Let $U \to X$ be a cover in $X_{\mathrm{\acute{e}t}}$ and $\mathcal{F}$ an abelian presheaf. The Čech complex of the cover is
with differentials the alternating sums of the maps induced by the projections. Its cohomology is $\check{H}^{i}(U/X, \mathcal{F})$. Passing to the colimit over all covers (ordered by refinement),
If $\mathcal{F}$ is a sheaf, the sheaf condition says exactly that $\check{H}^0(U/X, \mathcal{F}) = \mathcal{F}(X)$ for every cover. In higher degrees the two theories differ in general:
$\check{H}^i(X_{\mathrm{\acute{e}t}}, \mathcal{F})$ is not in general isomorphic to the derived functor cohomology $H^i(X_{\mathrm{\acute{e}t}}, \mathcal{F})$.
Suppose $X$ is quasi-compact and every finite subset of $X$ is contained in an affine open subset (true, for example, if $X$ is quasi-projective over an affine scheme). Then Čech cohomology is canonically isomorphic to derived functor cohomology:
There is a version of Čech cohomology in which covers are replaced by hypercovers; it computes derived functor cohomology with no hypothesis on $X$. We will not need it.
Choose an injective resolution of the sheaf $\mathcal{F}$ in $\mathrm{Sh}(X_{\mathrm{\acute{e}t}}, \mathbf{Ab})$:
Given a cover $U \to X$, apply the Čech complex to each $\mathcal{I}^s$. This gives a double complex
with the Čech differential in the $r$ direction and the resolution differential in the $s$ direction. The two ways of computing the cohomology of its total complex give two spectral sequences.
For a cover $U \to X$ and a sheaf $\mathcal{F}$ on $X_{\mathrm{\acute{e}t}}$ there is a spectral sequence
Passing to the colimit over covers gives the same statement with $\check{H}^{r}(X_{\mathrm{\acute{e}t}}, \underline{H}^{s}(\mathcal{F}))$ on the $E_2$ page.
The bottom row $s = 0$ is Čech cohomology of $\mathcal{F}$ itself. If the higher cohomology of $\mathcal{F}$ vanishes on every term $U^{\times_X(r+1)}$ of the cover, the spectral sequence collapses onto that row, and Čech cohomology of the cover computes $H^{\ast}(X_{\mathrm{\acute{e}t}}, \mathcal{F})$. This is Leray's acyclic-cover principle, and it is how the computations below go.
Litt uses the spectral sequence to prove a claim from the previous lecture: if $\check{C}^{\bullet}(X_{\mathrm{\acute{e}t}}, -)$ is exact on $\mathrm{Sh}^{\mathrm{ab}}(X_{\mathrm{\acute{e}t}})$, then $\check{H}^{\ast} = H^{\ast}$. This is Exercise 1 in §6.
Let $U = U_0 \cup U_1$, where $U_0, U_1 \subseteq U$ are Zariski-open subsets, and let $\mathcal{F}$ be a sheaf on $U_{\mathrm{\acute{e}t}}$.
There is a functorial long exact sequence of étale cohomology groups
where the second map is the difference of the two restrictions.
Apply the Čech-to-derived spectral sequence to the cover $U_0 \sqcup U_1 \to U$. For any sheaf $\mathcal{G}$ its Čech complex is
Claim (exercise): this is quasi-isomorphic to the two-term complex
This uses that the cover is Zariski: $U_i \times_U U_j = U_i \cap U_j$ and $U_i \times_U U_i = U_i$, so $(U_0 \sqcup U_1)^{\times(r+1)}$ is a disjoint union of intersections $U_{i_0} \cap \cdots \cap U_{i_r}$, and the complex is the full Čech complex of a two-set open cover, which is quasi-isomorphic to its alternating part.
Hence the $E_2$ page vanishes outside the two columns $r = 0, 1$. A two-column spectral sequence degenerates at $E_2$ and gives short exact sequences
with $E_2^{0,s} = \ker\big(H^s(U_0) \oplus H^s(U_1) \to H^s(U_0 \cap U_1)\big)$ and $E_2^{1,s-1} = \operatorname{coker}\big(H^{s-1}(U_0) \oplus H^{s-1}(U_1) \to H^{s-1}(U_0 \cap U_1)\big)$. Splicing these short exact sequences gives the long exact sequence.
