I'm studying the book Etale cohomology and the Weil conjecture by Freitag, Kiehl and I have some questions on the subchapter introducing the machinery associating to an étale presheaf a sheaf (that is the "sheafification" procedure in étale world): see pages 11-13.

Let $A$ be a commutative, unital, Noetherian ring and $\operatorname{Et}(A)$ the category of étale extensions of $A$ (étale means that the extensions $A \to B$ are flat, unramified $A$-algebras of finite type).
We follow the notation from the book and call as an étale presheaf (of set, abelian groups,...; for sake of simplicity say sets) a covariant functor

$$ \mathcal{F}: \operatorname{Et}(A) \to (Set), (Ab),... $$

Now we want to associate to $\mathcal{F}$ another étale presheaf $\widetilde{\mathcal{F}}$ as follows:
Let $B \in \operatorname{Et}(A)$ and $\mathcal{B}:= (B \to B_i, i \in I) $ an étale cover of $B$, that's a finite family $(B \to B_i, i \in I) $ of étale maps , when the images of $\operatorname{Spec} B_i$ cover all elements of $\operatorname{Spec} B$.

(These form naturaly a category; a morphism $\mathcal{B} \to \mathcal{B}'$ between

$$ \mathcal{B}= (B \to B_i, i \in I), \mathcal{B}'= (B \to B_j', j \in J) $$

is given by a map $\sigma:J \to I$ inducing $B_{\sigma(j)} \to B_j'$ (datails can be worked out easily).

Now we associate to an etale cover $\mathcal{B}= (B \to B_i)$ the object $\mathcal{F}(\mathcal{B})$ defined as set of all $I$-tuples $(s_i)_{i \in I}$ with $s_i \in \mathcal{F}(B_i)$ such that all $s_i$ fulfil the compatibility condition,
that is for every $i,j$ should hold

$$f^l_i(s_i)= f^r_j(s_j),$$

where the involved maps are $f^l_i: \mathcal{F}(B_i) \to \mathcal{F}(B_i \otimes_B B_j)$ (respectively $f^r_j: \mathcal{F}(B_j) \to \mathcal{F}(B_i \otimes_B B_j)$) are naturally induced by family of canonical maps $ l_i: B_i \to B_i \otimes_B B_j, b_i \mapsto b_i \otimes 1$,
(resp $ r_j: B_i \to B_i \otimes_B B_j, b_i \mapsto b_i \otimes 1$ )

We define the presheaf $\widetilde{\mathcal{F}}$ as direct limit over all coverings of $B$

$$ \tilde{\mathcal{F}}(B):= \varinjlim_{\mathcal{B}} \mathcal{F}(\mathcal{B})$$

(details like checking thats this is a functor etc. I omit here)
Now we come to my Questions. The book states some very important facts without giving a reference. And I'm looking for proofs of several statements there.

Question 1a: It is stated (p. 12) that for every $B \in \operatorname{Et}(A)$ and etale cover $\mathcal{B}= (B \to B_i)$ the induced map $\widetilde{\mathcal{F}}(B) \to \widetilde{\mathcal{F}}(\mathcal{B})$ is injective (note that $\widetilde{\mathcal{F}}$ is in general still a presheaf). How can it be proved?

Question 1b: Futhermore it is claimed that if for all $B$ and $\mathcal{B}= (B \to B_i)$ the maps

$$ \mathcal{F}(B) \to \mathcal{F}(\mathcal{B}) $$

are injective, then $\widetilde{\mathcal{F}}$ is already an étale sheaf. Why that's true? (Rmk.: combined with statement of Question 1a this implies that the application of double tilde $\widetilde{\widetilde{\mathcal{F}}}$ gives a sheaf)

Question 2: Why is the double tilde functor (= sheafification as we saw above modulo the two questions) $\widetilde{\widetilde{\mathcal{F}}}$ is exact? That is if we apply if to short exact sequence $0 \to \mathcal{A} \to \mathcal{B} \to \mathcal{C} \to 0$ of étale presheaves we obtain a ses of sheaves?
(recall, that by definition a sequence $0 \to \mathcal{A} \to \mathcal{B} \to \mathcal{C} \to 0$ of presheaves on site $\operatorname{Et}(A)$ is ses iff $0 \to \mathcal{A}(B) \to \mathcal{B}(B) \to \mathcal{C}(B) \to 0$ is ses for every $B \in \operatorname{Et}(A)$.)
Is $\widetilde{\mathcal{F}}$ exact as functor between etale presheaves?


