04.03.08 · algebraic-geometry / cohomology

Étale cohomology and $ℓ$ -adic cohomology of varieties

shipped3 tiersLean: none

Anchor (Master): Grothendieck et al. SGA 4 (Tôhoku-style site/topos foundations) and SGA 4½/SGA 5; Deligne 1974 *La conjecture de Weil I* (Publ. IHÉS 43); Milne *Étale Cohomology*

Intuition Beginner

Cohomology is a machine that turns a space into a sequence of vector spaces measuring its holes. Over the complex numbers a variety is also a topological space, so the usual cohomology of topology applies and counts holes correctly. Over a finite field, or over the rational numbers, there is no honest topology to feed the machine. The natural candidate, the Zariski topology, is far too coarse: its open sets are so large that a locally constant gadget with finite values has no room to wiggle, and every higher cohomology group it produces is zero. A theory that always returns zero sees nothing.

The fix is to enlarge what counts as an "open cover." Instead of covering a variety by Zariski opens, allow covering it by maps that are étale: maps that look, formally, like covering maps even though they need not be open inclusions. A small étale map onto a piece of the variety behaves like a many-sheeted cover, and this gives the missing room to wiggle. The resulting étale site is a finer notion of "open cover for varieties," and the cohomology built from it finally records arithmetic information.

What makes this remarkable is that the new cohomology carries a symmetry group. Over a finite field the absolute Galois group acts on the étale cohomology of a variety, and that action is a number-theoretic object you can compute with. Counting the solutions of equations over finite fields becomes a question about the eigenvalues of one special symmetry, the Frobenius. This is the bridge from geometry to the deepest counting problems in number theory.

Visual Beginner

Picture three pictures side by side. On the left, a circle covered by two fat Zariski-style arcs that overlap in big chunks: the cover is too crude, and any locally constant labelling with finitely many colours is forced to be one single colour everywhere, so no holes are detected. In the middle, the same circle wrapped by a genuine double cover, two sheets spiralling above the base like a helix projected down: now a labelling can swap colours as you go around, and the loop is finally visible. On the right, a finite field's worth of points with a single rotating arrow labelled Frobenius spinning the whole configuration, standing for the Galois symmetry that the finer cover unlocks.

The middle and right pictures are the heart of the matter: étale covers supply the spiral sheets that Zariski opens cannot, and the spinning arrow is the arithmetic symmetry you get for free.

Worked example Beginner

Take the simplest interesting base: a point with hidden arithmetic, namely $Spec k$ for a field $k$ that is not algebraically closed. As a topological space this is one single point, so ordinary topology declares it to have no structure at all. The étale viewpoint disagrees. The étale covers of this one point are the finite separable field extensions of $k$ , stacked above it like the sheets of a covering map. The bookkeeping of how these sheets permute is governed by the Galois group of the separable closure.

The upshot is a dictionary. Étale cohomology of $Spec k$ with values in a module equals the Galois cohomology of that module under the action of $Gal (\overset{ˉ}{k} / k)$ . A one-point space, invisible to topology, has a whole tower of cohomology groups recording the arithmetic of $k$ .

A second computation gives a geometric anchor. For any reasonable variety, the first étale cohomology group with values in the sheaf of invertible functions recovers the Picard group, the group of line bundles. So the new theory simultaneously sees arithmetic on a point and recovers a classical geometric invariant on a variety. Both facts fall out of the same finer notion of cover.

Check your understanding Beginner

Exercise (easy, multiple choice).

Why is the Zariski topology a poor input for a cohomology theory with finite coefficients?

A. It has too many open sets.
B. Its open sets are so large that a locally constant finite-valued sheaf has no higher cohomology.
C. It is not a topology at all.
D. It only works over the complex numbers.

Hint

Think about how a constant sheaf with finite values behaves when every open set is enormous and connected.

Answer

B. Feedback-correct: nonempty Zariski opens on an irreducible variety are huge and dense, so a locally constant finite-valued sheaf is forced to be a global constant and produces no higher cohomology — the theory returns zero and sees nothing. Feedback-wrong: A is backwards (the issue is too few, too large opens), C is false (it is a genuine topology), D confuses the failure with the comparison theorem that holds over the complex numbers.

Formal definition Intermediate+

Let $X$ be a scheme. A morphism $f : U \to X$ is étale 04.02.05 when it is flat and unramified, equivalently smooth of relative dimension zero; étale morphisms are the algebro-geometric analogue of local isomorphisms. The small étale site $X_{\overset{e}{ˊ} t}$ has as underlying category the étale morphisms $U \to X$ (the "open sets"), with morphisms the $X$ -morphisms between them, and a covering of $U$ is a family ${U_{i} \to U}$ of étale morphisms that is jointly surjective onto $U$ . This collection of coverings is a Grothendieck topology: it is stable under base change, refinement, and composition. A scheme is no longer a topological space here; the "opens" form a category and the "covers" are surjective étale families.

