Please ignore this, which hopefully should be a circle.
this is incomplete!
This post will largely consist of information which exists at nlab, and I will be a bit terse.
My goal here is to summarize the notions of a presheaf topos and a free category so as to use them later to do more interesting and original stuff.
Let \(\bC\) be a category.
The category of presheaves on \(\bC\) is \(\Set^{\bC^{\op}}\), i.e. the category of contravariant functors \(\bC^{\op} \rightarrow \Set\) and natural transformations between them.
The yoneda embedding is the functor \(\by:\bC \rightarrow \Set^{\bC^{\op}}\) sending \(X \mapsto \bC(-,X)\), and a functor in the essential image of the yoneda embedding is called representable.
Yoneda’s lemma implies that $\by$ is fully faithful.
For \(X \in \bC\), the category of subobjects of \(\bC\), called \(\Sub_\bC(X)\) is the full subcategory of the slice category \(\bC/X\) of bundles over \(X\), whose objects are the monics \(A \rightarrowtail X\).
Some elementary arguments yield that this is a poset, and morphisms correspond with factorization of monics \(A \rightarrowtail B \rightarrowtail X\).
In the case that \(\bC\) is locally small, taking the underlying set of this poset yields a presheaf \(\Sub:\bC^{\op} \rightarrow \Set\).
Suppose that \(\bC\) has a terminal object \(1\).
We say that a subobject classifier of \(\bC\) is a morphism \(\top:1 \rightarrow \Omega\) such that every monic \(A \rightarrowtail X\) is expressed as the pullback of a unique morphism along \(\top\);
diagramatically, this is expressed as
Now supposing that \(\bC\) is finitely cocomplete and locally small, this is equivalent to \(\Omega\) being a representing element for \(\Sub\), i.e. it is equivalent to requiring that there exists a natural isomorphism
$$\Sub(-) \simeq \bC(-,\Omega).$$
This is unique up to isomorphism by Yoneda’s lemma.
A topos is a cartesian closed, finitely cocomplete category possessing a subobject classifier.
We will care primarily about toposes as generalizations of \(\Set\) for doing logic, in a way which will become clear later.
We will give the relevant example of toposes now.
The presheaf topos
One early example of a topos is the presheaf category \(\Set^{\bC^{\op}}\).
This category is bicomplete and cartesian closed, as (co)limits and exponentials may be computed pointwise.
We will soon give the subobeject classifier for \(\Set^{\bC^{\op}}\) to verify that it is a topos, first characterizing what it must be in order for \(\Set^{\bC^{\op}}\) to have a subobject classifier.
Suppose such a subobject classifier \(\Omega\) exists;
then,
by Yoneda’s lemma, naturally in \(X\).
Hence characterizing subobjects reduces to characterizing subfunctors of \(\by X\).
Say that a sieve on \(X\) is a set of arrows \(S\) having codomain \(X\) such that, whenever \(g \in S\) and $gf$ is defined, we have \(gf \in S\).
This is sort of like a right ideal on the arrows with codomain \(X\).
There is a natural correspondence between sieves on \(X\) and subfunctors of \(\by X\), which gives the object map of the presheaf \(\Omega\).
Naturality gives that the restriction map of \(\Omega(X)\) is given by the pushforward map on \(\by X\), which is realized by the corresponding map of sieves.
Free categories
To consider a presheaf category, we need a category.
Why not choose a free one?
Let \(G:A \rightrightarrows N\) be a graph (by which, of course, I mean a quiver).
Then, the free category \(\Free(G)\) has as objects \(N\) and as morphisms \(\Free(G)(X,Y)\) the paths of finite (possibly zero) length from \(X\) to \(Y\) in \(G\), with composition given by concatenation.
The free category construction forms a left adjoint to the forgetful functor from categories to graphs.
Note that \(\Free(G)^{\op} \simeq \Free(G^{\op})\), where the latter graph has arrows reversed.
These are nicely structured;
for instance, note that every morphism in a free category is both epic and monic.
I’m more interested, however, in considering the presheaf topos on a free category.
Note that, presheaves on \(\Free(G)\) are equivalent to graphs homomorphisms \(G^{\op} \rightarrow \Set\);
that is, it is a set \(N(X)\) for each node \(X\) and a map \(N(Y) \rightarrow N(X)\) for each morphism \(f:X \rightarrow Y\) which is functorial.
We can come up with some examples;
note that \(\Set^{\Free\prn{\prn{\bullet \rightrightarrows \bullet}^{\op}}} = \Set^{\bullet \rightrightarrows \bullet}\) is the category of pairs of sets \(A,N\) with parallel arrows \(A \rightrightarrows N\);
that is, it forms the category of graph.
This was explored in this paper by Sebastiano Vigna.
For a similar feeling example, let \(\omega\) be the first countable ordinal, i.e. the natural numbers under the usual order \(\leq \subset \NN\times \NN\).
Note that \(\omega\) is a free category on the graph with nodes \(\NN\) and arcs \(i \rightarrow i+1\).
Using this, a presheaf on \(\omega\) corresponds with an \(\NN\)-indexed collection of sets \((X_i)_{i \in \NN}\) with morphisms \(p_i:X_i \rightarrow X_{i-1}\).
There is an equivalence of categories between \(\Set^{\omega^{\op}}\) and the category of forests of potentially infinite height;
the nodes at height \(i\) are given by \(X_i\), and the parent of a node \(x \in X_i\) is \(p(x_i)\).
For details, see this paper by Stefano Kasangian and Sebastiano Vigna.
On the next episode, I’ll go more in depth on some examples of what being a topos buys us for presheaves on particular free graphs.
In particular I’ll work some examples of topologies on presheaf toposes on free graphs, and we’ll see what those buy us.