MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Direct to the point. Since now I've looked a lot of presentations of $\infty$-categories, but it seems that the only way to do explicit computations on these objects is via model categories. Is that so, or maybe there are other way of doing computations in $\infty$-categories, and if there such different ways what are they?

Edit: I really want to thank all people that answered this question, unfortunately I can choose just one answer. Btw because here is yet midnight I wish everybody a happy new year.

share|cite|improve this question
Giorgio, could you clarify what kind of computations you're speaking of? If someone asked a question about "doing computations in [ordinary] categories", I'd have little idea what they meant. There are lots of different kinds of computation you can do in a category, and the question would be so general as to be almost meaningless. Similarly, I don't know what you have in mind here. – Tom Leinster Dec 27 '11 at 16:06
@Tom, Giorgio: I answered the question under the presumption that the Giorgio is interested in the homotopical invariants of the spaces of maps in an $\infty$-category, including what happens to these invariants under the composition product. – Justin Noel Dec 27 '11 at 16:39
It doesn't seem like the point of $\infty$-categories is to do computations. Rather, it seems to me that their purpose is to (1) set up a nice enough theory so as to create machines and theorems that justify and allow for computing and (2) to make it so that all the conceptual parts can be proven conceptually, neatly leaving the hard parts to be proven with hard work. – Dylan Wilson Dec 27 '11 at 18:49
@Giorgio: I don't think it's fair to describe the homotopy hypothesis as aiming to "reduce" the study of spaces to the study of $\infty$-groupoids. Usually we don't talk about "reducing" easy things to hard things! It might be more accurate to say that the homotopy hypothesis enables us to use spaces (which are things we have studied for a long time) to say things about $\infty$-groupoids and higher categories. – Mike Shulman Dec 28 '11 at 7:59
@Giorgio: Expanding on what Mike said, I think a good analogy is the relationship between category theory and sets. A lot of category theory is about translating abstract things to questions about sets, where we can do things (e.g. apply Hom(-,-) and get to work). Similarly, infinity-category theory allows one to translate infinity-categorical questions into questions about spaces, which we know a little about (not nearly as much as about sets). This is all, of course, a vast oversimplification, but still useful to think about, maybe. – Dylan Wilson Dec 28 '11 at 19:09
up vote 17 down vote accepted

(This is an answer to a question below from Akhil Mathew; he wanted examples of ``explicit computations'' since all he knew were classical 1950s calculations and abstract theory. My answer is too long for a comment and too short to do justice to the question.)

That is terrible!!! I don't know where to begin, since there are huge masses of explicit calculations in the past half century. You mention infinite loop spaces. There is a theorem that says that the ordinary mod $p$ homology of $\Omega^n\Sigma^n X$ is an explicit functor of the ordinary mod $p$ homology of $X$, with all information (product, Dyer-Lashof operations, coproduct, Steenrod operations) determined. That surely sounds conceptual, but try to find a conceptual proof! We have tons of explicit calculations in the classical and Novikov Adams spectral sequences, without which the solution of the Kervaire invariant problem would be unthinkable. Chromatic theory as a whole is informed by explicit calculations. In unstable homotopy theory, exponent theorems for the homotopy groups of spheres use a remarkable blend of theoretic and calculational techniques. There are tons of explicit calculations in ordinary and generalized cohomology in the past half century. Spin cobordism and its applications to curvature questions is an example of a blend of algebraic topology and differential geometry. I could go on for 100s of pages without pausing for breath, and none of these results use any model category theory, let alone $\infty$ categories. There is a subject out there, with real content.

