The Stacks project

59.75 Specializations and étale sheaves

Topological picture: Let $X$ be a topological space and let $x' \leadsto x$ be a specialization of points in $X$. Then every open neighbourhood of $x$ contains $x'$. Hence for any sheaf $\mathcal{F}$ on $X$ there is a specialization map

\[ sp : \mathcal{F}_ x \longrightarrow \mathcal{F}_{x'} \]

of stalks sending the equivalence class of the pair $(U, s)$ in $\mathcal{F}_ x$ to the equivalence class of the pair $(U, s)$ in $\mathcal{F}_{x'}$; see Sheaves, Section 6.11 for the description of stalks in terms of equivalence classes of pairs. Of course this map is functorial in $\mathcal{F}$, i.e., $sp$ is a transformation of functors.

For sheaves in the étale topology we can mimic this construction, see [Exposee VIII, 7.7, page 397, SGA4]. To do this suppose we have a scheme $S$, a geometric point $\overline{s}$ of $S$, and a geometric point $\overline{t}$ of $\mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{s}})$. For any sheaf $\mathcal{F}$ on $S_{\acute{e}tale}$ we will construct the specialization map

\[ sp : \mathcal{F}_{\overline{s}} \longrightarrow \mathcal{F}_{\overline{t}} \]

Here we have abused language: instead of writing $\mathcal{F}_{\overline{t}}$ we should write $\mathcal{F}_{p(\overline{t})}$ where $p : \mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{s}}) \to S$ is the canonical morphism. Recall that

\[ \mathcal{F}_{\overline{s}} = \mathop{\mathrm{colim}}\nolimits _{(U, \overline{u})} \mathcal{F}(U) \]

where the colimit is over all étale neighbourhoods $(U, \overline{u})$ of $(S, \overline{s})$, see Section 59.29. Since $\mathcal{O}^{sh}_{S, \overline{s}}$ is the stalk of the structure sheaf, we find for every étale neighbourhood $(U, \overline{u})$ of $(S, \overline{s})$ a canonical map $\mathcal{O}_{U, u} \to \mathcal{O}^{sh}_{S, \overline{s}}$. Hence we get a unique factorization

\[ \mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{s}}) \to U \to S \]

If $\overline{v}$ denotes the image of $\overline{t}$ in $U$, then we see that $(U, \overline{v})$ is an étale neighbourhood of $(S, \overline{t})$. This construction defines a functor from the category of étale neighbourhoods of $(S, \overline{s})$ to the category of étale neighbourhoods of $(S, \overline{t})$. Thus we may define the map $sp : \mathcal{F}_{\overline{s}} \to \mathcal{F}_{\overline{t}}$ by sending the equivalence class of $(U, \overline{u}, \sigma )$ where $\sigma \in \mathcal{F}(U)$ to the equivalence class of $(U, \overline{v}, \sigma )$.

Let $K \in D(S_{\acute{e}tale})$. With $\overline{s}$ and $\overline{t}$ as above we have the specialization map

\[ sp : K_{\overline{s}} \longrightarrow K_{\overline{t}} \quad \text{in}\quad D(\textit{Ab}) \]

Namely, if $K$ is represented by the complex $\mathcal{F}^\bullet $ of abelian sheaves, then we simply that the map

\[ K_{\overline{s}} = \mathcal{F}^\bullet _{\overline{s}} \longrightarrow \mathcal{F}^\bullet _{\overline{t}} = K_{\overline{t}} \]

which is termwise given by the specialization maps for sheaves constructed above. This is independent of the choice of complex representing $K$ by the exactness of the stalk functors (i.e., taking stalks of complexes is well defined on the derived category).

Clearly the construction is functorial in the sheaf $\mathcal{F}$ on $S_{\acute{e}tale}$. If we think of the stalk functors as morphisms of topoi $\overline{s}, \overline{t} : \textit{Sets} \to \mathop{\mathit{Sh}}\nolimits (S_{\acute{e}tale})$, then we may think of $sp$ as a $2$-morphism

\[ \xymatrix{ \textit{Sets} \rrtwocell ^{\overline{t}}_{\overline{s}}{\ sp} & & \mathop{\mathit{Sh}}\nolimits (S_{\acute{e}tale}) } \]

of topoi.

