Want to take part in these discussions? Sign in if you have an account, or apply for one below
Vanilla 1.1.10 is a product of Lussumo. More Information: Documentation, Community Support.
1 to 5 of 5
Let $\mathcal{E}$ be a local topos or sheaf topos over a concrete site and $X \in \mathcal{E}$ a concrete sheaf.
The concreteness-condition on a sheaf may be read as saying that $X$ has “enough points”, in a sense.
What about the slice topos $\mathcal{E}/X$? What can we say about its topos points? Under which conditions does it have enough?
Any slice topos of a topos with enough points has enough points. Proof: suppose $E$ has enough points. A point of $E/X$ is a point $e\colon Set \to E$ together with an element $x\in e^*(X)$. To prove that it has enough points, we must show that if $A,B\in E/X$ and $f:A\to B$ is a morphism in $E/X$ such that $(e,x)^*(f)$ is an isomorphism for every point $(e,x)$ of $E/X$, then $f$ is an isomorphism. But $(e,x)^*$ is given by applying $e^*$ and then taking the fiber over $x$, so if $(e,x)^*(f)$ is an isomorphism for all $e$ and $x$, then $e^*(f)$ is an isomorphism for all $e$, hence (since $E$ has enough points) $f$ is an isomorphism in $E$, hence also in $E/X$.
Thanks, Mike!
I have added some discussion to over-topos. I see the points of the form $(e,x)$, they are the composites
$(e,x) : Set \stackrel{x^*}{\to} Set/e^*(X) \stackrel{e/X}{\to} \mathcal{E}/X \,.$How do I see that every point of the slice topos arises this way?
But more importantly: in which generality can we assume $\mathcal{E}$ to have enough points in the first place? To talk about concrete sheaves we need a local topos. What can we say about local toposes having enough points? (Certainly they have one important canonical point, by definition.)
I need to think again about the standard example that I am interested in, the topos $Sh(CartSp) \simeq Sh(Mfd)$. It should have enough points, with one point $(n)$ per $n \in \mathbb{N}$ given by the stalk at any ordinary point in $\mathbb{R}^n$.
Together with the above this would also show where my intuition about concrete sheaves inducing “enough points” was wrong: the topos points of $Sh(CartSp)/X$ are not related to global points $* \to X$ but to stalks of maps of disks into $X$, hence manifestly do not distinguish between concrete and non-concrete $X$.
Is there maybe some other property of the slice $\mathcal{E}/X$ that witnesses the fact that $X$ is concrete?
Let’s see, if we start with a local topos $(p^* \dashv p_* \dashv p^!) : \mathcal{E} \to Set$ then we should maybe be looking at the canonical point $e_0 := (p_* \dashv p^!) : Set \to \mathcal{E}$ and the induced points of the special form
$(e_0, x) : Set \to \mathcal{E}/X \,.$Now these are really those given by the global points $x : * \to X$ of $X$. So concreteness of $X$ should say that $\mathcal{E}/X$ has in some sense enough of these $e_0$-points.
Hm…
Not sure if it helps, but just an observation: if $\mathcal{E}$ is a local topos and $X \in \mathcal{E}$ a concrete object, then the “global points”-monad corresponding to the adjunction $Set/p_*(X) \stackrel{\overset{p_*/X}{\leftarrow}}{\underset{p^!/X}{\to}} \mathcal{E}/X$ acts on an object $(A \to X) \in \mathcal{E}/X$ by concretifying $A$ relative to $X$:
by construction of the slice geometric morphism $(p_*/X \dashv p^!/X)$ we have that $p^!/X \circ p_*/X (A \to X)$ is the pullback $\tilde A$ in
$\array{ \tilde A &\to& p^! p_* A \\ \downarrow && \downarrow \\ X &\to& p^! p_* X } \,,$where the bottom morphism is the unit. By definition of concrete objects the bottom morphism is a mono precisely if $X$ is concrete. Since monos are stable under pullback, also the top morphism is a mono. Applying $p_*$ to the top morphism and using that $p^!$ is by definition full and faithful, we find that $\Gamma \tilde A \hookrightarrow \Gamma A$ is a mono. By the universal property of the unit this implies that $\tilde A \to p_! p^* \tilde A$ is a mono, hence that $\tilde A$ itself is concrete.
Now, on the objects $\tilde A \to X$ in the image of this “relative concretization”-monad it is certainly true that the $(e_0,x)$-points detect isomorphisms.
Hm…
How do I see that every point of the slice topos arises this way?
By pullback, it suffices to show that every section of $E/X\to E$ arises from a map $1\to X$ in E. You can regard that as a special case of the full embedding of internal locales in E into localic geometric morphisms over E, where X is regarded as a discrete internal locale in E.
1 to 5 of 5