D(B × 1) → D(1)B . We emphasize that, in general,also the functors diaB,A, which make diagrams partially incoherent, are far frombeing equivalences.

Page 74: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

74 MORITZ GROTH

Thus, given a derivator D and small categories A,B there is the diagram

(6.31)

D(B ×A)diaB,A

//

(πB)!

(πB)∗

D(A)B

D(A)

π∗B

OO

id// D(A).

∆B

OO

If the partial underlying diagram functor diaB,A : D(B×A)→ D(A)B is an equiv-alence, then also ∆B : D(A) → D(A)B has adjoints on both sides, which is to saythat D(A) has categorical (co)limits of shape B.

Lemma 6.32. Let D be a derivator and let A ∈ Cat.

(i) The category D(A) admits a terminal object ∗ and an initial object ∅.(ii) The category D(A) admits products and coproducts.

Proof. Let S be a set which we consider as a discrete category, i.e., a categorywith identity morphisms only. As a special case of axiom (Der1), the canonicalfunctor D(

∐s∈S A) →

∏s∈S D(A) induced from the inclusions is an equivalence.

One easily checks that the diagram

D(∐s∈S A)

' //

'

∏s∈S D(A)

'

D(S ×A)diaS,A

// D(A)S

commutes where the vertical maps are the canonical identifications. Thus, fordiscrete categories S the functors diaS,A are equivalences as well. It is immediatefrom the above discussion (see (6.31) in the case of B = S) that D(A) hence hasS-fold products and coproducts. Specializing to S = ∅ we deduce the existence ofterminal and initial objects.

Remark 6.33. (i) One way to refer to this result is by saying that in derivatorscategorical (co)products and homotopy (co)products both exist and that theyagree. This is particular to (co)products in that, in general, D(A) does notadmit further categorical (co)limits.

(ii) Using categories of diagrams one can also include examples of derivators suchthat only finite (co)products exist (this comment is related to Remark 6.3(ii)).

7. Homotopy exact squares and Kan extensions

In this section we discuss in more detail the formalism behind axiom (Der4),making precise that Kan extensions in derivators are pointwise. This formalismrelies on the calculus of mates, a certain calculus which applies to natural trans-formations. We begin in §7.1 by reviewing this calculus and illustrate it by someexamples.

In §7.2 we introduce the related notion of homotopy exact squares, arguably themain technical tool in the theory of derivators. While working with derivators, oneoften runs into the situation that outputs of certain constructions ‘obviously areisomorphic’. The formalism of homotopy exact squares often allows one to, first,construct canonical maps between such gadgets and, second, to deduce that these

Page 75: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 75

canonical maps actually are isomorphisms. To conclude this section, in §7.3 wecollect a few first applications of homotopy exact squares to the calculus of Kanextensions in derivators.

7.1. The calculus of mates. Before we define canonical mates, let us quicklyrecall the notion of pasting. This is a composition operation which applies to naturaltransformations living in ‘larger diagrams’. Here we refrain from giving a precisedefinition of such diagrams and instead only mention a few relevant examples. Letus consider the diagram

(7.1)

C1

C2

v∗1oo C3v∗2oo

D1

u∗1

OO

D2

u∗2

OO

w∗1

oo D3,w∗2

oo

u∗3

OO

which consists of categories, functors, and transformations α1 : v∗1u∗2 → u∗1w

∗1 and

α2 : v∗2u∗3 → u∗2w

∗2 . We define the natural transformation α1 α2 as

α1 α2 = (α1w∗2) (v∗1α2) : v∗1v

∗2u∗3 → v∗1u

∗2w∗2 → u∗1w

∗1w∗2 ,

and refer to it as the horizontal pasting of the squares in (7.1). Similarly, wedefine the vertical pasting of squares. The pasting operation also applies to morecomplicated such diagrams as long as all natural transformations ‘point in the samedirection’. As an additional simple example we include the following.

Example 7.2. Let (L,R) : C D be an adjunction with unit η : 1→ RL and counitε : LR → 1. We recall from §A.1 that the unit and the counit are subject to thetriangular identities,

LLη//

=++

LRL

εL

RηR//

=++

RLR

L, R.

Using the pasting operation we can rewrite the triangular identities concerning theleft adjoint L as

C L //

id ++

D

R

id

=

C

L

CL//

;Cη

D

;Cε

D,

and there is a similar picture for R.

As a special case of this pasting operation we now define mates. Let us considera natural transformation α : p∗u∗ → v∗q∗ living in a square of not necessarily smallcategories

(7.3)

C1 oop∗

OO

v∗

C2OOu∗

D1ooq∗D2.

Page 76: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

76 MORITZ GROTH

Having our applications in mind, in this section functors are often denoted likeprecomposition functors u∗ and left adjoints and right adjoints to such functors aredenoted by u! and u∗, respectively. If the necessary adjoint functors exist, then anynatural transformation α as in (7.3) has two associated canonical mates, namely

α! : v!p∗ η→ v!p

∗u∗u!α→ v!v

∗q∗u!ε→ q∗u! and(7.4)

α∗ : u∗q∗η→ p∗p

∗u∗q∗α→ p∗v

∗q∗q∗ε→ p∗v

∗.(7.5)

Here, η denotes the units of the adjunctions (u!, u∗) and (p∗, p∗), and ε the counits

of the adjunctions (v!, v∗) and (q∗, q∗), respectively. Thus, α! is defined as the

following pasting on the left, while α∗ is defined as the following pasting on theright,

C2

D1

C1v!oo

C2p∗oo

C1

p∗

OO

C2p∗oo

=kk

D1

v∗

OO

=

UU

D2q∗oo

u∗

OO

C2,u!oo

=kk

D1

v∗

OO

D2q∗oo

u∗

OO

D1.

q∗

OO

=

UU

Remark 7.6. (i) Note that the construction of the canonical mates depends onchoosing certain adjoint functors and adjunction (co)units. By Lemma A.11adjoint functors are unique up to unique isomorphism in a way that theisomorphism is compatible with adjunction units and counits. This impliesthat the canonical mates are well-defined up to some pasting with certaincanonical natural isomorphisms. In particular, the question if a canonicalmate is an isomorphism does not depend on the choices. This motivates usto refer to the transformations (7.4) and (7.5) as the canonical mates.

(ii) Let us again consider the square (7.3) and let us assume that the four functorshave adjoints on both sides. Using ‘conjugations by units and counits fromopposite sides’ the two natural transformations (7.4) and (7.5) are the onlyones which make sense for arbitrary α. Thus, the notation α 7→ α! andα 7→ α∗ is unambiguous.

Examples 7.7. (i) Associated to a functor u∗ : D → C there is the identity trans-formation u∗ → u∗, which populates the two squares

C

| id

Du∗oo C

id

C=oo

D

u∗

OO

D,

=

OO

=oo C

=

OO

D.

u∗

OO

u∗oo

If u∗ admits a left adjoint u! : C → D then the canonical mates (7.4) ofthe two squares yield the counit ε : u!u

∗ → id and the unit η : id → u∗u!,respectively. Similarly, if u∗ admits a right adjoint u∗ : C → D, then thecanonical mates (7.5) of these two squares are the unit η : id→ u∗u

∗ and thecounit ε : u∗u∗ → id, respectively.

Page 77: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 77

(ii) Let us consider functors u∗, v∗ : D → C and a natural transformation u∗ → v∗

which we want to rewrite as two different squares

C

| id

C=oo C

id

Du∗oo

D

v∗

OO

D,

u∗

OO

=oo C

=

OO

D.

=

OO

v∗oo

populated by this transformation. If u∗ and v∗ admit left adjoints u! andv!, respectively, then the canonical mate (7.4) of the square on the left isthe conjugate transformation or total mate v! → u! in the sense of (A.8).Similarly, if u∗ and v∗ admit right adjoints u∗ and v∗, respectively, then thecanonical mate (7.5) of the square on the right is the total mate v∗ → u∗ asdefined in (A.7).

(iii) Let A be a small category and let F : C → D be a functor. We abuse notationand simply write F : CA → DA for the induced functor. By the very definitionwe have the two commutative squares

DA F // CA DA F // CAYa

D

∆A

OO

F//

!

C,

∆A

OO

D

∆A

OO

F// C.

∆A

OO

Assuming the existence of the necessary (co)limits, the canonical mate (7.4)of the square on the left and the canonical mate (7.5) of the square on theright, respectively, are the canonical maps

colimA F → F colimA and F limA → limA F,

measuring whether F : C → D preserves (co)limits of shape A; see (A.19) and(A.20).

(iv) Let C be a complete and cocomplete category, let u : A → B be a functorbetween small categories, and let b ∈ B. In the background of ?? there werethe slice squares

(u/b)p

//

π(u/b)

A

u

1b

// B

and

(b/u)q

//

π(b/u)

A

u

1b

// B.

AI

Here, the component of the natural transformation in the square on the leftat the object (a, f : u(a) → b) is given by f , and similarly in the square onthe right. Passing to diagram categories we obtain induced natural transfor-mations

C(u/b)

CAp∗

oo

C

π∗(u/b)

OO

CBb∗

oo

u∗

OO

and

C(b/u) CAq∗

oo

C

π∗(b/u)

OO

CB .b∗

oo

AIu∗

OO

Page 78: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

78 MORITZ GROTH

The canonical mate (7.4) of the square on the left and the canonical mate(7.5) of the square on the right are the natural isomorphisms

colim(u/b) p∗ → b∗ LKanu and b∗ RKanu → lim(b/u) q∗

of ??, expressing that Kan extension in sufficiently (co)complete categoriesare pointwise.

Lemma 7.8. The passages to canonical mates are compatible with respect to hori-zontal and vertical pasting as expressed by the formulas

(α1 α2)! = (α2)! (α1)! and (α1 α2)∗ = (α2)∗ (α1)∗.

Proof. Let us consider the horizontal pasting α1α2 of two natural transformationsα1 : v∗1u

∗2 → u∗1w

∗1 and α2 : v∗2u

∗3 → u∗2w

∗2 ,

C1

C2

v∗1oo C3v∗2oo

D1

u∗1

OO

D2

u∗2

OO

w∗1

oo D3.w∗2

oo

u∗3

OO

Assuming that the functors u∗1, u∗2, and u∗3 have left adjoints, the pasting (α2)!(α1)!

is by definition given by

D1

C1

(u1)!oo C2

v∗1oo

D1

=

UU

u∗1

OO

D2

w∗1

oo

u∗2

OO

C2

=kk

(u2)!

oo C3

v∗2

oo

D2

=

UU

u∗2

OO

D3w∗2

oo

u∗3

OO

D3.

=kk

(u3)!

oo

Note that the two triangles in the middle cancel out by a triangular identity (seeExample 7.2) and we are left with (α1 α2)! as intended.

Let us consider a natural transformation (7.3) such that the canonical mate α!

is defined. This mate can then also be written as

(7.9)

D1oov!

OO

q∗

C1OOp∗

D2oou!C2,

and obviously the horizontal functors now have right adjoints. The canonical mate(α!)∗ is hence defined and can be chosen to be α.

Lemma 7.10. The two different formations of mates α 7→ α! and α 7→ α∗ areinverse to each other, i.e., we have α = (α!)∗ and α = (α∗)!.

Proof. This proof is left as an exercise.

Let us now consider a natural transformation (7.3) such that all four functorshave left adjoints. The canonical mate α! can again be written as in (7.9), and byour assumption it has a further canonical mate (α!)! : q!v! → u!p!.

Page 79: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 79

Lemma 7.11. Let α be a natural transformation (7.3) such that all functors haveleft adjoints. The canonical mate (α!)! : q!v! → u!p! is conjugate to α : p∗u∗ → v∗q∗.

Proof. This proof is left as an exercise.

Lemma 7.12. Given a natural transformation α as in (7.3). The canonical matesα! : v!p

∗ → q∗u! and α∗ : u∗q∗ → p∗v∗ are conjugate. In particular, α! is an iso-

morphism if and only if α∗ is an isomorphism.

Proof. The canonical mate α! can again be written as in (7.9). Note that for thatnatural transformation all four functors have right adjoints. By Lemma 7.10 wehence conclude that the canonical mate α∗ : u∗q∗ → p∗v

∗ is given by ((α!)∗)∗ andthe dual version of Lemma 7.11 then implies that α! and α∗ are conjugate. Finally,the second claim is guaranteed by Corollary A.9.

7.2. Homotopy exact squares. We are mostly interested in applications of thecalculus of mates to derivators. Let D be a derivator and let us consider a naturaltransformation in Cat which lives in a square

(7.13)

Cp//

v

| α

A

u

Dq// B.

Since a derivator is a 2-functor Catop → CAT we have an induced natural transfor-mation

D(C) oop∗

OO

v∗ α∗

D(A)OO

u∗

D(D) ooq∗

D(B).

As summarized in §7.1, such a transformation has canonical mates

α! : v!p∗ η→ v!p

∗u∗u!α∗→ v!v

∗q∗u!ε→ q∗u! and(7.14)

α∗ : u∗q∗η→ p∗p

∗u∗q∗α∗→ p∗v

∗q∗q∗ε→ p∗v

∗.(7.15)

(Again we generically write η for adjunction units and ε for adjunction counits.)Thus, α! is a natural transformation of functors D(A) → D(D) and α∗ a naturaltransformation of functors D(D) → D(A). It follows from Lemma 7.12 that α! isan isomorphism if and only if α∗ is an isomorphism.

Definition 7.16. Let α : up→ qv be a natural transformation in Cat as in (7.13).The square (7.13) is homotopy exact if the canonical mate (7.14) or, equivalently,the canonical mate (7.15) is an isomorphism for every derivator D .

Example 7.17. (i) By definition of a derivator, Kan extensions can be calculatedpointwise and this is made precise by axiom (Der4). Note that the tworelevant natural transformations (6.11) and (6.12) in that axiom are instancesof canonical mates, namely the ones associated to the two types of slicesquares (6.9). Thus, (Der4) is precisely saying that slice squares are homotopyexact.

Page 80: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

80 MORITZ GROTH

(ii) As a special case of the slice squares (6.9) we reconsider the following situationfrom Examples 5.35. Let A be a small category and let AB be the coconeon A, i.e., the category obtained from A by adjoining a new terminal object∞ ∈ AB. The maps a → ∞, a ∈ A, yield a natural transformation living inthe square

A1 //

πA

A

i

1 ∞// AB.

In fact, it is easily seen that this square is isomorphic to the slice squareassociated to i : A → AB and ∞ ∈ AB, and the square is hence homotopyexact.

(iii) The class of homotopy exact squares is closed under horizontal and verticalpasting. In fact, this is immediate from Lemma 7.8.

The point of this definition is that there is a battery of further examples ofhomotopy exact squares. These squares allow us to show that in the context ofabstract derivators (like homotopy derivators of model categories or complete andcocomplete∞-categories) certain canonical maps exist and are isomorphisms. A bitmore sloppy, these squares hence establish ‘rules which allow us to simultaneouslymanipulate categorical Kan extensions, derived Kan extensions, and homotopy Kanextensions’; see §6.3. For example, one ‘of course’ expects that given a categoryA ∈ Cat admitting a terminal object t ∈ A then for every derivator D and diagramX ∈ D(A) there should be a canonical isomorphism Xt

∼= colimAX in D(1). Thisand further first results along these lines are established in §7.3.

7.3. First applications to Kan extensions. In this section we illustrate the useof homotopy exact squares by extending a few results about the calculus of Kan ex-tensions from ordinary category theory to the framework of abstract derivators. Bythe duality principle, it suffices to make statements for left or right Kan extensions.

Lemma 7.18. For v : B → A a right adjoint between small categories the followingsquare is homotopy exact

Bv //

πB

| 1

A

πA

1id// 1,

i.e., for every derivator D and X ∈ D(A) the canonical mate

(πB)!v∗(X)→ (πA)!(X)

is a natural isomorphism.

Proof. For every adjunction (u, v, η, ε) : A B we obtain an induced adjunction

(v∗, u∗, η∗, ε∗) : D(B) D(A).

Thus, u∗ is a model for the right Kan extension functor v∗. By definition of ahomotopy exact square we hence have to show that

η∗π∗A : π∗A → u∗v∗π∗A

Page 81: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 81

is an isomorphism for every derivator D . But this natural transformation is givenby η∗π∗A = (πAη)∗ = 1∗ = 1, and hence clearly is an isomorphism.

In more standard notation, we thus have canonical isomorphisms

colimB v∗X → colimAX, X ∈ D(A),

saying that colimits of coherent diagrams are not affected by restrictions along rightadjoint functors. This is a derivator version of the ‘finality of right adjoints’.

Definition 7.19. A functor u : A → B in Cat is final if the canonical mate(πA)!u

∗ → (πB)! is an isomorphism.

Thus, u : A→ B is final if and only if the square

Au //

πA

| 1

B

πB

1 =// 1

is homotopy exact.

Examples 7.20. (i) Identity functors are final as are compositions of final func-tors.

(ii) Right adjoint functors and, in particular, equivalences are final.

Remark 7.21. If u : A→ B is final then the canonical mate colimA u∗ → colimB is

an isomorphism in every derivator D . Since this applies to represented derivators,such a functor is final in the usual sense of category theory, which, by Proposi-tion 5.12 can be characterized by the fact that for every b ∈ B the slice categories(b/u) are non-empty and connected. It can be shown that finality in the sense ofDefinition 7.19 is much more restrictive. In fact, it turns out that a functor u isfinal if and only if all slice categories (b/u) have weakly contractible nerves.

Corollary 7.22. Let A ∈ Cat admit a terminal object t. For X ∈ D(A) there is acanonical isomorphism

Xt

∼=→ colimAX.

Proof. This is immediate from Lemma 7.18 since t : 1 → A is right adjoint toπA : A→ 1.

Corollary 7.23. Let A ∈ Cat admit a terminal object t. There is a natural iso-morphism π∗A

∼= t∗ : D(1) → D(A). Moreover, t∗ is fully faithful and the essentialimage consists precisely of those X ∈ D(A) such that all morphisms in diaA(X)are isomorphisms.

Proof. This proof is left as an exercise.

Example 7.24. The inclusion of the terminal object 1 : 1 → [1] induces a fullyfaithful functor i∗ : D(1)→ D([1]). An object X ∈ D([1]) lies in the essential imageof i∗ if and only if the morphism dia[1](X) : X0 → X1 in D(1) is an isomorphism.We abuse terminology and refer to such a diagram X ∈ D([1]) as an isomorphism.

As in ordinary category theory, the following result justifies that we speak ofKan extensions. Since this is the first proof of this kind in this book, we give afairly detailed proof.

Page 82: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

82 MORITZ GROTH

Proposition 7.25. If u : A→ B is fully faithful, then the square

(7.26)

Aid //

id

| 1

A

u

Au// B

is homotopy exact. This is equivalent to saying that for every derivator D thefunctors u!, u∗ : D(A)→ D(B) are fully faithful.

Proof. The canonical mate (7.14) associated to this square is the adjunction unitη : 1 → u∗u! while the canonical mate (7.15) is the counit ε : u∗u∗ → 1. Thus, thesquare (7.26) is homotopy exact if and only if the Kan extension functors u!, u∗ arefully faithful.

Let us now show that this square is homotopy exact by showing that the unitη : 1→ u∗u! is an isomorphism. Since isomorphisms can be detected pointwise (byaxiom (Der2)) it is enough to show that all components ηa, a ∈ A, are isomorphisms.For this purpose, let us consider the following pasting

(7.27)

(A/a)p//

π(A/a)

~

A

1

id //

id

A

u

1a// A

u// B

where the square on the left is a slice square (6.9). The functoriality of mates withrespect to pasting implies that the canonical mate of this pasting is

(π(A/a))!p∗ ∼=→ a∗

ηa→ a∗u∗u!,

where the first map is an isomorphism by (Der4). Thus, ηa is an isomorphism ifand only if the canonical mate of the pasting (7.27) is an isomorphism.

Since u : A→ B is fully faithful, the reader easily verifies that the functor

(A/a)→ (A/u(a)) : (a′, f : a′ → a) 7→ (a′, u(f) : u(a′)→ u(a))

is an isomorphism of categories. Using this isomorphism we can rewrite the pasting(7.27) as the pasting

(7.28)

(A/a)∼= //

π(A/a)

(A/u(a))p//

π(A/u(a))

A

u

1 // 1u(a)

// B

in which the square on the right is a slice square (6.9). Since isomorphisms are finalfunctors (see Examples 7.20) the square on the left is homotopy exact. Thus, by(Der4) and Example 7.17(ii), the pasting (7.28)=(7.27) is homotopy exact, whichconcludes the proof.

Thus, Kan extensions along fully faithful functors u : A → B are fully faithful.In particular, an object X ∈ D(B) lies in the essential image of u! or u∗ if and onlyif the adjunction counit ε : u!u

∗(X) → X or the adjunction unit η : X → u∗u∗(X)

is an isomorphism, respectively. By (Der2) it is enough to check this for every

Page 83: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 83

object b ∈ B. The point of the following lemma is that it suffices to take care ofthe objects which do not lie in the image of u.

Lemma 7.29. Let u : A→ B be fully faithful and let X ∈ D(B).

(i) The diagram X lies in the essential image of u! if and only if the adjunctioncounit εb : u!u

∗(X)b → Xb is an isomorphism for all b ∈ B − u(A).(ii) The diagram X lies in the essential image of u∗ if and only if the adjunction

unit ηb : Xb → u∗u∗(X)b is an isomorphism for all b ∈ B − u(A).