An étale sheaf $F$ on $X_{\overset{e}{ˊ} t}$ is a contravariant functor from this category to abelian groups satisfying the sheaf axiom against étale covers: for every covering ${U_{i} \to U}$ the sequence $$ 0 \to \mathcal{F}(U) \to \prod_i \mathcal{F}(U_i) \to \prod_{i,j} \mathcal{F}(U_i \times_U U_j) $$ is exact. The category of abelian étale sheaves is an abelian category with enough injectives. Two families of coefficients are central: the constant sheaf $Z / n$ (the sheafification of the constant presheaf), and the multiplicative group sheaf $G_{m}$ defined by $G_{m} (U) = Γ (U, O_{U})^{\times}$ , the units of the global functions.

The global-sections functor $Γ (X, -) : F \mapsto F (X)$ is left exact. Its right derived functors are the étale cohomology groups $$ H^i_{\text{ét}}(X, \mathcal{F}) := R^i \Gamma(X, \mathcal{F}), $$ computed by resolving $F$ by injective étale sheaves and taking cohomology of the global sections of the resolution. As in the Zariski case, Čech cohomology against étale covers computes the same groups under mild hypotheses, and an étale-Čech-to-derived spectral sequence relates them.

Counterexamples to common slips

The étale site is not a topology on the point-set of $X$ : "opens" are étale schemes over $X$ , not subsets, and a single Zariski open can carry many distinct étale covers.
$H_{\overset{e}{ˊ} t}^{1} (X, G_{m}) = Pic (X)$ , but $H_{\overset{e}{ˊ} t}^{1} (X, Z / n)$ classifies $Z / n$ -torsors and is generally a different, finite group; do not conflate the multiplicative and constant coefficients.
$ℓ$ -adic cohomology $H^{i} (X_{\overset{ˉ}{k}}, Q_{ℓ})$ is not the étale cohomology of the constant sheaf $Q_{ℓ}$ ; it is built as an inverse limit of $Z / ℓ^{n}$ -cohomology, then inverted, because the naive constant-sheaf version is malformed.
The coefficient prime $ℓ$ must be invertible on $X$ (different from the residue characteristic $p$ ); $p$ -adic coefficients on a characteristic- $p$ variety give a degenerate theory and require crystalline cohomology instead.

Key theorem with proof Intermediate+

Theorem (foundational computations). Let $X$ be a scheme.

(i) For $X = Spec k$ with $k$ a field and separable closure $\overset{ˉ}{k}$ , and $M$ a discrete $Gal (\overset{ˉ}{k} / k)$ -module viewed as an étale sheaf, there is a canonical isomorphism $$ H^i_{\text{ét}}(\mathrm{Spec},k, M) ;\cong; H^i\big(\mathrm{Gal}(\bar k / k), M\big), $$ the continuous (profinite) Galois cohomology.

(ii) For any scheme $X$ , $$ H^1_{\text{ét}}(X, \mathbb{G}m) ;\cong; \mathrm{Pic}(X), $$ *the group of line bundles, and $H^0{\text{ét}}(X, \mathbb{G}_m) = \Gamma(X, \mathcal{O}_X)^\times$.*

Proof sketch. (i) The small étale site of $Spec k$ is equivalent to the site of finite discrete sets with continuous $Gal (\overset{ˉ}{k} / k)$ -action: a connected étale cover of $Spec k$ is $Spec k^{'}$ for a finite separable $k^{'} / k$ , and the fibre functor sending such a cover to its $\overset{ˉ}{k}$ -points is an equivalence onto finite $Gal (\overset{ˉ}{k} / k)$ -sets. Under this equivalence an abelian étale sheaf is the same datum as a discrete Galois module, the global-sections functor becomes the Galois-invariants functor $M \mapsto M^{Gal}$ , and right-derived invariants are by definition profinite group cohomology. The derived functors agree because the equivalence of sites induces an equivalence of the abelian sheaf categories and is compatible with the two left-exact global-section functors.

