Preliminaries: There are lots of variations on the settings in which we define sheaves. I am concerned with the details linking the general definition below to the Grothendieck topology it generates; particularly, the assertion that sheaves with respect to a coverage are the same as sheaves with respect to the generated Grothendieck coverage.

Let $\mathcal{C}$ be a (small) category. Do not assume $\mathcal{C}$ has pullbacks. Denote the Yoneda embedding of an object by $Yc$.

Descent can be phrased concisely in terms of subfunctors of Yoneda embeddings. We say, a presheaf $X$ satisfies descent with respect to a subfunctor $\iota \colon F \hookrightarrow Yc$ if and only if the precomposition $-\iota \colon \lbrack \mathcal{C}^{\operatorname{op}}, Set \rbrack (Yc,X) \rightarrow \lbrack \mathcal{C}^{\operatorname{op}}, Set \rbrack (F,X)$ is an isomorphism.

Any set of maps $f:=\{ f_i \colon c_i \rightarrow c \}_I$ generates a subfunctor $\iota_f \colon Y_f \hookrightarrow Yc$ by declaring $Y_f \bullet = \{ \alpha \colon \bullet \to c ~|~ \exists f_i ~ \alpha \mbox{ factors through } f_i \}$. In this manner, we can encode the usual "$X$ descends along the cover $f$" with "precomosition with $\iota_f$ is a isomorphism".

A site is a category equipped with a coverage $\mathcal{J}$ assigning to each object $c$ a collection of covering families $\mathcal{J}c$ which are sets of maps $f = \{ f_i \colon c_i \rightarrow c \}_I$ such that the covering families are supported under pullback i.e. $\forall f \in \mathcal{J}c,~\varphi \in \mathcal{C}(b,c)~ \exists g \in \mathcal{J}b: Y_g \hookrightarrow \varphi^*Y_f$ where $\varphi^*Y_f$ is the pullback of $Y_f \hookrightarrow Yc$ along composition with $\varphi$.

Then a sheaf with respect to the site $(\mathcal{C},\mathcal{J})$ is a presheaf $X \colon \mathcal{C}^{\operatorname{op}} \to Set$ satisfying descent for all the $Y_f$.

We can generate a Grothendieck coverage $\mathcal{G}$ from any coverage (for each $c$ intersect $\mathcal{G}_\lambda c$ all such coverages $\mathcal{G}_\lambda$ and note that the trivial coverage assigning all subsets of maps to $c$ as covering families is Grothendieck). What is not clear to me is how, "One can then show that for every coverage, there is a unique Grothendieck coverage having the same sheaves." as the nLab entry for coverage claims.


How is it that sheaves with respect to $\mathcal{G}$ are the same as those with respect to $\mathcal{J}$? In particular, given $X \in Sh(\mathcal{C},\mathcal{J})$, a cover $f \in \mathcal{J}$, and map $\varphi$; support under pullback gives us $g \in \mathcal{J}$ with:

$ Y_g \overset{\iota}{\hookrightarrow} \varphi^*Y_f \overset{\iota_{\varphi^*Y}}{\hookrightarrow} Yb$

That $X$ is a sheaf gives us precomposition with $\iota_g = \iota_{\varphi^*Y}\iota$ is an isomorphism. How does that imply precomposition with $\iota_{\varphi^*Y_f}$ is an isomorphism?

  • $\begingroup$ My guess is that the proof that shows that a pretopology generates a topology with the same sheaves also applies somewhat. The sheaf condition doesn't use closure under composition, so really you can ask the question for a pair of coverages, one containing the other: under what conditions do they have the same sheaves? $\endgroup$
    – David Roberts
    Apr 10, 2018 at 2:02
  • $\begingroup$ Sure. I am however, specifically asking about the relationship between support under pullback (the coverage $\mathcal{J}$) and stability under pullback (the generated Grothendieck coverage). $\endgroup$ Apr 10, 2018 at 13:03

1 Answer 1


Extend $g$ to any generating set for $\varphi^*F$, say $\widehat{f} := \{ b_k \overset{\widehat{f}_k}{\longrightarrow} b \} \supseteq g := \{ b_{k_j} \overset{\widehat{f}_{k_j}}{\longrightarrow} b \}$, we will need to demonstrate that a sheaf with respect to the coverage $\mathcal{J}$ must satisfy descent for $\varphi^*F$.

Let $(y_k) \in \prod_K Xb_k$ be a $\varphi^*F$-compatible tuple. Because $X$ descends along $Y_g$, and $(y_{k_j})$ is $G$-compatible, $\exists!y \in Xb$ such that $X\widehat{f}_{k_j}y = y_{k_j}$. The problem can be restated as: there may be $\widehat{f}_k$ not in $g$ and a priori we don't know that $X\widehat{f}_k y = y_k$.

Let $\widehat{f}_t \in \widehat{f} - g$. Support under pullback of $Y_g$ along $\widehat{f}_t$ guarantees there is a cover $Y_h \in \mathcal{J} b_t$ generated by elements of $\widehat{f}_t^*G$ whence we have commuting squares $\widehat{f}_t h_s = \widehat{f}_{k_{j_s}}\psi_s$ with h = $\{ h_s \}$. In particular note $(X\psi_s y_{k_{j_s}})$ is $Y_h$-compatible and therefore applying descent, we may conclude $\exists!z \in Xb_t$ such that $Xh_sz = X\psi_s y_{k_{j_s}}$.

But by $Y_g$-compatibility $\widehat{f}_t h_s = \widehat{f}_{k_{j_s}}\psi_s$, we know $Xh_s y_t = X\psi_s y_{k_{j_s}}$ and thus $z = y_t$. In other words, $X$ satisfies descent for $\varphi^*F$ as desired.

  • $\begingroup$ So, I figured it out eventually. Once I stopped believing it was true, I tried to prove it false and found this argument. $\endgroup$ May 10, 2018 at 0:23

Your Answer

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

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