Local Transition of Transport, Anafunctors and Descent of n-Functors

December 8, 2006

Local Transition of Transport, Anafunctors and Descent of n-Functors

Posted by Urs Schreiber

$MathML-enabled post (click for more details).$

In discussion here we discovered that we need to harmonize our use of the terms in the title of this post. Here is my take at what is going on. Please feel free to add your comment and correct me where I need correction.

$MathML-enabled post (click for more details).$

Notation is an issue here. I will take the liberty of using my own ideosyncratic notation in the following, the one I have been using all along here in discussions.

So, in my notation, the issue is the following:

we have a domain $n$ -category

(1)

P_{n} (X)

and a codomain

(2)

T

and we want to talk about $n$ -functors

(3)

tra : P_{n} (X) \to T

without actually doing so. Instead, we want to assume we can handle $tra$ “locally”. This means we assume a morphism

(4)

p : P_{n} (U) \to P_{n} (X)

and assume we know what $tra$ pulled back along $p$ is.

Or not even quite that. But consider this for a moment, and I will shortly discuss how to forget $tra$ alltogether, without actually losing information.

So, in general, the thing we have now locally, on $P_{n} (U)$ , might not even take values in $T$ anymore. Instead, there might be a morphism

(5)

i : T' \to T

of codomains and our local functor takes values in $T'$ :

(6)

{tra}_{U} : P_{n} (U) \to T' .

If you like, you can consider $T' = T$ and $i = {Id}_{T}$ in all of the following. But I claim that you will want to allow non-identity $i$ eventually.

Okay, so saying that $tra$ locally looks like something taking values in $T'$ is saying that there is an equivalence

(7)

\begin{matrix} P_{n} (U) & \overset{p}{\to} & P_{n} (X) \\ {tra}_{U} ↓ & \sim ⇓ t & ↓ tra \\ T' & \overset{i}{\to} & T \end{matrix} .

This equivalence is a local trivialization of our $n$ -functor. More precisely, in order to emphasize the assumptions that went into thi, I say it is a $p$ -local $i$ -trivialization.

(And I am claiming that, more generally, we may want to weaken the equivalence $t$ to a mere special ambidextrous adjunction. All we ever do in this setup can be done with just assuming special ambijunctions here. I talk about applications where this relaxed notion of local trivializaiton is important in On 2d QFT: From Arrows to Disks.)

Okay fine. But we may find ourselves in a situation where the globally defined $tra$ is just not available as a concrete object, but where we sort of reconstruct it from just knowing its local version ${tra}_{U}$ together with some data encoded in the trivialization morphism $t$ .

As I discuss in detail in Transport, Trivialization, Transition one can show that given a $p$ -local $i$ -trivialization as above one can derive the following data.

Let $(P_{n} (U))^{[n]}$ be the $(n - 1)$ -fold pullback of $P_{n} (U)$ along itself. If $U \to X$ is a cover of $X$ , this simply amounts to going to $(n - 1)$ -fold intersections of patches.

The data we get is this:

On double intersections an equivalence

(8)

g : p_{1}^{*} {tra}_{U} \to p_{2}^{*} {tra}_{U} .

on triple intersections an equivalence

(9)

f : p_{23}^{*} g \circ p_{12}^{*} g \to p_{13}^{*}

on quadruple intersections an equivalence

(10)

\begin{matrix} p_{2}^{*} {tra}_{U} & \overset{p_{23}^{*} g}{\to} & p_{3}^{*} {tra}_{U} \\ p_{12}^{*} g ↑ & ↗ & ↓ p_{34}^{*} g \\ p_{1}^{*} {tra}_{U} & \overset{p_{14}^{*} g}{\to} & p_{4}^{*} {tra}_{U} \end{matrix} ≃ \begin{matrix} p_{2}^{*} {tra}_{U} & \overset{p_{23}^{*} g}{\to} & p_{3}^{*} {tra}_{U} \\ p_{12}^{*} g ↑ & ↘ & ↓ p_{34}^{*} g \\ p_{1}^{*} {tra}_{U} & \overset{p_{14}^{*} g}{\to} & p_{4}^{*} {tra}_{U} \end{matrix}

and so on.

(Here in the last diagram the triangles are supposed to be filled by the corresponding $f$ s.)

This, I call the transition data of my original $n$ -functor $tra$ . Ross Street would call it the descent data, I think. Compare pages 2 and 3 of

Ross Street
Descent Theory

with definition 5 in TraTriTra, if you like.

Notice how the globally defined $tra$ does not appear anywhere anymore. Only the local ${tra}_{U}$ does and lots of morphisms between its pullbacks, all of which can be obtained from the local trivialization $t$ , if one was given.

But, and that’s the point, even if no $t$ was given to begin with, and no globally defined $tra$ , we can consider transition data as above. In fact, and that’s why all this is a good idea, under suitable circumstances we may reconstruct from the transition data the globally defined $tra$ .

If so, we have demonstrated that our $p$ -locally $i$ -trivializable $n$ -functors form an $n$ -stack. An $n$ -stack of $Q$ s on $X$ means, in words, that $Q$ s on $X$ are the same as glued $Q$ s on $U$ , where $U \to X$ is a cover.

More technically, we have an $n$ -category of descent data

- its objects are $n$ -simplices colored with transition morphisms as indicated above

- its morphisms are the “obvious” morphisms of $n$ -simplices. There are a couple of ways to see the same obvious structure here. For $n = 2$ I spell it out in TraTriTra. And it’s the same as what Ross Street uses in his theory of higher descent.

Before listing a couple of examples, I will now try to say what an anafunctor is and how the above, for $n = 1$ relates to anafunctors.