Remark 59.75.1 (Alternative description of sp). Let $S$, $\overline{s}$, and $\overline{t}$ be as above. Another way to describe the specialization map is to use that

\[ \mathcal{F}_{\overline{s}} = \Gamma (\mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{s}}), p^{-1}\mathcal{F}) \quad \text{and}\quad \mathcal{F}_{\overline{t}} = \Gamma (\overline{t}, \overline{t}^{-1}p^{-1}\mathcal{F}) \]

The first equality follows from Theorem 59.53.1 applied to $\text{id}_ S : S \to S$ and the second equality follows from Lemma 59.36.2. Then we can think of $sp$ as the map

\[ sp : \mathcal{F}_{\overline{s}} = \Gamma (\mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{s}}), p^{-1}\mathcal{F}) \xrightarrow {\text{pullback by }\overline{t}} \Gamma (\overline{t}, \overline{t}^{-1}p^{-1}\mathcal{F}) = \mathcal{F}_{\overline{t}} \]

Remark 59.75.2 (Yet another description of sp). Let $S$, $\overline{s}$, and $\overline{t}$ be as above. Another alternative is to use the unique morphism

\[ c : \mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{t}}) \longrightarrow \mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{s}}) \]

over $S$ which is compatible with the given morphism $\overline{t} \to \mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{s}})$ and the morphism $\overline{t} \to \mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{t, \overline{t}})$. The uniqueness and existence of the displayed arrow follows from Algebra, Lemma 10.154.6 applied to $\mathcal{O}_{S, s}$, $\mathcal{O}^{sh}_{S, \overline{t}}$, and $\mathcal{O}^{sh}_{S, \overline{s}} \to \kappa (\overline{t})$. We obtain

\[ sp : \mathcal{F}_{\overline{s}} = \Gamma (\mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{s}}), \mathcal{F}) \xrightarrow {\text{pullback by }c} \Gamma (\mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{t}}), \mathcal{F}) = \mathcal{F}_{\overline{t}} \]

(with obvious notational conventions). In fact this procedure also works for objects $K$ in $D(S_{\acute{e}tale})$: the specialization map for $K$ is equal to the map

\[ sp : K_{\overline{s}} = R\Gamma (\mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{s}}), K) \xrightarrow {\text{pullback by }c} R\Gamma (\mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{t}}), K) = K_{\overline{t}} \]

The equality signs are valid as taking global sections over the strictly henselian schemes $\mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{s}})$ and $\mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{t}})$ is exact (and the same as taking stalks at $\overline{s}$ and $\overline{t}$) and hence no subtleties related to the fact that $K$ may be unbounded arise.

Remark 59.75.3 (Lifting specializations). Let $S$ be a scheme and let $t \leadsto s$ be a specialization of point on $S$. Choose geometric points $\overline{t}$ and $\overline{s}$ lying over $t$ and $s$. Since $t$ corresponds to a point of $\mathop{\mathrm{Spec}}(\mathcal{O}_{S, s})$ by Schemes, Lemma 26.13.2 and since $\mathcal{O}_{S, s} \to \mathcal{O}^{sh}_{S, \overline{s}}$ is faithfully flat, we can find a point $t' \in \mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{s}})$ mapping to $t$. As $\mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{s}})$ is a limit of schemes étale over $S$ we see that $\kappa (t')/\kappa (t)$ is a separable algebraic extension (usually not finite of course). Since $\kappa (\overline{t})$ is algebraically closed, we can choose an embedding $\kappa (t') \to \kappa (\overline{t})$ as extensions of $\kappa (t)$. This choice gives us a commutative diagram

\[ \xymatrix{ \overline{t} \ar[d] \ar[r] & \mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{s}}) \ar[d] & \overline{s} \ar[l] \ar[d] \\ t \ar[r] & S & s \ar[l] } \]

of points and geometric points. Thus if $t \leadsto s$ we can always “lift” $\overline{t}$ to a geometric point of the strict henselization of $S$ at $\overline{s}$ and get specialization maps as above.

Lemma 59.75.4. Let $g : S' \to S$ be a morphism of schemes. Let $\mathcal{F}$ be a sheaf on $S_{\acute{e}tale}$. Let $\overline{s}'$ be a geometric point of $S'$, and let $\overline{t}'$ be a geometric point of $\mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S', \overline{s}'})$. Denote $\overline{s} = g(\overline{s}')$ and $\overline{t} = h(\overline{t}')$ where $h : \mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S', \overline{s}'}) \to \mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{s}})$ is the canonical morphism. For any sheaf $\mathcal{F}$ on $S_{\acute{e}tale}$ the specialization map

\[ sp : (f^{-1}\mathcal{F})_{\overline{s}'} \longrightarrow (f^{-1}\mathcal{F})_{\overline{t}'} \]

is equal to the specialization map $sp : \mathcal{F}_{\overline{s}} \to \mathcal{F}_{\overline{t}}$ via the identifications $(f^{-1}\mathcal{F})_{\overline{s}'} = \mathcal{F}_{\overline{s}}$ and $(f^{-1}\mathcal{F})_{\overline{t}'} = \mathcal{F}_{\overline{t}}$ of Lemma 59.36.2.