share|cite|improve this answer
Professor May, thank you for these clarifying remarks, and examples. – Akhil Mathew Dec 28 '11 at 18:56
Some more references for Akhil (mainly expanding "chromatic theory is informed by computations"): wherein computations prove the Conner-Floyd conjecture, where some self-maps of spectra are computed (useful for constructing nontrivial periodic families of homotopy elements), a nonexistence result proved using computations in the ANSS, wherein nontriviality of the gamma-family is proved (among other things), – Dylan Wilson Dec 28 '11 at 19:00
@Dylan: thanks! I'll have to look at these. – Akhil Mathew Dec 28 '11 at 19:01
etc., etc. All the above were published after the 1950s and are computation oriented. Also I posted Mitchell's proof of the CF conjecture- the original proof was even more computational, and consisted largely of computing the Morava K-theory of a bunch of spaces. – Dylan Wilson Dec 28 '11 at 19:03
@Akhil: You're welcome! – Dylan Wilson Dec 28 '11 at 19:03

It seems to me that $\infty$-categories are perfectly suited for a particular type of argument: namely to establish results in a homotopical context that would follow for purely formal reasons in a discrete context. There is quite a wide range of statements in algebraic geometry, representation theory etc which you know in a classical setting for purely abstract nonsense reasons, typically having to do with adjunctions, monadicity, commutation of limits with limits or colimits with colimits etc. and $\infty$-categories allow one to take such results and prove them in a derived setting. To put it in other words, if your "computation" is such that you can characterize the result by a universal property, then you can carry it out in the $\infty$-setting. On the other hand it seems to me that in this kind of mathematics many of the nontautological statements and real content tend to boil down to things like trying to pass a limit through a colimit. In this case you typically need some nonformal arguments and having models is likely inescapable.

These comments of course only reflect my limited experience with $\infty$-categories - luckily in my work with David Nadler in applying homotopical techniques to representation theoretic questions we have only needed formal statements, and $\infty$-categorical language made it quite easy to carry out the arguments, which would have been certainly impossible for me without it. The closest to a "calculation" we made in the $\infty$-context is in the paper Loop Spaces and Connections. We were able to prove the Hochschild-Kostant-Rosenberg theorem (identifying Hochschild homology of a commutative ring with symmetric algebra of differential forms) for any derived scheme (in characteristic zero) by this kind of abstract nonsense, but in order to show that the circle action on the former is "the same as" the de Rham differential (in the smooth case) was trickier - it's formal to see that the action map for the circle agrees with the de Rham differential but you still need to see all the higher structure of a circle action is uniquely determined (the action map only determines an action of the "free group" or James construction on the circle).. For this it appeared we'd need models until we saw that we could use a formal/structural argument (keeping track of the compatibility of everything with the action of the multiplicative group one finds there is no room for higher obstruction data). But I think this is a case of getting lucky in a sufficiently structured setting.. beyond that I doubt one can get away without models (forcing homotopical charlatans such as myself out of the business).

share|cite|improve this answer
@DavidBen-Zvi: Really interesting answer, btw I'm a little confused here by "models" do you mean "model categories"? – Giorgio Mossa Dec 28 '11 at 13:32
Yes - it seems to me that when you need to identify a particular differential or explicit relation or generally any argument that doesn't follow for purely structural reasons, you will need to get behind the gleaming facade and into the inner workings of the machine, which typically means working with model categories. This is analogous to needing to choose a basis or coordinates to do most explicit calculations in the "old world", unless you can obviate them by abstract nonsense. – David Ben-Zvi Dec 28 '11 at 16:55
I think this is a great point--one great thing about becoming familiar with oo-categories is that the kinds of "formal" arguments from classical category theory will carry over, really easily, into the oo-category world. To a listener it might be frustrating if Person A says the word "adjunction" instead of saying "oo-adjunction;" but one reason Person A can feel such license is because, really, so much of classical terminology and intuition can be transported into this framework. – Hiro Lee Tanaka Dec 28 '11 at 18:28

What a fun question!