Anafunctors were defined in

M. Makkai
Avoiding the axiom of choice in general category theory
Journal of Pure and Applied Algebra, Volume 108, Number 2, 22 April 1996, pp. 109-173(65)
.ps from M. Makkai’s site
anafun1.pdf (title and contents)
anafun2.pdf (introduction)
anafun3.pdf (main text)

Definition(M. Makkai) For $A$ and $B$ two categories, an anafunctor

(11)

F : A \to B

from $A$ to $B$ is a span

(12)

\begin{matrix} ∣ F ∣ & \overset{F_{1}}{\to} & B \\ F_{0} ↓ \\ A \end{matrix}

with $F_{0}$ surjective on objects and morphisms and such that every morphism in $A$ has at most one pre-image with given source and target object.

So, in words, instead of going directly from $A$ to $B$ we wirst kind of resolve $A$ in terms of $∣ F ∣$ and then go from $∣ F ∣$ to $B$ .

I will now try to indicate how from any object in the category of $p$ -local $i$ -transition data for a 1-functor we obtain an anafunctor in a canonical fashion.

In fact, I did already discuss this in On $n$ -transport: Universal Transition, albeit for the case $n = 2$ (which of course includes the case $n = 1$ we are restricting to right now.)

Namely, from a given cover

(13)

p : P_{1} (U) \to P_{1} (X)

we can form the category

(14)

P_{1} (U^{•})

of paths in the transition groupoid. Its morphism are generated from those in $P_{1} (U)$ together with a unique morphism between any two objects in $P_{1} (U)$ with the same projection, divided out by an obvious equivalence relation.

This comes equipped canonically with a projection

(15)

\begin{matrix} P_{1} (U^{•}) \\ ↓ \\ P_{1} (X) \end{matrix}

and, if you think about it, the relations mentioned above are precisely such that they ensure that every morphism in $P_{1} (X)$ has a unique lift with specified endpoints.

(I think I now understand that this is exactly what the pullback diagram (109) in

Toby Bartels
Higher gauge theory I: 2-Bundles
math.CT/0410328

achieves, too.)

Then, any $p$ -local transition data (descent data) $({tra}_{U}, g)$ canonically defines a functor

(16)

({tra}_{U}, g) : P_{1} (U^{•}) \to T .

In conclusion, from a given $p$ -local transition data (descent data) we construct the anafunctor

(17)

({tra}_{U}, g) : P_{1} (X) \to T

given by the span

(18)

\begin{matrix} P_{1} (U^{•}) & \overset{({tra}_{U}, g)}{\to} & T \\ ↓ \\ P_{1} (X) \end{matrix} .

I’d think that, conversely, every anafunctor can be interepreted as an object in the category of $p$ -local transition data of a 1-functor for given $p$ .

Examples:

The most natural examples are obtained by letting $X$ be a smooth space, letting $U \to X$ be a surjective submersion and letting $P_{n} (X)$ be some notion of smooth $n$ -paths in $X$ , letting $p : P_{n} (U) \to P_{n} (X)$ be the obvious induced morphisms on $n$ -paths and letting everything in sight be smooth.

Then all that remains to vary is the morphism $i : T' \to T$ . Just choosing different such $i$ we re-obtained an entire zoo of well-known structures.

For $G$ any ordinary Lie group and for

(19)

i = {Id}_{Σ (G)}

the category of $p$ -local $i$ -transition data is canonically isomorphic to the category of $G$ -cocycles describing locally trivialized $G$ -bundles with connection on $X$ . That’s an easy exercise.

For $G_{2}$ a strict Lie-2-group and for

(20)

i = {Id}_{Σ (G_{2})}

the category of $p$ -local $i$ -transition data is canonically isomorphic to the category of $G_{2}$ -cocycles describing locally trivialized $G_{2}$ -2-bundles with “fake flat” connection. This is the example discussed in John’s paper with me #. A greatly simplified proof is given here.

There is a rather obvious way generalizing this from a structure 2-group to a structure 2-groupoid. The resulting cocycle data is that discussed by Igor Bakovic # (he does not discuss connections, though).

The fake-flatness constraint is lifted as follows:

For $G_{2}$ a strict Lie-2-group and $INN (G_{2})$ its 3-group of inner automorphisms and for

(21)

i = {Id}_{Σ (INN (G_{2}))}

the category of $p$ -local $i$ -transition data is canonically isomorphic to the category of $G_{2}$ -cocycles describing locally trivialied $G_{2}$ -2-bundles with arbitrary connection, reproducing the data found by Breen-Messing. The proof is given here.

Just a special case of this, but worth mentioning is this:

For $G_{n} = Σ^{n} (U (1))$ the $n$ -fold suspension of $U (1)$ and

(22)

i = {Id}_{Σ^{n} (U (1))}

the category of $p$ -local $i$ -transition data is canonically isomorphic to the category of Deligne cocycles. I prove this for up to $n = 2$ here and I claim that it is clear that the statement holds for all $n$ (but I haven’t written down a proof for that).

In all these examples we had $i = Id$ the identity. So all these examples should correspond to anafunctors.

But, I claim, for $n > 1$ it is useful to have nontrivial $i$ . Identity $i$ will always give us cocycles for a “full local trivialization”. But often we want to do local pre-trivializations.

For instance: locally trivializing a gerbe yields as transition bundle also known as a bundle gerbe. Only if we, in turn, also locally trivialize this transition bundle do we get back to the Deligne cocycles mentioned above.

So, for

(23)

i : Σ (Σ (ℂ^{\times})) \overset{\subset}{\to} Σ (1 d {Vect}_{ℂ})

the canonical inclusion, the 2-category of $p$ -local $i$ -transition data is canonically isomorphic to the 2-category of line bundle gerbes with connection. The proof is here.