Proof. Omitted. $\square$

Lemma 59.75.5. Let $S$ be a scheme such that every quasi-compact open of $S$ has finite number of irreducible components (for example if $S$ has a Noetherian underlying topological space, or if $S$ is locally Noetherian). Let $\mathcal{F}$ be a sheaf of sets on $S_{\acute{e}tale}$. The following are equivalent

  1. $\mathcal{F}$ is finite locally constant, and

  2. all stalks of $\mathcal{F}$ are finite sets and all specialization maps $sp : \mathcal{F}_{\overline{s}} \to \mathcal{F}_{\overline{t}}$ are bijective.

Proof. Assume (2). Let $\overline{s}$ be a geometric point of $S$ lying over $s \in S$. In order to prove (1) we have to find an étale neighbourhood $(U, \overline{u})$ of $(S, \overline{s})$ such that $\mathcal{F}|_ U$ is constant. We may and do assume $S$ is affine.

Since $\mathcal{F}_{\overline{s}}$ is finite, we can choose $(U, \overline{u})$, $n \geq 0$, and pairwise distinct elements $\sigma _1, \ldots , \sigma _ n \in \mathcal{F}(U)$ such that $\{ \sigma _1, \ldots , \sigma _ n\} \subset \mathcal{F}(U)$ maps bijectively to $\mathcal{F}_{\overline{s}}$ via the map $\mathcal{F}(U) \to \mathcal{F}_{\overline{s}}$. Consider the map

\[ \varphi : \underline{\{ 1, \ldots , n\} } \longrightarrow \mathcal{F}|_ U \]

on $U_{\acute{e}tale}$ defined by $\sigma _1, \ldots , \sigma _ n$. This map is a bijection on stalks at $\overline{u}$ by construction. Let us consider the subset

\[ E = \{ u' \in U \mid \varphi _{\overline{u}'}\text{ is bijective}\} \subset U \]

Here $\overline{u}'$ is any geometric point of $U$ lying over $u'$ (the condition is independent of the choice by Remark 59.29.8). The image $u \in U$ of $\overline{u}$ is in $E$. By our assumption on the specialization maps for $\mathcal{F}$, by Remark 59.75.3, and by Lemma 59.75.4 we see that $E$ is closed under specializations and generalizations in the topological space $U$.

After shrinking $U$ we may assume $U$ is affine too. By Descent, Lemma 35.16.3 we see that $U$ has a finite number of irreducible components. After removing the irreducible components which do not pass through $u$, we may assume every irreducible component of $U$ passes through $u$. Since $U$ is a sober topological space it follows that $E = U$ and we conclude that $\varphi $ is an isomorphism by Theorem 59.29.10. Thus (1) follows.

We omit the proof that (1) implies (2). $\square$

Lemma 59.75.6. Let $S$ be a scheme such that every quasi-compact open of $S$ has finite number of irreducible components (for example if $S$ has a Noetherian underlying topological space, or if $S$ is locally Noetherian). Let $\Lambda $ be a Noetherian ring. Let $\mathcal{F}$ be a sheaf of $\Lambda $-modules on $S_{\acute{e}tale}$. The following are equivalent

  1. $\mathcal{F}$ is a finite type, locally constant sheaf of $\Lambda $-modules, and

  2. all stalks of $\mathcal{F}$ are finite $\Lambda $-modules and all specialization maps $sp : \mathcal{F}_{\overline{s}} \to \mathcal{F}_{\overline{t}}$ are bijective.

Proof. The proof of this lemma is the same as the proof of Lemma 59.75.5. Assume (2). Let $\overline{s}$ be a geometric point of $S$ lying over $s \in S$. In order to prove (1) we have to find an étale neighbourhood $(U, \overline{u})$ of $(S, \overline{s})$ such that $\mathcal{F}|_ U$ is constant. We may and do assume $S$ is affine.