□
Recall that a quasi-coherent sheaf $\mathcal{F}$ on $X$ gives a sheaf $\mathcal{F}^{\mathrm{\acute{e}t}}$ on $X_{\mathrm{\acute{e}t}}$ (and $\mathcal{F}^{\mathrm{fppf}}$ on $X_{\mathrm{fppf}}$) by $(V \xrightarrow{g} X) \mapsto \Gamma(V, g^{\ast}\mathcal{F})$; that this is a sheaf is fppf descent (Meetings 4–5). For $\mathcal{F} = \mathcal{O}_X$ this is the sheaf $\mathbb{G}_a$.
Let $X$ be a scheme and $\mathcal{F} \in \mathbf{QCoh}(X)$. Then
where the left side is the usual (Zariski) cohomology of $\mathcal{F}$.
Litt adds that, for $X$ quasi-compact and quasi-separated,
The theorem says that the étale and fppf topologies add nothing new for these coefficients: the new information in étale cohomology comes from sheaves that are not quasi-coherent, such as $\mathbb{F}_p$, $\mu_n$ and $\mathbb{G}_m$.
The proof below, following Litt, is for $X$ quasi-compact and separated, where Čech cohomology can be used directly. We write it for the fppf topology; the étale case is the same.
If $X$ is quasi-compact, every fppf (or étale) cover of $X$ can be refined by a cover $U = \bigsqcup_{j=1}^{n} U_j \to X$ with each $U_j$ affine: fppf maps are open, so finitely many affine opens of the source already cover $X$. If $X$ is moreover affine, then $U$ is affine.
Let $X = \operatorname{Spec} A$ and $U = \operatorname{Spec} B$ with $A \to B$ faithfully flat, and let $\mathcal{F} = \widetilde{M}$ for an $A$-module $M$. Then $U \times_X U = \operatorname{Spec}(B \otimes_A B)$, and so on, and the augmented Čech complex is the Amitsur complex
Claim: it is exact. This is the same argument as in the proof of fppf descent: by faithful flatness it is enough to check exactness after $- \otimes_A B$, and after this base change the map $A \to B$ acquires a section (the multiplication $B \otimes_A B \to B$), which can be used to write down a null-homotopy. Hence
If $X$ is affine and $\mathcal{F} \in \mathbf{QCoh}(X)$, then $\check{H}^{i}(X_{\mathrm{fppf}}, \mathcal{F})$ is $\mathcal{F}(X)$ for $i = 0$ and $0$ for $i > 0$: by Step 1 affine covers are cofinal among all covers, and by Step 2 each of them has vanishing higher Čech cohomology.
First, $H^{s}(V, \mathcal{F}) = 0$ for every affine $V$ and every $s > 0$. Induct on $s$. A class $c \in H^{s}(V, \mathcal{F})$ dies on some cover, which by Step 1 we may take to be affine, $W \to V$. All the $W^{\times_V (r+1)}$ are affine, so by induction $E_2^{r,t} = 0$ for $0 < t < s$ in the Čech-to-derived spectral sequence of $W/V$, and $E_2^{r,0} = 0$ for $r > 0$ by Step 2. So $H^{s}(V, \mathcal{F})$ injects into $E_2^{0,s} \subseteq H^{s}(W, \mathcal{F})$, and since $c|_W = 0$ we get $c = 0$.
Now let $X$ be quasi-compact and separated, and take a finite Zariski cover by affines $U_1, \dots, U_n$. Their intersections are affine because $X$ is separated, so in the spectral sequence of this cover $E_2^{r,s} = 0$ for $s > 0$, and
the second equality being the classical computation of Zariski cohomology of a quasi-coherent sheaf by an affine cover. (This step is Litt's “take an affine cover $U \to X$ and use the Čech-to-derived spectral sequence”, left as an exercise.)
□
For $X = \mathbb{P}^n_k$ and $\mathcal{F} = \mathcal{O}_X$, the classical computation of Zariski cohomology gives
Let $X$ be a scheme over $\mathbb{F}_p$, for instance a quasi-projective variety. Litt's question is
Here $\mathbb{F}_p$ is the constant sheaf. The tool is the additive group $\mathbb{G}_a = \operatorname{Hom}(-, \mathbb{A}^1)$, with $\mathbb{G}_a(U) = \mathcal{O}_U(U)$, together with the Artin–Schreier map $x \mapsto x^p - x$, which is additive in characteristic $p$.
On $X_{\mathrm{\acute{e}t}}$ the sequence
is exact.
Left exactness holds at the level of the representing objects: if $f \in \mathcal{O}(U)$ satisfies $f^p - f = 0$, then, since $x^p - x = \prod_{a \in \mathbb{F}_p}(x - a)$, $f$ is locally constant with values in $\mathbb{F}_p$, i.e. a section of the constant sheaf.