(ii) This is étale descent for line bundles together with Hilbert's Theorem 90. A line bundle is a $G_{m}$ -torsor; isomorphism classes of $G_{m}$ -torsors on a site are classified by the first cohomology of the structure group sheaf, giving $H_{\overset{e}{ˊ} t}^{1} (X, G_{m}) ≅ {G_{m} -torsors} / ≅$ . Étale descent says a line bundle that becomes the identity on an étale cover, with matching gluing data, descends to a genuine line bundle on $X$ ; the obstruction lives in $H^{1}$ . The identification with the classical Picard group uses that an étale-locally invertible sheaf is Zariski-locally invertible, which is exactly the vanishing $H_{\overset{e}{ˊ} t}^{1} (Spec A, G_{m}) = H_{Zar}^{1} (Spec A, G_{m})$ for local rings — Hilbert 90 in its cohomological form. $□$

Bridge. This pair of computations builds toward the entire $ℓ$ -adic story and appears again in the arithmetic of Galois representations 21.05.01, which consumes exactly the Galois action produced here. The Galois-cohomology identification is the foundational reason étale cohomology sees arithmetic at all: a one-point scheme, blind to topology, acquires the full cohomology of its absolute Galois group, and this is exactly the mechanism by which the cohomology of a variety over a finite field becomes a Frobenius module. The Picard-group computation is dual to this arithmetic statement on the geometric side: it shows the same machine that recovers Galois cohomology on a point recovers a classical geometric invariant on a variety, so the étale site generalises both the Zariski cohomology of line bundles 04.03.04 and the cohomology of Galois groups in one stroke. Putting these together, the central insight is that étale cohomology unifies arithmetic and geometry by replacing open subsets with étale covers, and the bridge to number theory is the Galois action that this finer topology makes visible.

Exercises Intermediate+

Exercise 1 (easy, short-answer).

State the sheaf axiom for an étale sheaf $F$ against a single étale cover ${U^{'} \to U}$ by one surjective étale morphism, and explain why the fibre product $U^{'} \times_{U} U^{'}$ replaces the pairwise intersection of the Zariski case.

Hint

In the étale world "intersection of two opens" is not a subset; it is a fibre product over the base.

Answer

The sequence $0 \to F (U) \to F (U^{'}) ⇉ F (U^{'} \times_{U} U^{'})$ is exact: a section over $U$ is the same as a section over $U^{'}$ whose two pullbacks to $U^{'} \times_{U} U^{'}$ agree. Because "opens" are étale schemes rather than subsets, there is no honest set-theoretic intersection; the correct gluing locus is the fibre product $U^{'} \times_{U} U^{'}$ , which records all the ways two points of $U^{'}$ can lie over the same point of $U$ . Rubric: full credit for the exact diagram plus the fibre-product explanation.

Exercise 2 (medium, short-answer).

Compute $H_{\overset{e}{ˊ} t}^{i} (Spec F_{q}, Z / n)$ for all $i$ , where $g cd (n, q) = 1$ , using the Galois-cohomology identification and the fact that $Gal (\overset{ˉ}{F}_{q} / F_{q}) ≅ Z$ with topological generator the Frobenius.

Hint

Continuous cohomology of $Z$ acting by the identity on a finite module $M$ is concentrated in degrees $0$ and $1$ , computed from $M^{Frob}$ and $M_{Frob}$ .

Answer

By the theorem, $H_{\overset{e}{ˊ} t}^{i} (Spec F_{q}, Z / n) = H^{i} (Z, Z / n)$ with the identity action. For a procyclic group with generator $F$ acting on a finite module $M$ , cohomology is $H^{0} = M^{F}$ (the invariants), $H^{1} = M_{F}$ (the coinvariants, here $coker (F - 1)$ ), and $H^{i} = 0$ for $i \geq 2$ . When $F$ acts as the identity $F - 1 = 0$ , so $H^{0} = Z / n$ , $H^{1} = Z / n$ , and $H^{i} = 0$ for $i \geq 2$ . Rubric: full credit for the degree- $0$ and degree- $1$ answers $Z / n$ and the vanishing above; partial credit for invoking the procyclic computation without the identity-action simplification.

Exercise 3 (medium, short-answer).

Explain why $H_{\overset{e}{ˊ} t}^{1} (X, G_{m}) = Pic (X)$ should be regarded as an instance of Hilbert's Theorem 90, by interpreting Theorem 90 cohomologically for a finite Galois extension $L / K$ .

Hint

Theorem 90 says $H^{1} (Gal (L / K), L^{\times}) = 0$ . Relate this vanishing to the descent of line bundles.

Answer

Hilbert 90 in cohomological form is the vanishing $H^{1} (Gal (L / K), L^{\times}) = 0$ , equivalently $H_{\overset{e}{ˊ} t}^{1} (Spec K, G_{m}) = 0$ . This says every $G_{m}$ -torsor over a field is the identity torsor, so étale-locally a line bundle is the identity and the only obstruction to gluing lives in $H^{1}$ . Globalising, $H_{\overset{e}{ˊ} t}^{1} (X, G_{m})$ classifies $G_{m}$ -torsors, which are line bundles, and the vanishing over local rings forces étale-local invertible sheaves to be Zariski-local, identifying the group with $Pic (X)$ . Rubric: full credit for stating the $H^{1} (Gal, L^{\times}) = 0$ vanishing and connecting it to torsor descent and the Picard identification.