Proof. By duality it suffices to take care of the case of left Kan extensions. ByProposition 7.25, u! is fully faithful and X lies in the essential image of u! if andonly if the adjunction counit ε : u!u

∗(X)→ X is an isomorphism. By (Der2) this isthe case if and only if we all components εb, b ∈ B, are isomorphisms, establishingone direction. The converse direction follows from the triangular identity (see (A.3))

1 = u∗ε · ηu∗ : u∗ → u∗u!u∗ → u∗.

In fact, since 1 and ηu∗ are isomorphisms this is also the case for u∗ε, and it ishence enough to check the objects which do not lie in the image of u.

As a further application of homotopy exact squares we show that derivators arestable under shifting.

Comments about dias

in dias.Example 7.30. Associated to B ∈ Cat there is the 2-functor B × − : Cat → Catwhich sends A to the product B × A. If we are also given a prederivator D , thenwe can define the shifted prederivator DB by setting

DB(A) = D(B ×A).

More precisely, the shifted prederivator is defined as

DB : Catop B×−→ Catop D→ CAT .

In particular, given u : A → A′ in Cat , the induced functor u∗ : DB(A′) → DB(A)is (1B×u)∗ : D(B×A′)→ D(B×A). The underlying category of DB is canonicallyisomorphic to D(B).

Let us illustrate the shifting operation by the following examples. In thoseexamples we already use the concept of an isomorphism of (pre)derivators – anotion which will be formally introduced in a later section.

Examples 7.31. (i) Represented case(ii) Let DA be the homotopy derivator of a Grothendieck abelian category A

and let B,A ∈ Cat . A combination of Lemma 3.10 with the categoricalexponential law yields canonical isomorphisms

DBA (A) = DA(B ×A)

= Ch(A)B×A[(WB×AA )−1]

∼= Ch(AB)A[(WAAB )−1]

= DAB (A).

Thus, there is an isomorphism of derivators DBA∼= DAB .

(iii) We now specialize to the derivator DR of a ring R (see also Examples 6.24(i)).(a) For every discrete group G there is an isomorphism DG

R∼= DRG, i.e., the

shifting operation encodes the passage to group algebras.

Page 84: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

84 MORITZ GROTH

(b) For every (finite) quiver Q there is an isomorphism DQR∼= DRQ, i.e., the

shifting operation encodes the passage to path algebras.(c) For every (finite) poset P there is an isomorphism DP

R∼= DRP , i.e., the

shifting operation encodes the passage to incidence algebras.(iv) Combinatorial model categories

In the above examples we saw instances of derivators D such that the associatedshifted prederivators DB are again derivators. This is fact always the case as weshow next.

Proposition 7.32. For every derivator D and every B ∈ Cat the prederivatorDB is again a derivator, the shifted derivator or the derivator of (coherent)B-shaped diagrams in D . The Kan extension functors of DB are calculated as

uDB

! = (1B × u)D! and uDB

∗ = (1B × u)D∗ .

Proof. The verification of the axioms (Der1)-(Der3) for DB is left to the reader. Itremains to check the pointwise formulas encoded by axiom (Der4). Thus, given afunctor u : A→ A′ and a′ ∈ A′ let us consider the square

(7.33)

B × (u/a′)1×p//

1×π

B ×A

1×u

B × 11×a′

// B ×A′

which is obtained from the slice square (6.9) by forming the product with B. Wehave to show that the canonical mate (1 × π)!(1 × p)∗ → (1 × a′)∗(1 × u)! is anisomorphism of functors D(B ×A)→ D(B × 1). Since isomorphisms are detectedpointwise by (Der2), it is enough to show that all components of the mate areisomorphisms. For b ∈ B we consider the following pasting diagram

(7.34)

((1× u)/(b, a′))

1

φ

∼=//

π

((1× π)/b)p′//

π

B × (u/a′)1×p//

1×π

B ×A

1×u

1 =// 1

b// B × 1

1×a′// B ×A′

in which the square in the middle is the slice square (6.9) associated to the functor1× π and b ∈ B ∼= B× 1. The square on the left is induced from the isomorphismsof categories

(7.35) ((1× π)/b) ∼= (B/b)× (u/a′) ∼= ((1× u)/(b, a′))

with (b, a′) : 1 → B × A′. The functoriality of mates with pasting (Lemma 7.8)implies that the canonical mate associated to (7.34) factors as

π!φ∗(p′)∗(1× p)∗

∼=→ π!(p′)∗(1× p)∗

∼=→ b∗(1× π)!(1× p)∗ → b∗(1× a′)∗(1× u)!.

Here, the second arrow is an isomorphism by (Der4) applied to D and the corre-sponding slice square. The first arrow is also an isomorphism because the isomor-phism (8.33) is final; see Examples 7.20. In order to conclude the proof it sufficesto show that (7.34) is homotopy exact for every b ∈ B. Since the pasting (7.34)agrees with the slice square (6.9) associated to 1 × u : B × A → B × A′ and theobject (b, a′) ∈ B ×A′, this follows by a further application of (Der4).

Page 85: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 85

8. Pointed derivators

In this section we introduce pointed derivators as derivators admitting zero ob-jects. Typical examples are homotopy derivators of abelian categories or homotopyderivators of pointed model categories, like the homotopy derivator of pointed topo-logical spaces.

We show in §8.1 that certain Kan extension functors along (co)sieves amountto extending diagrams by zero objects. This yields a convenient tool allowing usto ‘add zero objects where desired’. In §8.2 we illustrate this tool and constructsuspension, loop, cofiber, and fiber functors in pointed derivators, thereby generaliz-ing these classical constructions from homological algebra and homotopy theory. In§8.3 we include a short discussion of parametrized Kan extensions or Kan extensionsin shifted derivators, while in §8.4 we extend a few well-known results concerningpushout squares in ordinary categories to cocartesian squares in derivators. Thesetechniques are illustrated in our discussion of iterated cofiber constructions in §8.5.

8.1. Basics on pointed derivators. Let us recall from Lemma 6.32 that, given aderivator D and a small category A, the category D(A) admits an initial object ∅and a final object ∗.

Definition 8.1. A derivator D is pointed if the underlying category D(1) has azero object.

Thus, we ask axiomatically that the unique map ∅ → ∗ in D(1) is an iso-morphism. Following standard conventions, any zero object will be denoted by0 ∈ D(1). We again take up the examples from §6.3.

Examples 8.2. (i) A represented derivator y(C) is pointed if and only if the rep-resenting category C is pointed.

(ii) Homotopy derivators of Grothendieck abelian categories are pointed, hence,in particular, derivators of fields, rings, and schemes are pointed.

(iii) Homotopy derivators of pointed model categories are pointed. As specialcases, the homotopy derivator HoTop∗ of pointed topological spaces and thederivator Sp of spectra are pointed.

Lemma 8.3. (i) A derivator D is pointed if and only if Dop is pointed.(ii) If D is a pointed derivator, then D(A), A ∈ Cat , are pointed categories, i.e.,

the shifted derivators DA are again pointed.(iii) Let D be a pointed derivator and u : A→ B. The functors u∗ : D(B)→ D(A)

and u!, u∗ : D(A)→ D(B) preserve zero objects.

Proof. This proof is left as an exercise.

We know from Proposition 7.25 that Kan extensions along fully faithful functorsu : A → B are fully faithful and, by Lemma 7.29, that the essential images canbe characterized by the components of the adjunction (co)units lying in B − u(A).For special classes of fully faithful functors these characterizations of the essentialimages admit a particularly simple form.

Definition 8.4. Let u : A→ B be a fully faithful functor.

(i) The functor u is a cosieve if for every morphism u(a) → b in B it followsthat b lies in the image of u.

Page 86: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

86 MORITZ GROTH

(ii) Dually, u is a sieve if for every morphism b→ u(a) in B it follows that b liesin the image of u.

The following proposition will be of constant use in later sections, for example inthe construction of suspensions, cofibers, and cofiber sequences in pointed deriva-tors; see §8.2 and §8.5. To indicate its natural generality, we formulate the resultfor arbitrary derivators.

Proposition 8.5. Let D be a derivator and let u : A→ B be a functor.

(i) If u is a cosieve, then u! : D(A)→ D(B) is fully faithful and X ∈ D(B) liesin the essential image of u! if and only if Xb

∼= ∅ for all b ∈ B − u(A).(ii) If u is a sieve, then u∗ : D(A) → D(B) is fully faithful and X ∈ D(B) lies

in the essential image of u∗ if and only if Xb∼= ∗ for all b ∈ B − u(A).

Proof. We give a proof of the first statement. Proposition 7.25 guarantees thatu! : D(A)→ D(B) is fully faithful and by Lemma 7.29 a diagram X ∈ D(B) lies inthe essential image of u! if and only if the counit εb : u!u

∗(X)b → Xb, b ∈ B−u(A),is an isomorphism. By (Der4) there are canonical isomorphisms

u!u∗(X)b ∼= colim(u/b) p

∗u∗(X), b ∈ B.

We note that if b ∈ B−u(A) then the slice category (u/b) is empty, since u : A→ Bis a cosieve. By (Der1) the category D(∅) is equivalent to the terminal categoryand any object therein is hence a zero object. Since left adjoint functors preserveinitial objects, we deduce that for b ∈ B − u(A) there are isomorphisms

u!u∗(X)b ∼= colim(u/b) p

∗u∗(X) = colim∅ p∗u∗(X) ∼= ∅.

Thus, for b ∈ B − u(A) the map εb : u!u∗(X)b → Xb is an isomorphism if and only

if Xb∼= ∅, concluding the proof.

In the case of pointed derivators the two characterizations of the essential imagesagree. We say that X ∈ D(B) vanishes at an object b ∈ B if Xb

∼= 0, and wesimilarly speak of diagrams which vanish on a subcategory of B.

Corollary 8.6. Let D be a pointed derivator and let u : A→ B be a functor.

(i) If u is a cosieve, then u! : D(A) → D(B) is fully faithful and induces anequivalence onto the full subcategory of D(B) spanned by all diagrams whichvanish on B − u(A).

(ii) If u is a sieve, then u∗ : D(A)→ D(B) is fully faithful and induces an equiva-lence onto the full subcategory of D(B) spanned by all diagrams which vanishon B − u(A).

In the situation of the corollary, we refer to u! as left extension by zero andto u∗ as right extension by zero. These results already allow us to carry outsome interesting constructions in arbitrary pointed derivators, as we illustrate inthe §8.2.

8.2. Suspensions, loops, cofibers, and fibers. In this section we define sus-pensions, loops, cofibers, and fibers for arbitrary pointed derivators, thereby gener-alizing the classical constructions from homological algebra and homotopy theory.

Page 87: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 87

The following definition is motivated by Examples 5.35(iii). Let us recall that wedenote by = [1]× [1] the commutative square, i.e., the category

(0, 0) //

=

(1, 0)

(0, 1) // (1, 1)

and by ip : p→ and iy : y → the full subcategories obtained by removing thefinal object (1, 1) and the initial object (0, 0), respectively.

Definition 8.7. A square in a derivator D is an object in D(). A square in Dis cocartesian or cartesian if it lies in the essential image of (ip)! : D(p)→ D()or (iy)∗ : D(y)→ D(), respectively.

Remark 8.8. For every derivator D , the category of (coherent) spans D(p) and thecategory of cocartesian squares are equivalent categories, and dually for cospansand cartesian squares.

In §§8.3-8.4 we collect a few general results concerning (co)cartesian squares.Here, instead, we immediately want to apply this notion.

Example 8.9. A square in a represented derivator is cocartesian if and only if it is apushout square (Examples 5.35(iii)). In homotopy derivators of abelian categoriesor model categories the notion of cocartesian squares reduces to derived pushoutsquares and homotopy pushout squares.

We now reconsider the rather detailed description of the cone construction ofa morphism of chain complexes as given at the beginning of §5.1; see, in particu-lar, (5.1). Starting from a morphism of chain complexes f : X0 → X1, we first con-struct the span which is obtained by adding the inclusion in the cone i : X0 → CX0

and then pass to the pushout square. It is obvious how to mimic this second stepin the context of abstract pointed derivators. As for the first step, let us recall thatcones of chain complexes are contractible and are hence sent to zero objects in de-rived categories. As observed in Corollary 8.6, right Kan extensions along inclusionsof sieves are right extensions by zero. This suggests the following generalization ofthe cofiber construction to arbitrary pointed derivators.

Related to the span and the cospan we have the fully faithful functors

(8.10) i : [1]→ p and k : [1]→y

classifying the horizontal morphism (0, 0) → (1, 0) and the vertical morphism(1, 0)→ (1, 1), respectively. Combining these functors with the fully faithful inclu-sions ip : p→ and iy : y→ we obtain the fully faithful functors

(8.11) i′ = ip i : [1]→ p→ and k′ = iy k : [1]→y→ .

Definition 8.12. Let D be a pointed derivator.

(i) The cofiber functor cof : D([1])→ D([1]) is defined as

cof : D([1])i∗→ D(p)

(ip)!→ D()(k′)∗→ D([1]).

A further evaluation at 1 ∈ [1] yields C = 1∗ cof : D([1])→ D(1).

Page 88: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

88 MORITZ GROTH

(ii) The fiber functor fib : D([1])→ D([1]) is defined as

fib : D([1])k!→ D(y)

(iy)∗→ D()(i′)∗→ D([1]).

A final evaluation at 0 ∈ [1] yields F = 0∗ fib : D([1])→ D(1).

Note that i : [1] → p is a sieve while k : [1] →y is a cosieve. As a special case ofCorollary 8.6 it follows that i∗ is right extension by zero and k! left extension by zero.Given f ∈ D([1]) with underlying diagram X0 → X1 there are thus a cocartesiansquare and a cartesian square in D with respective underlying diagrams

X0f//

X1

cof(f)

Fffib(f)

//

X0

f

0 // Cf, 0 // X1.

Proposition 8.13. For every pointed derivator D there is an adjunction

(cof, fib) : D([1]) D([1]).

Proof. By definition, the functors cof and fib respectively factor as indicated in

D([1])i∗ // D(p)

(ip)!//

i∗oo D()

(iy)∗//

(ip)∗oo D(y)

k∗ //

(iy)∗

oo D([1]);k∗

oo

see (8.10) and (8.11). Thus, there are four adjunctions of which the outer ones‘point in the bad direction’. In order to adress this issue, let us denote by

D(p)ex ⊆ D(p), D(y)ex ⊆ D(y), and D()ex ⊆ D()

the respective full subcategories spanned by the coherent diagrams vanishing atthe lower left corner (0, 1). As a special case of Corollary 8.6 we obtain adjointequivalences of categories

(i∗, i∗) : D([1]) ' D(p)ex and (k∗, k!) : D(y)ex ' D([1]).

Moreover, the fully faithfulness of Kan extensions along fully faithful functors(Proposition 7.25) implies that all four functors (ip)!, (ip)

∗ and (iy)∗, (iy)∗ preserve

the vanishing condition at the lower left corner (0, 1). As an upshot, the functorscof and fib respectively factor as

D([1])i∗ // D(p)ex

(ip)!//

i∗oo D()ex

(iy)∗//

(ip)∗oo D(y)ex

k∗ //

(iy)∗

oo D([1]).k∗

oo

Thus, (cof, fib) is obtained by composing four adjunctions (two of which actuallyare adjoint equivalences), concluding the proof.

Remark 8.14. Given a pointed derivator D we denote by D()cof ⊆ D() thefull subcategory spanned by the cofiber squares, i.e., the cocartesian squareswhich vanish on the lower left corner. Similarly, let D()fib ⊆ D() be the fullsubcategory given by the fiber squares, i.e., the cartesian squares which vanishon the lower left corner. The proof of Proposition 8.13 together with Remark 8.8imply that there are equivalences of categories

D([1]) ' D()cof and D([1]) ' D()fib.

Page 89: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 89

Thus, using coherent formulations, a morphism is simply ‘as good as a (co)fibersquare’. We will come back to this in §8.5; see Proposition 8.37.

In a similar way we can define suspension and loop functors in pointed derivators.Naively, for a pointed derivator D and X ∈ D(1) we would like to set

ΣX = C(X → 0) and ΩX = F (0→ X).

This can be made precise using the functors

0∗ : D(1)→ D([1]) and 1! : D(1)→ D([1]).

In fact, since 0: 1 → [1] is a sieve and 1: 1 → [1] is a cosieve, the above two Kanextension functors are extensions by zero (Corollary 8.6).

For the convenience of the reader, we include a more detailed construction ofthese functors. Let us consider the fully faithful functors

(8.15) i = (0, 0) : 1→ p and k = (1, 1) : 1→y

classifying the initial and the final object, respectively. Postcomposition with thefunctors ip : p→ and iy : y→ yields the fully faithful functors

(8.16) i′ = ip i : 1→ p→ and k′ = iy k : 1→y→ .

Definition 8.17. Let D be a pointed derivator.

(i) The suspension functor Σ: D(1)→ D(1) is defined as

Σ: D(1)i∗→ D(p)

(ip)!→ D()(k′)∗→ D(1).

(ii) The loop functor Ω: D(1)→ D(1) is defined as

Ω: D(1)k!→ D(y)

(iy)∗→ D()(i′)∗→ D(1).

We note again that i : 1 → p is a sieve while k : 1 →y is a cosieve so that i∗ isright extension by zero and k! left extension by zero (Corollary 8.6). As a summaryof these constructions, for X ∈ D(1) there is a cocartesian square and a cartesiansquare in D with respective underlying diagrams

X //

0

ΩX //

0

0 // ΣX, 0 // X.

Blabla.Remark 8.18. Only interesting in derived contexts. C as homotopy cokernel andF as homotopy kernel. One-categorical these are boring on objects X → 0 and0→ X, respectively.

Proposition 8.19. For every pointed derivator D there is an adjunction

(Σ,Ω): D(1) D(1).

Proof. It follows from the definition as well as equations (8.15) and (8.16) that thefunctors Σ and Ω respectively factor as

D(1)i∗ // D(p)

(ip)!//

i∗oo D()

(iy)∗//

(ip)∗oo D(y)

k∗ //

(iy)∗

oo D(1).k∗

oo

Page 90: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

90 MORITZ GROTH

Again, there are four adjunctions of which only the outer ones ‘point in the baddirection’. We denote by

D(p)ex ⊆ D(p), D(y)ex ⊆ D(y), and D()ex ⊆ D()

the respective full subcategories spanned by the coherent diagrams vanishing atthe lower left corner (0, 1) and the upper right corner (1, 0). As a special case ofCorollary 8.6 we obtain adjoint equivalences of categories

(i∗, i∗) : D(1) ' D(p)ex and (k∗, k!) : D(y)ex ' D(1).

Since the functors (ip)!, (ip)∗ and (iy)

∗, (iy)∗ preserve the vanishing conditions atthe corners (0, 1), (1, 0), the functors Σ and Ω respectively factor as

D(1)i∗ // D(p)ex

(ip)!//

i∗oo D()ex

(iy)∗//

(ip)∗oo D(y)ex

k∗ //

(iy)∗

oo D(1).k∗

oo

As a composition of two adjoint equivalences and two adjunctions also (Σ,Ω) is anadjunction.

The remaining main goal in this section is to study iterated constructions ofcofibers which, in stable derivators, will lead to canonical triangulations. For thispurpose however we need some basic results concerning (co)cartesian squares whichwe collect in §8.4. Some of these results in turn rely on a basic understanding ofparametrized Kan extensions which we study next.

8.3. Parametrized Kan extensions. The goal of this subsection is to establish aresult making precise that restriction functors and Kan extensions in unrelated vari-ables commute (Proposition 8.20) and to deduce some interesting consequences. Inparticular, this allows us to understand (co)cartesian squares in shifted derivators.

To begin with, let us recall the following observation from ordinary categorytheory. If C is a cocomplete category and B ∈ Cat , then the diagram categoryCB is also cocomplete and colimits in CB are constructed coordinatewise. Moreprecisely, for every small category A ∈ Cat and diagram X : A → CB the colimitcolima∈AX(a) ∈ CB exists and for every b ∈ B there is an isomorphism(

colima∈AX(a))(b) ∼= colima∈A

(X(a)(b)

).

Actually, a typical proof of the cocompleteness is given by showing that the righthand side of the above isomorphisms defines a diagram B → C and that it, togetherwith a suitably defined cocone, yields a colimit colimAX ∈ CB . We sometimes referto the colimits in CB as colimits with parameters or parametrized colimits.It can be shown that every evaluation morphism b∗ : CB → C preserves colimits,i.e., that the corresponding canonical map is an isomorphism.

More generally, as a cocomplete category CB admits left Kan extensions alongfunctors u : A → A′ (??) which we also refer to as left Kan extensions withparameters or parametrized left Kan extensions. It turns out that evaluationfunctors also commute with (parametrized) Kan extensions, i.e., that for everyb ∈ B the following square

(CB)Au! //

b∗

(CB)A′

b∗

CAu!

// CA′

@H∼=

Page 91: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 91

commutes up to a certain canonical isomorphism.As a final generalization, we can pass from evaluation functors to general restric-

tion functors. In fact, it turns out that, given a functor v : B → B′, then the inducedrestriction functor v∗ : CB′ → CB preserves colimits and left Kan extensions. Theseresults from ordinary category theory can be obtained by specializing the followingproposition to represented derivators. At the same time, that proposition yieldssimilar results for arbitrary derivators, hence, for example, for homotopy derivatorsof abelian categories or model categories.

