I am currently trying to understand Grothendieck Topologies and coverages and want to endow the category Prob, consisting of finite probability spaces and measure preserving maps, with a Grothendieck Topology. I have already tried endowing it with the trivial topology and dense topology but got no interesting results (yet). Is there an interesting topology that can be put onto Prob, or is there a general way to endow a given category with an interesting Grothendieck Topology?
-
3$\begingroup$ I don't know about Prob specifically, but I strongly suspect "is there a general way to endow a given category with an interesting Grothendieck topology?" is likely to be as hard to answer as "is there a general way to endow a given set with an interesting topology?". $\endgroup$– LSpiceCommented Oct 26, 2023 at 20:42
-
3$\begingroup$ It can help to first attempt to describe the topos of sheaves you want to end up with, and then work backwards to find the appropriate Grothendieck topology. The finite probability spaces will be the representable sheaves. So you can ask yourself: if I take the coproduct of two representable sheaves, do I want that the result is again a representable sheaf, and which one? The maps between sheaves will be "locally measure preserving maps" in some sense, similar to how maps between schemes are described locally by maps between affine schemes. $\endgroup$– Jens HemelaerCommented Oct 30, 2023 at 8:59
1 Answer
This answer is motivated by the related question here.
I think it can help to write $\mathbf{Prob}$ as a disjoint union of the categories $\mathbf{Prop}_\lambda$, with $\mathbf{Prop}_\lambda$ the full subcategory of spaces with total measure $\lambda$, with $\lambda \geq 0$. This works because for any measure-preserving map $f: X \to Y$, we have that $\mu(X)=\mu(Y)$.
You can also use rescaling to show that all categories $\mathbf{Prop}_\lambda$ with $\lambda\neq 0$ are isomorphic to each other. This doesn't work for $\lambda=0$, instead $\mathbf{Prop}_0$ is isomorphic (or at least equivalent) to the category of finite sets.
Defining a Grothendieck topology on $\mathbf{Prop}$ amounts to defining a Grothendieck topology on $\mathbf{Prop}_\lambda$ for each $\lambda \geq 0$. In this way, in order to classify the Grothendieck topologies on $\mathbf{Prop}$, it is enough to classify them on $\mathbf{Prop}_1$ and on $\mathbf{Prop}_0$.
I'm not sure which one of these two will be the most difficult. On $\mathbf{Prop}_1$ you have for example the chaotic, trivial and dense topologies, and since you already computed that the dense topology agrees with the atomic topology, they all have a concrete description.
Other than that, maybe you can use that any full subcategory $\mathcal{C} \subseteq \mathbf{Prop}_1$ defines a Grothendieck topology, see the example on slide 9 here (and more generally any Grothendieck topology on $\mathcal{C}$ then extends to one on $\mathbf{Prop}_1$). A full subcategory of $\mathbf{Prop}_1$ that looks particularly interesting to me is the one consisting of the spaces in which all points have measure $>0$.
-
1$\begingroup$ I did not know this was a way of constructing Grothendieck Topologies thanks! $\endgroup$– MaatCommented Nov 23, 2023 at 11:18