Exercise 4 (hard, short-answer).

For an algebraically closed field $\overset{ˉ}{k}$ and $g cd (n, char \overset{ˉ}{k}) = 1$ , the Kummer sequence $1 \to μ_{n} \to G_{m} x \mapsto x^{n} G_{m} \to 1$ is exact on the étale site. Use its long exact cohomology sequence to compute $H_{\overset{e}{ˊ} t}^{1} (X, μ_{n})$ in terms of $Pic (X)$ and global units, and identify the appearance of the $n$ -torsion of the Picard group.

Hint

Feed $H_{\overset{e}{ˊ} t}^{0} (G_{m}) = Γ^{\times}$ and $H_{\overset{e}{ˊ} t}^{1} (G_{m}) = Pic$ into the long exact sequence; the boundary maps come from the $n$ -th-power map.

Answer

The long exact sequence reads $$ 0 \to \mu_n(X) \to \Gamma(X,\mathcal{O})^\times \xrightarrow{n} \Gamma(X,\mathcal{O})^\times \to H^1_{\text{ét}}(X, \mu_n) \to \mathrm{Pic}(X) \xrightarrow{n} \mathrm{Pic}(X) \to \cdots, $$ so there is a short exact sequence $$ 0 \to \Gamma(X,\mathcal{O})^\times / \big(\Gamma(X,\mathcal{O})^\times\big)^n \to H^1_{\text{ét}}(X, \mu_n) \to \mathrm{Pic}(X)[n] \to 0, $$ where $Pic (X) [n]$ is the $n$ -torsion subgroup. The Kummer sequence thus presents $H_{\overset{e}{ˊ} t}^{1} (X, μ_{n})$ as an extension of the $n$ -torsion line bundles by the units modulo $n$ -th powers. For a proper connected variety over $\overset{ˉ}{k}$ the unit term reduces to the constants, and the $n$ -torsion of $Pic$ is the geometric content. Rubric: full credit for the short exact sequence with both outer terms identified; partial credit for writing the long exact sequence without extracting the $n$ -torsion.

Exercise 5 (hard, short-answer).

State the Grothendieck-Lefschetz trace formula for a smooth projective variety $X$ over $F_{q}$ and explain, in one paragraph, how it recovers the rationality of the local zeta function $Z (X, t) = exp (\sum_{m \geq 1} # X (F_{q^{m}}) t^{m} / m)$ .

Hint

Write $# X (F_{q^{m}})$ as an alternating sum of traces of $Frob^{m}$ on the $ℓ$ -adic cohomology groups, then sum the exponential.

Answer

The trace formula is $# X (F_{q^{m}}) = \sum_{i = 0}^{2 d i m X} (- 1)^{i} tr (Frob^{m} ∣ H_{\overset{e}{ˊ} t}^{i} (X_{\overset{ˉ}{F}_{q}}, Q_{ℓ}))$ . Substituting into the zeta exponential and using $exp (\sum_{m} tr (Φ^{m}) t^{m} / m) = det (1 - t Φ)^{- 1}$ for the action $Φ = Frob$ on each finite-dimensional $H^{i}$ gives $Z (X, t) = \prod_{i = 0}^{2 d i m X} det (1 - t Frob ∣ H^{i})^{(- 1)^{i + 1}}$ , an alternating product of polynomials with $Q$ coefficients, hence a rational function. The cohomology groups being finite-dimensional is what makes each determinant a polynomial, and that finiteness is the structural payoff of the étale theory. Rubric: full credit for the trace formula and the determinant identity yielding the rational product expression.

Lean formalization Intermediate+

lean_status: none. Mathlib provides abstract Grothendieck topologies and sheaf cohomology via CategoryTheory.Sites, and profinite-group cohomology is partially available, but the étale site of a scheme is not constructed: there is no packaged category of étale morphisms over a base with the surjective-family covering topology, hence no étale topos and no étale cohomology functor. A schematic target reads:

import Mathlib.CategoryTheory.Sites.Sheaf
import Mathlib.AlgebraicGeometry.Morphisms.Etale
import Mathlib.CategoryTheory.Sites.Cohomology

namespace Codex.AlgebraicGeometry.Cohomology.EtaleCohomology

variable {X : Scheme}

/-- The small étale site of a scheme: étale morphisms over X with the
    jointly-surjective étale covering topology. -/
def etaleSite (X : Scheme) : GrothendieckTopology (Over X) :=
  sorry