Proposition 8.20. For functors v : B → B′ and u : A → A′ between small cate-gories the square

(8.21)

B ×A v×1//

1×u

1

B′ ×A

1×u

B ×A′v×1// B′ ×A′

is homotopy exact, i.e., in every derivator the canonical mate transformations

(1× u)!(v × 1)∗ → (v × 1)∗(1× u)! and (v × 1)∗(1× u)∗ → (1× u)∗(v × 1)∗

are isomorphims.

Proof. We begin by reducing to the special case of u = πA : A → 1. For thispurpose, we consider the following two pastings

B × (u/a′)1×p//

π

B ×A v×1//

1×u

1

B′ ×A1×u

=

B × (u/a′)v×1//

π

1

B′ × (u/a′)1×p//

π

B′ ×A1×u

B1×a′

// B ×A′v×1// B′ ×A′ B

v// B′

1×a′// B′ ×A′

in which the square to the very left and to the very right are obtained from slicesquares be forming the product with a fixed category. We recall from the proof ofProposition 7.32 that such such squares are homotopy exact (see (7.33)).

By (Der2) and the compatibility of mates with respect to pasting, in order toconclude that (8.21) is homotopy exact it suffices to show that the pasting on theleft is homotopy exact. Since these two pastings agree, a further application of thefunctoriality of mates with pasting implies that it suffices to show that the secondsquare from the right is homotopy exact. This completes the reduction to the caseof u = πA : A→ 1.

As an additional reduction we show that is is enough to consider evaluationfunctors instead of more general restriction functors. To this end, let us considerthe pasting on the left in

(8.22)

At×1//

π

1

(B/b)×Ap×1//

π

B ×A v×1//

π

1

B′ ×A

π

=

Avb×1

//

π

1

B′ ×A

π

11

// 1b

// Bv

// B′ 1vb

// B′.

We have to show that in that pasting the right square is homotopy exact. As a slicesquare, we know that the square in the middle is homotopy exact. Moreover, thesquare on the left is induced from the functor t : 1→ (B/b) classifying the terminal

Page 92: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

92 MORITZ GROTH

object (b, 1: b→ b) ∈ (B/b). Since right adjoint funtors are final, the left square ishomotopy exact as well (Lemma 7.18). By the functoriality of mates with pastingand (Der2) it hence suffices to show that the pasting is homotopy exact. Since thepasting is easily seen to agree with the commutative square on the right in (8.22),this concludes the reduction to evaluation functors.

Finally, let us consider an object b′ ∈ B′. Our remaining task is to show thatthe commutative square to the left in

Ab′×1

//

π

1

B′ ×A

π

=

At×1//

π

1

(B′/b′)×A

π

p×1//

B′ ×A

π

1b′

// B′ 1 =// 1

b′// B′

is homotopy exact. In the pasting to the right, the right square is a slice squareand hence homotopy exact. The remaining square is induced by t : 1 → (B′/b′)pointing at the terminal object (b′, 1: b′ → b′). Since the pasting on the right iseasily seen to agree with the square to the very left, the finality of right adjoints(Lemma 7.18) and the compatibility of homotopy exact squares with respect topasting (Example 7.17) conclude the proof.

For Kan extensions along fully faithful functors there is the following conse-quence.

Corollary 8.23. Let D be a derivator, let u : A → A′ be fully faithful, and letB ∈ Cat. A coherent diagram X ∈ D(B×A′) lies in the essential image of (1×u)!

if and only if each Xb ∈ D(A′), b ∈ B, lies in the essential image of u!.

Proof. Let us consider the following pasting of commutative squares in which thesquare to the very left is homotopy exact as an instance of (8.21),

A

u

b×1// B ×A

1×u

1×u// B ×A′

=

=

A

u

u // A′

=

b×1// B ×A′

=

A′b×1// B ×A′ =

// B ×A′ A′ =// A′

b×1// B ×A′.

Since 1 × u is fully faithful, X ∈ D(B × A′) lies in the essential image of (1 × u)!

if and only if the adjunction counit (1 × u)!(1 × u)∗ → 1 is an isomorphism onX (Proposition 7.25). We note that this counit is the canonical mate associatedto the second square from the left. Thus, (Der2), the homotopy exactness of thesquare to the very left, and the functoriality of mates with pasting (Lemma 7.8)imply that X lies in the essential image of (1 × u)! if and only if for each b ∈ Bthe canonical mate of the pasting on the left is an isomorphism. Since the abovetwo pastings agree, this is the case if and only if for every b ∈ B the canonicalmate of the pasting on the right is an isomorphism on X. To conclude the proof itsuffices to observe that these canonical mates are the counits u!u

∗ → 1 applied toXb ∈ D(A′), b ∈ B.

Note that we can read X ∈ D(B×A′) as an object in DB(A′) while the objectsXb, b ∈ B, all live in D(A′). In this sense, the essential image of a fully faithfulKan extension functor in a shifted derivator DB can be detected by evaluating at

Page 93: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 93

all b ∈ B (see Proposition 7.32 for the construction of Kan extensions in shiftedderivators).

Corollary 8.24. Let D be a derivator and let B ∈ Cat. A square X ∈ DB() iscocartesian if and only if Xb ∈ D() is cocartesian for all b ∈ B.

Proof. This is an immediate application of Corollary 8.23 to the fully faithful func-tor ip : p→ .

We recall that an object X ∈ D([1]) is called an isomorphism if the underlyingdiagram X0 → X1 is an isomorphism.

Corollary 8.25. Let D be a derivator and let B ∈ Cat. The following are equiva-lent for a coherent morphism X ∈ DB([1]).

(i) The morphism X is an isomorphism.(ii) The morphisms Xb ∈ D([1]), b ∈ B, are isomorphisms.

(iii) The morphism X lies in the essential image of 0! : DB(1)→ DB([1]).

Proof. This follows from Example 7.24 and (Der2). Alternatively, we can alsoinvoke Example 7.24 and Corollary 8.23.

8.4. Cartesian and cocartesian squares. In this subsection we extend a fewwell-known results concerning pushouts and pullbacks in ordinary categories tocartesian and cocartesian squares in arbitrary derivators. Recall the definition ofthese squares given in Definition 8.7.

We begin by establishing some terminology. Let D be a derivator and let usconsider a square X ∈ D() with underlying diagram

X(0,0)//

X(1,0)

X(0,1)// X(1,1).

The square X ∈ D() is vertically constant if the morphisms X(0,0) → X(0,1) andX(1,0) → X(1,1) both are isomorphisms. There is a similar notion of horizontallyconstant squares and also the combined notion of constant squares.

We now consider the horizontal inclusion (id[1] × 0) : [1] → and the inclusion(0, 0) : 1→ of the initial object.

Corollary 8.26. Let D be a derivator.

(i) The functor (id[1]×0)! : D([1])→ D() is fully faithful and induces an equiv-alence onto the full subcategory of D() spanned by the vertically constantsquares.

(ii) The functor (0, 0)! : D(1)→ D() is fully faithful and induces an equivalenceonto the full subcategory of D() spanned by the constant squares.

Proof. The first statement is immediate from Corollary 8.25 and the second state-ment is a special case of (the dual of) Corollary 7.23.

The horizontal inclusion id[1] × 0: [1]→ factors as ip i : [1]→ p→ .

Lemma 8.27. Let D be a derivator. The functor i! : D([1])→ D(p) is fully faithfuland induces an equivalence onto the full subcategory of D(p) spanned by those Xsuch that X(0,0) → X(0,1) is an isomorphism.

Page 94: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

94 MORITZ GROTH

Proof. By Proposition 7.25 the functor i! is fully faithful with essential image thoseX such that ε : i!i

∗(X)→ X is an isomorphism at (0, 1). To reformulate this let usconsider the pasting on the left

1∼= //

1

(i/(0, 1)) //

[1]i //

i

p

1

=

1(0,0)

//

1

z

p

1

11

// 1(0,1)

// p1// p 1

(0,1)// p,

in which the square in the middle is a slice square and the square to the left is givenby an isomorphism. Since these two squares are homotopy exact, the functoriality ofcanonical mates with pasting implies that ε(0,1) is an isomorphism on X if and onlyif the canonical mate associated to this pasting is an isomorphism at X. Since thispasting agrees with the square to the right given by the morphism (0, 0) → (0, 1).The canonical mate associated to this square is X(0,0) → X(0,1), concluding theproof.

The relation of suitably constant squares to (co)cartesian squares is as follows.One of the implications is a derivator version of the classical fact that pushouts ofisomorphisms are isomorphisms.

Proposition 8.28. Let D be a derivator and let X be a square in D such thatX(0,0) → X(0,1) is an isomorphism. The square X is cocartesian if and only ifX(1,0) → X(1,1) is an isomorphism, i.e., the square is vertically constant.

Proof. The inclusion id[1] × 0: [1]→ factors as i : [1]→ p followed by ip : p→ .Hence, there is a canonical isomorphism (id[1]×0)!

∼= (ip)!i!, and, as a consequenceof Corollary 8.26, vertically constant squares are thus cocartesian. Conversely, ifX is a cocartesian square such that X(1,0) → X(1,1) is an isomorphism, then, byLemma 8.27, X lies in the essential image of (ip)! i! ∼= (id[1] × 0)! and is byCorollary 8.26 vertically constant.

Let [2] be the poset (0 < 1 < 2) considered as a category. For 0 ≤ i ≤ j ≤ 2there is the functor ιi,j : [1] → [2] which sends 0 to i and 1 to j. The productcategory [2]× [1] will be denoted by . Given an object X ∈ D(),

X(0,0)//

X(1,0)//

X(2,0)

X(0,1)// X(1,1)

// X(2,1),

by restriction we obtain the left square ι∗01X ∈ D(), the right square ι∗12X ∈ D(),and the composite square ι∗02X ∈ D(). (Here we abused notation and simplywrote ιi,j instead of ιi,j × id[1].)

Cocartesian squares enjoy a composition and cancelation property as made pre-cise by Proposition 8.30. The following lemma is a first step towards that propo-sition. Let A ⊆ be the full subcategory obtained by removing the objects(1, 1), (2, 1) and let B ⊆ be the full subcategory obtained by removing (2, 1)only. Related to these subcategories there are fully faithful inclusion functors

k = j i : A i→ Bj→ .

Page 95: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 95

Lemma 8.29. Let D be a derivator.

(i) The functor i! : D(A) → D(B) is fully faithful and induces an equivalenceonto the full subcategory spanned by all X ∈ D(B) such that ι∗01(X) ∈ D()is cocartesian.

(ii) The functor j! : D(B) → D() is fully faithful and induces an equivalenceonto the full subcategory spanned by all X ∈ D() such that ι∗12(X) ∈ D()is cocartesian.

(iii) The functor k! : D(A) → D() is fully faithful and induces an equivalenceonto the full subcategory spanned by all diagrams X ∈ D() such that bothsquares ι∗01X, ι

∗12X are cocartesian.

Proof. We begin by statement (i). Since i : A→ B is fully faithful, i! is fully faithfuland X ∈ D(B) lies in the essential image if and only if the counit ε : i!i

∗(X)→ Xis an isomorphism at (1, 1) (Proposition 7.25). Using the homotopy exactness ofslice squares and the functoriality of mates with respect to pasting, this is the caseif and only if the canonical mate associated to the pasting on the left in

pι01 //

π

Ai //

i

B

1

=

p1 //

π

pip //

ip

1

ι01 // B

1

1(1,1)

// B1// B 1

(1,1)//

1//

ι01// B

is an isomorphism on X. In the above pasting on the right, the square to the left is aslice square. Since the above two pastings coincide, using again the functoriality ofmates with pasting, the homotopy exactness of slice squares, and Proposition 7.25,we deduce that X lies in the essential image of i! if and only if ι∗01(X) is cocartesian.In fact, the corresponding canonical mate factors as

colimp i∗pι∗01(X)

∼=→ (1, 1)∗(ip)!i∗pι∗01(X)

ε1,1→ (1, 1)∗ι∗01(X)

and is hence an isomorphism if and only if the counit ε1,1 is an isomorphism onι∗01(X) which in turn is the case if and only if the square ι∗01(X) is cocartesian(again by Proposition 7.25).

As for statement (ii), we note that the essential image of j! : D(A) → D(B)consists precisely of those X ∈ D(B) such that the counit ε : j!j

∗(X) → X isan isomorphism at (2, 1) (Proposition 7.25). To re-express this differently let usconsider the pasting on the left in

pι12 //

π

B

π

1 //

Bj//

j

1

=

p 1 //

π

pip //

ip

1

ι12 //

1

11// 1

(2,1)//

1// 1

(1,1)//

1//

ι12// ,

and for now let us focus on the two squares to the right. Proposition 7.25, thehomotopy exactness of slice squares, and the compatibility of mates with pastingimply that X lies in the essential image of j! if and only if the canonical mate ofthe pasting of those two squares is an isomorphism on X. We next observe that thefunctor ι01 : p→ B is a right adjoint. In fact, the functor s0 : [2] → [1] determinedby 0, 1 7→ 0 and 2 7→ 1 induces a left adjoint to ι01 : p→ B. Hence, by Lemma 7.18the corresponding square is homotopy and, again by the functoriality of mates

Page 96: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

96 MORITZ GROTH

with pasting, we conclude that X lies in the essential image of j! if and only ifthe canonical mate of the pasting to the left is an isomorphism on X. Since thatpasting can be rewritten as the above pasting on the right, we can now conclude asin the case of (ii).

One easily combines these two statements to deduce (iii).

This lemma will again show up in §8.5. The composition and cancelation prop-erty is as follows.

Proposition 8.30. Let D be a derivator and let X ∈ D() such that ι∗01X iscocartesian. Then ι∗12X is cocartesian if and only if ι∗02X is.

Proof. We consider X ∈ D() such that the square ι∗01(X) is cocartesian. ByLemma 8.29 and Proposition 7.25 for such a diagram X also the square ι∗12(X) iscocartesian if and only if X lies in the essential image of k! : D(A)→ D() if andonly if the counit ε : k!k

∗(X) → X is an isomorphism at (2, 1). In more detail, byProposition 7.25 the essential image of k! is characterized by ε : k!k

∗ → 1 beingan isomorphism at (1, 1) and (2, 1). Under our assumption the component ε1,1 isalways an isomorphism as the counit factors as

k!k∗(X) ∼= j!i!i

∗j∗(X)ε→ j!j

∗(X)ε→ X

and the component at (1, 1) of the first morphism is an isomorpism by our as-sumption on X (use Proposition 7.25 and Lemma 8.29) while the correspondingcomponent of the second morphism is always an isomorphism (Proposition 7.25).

To reformulate this we consider the pasting on the left in

pι02 //

π

A

π

1 //

Ak //

k

1

=

p1 //

π

pip //

ip

1

ι02 //

1

11// 1

(2,1)//

1// 1

(1,1)//

1//

ι02// .

Here, the functor ι02 : p→ A is a right adjoint so that the square to the very left ishomotopy exact by Lemma 7.18. In fact, a left adjoint is induced by the functors1 : [2] → [1] determined by 0 7→ 0 and 1, 2 7→ 1. Together with the homotopyexactness of slice squares and the functoriality of mates with pasting this showsthat for our diagram X ∈ D() the square ι∗12(X) is cocartesian if and only if thecanonical mate of the pasting on the left is an isomorphism on X. In the pasting onthe right, the square to the very left is a slice square and one notes that the abovetwo pasting coincide. Using standard arguments we thus conclude that ι∗12(X) iscocartesian if and only if ι∗02(X) is cocartesian.

8.5. Iterated cofiber constructions. In this subsection we apply our above re-sults on cocartesian squares and briefly discuss cofiber sequences. We will also showthat if we iterate the cofiber construction three times then we obtain the suspensionmorphism up to natural isomorphism.

But let us begin with the cofiber sequences themselves.

Definition 8.31. Let D be a pointed derivator. A cofiber sequence in D is adiagram X ∈ D() such that the following two conditions are satisfied.

(i) The squares ι∗01(X), ι∗12(X) ∈ D() are cocartesian.(ii) The diagram vanishes at (2, 0) and (0, 1), X2,0

∼= X0,1∼= 0.

Page 97: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 97

We denote by D()cof ⊆ D() the full subcategory spanned by all cofiber se-quences.

Thus, given a cofiber sequence X, we obtain an underlying incoherent diagramdia(X) : → D(1) looking like

(8.32)

xf//

y //

g

0

0 // z

h// w.

By Proposition 8.30 also the composite square ι∗02(X) is cocartesian and it has anunderlying diagram

x //

0

0 // w.

Let us recall from Definition 8.17 that such squares were used in the constructionof the suspension functor Σ: D(1)→ D(1). In fact, let D()ex ⊆ D() be the fullsubcategory spanned by all cocartesian squares which vanish at (1, 0) and (0, 1).As observed in the proof of Proposition 8.13 there is an equivalence of categoriesD(1) ' D()ex and we just saw that ι∗02(X) ∈ D()ex. It follows from thisobservation that for every cofiber sequence X with underlying diagram (8.32) thereis a canonical isomorphism

(8.33) φ : w ∼= Σx.

Let [3] be the poset (0 < 1 < 2 < 3) and let k : [3] → be the ‘diagonalembedding‘ pointing at the objects (0, 0) < (1, 0) < (1, 1) < (2, 1). Given a cofibersequence X ∈ D()cof , we can pass to the underlying diagram of k∗(X) ∈ D([3])which is an ordinary incoherent diagram [3] → D(1). Combining this diagramwith the canonical isomorphism (8.33) we obtain, as an upshot, for every cofibersequence X as in (8.32) an underlying incoherent cofiber sequence in D(1)

(8.34) xf→ y

g→ zφh→ Σx.

In the context of stable derivators, these incoherent cofiber sequences lead to canon-ical triangulations as we will see in §9.6.

Remark 8.35. In the construction of incoherent cofiber sequences we applied theunderlying diagram functor dia[3] : D([3]) → D(1)[3] and this step results in a lossof information. These incoherent cofiber sequences are some non-canonical shad-ows of certain universal constructions which were applied to a coherent morphism.In particular, given an incoherent cofiber sequence (8.34) it is true that the com-positions x → z and y → Σx are zero morphisms. But at that level we do notremember the reason why these compositions are zero, namely that they belong tocertain cofiber squares (Remark 8.14) in the background (see (8.32)).

As a related remark, in general, it is not possible to canonically reconstruct a co-herent version of the morphism f : x→ y starting with the given incoherent cofibersequence. This is to be seen in contrast to the following result (Proposition 8.37).

Page 98: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

98 MORITZ GROTH

We now show that every coherent morphism can be extended to a cofiber se-quence. For this purpose, let A ⊆ again denote the full subcategory obtained byremoving the objects (2, 0), (0, 1) and let k : A→ be the corresponding inclusion.Moreover, let i : [1]→ A classify the horizontal morphism (0, 0)→ (1, 0). For everyderivator D there are related Kan extension functors

(8.36) D([1])i∗→ D(A)

k!→ D().

Proposition 8.37. For every pointed derivator D the functors (8.36) induce anequivalence of categories

D([1]) ' D()cof .

Proof. The functors i : [1]→ A and k : A→ are both fully faithful and hence thesame is true for the Kan extension functors i∗ : D([1])→ D(A), k! : D(A)→ D().Hence, both functors restrict to equivalences onto their essential images and wedeal with both cases individually.

We note first that the functor i is a sieve and that i∗ hence is right extensionby zero (Corollary 8.6). Thus, the essential image of i∗ consists of precisely thoseX ∈ D(A) which vanish at (2, 0) and (0, 1). As for k! : D(A) → D() we alreadyknow by Lemma 8.29 that the essential image of k! consists precisely of thoseX ∈ D() such that both squares ι∗01(X), ι∗12(X) are cocartesian. Since k! is a fullyfaithful Kan extension functor it respects the vanishing condition at the objects(2, 0), (0, 1) ∈ A. In particular, k! restricts to an equivalence between the fullsubcategory D(A)ex of D(A) spanned by all diagrams which vanish at these twoobjects on the one-hand-side and the category D()cof of cofiber sequences on theother-hand-side.

As an upshot, we see that i∗ and k! respectively restrict to equivalences

D([1]) ' D(A)ex ' D()cof ,

concluding the proof.

Thus, if we stay at the level of coherent diagrams, then a morphism in a pointedderivator is simply as good as a cofiber sequence. Compare this observation toRemark 8.14.

We now add one more iteration of the construction of the cofiber. To make thisprecise, let us consider the full subcategory B ⊆ [2] × [2] obtained by removing(0, 2),

(8.38)

(0, 0) //

(1, 0) //

(2, 0)

(0, 1) // (1, 1) //

(2, 1)

(1, 2) // (2, 2).

Similarly to the case of cofiber sequences, we would like to extend a coherent mor-phism X ∈ D([1]) to a diagram of the above shape by first adding certain zeroobjects and then adding cocartesian squares.

To this end, let us denote by A ⊆ B the full subcategory spanned by (0, 0), (1, 0)and (0, 1), (2, 0), (1, 1). Related to this there is the fully faithful functor i : [1]→ A

Page 99: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 99

classifying the horizontal morphism (0, 0) → (1, 0) and the obvious fully faithfulinclusion j : A→ B. At the level of coherent diagrams in a derivator D we obtainthe corresponding Kan extension functors

(8.39) D([1])i∗→ D(A)

j!→ D(B).

