5
$\begingroup$

I want to prove that $0 \to F\to G\to H \to 0$ is an exact sequence of étale sheaves. I understand that it is enough to show that $0\to F_{\bar{x}}\to G_{\bar{x}}\to H_{\bar{x}}\to 0$ is exact at every geometric point $\bar{x}\to X$. Why is it then enough to show that $0\to F(U)\to G(U)\to H(U)\to 0$ is exact for every strictly Hensel local scheme $U$?

I ask this because I don't understand the proof of proposition 6.12 of Voevodsky's "Lecture notes on motivic cohomology". Specifically I'm trying to understand the first two paragraphs here http://www2.maths.ox.ac.uk/cmi/library/MotivicCohomologyold.pdf#page=52 and use that argument to understand the proof of theorem 7.20 ( Suslin's rigidity theorem ) here http://www2.maths.ox.ac.uk/cmi/library/MotivicCohomologyold.pdf#page=63. It seems the arguments are quite similar in that the book states that it is enough to show an isomorphism on stalks and then goes to show said isomorphism on the sheaf evaluated on some henselian schemes. Is this correct?

$\endgroup$
1
  • 2
    $\begingroup$ It isn't, unless you tell us something about the sheaves. By definition, the stalk of a sheaf F at a geometric point is the direct limit of F(U) as U ranges of the etale nhds of the geometric point. Sometimes you can move the limit past the F, in which case it becomes an inverse limit, whose value is the spec of the strictly local ring at the geometric point (a strictly henselian ring). $\endgroup$
    – anon
    Sep 20, 2014 at 23:59

1 Answer 1

5
$\begingroup$

Your statement is correct, in both cases, the exactness of the sequence is established by looking at the stalks. Strictly henselian local rings show up because the stalk of the structure sheaf on the étale site at a geometric point is the strict henselization of the corresponding local ring, cf. Section 4 of Milne's lecture notes on étale cohomology, or EGA IV.

To identify stalks of the sheaf with sections over the local ring, you need some condition - the limit needs to commute with the functor as mentioned in anon's comment. (This can be seen as a continuity condition). In both cases mentioned in the question, continuity works:

  • in the case of Prop 6.12 of the lecture notes on motivic cohomology, $F=\mathbb{Z}_{\operatorname{tr}}(T)$ for $T$ a scheme of finite type. Because of the finite type assumption, every correspondence $\operatorname{Spec}\mathcal{O}_{X,x}^{\operatorname{sh}}\to T$ is in fact represented by a correspondence $U\to T$ where $U$ is an étale neighbourhood of $x$ in $X$; therefore the stalk of $\mathbb{Z}_{\operatorname{tr}}(T)$ at $x$ can be computed as $\mathbb{Z}_{\operatorname{tr}}(T)(\operatorname{Spec}\mathcal{O}_{X,x}^{\operatorname{sh}})$. (See also Exercise 1.13 in the lecture notes.)

  • in the argument for the rigidity theorem, the functors that are being considered (i.e. presheaves with transfers) live on a category of schemes of finite type - in some sense, there is actually no definition of $F(\operatorname{Spec}\mathcal{O}_{X,x}^{\operatorname{sh}})$. So the functors are extended to local rings (or more general, essentially finite type schemes) by the definition $F(\operatorname{Spec}\mathcal{O}_{X,x}^{\operatorname{sh}})=\operatorname{colim}F(U)$ where $U$ runs over all étale neighbourhoods of $x$ in $X$. The necessary continuity problem is defined away... (As a side note: this is not possible in the original application of the rigidity theorem to K-theory because K-theory is defined for all rings. Fortunately, K-theory commutes with directed colimits as in the definition of the local rings, so the continuity requirement is again met.)

$\endgroup$
1
  • $\begingroup$ Thank you very much!, that was very helpful. I'm also confused about the usage of a similar argument in Suslin's Rigidity theorem. I just edited the question with more information. $\endgroup$
    – AT0
    Sep 21, 2014 at 10:13

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.