/-- Étale cohomology as the derived functor of global sections on the
    étale site. -/
def etaleCohomology (X : Scheme) (F : Sheaf (etaleSite X) Ab) (i : ℕ) : Ab :=
  sorry

/-- Foundational computation: H¹_ét(X, 𝔾ₘ) ≅ Pic(X). -/
theorem H1_Gm_eq_Pic (X : Scheme) :
    etaleCohomology X gmSheaf 1 ≅ Picard X :=
  sorry

end Codex.AlgebraicGeometry.Cohomology.EtaleCohomology

The gap is foundational on three fronts. First, the covering topology generated by surjective étale families must be defined and shown to satisfy the Grothendieck-topology axioms, drawing on the étale-morphism API. Second, the structure sheaf $G_{m}$ and constant sheaves $Z / ℓ^{n}$ must be built as sheaves for that topology, with their sheaf condition verified against fibre products. Third, the comparison of étale-site cohomology with profinite Galois cohomology on $Spec k$ , and the inverse-limit construction of $ℓ$ -adic cohomology, require joining Mathlib's continuous-cohomology machinery to the new site. Each is a substantial programme; the present unit is prose-survey-tier with lean_status: none.

Advanced results Master

$ℓ$ -adic cohomology. Constant finite coefficients $Z / ℓ^{n}$ on the étale site, with $ℓ$ invertible on $X$ , give well-behaved finite cohomology groups, but a single such group is too small to be a Weil cohomology. The remedy is to assemble all torsion levels and pass to characteristic zero. For $X$ a variety over a field $k$ with separable closure $\overset{ˉ}{k}$ , set $X_{\overset{ˉ}{k}} = X \times_{k} \overset{ˉ}{k}$ and define $$ H^i(X_{\bar k}, \mathbb{Z}\ell) := \varprojlim_n H^i{\text{ét}}(X_{\bar k}, \mathbb{Z}/\ell^n), \qquad H^i(X_{\bar k}, \mathbb{Q}\ell) := H^i(X{\bar k}, \mathbb{Z}\ell) \otimes{\mathbb{Z}\ell} \mathbb{Q}\ell. $$ For $X$ smooth and proper these are finite-dimensional $Q_{ℓ}$ -vector spaces, vanishing outside the range $0 \leq i \leq 2 dim X$ . Crucially, passing to $X_{\overset{ˉ}{k}}$ does not erase the arithmetic: the Galois group $Gal (\overset{ˉ}{k} / k)$ acts on $X_{\overset{ˉ}{k}}$ over $X$ , hence acts continuously on every $H^{i} (X_{\overset{ˉ}{k}}, Q_{ℓ})$ . This action is the source of the $ℓ$ -adic Galois representations studied in 21.05.01.

Comparison with singular cohomology. When $k ↪ C$ and $X_{C}$ is the associated complex analytic space, the comparison theorem gives canonical isomorphisms $$ H^i_{\text{ét}}(X_{\bar k}, \mathbb{Z}/n) \cong H^i_{\text{sing}}(X_{\mathbb{C}}^{\mathrm{an}}, \mathbb{Z}/n), \qquad H^i(X_{\bar k}, \mathbb{Q}\ell) \cong H^i{\text{sing}}(X_{\mathbb{C}}^{\mathrm{an}}, \mathbb{Q}) \otimes_{\mathbb{Q}} \mathbb{Q}_\ell. $$ So étale cohomology numerically reproduces the topological Betti numbers it was designed to imitate, confirming that the étale theory is the right arithmetic refinement. The analytic side links to GAGA 04.09.03, the bridge that lets coherent and constructible comparisons pass between the algebraic and complex-analytic categories.

Poincaré duality. For $X$ smooth proper of pure dimension $d$ over a separably closed field, there is a perfect pairing $$ H^i(X, \mathbb{Q}\ell) \times H^{2d-i}(X, \mathbb{Q}\ell(d)) \to H^{2d}(X, \mathbb{Q}\ell(d)) \cong \mathbb{Q}\ell, $$ where $Q_{ℓ} (d)$ is the $d$ -fold Tate twist recording the Frobenius weight. This is the étale analogue of classical Poincaré duality and is compatible with the Galois action, so the duality intertwines Frobenius eigenvalues $α \leftrightarrow q^{d} / α$ on a variety over $F_{q}$ .

Lefschetz trace formula and the Weil conjectures. For $X$ smooth proper over $F_{q}$ the Grothendieck-Lefschetz fixed-point formula reads $$ #X(\mathbb{F}{q^m}) = \sum{i=0}^{2d} (-1)^i , \mathrm{tr}!\big(\mathrm{Frob}^m \mid H^i(X_{\bar{\mathbb{F}}q}, \mathbb{Q}\ell)\big). $$ Rationality of the zeta function follows from finite-dimensionality (Grothendieck 1965); the functional equation follows from Poincaré duality; and Deligne's 1974 Weil I proves the Riemann Hypothesis part, that the Frobenius eigenvalues on $H^{i}$ have absolute value $q^{i /2}$ . Together these are the Weil conjectures, whose statement is cross-referenced in 21.05.01. The trace formula is the engine: it converts the arithmetic question "how many points" into the linear-algebra question "what are the Frobenius eigenvalues on cohomology."