Lemma 8.40. Let D be a pointed derivator and let D(B)cof ⊆ D(B) be the fullsubcategory spanned by all X ∈ D(B) satisfying the following conditions.

(i) The diagram X vanishes at (0, 1), (2, 0), and (1, 2).(ii) The restrictions of X to the left square, to the top right square, and to the

bottom right square are cocartesian.

The functors (8.39) induce an equivalence of categories

D([1]) ' D(B)cof .

Proof. Since the proof is similar to the one of Proposition 8.37 we allow ourselvesto be a bit sketchy. Both functors i : [1]→ A and j : A→ B are fully faithful, hencethe same is true for the Kan extension functors i∗ : D([1])→ D(A) and j! : D(A)→D(B). The functor i : [1] → A is a sieve and i∗ is hence right extension by zero(Corollary 8.6) and as a such it induces an equivalence onto the full subcategoryD(A)ex of D(A) spanned by all diagrams which vanish at (0, 1), (2, 0), and (1, 2).

We next analyze the functor j : A→ B and begin by observing that it factors asthe composition of fully faithful inclusions

j : Aj1→ A1

j2→ A2j3→ B

where the individual steps add the objects (1, 1), (2, 1), and (2, 2), respectively. Theleft Kan extension functor j! : D(A) → D(B) is accordingly naturally isomorphicto the composition

D(A)(j1)!→ D(A1)

(j2)!→ D(A2)(j3)!→ D(B).

Similarly to Lemma 8.29, one now observes that each of these three Kan extensionsprecisely amounts to adding a new cocartesian square in the obvious sense.

As an upshot, the functor j! hence amounts to adding three cocartesian squaresand combining this with the first part of the proof we are hence done.

The equivalence D([1]) ' D(B)cof sends a morphism (f : x → y) ∈ D([1]) to acoherent diagram looking like

(8.41)

xf//

y //

g

02

01// z

h //

x′

f ′

03// y′

and making all three squares cocartesian. In this diagram, the objects 01, 02, 03

denote zero objects in D(1) and the subscripts can be ignored for now. We nextshow that the vertical morphism (f ′ : x′ → y′) is naturally isomorphic to Σf .

Page 100: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

100 MORITZ GROTH

To make this precise, let us recall from Lemma 8.3 that pointed derivators arestable under shifting. In particular, if D is pointed then so is D [1], and the pointedderivator D [1] comes with a suspension functor Σ: D([1])→ D([1]).

Proposition 8.42. For every pointed derivator D there is a natural isomorphism

cof3 ∼= Σ: D([1])→ D([1]).

Proof. Let B be the category (8.38) and let us consider the functors k : [1] → Band l : [1]→ B classifying the morphisms

(0, 0)→ (1, 0) and (2, 1)→ (2, 2),

respectively. The main step consists of showing that for every Y ∈ D(B)ex thereis a natural isomorphism Σ(k∗Y ) ∼= l∗(Y ). To this end, let Y ∈ D(B)ex withunderlying diagram (8.41). There is a unique functor q : [1] × → B such thatq∗(Y ) ∈ D([1]×) has underlying diagram looking like

(8.43)

x //

f

02

y //

02

01

// x′

f ′

03// y′.

In this diagram, the [1]-coordinate is drawn diagonally while the -coordinate isas before and the decoration of the objects implies that there is at most one suchfunctor (the categories under consideration are posets). We leave it to the readerthat such a functor q : [1]×→ B actually exists, and we note that q satisfies

q(−, (0, 0)) = k : [1]→ B and q(−, (1, 1)) = l : [1]→ B.

The diagram Z = q∗(Y ) ∈ D([1] × ) has the property that both squaresZ0, Z1 ∈ D() are cocartesian. In fact, Z0 is cocartesian as a horizontal pastingof two cocartesian squares while Z1 is a vertical pasting of two cocartesian squares(Proposition 8.30). Thus, by Corollary 8.24, the square Z ∈ D [1]() is cocartesianwhen considered as an object of D [1]. Morever, this cocartesian square vanishesat (1, 0), (0, 1), implying that it qualifies for a construction of Σ: D([1]) → D([1]).More precisely, we obtain a canonical isomorphism

φ : Z1,1∼= Σ(Z0,0).

Unraveling definitions we see that Z1,1 = (1, 1)∗q∗Y = l∗Y and similarly we seethat Z0,0 = (0, 0)∗q∗Y = k∗Y , yielding the desired canonical isomorphism

Σ(k∗Y ) ∼= l∗(Y ).

We leave it to the reader to apply Lemma 8.40 to conclude the proof of thisproposition.

We will see in §9.6 that stable derivators canonically take values in triangulatedcategories. Ignoring signs for the moment, this proposition is a coherent version ofthe following observation. Recall the rotation axiom (T2) for triangulated categories(Definition 4.9) by which a distinguished triangle can be rotated backwards. At thelevel of coherent morphisms this amounts to passing to the cofiber of the coherent

Page 101: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 101

morphism. If we take a distinguished triangle and rotate it three times then weobtain a distinguished triangle which is the (negative of the) suspension of theoriginal triangle. The corresponding statement at the level of coherent morphismsis Proposition 8.42.

9. Stable derivators

Stable derivators are obtained from pointed derivators by imposing the linearitycondition that a square is cocartesian if and only if it is cartesian. This linearityaxiom allows us to establish derivator versions of well-known results from homo-logical algebra and stable homotopy theory, i.e., the study of spectra in the senseof topology. (In the next section we will see less well-known such results.) Stablederivators canonically take values in triangulated categories (based on the incoher-ent cofiber sequences introduced in §8), and stable derivators hence belong to thefamily of different enhancements of triangulated categories.

We begin in §9.1 with the definition of stable derivators and some first conse-quences of these axioms. In §9.2 we show that the values of stable derivators arepreadditive categories and in §9.3 we sketch ideas involved in a proof that thesecategories actually are additive. In §9.4 we include a short digression on morphismsof derivators and natural transformations between such morphisms. We also brieflytalk about adjunctions and equivalences of derivators and related notions. Finally,in §9.6 we show that (strong) stable derivators take values in triangulated categorieswhich are compatible with restriction functors and Kan extension functors.

9.1. Basics on stable derivators. Similarly to the theory of model categoriesand ∞-categories, we obtain stable derivators from pointed ones by imposing thefollowing linearity condition.

Definition 9.1. A pointed derivator D is stable if a square in D is cartesian ifand only if it is cocartesian. A bicartesian square is square which is both cartesianand cocartesian.

Examples 9.2. (i) The homotopy derivator of a Grothendieck abelian categoryis stable. In particular, we obtain stable derivators associated to fields, rings,and schemes.

(ii) The derivator Sp of spectra is stable. In a certain precise sense this is theuniversal example of a stable derivator, namely the free stable derivator gen-erated by the sphere spectrum.

(iii) The homotopy derivator of a stable model category or stable ∞-category isstable, giving rise to a plethora of stable derivators. We refer the reader to[GS14c, Examples 5.5] for a fairly long list of additional examples.

Lemma 9.3. A derivator D is stable if and only if the opposite Dop is stable.

Proof. This is immediate from the definition.

Thus, the duality principle also extends to stable derivators.

Lemma 9.4. Let D be a stable derivator and let B ∈ Cat. The shifted derivatorDB is stable.

Proof. By Lemma 8.3 the shifted derivator DB is pointed, thus it remains to showthat a square in DB is cocartesian if and only if it is cartesian. This follows from

Page 102: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

102 MORITZ GROTH

our discussion of parametrized Kan extensions in §8.3. In fact, by Corollary 8.24and its dual we can use the stability of D to see that X ∈ DB() = D(B × ) iscocartesian if and only if all Xb ∈ D(), b ∈ B, are cocartesian if and only if allXb ∈ D(), b ∈ B, are cartesian if and only if X ∈ DB() = D(B×) is cartesian.Thus, DB is also stable.

We begin by a few sanity checks, indicating that Definition 9.1 captures some ofthe expected phenomenons.

Proposition 9.5. Let D be a stable derivator. A morphism f ∈ D([1]) is anisomorphism if and only if Cf ∼= 0 if and only if Ff ∼= 0.

Proof. It follows from the construction of C : D([1]) → D(1) that for every mor-phism f ∈ D([1]) there is a cocartesian square

xf//

y

0 // Cf.

If f is an isomorphism then the same is true for 0 → Cf by Proposition 8.28, es-tablishing one direction. Conversely, let us assume that Cf ∼= 0. Thus, the bottomhorizontal morphism in the above cocartesian square is an isomorphism. Since D isstable, the square is also cartesian and it follows from the dual of Proposition 8.28that f is an isomorphism. By duality this concludes the proof.

Proposition 9.6. Let D be a stable derivator and let X ∈ D(). If two of thesquares ι∗01(X), ι∗12(X), and ι∗02(X) are bicartesian then so is the third one.

Proof. This is immediate from Proposition 8.30 and its dual.

We now establish a derivator version of the 5-lemma. In that result we considertwo cofiber squares X,X ′ ∈ D()cof looking like

x //

y

x′ //

y′

0 // z, 0 // z′

and a morphism X → X ′ with components f : x→ x′, g : y → y′, and h : z → z′.

Proposition 9.7. Let D be a stable derivator and let F : X → X ′ be a morphismof cofiber squares with components f : x → x′, g : y → y′, and h : z → z′. If two ofthe morphisms f, g, and h are isomorphisms then so is the third one.

Proof. Let us assume that f and g are isomorphisms. It follows from axiom (Der2)that the restriction i∗p(F ) is an isomorphism. The fully faithfulness of (ip)! impliesthat also F ∼= (ip)!i

∗p(F ) is an isomorphism hence so is F(1,1) = h. The dual proof

shows that if g, h are isomorphisms then so is f (use that the squares are also fibersquares).

We attack the remaining case and assume that f and h are isomorphisms. Aright Kan extension by zero followed by a left Kan extension allows us to extend

Page 103: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 103

X and X ′ to cofiber sequences X and X ′,

x //

y

// 0

x′ //

y′

// 0

0 // z // w, 0 // z′ // w′,

and to obtain an induced morphism of cofiber sequences F : X → X ′. The restric-tion ι∗0,2(F ) to the cocartesian composite squares is an isomorphism. In fact, it isan isomorphism if we restrict it further along ip : p→ since f is an isomorphismso that axiom (Der2) allows us to conclude. By the first part of this proof it follows

that ι∗0,2(F ) is an isomorphism. Thus, the restriction ι∗1,2(F ) to the squares on theright has the property that the two components at (0, 1), (1, 1) are isomorphisms.Since D is stable, this restriction is a morphism of fiber squares and we henceconclude by the first part of the proof.

Proposition 9.8. For every stable derivator D there is an equivalence

(cof, fib) : D([1]) ' D([1]).

Proof. We note again that the functors cof and fib respectively factor as indicatedin

D([1])i∗ // D(p)

(ip)!//

i∗oo D()

(iy)∗//

(ip)∗oo D(y)

k∗ //

(iy)∗

oo D([1]).k∗

oo

These four adjunctions restrict to equivalences as follows. Let us denote by

D(p)ex ⊆ D(p), D(y)ex ⊆ D(y), and D()ex ⊆ D()

the respective full subcategories spanned by the coherent diagrams satisfying thefollowing exactness conditions. In the first two cases we impose the vanishingcondition at (0, 1) while in the third case we consider only the squares which arebicartesian and satisfy this vanishing condition. It follows from Corollary 8.6 andProposition 7.25 that the functors cof and fib respectively factor as

D([1])i∗ // D(p)ex

(ip)!//

i∗oo D()ex

(iy)∗//

(ip)∗oo D(y)ex

k∗ //

(iy)∗

oo D([1]).k∗

oo

and that each individual step is an equivalence.

As already mentioned, we will see in §9.6 that stable derivators D canonicallytake values in triangulated categories. In particular, the underlying category D(1)can be endowed with a triangulated structure. The proposition above establishesthat we can rotate triangles back and forth without a loss of information. And thefollowing proposition gives rise to the suspension equivalences Σ: D(1) → D(1)belonging to the canonical triangulations.

Proposition 9.9. For every stable derivator D there is an equivalence

(Σ,Ω): D(1) ' D(1).

Proof. The proof is very similar to the previous case, the main modification beingthat we impose an additional vanishing condition at the upper right corner (1, 0).The details are left as an exercise to the reader.

Page 104: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

104 MORITZ GROTH

We conclude by mentioning the following theorem, offering different characteri-zations of stable derivators. A proof of this result can be found in [GPS14].

Theorem 9.10. The following are equivalent for a pointed derivator D .

(i) The adjunction (Σ,Ω): D(1) D(1) is an equivalence.(ii) The adjunction (cof, fib) : D([1]) D([1]) is an equivalence.

(iii) A square in D is cocartesian if and only if it is cartesian, i.e., D is stable.

9.2. The preadditivity of stable derivators. In this subsection we will see thatthe values D(A), A ∈ Cat , of a stable derivator D are preadditive categories. Asa preparation we generalize the following result from ordinary category theory inwhich we consider a pushout square

∅ //

Y

X // W

in a cocomplete category such that the upper left corner is populated by an initialobject. In this situation the object W is isomorphic to the coproduct X t Y , moreprecisely, the cospan X →W ← Y is a coproduct cocone for the pair (X,Y ).

To extend this to derivators let us consider the functor ((1, 0), (0, 1)) : 1t1→ which factors as compositions of fully faithful functors

(9.11) 1 t 1i→ p ip→ and 1 t 1

j→y iy→ .

Definition 9.12. Let D be a derivator. A coherent cospan X ∈ D(y) is a coprod-uct cocone if it lies in the essential image of j! : D(1 t 1)→ D(y).

Lemma 9.13. For every derivator D the category D(1 t 1) is equivalent to thefull subcategory D()copr ⊆ D() spanned by the cocartesian squares X such thatX(0,0)

∼= ∅. Moreover, a square X lies in D()copr if and only if X(0,0)∼= ∅ and if

the restriction (iy)∗(X) ∈ D(y) is a coproduct cocone.

Proof. The functor k = ((1, 0), (0, 1)) : 1 t 1 → classifying the upper right andthe lower left corner is fully faithful. By Proposition 7.25 the same is true fork! : D(1t 1)→ D() which hence induces an equivalence onto the essential image.Since k factors as indicated in (9.11), there are natural isomorphisms

k!∼= (ip)!i! ∼= (iy)!j!.

All of these functors are fully faithful and these induced factorizations allow us toobtain a different description of the essential image. Using the natural isomorphismk!∼= (ip)!i! we see that X ∈ D() lies in the essential image of k! if and only if

X is cocartesian and X(0,0)∼= ∅. In fact, since i is a cosieve this follows from

Proposition 8.5 and Proposition 7.25. Similarly, using the isomorphism k!∼= (iy)!j!

and the fact that iy is a cosieve it follows from the same two propositions that theessential image of k! consists precisely of those X with X(0,0)

∼= ∅ and such that(iy)∗(X) is a coproduct cocone.

A combination of this lemma with its dual allows us to establish the preadditivityof stable derivators.

Proposition 9.14. Let D be a stable derivator and A ∈ Cat. The category D(A)is preadditive.

Page 105: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 105

Proof. Let D be a stable derivator and A ∈ Cat . The derivator DA is again stable(Lemma 9.4) and there is an isomorphism of categories DA(1) ∼= D(A). Thus, it isenough to show that the underlying category D(1) is preaddive. By Lemma 6.32,the category D(1) admits small (co)products. Hence, in order to conclude it isenough to show that for X,Y ∈ D(1) the canonical map X t Y → X × Y is anisomorphism. To this end, we begin by observing that (Der1) yields a canonicalequivalence of categories D(1 t 1) ' D(1) × D(1). For X,Y ∈ D(1) we denoteby (X,Y ) ∈ D(1 t 1) the object corresponding to the pair (X,Y ) ∈ D(1) × D(1)under this equivalence.

Associated to (X,Y ) ∈ D(1 t 1) we construct a coherent diagram of shape[2]× [2]. Here, [2] is again the poset (0 < 1 < 2) and the product category [2]× [2]hence looks like

(0, 0) //

(1, 0) //

(2, 0)

(0, 1) //

(1, 1) //

(2, 1)

(0, 2) // (1, 2) // (2, 2).

Related to [2]× [2] there are categories and fully faithful functors

1 t 1i→ A

j→ Bk→ [2]× [2]

which are defined as follows.

(i) The category A ⊆ [2] × [2] is the full subcategory spanned by the objects(2, 0), (2, 1) and (0, 2), (1, 2). (Thus, A is isomorphic to [1] t [1].) We definethe functor i : 1 t 1→ A to be the cosieve classifying (1, 2) and (2, 1).

(ii) The category B ⊆ [2]× [2] is the full subcategory obtained from A by addingthe object (2, 2) and the functor j : A→ B is the obvious fully faithful inclu-sion. Note that this inclusion is a sieve.

(iii) Finally, the functor k : B → [2]×[2] is also the obvious fully faithful inclusion.

Associated to these fully faithful functors there are by Proposition 7.25 the fullyfaithful Kan extension functors

(9.15) D(1 t 1)i!→ D(A)

j∗→ D(B)k∗→ D([2]× [2]).

Moreover, since i is a cosieve and j a sieve it follows from Corollary 8.6 that i! andj∗ are left and right extension by zero, respectively. To understand the remainingfunctor k∗ we note that k : B → [2] × [2] factors as a composition of four fullyfaithful inclusions,

k : Bk1→ B1

k2→ B2k3→ B3

k4→ [2]× [2],

which are obtained by adding the objects (1, 1), (1, 0), (0, 1), and (0, 0) one by onein turn. Using arguments similar to the proof of Lemma 8.29, the reader checksthat each of the corresponding right Kan extension functors precisely amounts toadding a new cartesian square.

As an upshot, we have shown that (9.15) induces an equivalence between D(1t1)and the full subcategory D([2] × [2])ex ⊆ D([2] × [2]) spanned by all diagramsQ ∈ D([2]× [2]) satisfying the following exactness properties.

(i) The diagram Q vanishes at the three corners (0, 2), (2, 0), and (2, 2).

Page 106: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

106 MORITZ GROTH

(ii) The diagram Q makes the four squares cartesian.

Now, let (X,Y ) ∈ D(1t1) and let Q ∈ D([2]× [2])ex be the corresponding coherentdiagram. The diagram Q looks like

(9.16)

Z //

1

X ′ //

2

0

Y ′ //

3

B //

4

Y

0 // X // 0.

Considering the squares 2 , 4 only, Proposition 9.6 and (the dual of) Proposi-tion 8.28 imply that the composite square is cartesian and that the structure mapX ′ → X is an isomorphism. Similarly, we obtain that the structure map Y ′ → Y isan isomorphism and also that the upper left corner in (9.16) is populated by a zero

object Z ∼= 0. An application of (the dual of) Lemma 9.13 to 4 implies that B is

the product of X,Y . Finally, since D is stable the square 1 is also cocartesian andLemma 9.13 hence implies that B is the coproduct of X ′, Y ′. A final combinationof these observations with the isomorphisms X ′ ∼= X and Y ′ ∼= Y concludes theproof.

9.3. The additivity of stable derivators. In this subsection we mention ingre-dients involved in a proof that stable derivators take values in additive categories.Let us recall from §2.1 (see, in particular, Lemma 2.4) that it suffices to show thatthe abelian monoid structures on sets of morphisms in the preadditive categoriesD(A), A ∈ Cat , actually are group structures.

To this end, let D be a pointed derivator and let X ∈ D(1). Mimicking a classicalconstruction from topology, one can show that the loop object ΩX ∈ D(1) can beturned into a group object. The corresponding multiplication map

(9.17) ∗ = ∗X : ΩX × ΩX → ΩX, X ∈ D(1),

is an abstraction of the classical concatenation of loops in a pointed topologicalspace. We do not include the construction of these concatenation maps hereand instead refer the reader to [Gro13]. Nevertheless we collect the following resultmaking precise that loop objects are group objects.

Theorem 9.18. Let D be a pointed derivator and let X ∈ D(1). The concatenationmap (9.17) endows ΩX with the structure of a group object.

While the precise definition of the concatenation (9.17) is not relevant here, wewill later need some hint where the inversion map comes from. This hint is obtainedfrom the following heuristics in the context of pointed topological spaces.

Let (X,x0) be a pointed topological space. Then a typical model for the loopspace is given by the space of maps [0, 1] → X which send the boundary pointsto x0 (the topology is the compact-open topology). Here it is more convenient touse a different description of the same homotopy type. To begin with, let PX bethe space of paths [0, 1] → X starting at x0 and let ev1 : PX → X be the mapwhich evaluates a path at its target 1 ∈ [0, 1]. We define the loop space ΩX to be

Page 107: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 107

the pullback

(9.19)

ΩXp2 //

p1

PX

ev1

PXev1

// X.

Thus, up to homeomorphism, ΩX is the space of maps [−1, 1] → X which sendthe boundary points −1, 1 to the base point x0. The base point of this space is theconstant map [−1, 1]→ X with value x0.

In this model for the loop space, the inversion of loops ι : ΩX → ΩX is given by areparametrization via the reflection at the origin 0 ∈ [−1, 1] and this reparametriza-tion can be obtained by interchanging the two copies of PX in the pullback square(9.19). More formally, the outer commutative square in the diagram

ΩXp1

))p2

""

ΩXp2//

p1

PX

ev1

PXev1

// X

induces by the universal property of the pullback square a canonical dashed mor-phism making everything commute. The reader easily checks that this morpism isthe above reparametrization ι.

