I’m almost done explaining why any presheaf category is an **elementary topos**, meaning that

• it has finite colimits;

• it has finite limits;

• it’s cartesian closed.

• it has a subboject classifier.

Last time I explained why such categories are cartesian closed; now let’s talk about the subobject classifier!

### Subobject classifiers

In the category of sets, 0-element and 1-element sets play special roles. The 0-element set is the initial object, and any 1-element set is a terminal object. But 2-element sets are also special! They let us talk about ‘truth values’.

We can take any such set, call it 2, and call its elements ‘true’ and ‘false’. Then subsets of any set correspond in a one-to-one way with functions

by associating to any subset the function taking the value ‘true’ on and ‘false’ elsewhere: its **characteristic function**.

The idea of a subobject classifier is simply to copy this idea in other categories. A subobject classifier, which we’ll call will serve as an ‘object of truth values’ in the category at hand. To make this precise, we’ll demand that subobjects of any object correspond bijectively to morphisms

That’s the idea. But to make the idea useful we need to say *how* subobjects of correspond to morphisms And of course before that we’ll need to say exactly what a subobject is!

In we can pick out a subset of using a monomorphism where I use the funny arrow to denote a monomorphism (affectionately known as a ‘mono’). The idea is that the *image* of is the subset of we care about. But two different monos can have the same image, so we have to be careful!

Suppose we have two different monos

and

in When do they have the same image? The answer is: precisely when there’s an isomorphism

such that

**Puzzle.** Prove this.

So, more generally, in any category, we shall say that two monos and are **isomorphic** iff there’s an isomorphism such that

**Definition.** A **subobject** of an object in a category is an isomorphism class of monos into

We will eventually study subobjects to see how much they behave like subsets—there’s a lot to say about that! But first let’s see what a subobject classifier is.

Again we return to the category to see how this should work. We need to know how subobjects of arise from functions For this we need to know which elements of get mapped by to ‘true’, so we need to specify which element of the set counts as ‘true’. We do this using a function

which of course is a mono. Then, given any function

we can take the pullback of along and get a mono

Please see diagram (2) in Section I.3 of Mac Lane and Moerdijk to see this pullback in all its glory.

The usual description of pullbacks in assures us that

and the map is the obvious inclusion.

We can copy this procedure in any category with pullbacks, leading to the definition we seek:

**Definition.** Let be a category with pullbacks. A **subobject classifier** is an object equipped with a mono

such that for any object there is a bijection between morphisms

and subobjects of given as follows: for any such morphism , we pull back along obtaining a mono

and then take the subobject of corresponding to this.

Here we are using a fact:

**Puzzle.** Show that the pullback of any mono is a mono.

Thus, is automatically a mono, because we are assuming is.

### Subobjects in presheaf categories

Let’s see what subobjects and subobject classifiers look like in presheaf categories. So now let be any category and let

be the category of presheaves on

To get started: what are monos in like? Remember from Part 5 pullbacks are computed pointwise in presheaf categories. Furthermore, monos can be defined in terms of pullbacks:

**Puzzle.** Show that a morphism is a mono iff its pullback along itself is the identity

This implies that determining whether a morphism in is a mono must also be a ‘pointwise’ matter: that is, one that you can check by looking at one object at a time. But you can prove this directly:

**Puzzle.** Let be a morphism between presheaves that is, a natural transformation between functors Show that is a mono iff for each

is a mono in that is, a one-to-one function.

Thus, people say a morphism of presheaves makes into a **subpresheaf** of when each function is one-to-one, so corresponds to a subset of It’s good to look at this in examples, like the case of graphs, where we get the concept of ‘subgraph’: one graph included in another.

Of course, we should remember that a subobject of really an *equivalence class* of monos. Two different monos and give the same subobject of if there’s an isomorphism between and that makes the obvious triangle commute. But sometimes people slack off and say is a subobject of if it’s equipped with a mono I was coming close to doing that in the last paragraph.

### Subobject classifiers in presheaf categories

Now we’re ready to understand the subobject classifier in a presheaf category. I’ll just tell you what it is. The subobject classifier in the presheaf category assigns to each object the set of ‘sieves’ on So what’s a sieve?

**Definition.** Given a category , a **sieve** on an object is a collection of morphisms to such that if is in the sieve and is any morphism, then is in the sieve.

In other words, a sieve is a collection of morphisms with a fixed target that’s closed under precomposition. The name ‘sieve’ should remind you that if a piece of grain can get through a sieve, any smaller piece of grain can also get through. You can think of as ‘smaller’ than in some sense.

Here’s a slick way to think about sieves. Remember that the Yoneda embedding

sends any object to a presheaf

called a **representable** presheaf.

Here’s the cool fact: a sieve on is just the same as a subobject of For each object it gives a subset of and for each morphism it gives a map from to namely the map sending each to

The subobject classifier in is a beautiful thing: it assigns to each object the set of all sieves on that object!

That is, for each object the set is the set of all sieves on But we also need to say what does to morphisms. Given a morphism the map

sends sieves on to sieves on as follows. For any sieve on we say a morphism is in if its composite with is in

We also need to describe

A terminal presheaf sends each object of to a one-element set, so must pick out an element of for each object in a natural way (it’s a natural transformation.)

In other words, must choose a sieve on for each in a natural way. The naturality condition here says that if a morphism is in the chosen sieve on iff its composite with is in the chosen sieve on

How does do this wonderful thing? Simple: for each object it chooses the sieve containing *all* morphisms to Then the naturality condition holds trivially.

Of course, we have to check that

really *is* a subobject classifier for Only this will let us really understand what we’ve done here.

I’ll talk about this more next time, perhaps focusing on examples to build up intuition. For now I recommend that you read Section I.4 of Mac Lane and Moerdijk’s book for a general proof—and also a look at some examples!

The series so far:

• Part 1: sheaves, elementary topoi, Grothendieck topoi and geometric morphisms.

• Part 2: turning presheaves into bundles and vice versa; turning sheaves into etale spaces and vice versa.

• Part 3: sheafification; the adjunction between presheaves and bundles.

• Part 4: direct and inverse images of sheaves.

• Part 5: why presheaf categories are elementary topoi: colimits and limits in presheaf categories.

• Part 6: why presheaf categories are elementary topoi: cartesian closed categories and why presheaf categories are cartesian closed.

• Part 7: why presheaf categories are elementary topoi: subobjects and subobject classifiers, and why presheaf categories have a subobject classifier.

• Part 8: an example: the topos of time-dependent sets, and its subobject classifier.

Typos: should be , and $\Omega(f)(S)$ should be .

Thanks — fixed!

Let’s look at an example of a presheaf topos, to see what various things I’ve been talking about actually look like—especially the subobject classifier.