In fact, we should be thinking in terms of the chain of inclusions

(24)

Σ (Σ (ℂ^{\times})) \overset{\subset}{\to} Σ (1 d {Vect}_{ℂ}) \overset{\subset}{\to} Bim ({Vect}_{ℂ})

which manifestly makes line bundle gerbes the transition data of rank-1 line-2-bundles. This aspect is discussed here.

This has an obvious generalization. For $H$ any ordinary group, for $H BiTor$ the monoidal category of $H$ -bitorsors, and for the canonical inclusion

(25)

i : Σ (AUT (H)) \overset{\subset}{\to} Σ (H BiTor)

the 2-category of $p$ -local $i$ -transition data is canonically isomorphic to the 2-category of principal (nonabelian ) bibundle gerbes with connection as defined (on objects) by Aschieri-Jurčo. Again, the connection is “fake flat” as long as we work just with 2-groups. The proof is here.

Generally, if we take $i$ to be a representation of an $n$ -group

(26)

i : Σ (G_{n}) \to n Vect

we get the local data of an associated $n$ -bundle.

The discussion of line bundle gerbes above can be understood as an example for that for the canonical representation of $Σ (U (1))$ on $Bim (Vect) \subset 2 Vect$ . This is explained here.

For 3-bundles and 2-gerbes one would consider even longer chains of inclusions of codomains, corresponding to more steps of local trivializations: a 2-gerbe has a transition 1-gerbe, which has a transition bundle, which has a transition function.

A chain of inclusions for instance relevant for 1-dimensional vector 3-bundles I have discussed here.

All these examples assume that the transition morphisms are equivalences. If we instead consider transition data (anafunctors) and require the transitions to be just special ambidextrous adjunctions, we pass from the world of classical transport ( $n$ -bundles with connection) to that of quantum transport (evolution/propagation in $n$ -dimensional quantum field theory).

For instance, the inclusion

(27)

i = {Id}_{Σ (Vect)}

and demanding transitions to be just special ambijunctions yields the local data that is the Fukuma-Hosono-Kawai state sum model of 2-dimensional topological field theory. This I discussed here.

2-dimensional rational conformal field theory is, essentially, just topological field theory internalized in a modular tensor category $C$ more sophisticated than $Vect$ . Accordingly, transition data with respect to

(28)

i : Σ (C) \overset{\subset}{\to} Bim (C) \overset{\subset}{\to} TwBim (C)

can be understood as underlying the decoration prescription (“state sum model”) used in the FRS description of rational conformal field theory. This I talk about here and here.

So much for now. I’d be content with addressing all these examples of descent data for $n$ -functors as $n$ -anafunctors. Maybe “generalized” $n$ -anafunctors. Please let me know if that would be reasonable use of terminology.

Further notes on this topic, discussed in the comment section below, are these:

Morphisms of Anafunctors

A note on the peculiarity encountered in composing morphisms of anafunctors. Compare with Toby’s remarks below.

On Anafunctors and Transitions

A note on how anafunctors and functors with transition data are equivalent, followed by a proposal for defining 2-anafunctors in terms of transition data of 2-functors. Essentially a reformulation of what I had written about transitions and the universal transition before, now with an eye on the language of anafunctors.

Posted at December 8, 2006 5:45 AM UTC

TrackBack URL for this Entry: http://golem.ph.utexas.edu/cgi-bin/MT-3.0/dxy-tb.fcgi/1069

51 Comments & 19 Trackbacks

Re: Local Transition of Transport, Anafunctors and Descent of n-Functors

$MathML-enabled post (click for more details).$

In his article on anafunctors, M. Makkai presents almost everything in terms of two equivalent definitions. The one exception is the composition of morphisms of anafunctors, which is not presented in the otherwise more elegant definition in terms of spans.

I would like to understand composition of their morphisms in the more elegant span form.

I understand that Toby Bartels writes down exactly that, even internally, in his thesis. But for my own benefit, I want to see the relevant structure stripped off the complexity introduced by writing down everything internalized.

In the little time I have left today, I came up with this.

I’d be grateful if Toby, or anyone else, could quickly check if the composition of morphisms of anafunctors that I write down is correct and coincides with the standard definition.

Thanks!

Posted by: urs on December 8, 2006 4:52 PM | Permalink | Reply to this

Re: Local Transition of Transport, Anafunctors and Descent of n-Functors

I’m missing something here - probably an unspoken assumption. In the example $A=P_1(X)$ we have a functor with values
in Cat. Definition 3 seems to depend crucially on X and sigma seems to be a
________ from X to |G|.

Posted by: jim stasheff on December 9, 2006 3:01 AM | Permalink | Reply to this

Re: Local Transition of Transport, Anafunctors and Descent of n-Functors

$MathML-enabled post (click for more details).$

I’m missing something here […]

Oops, thanks. I wrote $X$ where it should have written $A$ .

Posted by: urs on December 9, 2006 12:54 PM | Permalink | Reply to this

Re: Local Transition of Transport, Anafunctors and Descent of n-Functors

Yeah, that’s right. (There’s a typo at the end of the linear diagram on the last page, but that’s obvious.)

Notice that neither I nor Makkai would allow ourselves to discuss σ (hence nor t). This is because you need the Axiom of Choice to prove that a (strict) functor σ exists, and even then it won’t be smooth. (Or to put it another way, it exists in the strict 2category of 2spaces and 2functors only if the axiom of choice holds in the category of spaces.) That may be one reason why Makkai doesn’t include this version of the definition.