To mimick this in arbitrary pointed derivators, let

σ : y→y

be the swap symmetry which interchanges the objects (1, 0) and (0, 1). Givena pointed derivator D and X ∈ D(1) it can be shown that this swap symmetryinduces a loop inversion ι : ΩX → ΩX, yielding the inverses for the group structureof Theorem 9.18. In more detail, ι is essentially the canonical mate limy σ

∗ → limyassociated to

yσ //

π

y

π

1 =// 1

;C1

evaluated on the coherent cospan (1, 1)!(X) ∈ D(y). And this induces the loopinversion map.

Adapt, simplify, lessdetailed.Theorem 9.20. Let D be a stable derivator and let A ∈ Cat. The category D(A)

is additive.

Proof. By Lemma 9.4 it is enough to show that the underlying category D(1) isadditive. By Proposition 9.14 the category D(1) is preadditive and we have toshow that the abelian monoid structure on homD(1)(Y, Y ), Y ∈ D(1), is a groupstructure. The stability of D implies by Proposition 9.9 that there is an equivalenceof categories (Σ,Ω): D(1) ' D(1), yielding a canonical isomorphism η : Y ∼= ΩΣY .Thus, it suffices to consider the case of loop objects ΩX. In this case one can show

Page 108: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

108 MORITZ GROTH

that the loop inversion map (??) is an additive inverse to id : ΩX → ΩX, concludingthe proof by Lemma 2.4. We refer the reader to [Gro13] for more details.

We conclude this subsection by a remark sketching an alternative proof of theadditivity of stable derivators.

Remark 9.21. (i) Abstracting a classical result concerning the homotopy cate-gory Ho(Top∗) of pointed topological spaces, one can show that in arbitrarypointed derivators 2-fold loop objects are abelian group objects. More pre-cisely, let D be a pointed derivator and let X ∈ D(1). The object Ω2X canbe endowed with two structures of a group object, respectively coming withmultiplication morphisms

∗1 : Ω2X × Ω2X → Ω2X and ∗2 : Ω2X × Ω2X → Ω2X.

These morphisms are abstractions of the loop concatenations with respect todifferent sphere coordinates in topology. One can show that these two groupstructures agree and actually are abelian group structures.

(ii) The previous point yields a different proof of the additivity of stable derivators(Theorem 9.20). Again, it suffices to show that the preadditive category D(1)is additive. The preadditivity implies that every object X ∈ D(1) can beessentially uniquely endowed with the structure of a commutative monoidobject, and by Lemma 2.4 it remains to show that this unique commutativemonoid structure is a group structure. Since D is stable there is a naturalisomorphism X ∼= Ω2Σ2X and the concatenation morphism on 2-fold loopobjects turns X into an abelian group object. Thus, the unique abelianmonoid structures are abelian group structures, concluding the proof.

9.4. Morphisms and natural transformations. In this subsection we definemorphisms of derivators and natural transformations between such morphisms.These definitions are special cases of the more general concepts of pseudo-naturaltransformations and modifications.

Definition 9.22. A morphism of prederivators is a pseudo-natural transfor-mation.

Let us unravel this definition. If D and E are prederivators, then a morphismF : D → E consists of

(i) functors FA : D(A)→ E (A), A ∈ Cat , and(ii) natural isomorphisms γF,u = γu : u∗ FB → FA u∗ for u : A→ B in Cat .

This datum has to satisfy the following three coherence properties.

(a) For A ∈ Cat the transformation γ1A : FA → FA is the identity.

(b) For Au→ B

v→ C in Cat the equation γvu = γuv∗ · u∗γv holds,

D(C)(vu)∗

//

FC

D(A)

FA

=

D(C)v∗ //

FC

D(B)u∗ //

FB

D(A)

FA

E (C)(vu)∗

// E (A)

?G∼=

E (C)v∗// E (B)

u∗//

@H∼=

E (A).

@H∼=

(c) For functors u1, u2 : A → B and a natural transformation α : u1 → u2 wehave the equality γu2

· (α∗FB) = (FAα∗) · γu1

.

Page 109: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 109

Thus, a morphism of prederivators is given by a collection of functors which arecompatible with restrictions up to specified, coherent isomorphisms. A morphismof derivators is simply a morphism of underlying prederivators. We say that amorphism is strict if all γu are identity transformations (which is to say that wehave a 2-natural transformation).

Examples 9.23. (i) Let F : C → D be a functor between ordinary categories.Associativity of composition of functors implies that the induced functorsCA → DA : X 7→ F X commute with restriction functors on the nose.Hence, they assemble to a strict morphism of represented prederivators

y(F ) : y(C)→ y(D).

(ii) Let D be a prederivator and let v : B → B′ be in Cat . The restrictionfunctors (v × 1A)∗ : D(B′ × A) → D(B × A), A ∈ Cat , assemble to a strict

restriction morphism v∗ : DB′ → DB . In particular, there are evaluationmorphisms b∗ : DB → D , b ∈ B.

(iii) Left Quillen functors between combinatorial model categories have associatedleft derived morphisms between homotopy derivators, and similarly for rightQuillen functors; see §B.2 for more details.

Definition 9.24. Let F,G : D → E be morphisms of prederivators. A naturaltransformation φ : F → G is a modification.

Thus, such a φ : F → G consists of natural transformations φA : FA → GA whichfor each u : A→ B make the following diagram commute

u∗FBγF,u

//

u∗φB

FAu∗

φAu∗

u∗GB γG,u// GAu

∗.

The same definition also applies to derivators. The reader easily defines compositionlaws for morphisms and natural transformations of (pre)derivators and checks thatthese yield 2-categories.

Notation 9.25. We write PDER for the 2-category of prederivators, morphisms,and natural transformations and DER for the 2-category of derivators, morphisms,and natural transformations. Thus, the 2-category DER of derivators is a full sub-2-category of PDER. Often we simplify notation and drop some of the decorationsof the components of morphisms and natural transformations of (pre)derivators.

Examples 9.26. (i) Let F, F ′ : C → D be functors between ordinary categoriesand let φ : F → F ′ be a natural transformation. For A ∈ Cat there is the2-functor (−)A : CAT → CAT and φ hence yields natural transformationsφA : FA → F ′A of functors CA → DA, defining a natural transformationy(φ) : y(F )→ y(F ′). In fact, the passage to represented prederivators definesan embedding

y : CAT → PDER.

By restriction, there is a similar embedding of the 2-category of complete andcocomplete categories into the 2-category DER of derivators.

Page 110: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

110 MORITZ GROTH

(ii) Let D be a prederivator, let u, v : A → B be in Cat , and let φ : u → v be anatural transformation. The transformations φ× 1: u× 1C → v× 1C inducea natural transformation φ∗ : u∗ → v∗ of strict morphisms DB → DA. Infact, the following stronger result is true: the passage to shifted prederivatorsdefines a 2-functor

(−)(−) : Catop × PDER → PDER : (A,D) 7→ DA.

In particular, for every morphism F : D → E and B ∈ Cat there is an inducedshifted morphism

(9.27) FA : DA → EA.

Recall that a morphism of derivators is a pseudo-natural transformation comingwith structure isomorphisms. We can use the isomorphisms γ−1

u : FAu∗ → u∗FB

and γu : u∗FB → FAu∗,

D(B)u∗ //

FB

∼=

D(A)

FA

D(B)u∗ //

FB

D(A)

FA

E (B)u∗// E (A), E (B)

u∗// E (A),

@H∼=

in order to talk about morphisms of derivators which preserve certain Kan exten-sions.

Definition 9.28. (i) A morphism of derivators F : D → E preserves left Kanextensions along u : A→ B if the canonical mate transformation

(9.29) u!FAη→ u!FAu

∗u!γ−1u→ u!u

∗FBu!ε→ FBu!

is an isomorphism. A morphism of derivators is cocontinuous if it preservesarbitrary left Kan extensions.

(ii) A morphism of derivators F : D → E preserves right Kan extensionsalong u : A→ B if the canonical mate transformation

FBu∗η→ u∗u

∗FBu∗γu→ u∗FAu

∗u∗ε→ u∗FA

is an isomorphism. A morphism of derivators is continuous if it preservesarbitrary right Kan extensions.

Specializing to Kan extensions along πA : A → 1, there are obvious variants ofmorphisms preserving limits of a fixed shape or arbitrary limits, and dually.

Example 9.30. Let F : C → D be a functor between complete and cocompletecategories and let A ∈ Cat . If we consider the induced morphism of derivatorsy(F ) : y(C) → y(D) then the canonical mate (9.29) associated to u = πA : A → 1can be identified with the canonical map colimA F → F colimA defined in (A.19)(the reader is invited to verify this as an exercise).

Lemma 9.31. A morphism of derivators is cocontinuous if and only if it preservescolimits.

Proof. This proof is left as an exercise.

As a consequence one can deduce the following.

Page 111: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 111

Remark 9.32. Let C and D be complete and cocomplete categories. A functorF : C → D is cocontinuous in the usual sense of Definition A.21 if and only if thestrict morphism y(F ) : y(C)→ y(D) is cocontinous in the sense of Definition 9.28.

Our discussion of parametrized Kan extensions in §8.3 immediately establishesthe following result.

Proposition 9.33. Let D be a derivator and let v : B → B′ be a functor. Therestriction morphism v∗ : DB′ → DB is continuous and cocontinuous.

Proof. By duality it suffices to show that restriction morphisms are cocontinuous.Recall from Proposition 7.32 the construction of left Kan extensions in shiftedderivators. From this it follows that, given a functor u : A → A′, we have to showthat a certain canonical map

(1× u)!(v × 1)∗ → (v × 1)∗(1× u)!

is an isomorphism. But unraveling definitions the reader checks that this is preciselythe canonical mate expressing that the square (8.21) is homotopy exact and wehence conclude by Proposition 8.20.

As in ordinary category theory, examples of continuous or cocontinuous mor-phisms come from adjunctions. To this end we recall from §A.1 that an adjunctioncan be specified by two functors and two natural transformations satisfying thetriangular identities (A.3). This description allows us to extend to concept of anadjunction to arbitrary 2-categories.

Definition 9.34. An adjunction of derivators is an adjunction internal to the2-category DER.

Thus, an adjunction consists of morphisms L : D1 → D2 and R : D2 → D1

together with natural transformations η : 1 → RL and ε : LR → 1 satisfying thetriangular identities

1 = εL · Lη : L→ L and 1 = Rε · ηR : R→ R.

This general notion can be re-expressed in more elementary terms.

Proposition 9.35. (i) A morphism of derivators is a left adjoint if and only ifeach component is a left adjoint and the morphism is cocontinuous.

(ii) A morphism of derivators is a right adjoint if and only each component is aright adjoint and the morphism is continuous.

Proof. We refer the reader to [Gro13] for a proof.

The proof of this result shows the following. Let L : D1 → D2 be a cocontin-uous morphism such that all components LA : D1(A) → D2(A) are left adjoints.Then one can choose right adjoint functors RA : D2(A)→ D1(A) and these can beassembled to a right adjoint morphism of derivators R : D2 → D1.

We specialize this to the following situation. Let D be a derivator and let usconsider a restriction morphism v∗ : DB′ → DB which is cocontinuous by Proposi-tion 9.33. By (Der3) each component of v∗ has a left adjoint (v × 1)! which hence

assemble to a left Kan extension morphism of derivators v! : DB → DB′ .

Page 112: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

112 MORITZ GROTH

Corollary 9.36. Let D be a derivator and let u : A → B be a functor. There areadjunctions of derivators

(u!, u∗) : DA DB and (u∗, u∗) : DB DA.

In particular, the left Kan extension morphism u! : DA → DB is cocontinuous andthe right Kan extension morphism u∗ : DA → DB is continuous.

Definition 9.37. An equivalence of derivators is an equivalence internal to the2-category DER.

Thus, an equivalence consists of morphisms F : D1 → D2 and G : D2 → D1

together with natural isomorphisms η : 1 ∼= GF and ε : FG ∼= 1. In that case wealso say that the morphism F or G alone is an equivalence. Similarly to the caseof adjunctions there is the following description of equivalences in more elementaryterms.

Proposition 9.38. A morphism of derivators is an equivalence if and only if eachcomponent is an equivalence.

We conclude this subsection with the following remark.

Remark 9.39. In earlier sections of this book we established results about Kanextension functors in arbitrary derivators or suitable classes of derivators. If wetake the philosophy of derivators more serious, then we should not content ourselveswith such results concerning categories of coherent diagrams. In fact, based onthe shifting operation of derivators (Proposition 7.32), we should instead aim forsimilar such results about derivators of coherent diagrams. More specifically, in thecontext of homotopy derivators of model categories we would obtain statementsabout homotopy theories of diagrams which, in general, are stricter than merestatements about homotopy categories of diagrams.

It turns out that with a bit more work one can show that basically all our aboveresults extend to statements about Kan extension morphisms of derivators. Forexample, given a derivator D and a fully faithful functor u : A → B, then theKan extension morphisms u!, u∗ : DA → DB are fully faithful and hence induceequivalences onto their essential images.

These refined versions are central to our applications in abstract representationtheory in the final section, and in that section we freely use these derivator versionsof our earlier results. (In a more systematic treatment one would hence right awayestablish these results in their derivator version while for pedagogic reasons werefrained from doing so here.)

Todo 9.5. Strongness of derivators. Discuss definition and variants. Refer to ap-pendix for proof in case of combinatorial model categories. Discuss the functorsD(k)→ D([1])[1] → D(k) for a field k, AR-quivers

9.6. Canonical triangulations in stable derivators. In this subsection we con-struct canonical triangulations on strong stable derivators. Recall from Lemma 6.30that for every prederivator D and categories A,B ∈ Cat there is the partial under-lying diagram functor

diaB,A : D(B ×A)→ D(A)B ,

making a coherent diagram incoherent in the B-direction.

Page 113: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 113

Definition 9.40. A derivator is strong if dia[1],A : D([1] × A) → D(A)[1] is fulland essentially surjective for every A ∈ Cat .

Note that we do not ask that the functor is faithful, so the partial underlyingdiagram functors dia[1],A are not asked to be equivalences.

Examples 9.41. (i) Represented derivators are strong. In fact, in that case allpartial underlying diagram functors are equivalences.

(ii) Homotopy derivators of model categories are strong (see §B.2 for a proof).Thus, typical derivators arising in homological algebra and homotopy theory(like the ones mentioned in Examples 6.24) are strong.

(iii) A derivator D is strong if and only if Dop is strong.(iv) If D is a strong derivator and if B ∈ Cat , then the shifted derivator DB is

also strong.

Remark 9.42. The precise form of Definition 9.40 depends on the author. Forexample, Heller [Hel88] asks that the functors diaF,A : D(F × A) → D(A)F arefull and essentially surjective for all finite free categories F and all A ∈ Cat . Itcan be shown that the derivators mentioned in Examples 9.41 all satisfy this morerestrictive axiom.

The property of being strong is important when one wants to relate propertiesof stable derivators to structure on its values. This is illustrated in the proof of thefollowing theorem.

As a preparation recall from §8.5 the construction of (coherent) cofiber sequencesin a pointed derivator D . By Proposition 8.37 there is an equivalence of categoriesD([1]) ' D()cof sending a morphism to its cofiber sequence. Every coherentcofiber sequence has an underlying incoherent cofiber sequence, yielding the functorD()cof → D(1)[3] defined in (8.34). In the case of a stable derivator we denotethe following composition by

(9.43) tria : D([1]) ' D()cof → D(1)[3]

and refer to tria(X), X ∈ D([1]), as the standard triangle associated to X. Atriangle x → y → z → Σx in D(1) is distinguished if it is naturally isomorphicto a standard triangle.

Passing to shifted derivators this also defines a class of distinguished triangles onD(A), A ∈ Cat . In fact, with D also the shifted derivator DA is stable (Lemma 9.4)and the underlying category of DA is canonically isomorphic to D(A).

Theorem 9.44. Let D be a strong, stable derivator and A ∈ Cat. The suspensionfunctor Σ: D(A) → D(A) together with the above class of distinguished trianglesturn D(A) into a triangulated category.

Proof. Passing to shifted derivators we may assume that A = 1. By Theorem 9.20the category D(1) is additive and the suspension functor Σ: D(1) → D(1) is anequivalence by Proposition 9.9. It remains to show that the above class of distin-guished triangles satisfies the axioms (T1)-(T4) of Definition 4.9.

(T1): Let x ∈ D(1) and let us consider the corresponding identity morphismid: x→ x as an object in D(1)[1]. Since D is strong, the underlying diagram functorD([1]) → D(1)[1] is essentially surjective and we can find a coherent morphismX ∈ D([1]) and an isomorphism dia[1](X) ∼= id (one can also avoid using thestrength here by restricting along π : [1] → 1 or by invoking Example 7.24). To

Page 114: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

114 MORITZ GROTH

see that x → x → 0 → Σx is distinguished the reader easily checks that it sufficesto show that the third object in tria(X) is a zero object. And this claim followsimmediately from Proposition 9.5 since the third object in tria(X) is isomorphic toCX which is trivial as X is an isomorphism.

By definition the class of distinguished triangles is closed under isomorphismsand for axiom (T1) it hence remains to show that every morphism f : x → yin D(1) extends to a distinguished triangle. By assumption the diagram functorD([1]) → D(1)[1] is essentially surjective and we can hence find X ∈ D([1]) anda natural isomorphism dia[1](X) ∼= f . The reader checks that this isomorphismtogether with the distinguished triangle tria(X) yields the desired distinguishedtriangle extending f (using once more that the class of distinguished triangles isclosed under isomorphisms).

(T2): We leave it to the reader to verify that it suffices to establish the followingclaim: if we rotate backwards a standard triangle then we obtain a distinguishedtriangle. Since our distinguished triangles are obtained from two iterations of thecofiber construction, here it is convenient to consider coherent diagrams encodingthree iterations of cofiber constructions as it was already done in §8.5 (we suggestthe reader to recall the statement of Lemma 8.40 as well as Proposition 8.42 togetherwith its proof).

By this lemma, every coherent morphism X = (f : x → y) ∈ D([1]) can beextended to a coherent diagram looking like (8.41),

(9.45)

xf//

y //

g

02

01// z

h //

x′

f ′

03// y′,

vanishing as indicated and making all squares cocartesian (we can ignore the sub-scripts in the objects 01, 02, 03 for now). Note that the cofiber sequence consistingof the two top squares in (9.45) is used to construct the standard triangle tria(f).In more detail, one considers the composition of these two squares, looking like

(9.46)

x //

02

01// x′,

and the associated identification φ : x′ ∼= Σx in order to obtain the standard triangle

(9.47) xf→ y

g→ zφh→ Σx.

We next collect a detailed description of the standard triangle tria(g). To beginwith we note that if we reflect the two squares on the right in (9.45) in order to

Page 115: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 115

draw them horizontally, then we obtain the cofiber sequence on the left in

(9.48)

yg//

z

h

// 03

y //

03

02// x′

f ′// y′, 02

// y′,

which can hence be used to calculate tria(g). In fact, the cocartesian square on theright in (9.48) obtained by composing the cocartesian squares on the left yields anidentification ψ : y′ ∼= Σy and tria(g) hence looks like

(9.49) yg→ z

h→ x′ψf ′→ Σy.

It remains to show that this triangle is isomorphic to the rotation of (9.47). Tothis end, we recall that f ′ is naturally isomorphic to Σf . Since we need some detailsabout this identification, we repete from the proof of Proposition 8.42 that there isa coherent cube (8.43),

x //

f

02

y //

02

01

// x′

f ′

03// y′,

making the back face and the front face cocartesian. And the natural isomorphismf ′ ∼= Σf is obtained from this cube. Thus, in more detail, we use the identificationsx′ ∼= Σx and y′ ∼= Σy induced from the cocartesian squares

(9.50)

x //

02

y //

02

01// x′, 03

// y′.

Since the square on the left is precisely the square (9.46), the first identificationis again the morphism φ : x′ ∼= Σx showing up in (9.47). However, if we comparethe square on the right in (9.50) to the square on the right in (9.48), then we seethat they differ by a restriction along the automorphism swapping (1, 0) ↔ (0, 1).It follows from the sketch proof of Theorem 9.20 that the second identificationinduced by (9.50) hence is −ψ : y′ ∼= Σy, where ψ is as in (9.49). To summarize,the natural isomorphism f ′ ∼= Σf has components

x′f ′//

φ ∼=

y′

−ψ∼=

ΣxΣf// Σy.

Page 116: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

116 MORITZ GROTH

This finally gives us the desired isomorphism of triangles

y

=

g// z

=

h // x′

φ

ψf ′// Σy

=

yg// z

φh// Σx

−Σf// Σy

from the standard triangle tria(g) (see (9.49)) to the triangle which is obtainedfrom tria(f) (see (9.47)) by rotation.

(T3): We leave it to the reader to show that this axiom follows rather directlyfrom the assumption that D is strong.