1 Answer 1


A good trick for answering questions of this type is that there are multiple foundational resources in étale cohomology. If a detail isn't explained in one you're reading, you can quickly check another one - the notation should hopefully be similar enough that you can transfer the proof over. In particular, the Stacks project tends to give a lot of details, and I found an answer to question 1 and a partial answer to question 2 there. (You should probably try to solve them as an exercise before doing this.)

The way I would say it.

Question 1a: Suppose we are given two elements of $\widetilde{\mathcal F}(B)$ whose image in $\widetilde{\mathcal F}(\mathcal B)$ is equal. We must show they are equal. By assumption, the two elements arise from covers $\mathcal A, \mathcal C$ of $\mathcal B$ and families of sections on these covers satisfying a compatibility condition. Their pullbacks to $\widetilde{\mathcal F}(\mathcal B)$ are given by the pullbacks of those families of sections to $\mathcal B \times_{B} \mathcal A$ and $\mathcal B\times_B \mathcal C$ respectively. If those are equal, then by the definition of direct limit, there is a common refinement where they are equal. But this common refinement is also a common refinement of $\mathcal A$ and $\mathcal C$ where our two families of sections are equal, so again by the definition of directly limit, the two starting elements of $\mathcal F(B)$ are equal.

Question 1b: Well, one must show that given a cover $U_1,\dots, U_n$ of $B$, the map from sections of $\widetilde{\mathcal F}$ on $B$ to compatible sections on $U_1,\dots, U_n$ is an isomorphism. So let's construct an inverse map. Given a compatible section of $\widetilde{\mathcal F}$ on a cover $U_1,\dots, U_n$, we obtain for each $i$ a cover $\mathcal U_i$ of $U_i$ and a compatible family of sections of $\mathcal F$ on that cover. We would like to make $\bigcup_i \mathcal U_i$ a cover of $B$ and this family of sections a compatible family. The only issue is that when we pass to an intersection $\mathcal U_i \times \mathcal U_j$, the pullbacks of the sections from $U_i$ and $U_j$ are only necessarily equal in $\widetilde{\mathcal F}$, not in $\mathcal F$. But given that $\mathcal F \to \widetilde{\mathcal F}$ is injective, we get the desired compatibility also in $\mathcal F$.

Question 2: No, $\tilde{F}$ is not exact as a functor between etale presheaves. To make a counterexample, take any exact sequence $0 \to \mathcal A \to \mathcal B \to \mathcal C \to 0$ of sheaves whose sequence of global sections $0 \to H^0( X, \mathcal A) \to H^0(X,\mathcal B)\to H^0(X,\mathcal C)$ is not right exact. Let $\mathcal C'$ be the quotient of $\mathcal B$ by $\mathcal A$ as presheaves, i.e. $\mathcal C'(U) = \mathcal B(U)/\mathcal A(U)$.

Then $0 \to \mathcal A \to \mathcal B \to \mathcal C' \to 0$ is a short exact sequence of presheaves. Its sheafification is simply $ \mathcal A \to \mathcal B \to \mathcal C$, which is a short exact sequence of sheaves, but not a short exact sequence of presheaves, since its global sections are not short exact.

Instead, the proof that sheafification is exact makes use of the fact that $\widetilde{\widetilde{\mathcal F}}$ is a left adjoint and therefore right exact, without proving right exactness for $\widetilde{\mathcal F}$.

  • $\begingroup$ Thank you very much for the answer. A remark: when you proved right exactness of $\widetilde{\widetilde{\mathcal F}}$ you exploted it's left adjointness (which $\widetilde{\mathcal F}$ not has). On the other hand following the proof in stacks.math.columbia.edu/tag/00WJ the proof of left exactness uses esentially just the category theoretically fact that finite limits commute with filtered colimits. But this can be also applied directly to $\widetilde{\mathcal F}$ because of it's construction. So left exactness works unproblematically already for $\widetilde{\mathcal F}$, right? $\endgroup$
    – user267839
    Aug 5, 2021 at 19:54
  • $\begingroup$ @katalaveino I believe so, yes. (And it's even not so hard to check directly, without calling on the category-theoretic fact, by expanding the definitions and going step by step.) $\endgroup$
    – Will Sawin
    Aug 5, 2021 at 19:57

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.