Synthesis. The étale theory is the foundational reason a single cohomology machine serves both topology and arithmetic. The same site that recovers Galois cohomology on a point and the Picard group on a variety assembles, through the inverse-limit $ℓ$ -adic construction, into a Weil cohomology with a Galois action; this is exactly the structure the Weil conjectures demand. The comparison theorem shows the new theory is dual to the classical topological one over $C$ , reproducing Betti numbers, while Poincaré duality on the étale site generalises the classical duality and pins down the Frobenius-weight symmetry. Putting these together, the central insight is that the Lefschetz trace formula is the bridge from geometry to point-counting: the foundational reason $# X (F_{q^{m}})$ is governed by eigenvalues is that finite-dimensional $ℓ$ -adic cohomology with its Frobenius action exists at all, and Deligne's weight bound is what makes those eigenvalues as constrained as the Riemann Hypothesis predicts. This pattern recurs in the Galois-representation machinery of 21.05.01 and builds toward the entire Langlands-era arithmetic of varieties.

Full proof set Master

Proposition (Galois-cohomology identification, full statement and proof). Let $k$ be a field with separable closure $\overset{ˉ}{k}$ and absolute Galois group $G = Gal (\overset{ˉ}{k} / k)$ , a profinite group. The global-sections functor on the small étale site of $Spec k$ corresponds, under the equivalence of categories between abelian étale sheaves and discrete $G$ -modules, to the $G$ -invariants functor; hence $H_{\overset{e}{ˊ} t}^{i} (Spec k, F) ≅ H_{cont}^{i} (G, M_{F})$ for the corresponding discrete $G$ -module $M_{F}$ .

Proof. A connected scheme étale over $Spec k$ is $Spec k^{'}$ for a finite separable extension $k^{'} / k$ inside $\overset{ˉ}{k}$ ; a general object of the small étale site is a finite disjoint union of such. Sending $Spec k^{'} \mapsto Hom_{k} (k^{'}, \overset{ˉ}{k})$ defines the fibre functor to finite sets, and the $G$ -action by post-composition makes the target a finite discrete $G$ -set. This functor is an equivalence from the small étale site of $Spec k$ to the category of finite discrete $G$ -sets: full faithfulness is the primitive-element / Galois-theory dictionary, and essential surjectivity expresses every finite $G$ -set as a disjoint union of orbits $G / H$ realised by the fixed field of $H$ . An abelian sheaf for the canonical topology on finite $G$ -sets is precisely a discrete $G$ -module $M$ , with sections over $G / H$ given by $M^{H}$ . Under this dictionary $Γ (Spec k, -)$ , sections over the terminal object $G / G$ , becomes $M \mapsto M^{G}$ . The functor $M \mapsto M^{G}$ on discrete $G$ -modules is left exact, and its right derived functors are by definition the continuous cohomology $H_{cont}^{i} (G, M) = lim_{U} H^{i} (G / U, M^{U})$ over open normal subgroups $U$ . Since equivalences of abelian categories preserve injective resolutions and derived functors of corresponding left-exact functors, $R^{i} Γ (Spec k, F) ≅ H_{cont}^{i} (G, M_{F})$ . $□$

Proposition ( $H_{\overset{e}{ˊ} t}^{1} (X, G_{m}) = Pic (X)$ , proof via torsors and descent). For any scheme $X$ , first étale cohomology of $G_{m}$ is canonically the Picard group.

Proof. On any site, $H^{1}$ of a sheaf of groups $G$ classifies $G$ -torsors up to isomorphism; for the abelian sheaf $G_{m}$ this is $H_{\overset{e}{ˊ} t}^{1} (X, G_{m}) ≅ {G_{m} -torsors on X_{\overset{e}{ˊ} t}} / ≅$ . A $G_{m}$ -torsor is the same data as an invertible sheaf together with a chosen $G_{m}$ -action, so torsors correspond to line bundles; the group structure on $H^{1}$ matches tensor product of line bundles. It remains to see that étale-locally split invertible sheaves are Zariski-locally split, so that the étale and Zariski Picard groups coincide. This is Hilbert 90: for a local ring $A$ , $H_{\overset{e}{ˊ} t}^{1} (Spec A, G_{m}) = 0$ , because any $G_{m}$ -torsor over a strictly henselian (or just local) base reduces to the identity torsor — concretely, descent data for a line bundle along a finite Galois étale cover $Spec B / Spec A$ is a $1$ -cocycle in $B^{\times}$ , and $H^{1} (Gal (B / A), B^{\times}) = 0$ is the classical Theorem 90. Therefore the comparison map $H_{Zar}^{1} (X, G_{m}) \to H_{\overset{e}{ˊ} t}^{1} (X, G_{m})$ is an isomorphism, and the left side is $Pic (X)$ by the standard Čech computation. $□$

