Skip to the Main Content

Note:These pages make extensive use of the latest XHTML and CSS Standards. They ought to look great in any standards-compliant modern browser. Unfortunately, they will probably look horrible in older browsers, like Netscape 4.x and IE 4.x. Moreover, many posts use MathML, which is, currently only supported in Mozilla. My best suggestion (and you will thank me when surfing an ever-increasing number of sites on the web which have been crafted to use the new standards) is to upgrade to the latest version of your browser. If that's not possible, consider moving to the Standards-compliant and open-source Mozilla browser.

March 11, 2018

Stabilization of Derivators

Posted by Mike Shulman

(guest post by Ian Coley)

I recently published a paper to the arXiv which reconstructs an old paper of Alex Heller. Heller’s Stable homotopy theories and stabilization is one of a few proto-derivator papers that are still oft-cited by those of us studying derivators — a subject absent from this website since the two papers of Mike Shulman and Kate Ponto were published in 2014! Therefore before getting into the paper itself, it’s worth recalling what a derivator is supposed to be and do. For those interested in the long version, check out the nLab article or Moritz Groth’s excellent paper.

But for the short version: a prederivator is a strict 2-functor 𝔻:Cat opCAT\mathbb{D}\colon\mathbf{Cat}^\text{op}\to\mathbf{CAT}. For J,KCatJ,K\in\mathbf{Cat} and a functor u:JKu\colon J\to K, we obtain a restriction functor u *:𝔻(K)𝔻(J)u^\ast\colon\mathbb{D}(K)\to\mathbb{D}(J). A natural transformation α:uv\alpha\colon u\to v is mapped contravariantly to α *:u *v *\alpha^\ast\colon u^\ast\to v^\ast.

We think of the domain of a derivator as diagrams in the shape of small categories, and the outputs coherent diagrams with values in some category 𝒞\mathcal{C}. Specifically, that category 𝒞\mathcal{C} is the value of 𝔻\mathbb{D} at the one-point category ee. We’ll call 𝔻(e)\mathbb{D}(e) the underlying category of 𝔻\mathbb{D}.

A derivator is a prederivator satisfying some more axioms, most pertinent among them being the following: each u *u^\ast admits a left adjoint u !u_! and a right adjoint u *u_\ast (called the homotopy Kan extensions along uu). As a consequence, 𝔻(e)\mathbb{D}(e) admits all homotopy limits and homotopy colimits. For one final axiom, a pointed derivator is a derivator with a pointed base, which implies that each 𝔻(K)\mathbb{D}(K) is also pointed. In this situation, we obtain a suspension-loop adjunction on 𝔻(e)\mathbb{D}(e), where suspension Σ\Sigma is defined as the (homotopy) pushout of 0X00\leftarrow X\rightarrow 0 and loop Ω\Omega is the (homotopy) pullback of 0X00\rightarrow X\leftarrow 0. It’s worth emphasizing that, once we have a derivator whose base happens to be pointed, the suspension and loop are canonical adjoints and are determined by the higher structure of the derivator. Other interesting properties are also automatically satisfied, but we needn’t get into that here.

What examples in real life give rise to a pointed derivator? For any pointed combinatorial model category \mathcal{M}, we can consider its homotopy category Ho\operatorname{Ho}\mathcal{M}. It can be shown that, for any small category KK, K\mathcal{M}^K admits both a projective and an injective model structure. This leads us to a pointed derivator 𝔻 \mathbb{D}_{\mathcal{M}}, defined by 𝔻 (K):=Ho( K)\mathbb{D}_{\mathcal{M}}(K):=\operatorname{Ho}(\mathcal{M}^K). For 𝒞\mathcal{C} a pointed \infty-category, we have a similar derivator 𝔻 𝒞(K):=Ho(𝒞 N(K))\mathbb{D}_{\mathcal{C}}(K):=\operatorname{Ho}(\mathcal{C}^{N(K)}), where N(K)N(K) is the nerve of KK.

The next question is of stability. A pointed derivator is stable if the (Σ,Ω)(\Sigma,\Omega) adjunction is an adjoint equivalence. The stable homotopy category of pointed spaces is constructed for precisely this purpose, that is, so that reduced suspension and loop space become inverse equivalences. There are many (classical) ways to perform this stabilization, and we aim for the abstract formulation in derivators.

If we are given a pointed derivator 𝔻\mathbb{D}, is there some universal nearest stable derivator? Put another way, is there some derivator St𝔻\operatorname{St}\mathbb{D} so that morphisms from 𝔻\mathbb{D} into a stable derivator 𝕊\mathbb{S} are the “same” as morphisms St𝔻𝕊\operatorname{St}\mathbb{D}\to \mathbb{S}? The answer is yes, and we have the following strong result.

Theorem 7.14. Let Der !\mathbf{Der}_! be the 2-category with objects regular pointed derivators, maps cocontinuous morphisms of derivators, and natural transformations modifications. Let StDer !\mathbf{StDer}_! be the full sub-2-category of stable derivators. There is a pseudofunctor St:Der !StDer !\operatorname{St}\colon\mathbf{Der}_!\to\mathbf{StDer}_! which is left adjoint to the inclusion. Specifically, there is a universal cocontinuous morphism of derivators stab:𝔻St𝔻 \operatorname{stab}\colon\mathbb{D}\to\operatorname{St}\mathbb{D} and precomposition with this morphism give an equivalence of categories of cocontinuous morphisms Hom !(St𝔻,𝕊)stab *Hom !(𝔻,𝕊) \operatorname{Hom}_!(\operatorname{St}\mathbb{D},\mathbb{S})\overset{\operatorname{stab}^\ast}{\longrightarrow}\operatorname{Hom}_!(\mathbb{D},\mathbb{S}) for any stable derivator 𝕊\mathbb{S}.

