MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $X$ be a smooth variety of finite type over a number field $k$. Let $\overline{X} = X \times_{k} \overline{k}$, and let $\ell$ be a prime. It's well known that if $X$ is proper, then the étale cohomology groups $H^i_{et}(\overline{X}, \mathbb{Z}_{\ell})$ are unramified at any prime $\mathfrak{p} \nmid \ell$ at which $X$ has good reduction (and in fact are isomorphic as representations of $\operatorname{Gal}(\overline{K}_\mathfrak{p} / K_\mathfrak{p})$ to the étale cohomology groups of the special fibre).

Does this statement also hold if $X$ is not assumed to be proper? (I'm interested in the case of smooth affine varieties.) What about the weaker statement that $H^i_{et}(\overline{X}, \mathbb{Z}_{\ell})$ is unramified almost everywhere?

(I don't know a reference for the proof of the "well known" statement -- I couldn't find it in Milne's books or in SGA 4.5.)

share|cite|improve this question
This is just in reference to your parenthetical comment. For proper smooth maps, the unramifiedness is a consequence of the general fact that local constancy for $\ell$-adic sheaves is stable under push-forward by proper smooth maps. This is Thm 20.2 in Milne's online notes. – Keerthi Madapusi Pera Jun 18 '12 at 17:36
@Keerthi: Can you elaborate a little on why unramifiedness follows from this pushforward result? – David Loeffler Jun 18 '12 at 18:42
Thanks! This seems to use the properness in a completely essential way, though. – David Loeffler Jun 18 '12 at 19:22
I think the weaker statement is true, but for the main question it depends what you mean by "good reduction". Could you specify? – Dustin Clausen Jun 18 '12 at 22:16
Yes, you really need to say what you mean by good reduction. If you mean "the generic fiber of a smooth morphism", then every smooth variety has this property (just take a model over the good primes), so anything with ramified cohomology is a counterexample. If you mean the complement of a proper flat closed subscheme in a smooth proper scheme, then deleting a ramified closed point from the projective line should give a counterexample. If you mean the complement of a smooth proper closed subscheme in a smooth proper scheme, then it's probably true. – JBorger Jun 19 '12 at 1:02
up vote 9 down vote accepted

I think the weaker statement should be true. Here's a sketch of an argument: by compactification theorems and resolution of singularities, there is a smooth proper scheme $Y$ over $k$ containing $X$ as an open subscheme, such that $Y \setminus X$ is a divisor $D$ with simple normal crossings. Let $D_1, \dots, D_r$ be the irreducible components of $D$. Then any $p$-fold intersection of the $D_i$'s is smooth and proper over $k$.

There should be a spectral sequence, in terms of the etale cohomology of $\overline{Y}$ and that of the intersections of the $D_i$'s, that abuts to the etale cohomology of $\overline{X}$. Thus the etale cohomology of $\overline{X}$ should be unramified at any prime of good reduction for $\overline{Y}$ and all of the intersections of the $D_i$'s. I imagine you could also use this to show that at such primes the cohomology of $\overline{X}$ was isomorphic to the cohomology of the reduction.


share|cite|improve this answer
So what's the definition of 'good reduction' used in this answer? – JBorger Jun 19 '12 at 7:36
The answer only considers "good reduction" for $Y$ and the intersections of the $D_i$'s which are all proper, so we know what good reduction means there, don't we? – David Loeffler Jun 20 '12 at 10:56
Oh, sorry. I missed that this was an answer to the weaker question. FYI it's also in SGA 4.5. It follows from constructibility of higher direct images. – JBorger Jun 21 '12 at 9:44

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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