Proposition (rationality of the zeta function from finite-dimensional cohomology). Let $X$ be smooth proper over $F_{q}$ with $ℓ \neq = char F_{q}$ . Granting the trace formula, the zeta function $Z (X, t)$ is a rational function with $Q$ coefficients.

Proof. For a single finite-dimensional $Q_{ℓ}$ -vector space $V$ with endomorphism $Φ$ , the identity $\sum_{m \geq 1} tr (Φ^{m}) t^{m} / m = - lo g det (1 - t Φ)$ holds as formal power series (diagonalise over $\overset{ˉ}{Q}_{ℓ}$ and sum the geometric series eigenvalue-by-eigenvalue). Exponentiating, $exp (\sum_{m} tr (Φ^{m}) t^{m} / m) = det (1 - t Φ)^{- 1}$ . Apply this with $V = H^{i} (X_{\overset{ˉ}{F}_{q}}, Q_{ℓ})$ and $Φ = Frob$ for each $i$ . Inserting the trace formula $# X (F_{q^{m}}) = \sum_{i} (- 1)^{i} tr (Frob^{m} ∣ H^{i})$ into the definition of $Z (X, t)$ and using additivity of $exp$ over the alternating sum gives $$ Z(X, t) = \prod_{i=0}^{2d} \det!\big(1 - t,\mathrm{Frob} \mid H^i(X_{\bar{\mathbb{F}}q}, \mathbb{Q}\ell)\big)^{(-1)^{i+1}}. $$ Each determinant is a polynomial in $t$ because each $H^{i}$ is finite-dimensional; the alternating product is therefore a rational function. That the polynomials have $Q$ (indeed $Z$ ) coefficients independent of $ℓ$ is the deeper part, established by Grothendieck's formalism together with Deligne's integrality results. $□$

Connections Master

Smooth, étale, and unramified morphisms 04.02.05. The étale morphisms defined there are precisely the "open sets" of the étale site: flat unramified maps, smooth of relative dimension zero, the algebraic analogue of local isomorphisms. The covering families that generate the Grothendieck topology are jointly surjective étale families, so the entire site rests on that morphism class; without the étale-morphism theory there is no finer notion of cover to replace Zariski opens.
Cohomology of line bundles on projective space 04.03.04. The Zariski-side computation $H_{Zar}^{1} (X, G_{m}) = Pic (X)$ and the explicit cohomology of $O (d)$ on $P^{n}$ are the classical shadow that étale cohomology must reproduce and extend. The étale theory recovers the same Picard group through descent and Hilbert 90, and for coherent coefficients the étale and Zariski cohomologies agree, so the projective-space table is the base case the finer theory generalises.
$ℓ$ -adic Galois representations and the Weil conjectures 21.05.01. This is the consumer of the present unit: the Galois action on $H^{i} (X_{\overset{ˉ}{k}}, Q_{ℓ})$ constructed here is exactly the source of $ℓ$ -adic Galois representations, and the Lefschetz trace formula plus Deligne's weight bound are the engine of the Weil conjectures whose statement is cross-linked there. The forward reference that unit makes to a "forthcoming étale-cohomology pointer" is resolved by this unit.
Serre's GAGA comparison theorem 04.09.03. GAGA is the algebraic-to-analytic bridge that runs parallel to the étale comparison theorem: where GAGA compares coherent algebraic and analytic sheaves on a projective variety over $C$ , the étale comparison theorem compares torsion étale cohomology with singular cohomology of the analytification. Both express that the algebraic category over $C$ loses no cohomological information, and together they let constructible and coherent data pass between the algebraic and complex-analytic halves of the corpus.

Historical & philosophical context Master

The étale topology was conceived by Alexander Grothendieck around 1958-1963 as the technical means to prove the Weil conjectures, which André Weil had formulated in 1949 by analogy between point-counting over finite fields and the topology of complex varieties ^{[Weil 1949]}. Weil had observed that if there were a cohomology theory for varieties over finite fields with a Lefschetz fixed-point formula, then rationality, the functional equation, and the Riemann-Hypothesis bound for zeta functions would follow from formal properties — finite-dimensionality, Poincaré duality, and a weight structure. The missing ingredient was the cohomology theory itself, since the Zariski topology was visibly too coarse to supply it. Grothendieck's radical move, developed with Michael Artin and Jean-Louis Verdier in the seminar SGA 4 ^{[Grothendieck-Artin-Verdier 1972]}, was to abandon point-set topology entirely and replace open covers by the abstract notion of a Grothendieck topology, with étale morphisms playing the role of open immersions.

