The Stacks project

Lemma 62.3.11. Let $f : X \to Y$ be a morphism of schemes which is separated and locally of finite type. Let $\mathcal{F}$ be an abelian sheaf on $X_{\acute{e}tale}$. Then there is a canonical isomorphism

\[ (f_!\mathcal{F})_{\overline{y}} \longrightarrow H^0_ c(X_{\overline{y}}, \mathcal{F}|_{X_{\overline{y}}}) \]

for any geometric point $\overline{y} : \mathop{\mathrm{Spec}}(k) \to Y$.

Proof. Recall that $(f_*\mathcal{F})_{\overline{y}} = \mathop{\mathrm{colim}}\nolimits f_*\mathcal{F}(V)$ where the colimit is over the étale neighbourhoods $(V, \overline{v})$ of $\overline{y}$. If $s \in f_*\mathcal{F}(V) = \mathcal{F}(X_ V)$, then we can pullback $s$ to a section of $\mathcal{F}$ over $(X_ V)_{\overline{v}} = X_{\overline{y}}$. Thus we obtain a canonical map

\[ c_{\overline{y}} : (f_*\mathcal{F})_{\overline{y}} \longrightarrow H^0(X_{\overline{y}}, \mathcal{F}|_{X_{\overline{y}}}) \]

We claim that this map induces a bijection between the subgroups $(f_!\mathcal{F})_{\overline{y}}$ and $H^0_ c(X_{\overline{y}}, \mathcal{F}|_{X_{\overline{y}}})$. The claim implies the lemma, but is a little bit more precise in that it describes the identification of the lemma as given by pullbacks of sections of $\mathcal{F}$ to the geometric fibre of $f$.

Observe that any element $s \in (f_!\mathcal{F})_{\overline{y}} \subset (f_*\mathcal{F})_{\overline{y}}$ is mapped by $c_{\overline{y}}$ to an element of $H^0_ c(X_{\overline{y}}, \mathcal{F}|_{X_{\overline{y}}}) \subset H^0(X_{\overline{y}}, \mathcal{F}|_{X_{\overline{y}}})$. This is true because taking the support of a section commutes with pullback and because properness is preserved by base change. This at least produces the map in the statement of the lemma. To prove that it is an isomorphism we may work Zariski locally on $Y$ and hence we may and do assume $Y$ is affine.

An observation that we will use below is that given an open subscheme $X' \subset X$ and if $f' = f|_{X'}$, then we obtain a commutative diagram

\[ \xymatrix{ (f'_!(\mathcal{F}|_{X'}))_{\overline{y}} \ar[r] \ar[d] & H^0_ c(X'_{\overline{y}}, \mathcal{F}|_{X'_{\overline{y}}}) \ar[d] \\ (f_!\mathcal{F})_{\overline{y}} \ar[r] & H^0_ c(X_{\overline{y}}, \mathcal{F}|_{X_{\overline{y}}}) } \]

where the horizontal arrows are the maps constructed above and the vertical arrows are given in Remarks 62.3.5 and 62.3.9. The reason is that given an étale neighbourhood $(V, \overline{v})$ of $\overline{y}$ and a section $s \in f_*\mathcal{F}(V) = \mathcal{F}(X_ V)$ whose support $Z$ happens to be contained in $X'_ V$ and is proper over $V$, so that $s$ gives rise to an element of both $(f'_!(\mathcal{F}|_{X'}))_{\overline{y}}$ and $(f_!\mathcal{F})_{\overline{y}}$ which correspond via the vertical arrow of the diagram, then these elements are mapped via the horizontal arrows to the pullback $s|_{X_{\overline{y}}}$ of $s$ to $X_{\overline{y}}$ whose support $Z_{\overline{y}}$ is contained in $X'_{\overline{y}}$ and hence this restriction gives rise to a compatible pair of elements of $H^0_ c(X'_{\overline{y}}, \mathcal{F}|_{X'_{\overline{y}}})$ and $H^0_ c(X_{\overline{y}}, \mathcal{F}|_{X_{\overline{y}}})$.

Suppose $s \in (f_!\mathcal{F})_{\overline{y}}$ maps to zero in $H^0_ c(X_{\overline{y}}, \mathcal{F}|_{X_{\overline{y}}})$. Say $s$ corresponds to $s \in f_*\mathcal{F}(V) = \mathcal{F}(X_ V)$ with support $Z$ proper over $V$. We may assume that $V$ is affine and hence $Z$ is quasi-compact. Then we may choose a quasi-compact open $X' \subset X$ containing the image of $Z$. Then $Z$ is contained in $X'_ V$ and hence $s$ is the image of an element $s' \in f'_!(\mathcal{F}|_{X'})(V)$ where $f' = f|_{X'}$ as in the previous paragraph. Then $s'$ maps to zero in $H^0_ c(X'_{\overline{y}}, \mathcal{F}|_{X'_{\overline{y}}})$. Hence in order to prove injectivity, we may replace $X$ by $X'$, i.e., we may assume $X$ is quasi-compact. We will prove this case below.

Suppose that $t \in H^0_ c(X_{\overline{y}}, \mathcal{F}|_{X_{\overline{y}}})$. Then the support of $t$ is contained in a quasi-compact open subscheme $W \subset X_{\overline{y}}$. Hence we can find a quasi-compact open subscheme $X' \subset X$ such that $X'_{\overline{y}}$ contains $W$. Then it is clear that $t$ is contained in the image of the injective map $H^0_ c(X'_{\overline{y}}, \mathcal{F}|_{X'_{\overline{y}}}) \to H^0_ c(X_{\overline{y}}, \mathcal{F}|_{X_{\overline{y}}})$. Hence in order to show surjectivity, we may replace $X$ by $X'$, i.e., we may assume $X$ is quasi-compact. We will prove this case below.

In this last paragraph of the proof we prove the lemma in case $X$ is quasi-compact and $Y$ is affine. By More on Flatness, Theorem 38.33.8 there exists a compactification $j : X \to \overline{X}$ over $Y$. Set $\mathcal{G} = j_!\mathcal{F}$ so that $\mathcal{F} = \mathcal{G}|_ X$ by Étale Cohomology, Lemma 59.70.4. By the disussion above we get a commutative diagram

\[ \xymatrix{ (f_!\mathcal{F})_{\overline{y}} \ar[r] \ar[d] & H^0_ c(X_{\overline{y}}, \mathcal{F}|_{X_{\overline{y}}}) \ar[d] \\ (\overline{f}_!\mathcal{G})_{\overline{y}} \ar[r] & H^0_ c(\overline{X}_{\overline{y}}, \mathcal{G}|_{\overline{X}_{\overline{y}}}) } \]

By Lemmas 62.3.6 and 62.3.10 the vertical maps are isomorphisms. This reduces us to the case of the proper morphism $\overline{X} \to Y$. For a proper morphism our map is an isomorphism by Lemmas 62.3.4 and 62.3.8 and proper base change for pushforwards, see Étale Cohomology, Lemma 59.91.4. $\square$


Comments (0)


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 0F76. Beware of the difference between the letter 'O' and the digit '0'.