## This Week's Finds in Mathematical Physics (Week 75)

#### John Baez

The definition of n-categories - or maybe I should say the problem of defining n-categories - is actually surprisingly subtle. Since I want to proceed at a gentle pace here, I think I should first get everyone used to the 2-category Cat, then define 2-categories in general, then play around with those a bit, and then move on to n-categories for higher n.

So recall what the objects, morphisms and 2-morphisms in Cat are: they are categories, functors and natural transformations. A functor F: C → D assigns to each object x of C an object F(x) of D, and to each morphism f of C a morphism F(f) of D, and has

1. If f: x → y, then F(f): F(x) → F(y).
2. If fg = h, then F(f)F(g) = F(h).
3. If 1x is the identity morphism of x, then F(1x) is the identity morphism of F(x).

Given two functors F: C → D and G: C → D, a "natural transformation" T: F → G assigns to each object x of C a morphism Tx: F(x) → G(x), such that for any morphism f: x → y in C, the diagram

```                               F(f)
F(x) -----> F(y)
|           |
Tx|           |Ty
v           v
G(x) -----> G(y)
G(f)

```
commutes.

Let me give a nice example. Let Top be the category with topological spaces and continuous functions between them as morphisms. Let Gpd be the category with groupoids as objects and functors between them as morphisms. (Remember from "week74" that a groupoid is a category with all morphisms invertible.) Then there is a functor

Π1: Top → Gpd

called the "fundamental groupoid" functor, which plays a very basic role in algebraic topology.

Here's how we get from any space X its "fundamental groupoid" Π1(X). (If perchance you already know about the "fundamental group", fear not, what we're doing now is very similar.) To say what the groupoid Π1(X) is, we need to say what its objects and morphisms are. The objects in Π1(X) are just the POINTS of X and the morphisms are just certain equivalence classes of PATHS in X. More precisely, a morphism f: x → y in Π1(X) is just an equivalence class of continuous paths from x to y, where two paths from x to y are decreed equivalent if one can be continuously deformed to the other while not moving the endpoints. (If this equivalence relation holds we say the two paths are "homotopic", and we call the equivalence classes "homotopy classes of paths.")

This is a truly wonderful thing! Recall the idea behind categories. A morphism f: x → y is supposed to be some abstract sort of "process going from x to y." The human mind often works by visual metaphors, and our visual image of a "process" from x to y is some sort of "arrow" from x to y:

```                         f
x ----------------> y

```
That's why we write f: x → y, of course. But now what we are doing is taking this visual metaphor literally! We have a space X, like the piece of the computer screen on which you are actually reading this text. The objects in Π1(X) are then points in X, and a morphism is basically just a path from x to y:
```                         f
x ----------------> y !

```
Well, not quite; it's a homotopy class of paths. But still, something very interesting is going on here: we are turning something "concrete", namely the space X, into a corresponding "abstract" thing, namely the groupoid Π1(X), by thinking of "concrete" paths as "abstract" morphisms. Here I am thinking of geometrical concepts as "concrete", and algebraic ones as "abstract".

You may wonder why we use homotopy classes of paths, rather than paths. One reason is that topologists want to use Π1 to distill a very abstract "essence" of the topological space X, an essence that conveys only information that's invariant under "homotopy equivalence". Another reason is that we can easily get homotopy classes of paths to compose associatively, as they must if they are to be morphisms in a category. We simply glom them end to end:

```                         f                 g
x ----------------> y ---------------> z

```
and there is no problem with associativity, since we can easily reparametrize the paths to make (fg)h = f(gh). (If you haven't thought about this, please do!) If we do not work with homotopy classes, it's a pain to define composition in such a way that (fg)h = f(gh). Unless you are sneaky, you only get that (fg)h is homotopic to f(gh); there are ways to get composition to come out associative, but they are all somewhat immoral. As we shall see, if we want to preserve the "concreteness" of X as much as possible, and work with morphisms that are actual paths in X rather than homotopy equivalence classes, the best thing is to work with n-categories. More on that later.

Let's see; I claimed there is a functor Π1: Top → Gpd, but so far I have only defined Π1 on the objects of Top; we also need to define it for morphisms. That's easy. A morphism F: X → Y in Top is a continuous map from the space X to the space Y; this is just what we need to take points in X to points in Y, and homotopy classes of paths in X to homotopy classes of paths in Y. So it gives us a morphism in Gpd from the fundamental groupoid Π1(X) to the fundamental groupoid Π1(Y). There are various things to check here, but it's not hard. We call this morphism Π1(F): Π1(X) → Π1(Y). With a little extra work, we can check that Π1: Top → Gpd, defined this way, is really a functor.