(T4): It remains to establish the octahedron axiom, saying that for every pairof composable morphisms in D(1) there is an associated octahedron diagram. Wefirst show that every coherent pair of composable morphisms X ∈ D([2]) with

underlying diagram f3 = f2 f1 : xf1→ y

f2→ z gives rise to such a diagram. Choosinga slightly different notation than in Definition 4.9, we have to show that there is acommutative diagram

(9.51)

xf1 //

=

yg1 //

f2

0

uh1 //

f4

1

Σx

=

xf3 // z

g2

g3 //

2

vh3 //

g4

3

Σx

Σf1

w

h2

=//

4

w

h4

h2

// Σy

ΣyΣg1

// Σu

such that the rows and colums are distinguished triangles, i.e., such that

(i) xf1→ y

g1→ uh1→ Σx,

(ii) yf2→ z

g2→ wh2→ Σy,

(iii) xf3→ z

g3→ vh3→ Σx, and

(iv) uf4→ v

g4→ wh4→ Σu

are distinguished triangles. Thus, these four distinguished triangles have to be

constructed such that the squares 0 − 4 in (9.51) all commute.This can be achieved by considering coherent diagrams looking like

(9.52)

xf1 //

yf2 //

z //

0

0 // u //

v //

x′ //

0

0 // w // y′ // u′.

Page 117: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 117

In more detail, let B ⊆ [4] × [2] be the full subcategory obtained by removing theobjects (4, 0), (0, 2). The functor [2] → B classifying the composable morphisms

(0, 0) → (1, 0) → (2, 0) factors as a composition [2]i→ A

j→ B of fully faithfulfunctors where A ⊆ B contains the objects (0, 0), (1, 0), (2, 0) and the four objectspopulated by zeros in (9.52). Similarly to Lemma 8.40 one checks that the fully

faithful functors D([2])i∗→ D(A)

j!→ D(B) induce an equivalence

(9.53) D([2]) ' D(B)cof ,

where D(B)cof ⊆ D(B) is the full subcategory spanned by all diagrams

(i) which vanish at the four objects (0, 1), (3, 0), (1, 2), (4, 1) and(ii) which make all squares cocartesian.

Using implicitly certain identifications obtained by similar methods as in §8.5, thisshows that every X ∈ D([2]) with underlying diagram f3 = f2 f1 : x → y → zgives rise to a coherent diagram Q ∈ D(B)cof with underlying diagram

(9.54)

xf1 //

yf2 //

g1

z //

g3

0

0 // uf4 //

vh3 //

g4

Σx //

Σf1

0

0 // wh2

// ΣyΣg1

// Σu.

(Of course the notation is slightly abusive in that the respective squares in theunderlying diagram are no pushout squares anymore; the decoration is only meantto indicate that they come from coherent cocartesian squares.) A ‘diagram chase’in (9.52) combined with the identifications made in (9.54) yields the desired four

distinguished triangles satisfying the above five relations 0 − 4 . To begin with, ifone considers the top three cocartesian squares and glues the two to the right, thenone obtains the first distinguished triangle together with the relation h3f4 = h1,

settling equation 1 . Let us next consider the copy of [2] × [2] in the middle ofthe diagram and compose vertically. This gives the second distinguished triangle

together with the relation g2 = g4g3, hence taking care of 2 . If we again considerthe top three cocartesian squares but this time glue the two squares to the left,then we construct the third distinguished triangle. For the remaining distinguishedtriangle we consider the bottom three cocartesian squares and compose the two to

the right. This also yields the relation h4 = (Σg1)h2, hence making 4 commute.

Finally, in order to show that also the squares 0 and 3 commute it sufficesto consider the middle square in the top and bottom row of (9.52), respectively.This completes the construction of an octahedron diagram for a coherent pair ofcomposable morphisms.

Let us now consider a pair of composable morphisms x→ y → z in the underlyingcategory D(1) and show that it has an associated octahedron diagram. The readereasily checks that it suffices to construct a coherent diagram X ∈ D([2]) such thatthe underlying diagram is naturally isomorphic to x → y → z. If we use thestronger variant of strength (see Remark 9.42) then this is immediate since [2] is afinite, free category. The reader who does not want to use this stronger axiom is

Page 118: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

118 MORITZ GROTH

invited to come up with an independent construction of such a natural isomorphismdia(X) ∼= (x→ y → z), using Definition 9.40 only.

The triangulations of the theorem are referred to as canonical triangulations.As a special case the theorem yields canonical triangulations on underlying cate-gories of strong, stable derivators. We illustrate this by a few examples.

Examples 9.55. (i) If A is an abelian category, such that the homotopy derivatorDA exists (for example, if A is a Grothendieck abelian category; see Defini-tion 6.19), then the underlying category DA(1) ∼= D(A) is a triangulatedcategory. Taking up again Examples 6.20, Theorem 9.44 thus reproduces theclassical triangulations on derived categories of fields, rings, and schemes.

(ii) If we apply Theorem 9.44 to the stable derivator of spectra (6.25), then weobtain the classical triangulation on the stable homotopy category SHC.

(iii) More generally, homotopy derivators of stable model categories and sta-ble ∞-categories are strong, stable derivators. Applied to such examples,Theorem 9.44 reproduces the triangulations on homotopy categories of sta-ble model categories ([Hov99, §7]) and homotopy categories of stable ∞-categories ([Lur11, §1]).

We conclude this subsection by a short remark on the octahedron axiom (relatedto this see also Remark 8.35).

Remark 9.56. Let us recall the main steps of the proof of the octahedron axiom(T4) for the canonical triangulations from Theorem 9.44. Let D be a strong, stablederivator and let X ∈ D([2]) be a coherent pair of composable morphisms. As-sociated to X we obtain a coherent diagram looking like (9.52) satisfying certainexactness and vanishing properties. More precisely, there is an equivalence of cate-gories (9.53), and one might refer to the target category D(B)cof as the categoryof coherent octahedron diagrams. Thus, as long as we stay at the level ofcoherent diagrams, having a pair of composable morphisms is equivalent to havingan octahedron diagram.

Now, given a coherent octahedron diagram X ∈ D(B)cof , making some identifi-cations as indicated in (9.54) we obtain an incoherent octahedron diagram lookinglike (9.51), and this step amounts to a loss of information since we pass to un-derlying diagrams. If we denote the shape of an incoherent octahedron diagram(9.51) by O, then these two steps hence amount to the functorial construction ofoctahedron diagrams

octa: D([2]) ' D(B)cof → D(1)O.

If we start with an incoherent pair of composable morphisms instead, then we firsthave to lift against dia[2] : D([2])→ D(1)[2], and this step destroys the functoriality.

This reminds us of the discussion of the non-functoriality of the constructionof cones and distinguished triangles at the level of triangulated categories; see thebeginning of §4.4. Also in these cases we can obtain functorial such constructionsif we stay at the level of coherent diagrams as made precise by Proposition 8.37and the functor tria : D([1]) → D(1)[3] defined in (9.43). Again, if we insteadbegin with incoherent morphisms, then we first have to lift against the functordia[1] : D([1]) → D(1)[1], resulting in the loss of functoriality. These two similar

Page 119: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 119

observations can be summarized by the following two diagrams

D([1])' //

tria%%

dia[1]

D()cof

D([2])' //

octa%%

dia[2]

D(B)cof

D(1)[1] D(1)[3], D(1)[2] D(1)O,

in which the vertical functors result in a loss of information.

9.7. Exact morphisms of stable derivators. Our next goal is to show thatthe triangulations D(A), A ∈ Cat , of Theorem 9.44 are compatible with restrictionfunctors u∗ : D(B) → D(A) and Kan extension functors u!, u∗ : D(A) → D(B).This actually is a special case of a more general result concerning exact morphismsof (strong) stable derivators.

Recall from Definition 9.28 what it means that a morphism of derivators preservescertain Kan extensions. Specializing this to Kan extensions along πA : A → 1 forsuitable A we obtain the following definition.

Definition 9.57. (i) A morphism of derivators is left exact if it preserves ter-minal objects and cartesian squares.

(ii) A morphism of derivators is right exact if it preserves initial objects andcocartesian squares.

(iii) A morphism of derivators is exact if it is left exact and right exact.

The following is immediate from Remark 9.32 and Proposition A.22.

Example 9.58. A functor F : C → D between complete and cocomplete categoriesis left exact in the sense of Definition A.21 if and only if the induced morphism ofderivators y(F ) : y(C)→ y(D) is left exact in the sense of Definition 9.57.

Lemma 9.59. (i) A morphism of stable derivators is left exact if and only if itis right exact if and only if it is exact.

(ii) Let (F,G) : D E be an adjunction between stable derivators. The mor-phisms F,G both are exact.

Lemma 9.60. Let F : D → E be a morphism of derivators, let u : A → A′, andlet B ∈ Cat. If F preserves left Kan extensions along u, then F also preserves leftKan extensions along 1× u : B ×A→ B ×A′.

Proof. TODO.

Let us recall from Examples 9.26 that the shifting operation also applies tomorphisms of derivators (see (9.27)). A different way of phrasing Lemma 9.60 isby saying that if F : D → E preserves left Kan extensions along u then so does theshifted morphism FB : DB → E B .

Corollary 9.61. Let F : D → E be an exact morphism of stable derivators and letB ∈ Cat. The induced morphism F : DB → E B is again exact.

Proof. This is immediate from Lemma 9.60 and its dual.

Recall from Definition 4.17 that an exact functor between triangulated categoriesis an additive functor endowed with exact structure.

Page 120: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

120 MORITZ GROTH

Proposition 9.62. Let F : D → E be an exact morphism of strong, stable deriva-tors and let A ∈ Cat. The functor FA : D(A) → E (A) can be turned into an exactfunctor with respect to canonical triangulations.

Proof. By Corollary 9.61 we may assume without loss of generality that A = 1.We begin by showing that the underlying functor F1 : D(1) → E (1) is additive.TODO.

Corollary 9.63. Let D be a strong, stable derivator and let u : A→ B be in Cat.The functors u∗ : D(B) → D(A), u! : D(A) → D(B), and u∗ : D(A) → D(B) canbe turned into exact functors with respect to canonical triangulations.

Proof. By Corollary 9.36 there are adjunctions of strong, stable derivators

(u!, u∗) : DA DB and (u∗, u∗) : DB DA,

given by restriction and Kan extension morphisms. It follows from Lemma 9.59that these three morphisms are exact and the underlying functors can hence beturned into exact functors (Proposition 9.62). We note that these underlyingfunctors respectively are the functors u∗ : D(B) → D(A), u! : D(A) → D(B), andu∗ : D(A)→ D(B), concluding the proof.

Remark 9.64. Given a strong, stable derivator D , by Theorem 9.44 there are canon-ical triangulations on D(A), A ∈ Cat , and by Proposition 9.62 the restriction func-tors u∗ : D(B) → D(A) can be turned into exact functors. There is also such anobservation for natural transformations.

To make this precise, we recall the following definition. Let T , T ′ be trian-gulated categories and let (F, σ), (F ′, σ′) be exact functors T → T ′. A naturaltransformation α : F → F ′ is exact if the diagram

FΣσ∼=//

αΣ

ΣF

Σα

F ′Σσ′

∼= // ΣF ′

commutes. Given a strong, stable derivator D , functors u, v : A→ B, and a trans-formation α : u→ v, one can show that the natural transformation α∗ : u∗ → v∗ isexact in this sense.

A concise way of summarizing this is as follows. Triangulated categories, exactfunctors, and exact natural transformations are assembled in a (again very large)2-category T riaCAT of triangulated categories. Every strong, stable derivatorD : Catop → CAT admits a lift against the forgetful 2-functor T riaCAT → CAT ,

T riaCAT

Catop

D//

∃D88

CAT .

10. Towards abstract representation theory

10.1. Morita equivalences.

10.2. Derived equivalences.

Page 121: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 121

10.3. Strong stable equivalences. Given a ring R we denote by Mod(R) theabelian category of R-modules. Attributing credit to [?] there is the followingdefinition.

Definition 10.1. Two rings R,S are Morita equivalent if there is an equivalenceof categories Mod(R) ' Mod(S).

It can be shown that such an equivalence is necessarily given by the tensor prod-uct with an (R,S)-bimodule. In representation theory people are often interestedin derived categories of rings and algebras; see for example [Hap88, AHHK07]. Thederived analogue of Definition 10.1 is the following.

Definition 10.2. Two rings R,S are derived equivalent if there is an exact

equivalence D(R)∆' D(S).

We now specialize to particular types of rings, namely path algebras of finitequivers. Let Q be a finite quiver and let k be a field. We recall that the pathalgebra kQ is constructed as follows. The underlying k-vector space has as basisthe set of all paths in Q. More precisely, if Qn, n ≥ 0, denotes the set of all pathsof length precisely n, then we set

kQ =⊕n≥0

kQn.

The multiplication is the bilinear extension of the concatenation of two such paths(the product of two basis elements is set to be zero if the source and targets donot match appropriately). A key feature about this k-algebra is that there is anequivalence of categories

(10.3) Mod(kQ) ' Mod(k)Q.

Here, Q also denotes the free category associated to the oriented graph Q, andrepresentations can hence equivalently be specified by Q-shaped diagrams of vectorspaces. Of course, instead of working over a field k we can consider a ring R andobtain a similarly defined path algebra RQ and a corresponding equivalence (10.3).

Definition 10.4. Let Q,Q′ be finite quivers and let R be a ring. The quivers Q,Q′

are derived equivalent over R if there is an exact equivalence

D(RQ)∆' D(RQ′).

Note that in principle the property of being derived equivalent might depend onthe ring R. Classically, such derived equivalences are often established over fields.To mention one such result we recall the following definition.

Definition 10.5. A quiver is acyclic if it admits no non-trivial, oriented loops. Aquiver is a tree if it admits no non-trivial, unoriented loops.

A reorientation of a quiver Q is a quiver Q′ obtained from Q by changing theorientation of some edges. For example the quivers

1→ 2→ 3 and 1← 2→ 3

are reorientations of each other. We leave it to the reader to come up with a directargument showing that the corresponding path algebras over fields are not Moritaequivalent. Nevertheless, it turns out that these quivers are derived equivalent overarbitrary fields. In fact, this is only a special case of the following more generalresult of Happel.

Page 122: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

122 MORITZ GROTH

Theorem 10.6. Let T be a finite tree and let T ′ be a reorientation of T . Thequivers T, T ′ are derived equivalent over arbitrary fields.

By a purely combinatorial argument the task is reduced to the following situa-tion. Let T be a finite tree and let q0 ∈ T be a source, i.e., a vertex q0 such thatall edges adjacent to it start at q0. The reflected tree T ′ of T at the source q0 isobtained by changing the orientation of all edges starting at q0 and thereby turningthe source q0 into a sink. Now, in order to establish Theorem 10.6 it suffices to showthat finite trees and their reflections at sources and sinks are derived equivalent overarbitrary fields.

It turns out that such derived equivalences even exist more generally. The notionof a reflection at a soure or a sink makes perfectly well sense for arbitrary quivers.And in the case of acyclic quivers there is the following result of Happel [Hap87].

Theorem 10.7. Let Q be a finite, acylic quiver and let Q′ be the reflection of Q ata source or a sink. The quivers Q,Q′ are derived equivalent over arbitrary fields.

In fact, such derived equivalences can be explicitly constructed as derived reflec-tion functors in the sense of Bernsteın, Gel′fand, and Ponomarev (see [BGP73] and[Hap87]). Using the formalism of derivators, one can show that such equivalencesexist in much broader generality. To this end we begin by a minor reformulationof Definition 10.4. As a special case of Lemma 3.10 we observe that for a finitequiver Q there is an isomorphism of categories

Ch(Mod(R)Q) ∼= Ch(R)Q

which sends quasi-isomorphisms to levelwise quasi-isomorphisms. If we combinethis with a degreewise application of (10.3) then we obtain an equivalence of cate-gories

Ch(RQ) ' Ch(R)Q,

again sending quasi-isomorphisms to levelwise quasi-isomorphisms. As an upshot,if two quivers Q,Q′ are derived equivalent over a ring R, then there is the tophorizontal and the two vertical equivalences

(10.8)

D(RQ)' //

'

D(RQ′)

'

Ho(Ch(R)Q) '// Ho(Ch(R)Q

′),

hence inducing the bottom equivalence. Thus, to put it in words, the fact that twoquivers are derived equivalent over R means that if we pass to the correspondinghomotopy theories of diagrams in Ch(R) then the associated homotopy categoriesare equivalent. In some cases one can obtain similar results for homotopy theoriesof representations in arbitrary stable homotopy theories in the sense of the followingdefinition (see [GS14c]).

Definition 10.9. Two quivers Q,Q′ are strongly stably equivalent if for everystable derivator D there is an equivalence of derivators

DQ ' DQ′

which is natural with respect to exact morphisms of derivators.

We include a few comments about the definition.

Page 123: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 123

Remark 10.10. (i) This definition of course makes perfectly well sense for smallcategories A,A′ as well, but here we focus on quivers.

(ii) If two finite quivers Q,Q′ are strongly stably equivalent then they are alsoderived equivalent over arbitrary rings. In fact, considering the special caseof the stable derivator DR of a ring (Examples 9.2), we obtain an equivalence

of derivators DQR ' DQ′

R . Passing to underlying categories this yields an

equivalence of categories DQR (1) ' DQ′

R (1) which is exact with respect tothe canonical triangulations of Theorem 9.44. Combining this with (10.8) wededuce that Q,Q′ are derived equivalent over R. (One can also invoke the

equivalence of derivators DQR ' DRQ; see Examples 7.31.)

(iii) However, the notion of being strongly stably equivalent is a priori much morerestrictive for the following three reasons.(a) First, simply by choosing other examples of stable derivators (Exam-

ples 9.2) and passing to underlying categories we obtain further exactequivalences of triangulated categories of representations. For example,considering the stable derivator DX of a scheme X we obtain equiva-lent derived categories of representations in quasi-coherent OX -modules.Similarly, if we pass to the stable derivator DA of a differential-gradedalgebra A, then we get exact equivalences of triangulated categories ofdifferential-graded representations. Moreover, there are further such re-sults for spectral representations, simply by passing to the stable deriva-tor of spectra or the stable derivator DE of a —say— symmetric ringspectrum E. Still more generally, we can consider homotopy derivatorsassociated to arbitrary stable model categories or stable ∞-categories,concluding that there are exact equivalences of triangulated categories ofabstract representations of the quivers under consideration. (We againrefer the reader to [GS14c, Examples 5.5] for a fairly long list of addi-tional examples.)

(b) Second, by the very definition we ask for equivalences of derivators asopposed to merely asking for exact equivalences of the underlying trian-gulated categories. Recall that, in general, it is a stronger statement tohave an equivalence of homotopy theories and not only an equivalence ofhomotopy categories. Thus, there are equivalences of homotopy theoriesof abstract representations of the quivers under consideration.

(c) Third, these equivalences are compatible with respect to exact mor-phisms. In fact, by the very definition suitably restricted corepresentedcopresheaves (−)Q, (−)Q

′are asked to be equivalent. In particular, the

various equivalences listed above are compatible with (derived) restric-tions of scalars, inductions and coinductions of scalars, (derived) tensorand hom functors, as well as localization and colocalization functors.

With these remarks in mind, let us mention the following generalization of The-orem 10.6; see [GS14b].

Theorem 10.11. Let T be a finite tree and let T ′ be a reorientation of T . Thequivers T, T ′ are strongly stably equivalent.

In §10.4 we give a sketch proof in the special case of Dynkin quivers of type A.Similar techniques can also be used to take care of trees with at most one branchingpoint, i.e., vertex of valences at least three. With more refined techniques one can

Page 124: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

124 MORITZ GROTH

mimic the classical construction of reflection functors in the sense of Bernsteın,Gel′fand, and Ponomarev [BGP73] in order to obtain the strong stable equivalencesof Theorem 10.11; see [GS14b]. Theorem 10.7 can be refined to yield the followingresult.

Theorem 10.12. Let Q be a finite, acylic quiver and let Q′ be the reflection of Qat a source or a sink. The quivers Q,Q′ are derived equivalent over arbitrary fields.

In fact, in [GS15] there will be the construction of abstract reflection functorsestablishing this result. The techniques of that paper apply even more generally,yielding additional examples of strong stable equivalences. A further specializingto derivators of rings and fields gives rise to new examples of derived equivalencesof category algebras, generalizing Theorem 10.7.

10.4. Abstract tilting theory of An-quivers. In this subsection we give a sketchproof that different Dynkin quivers of type A of the same length are strongly stablyequivalent. These observations are closely related to the construction of canonicalhigher triangulations on (strong) stable derivators.

To begin with let us recall that a Dynkin quiver of type A is simply a finitezig-zag, i.e., an oriented graph obtained by endowing the unoriented graph

1 2 . . . n− 1 n

with an arbitrary orientation. More precisely, fixing the length n this way we obtainthe An-quivers. Thus, the goal is to see that all An-quivers are strongly stablyequivalent. As a warmup let us consider the four different An-quivers

1→ 2→ 3, 1← 2→ 3, 1→ 2← 3, and 1← 2← 3.

As a preparation for the proofs in this subsection we suggest the reader to againtake a look at Remark 9.39.

Proposition 10.13. All A3-quivers are strongly stably equivalent.

Proof. Let D be a stable derivator and let us show that the relevant stable derivatorsof representations in D are naturally equivalent. To begin with, since the twocategories (1→ 2→ 3) and (1← 2← 3) are isomorphic these quivers are stronglystably equivalent.