Since $M = \mathcal{F}_{\overline{s}}$ is a finite $\Lambda $-module and $\Lambda $ is Noetherian, we can choose a presentation

\[ \Lambda ^{\oplus m} \xrightarrow {A} \Lambda ^{\oplus n} \to M \to 0 \]

for some matrix $A = (a_{ji})$ with coefficients in $\Lambda $. We can choose $(U, \overline{u})$ and elements $\sigma _1, \ldots , \sigma _ n \in \mathcal{F}(U)$ such that $\sum a_{ji}\sigma _ i = 0$ in $\mathcal{F}(U)$ and such that the images of $\sigma _ i$ in $\mathcal{F}_{\overline{s}} = M$ are the images of the standard basis element of $\Lambda ^ n$ in the presentation of $M$ given above. Consider the map

\[ \varphi : \underline{M} \longrightarrow \mathcal{F}|_ U \]

on $U_{\acute{e}tale}$ defined by $\sigma _1, \ldots , \sigma _ n$. This map is a bijection on stalks at $\overline{u}$ by construction. Let us consider the subset

\[ E = \{ u' \in U \mid \varphi _{\overline{u}'}\text{ is bijective}\} \subset U \]

Here $\overline{u}'$ is any geometric point of $U$ lying over $u'$ (the condition is independent of the choice by Remark 59.29.8). The image $u \in U$ of $\overline{u}$ is in $E$. By our assumption on the specialization maps for $\mathcal{F}$, by Remark 59.75.3, and by Lemma 59.75.4 we see that $E$ is closed under specializations and generalizations in the topological space $U$.

After shrinking $U$ we may assume $U$ is affine too. By Descent, Lemma 35.16.3 we see that $U$ has a finite number of irreducible components. After removing the irreducible components which do not pass through $u$, we may assume every irreducible component of $U$ passes through $u$. Since $U$ is a sober topological space it follows that $E = U$ and we conclude that $\varphi $ is an isomorphism by Theorem 59.29.10. Thus (1) follows.

We omit the proof that (1) implies (2). $\square$

Lemma 59.75.7. Let $f : X \to S$ be a quasi-compact and quasi-separated morphism of schemes. Let $K \in D^+(X_{\acute{e}tale})$. Let $\overline{s}$ be a geometric point of $S$ and let $\overline{t}$ be a geometric point of $\mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{s}})$. We have a commutative diagram

\[ \xymatrix{ (Rf_*K)_{\overline{s}} \ar[r]_{sp} \ar@{=}[d] & (Rf_*K)_{\overline{t}} \ar@{=}[d] \\ R\Gamma (X \times _ S \mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{s}}), K) \ar[r] & R\Gamma (X \times _ S \mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{t}}), K) } \]

where the bottom horizontal arrow arises as pullback by the morphism $\text{id}_ X \times c$ where $c : \mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{t}}) \to \mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{S}})$ is the morphism introduced in Remark 59.75.2. The vertical arrows are given by Theorem 59.53.1.

Proof. This follows immediately from the description of $sp$ in Remark 59.75.2. $\square$

Remark 59.75.8. Let $f : X \to S$ be a morphism of schemes. Let $K \in D(X_{\acute{e}tale})$. Let $\overline{s}$ be a geometric point of $S$ and let $\overline{t}$ be a geometric point of $\mathop{\mathrm{Spec}}(\mathcal{O}^{sh}_{S, \overline{s}})$. Let $c$ be as in Remark 59.75.2. We can always make a commutative diagram

\[ \xymatrix{ (Rf_*K)_{\overline{s}} \ar[r] \ar[d]_{sp} & R\Gamma (X \times _ S \mathop{\mathrm{Spec}}(\mathcal{O}_{S, \overline{s}}^{sh}), K) \ar[r] \ar[d]_{(\text{id}_ X \times c)^{-1}} & R\Gamma (X_{\overline{s}}, K) \\ (Rf_*K)_{\overline{t}} \ar[r] & R\Gamma (X \times _ S \mathop{\mathrm{Spec}}(\mathcal{O}_{S, \overline{t}}^{sh}), K) \ar[r] & R\Gamma (X_{\overline{t}}, K) } \]

where the horizontal arrows are those of Remark 59.53.2. In general there won't be a vertical map on the right between the cohomologies of $K$ on the fibres fitting into this diagram, even in the case of Lemma 59.75.7.


Comments (2)

Comment #8639 by Emmanuel Kowalski on

The reference to EGA should be to Exposé VIII, instead of VII.


Post a comment

Your email address will not be published. Required fields are marked.

In your comment you can use Markdown and LaTeX style mathematics (enclose it like $\pi$). A preview option is available if you wish to see how it works out (just click on the eye in the toolbar).

Unfortunately JavaScript is disabled in your browser, so the comment preview function will not work.

All contributions are licensed under the GNU Free Documentation License.




In order to prevent bots from posting comments, we would like you to prove that you are human. You can do this by filling in the name of the current tag in the following input field. As a reminder, this is tag 0GJ2. Beware of the difference between the letter 'O' and the digit '0'.