Now, we can include everything in the diagram at the top of the last page except for the map t. Then we could argue that, because of the properties of the functor from |F| ×_X |G| ×_X |H| to |F| ×_X |H| is a surjection (or more generally cover) on objects and morphisms, there is still a natural transformation between the functors that we want (which, as you point out, no longer have |G| in them). Indeed, this is precisely what the sentence directly after diagram (118) in my thesis (version 3) is about.

I can’t read Makkai right now, since I don’t have a PostScript reader on this machine (and its hard drive is full to the bursting point), so I can’t give an intelligent reason as to why Makkai didn’t make the same argument. But I’m sure that he knew how to do it.

Posted by: Toby Bartels on December 10, 2006 1:24 AM | Permalink | Reply to this

morphisms of anafunctors

$MathML-enabled post (click for more details).$

Many thanks indeed to Jim Stasheff and Toby Bartels for pointing out typos! I think I have corrected them now, the corrected file is here.

Toby wrote:

Notice that neither I nor Makkai would allow ourselves to discuss $σ$ (hence nor $t$ ). This is because you need the Axiom of Choice […]

That may be one reason why Makkai doesn’t include this version of the definition.

Okay, I understand that. Or at least the idea. Since the diagrams I drew were supposed to be just a translation of what Makkai does on p.15 , when he says

[…] with any $t \in ∣ G ∣ X;$

I am not sure how choosing that $t$ does not involve the axiom of choice. I guess the answer is implicit somewhere in your (Toby’s) diagram (118). Indeed, as you write

Indeed, this is precisely what the sentence directly after diagram (118) in my thesis (version 3) is about.

I’ll need to think about this.

Posted by: urs on December 11, 2006 10:54 AM | Permalink | Reply to this

Re: morphisms of anafunctors

I am not sure how choosing that t does not involve the axiom of choice.

But t is just a single element of a set (known to be occupied). You don’t need choice to choose a singleton!

The diagram in 3.(iii) is making the same point (in its own way) as your diagram in Morphisms of Anafunctors. But it’s doing so point by point, with no need for a global choice function.

Here is the general situation: Let A, B, and C be sets, let f be a surjection from A to B, and let g be a function from A to C. Suppose that one wants to define a function h from B to C such that the triangle commutes. One might (using Choice) pick a section f′ of f, then let h be the composite of f′ and g. Under certain circumstances, one can prove that this composite is independent of the section chosen. This is what you do.

Alternatively, one might look at each element x of B and consider how g(y) varies as y varies over the fibre of f over x. Under certain circumstances, one can prove g(y) is independent of y, so long as f(y) is fixed. Then one may define h(x) to be g(y) for any y over x. This is what Makkai does; it does not require Choice, but it makes explicit reference to elements.

Finally, one can consider the kernel of f, which is a subset of the Cartesian square of A. (Membership-theoretically, it consists of pairs (y, z) such that f(y) = f(z). But it is defined by a universal property; category theorists call its inclusion map ‘kernel pair’.) There are two projection maps from this kernel to A, and under certain circumstances, one can prove that their composites with g are equal. In that case, it is a theorem of set theory that a unique function h exists as desired. And this theorem does not require Choice; in fact, it holds in any category, as long as f is a regular epimorphism and has a kernel pair. This is what I do.

Posted by: Toby Bartels on December 11, 2006 10:30 PM | Permalink | Reply to this

Re: morphisms of anafunctors

$MathML-enabled post (click for more details).$

In that case, it is a theorem of set theory that a unique function h exists as desired.

That’s very good. Have you ever thought about lifting the corresponding construction in your thesis from 1- to 2-categories?

Posted by: urs on December 12, 2006 10:15 AM | Permalink | Reply to this

Re: Local Transition of Transport, Anafunctors and Descent of n-Functors

This looks reasonable. As far as checking for errors, I’ve only looked at the references to anafunctors; but on the level of rough intuition, I like all of it.

I do want to stress one point, however: Anafunctors should be seen as the standard notion of morphism between categories (or at least internal categories). You can get away with using (strict) functors only if the Axiom of Choice holds in your context category. (In particular, Mac Lane and Eilenberg were able to get away with this in their original definition of functor, because they used Choice in the category of sets.) Thus Makkai wrote his paper originally because he wanted to allow for the possibility that Choice fails in the category of sets, and I used anafunctors because I was working in the category of smooth manifolds, where Choice definitely fails. (The precise statement of Choice for this purpose, incidentally, is that every cover has a section; you need to equip the category with a notion of what a cover is to even discuss what an anafunctor is. In the category of sets, a cover is a surjection; in the category of smooth manifolds, it’s probably best to let a cover be a surjective submersion, although my thesis required a surjective local diffeomorphism, which fortunately yields an equivalent 2category.)

So, when considering whether some descent data define a morphism between categories, it should be enough that they define an anafunctor.

Posted by: Toby Bartels on December 10, 2006 1:41 AM | Permalink | Reply to this

Re: Local Transition of Transport, Anafunctors and Descent of n-Functors

By the way, if you’re interested in Makkai’s notion of saturated anafunctor, I’ve fairly well convinced myself that these are (in the contenxt of Lie groupoids) Hilsum–Skandalis morphisms, which are fairly well established. So the Lie groupoid theorists are probably doing things correctly.

Posted by: Toby Bartels on December 10, 2006 1:49 AM | Permalink | Reply to this

Re: Local Transition of Transport, Anafunctors and Descent of n-Functors

I’d be interested in any *conceptual* comments on `span’ in anafunctor
in comparison to `roof’ as used in e.g. inverting equivalences in alg top/
homotopy theory.