We next show that the quivers p= (1← 2→ 3) and y = (1→ 2← 3) are stronglystably equivalent. To this end, we consider the functors ip : p→ and iy : y → and the associated fully faithful Kan extension morphisms (ip)! : Dp → D and(iy)∗ : Dy → D. The essential image of (ip)! consists precisely of those X ∈ D(A)such that every Xa ∈ D(), a ∈ A, is cocartesian (Corollary 8.23) and similarly forthe essential image of (iy)∗. Since D is stable these two essential images coincideand will be denoted by D,ex ⊆ D. As an upshot we obtain equivalences ofderivators

Dp ' D,ex ' Dy,

showing that these two A3-quivers are strongly stably equivalent.It remains to show that the quivers [2] ∼= (1 → 2 → 3) and p= (1 ← 2 → 3)

are strongly stably equivalent. Let X = (X0 → X1 → X2) ∈ D [2] be an abstractrepresentation of [2]. Then such an equivalence is obtained by passing to the cofiber

Page 125: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 125

of (X0 → X1). In more detail, us consider the full subcategory B ⊆ [2]×[1] obtainedby removing the final object (2, 1),

(0, 0) //

(1, 0) //

(2, 0)

(0, 1) // (1, 1).

The functor [2]→ B classifying (0, 0)→ (1, 0)→ (2, 0) factors as a composition offully faithful functors

[2]i1→ A1

j1→ B,

where A1 is obtained from B by also removing the object (1, 1). Associated tothese functors there are fully faithful Kan extension morphisms (i1)∗ : D [2] → DA1

and (j1)! : DA1 → DB . Since i1 is a sieve, the right Kan extension morphism (i1)∗is right extension by zero (Corollary 8.6). Moreover, the essential image of (j1)! iseasily checked to consist of precisely those coherent diagrams such that the obvioussquare is cocartesian. As an upshot, (j1)!(i1)∗ : D [2] → DB induces an equivalenceonto the full subderivator of DB spanned by all Z satisfying the following exactnessproperties.

(i) The diagram Z makes the obvious square cocartesian.(ii) The diagram Z vanishes at the lower left corner, Z(0,1)

∼= 0.

Let us now consider the functor p→ B classifying (1, 1) ← (1, 0) → (2, 1). Thisfunctor factors as a composition of fully faithful functors

pi2→ A2

j2→ B,

where A2 is obtained from B by also removing the object (0, 0). Using simi-lar arguments as above and an additional cofinality argument, one checks that(j2)∗(i2)! : Dp → DB induces an equivalence onto the full subderivator of DB

spanned by all Z satisfying the following exactness properties.

(i) The diagram Z makes the obvious square cartesian.(ii) The diagram Z vanishes at the lower left corner, Z(0,1)

∼= 0.

Note that since D is stable in both cases the essential images coincide (Corol-lary 8.23), and let us denote it by DB,ex ⊆ DB . As an upshot, the above construc-tions yield equivalences of derivators

D [2] ' DB,ex ' Dp,

showing that the A3-quivers [2] and p are strongly stably equivalent. Since beingstrongly stably equivalent is an equivalence relation, this concludes the proof.

Similar techniques can be used to also show that all An-quivers for a fixed n ≥ 1are strongly stably equivalent. It will pay off to approach these constructions a bitmore systematically. To motivate the construction, let us emphasize that in the caseof A3-quivers, once the dust settles, everything essentially boiled down to expandingour representations by passing to (co)fibers and adding certain bicartesian squares.In fact, similar arguments were already used in the proof of Theorem 9.44, namelyin the proof of the octahedron axiom by means of the construction of coherentoctahedron diagrams; see (9.54) and Remark 9.56.

It turns out that one convenient strategy consists of expanding coherent diagramslooking like (9.54) doubly infinitely by adding more zeros and bicartesian squares

Page 126: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

126 MORITZ GROTH

as we explain next. From now on, let us write ~An = [n−1] for the linearly oriented

An-quiver. Given an abstract representation X = (x → y → z) ∈ D~A3 , suitable

combinations of Kan extensions yield a coherent diagram looking like

(10.14)

· · · 0

##

0

0

!!

0

""

· · ·<<

!!

Ωw

==

z

@@

Σx

<<

""

Σu

<<

""· · · Ωv

<<

""

y

AA

v

>>

Σy

<<

""

· · ·==

""

x

>>

!!

u

AA

w

<<

""

Σz

<<

""· · · 0

;;

0

@@

0

==

0

<<

· · ·The defining exactness properties of such coherent diagrams are that they vanishon the boundary stripes and that all squares are bicartesian. We denote the cor-responding shape by M3 ∈ Cat and note that there is the fully faithful inclusion

functor ~A3 →M3 pointing at the objects populated by (x→ y → z). More gener-ally, for n ≥ 1 there is a similar category Mn ∈ Cat and a fully faithful inclusion

functor ~An → Mn. Note that this inclusion functor factors as a composition offully faithful functors

~Ani1→ B1

i2→ B2i3→ B3

i4→Mn,

where the respective full subcategories of Mn are obtained as follows.

(i) The category B1 is obtained from the image of ~An → Mn by adjoining allobjects on the boundary stripes sitting below this image.

(ii) The category B2 is obtained from B1 by adding the remaining objects belowthis image.

(iii) The category B3 is obtained from B2 by adjoining the remaining objects onthe boundary stripes.

Given a derivator D , associated to these inclusions there are the following fullyfaithful Kan extension morphisms,

(10.15) D~An (i1)∗→ DB1

(i2)!→ DB2(i3)!→ DB3

(i4)∗→ DMn .

Theorem 10.16. Let D be a stable derivator and let n ≥ 1. The fully faithful Kan

extension morphisms (10.15) induce an equivalence of derivators D~An ' DMn,ex

where DMn,ex ⊆ DMn is the full subderivator spanned by all coherent diagramsZ ∈ DMn satisfying the following exactness properties.

(i) The diagram Z vanishes on all objects in the boundary stripes.(ii) The diagram Z makes all squares bicartesian.

Moreover, this equivalence is natural with respect to exact morphisms of derivators.

Proof. The idea of the proof is obvious. The first morphism in (10.15) is rightextension by zero, the second one amounts to adding cocartesian squares in onedirection, the third morphism is left extension by zero, and the fourth morphismadds cartesian squares in the other direction. Thus, in stable derivators the essentialimage is as claimed. For a detailed proof see [GS14a, §4].

One reason for us to consider these shapes Mn is the following observation.

Besided the inclusions ~An → Mn there are many additional embeddings Q → Mn

Page 127: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 127

for arbitrary An-quivers Q of the same length. And related to these embedding oneconstructs, for every stable derivator D , similar equivalences of derivators

DQ ' DMn,ex,

which again are natural with respect to exact morphisms of stable derivators. Notethat in this case as well as in the equivalence of Theorem 10.16 an inverse equiva-lence is simply given by restriction along the respective embedding.

To illustrate this, in the case of n = 3, let us consider the inclusion p→ M3

pointing at the objects populated by (u← y → z) in (10.14). Similar constructionsas in (10.15) induce natural equivalences Dp ' DM3,ex. If we combine this withthe equivalence of Theorem 10.16, then we obtain natural equivalences

D~A3 ' DMn,ex ' Dp,

showing that ~A3 and p are strongly stably equivalent. The reader easily checks thatthis reproduces the strong stable equivalence constructed in the proof of Proposi-tion 10.13.

Theorem 10.17. For every fixed n ≥ 1 all An-quivers are strongly stably equiva-lent.

Proof. Let Q,Q′ be two An-quivers of the same length and let D be a stable deriva-tor. Then we can choose embeddings iQ : Q → Mn and iQ′ : Q

′ → Mn as above.Related to these embedding, the abovementioned more general version of Theo-rem 10.16 yields natural equivalences

DQ ' DMn,ex ' DQ′ ,

showing that Q,Q′ are strongly stably equivalent.

An additional reason for us to consider the shapes Mn ∈ Cat is that they showup in the definition of strong triangulations in the sense of Maltsiniotis; see [Mal05].The rough idea is as follows. A strong triangulation on an additive category Aconsists of a self-equivalence Σ: A ' A and chosen classes of distinguished n-triangles. Here, an n-triangle in A is a diagram Mn → A together with suit-ably natural isomorphisms making precise that ‘suspensions show up as intended’;see (10.14) to get an idea. This structure has to satisfy a list of compatibilityaxioms, including relations between distinguished n-triangles for varying n.

Thus, 2-triangles look like a more refined version of classical triangles and 3-triangles like refined octahedron diagrams; see Remark 9.56. The distinguishedn-triangles for n ≥ 4 can be thought of as ‘higher octahedron diagrams’. A preciseformulation and a proof of the following result can be found in [GS14a, §13].

Theorem 10.18. Let D be a strong stable derivator and let A ∈ Cat. The categoryD(A) admits a canonical strong triangulation.

As in the case of triangulations (see Proposition 9.62 and Remark 9.64) one canshow that these strong triangulations are natural with respect to exact morphismsand that a strong stable derivator D : Catop → CAT lifts against the forgetfulfunctor from a certain 2-category of strong triangulated categories, exact functors,and exact transformation. We again refer the reader to [GS14a, §13] for moredetails.

Page 128: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

128 MORITZ GROTH

10.5. Universal tilting modules. Will be added soon.Topics to be briefly mentioned:

(i) Universality of spaces and spectra(ii) implications for (co)limits: cofinality, characterization of htpy exact squares:

depends only on spaces, ends and coends, (co)simplicial bar constructions(iii) bicategory of profunctors(iv) induced canceling action, universal tilting modules

We include a short remark which is intended for the reader who is familiar withsimplicial sets. It can be safely skipped without any loss of continuity.

Remark 10.19. The characterization of final functors given in the above propositionsounds like a geometric statement. To make it precise, let us recall that there isthe nerve functor associating a simplicial set NA to every small category A. Theabove statement can be reformulated by saying that u : A→ B is final if and onlyif all the nerves N(b/u), b ∈ B, are non-empty and have a vanishing π0.

There is a similar theorem for functors u : A → B which tell us that A-shapedand B-shaped homotopy colimits are the same. In that case, the combinatorialcharacterization of such functors is that all nerves N(b/u) are weakly contractiblesimplicial sets.

Appendix A. Some category theory

In this section we review some basics from category theory and refer the readerto [ML98, Bor94a, Bor94b] for more details.

A.1. Adjunctions. Let us begin by recalling that an adjunction (L,R) : C Dbetween two categories C,D consists of two functors L : C → D and R : D → Ctogether with the choice of an isomorphism

(A.1) φ = φc,d : homD(Lc, d) ∼= homC(c,Rd)

which is natural in c ∈ C and d ∈ D. There are various equivalent ways of encodingsuch a datum, one of them uses the notion of adjunction (co)units. Specializing(A.1) to objects d = Lc and applying φ to the identity morphisms 1: Lc → Lc weobtain a natural transformation η : 1 → RL, the (adjunction) unit. Dually, ap-plying φ−1 to identity morphisms 1: Rd→ Rd we obtain a natural transformationε : LR→ 1, the (adjunction) counit.

These two transformations are not unrelated since they are constructed using theisomorphisms (A.1) and their inverses. First, let us note that we can reconstruct φfrom R and η. In fact, the naturality of φ implies that for every f : Lc → d thereis a commutative diagram

homD(Lc, Lc)φ//

f∗

homC(c,RLc)

(Rf)∗

homD(Lc, d)φ// homC(c,Rd).

Tracing 1: Lc → Lc through the diagram and using the definition of η we seethat φ(f) = (Rf)∗(ηc) = Rf ηc. Similarly, the inverse isomorphism φ−1 can be

Page 129: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 129

reconstructed from L and ε, and we have the commutative diagram

(A.2)

homC(RLc,Rd)

η∗

((

homD(Lc, d)

R

66

φ//homC(c,Rd)

φ−1

oo

Lvv

homD(Lc, LRd).

ε∗

hh

In particular, starting with any morphism Lc → d or c → Rd an application ofthe four maps in the clockwise direction gives us back the same morphism. If wespecialize this to the identity morphisms 1: Lc → Lc and 1: Rd → Rd, then wededuce the following two relations between η and ε,

(A.3)

LLη//

=++

LRL

εL

RηR//

=++

RLR

L, R,

called the triangular identities. It can be shown that an adjunction is determinedby the datum of two functors L : C → D, R : D → C and two natural transformationsη : 1→ RL, ε : LR→ 1 such that the triangular identities are satisfied.

In this book we denote adjunction units generically by η and adjunction counitsby ε. If we want to emphasize that there are different adjunctions in a given context,then we use obvious notational variations like η′, ε′. We recall the following factabout adjunctions.

Lemma A.4. Let (L,R) : C D be an adjunction.

(i) The left adjoint L is fully faithful if and only if the unit η : 1 → RL is anatural isomorphism. An object d ∈ D lies in the essential image of L if andonly if the counit εd : LRd→ d is an isomorphism.

(ii) The right adjoint R is fully faithful if and only if the counit ε : LR → 1 is anatural isomorphism. An object c ∈ C lies in the essential image of R if andonly if the unit ηc : c→ RLc is an isomorphism.

(iii) The adjunction (L,R) is an adjoint equivalence if and only if L and R arefully faithful if and only if η and ε are natural isomorphisms.

It is immediate from (A.1) that adjunctions can be composed. Given adjunc-tions (L′, R′) : C D and (L′′, R′′) : D E then composing the respective ad-junction isomorphisms we obtain an adjunction (L,R) : C E with L = L′′L′ andR = R′R′′. The adjunction unit and counit of this composite adjunction can becalculated as

η : 1η′→ R′L′

R′η′′L′→ R′R′′L′′L′ and ε : L′′L′R′R′′L′′ε′R′′→ L′′R′′

ε′′→ 1.

Natural transformations between left adjoint functors are closely related to nat-ural transformations between associated right adjoint functors.

Lemma A.5. Let (L,R), (L′, R′) : C D be adjunctions and let α : L → L′ be anatural transformation. There is a unique natural transformation α′ : R′ → R such

Page 130: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

130 MORITZ GROTH

that the following diagram commutes

(A.6)

homD(Lc, d)∼=φ//

OO

α∗

homC(c,Rd)OO

(α′)∗

homD(L′c, d) ∼=

φ′// homC(c,R

′d).

This defines a bijection between natural transformations α : L → L′ and naturaltransformations α′ : R′ → R.

Proof. Let us choose c = R′d and trace the counit ε′d : L′R′d → d through thediagram. This implies α′d = φ(ε′d αR′d), showing that there is at most one suchnatural transformation. Note that using the description of the adjunction isomor-phism φ in terms of R and η as in (A.2) we obtain α′ = Rε′ RαR′ ηR′ and thisshows that we actually constructed a natural transformation α′.

Natural transformations α and α′ making (A.6) commute are conjugate or totalmates. The proof and its converse show that conjugate natural transformationsα, α′ determine each other by means of the relations

(A.7) α′ = Rε′ RαR′ ηR′ : R′ → RLR′ → RL′R′ → R

and

(A.8) α = εL′ Lα′L′ Lη′ : L→ LR′L′ → LRL′ → L′.

The uniqueness of conjugate transformations implies that the construction is com-patible with compositions and identity transformations.

Corollary A.9. If α : L → L′ and α′ : R′ → R are conjugate natural transforma-tions, then α is a natural isomorphism if and only if α′ is a natural isomorphism.

A more systematic way of putting this is as follows. Given two categories C,D,let LAdj(C,D) be the category of left adjoint functors from C to D. An object inLAdj(C,D) is a left adjoint L : C → D and morphisms are natural transformationsbetween left adjoints. Dually, we define the category RAdj(C,D).

Corollary A.10. Let C and D be categories. There is an equivalence of categoriesLAdj(C,D) ' RAdj(D, C)op.

Proof. We describe a functor LAdj(C,D) → RAdj(D, C)op. For each left adjointfunctor L : C → D we choose a right adjoint R : D → C. Given these choices, forevery pair of left adjoint functors L,L′ the passage to conjugate transformationsdefines a bijection between natural transformations L→ L′ and R′ → R, which, asalready observed, is compatible with compositions and identities. Thus, we obtaina fully faithful functor LAdj(C,D) → RAdj(D, C)op. Clearly this functor is alsoessentially surjective and hence an equivalence.

Thus, the specification of a left adjoint is essentially equivalent to the specifi-cation of a right adjoint. We will later need the following more precise statementalong these lines.

Page 131: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 131

Lemma A.11. Let (L,R, η, ε) : C D and (L′, R, η′, ε′) : C D be adjunctionswith the same right adjoint R. There is a unique natural isomorphism α : L → L′

making the following diagrams commute

1η//

η′ ))

RL

∼= Rα

LRε //

αR ∼=

1

RL′, L′R.ε′

JJ

Proof. The existence of a unique natural isomorphism α making the diagram onthe left commute is immediate from the definition of units as certain initial objects.In order to check that also the remaining triangle commutes it is enough to showthat the two compositions have the same image under the natural isomorphismφ : homD(LRd, d) ∼= homC(Rd,Rd). By definition of the adjunction counit ε wehave to show that φ(ε′ αR) = 1. By (A.2) we know that φ(ε′ αR) is given by

RηR→ RLR

RαR→ RL′RRε′→ R.

Since we already showed that Rα η = η′, we see that φ(ε′ αR) = Rε′ η′R whichis the identity 1: R→ R by a triangular identity (A.3).

A.2. Limits and colimits. We assume the reader to have some basic familiar-ity with limits and colimits, including a discussion of terminal objects, products,pullbacks and the dual notions of initial objects, coproducts, and pushouts. Nev-ertheless we recall some key definitions and results about (co)limits, mainly toestablish notation and to prepare the ground for homotopical generalizations asdiscussed in the main body of the text.

Definition A.12. Let A be a small category, let C be a category, and let X : A→ Cbe a functor. A cone on X is a pair (l, α) consisting of an object l ∈ C andmorphisms αa : l → Xa, a ∈ A, such that for every morphism f : a → a′ in A thediagram

lαa //

αa′

Xa

f∗

Xa′

commutes. A morphism of cones (l, α)→ (l′, α′) is a morphism l→ l′ in C suchthat

l

αa

l′α′a

// Xa

commutes for every a ∈ A.

With the obvious composition law and identity morphisms this defines the cat-egory of cones on X. Passing to the formal dual, there is the notion of a cocone(c, β) on X : A→ C. In this case the maps βa : Xa→ c are compatible in the sense

Page 132: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

132 MORITZ GROTH

that

Xaβa //

f∗

c

Xa′βa′

>>

commutes for every morphism f : a → a′. Defining morphisms of cocones in theobvious way, associated to a functor X : A→ C there are the two categories

cone(X), cocone(X) ∈ CAT ,

both coming with obvious forgetful functors cone(X)→ C and cocone(X)→ C.Universal examples of (co)cones deserve particular names.

Definition A.13. Let A be a small category, let C be a category, and let X : A→ Cbe a functor.

(i) A limit of X is a terminal object limAX of cone(X).(ii) A colimit of X is an initial object colimAX of cocone(X).

Thus, to emphasize, such a limit is a pair consisting of an underlying object,also denoted by limAX, together with a universal cone which we also refer to as alimiting cone. Similarly, we abuse notation and write colimAX for the underlyingobject of a colimit and refer to the corresponding cocone as a colimiting cocone.

As always with universal construction, if (co)limits exist, then they are unique upto unique isomorphisms and this uniqueness implies functoriality. In the followingproposition we write ∆ = ∆A : C → CA for the diagonal functor which associatesto x ∈ C the constant diagram ∆(x) : A→ C with value x.

Proposition A.14. Let A be a small category and let C be a category.

(i) If every X : A→ C has a limit, then the assignment X 7→ limAX extends toa limit functor limA : CA → C which is right adjoint to ∆A,

(∆A, limA) : C CA.(ii) If every X : A→ C has a colimit, then the assignment X 7→ colimAX extends

to a colimit functor colimA : CA → C which is left adjoint to ∆A,

(colimA,∆A) : CA C.

For particular choices of small categories A, there are special names for theresulting limits and colimits, namely,

(i) final objects and initial objects if A is empty,(ii) products and coproducts if A is discrete (all morphisms are identities),(iii) equalizers and coequalizers if A = (0⇒ 1), and(iv) pullbacks for limits over A = ((0, 1) → (1, 1) ← (1, 0)) and pushouts for

colimits over A = ((0, 1)← (0, 0)→ (1, 0)).

We next recall that these are the basic building blocks for arbitrary (co)limits inthe following precise sense. A category is complete or cocomplete if it admitslimits or colimits of all small diagrams, respectively.

Proposition A.15. (i) A category is complete if and only if it admits equalizersand (small) products.

(ii) A category is cocomplete if and only if it admits coequalizers and (small)coproducts.

Page 133: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 133

Proof. By duality it is enough to take care of the first statement. The task is toshow that an arbitrary limit can be obtained by combining products and equalizers.To this end, let A be a small category and X : A → C such that C admits smallproducts and equalizers. We consider the following pair of parallel morphisms

d0, d1 :∏a∈A

Xa→∏

f : a0→a1

Xa1

in C. Using the universal property of the product∏f : a0→a1 Xa1, in order to define

the morphisms d0, d1 it suffices to specify the respective compositions with theprojections onto the factor Xa1 associated to an arbitrary f : a0 → a1. In the caseof d0 this composition is chosen to be∏

a∈AXa→ Xa0

f∗→ Xa1

while in the case of d1 we simply take the projection∏a∈A

Xa→ Xa1.

The underlying object of the limit limAX is now defined to be the equalizer ofthese two morphisms,

(A.16) limAX = eq

( ∏a∈A

Xa⇒∏

f : a0→a1

Xa1

).