Surjectivity: we need $x^p - x : \mathbb{G}_a \to \mathbb{G}_a$ to be an epimorphism of étale sheaves. Given $f \in \mathcal{O}_U(U) = \mathbb{G}_a(U)$, we must solve $x^p - x = f$ étale-locally on $U$. Form the fiber product
It is finite and free of rank $p$ over $U$, hence surjective, and it is étale because the derivative of $x^p - x - f$ in $x$ is $-1$, a unit. So $U' \to U$ is an étale cover, and on it $f$ acquires the root $x$.
□
By the theorem of §4, $H^{i}(X_{\mathrm{\acute{e}t}}, \mathbb{G}_a) = H^{i}(X, \mathcal{O}_X)$. So the long exact sequence of the Artin–Schreier sequence is written in coherent cohomology:
In particular, if $X$ is a quasi-projective variety, then $H^{i}(X_{\mathrm{\acute{e}t}}, \mathbb{F}_p) = 0$ for $i > \dim X + 1$, because coherent cohomology vanishes above the dimension.
Let $X = \mathbb{A}^1 = \operatorname{Spec} \mathbb{F}_p[t]$. Since $X$ is affine, $H^{i}(X, \mathcal{O}_X) = 0$ for $i > 0$, and the long exact sequence becomes
So $H^0 = \mathbb{F}_p$, $H^i = 0$ for $i \geq 2$, and
which is huge: infinite-dimensional over $\mathbb{F}_p$ (Exercise 5). A class represented by $h(t)$ corresponds to the Artin–Schreier cover $y^p - y = h(t)$ of the line.
Over $\mathbb{C}$ the affine line is contractible. In characteristic $p$, $\mathbb{F}_p$-coefficients see wild ramification at infinity, and $H^1(\mathbb{A}^1, \mathbb{F}_p)$ is infinite. This is why the Weil conjectures are approached with $\mathbb{Z}/\ell$ and $\mathbb{Q}_\ell$ coefficients for $\ell \neq p$, where the Kummer sequence (Litt's Lecture 10) takes the place of Artin–Schreier.
Exercises 1–4 are the ones Litt leaves in Lecture 8. Exercises 5–6 are seminar additions.
Suppose $\check{C}^{\bullet}(X_{\mathrm{\acute{e}t}}, -)$ is exact on $\mathrm{Sh}^{\mathrm{ab}}(X_{\mathrm{\acute{e}t}})$, i.e. takes short exact sequences of sheaves to short exact sequences of complexes. Prove that $\check{H}^{\ast}(X_{\mathrm{\acute{e}t}}, -) = H^{\ast}(X_{\mathrm{\acute{e}t}}, -)$ using the Čech-to-derived spectral sequence.
Let $U = U_0 \cup U_1$ be a Zariski open cover. Show that the Čech complex of $U_0 \sqcup U_1 \to U$ with coefficients in a sheaf $\mathcal{G}$ is quasi-isomorphic to $\mathcal{G}(U_0) \times \mathcal{G}(U_1) \to \mathcal{G}(U_0 \cap U_1)$. Where is it used that the $U_i$ are Zariski open rather than merely étale over $U$?
Let $A \to B$ be faithfully flat and $M$ an $A$-module. Show that $0 \to M \to M \otimes_A B \to M \otimes_A B^{\otimes 2} \to \cdots$ is exact. Hint (Litt): base change to $B$ and write down a null-homotopy using a section.
Let $X$ be quasi-compact and separated and $\mathcal{F}$ quasi-coherent. Take an affine cover $U \to X$ and use the Čech-to-derived spectral sequence to show that $H^{i}(X_{\mathrm{\acute{e}t}}, \mathcal{F}^{\mathrm{\acute{e}t}}) = H^{i}(X, \mathcal{F})$, filling in the details of Step 4 of §4.
Show that the classes of $t^n$, for $n \geq 1$ prime to $p$, are linearly independent in $\mathbb{F}_p[t] / \{g^p - g\}$. Conclude that $H^{1}(\mathbb{A}^1_{\mathbb{F}_p,\mathrm{\acute{e}t}}, \mathbb{F}_p)$ is infinite-dimensional. What changes over $\overline{\mathbb{F}}_p$?
Let $k$ be an algebraically closed field of characteristic $p$. Use the Artin–Schreier sequence and the example of §4 to compute $H^{i}(\mathbb{P}^n_{k,\mathrm{\acute{e}t}}, \mathbb{F}_p)$ for all $i$. Compare with $H^{i}(\mathbb{CP}^n, \mathbb{F}_p)$, and explain why $\mathbb{F}_p$-coefficients cannot give a Weil cohomology theory.