The adjective regular will be explained in a few paragraphs. To give a sketch of the proof, we first develop a theory of prespectrum objects in a derivator 𝔻\mathbb{D} (Notation 5.8). Consider the poset V 2V\subset\mathbb{Z}^2 on the objects (i,j)(i,j) such that |ij|1|i-j|\leq 1. We construct a pointed derivator which we name Sp𝔻\operatorname{Sp}\mathbb{D} whose underlying category is the subcategory of 𝔻(V)\mathbb{D}(V) that vanish off the diagonal. That this is a pointed derivator is a consequence of Theorem 4.10, which was stated without proof by Heller (his Proposition 7.4). The proof is nontrivial but straightforward, and it’s a fantastic exercise in the calculus of mates.

An object XSp𝔻(e)X\in\operatorname{Sp}\mathbb{D}(e) looks like 0 X 1 0 X 0 0 X 1 0 \array{&&0&\to&X_1\\ &&\uparrow&&\uparrow\\ 0&\to&X_0&\to&0\\ \uparrow&&\uparrow&&\\ X_{-1}&\to&0&&} extending infinitely in both directions. The higher structure of the derivator encodes the comparison maps σ n:X nΩX n+1\sigma_n\colon X_n\to \Omega X_{n+1} for all nn\in\mathbb{Z} naturally in such an object, so we rightfully call such an XX a prespectrum object. If all these comparison maps are isomorphisms, we call XX a (stable) spectrum object. We let St𝔻Sp𝔻\operatorname{St}\mathbb{D}\subset\operatorname{Sp}\mathbb{D} be the full subprederivator (i.e. full subcategories at each KCatK\in\mathbf{Cat}) consisting of the stable spectrum objects.

There is no guarantee that the stable spectrum prederivator is actually a derivator, let alone a stable derivator. The first big theorem of the paper (Theorem 6.12) is to show that there is a localization Sp𝔻St𝔻\operatorname{Sp}\mathbb{D}\to\operatorname{St}\mathbb{D}. Lemme 4.2 of Denis-Charles Cisinski in Catégories Dérivables says that the localization of any derivator is still a derivator, so that takes care of the first concern.

To prove that St𝔻\operatorname{St}\mathbb{D} is stable, we now need to define the adjective regular. A derivator is called regular if filtered colimits commute with finite limits. The derivator associated to any nn-topos for n[0,]n\in[0,\infty] is regular, as is the derivator associated to a Grothendieck abelian category. One of the equivalent definitions of stability for a derivator is that all colimits commute with finite limits, so regularity is a kind of pre-stability assumption on 𝔻\mathbb{D}. In this case, St𝔻\operatorname{St}\mathbb{D} is stable (Lemma 6.19 and Proposition 6.23). (As a side note: I am in the market for an alternative to regular as a descriptor for this situation and welcome audience suggestions.)

The last thing to settle is the universal property of the stabilization. Most specifically, the pseudonaturality of St\operatorname{St} on morphisms in the domain. Heller’s original proof has a critical error that we take great pains to fix in §7 of the paper, specifically in Lemma 7.4. Heller writes down a diagram incoherently (his Diagram 9.3) that he lifts to a coherent object. Perhaps the crowning achievement of the paper is the construction of this diagram coherently, which takes place through a subposet of 5\mathbb{Z}^5. Diagram yoga is absent in Heller’s original work, but our proof methodology has a diagrammatic flavor that is emblematic of the theory of derivators in modern scholarship. This makes it difficult to recreate the more ambitious diagrams within this blog, and so interested readers should check out the paper itself. Specifically, most of the subsections titled “Construction” repair baseless (though ultimately provable) claims by Heller and contain the most complex diagram shapes.

Posted at March 11, 2018 8:18 PM UTC

TrackBack URL for this Entry:

3 Comments & 0 Trackbacks

Re: Stabilization of Derivators

If “regularity is a kind of pre-stability assumption”, then could you just say “pre-stable” instead of “regular”?

Is there an established name for this property in (,1)(\infty,1)-category theory? Or does the existing literature in that case make a stronger assumption like local presentability?

Posted by: Mike Shulman on March 14, 2018 8:05 AM | Permalink | Reply to this

Re: Stabilization of Derivators

In (\infty-)category theory, if you’re locally presentable then filtered colimits commute with finite limits automatically. But using the adjective ‘locally presentable’ for this phenomenon alone seems reckless, and I think some other researchers are working on a theory of locally presentable derivators per se.

Pre-stable isn’t a bad name but it doesn’t imply an adjective for the dual notion. A derivator in which filtered limits commute with finite colimits also admits a (co?)stabilization with a universal property for continuous morphisms. Though the name ‘locally corepresentable’ is ugly, at least there’s a natural place to put ‘co-’ in the mix. Maybe one could call it ‘left pre-stable’, but the issue of left/right is already a nomenclature debacle in derivator circles.

Posted by: Ian Coley on March 14, 2018 10:45 PM | Permalink | Reply to this

Re: Stabilization of Derivators

I certainly wasn’t suggesting to use “locally presentable” for this property; that would be a terrible idea. I was just wondering whether people working in (,1)(\infty,1)-category theory generally assume local presentability and therefore don’t need a name for this property since it automatically holds in that case.

Are there naturally-occurring examples of “co-pre-stable” derivators?

Posted by: Mike Shulman on March 14, 2018 10:51 PM | Permalink | Reply to this

Post a New Comment