Posted by: jim stasheff on December 10, 2006 7:25 PM | Permalink | Reply to this

Re: Local Transition of Transport, Anafunctors and Descent of n-Functors

I’d be interested in any conceptual comments on ‘span’ in anafunctor in comparison to ‘roof’ as used in e.g. inverting equivalences in alg top/ homotopy theory.

As far as I know, ‘span’ is just a general term in (elementary) category theory for a pair of morphisms with common domain. I don’t know what ‘roof’ means.

I would argue that the proper notion of morphism between spaces considered up to homotopy is not a (continuous) map, nor a homotopy-equivalence class of such (a useless step anyway if you’re going to use ωgroupoids), but a span whose first leg induces isomorphisms on all homotopy groups. Is this relevant?

Posted by: Toby Bartels on December 11, 2006 3:35 AM | Permalink | Reply to this

spans and roofs

$MathML-enabled post (click for more details).$

I’d be interested in any conceptual comments on ‘span’ in anafunctor in comparison to ‘roof’ as used in e.g. inverting equivalences in alg top/ homotopy theory.

As far as I know, ‘span’ is just a general term in (elementary) category theory for a pair of morphisms with common domain. I don’t know what ‘roof’ means.

I guess Jim Stasheff is referring to the use of “roof” as common in the theory of localization (many-object version of localization of a ring).

See for instance p. 19-20 of this.

So, a “roof” is a “span” in a certain context.

Posted by: urs on December 11, 2006 1:19 PM | Permalink | Reply to this

2-anafunctors?

$MathML-enabled post (click for more details).$

So is there any literature on 2-anafunctors?

Given all that I wrote above about how anafunctors are like descent data of functors I would think I know what an 2-anafunctor should be. But does anyone talk about 2-anafunctors?

One reason why I was interested in the composition of morphisms of anafunctors that we dicussed above is that for 2-anafunctors the analog problem is an old one in the theory of bundle gerbes:

What is a morphism between bundle gerbes that live on different surjective submersions $Y$ ?

There are proposed solutions to that, as in Danny Stevenson’s thesis, but it would be nice to have a general understanding.

Namely, as I have tried to indicate, a bundle gerbe is nothing but what should be called a 2-anafunctor. Morphisms of bundle gerbes correspond to (pseudo-)natural transformations of these 2-anafunctors, and the subtlety concerning morphsims of gerbes on different $Y$ correspond precisely to the subtlety we discussed above.

Posted by: urs on December 11, 2006 11:09 AM | Permalink | Reply to this

Re: 2-anafunctors?

[A] bundle gerbe is nothing but what should be called a 2-anafunctor.

Under certain circumstances, a bundle is nothing what should be called an anafunctor. Specifically, a principal Gbundle over B is nothing but what should be called an anafunctor from the (categorially) discrete groupoid of B to the one-object groupoid G. More generally, an anafunctor (at least is properly saturated?) from H to G defines an Hprincipal (H,G)bibundle (that is a Hilsum–Skandalis morphism).

Since a bundle gerbe is nothing but what should be called a 2bundle, you seem to be right on track!

Posted by: Toby Bartels on December 11, 2006 10:38 PM | Permalink | Reply to this

Re: 2-anafunctors?

$MathML-enabled post (click for more details).$

[A] bundle gerbe is nothing but what should be called a 2-anafunctor.

Under certain circumstances, a bundle is nothing what should be called an anafunctor.

There is a slight difference in these two statements. If by “is” we mean anything stronger than “is equivalent to” then we should say that

an anafunctor (with values in a group) is the transition function of a principal bundle (possibly with connection).

Same for bundle gerbes. A bundle gerbe is not a 2-bundle (at least not on base space) - it is the “transition 2-function” of a 2-bundle.

Sorry for being nitpicky about this, but this is essentially the point which got our discussion here started.

The category of anafunctors from $P_{1} (X)$ to $Σ (G)$ is canonically isomorphic to the category of “differential $G$ -1-cocycles” on $X$ . And it is (only, and non-canonically) equivalent to the category of $G$ -bundles (in the sense of total spaces $B$ with projection $B \to X$ , etc.) with connection on $X$ .

Of course in the end we don’t need to care about anything beyond equivalence, and probably that’s what you are doing already. But I think for the moment, where we are concerned with the peculiarities of the details of anafunctors, it pays to point out that some equivalences in the game here are actually isomorphisms, while others are not.

More generally, an anafunctor (at least is properly saturated?) from $H$ to $G$ defines an $H$ principal $(H, G)$ bibundle (that is a Hilsum-Skandalis morphism).

Interesting! Let me see if I understand that:

To me, the cocylce of an $(H, G)$ -bibundle on $X$ for $H$ and $G$ groups with $G$ acting on $H$ , forming a crossed module of groups, is an anfunctor

(1)

g : Disc (X) \to (H \to G)

from the discrete category of $X$ to $(H \to G)$ , where $(H \to G)$ denotes the 2-group corresponding to the above crossed module, regarded here as a mere groupoid (i.e. not using the monoidal structure on that).

In other words, this is the cocycle of an $H$ -bundle with the special property that its image under the map $t : H \to G$ is a trivial $G$ -cocycle (with fixed trivialization).

