For a scheme $X$, fppf descent gives equivalences of categories
A quasi-coherent sheaf does not care which of these topologies we use.
Étale descent data for schemes need not be effective: descending a scheme along an étale cover can produce an object that is not a scheme. What one gets in general is an algebraic space.
Show that the category of abelian sheaves on $X_{\mathrm{\acute{e}t}}$ is abelian, with enough injectives. Then cohomology can be defined as a derived functor, for example $$H^{i}(X_{\mathrm{\acute{e}t}}, \mathbb{Z}/\ell\mathbb{Z}) \;:=\; R^{i}\Gamma(X_{\mathrm{\acute{e}t}}, \underline{\mathbb{Z}/\ell\mathbb{Z}}).$$
Let $\tau$ be a site. The forgetful functor $\mathrm{Sh}(\tau) \to \mathrm{PreSh}(\tau)$ has a left adjoint, sheafification. Litt proves this for $X_{\mathrm{\acute{e}t}}$ (§5).
Let $f : \tau_1 \to \tau_2$ be a continuous map of sites, given by a functor $f^{-1}$ from the underlying category of $\tau_2$ to that of $\tau_1$. For $\mathcal{G} \in \mathrm{Sh}(\tau_1)$ define
It is a sheaf on $\tau_2$ (Exercise 1).
A morphism of schemes $f : X \to Y$ gives a continuous map of sites $X_{\mathrm{\acute{e}t}} \to Y_{\mathrm{\acute{e}t}}$, with $f^{-1}(U/Y) = U \times_Y X$. Hence a functor $f_{\ast} : \mathrm{Sh}(X_{\mathrm{\acute{e}t}}) \to \mathrm{Sh}(Y_{\mathrm{\acute{e}t}})$, with $(f_\ast\mathcal{G})(U) = \mathcal{G}(U \times_Y X)$.
Let $k$ be an algebraically closed field and $\iota_{\bar{x}} : \operatorname{Spec} k \to X$ a geometric point. Every étale cover of $\operatorname{Spec} k$ is a disjoint union of copies of $\operatorname{Spec} k$, so
(Exercise 2). Given $\mathcal{F} \in \mathrm{Sh}((\operatorname{Spec} k)_{\mathrm{\acute{e}t}})$, i.e. a set, and $U \to X$ étale,
Here $U \times_X \bar{x}$ is a disjoint union of copies of $\operatorname{Spec} k$, one for each geometric point of $U$ lying over $\bar{x}$.
$(\iota_{\bar{x}})_{\ast}\mathcal{F}$ is called a skyscraper sheaf at $\bar{x}$.
If $X$ lies over an algebraically closed field $k$ and $\pi : X \to \operatorname{Spec} k$ is the structure map, then $\Gamma(\operatorname{Spec} k, \pi_{\ast}\mathcal{F}) = \Gamma(X, \mathcal{F})$. So under $\mathrm{Sh}((\operatorname{Spec} k)_{\mathrm{\acute{e}t}}) \simeq \mathbf{Sets}$, $\pi_\ast = \Gamma$.
In the Zariski topology a stalk is taken at a point. In the étale topology points are replaced by geometric points $\iota_{\bar{x}} : \operatorname{Spec} k \to X$ with $k = \bar{k}$.
For $\mathcal{F} \in \mathrm{Sh}(X_{\mathrm{\acute{e}t}})$, the stalk of $\mathcal{F}$ at $\bar{x}$ is the set $\mathcal{F}_{\bar{x}} = \iota_{\bar{x}}^{\ast}\mathcal{F}$, a sheaf on $(\operatorname{Spec} k)_{\mathrm{\acute{e}t}}$. Concretely,
where the colimit runs over étale neighbourhoods of $\bar{x}$: étale maps $U \to X$ together with a geometric point $\bar{u} : \bar{x} \to U$ lifting $\iota_{\bar{x}}$.
The adjunction between $\iota_{\bar{x}}^{\ast}$ and $(\iota_{\bar{x}})_{\ast}$ reads $\operatorname{Hom}(\mathcal{G}, (\iota_{\bar{x}})_{\ast}\Lambda) = \operatorname{Hom}(\mathcal{G}_{\bar{x}}, \Lambda)$. Every map from a sheaf to a skyscraper sheaf factors through the stalk; this is used in §4.
Let $\varphi : \mathcal{F} \to \mathcal{G}$ be a map of sheaves of abelian groups on $X_{\mathrm{\acute{e}t}}$. The following are equivalent:
(ii) ⇒ (i). Let $a, b : \mathcal{G} \rightrightarrows \mathcal{H}$ with $a \circ \varphi = b \circ \varphi$; we want $a = b$. Given $s \in \mathcal{G}(U)$, choose a cover $U' \to U$ with $s|_{U'} = \varphi(s')$. Then $a(s)|_{U'} = a\varphi(s') = b\varphi(s') = b(s)|_{U'}$, and since $\mathcal{H}$ is a sheaf, $a(s) = b(s)$.