I'd like to first mention a speculation on my part: When most people think of $\infty$-categories, the categories they think of probably exist in another formulation. For instance, the category of spaces (or Kan complexes), of chain complexes, of commutative DGAs, of spectra, et cetera, all have well-developed theory. And we're used to doing computations in these categories based on approaches that predate quasi-categories or Lurie's HTT or Higher Algebra. For such well-studied categories, I think you'll almost always find some pre-quasi-category (e.g., model categorical) "computation" that'll get the job done. There's no need to think of them as weak Kan complexes first.

(To compute a homotopy fiber of a map, for instance, we'll probably just replace the map by a standard fibration and compute its fiber thereafter. This is a silly example, but one in which I think working with a model-categorical framework is easier. This is much faster than proving that some over-category has a terminal object.)

But it would be a lie to say that most "computations" one needs to perform in an $\infty$-category can already be done in a non-quasi-category world. Here are two examples:

(1) If you're working with a new $\infty$-category for which someone hasn't done the model-categorical prep work for you. Depending completely on your math-path, you might come across (or define) a category which is most naturally defined as a weak Kan complex. One degree of separation away, you might define a category which is most naturally enriched over Kan complexes, and apply the simplicial nerve construction to obtain a quasi-category. But if you're just a tramp like me, suddenly face-to-face with a new category with some homotopical flavor, you may not have a natural candidate for a model structure, nor have any intuition for how to prove that something really is a "homotopy fiber" for some map, or more generally a homotopy (co)limit for a homotopy coherent diagram. (Whatever a "homotopy (co)limit" means for your category.)

With such a new category, all you a priori have is whatever led you to define this combinatorial gadget (a simplicially enriched category, or a quasi-category) and maybe some interpretations of your morphisms depending on what motivated your definition. So when trying to prove that some object is the homotopy fiber for some morphism in your quasi-category, it might be easiest for you to simply prove that an over-category has a terminal vertex.

I should admit that, while I imagine that examples like this will come up more and more, the only example I have in mind comes from joint work with David Nadler, where we compute kernels in a category we define, which happens to be most naturally a quasi-category.

(By the way, if someone has techniques that make it very easy to prove something is a limit of a diagram in a fibrant simplicial category, please post it as a comment to this post! I'd love to know more techniques.)

(2) If the diagrams you're working with are homotopy coherent but hard to make sense of at a level of strict commutativity. Another example I've come across is to prove that two homotopy coherent diagrams have the same homotopy colimit. Again because there is a concrete model for over/under categories in the $\infty$-categorical model, I could write down an equivalence between the over-categories associated to the two diagrams.


If the non-$\infty$-categorical computations are made possible by model-categorical ideas (like knowing how to replace morphisms), then $\infty$-categorical computations are possible because you reduce your computations to simplicial-set ideas, often proving that a simplicial set is contractible or that it has a terminal vertex. In some mathematical universe both paths may amount to the same thing, but it's my impression that the former approach is hard to follow if you're working in a new category without model structure, and is also not as easy to work with when the higher homotopy coherences of your diagrams are subtle. (Please feel free to let me know if I'm mistaken on this point, I'd love to hear more views.) The quasi-category framework allows you to avoid much of that difficulty by passing the buck to the geometry of simplicial sets.

By the way, you also mentioned something about computing spaces of $k$-morphisms. I have no idea how to do such a thing in the $\infty$-category world, or in either world, really. (Please educate me if someone does know how.)

share|cite|improve this answer

I believe that many computations could be done in the context of $\infty$-categories given enough outside input. For example, I would expect that most of the spectral sequences arising in topology could be constructed in the context of $\infty$-categories (indeed there are some basic spectral sequences constructed in Lurie's Higher Algebra). However, knowing that they converge to the right thing, might require you to know very specific details about limits and colimits in your $\infty$-category. Since the existence of these limits and colimits is typically deduced from the existence of homotopy colimits and limits in some model category we are often forced to consider the constructions in some model. For the issues of convergence, I am thinking of the Eilenberg-Moore and unstable Adams spectral sequences.