Is this roughly what you have in mind when you say “ $H$ -principal $(H, G)$ -bibundle”?

a Hilsum-Skandalis morphism

I had to look this up. Found one definition in Henriques & Metzelder: Presentation of noneffective orbifolds on p. 7:

Definition 2.6. Let $G$ and $H$ be smooth groupoids. A (Hilsum-Skandalis) morphism from $G$ to $H$ consists of a manifold $P$ , maps $s_{P} : P \to G_{0}$ , $t_{P} : P \to H_{0}$ , a left action of $G$ on $P$ with the base map $s_{P}$ , a right action of $H$ on $P$ with base map $t_{P}$ , such that:

(1) $s_{P}$ is $H$ -invariant, $t_{P}$ is $G$ -invariant;

(2) the actions of $G$ and $H$ on $P$ are compatible: given $p \in P$ , $(gp) h = g (ph)$ ;

(3) $s_{P} : P \to G_{0}$ , as an $H$ -bundle with base map $t_{P}$ , is principal. We will often abuse notation and denote such a morphism simply by $P$ .

Posted by: urs on December 12, 2006 10:02 AM | Permalink | Reply to this

Re: 2-anafunctors?

Henriques & Metzelder

(Notice that (1) and (2) make it a bibundle, while (3) makes it Hprincipal.)

Posted by: Toby Bartels on December 14, 2006 1:51 AM | Permalink | Reply to this

Re: 2-anafunctors?

$MathML-enabled post (click for more details).$

Henriques & Metzelder

(Notice that (1) and (2) make it a bibundle, while (3) makes it $H$ principal.)

Okay. So morally, a Hilsum-Skandalis morphism is to groupoids like a bimodule is to algebras. It is a groupoid bi-torsor.

The only subtlety is that where for a bimodule we just demand compatible left and right action, for a bitorsor we in addition want to demand one of the two actions to be principal. (They may both happen to be principal if $H = G$ , but need not.)

So then there should be a more or less obvious composition of HS-morphisms by a coequalizer construction, and a 2-morphism between two parallel HS morphisms should be any groupoid bundle homomorphism compatible with the left and right actions.

Very nice. This is now getting close to a question I meant to ask: what is the relation between anafunctors and profunctors?

In his introduction Makkai mentions that anafunctor alludes to profunctor by the analogy with mitosis, where we have, among others, an anaphase and prophase.

I haven’t really understood this analogy at all so far!

But given that you claim that a saturated anafunctor is, in the context of Lie groups, a Hilsum-Skandalis morphism, and that it seems as if such a morphism is nothing but a bitorsor, hence something like a functor on $H \times G^{op}$ , it’s now beginning to look as if

saturated anafunctors are profunctors

Or is it? I’d need to think more about this.

Posted by: urs on December 14, 2006 11:21 AM | Permalink | Reply to this

Re: 2-anafunctors?

But given that you claim that a saturated anafunctor is, in the context of Lie groups, a Hilsum-Skandalis morphism.

I’m not sure that this is really true. But something like it is.

Posted by: Toby Bartels on December 14, 2006 11:22 PM | Permalink | Reply to this

Re: 2-anafunctors?

what is the relation between anafunctors and profunctors?

On the one hand, only saturated anafunctors can be described this way as profunctors. On the other hand, only representable profunctors define anafunctors. However, up to equivalence, it’s fair to say that anafunctors are the same as representable profunctors. (That’s because every anafunctor is naturally ismorphic to a saturated one, and this respects composition, given an equivalence of bicategories.)

The idea that H-S morphisms correspond to saturated anafunctors is therefore akin to the idea that representability of a functor corresponds to principality of a bundle. I’m not sure if that’s true, but something like it should be.

I haven’t really understood this analogy at all so far!

As far as I can tell, it’s not meant to be anything more precise than that “pro” and “ana” can appear together to describe different variations on the same theme.

Posted by: Toby Bartels on December 14, 2006 11:39 PM | Permalink | Reply to this

Re: 2-anafunctors?

$MathML-enabled post (click for more details).$

I have a proposal for what a $n$ -anafunctor should be. Explicit for $n = 2$ with a rather straightforward generalization to higher $n$ .

My proposal is that an $n$ -anafunctor is the unique morphism obtained from an $n$ -functor with transition data under the construciton of what I call the “universal transition” # – a certain pushout.

For this to make sense, it should reduce to Makkai’s definition of anafunctors for $n = 1$ . And it does. I have written up a short note on that here:

On Anafunctors and Transitions .

Comments are very welcome. In essence, I am just restating what I wrote elsewhere concerning “transition of $n$ -transport”, but now trying to make direct contact with the anafunctor language.

Posted by: urs on December 12, 2006 2:40 PM | Permalink | Reply to this

Re: 2-anafunctors?

On Anafunctors and Transitions.

OK. I need to think about this.

Posted by: Toby Bartels on December 14, 2006 1:52 AM | Permalink | Reply to this

Re: 2-anafunctors?

$MathML-enabled post (click for more details).$

On Anafunctors and Transitions.

OK. I need to think about this.

I have now corrected some funny blunders in labelling some of the diagrams. Sorry for that. But I guess you got the idea nevertheless.

Posted by: urs on December 14, 2006 2:12 PM | Permalink | Reply to this

forgetting structure, remembering choice

$MathML-enabled post (click for more details).$

I hate myself for writing the fourth comment in a row here - but remember that it could be worse: I could be writing a new entry instead! ;-)

Anyway: now that we have made some progress in understanding each other’s terminology, I would like to hear in more detail Toby’s and/or John’s opinion on the way I did approach this issue.

It seems that I did implicitly use anafunctors all along, addressing them (originally, in my thesis) as functors on the “Čech path groupoid” and (later) as functors on paths in the transition groupoid of a cover.

But I always considered these functors as possibly being obtained from having “locally trivialized” a globally defined functor, i.e. one that was literally defined on the domain, not just on a surjection of the domain.

Of course, as we have all emphasized various times, that “globally defined” thing is usually not itself continuous/smooth in a direct sense.

