## 59.63 The Artin-Schreier sequence

Let $p$ be a prime number. Let $S$ be a scheme in characteristic $p$. The Artin-Schreier sequence is the short exact sequence

$0 \longrightarrow \underline{\mathbf{Z}/p\mathbf{Z}}_ S \longrightarrow \mathbf{G}_{a, S} \xrightarrow {F-1} \mathbf{G}_{a, S} \longrightarrow 0$

where $F - 1$ is the map $x \mapsto x^ p - x$.

Lemma 59.63.1. Let $p$ be a prime. Let $S$ be a scheme of characteristic $p$.

1. If $S$ is affine, then $H_{\acute{e}tale}^ q(S, \underline{\mathbf{Z}/p\mathbf{Z}}) = 0$ for all $q \geq 2$.

2. If $S$ is a quasi-compact and quasi-separated scheme of dimension $d$, then $H_{\acute{e}tale}^ q(S, \underline{\mathbf{Z}/p\mathbf{Z}}) = 0$ for all $q \geq 2 + d$.

Proof. Recall that the étale cohomology of the structure sheaf is equal to its cohomology on the underlying topological space (Theorem 59.22.4). The first statement follows from the Artin-Schreier exact sequence and the vanishing of cohomology of the structure sheaf on an affine scheme (Cohomology of Schemes, Lemma 30.2.2). The second statement follows by the same argument from the vanishing of Cohomology, Proposition 20.22.4 and the fact that $S$ is a spectral space (Properties, Lemma 28.2.4). $\square$

Lemma 59.63.2. Let $k$ be an algebraically closed field of characteristic $p > 0$. Let $V$ be a finite dimensional $k$-vector space. Let $F : V \to V$ be a frobenius linear map, i.e., an additive map such that $F(\lambda v) = \lambda ^ p F(v)$ for all $\lambda \in k$ and $v \in V$. Then $F - 1 : V \to V$ is surjective with kernel a finite dimensional $\mathbf{F}_ p$-vector space of dimension $\leq \dim _ k(V)$.

Proof. If $F = 0$, then the statement holds. If we have a filtration of $V$ by $F$-stable subvector spaces such that the statement holds for each graded piece, then it holds for $(V, F)$. Combining these two remarks we may assume the kernel of $F$ is zero.

Choose a basis $v_1, \ldots , v_ n$ of $V$ and write $F(v_ i) = \sum a_{ij} v_ j$. Observe that $v = \sum \lambda _ i v_ i$ is in the kernel if and only if $\sum \lambda _ i^ p a_{ij} v_ j = 0$. Since $k$ is algebraically closed this implies the matrix $(a_{ij})$ is invertible. Let $(b_{ij})$ be its inverse. Then to see that $F - 1$ is surjective we pick $w = \sum \mu _ i v_ i \in V$ and we try to solve

$(F - 1)(\sum \lambda _ iv_ i) = \sum \lambda _ i^ p a_{ij} v_ j - \sum \lambda _ j v_ j = \sum \mu _ j v_ j$

This is equivalent to

$\sum \lambda _ j^ p v_ j - \sum b_{ij} \lambda _ i v_ j = \sum b_{ij} \mu _ i v_ j$

in other words

$\lambda _ j^ p - \sum b_{ij} \lambda _ i = \sum b_{ij} \mu _ i, \quad j = 1, \ldots , \dim (V).$

The algebra

$A = k[x_1, \ldots , x_ n]/ (x_ j^ p - \sum b_{ij} x_ i - \sum b_{ij} \mu _ i)$

is standard smooth over $k$ (Algebra, Definition 10.137.6) because the matrix $(b_{ij})$ is invertible and the partial derivatives of $x_ j^ p$ are zero. A basis of $A$ over $k$ is the set of monomials $x_1^{e_1} \ldots x_ n^{e_ n}$ with $e_ i < p$, hence $\dim _ k(A) = p^ n$. Since $k$ is algebraically closed we see that $\mathop{\mathrm{Spec}}(A)$ has exactly $p^ n$ points. It follows that $F - 1$ is surjective and every fibre has $p^ n$ points, i.e., the kernel of $F - 1$ is a group with $p^ n$ elements. $\square$

Lemma 59.63.3. Let $X$ be a separated scheme of finite type over a field $k$. Let $\mathcal{F}$ be a coherent sheaf of $\mathcal{O}_ X$-modules. Then $\dim _ k H^ d(X, \mathcal{F}) < \infty$ where $d = \dim (X)$.

Proof. We will prove this by induction on $d$. The case $d = 0$ holds because in that case $X$ is the spectrum of a finite dimensional $k$-algebra $A$ (Varieties, Lemma 33.20.2) and every coherent sheaf $\mathcal{F}$ corresponds to a finite $A$-module $M = H^0(X, \mathcal{F})$ which has $\dim _ k M < \infty$.