Even if we can overcome these problems, we usually need a small stable of existing computations to get us started (e.g. $\pi_1 S^1\cong \mathbb{Z}$, the ring structure on $K(S^2)$, the group cohomology of cyclic groups,...). The $\infty$-category context is an unnatural place to do such computations. It is more natural to work in an actual category with actual point-set models for our objects that let us compute something from first principles.

When using $\infty$-categories that arise from taking the nerves of the subcategories of bifibrant objects in some model categories, we find, after unwinding the definitions, that many of our $\infty$-category constructions involve taking lots of cofibrant/fibrant replacements. Sometimes we don't need to take a bifibrant model for a computation and we could get by with a cofibrant model. Sometimes we have no concrete idea what the fibrant replacement does to our objects. This is particularly true when we have taken a left Bousfield localization of a model category (in the combinatorial case, this corresponds to presentable $\infty$-categories). In such a case, the $\infty$-category context (unnecessarily) forces us to lose control of our point-set models.

share|cite|improve this answer
Can you elaborate on what sort of "very specific details" you need to know about limits and colimits? – Mike Shulman Dec 28 '11 at 7:54
Mike: I was thinking of the EMSS and the UASS. These two spectral sequences are associated to the Tot filtration on a cosimplicial space, where you want the totalization to be identified with something else. For example, in the UASS we try to compute the homotopy groups of $Tot(Top(X,\mathbb{F}_p^{\bullet} X))$ where $\pi_*\mathbb{F}_p X\cong \tilde{H}_*(X,\mathbb{F}_p)$. Our interest in this spectral sequence lies in the cases where this Tot has the homotopy type of the space of maps from $X$ to the $p$-completion of $Y$. – Justin Noel Dec 28 '11 at 8:31
This works if $Y$ is a simply connected product of Eilenberg-Maclane spaces. Using this and the 'fibre lemmas' of Bousfield-Kan we see that it works more generally, by induction up the Postnikov tower, for $Y$ $p$-nilpotent. I was thinking of these fibre lemmas as requiring this sort of hands on data. – Justin Noel Dec 28 '11 at 8:46

See my answer to the question "Do we still need model categories?" here.

share|cite|improve this answer
@Peter: Please don't take this as asserting disagreement with what you wrote there. I'm genuinely curious: do you intend this as an answer to the question, or more like commentary on the question? If it is meant as an answer to the actual question (whatever it means exactly), then my guess is that your answer is 'no'. But then, with Tom Leinster above, I don't know what you'd be saying no to, exactly. – Todd Trimble Dec 27 '11 at 16:25
Except that he perhaps exaggerates the utility of $\infty$-categories, I agree with Dylan's answer above. His point (1) applies to certain contexts, and the all'' in his (2) is way too strong. But his point is right: that is not what $\infty$ categories (or categories for that matter) are there for. One can of course specialize either to more calculable contexts (as in Ronnie's answer) but that misses the thrust of the question, I think. It would be nice if there were more young mathematicians who felt at home with both explicit computations'' and abstract theory. – Peter May Dec 27 '11 at 19:41
Professor May -- I realize you've said this multiple times now, but I'm unsure: could you give an example of a serious "explicit computation"? Most of what I've come across in algebraic topology has been either very classical computations (i.e. 1950s computation of cohomology rings) or abstract theoretical material (from then onwards) such as model categories or infinite loop spaces. – Akhil Mathew Dec 28 '11 at 14:37
For explicit computations see for example Ellis, G.J., About HAP: Third Homotopy Groups Of Suspensions Of Classifying Spaces:… Here the homotopy 3-type of the suspension of a classifying space of a group $G$ is determined by the nonabelian tensor product $G \otimes G$ and a "commutator" morphism from that to $G$, whose kernel is the 3rd homotopy group; and one often needs a computer to do the computations. The theory does not use cohomology or model categories. – Ronnie Brown Jan 15 '12 at 15:46

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.