In the language of anafunctors, I think the language that I adopted amounts to the following, and I’d be interested in anyone’s comment on that strategy:

While ambient categories like $Top$ or $C^{∞}$ don’t have sections for every epimorphism (“the axiom of choice fails”), there is a forgetful functor to $Set$ and - if we allow ourselves to use the axiom of choice on bare sets, which I am willing to do - after applying that functor the sections do exist.

So, for instance, for $A$ some domain and $∣ F ∣ \to A$ some surjection, all in our ambient category, we don’t in general have any section

(1)

s : A \to ∣ F ∣

but we do as soon as we forget the extra structure (e.g. the topology, or the smooth structure).

This allows to consider functors on $A$ , not internal to our ambient category. Instead of them being continuous/smooth by way of being internal to $Top$ or to $C^{∞}$ , it makes sense (I think) to equip them with a smooth structure by specifying that and how their pullback to $∣ F ∣$ is equivalent (internal just to $Set$ ) to a functor (on $∣ F ∣$ ) which is continuous/smooth in the sense that it lives internal to $Top$ / $C^{∞}$ .

It seems to me that this amplifies a property of anafunctors which is present, but which is only implicit in the way Makkai defines them and even less obvious in the way Toby defined them (so implicit, in fact, that I completely missed it until recently): and that’s the relation to descent data.

As the name suggests, this is precisely about descending down along the surjection

(2)

\begin{matrix} ∣ F ∣ \\ ↓ \\ A \end{matrix},

expressing the idea that the entity upstairs is like something downstairs having been pulled up.

That’s important, isn’t it? For instance this is, to my mind, the very justification for the precise definition of anafunctor that Makkai gives:

one might ask why exactly we demand an anafunctor to be a span with the extra property that the surjection of the left leg has unique lifts with specified source and targets.

This is the crucial point. I expect this becomes quite amplified once one tries to talk about $n$ -anafunctors: one will have to figure out what this condition really means.

And, I think, what it really means is that there is a certain descent property around. For $n = 1$ this just happens to be encoded in codiscrete categories.

Posted by: urs on December 11, 2006 5:38 PM | Permalink | Reply to this

Re: forgetting structure, remembering choice

Urs wrote:

… even less obvious in the way Toby defined them…

Okay, Toby — you’ve got to fix this while preparing your thesis for publication. Talk about descent!

Posted by: John Baez on December 28, 2006 8:06 AM | Permalink | Reply to this

Read the post Local Transition of Transport, Anafunctors and Descent of n-Functors
Weblog: The n-Category Café
Excerpt: Concepts and examples of what would be called transition data or descent data for n-functors.
Tracked: December 12, 2006 3:00 PM

Michael Makkai’s text on anafunctors

$MathML-enabled post (click for more details).$

On some systems the postscript files provided on Michael Makkai’s site don’t display properly. Since I was asked, I have now converted the PS files to PDFs, which do seem to be more well-behaved:

anafun1.pdf (title and contents)

anafun2.pdf (introduction)

anafun3.pdf (main text)

Since the original files are freely available, I guess it is okay that I provide the above links.

Posted by: urs on December 14, 2006 12:48 PM | Permalink | Reply to this

Re: Michael Makkai’s text on anafunctors

Since the original files are freely available, I guess it is okay that I provide the above links.

I’ve brought this discussion to Makkai’s attention already by email, so perhaps he will say something if he objects.

Posted by: Toby Bartels on December 14, 2006 11:21 PM | Permalink | Reply to this

Re: Michael Makkai’s text on anafunctors

As Makkai has pointed out to me by email, there is already a brief discussion of 2-anafunctors in his paper.

Posted by: Toby Bartels on December 25, 2006 1:14 PM | Permalink | Reply to this

Re: Michael Makkai’s text on anafunctors

$MathML-enabled post (click for more details).$

As Makkai has pointed out to me by email, there is already a brief discussion of 2-anafunctors in his paper.

I assume you are referring to what is called morphisms of anabicategories and anafunctors of anabicategories in the paper, the definition of which begins on p. 55 (part 4 of chapter 3).

This definition is given in the componentwise style in which also the first version of the definition of anafunctors themselves was presented. For the latter, there is the equivalent but more concise formulation in terms of spans. Is an equivalent definition of morphisms of anabicategories known which is analogous to the definition of anafunctors in terms of spans?

Posted by: urs on December 27, 2006 10:21 PM | Permalink | Reply to this

Re: Michael Makkai’s text on anafunctors

Is an equivalent definition of morphisms of anabicategories known which is analogous to the definition of anafunctors in terms of spans?

I’m sure that there is one, but if it’s not in Makkai’s paper, then it probably hasn’t been written down, perhaps never even fully worked out. So you should probably just develop your own idea and then compare it with Makkai’s componentwise definition afterwards.

Perhaps I should have quoted Makkai’s email more fully. I can’t get at it right now, but he said something like ‘The discussion was very brief and not very explicit.’. So hardly a reason to stop working on your own approach!

Posted by: Toby Bartels on December 28, 2006 5:49 AM | Permalink | Reply to this

Re: Michael Makkai’s text on anafunctors

$MathML-enabled post (click for more details).$

but if it’s not in Makkai’s paper

At least I haven’t seen it.

So you should probably just develop your own idea and then compare it with Makkai’s componentwise definition afterwards.

I am sort of on vacation currently, but I thought about this a little. I’d be surprised if the “descent of 2-functors” that I talked about would not give rise to a morphism of anabicategories as defined by Makkai. (The converse statement, if true, might be harder to see.)

Essentially, it amounts to noticing that the descent data of a 2-functor, while not ensuring that any two lifts with fixed enpoints of a given 1-morphism are equal, do precisely ensure that any two lifts are connected by a unique 2-isomorphism.