Using the universal cone belonging to this equalizer, we can define the universalcone of limAX to have components

limAX →

∏a∈A

Xa→ Xa′, a′ ∈ A,

where the second morphism is the projection onto the a′-component. We leave itto the reader to verify that this is a limiting cone for X.

In the case of colimits the key step consists of forming the coequalizer

(A.17) colimAX = coeq( ∐f : a0→a1

Xa0 ⇒∐a∈A

Xa)

of two similarly defined morphisms d0, d1 :∐f : a0→a1 Xb→

∐a∈AXa. The details

are left to the reader.

This proposition admits a few variants, one of which we include here. A categoryis finitely complete if it admits finite limits, i.e., limits of diagrams defined onfinite categories (recall that a category is finite if it has finitely many objects andmorphisms only).

Corollary A.18. The following are equivalent for a category C.

(i) The category C is finitely complete.(ii) The category C admits finite products and equalizers.

(iii) The category C admits terminal objects and pullbacks.

Proof. The equivalence of the first two statements follows as above, so that it issuffices to show that terminal objects and pullbacks generate finite products andequalizers. Since equalizers are special cases of pullbacks it remains to constructfinite products. The reader easily checks that products of two objects are obtained

Page 134: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

134 MORITZ GROTH

by forming pullbacks over terminal objects, and the case of finite products hencefollows by induction.

We conclude this section by a short discussion of the preservation of (certain) lim-its by functors. Let us consider a functor X : A→ C admitting a colimit colimAXand let F : C → D be a functor such that also F X : A → D admits a colimitcolimA F X. If we apply F to the colimiting cocone of X, then for every f : a→ a′

in A there is a commutative diagram

F (Xa)

// F (colimAX)

F (Xa′)

88

in D, which is to say that we obtain a cocone on F X. The universality of thecolimiting cocone of F X implies that there is a unique morphism

(A.19) colimA(FX)→ F (colimAX)

compatible with these two cocones. We refer to this morphism and a dually definedmorphism

(A.20) F (limAX)→ limA(FX)

as the canonical morphisms.Let F : C → D be a functor between categories admitting colimits of shape A. We

say that F preserves colimits of shape A if for every X : A → C the canonicalmorphism colimA(FX)→ F (colimAX) is an isomorphism.

Definition A.21. (i) A functor between complete categories is continuous ifit preserves all limits.

(ii) A functor between cocomplete categories is cocontinuous if it preserves allcolimits.

(iii) A functor between finitely complete categories is left exact if it preservesfinite limits.

(iv) A functor between finitely cocomplete categories is right exact if it preservesfinite colimits.

(v) A functor between finitely complete and finitely cocomplete categories is ex-act if it is left exact and right exact.

Proposition A.15 and Corollary A.18 have the following variants for functors.

Proposition A.22. (i) A functor between complete categories is continuous ifand only if it preserves equalizers and products.

(ii) A functor between finitely complete categories is left exact if and only if itpreserves equalizers and finite products if and only if it preserves pullbacksand terminal objects.

The following lemma provides a large class of (co)continuous functors.

Lemma A.23. (i) A right adjoint functor between complete categories is con-tinuous and hence left exact.

(ii) A left adjoint functor between cocomplete categories is cocontinuous and henceright exact.

A.3. Basic 2-categorical terminology. Will be added soon.

Page 135: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 135

Appendix B. Examples of derivators

B.1. Represented derivators. Will be added soon.

B.2. Homotopy derivators of model categories. In this subsection we includea detailed proof that combinatorial model categories have underlying homotopyderivators. The proof relies on the non-trivial results that diagram categories incombinatorial model categories can be endowed with both the projective and theinjective model structures.

Let M be a Quillen model category and let A ∈ Cat . In general, it is not truethat we can endow the diagram categoryMA with a model structure such that theweak equivalences are precisely the levelwise weak equivalences, i.e., those naturaltransformations f : X → Y : A→M such that all components fa : Xa → Ya, a ∈ A,are weak equivalences in M. However, if one imposes additional conditions on M,then there are two such model structures for arbitrary A ∈ Cat . We begin by theone which is adapted to the study of homotopy colimits and homotopy left Kanextensions.

Definition B.1. LetM be a model category, let A ∈ Cat , and let X,Y : A→M.

(i) A morphism f : X → Y is a projective fibration if it is levelwise fibration.(ii) A morphism f : X → Y is a projective weak equivalence if it is a levelwise

weak equivalence.(iii) A morphism f : X → Y is a projective cofibration if it has the LLP with

respect to acyclic projective fibrations.

If these three classes define a model structure onMA, then we refer to it as theprojective model structure. For the notion of a cofibrantly generated modelstructure we refer the reader to [Hov99]. Let us only mention that most modelstructure showing up in nature (like homotopy theory, homological algebra, andhigher category theory) enjoy this property, so that the following result often ap-plies.

Theorem B.2. Let M be a cofibrantly generated model category and let A ∈ Cat.The projective model structure exists on MA.

The category MA endowed with the projective model structure will be denotedby MA

proj. Recall that the category of simplicial sets can be endowed with thecofibrantly generated Kan–Quillen model structure. In this case, the correspondingprojective model structures on diagram categories were first established in [].

Projective model structures have good functorial properties. For the study ofhomotopy left Kan extensions we observe the following.

Proposition B.3. LetM be a cofibrantly generated model category and let u : A→B be in Cat. The adjunction (u!, u

∗) : MAproj MB

proj is a Quillen adjunction. In

particular, (colimA,∆A) : MAproj M is a Quillen adjunction.

Proof. We have to show that u∗ : MB → MA preserves projective fibrations andacyclic projective fibrations which is immediate since both classes are defined lev-elwise.

Passing to left derived functors, this proposition yields homotopy left Kan ex-tensions and homotopy colimits. Dualizing Definition B.1 we obtain the following.

Page 136: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

136 MORITZ GROTH

Definition B.4. LetM be a model category, let A ∈ Cat , and let X,Y : A→M.

(i) A morphism f : X → Y is an injective cofibration if it is levelwise cofibra-tion.

(ii) A morphism f : X → Y is an injective weak equivalence if it is a levelwiseweak equivalence.

(iii) A morphism f : X → Y is an injective fibration if it has the RLP withrespect to acyclic injective cofibrations.

As in the projective case, if the above three classes define a model structure onMA, then we refer to it as the injective model structure and denote the resultingmodel category by MA

inj. One of the first examples was established by Heller in

[Hel88] where he shows that diagram categories in simplicial sets admit the injectivemodel structure. In general, as of this writing, we only know that injective modelstructures exist on diagram categories in combinatorial model categories. Let usrecall that a model category is combinatorial if

(i) the model structure is cofibrantly generated and if(ii) the underlying category is locally presentable.

We refer the reader to [Gro10, §3.2] for a short introduction to combinatorial modelcategories and additional references. Here, we only mention that many modelcategories arising in nature are combinatorial.

Theorem B.5. Let M be a combinatorial model category and let A ∈ Cat. Theinjective model structure exists on MA.

Injective model structures enjoy the following dual functorial properties.

Proposition B.6. Let M be a combinatorial model category and let u : A → Bbe in Cat. The adjunction (u∗, u∗) : MB

inj MAinj is a Quillen adjunction. In

particular, (∆A, limA) : MMAinj is a Quillen adjunction.

Proof. The functor u∗ : MB → MA preserves injective cofibrations and acyclicinjective cofibrations since both classes are defined levelwise.

Thus, for combinatorial model categories we can use the projective model struc-tures of Theorem B.2 to establish the existence of homotopy left Kan extensionsand the injective model structures of Theorem B.5 to obtain homotopy right Kanextensions. To relate the domains of the respective functor we make the followingobservation.

Proposition B.7. Let M be a combinatorial model category and let A ∈ Cat. Theidentity adjunction (id, id) : MA

proj MAinj is a Quillen equivalence.

Proof.

The main goal of this subsection is to show that combinatorial model categorieshave homotopy derivators. The proof that homotopy Kan extensions can be calcu-lated pointwise needs some additional preparation. For this purpose, let us consideru : A→ B, b ∈ B, and the associated slice squares

(u/b)p

//

π(u/b)

A

u

1b

// B

and

(b/u)q

//

π(b/u)

A

u

1b

// B.

AI

Page 137: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 137

Proposition B.8. LetM be a model category, let u : A→ B in Cat, and let b ∈ B.

(i) If M is a cofibrantly generated model category, then the restriction functors

p∗ : MAproj →M

(u/b)proj and b∗ : M→MB

proj are left Quillen functors.

(ii) IfM is a combinatorial model category, then the functors q∗ : MAinj →M

(b/u)inj

and b∗ : M→MBinj are right Quillen functors.

Proof. By Theorem B.2 and Theorem B.5 the respective model structures exist andit suffices by duality to take care of (i). We begin by showing that b∗ : MB

proj →M isa left Quillen functor. Since b∗ preserves weak equivalences, it suffices to show thatb∗ preserves projective cofibrations. Considering the adjunction (b∗, b∗) : MB M this is equivalent to b∗ preserving acyclic projective fibrations, i.e., levelwiseacyclic fibrations. Let us recall from the explicit construction of cofree diagrams inExample 5.29 that there are natural isomorphisms

b∗(X)b′ ∼=∏

homB(b′,b)

X, X ∈M, b′ ∈ B.

Since products of acyclic fibrations are again acyclic fibration, b∗ sends acyclicfibrations to acyclic projective fibrations.

We now show that p∗ : MAproj →M

(u/b)proj also is a left Quillen functor. TODO

Proposition B.9. Let M,N be cofibrantly generated model categories, and let(L,R) : M N be a Quillen adjunction.

(i) The induced adjunction (FA, GA) : MA NA is a Quillen adjunction.(ii) If (F,G) : M N is a Quillen equivalence then so is (FA, GA) : MA NA.

Proposition B.10. Let F : M → N be a left Quillen functor beween combinato-rial model categories and let y(F ) : y(M) → y(N ) be the induced strict morphismof derivators. Passing to left derived functors of the components of y(F ), we ob-tain an induced morphism of derivators LF : Ho(M) → Ho(N ). Similarly, aright Quillen functor G : N → M between combinatorial model categories inducesa morphism RG : Ho(N )→Ho(M).

Note that, in general, the morphisms LF and RG are not strict, since Quillenfunctors do not necessarily preserve weak equivalences. This fact is part of the mo-tivation behind defining morphisms of derivators as pseudo-natural transformations(as opposed to as 2-natural transformations).

References

[Ada74] J.F. Adams. Stable homotopy and generalised homology. Chicago Lectures in Mathe-matics. University of Chicago Press, Chicago, Ill.-London, 1974. 7, 72

[AHHK07] Lidia Angeleri Hugel, Dieter Happel, and Henning Krause, editors. Handbook of tiltingtheory, volume 332 of London Mathematical Society Lecture Note Series. Cambridge

University Press, Cambridge, 2007. 9, 121

[AR94] Jirı Adamek and Jirı Rosicky. Locally presentable and accessible categories, volume189 of London Mathematical Society Lecture Note Series. Cambridge University Press,

Cambridge, 1994. 70

[BBD82] Alexander Beılinson, Joseph Bernstein, and Pierre Deligne. Faisceaux pervers. In Anal-ysis and topology on singular spaces, I (Luminy, 1981), volume 100 of Asterisque,

pages 5–171. Soc. Math. France, Paris, 1982. 49

[BGP73] I. N. Bernsteın, I. M. Gel′fand, and V. A. Ponomarev. Coxeter functors, and Gabriel’stheorem. Uspehi Mat. Nauk, 28(2(170)):19–33, 1973. 122, 124

Page 138: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

138 MORITZ GROTH

[BK72] Aldridge Knight Bousfield and Daniel Marinus Kan. Homotopy limits, completions

and localizations. Lecture Notes in Mathematics, Vol. 304. Springer-Verlag, Berlin,

1972. 7[Bor94a] Francis Borceux. Handbook of categorical algebra. 1, volume 50 of Encyclopedia of

Mathematics and its Applications. Cambridge University Press, Cambridge, 1994. Ba-

sic category theory. 70, 128[Bor94b] Francis Borceux. Handbook of categorical algebra. 2, volume 51 of Encyclopedia of

Mathematics and its Applications. Cambridge University Press, Cambridge, 1994. Cat-

egories and structures. 70, 128[Bro94] K. S. Brown. Cohomology of groups, volume 87 of Graduate Texts in Mathematics.

Springer-Verlag, New York, 1994. Corrected reprint of the 1982 original. 18

[CE99] Henri Cartan and Samuel Eilenberg. Homological algebra. Princeton Landmarks inMathematics. Princeton University Press, Princeton, NJ, 1999. With an appendix by

David A. Buchsbaum, Reprint of the 1956 original. 10[Cis03] Denis-Charles Cisinski. Images directes cohomologiques dans les categories de modeles.

Ann. Math. Blaise Pascal, 10(2):195–244, 2003. 7, 65, 71

[Cis08] Denis-Charles Cisinski. Proprietes universelles et extensions de Kan derivees. TheoryAppl. Categ., 20:No. 17, 605–649, 2008. 65

[DS95] William Gerard Dwyer and Jan Spalinski. Homotopy theories and model categories.

In Handbook of algebraic topology, pages 73–126. North-Holland, Amsterdam, 1995. 7,28, 65

[Fai73] Carl Faith. Algebra: rings, modules and categories. I. Springer-Verlag, New York,

1973. Die Grundlehren der mathematischen Wissenschaften, Band 190. 20, 28[Fra96] Jens Franke. Uniqueness theorems for certain triangulated categories with an Adams

spectral sequence, 1996. Preprint. 3, 65

[Fre03] Peter J. Freyd. Abelian categories. Repr. Theory Appl. Categ., (3):1–190, 2003. 28[GM03] Sergei I. Gelfand and Yuri I. Manin. Methods of homological algebra. Springer Mono-

graphs in Mathematics. Springer-Verlag, Berlin, second edition, 2003. 3, 10, 30, 37[GPS14] Moritz Groth, Kate Ponto, and Michael Shulman. Mayer–Vietoris sequences in stable

derivators. Homology, Homotopy Appl., 16:265–294, 2014. 7, 71, 104

[Gro] A. Grothendieck. Les derivateurs.www.math.jussieu.fr/~maltsin/groth/Derivateurs.html. Manuscript, edited by

M. Kunzer, J. Malgoire, and G. Maltsiniotis. 3, 65

[Gro57] Alexander Grothendieck. Sur quelques points d’algebre homologique. Tohoku Math.J. (2), 9:119–221, 1957. 4, 10, 20

[Gro10] Moritz Groth. A short course on ∞-categories. http://arxiv.org/abs/1007.2925,

2010. Preprint. 7, 70, 136[Gro13] Moritz Groth. Derivators, pointed derivators, and stable derivators. Algebr. Geom.

Topol., 13:313–374, 2013. 7, 70, 106, 108, 111

[GS14a] Moritz Groth and Jan Stovıcek. Abstract representation theory of Dynkin quivers oftype A. arXiv:1409.5003, 2014. 9, 126, 127

[GS14b] Moritz Groth and Jan Stovıcek. Tilting theory for trees via stable homotopy theory.

arXiv:1402.6984, 2014. 9, 123, 124[GS14c] Moritz Groth and Jan Stovıcek. Tilting theory via stable homotopy theory.

arXiv:1401.6451, 2014. 9, 101, 122, 123[GS15] Moritz Groth and Jan Stovıcek. Towards abstract representation theory of acyclic

quivers. In preparation, 2015. 9, 124

[GU71] Peter Gabriel and Friedrich Ulmer. Lokal prasentierbare Kategorien. Lecture Notes inMathematics, Vol. 221. Springer-Verlag, Berlin, 1971. 70

[GZ67] Peter Gabriel and Michel Zisman. Calculus of fractions and homotopy theory. Ergeb-

nisse der Mathematik und ihrer Grenzgebiete, Band 35. Springer-Verlag New York,Inc., New York, 1967. 27

[Hap87] Dieter Happel. On the derived category of a finite-dimensional algebra. Comment.

Math. Helv., 62(3):339–389, 1987. 122[Hap88] Dieter Happel. Triangulated categories in the representation theory of finite-

dimensional algebras, volume 119 of London Mathematical Society Lecture Note Se-

ries. Cambridge University Press, Cambridge, 1988. 9, 19, 121

Page 139: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

INTRODUCTION TO THE THEORY OF DERIVATORS 139

[Hel68] Alex Heller. Stable homotopy categories. Bull. Amer. Math. Soc., 74:28–63, 1968. 5,

47

[Hel88] Alex Heller. Homotopy theories. Mem. Amer. Math. Soc., 71(383):vi+78, 1988. 3, 65,113, 136

[Hir03] Philip Steven Hirschhorn. Model categories and their localizations, volume 99 of Math-

ematical Surveys and Monographs. American Mathematical Society, Providence, RI,2003. 65

[HJR10] Thorsten Holm, Peter Jørgensen, and Raphael Rouquier, editors. Triangulated cate-

gories, volume 375 of London Mathematical Society Lecture Note Series. CambridgeUniversity Press, Cambridge, 2010. 5, 44, 46, 47

[Hov99] Mark Hovey. Model categories, volume 63 of Mathematical Surveys and Monographs.

American Mathematical Society, Providence, RI, 1999. 7, 28, 65, 118, 135[Hov01] Mark Hovey. Model category structures on chain complexes of sheaves. Trans. Amer.

Math. Soc., 353(6):2441–2457 (electronic), 2001. 28, 71[Joy08] Andre Joyal. The theory of quasi-categories and its applications. Lectures at the CRM

(Barcelona). Preprint, 2008. 7

[Kra10] Henning Krause. Localization theory for triangulated categories. In Triangulated cate-gories, volume 375 of London Math. Soc. Lecture Note Ser., pages 161–235. Cambridge

Univ. Press, Cambridge, 2010. 28, 47

[KS06] Masaki Kashiwara and Pierre Schapira. Categories and sheaves, volume 332 ofGrundlehren der Mathematischen Wissenschaften [Fundamental Principles of Math-

ematical Sciences]. Springer-Verlag, Berlin, 2006. 4, 10, 54, 71

[KV87] Bernhard Keller and Dieter Vossieck. Sous les categories derivees. C. R. Acad. Sci.Paris Ser. I Math., 305(6):225–228, 1987. 46

[Lur09] Jacob Lurie. Higher topos theory, volume 170 of Annals of Mathematics Studies.

Princeton University Press, Princeton, NJ, 2009. 7[Lur11] Jacob Lurie. Higher algebra. http://www.math.harvard.edu/~lurie/, 2011. Preprint.

118[Mal01] Georges Maltsiniotis. Introduction a la theorie des derivateurs (d’apres Grothendieck).

http://people.math.jussieu.fr/~maltsin/textes.html, 2001. Preprint. 65

[Mal05] Georges Maltsiniotis. Categories triangulees superieures. http://people.math.

jussieu.fr/~maltsin/textes.html, 2005. Preprint. 9, 49, 127

[Mal07] Georges Maltsiniotis. La K-theorie d’un derivateur triangule. In Categories in algebra,

geometry and mathematical physics, volume 431 of Contemp. Math., pages 341–368.Amer. Math. Soc., Providence, RI, 2007. 65

[ML98] Saunders Mac Lane. Categories for the working mathematician, volume 5 of Graduate

Texts in Mathematics. Springer-Verlag, New York, second edition, 1998. 6, 128[Mor58] Kiiti Morita. Duality for modules and its applications to the theory of rings with

minimum condition. Sci. Rep. Tokyo Kyoiku Daigaku Sect. A, 6:83–142, 1958. 9

[MP89] Michael Makkai and Robert Pare. Accessible categories: the foundations of categoricalmodel theory, volume 104 of Contemporary Mathematics. American Mathematical

Society, Providence, RI, 1989. 70[Nee01] Amnon Neeman. Triangulated categories, volume 148 of Annals of Mathematics Stud-

ies. Princeton University Press, Princeton, NJ, 2001. 5, 44, 46

[Pup67] D. Puppe. Stabile Homotopietheorie. I. Math. Ann., 169:243–274, 1967. 5, 45[Qui67] Daniel Gray Quillen. Homotopical algebra. Lecture Notes in Mathematics, No. 43.

Springer-Verlag, Berlin, 1967. 7, 28, 65[Rot79] Joseph J. Rotman. An introduction to homological algebra, volume 85 of Pure and

Applied Mathematics. Academic Press Inc. [Harcourt Brace Jovanovich Publishers],

New York, 1979. 10

[Ver96] Jean-Louis Verdier. Des categories derivees des categories abeliennes. Asterisque,(239):xii+253 pp. (1997), 1996. With a preface by Luc Illusie, Edited and with a

note by Georges Maltsiniotis. 4, 5, 45, 48[Vog70] R. Vogt. Boardman’s stable homotopy category. Lecture Notes Series, No. 21. Matem-

atisk Institut, Aarhus Universitet, Aarhus, 1970. 7, 72

[Wei94] Charles A. Weibel. An introduction to homological algebra, volume 38 of CambridgeStudies in Advanced Mathematics. Cambridge University Press, Cambridge, 1994. 3,

10, 30, 45, 47

Page 140: INTRODUCTION TO THE THEORY OF DERIVATORS · INTRODUCTION TO THE THEORY OF DERIVATORS MORITZ GROTH Abstract. In this book we give an introduction to the theory of derivators, a purely

140 MORITZ GROTH

MPIM, Vivatsgasse 7, 53111 Bonn, Germany

E-mail address: [email protected]