Assume $d > 0$ and the result has been shown for separated schemes of finite type of dimension $< d$. The scheme $X$ is Noetherian. Consider the property $\mathcal{P}$ of coherent sheaves on $X$ defined by the rule

$\mathcal{P}(\mathcal{F}) \Leftrightarrow \dim _ k H^ d(X, \mathcal{F}) < \infty$

We are going to use the result of Cohomology of Schemes, Lemma 30.12.4 to prove that $\mathcal{P}$ holds for every coherent sheaf on $X$.

Let

$0 \to \mathcal{F}_1 \to \mathcal{F} \to \mathcal{F}_2 \to 0$

be a short exact sequence of coherent sheaves on $X$. Consider the long exact sequence of cohomology

$H^ d(X, \mathcal{F}_1) \to H^ d(X, \mathcal{F}) \to H^ d(X, \mathcal{F}_2)$

Thus if $\mathcal{P}$ holds for $\mathcal{F}_1$ and $\mathcal{F}_2$, then it holds for $\mathcal{F}$.

Let $Z \subset X$ be an integral closed subscheme. Let $\mathcal{I}$ be a coherent sheaf of ideals on $Z$. To finish the proof we have to show that $H^ d(X, i_*\mathcal{I}) = H^ d(Z, \mathcal{I})$ is finite dimensional. If $\dim (Z) < d$, then the result holds because the cohomology group will be zero (Cohomology, Proposition 20.20.7). In this way we reduce to the situation discussed in the following paragraph.

Assume $X$ is a variety of dimension $d$ and $\mathcal{F} = \mathcal{I}$ is a coherent ideal sheaf. In this case we have a short exact sequence

$0 \to \mathcal{I} \to \mathcal{O}_ X \to i_*\mathcal{O}_ Z \to 0$

where $i : Z \to X$ is the closed subscheme defined by $\mathcal{I}$. By induction hypothesis we see that $H^{d - 1}(Z, \mathcal{O}_ Z) = H^{d - 1}(X, i_*\mathcal{O}_ Z)$ is finite dimensional. Thus we see that it suffices to prove the result for the structure sheaf.

We can apply Chow's lemma (Cohomology of Schemes, Lemma 30.18.1) to the morphism $X \to \mathop{\mathrm{Spec}}(k)$. Thus we get a diagram

$\xymatrix{ X \ar[rd]_ g & X' \ar[d]^-{g'} \ar[l]^\pi \ar[r]_ i & \mathbf{P}^ n_ k \ar[dl] \\ & \mathop{\mathrm{Spec}}(k) & }$

as in the statement of Chow's lemma. Also, let $U \subset X$ be the dense open subscheme such that $\pi ^{-1}(U) \to U$ is an isomorphism. We may assume $X'$ is a variety as well, see Cohomology of Schemes, Remark 30.18.2. The morphism $i' = (i, \pi ) : X' \to \mathbf{P}^ n_ X$ is a closed immersion (loc. cit.). Hence

$\mathcal{L} = i^*\mathcal{O}_{\mathbf{P}^ n_ k}(1) \cong (i')^*\mathcal{O}_{\mathbf{P}^ n_ X}(1)$

is $\pi$-relatively ample (for example by Morphisms, Lemma 29.39.7). Hence by Cohomology of Schemes, Lemma 30.16.2 there exists an $n \geq 0$ such that $R^ p\pi _*\mathcal{L}^{\otimes n} = 0$ for all $p > 0$. Set $\mathcal{G} = \pi _*\mathcal{L}^{\otimes n}$. Choose any nonzero global section $s$ of $\mathcal{L}^{\otimes n}$. Since $\mathcal{G} = \pi _*\mathcal{L}^{\otimes n}$, the section $s$ corresponds to section of $\mathcal{G}$, i.e., a map $\mathcal{O}_ X \to \mathcal{G}$. Since $s|_ U \not= 0$ as $X'$ is a variety and $\mathcal{L}$ invertible, we see that $\mathcal{O}_ X|_ U \to \mathcal{G}|_ U$ is nonzero. As $\mathcal{G}|_ U = \mathcal{L}^{\otimes n}|_{\pi ^{-1}(U)}$ is invertible we conclude that we have a short exact sequence

$0 \to \mathcal{O}_ X \to \mathcal{G} \to \mathcal{Q} \to 0$

where $\mathcal{Q}$ is coherent and supported on a proper closed subscheme of $X$. Arguing as before using our induction hypothesis, we see that it suffices to prove $\dim H^ d(X, \mathcal{G}) < \infty$.