Perhaps you are finding this confusing. If so, it could easily be because there are several levels of categories and such going on here. For example, Π1(X) is a groupoid, and thus a category, so there are morphisms like f: x → y in it; but on the other hand Gpd itself is a category, and there are morphisms like Π1(F): Π1(X) → Π1(Y) in it, which are functors! If you find this confusing, take heart. Getting confused this way is crucial to learning n-category theory! After all, n-category theory is all about how every "process" is also a "thing" which can undergo higher-level "processes". Complex, interesting structures emerge from very simple ones by the interplay of these different levels. It takes work to mentally hop up and down these levels, and to weather the inevitable "level slips" one makes when one screws up. If you expect it to be easy and are annoyed when you mess up, you will hate this subject. When approached in the right spirit, it is very fun; it teaches one a special sort of agility. (We are, by the way, nowhere near the really tricky stuff yet.)

Okay, so we have seen an interesting example of a functor

Π1: Top → Gpd.

As I said, we can think of this as going from the concrete world of spaces to the abstract world of groupoids, turning concrete paths into abstract "morphisms". Wonderfully, there is a kind of "reverse" functor,

K: Gpd → Top

which turns the abstract into the concrete, by making abstract morphisms into concrete paths! Basically, it goes like this. Say we have a groupoid G. We will build the space K(G) out of simplices as follows. Start with one 0-simplex for each object in G. A 0-simplex is simply a point. We can draw the 0-simplex for the object x as follows:

```                             x

```
Then put in one 1-simplex for each morphism in G. A 1-simplex is just a line segment. We can draw the 1-simplex for the morphism f: x → y as follows:
```                           x--f--y

```
Really we should draw an arrow going from left to right, but soon things will get too messy if I do that, so I won't. Then, whenever we have fg = h in the groupoid, we stick in a 2-simplex. A 2-simplex is just a triangle and we visualize it as follows:
```
y
/ \                       f: x → y
f   g                      g: y → z
/     \                     h: x → z
x---h---z

```
Then, whenever we have fgh = k in the groupoid, we stick in a 3-simplex, which we can visualize as a tetrahedron like this
```
x
/|\
/ | \
/  g  \                   f: w → x
f   |   gh                 g: x → y
/   _y_   \                 h: y → z
/   /   \_  \                k: w → z
/ _fg      h_ \
/_/           \_\
w--------k--------z

```
And so on... we do this forever and get a big "simplicial complex," which we can think of as the topological space KG. The reader might want to compare "week70", where do the same thing for a monoid instead of a groupoid. Really, one can do it for any category.

That's how we define K on objects; it's not hard to define K on morphisms too, so we get

K: Gpd → Top

In case you study this in more detail at some point, I should add that folks often think of simplicial complexes as somewhat abstract combinatorial objects in their own right, and then they break down K into two steps: first they take the "nerve" of a groupoid and get a simplicial complex, and then they take the "geometrical realization" of the simplicial complex to get a topological space. For more on simplicial complexes and the like, try:

1) J. P. May, Simplicial Objects in Algebraic Topology, Van Nostrand, Princeton, 1968.

Now, in what sense is the functor K: Gpd → Top the "reverse" of the functor Π1: Top → Gpd? Is it just the "inverse" in the traditional sense? No! It's something more subtle. As we shall see, the fact that Cat is a 2-category means that a functor can have a more subtle and interesting sorts of "reverse" than one might expect if one were used to the simple "inverse" of a function. This is something I alluded to in "week74": inverses become subtler as we march up the n-categorical hierarchy.

I'll talk about this more later. But let me just drop a teaser... Quantum mechanics is all about Hilbert spaces and linear operators between them. Given any (bounded) linear operator F: H → H' from one Hilbert space to another, there is a subtle kind of "reverse" operator, called the "adjoint" of F and written F*: H' → H, defined by

<x,F*y> = <Fx,y>

for all x in H and y in H'. This is not the same as the "inverse" of F; indeed, it exists even if F is not invertible. This sort of "reverse" operator is deeply related to the "reverse" functors I am hinting at above, and for this reason those "reverse" functors are also called "adjoints". This is part of a profound and somewhat mysterious relationship between quantum theory and category theory... which I eventually need to describe.