(i) ⇒ (iii). Suppose $\mathcal{F}_{\bar{x}} \to \mathcal{G}_{\bar{x}}$ is not surjective for some $\bar{x}$, with cokernel $\Lambda \neq 0$. Consider the two maps
namely $0$ and “the natural map” $\mathcal{G} \to (\iota_{\bar{x}})_\ast\mathcal{G}_{\bar{x}} \to (\iota_{\bar{x}})_\ast\Lambda$, which corresponds under adjunction to $\mathcal{G}_{\bar{x}} \to \Lambda$. Both composites with $\varphi$ are $0$, but the two maps differ. So $\varphi$ is not an epimorphism, a contradiction.
(iii) ⇒ (ii). Given $s \in \mathcal{G}(U)$, we want a cover $U' \to U$ on which $s$ comes from $\mathcal{F}$. Choose a geometric point $\bar{x}$ of $U$. Since $\mathcal{F}_{\bar{x}} \to \mathcal{G}_{\bar{x}}$ is surjective, there is an étale neighbourhood $(V, \bar{v})$ of $\bar{x}$ such that $s|_V$ is in the image of $\mathcal{F}(V)$. Now choose a geometric point $\bar{x}'$ not in the image of $V$ and keep going. The disjoint union of all the $V$'s obtained is an étale cover $U' \to U$ with the required property.
□
Let $0 \to \mathcal{F} \to \mathcal{G} \xrightarrow{f} \mathcal{H}$ be a sequence of abelian sheaves on $X_{\mathrm{\acute{e}t}}$. The following are equivalent:
Proof: Exercise 3. It is the same proof as for topological spaces; for (ii) ⇒ (iii), use that filtered colimits preserve exactness.
We now prove that sheafification exists for $\mathrm{PreSh}(X_{\mathrm{\acute{e}t}})$. The construction imitates the espace étalé of a presheaf on a topological space: a sheaf is determined by its stalks, so we put the stalks together first and then cut down.
For each point $x \in X$ choose a geometric point $\bar{x}$ lying over $x$. Given $\mathcal{F} \in \mathrm{PreSh}(X_{\mathrm{\acute{e}t}})$, define
This is a sheaf, being a product of skyscraper sheaves, and there is a natural map of presheaves $\mathcal{F} \to \mathrm{Esp}(\mathcal{F})$. (The stalk $\mathcal{F}_{\bar{x}}$ of a presheaf is defined by the same colimit as in §3.)
$\mathcal{F}^{a}$ is the subsheaf of $\mathrm{Esp}(\mathcal{F})$ generated by $\mathcal{F}$:
The Meeting 5 notes wrote $\mathcal{F}^{+}$ (or $\mathcal{F}^{++}$) for the sheafification built from covers. Litt's $\mathcal{F}^{a}$ is built from stalks instead. By the universal property the two constructions agree.
Colimits exist in $\mathrm{Sh}(X_{\mathrm{\acute{e}t}})$, and
Colimits exist for presheaves (computed objectwise), and left adjoints send colimits to colimits.
□
$\mathrm{Sh}^{\mathrm{ab}}(X_{\mathrm{\acute{e}t}})$ is an abelian category.
□
Enough injectives, derived functor cohomology, Čech cohomology — and then start computing!
Exercises 1–5 are the ones Litt leaves in Lecture 6. Exercise 6 is a seminar addition.
Let $f : \tau_1 \to \tau_2$ be a continuous map of sites and $\mathcal{G} \in \mathrm{Sh}(\tau_1)$. Show that $U \mapsto \mathcal{G}(f^{-1}(U))$ is a sheaf on $\tau_2$. Which properties of a continuous map of sites do you use?
Let $k$ be an algebraically closed field. Show that $\mathcal{F} \mapsto \mathcal{F}(\operatorname{Spec} k)$ is an equivalence $\mathrm{Sh}((\operatorname{Spec} k)_{\mathrm{\acute{e}t}}) \simeq \mathbf{Sets}$.
Prove the left-exactness lemma of §4, following the proof for sheaves on a topological space.
Show that $\mathcal{F}^{a}$ is a sheaf, and that for every presheaf $\mathcal{F}$ and sheaf $\mathcal{G}$, composing with $\mathcal{F} \to \mathcal{F}^{a}$ gives a bijection $\operatorname{Hom}_{\mathrm{Sh}}(\mathcal{F}^{a}, \mathcal{G}) \cong \operatorname{Hom}_{\mathrm{PreSh}}(\mathcal{F}, \mathcal{G})$.
For a map $f : \mathcal{F} \to \mathcal{G}$ in $\mathrm{Sh}^{\mathrm{ab}}(X_{\mathrm{\acute{e}t}})$, show that the natural map $\operatorname{coim} f \to \operatorname{im} f$ is an isomorphism by checking it on stalks.
Let $K$ be any field, with separable closure $K^{s}$ and $G = \operatorname{Gal}(K^{s}/K)$. Show that $\mathcal{F} \mapsto \varinjlim_{L} \mathcal{F}(\operatorname{Spec} L)$, over finite separable extensions $K \subseteq L \subseteq K^{s}$, is an equivalence between $\mathrm{Sh}((\operatorname{Spec} K)_{\mathrm{\acute{e}t}})$ and sets with a continuous $G$-action. What does Exercise 2 become in this language?