This should exactly mean that a 2-functor on these lifts restricts to a 1-anafunctor on the respective Hom-categories. And that, in turn, is the main aspect of M. Makkai’s definition of “anafunctor of anabicategoies”.

So hardly a reason to stop working on your own approach!

Okay, good! :-)

But I’d be very interested in learning more about what other poeple do and did in this direction.

Posted by: urs on December 28, 2006 2:28 PM | Permalink | Reply to this

Re: Michael Makkai’s text on anafunctors

$MathML-enabled post (click for more details).$

Urs wrote:

In this paper by Eugenia Cheng an equivalence between “ordinary” bicategories and opetopic bicategories is shown.

Doesn’t this indicate that opetopic bicategories are rather not like anabicategories?

No. Anafunctors are just like functors, given the axiom of choice. So, anabicategories are just like bicategories, given the axiom of choice. And, similarly, opetopic bicategories are just like bicategories, given the axiom of choice.

You’re interested in applications ‘internal to’ categories like $Top$ or $C^{∞}$ , where the axiom of choice fails. In this situation, I guess that opetopic bicategories will still be just like anabicategories. But, they’ll be different from ordinary bicategories.

You already know the following stuff, but in case anyone out there is falling behind:

A functor $f : X \to Y$ assigns to each object of $X$ an object of $Y$ . An anafunctor $f : X \to Y$ assigns to each object of $X$ an object of $Y$ up to canonical isomorphism.

Anafunctors show up all over the place in category theory: for example, the ‘product’ operation in a category with products is an anafunctor, since the product of objects is defined up to canonical isomorphism.

We can convert an anafunctor to a functor by choosing a specific object $f (x) \in Y$ for each object of $x \in X$ . The choice doesn’t matter much, because $f (x)$ is defined up to canonical isomorphism. So, we can get different functors from different choices, but they’re all naturally isomorphic.

However, to actually do the choice we typically need the axiom of choice!

In contexts where our categories have not a set of objects but (for example) a space of objects, the axiom of choice fails — so anafunctors become significantly more general than functors.

Posted by: John Baez on December 28, 2006 8:44 PM | Permalink | Reply to this

Re: Michael Makkai’s text on anafunctors

$MathML-enabled post (click for more details).$

I’m not sure what an ‘anabicategory’ is. But, I’ll guess that it’s like a bicategory, but where the composition functor:

$\circ : hom (x_{1}, x_{2}) \times hom (x_{2}, x_{3}) \to hom (x_{1}, x_{3})$

has been replaced by an anafunctor.

If so, it should be fairly similar to what James Dolan and I call simply a ‘2-category’ in our opetopic approach to n-categories. The main difference is that instead of treating binary composition and identities

$i : 1 \to hom (x_{1}, x_{1})$

as fundamental, we treat all $k$ -ary compositions

$\circ_{k} : hom (x_{1}, x_{2}) \times \dots \times hom (x_{k - 1}, x_{k}) \to hom (x_{1}, x_{k})$

on an equal footing.

I similarly suspect that an ‘anabifunctor’ may be a lot like what James and I called a ‘virtual functor between 2-categories’.

This is part of a much bigger story you probably don’t want to hear. Briefly, what we did is take Makkai’s anafunctor idea together with the theory of operads and use them to develop a general ‘opetopic’ definition of $n$ -category. Then, Makkai took our ideas and developed them into $n$ -categorical foundation of mathematics! But, for technical reasons, he used the word ‘multitope’ instead of ‘opetope’.

Most of this work will not be useful to you in the short run, because it focuses on general $n$ instead of $n = 2$ .

But, here’s something that focuses on the $n = 2$ case:

Eugenia Cheng, Opetopic bicategories: comparison with the classical case.

Tom Leinster may also have written some useful stuff about the $n = 2$ case.

(You’ll meet Eugenia and Tom the week after next!)

Posted by: John Baez on December 28, 2006 7:38 AM | Permalink | Reply to this

Re: Michael Makkai’s text on anafunctors

$MathML-enabled post (click for more details).$

I’m not sure what an ‘anabicategory’ is. But, I’ll guess that it’s like a bicategory, but where the composition functor […] has been replaced by an anafunctor.

Yes, exacxtly.

I similarly suspect that an “anabifunctor” […]

The main idea is, rather naturally, that a “morphism of anabicategories” is like a morphism of bicatgeories, but with functors on the Hom-categories replaced by anafunctors.

BTW, Makkai uses the term “anabifunctor” for something else, namely for a functor on a product category $A \times X$ that restricts to an anafunctor in each of its two arguments. The ana-version of a 2-functor he calls a “morphism/anafunctor of anabicategories”.

Just one instance of the general problem with “bi”. What is a “bitorsor”? A 2-sided torsor or a torsor for a 2-group? The former.

But then what is a “bigroupoid bitorsor”? Now this is both: the groupoid is categorified, hence is its torsor, which in addition is 2-sided.

The usage of the prefix “bi” here leads to ambiguous terminology.

[…] like what James and I called a ‘virtual functor between 2-categories’.

Oh, interesting. So: in your and my work on 2-connections we work out the descent data for a 2-functor with values in a 2-group.

This defines a 2-functor on 2-paths in the transition 2-groupoid of the chosen cover.

Does this define a “virtual functor between 2-categories”?

As I have mentioned in my last reply to Toby # it seems like this does define a “anafunctor of anabicategories”.

Eugenia Cheng, Opetopic Bicategories

Thanks for that! I’ll take a look at it.

Posted by: urs on December 28, 2006 2:52 PM | Permalink | Reply to this

The n-Category Café

Skip to the Main Content

December 8, 2006