By the Leray spectral sequence (Cohomology, Lemma 20.13.6) we see that $H^ d(X, \mathcal{G}) = H^ d(X', \mathcal{L}^{\otimes n})$. Let $\overline{X}' \subset \mathbf{P}^ n_ k$ be the closure of $X'$. Then $\overline{X}'$ is a projective variety of dimension $d$ over $k$ and $X' \subset \overline{X}'$ is a dense open. The invertible sheaf $\mathcal{L}$ is the restriction of $\mathcal{O}_{\overline{X}'}(n)$ to $X$. By Cohomology, Proposition 20.22.4 the map

$H^ d(\overline{X}', \mathcal{O}_{\overline{X}'}(n)) \longrightarrow H^ d(X', \mathcal{L}^{\otimes n})$

is surjective. Since the cohomology group on the left has finite dimension by Cohomology of Schemes, Lemma 30.14.1 the proof is complete. $\square$

Lemma 59.63.4. Let $X$ be separated of finite type over an algebraically closed field $k$ of characteristic $p > 0$. Then $H_{\acute{e}tale}^ q(X, \underline{\mathbf{Z}/p\mathbf{Z}}) = 0$ for $q \geq dim(X) + 1$.

Proof. Let $d = \dim (X)$. By the vanishing established in Lemma 59.63.1 it suffices to show that $H_{\acute{e}tale}^{d + 1}(X, \underline{\mathbf{Z}/p\mathbf{Z}}) = 0$. By Lemma 59.63.3 we see that $H^ d(X, \mathcal{O}_ X)$ is a finite dimensional $k$-vector space. Hence the long exact cohomology sequence associated to the Artin-Schreier sequence ends with

$H^ d(X, \mathcal{O}_ X) \xrightarrow {F - 1} H^ d(X, \mathcal{O}_ X) \to H^{d + 1}_{\acute{e}tale}(X, \mathbf{Z}/p\mathbf{Z}) \to 0$

By Lemma 59.63.2 the map $F - 1$ in this sequence is surjective. This proves the lemma. $\square$

Lemma 59.63.5. Let $X$ be a proper scheme over an algebraically closed field $k$ of characteristic $p > 0$. Then

1. $H_{\acute{e}tale}^ q(X, \underline{\mathbf{Z}/p\mathbf{Z}})$ is a finite $\mathbf{Z}/p\mathbf{Z}$-module for all $q$, and

2. $H^ q_{\acute{e}tale}(X, \underline{\mathbf{Z}/p\mathbf{Z}}) \to H^ q_{\acute{e}tale}(X_{k'}, \underline{\mathbf{Z}/p\mathbf{Z}}))$ is an isomorphism if $k'/k$ is an extension of algebraically closed fields.

Proof. By Cohomology of Schemes, Lemma 30.19.2) and the comparison of cohomology of Theorem 59.22.4 the cohomology groups $H^ q_{\acute{e}tale}(X, \mathbf{G}_ a) = H^ q(X, \mathcal{O}_ X)$ are finite dimensional $k$-vector spaces. Hence by Lemma 59.63.2 the long exact cohomology sequence associated to the Artin-Schreier sequence, splits into short exact sequences

$0 \to H_{\acute{e}tale}^ q(X, \underline{\mathbf{Z}/p\mathbf{Z}}) \to H^ q(X, \mathcal{O}_ X) \xrightarrow {F - 1} H^ q(X, \mathcal{O}_ X) \to 0$

and moreover the $\mathbf{F}_ p$-dimension of the cohomology groups $H_{\acute{e}tale}^ q(X, \underline{\mathbf{Z}/p\mathbf{Z}})$ is equal to the $k$-dimension of the vector space $H^ q(X, \mathcal{O}_ X)$. This proves the first statement. The second statement follows as $H^ q(X, \mathcal{O}_ X) \otimes _ k k' \to H^ q(X_{k'}, \mathcal{O}_{X_{k'}})$ is an isomorphism by flat base change (Cohomology of Schemes, Lemma 30.5.2). $\square$

Comment #2332 by Zili Zhang on

Do we need some integrability assumption on $S$ for the Artin-Schreier sequence? There might be some issue for the exactness in the middle. It occurs to me that to get $t\in \mathbb{F}_p$ from $t^p-t=\prod_{j\in\mathbb{F}_p}{(t-j)}=0$, some integrability (or irreducibility) should be used.

Comment #2403 by on

No condition is needed for $S$ except that it be a scheme over $\Spec(\mathbf{F}_p)$. It is true that the proof of exactness of the Artin-Schreier sequence in the \'etale topology is missing. (It is not exact in the Zariski topology; this may be related to your question...)

Comment #4605 by Rex on

Some typos:

"Thus if P holds for F1 and F2, then it hods for F"

"To finish the proof have to show that Hd(X,i∗I)=Hd(Z,I) ..."

"As G|U=KL⊗n|π−1(U) is invertible ..." (extra mathcal K here)

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).