The philosophical shift is that "space" is dethroned in favour of "site": the cohomologically relevant data of a scheme is not its underlying point-set but the category of étale maps into it together with a notion of covering. This is the topos-theoretic viewpoint, and étale cohomology is its first decisive success. Grothendieck proved rationality of the zeta function and the Lefschetz formula by 1965 ^{[Grothendieck 1965]}, but the Riemann-Hypothesis part resisted his program of standard conjectures. It was Pierre Deligne who, in La conjecture de Weil I (1974) ^{[Deligne 1974]}, proved the weight bound by an ingenious argument combining Lefschetz pencils, the monodromy of vanishing cycles, and a Rankin-Selberg positivity trick, sidestepping the standard conjectures. Deligne's proof is widely regarded as one of the deepest achievements of twentieth-century mathematics, and it vindicated Grothendieck's two-decade gamble that a wholly new notion of topology was the right foundation for arithmetic geometry.

Bibliography Master

@book{SGA4,
  author    = {Artin, Michael and Grothendieck, Alexander and Verdier, Jean-Louis},
  title     = {Th{\'e}orie des Topos et Cohomologie {\'E}tale des Sch{\'e}mas (SGA 4)},
  publisher = {Springer-Verlag},
  series    = {Lecture Notes in Mathematics},
  volume    = {269, 270, 305},
  year      = {1972--1973}
}

@book{SGA4demi,
  author    = {Deligne, Pierre},
  title     = {Cohomologie {\'E}tale (SGA 4 1/2)},
  publisher = {Springer-Verlag},
  series    = {Lecture Notes in Mathematics},
  volume    = {569},
  year      = {1977}
}

@article{Deligne1974WeilI,
  author  = {Deligne, Pierre},
  title   = {La conjecture de {W}eil. {I}},
  journal = {Publ. Math. Inst. Hautes {\'E}tudes Sci.},
  volume  = {43},
  year    = {1974},
  pages   = {273--307}
}

@incollection{Grothendieck1965Lefschetz,
  author    = {Grothendieck, Alexander},
  title     = {Formule de {L}efschetz et rationalit{\'e} des fonctions {L}},
  booktitle = {S{\'e}minaire Bourbaki, Vol. 9, Exp. No. 279},
  publisher = {Soc. Math. France},
  year      = {1965},
  pages     = {41--55}
}

@book{MilneEtale,
  author    = {Milne, James S.},
  title     = {{\'E}tale Cohomology},
  publisher = {Princeton University Press},
  series    = {Princeton Mathematical Series},
  volume    = {33},
  year      = {1980}
}

@article{Weil1949,
  author  = {Weil, Andr{\'e}},
  title   = {Numbers of solutions of equations in finite fields},
  journal = {Bull. Amer. Math. Soc.},
  volume  = {55},
  year    = {1949},
  pages   = {497--508}
}

Prerequisites

04.02.05
04.03.04
04.03.01

Tier anchors

beginner: Milne *Étale Cohomology* §I informal — a finer notion of open cover that finally sees arithmetic
intermediate: Milne *Étale Cohomology* (PUP 1980) Chapters II–III; Deligne SGA 4½ *Arcata*
master: Grothendieck et al. SGA 4 (Tôhoku-style site/topos foundations) and SGA 4½/SGA 5; Deligne 1974 *La conjecture de Weil I* (Publ. IHÉS 43); Milne *Étale Cohomology*

References

Grothendieck, Artin, Verdier — SGA 4: Théorie des topos et cohomologie étale des schémas · Lecture Notes in Math. 269, 270, 305 (1972-73), site/topos foundations and étale cohomology
Deligne — SGA 4½ (Cohomologie étale) · Lecture Notes in Math. 569 (1977), Arcata notes and the streamlined ℓ-adic formalism
Deligne — La conjecture de Weil. I · Publ. Math. IHÉS 43 (1974) 273-307, Weil Riemann Hypothesis / weight bound
Milne — Étale Cohomology · Princeton Univ. Press (1980); also the online lecture notes, étale site / cohomology / comparison / duality
Grothendieck — Formule de Lefschetz et rationalité des fonctions L · Séminaire Bourbaki 279 (1965), Lefschetz trace formula and rationality of the zeta function

Estimated time

beginner: 20m
intermediate: 55m
master: 90m