7
$\begingroup$

It is well known that the category $\mathit{Sh}(X)$ of sheaves of sets on a topological space $ X $ is a topos. On the other hand, there exists a natural equivalence of categories between $\mathit{Sh}(X)$ and the category $\mathit{Et}(X)$ of étalé spaces over $ X $, so that the category of etale spaces over $ X $ is a topos, too. The constructions in $\mathit{Sh}(X)$ are translatable to $\mathit{Et}(X)$, and vice versa. For example, products of sheaves correspond to fiber products of étalé spaces. The description of the exponentials in $\mathit{Sh}(X)$ is given, e.g., in [Mac Lane and Moerdijk, Sheaves in Geometry and Logic]. The question that arises is

What is the explicit construction of exponentials in $Et(X)$?

$\endgroup$
2
  • 4
    $\begingroup$ Have you tried to compute this by yourself? All the ingridients are explictly available: the functors passing between sheaves and etale spaces, as well as the exponentials in sheaves. Where did you get stuck? $\endgroup$ Aug 30, 2020 at 18:53
  • $\begingroup$ It is basically a nonlinear variant of the well-known description of inner homs in the category of (geometric) vector bundles over X. $\endgroup$ May 31, 2021 at 23:30

2 Answers 2

3
$\begingroup$

Let $[E,E’]$ be the internal hom of two étale spaces $\pi : E \to X$ and $\pi’: E’ \to X$. I will write $\mathrm{Hom}_X(A,B)$ for the set of morphisms of étale spaces $A$ and $B$ over $X$, and $A \times_X B$ for the product of $A$ and $B$ as étale spaces over $X$.

To compute the sections of $[E,E’]$, you can use $$\mathrm{Hom}_X(U,[E,E’]) \simeq \mathrm{Hom}_X(U \times_X E, E’) \simeq \mathrm{Hom}_X(U \times_X E, U \times E’).$$ The first isomorphism uses the universal property of the exponential, the second isomorphism uses that the image of $U \times E$ is always contained in $(\pi’)^{-1}(U) =U \times E’$, so there is a unique factorization through $U \times E’$.

The elements of the total space correspond to stalks, so the fiber of $[E,E’]$ above $x \in X$ is given by the set $$[E,E’]_x \simeq \varinjlim_i \mathrm{Hom}_X(U_i \times_X E, U_i \times_X E’)$$ where $(U_i)_{i \in I}$ is a projective system of open neighborhoods of $x$ such that every open neighborhood of $x$ contains some $U_i$ (for example, if $X$ is a metric space, you can take open balls of shrinking radius).

A basis of open sets on $[E,E’] = \bigsqcup_x [E,E’]_x$ is given by the sets $A_{(U,s)} = \{ s_x : x \in U \}$, for $U$ in some basis of open sets of $X$, $s \in \mathrm{Hom}_X(U \times_X E, U \times_X E’)$, and $s_x$ the element of $[E,E’]_x$ corresponding to $s$. This topology is often not easy to visualize. The spaces $E$ and $E’$ are already typically non-Hausdorff, and even if they are Hausdorff there is no guarantee that $[E,E’]$ will be as well. But maybe it helps to keep in mind that the subspace topology on the fibers is always the discrete topology (as for any étale space).

For example, take $X= \mathbb{R}$, $E = \mathbb{R}-\{0\}$ and $E’ = \mathbb{R}\sqcup\mathbb{R}$ with $\pi$ and $\pi’$ the natural projections. We first look at $x \neq 0$. For a small open interval $U$ containing $x\neq 0$, there are exactly two maps $E \times_X U \to E’ \times_X U$ (because $E \times_X U \simeq U$ and $E’ \times_X U \simeq U \sqcup U$). However, if $U$ is a small open interval containing $x=0$, then $E \times_X U = U -\{0\}$ has two components, and because of this there are four maps to $E’ \times_X U \simeq (U-\{0\}) \sqcup (U - \{0\})$.

So in this example, $[E,E’]$ has four points above $x=0$ and two points above each $x \neq 0$. Further, $[E,E’]$ has four global sections corresponding to the four continuous maps $E \to E’$ over $X$. This gives four open sets $A_{(X,s)}$ in $[E,E’]$, with each of these homeomorphic to $\mathbb{R}$. But these four lines are glued together to a single line on $\{ x \in \mathbb{R} : x > 0\}$ iff the corresponding sections agree on $\{ x \in \mathbb{R} : x > 0\}$, and similarly over $\{x \in \mathbb{R} : x < 0 \}$.

Here is a picture:

The exponential [E,E']

The gluing of different lines is supposed to be “instant” (just as in the case of the line with two origins) and as a result the space is non-Hausdorff.

$\endgroup$
0
$\begingroup$

Depending on what you mean by "explicit", the following answer will be satisfying or not:

Let $F : {\sf Et}(X) \leftrightarrows {\sf Sh}(X) : G$ be the equivalence in subject. Let $[-,-]$ be the exponential of sheaves, i.e. on presheaves, i.e. the sheaf $U\mapsto {\sf Sh}(X)(yU \times A,B)$ (if I remember well, it is enough that $B$ is a sheaf for this to be a sheaf). Then, the exponential $\langle -,-\rangle$ on étale spaces is the functor $(E,E')\mapsto G[FE, FE']$.

Indeed, let's prove that ${\sf Et}(X)(E\times E', E'')\cong {\sf Et}(X)(E, \langle E',E''\rangle)$: $$ \begin{align*} {\sf Et}(X)(E\times E', E'') & \cong {\sf Sh}(X)(F(E\times E'), FE'') \\ & \cong {\sf Sh}(X)(FE\times FE', FE'')\\ & \cong {\sf Sh}(X)(FE, [FE', FE''])\\ & \cong {\sf Et}(X)(E, G[FE', FE'']) \\ & = {\sf Et}(X)(E, \langle E', E''\rangle). \end{align*} $$ I have used the following facts: 1. $F$ preserves products (and for that matter, all limits: it is an equivalence). 2. ${\sf Sh}(X)$ is cartesian closed.

$\endgroup$
6
  • $\begingroup$ By explicit I mean an object in $Et(X)$ that represents the exponential of $E,E'\in Et(X)$. $\endgroup$
    – ARA
    Apr 3, 2020 at 8:22
  • $\begingroup$ So what you don't like in my answer? $F,G$ and $[-,-]$ are explicitly determined(in McLM, or in other books on sheaf theory). $F$ sends an étale space to its sheaf of sections. $G$ is more difficult to write down properly, but you can do it with a little effort. ${\sf Et}(X)$ is cocomplete, as well as $G$, so it is enough to define it on representables in ${\sf Sh}(X)$ and then take a Kan extension. $\endgroup$
    – fosco
    Apr 3, 2020 at 11:28
  • $\begingroup$ In your answer, what I like to know is just the description of the etale space $ G[FE, FE'] $ with respect to $E$ and $E'$. At least, what is the total space of $ G[FE, FE'] $? $\endgroup$
    – ARA
    Apr 3, 2020 at 13:33
  • 3
    $\begingroup$ The total space of $G[FE,FE']$ is the disjoint union of all stalks $[FE, FE']_x$ at points of $X$. Properly topologized, this is étale over $X$. $\endgroup$
    – fosco
    Apr 3, 2020 at 14:16
  • 5
    $\begingroup$ @Fosco: I think the OP is asking precisely for the description you indicated in the comment (stalks). It would probably be helpful if you took thre trouble of amending your answer. which right now looks a bit like you're trying to play a joke on the OP. $\endgroup$ Aug 31, 2020 at 7:42

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.