ETALE COHOMOLOGY AND THE WEIL CONJECTURES

BRUNO KLINGLER

Sommaire
1. What is the “shape” of a scheme? 3
1.1. Characteristic zero 4
1.2. Positive characteristic 4
2. The Weil conjectures, first version 5
2.1. Reminder on finite fields 5
2.2. Schematic points 5
2.3. Scheme-valued points 6
2.4. Counting points for schemes of finite type over Fq 7
2.5. Weil conjectures, first version 8
3. Zeta functions 9
3.1. Riemann zeta function 9
3.2. Zeta functions for schemes of finite type over Z 14
3.3. Geometric zeta function for a scheme of finite type over Fq 15
3.4. Properties of zeta functions 16
3.5. Some examples 17
3.6. Some questions and conjectures 17
3.7. Weil conjectures 18
4. The Weil conjectures for curves 19
4.1. Heuristics 20
4.2. The Riemann-Roch’s formula 20
4.3. Rationality 21
4.4. Functional equation 23
4.5. The geometric Frobenius 24
4.6. The Riemann hypothesis for curves over finite fields 25
5. Transition to étale cohomology 26
5.1. Heuristic for the Weil conjectures: about the Lefschetz trace formula 26
5.2. Weil cohomologies 28
5.3. Applications to the Weil conjectures 30
6. Differential calculus 32
6.1. Thickenings 32
6.2. First infinitesimal neighborhood 32
6.3. Conormal subsheaf of an immersion 33
6.4. Cotangent sheaf: definition 35
6.5. Cotangent sheaf: basic properties 36
6.6. Digression: the De Rham complex 37
1
2 BRUNO KLINGLER

7. Smooth, net and étale morphisms 38

7.1. Definitions 38
7.2. Main properties 38
7.3. Local coordinates 39
7.4. Jacobian criterion 39
7.5. Implicit functions theorem 39
7.6. Proof of proposition 7.2.1 40
7.7. A remark on smoothness 45
7.8. Smoothness, flatness and regularity 46
7.9. Examples of étale morphisms 54
8. Etale fundamental group 54
8.1. Reminder on the topological fundamental group 54
8.2. The étale fundamental group 55
9. Sites and sheaves 56
9.1. Presheaves 56
9.2. Sheaves on topological spaces 56
9.3. Sites 57
9.4. Sheaves on a site 59
9.5. The abelian category of abelian sheaves; cohomology 61
9.6. Functoriality 61
10. Sheaves on schemes; fpqc sheaves 63
10.1. Cohomology of sheaves on schemes 63
10.2. A criterion to be a sheaf on (Sch/S)τ 63
10.3. fpqc sheaves and faithfully flat descent 64
10.4. The fpqc sheaf defined by a scheme 66
10.5. Constant sheaf 67
11. Etale sheaves 67
11.1. Neighborhoods and stalks 67
11.2. Strict localisation 68
11.3. Direct image and inverse image, the étale case 69
11.4. Extension by zero 71
11.5. The fundamental triangle 72
12. Etale cohomology 74
12.1. Cohomology with support 74
12.2. Nisnevich excision 75
13. Čech cohomology and étale cohomology of quasi-coherent sheaves 75
13.1. Čech cohomology for coverings 75
13.2. Čech to cohomology spectral sequence 77
13.3. Proof of Theorem 13.0.1 78
13.4. Other applications of Čech cohomology 78
13.5. Flasque sheaves 79
13.6. The Leray spectral sequence 80
13.7. Cohomology of points: Galois cohomology 82
14. Cohomology of curves over an algebraically closed field 82
14.1. The divisorial exact sequence 83
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 3

14.2. Proof of Theorem 14.0.1 84

14.3. Brauer groups and the proof of proposition 14.2.4 85
15. Constructible sheaves 87
15.1. Pathology of the étale constant sheaf Z 88
15.2. Locally constant constructible sheaves 89
15.3. Constructible sheaves 90
15.4. Properties of constructible sheaves on Noetherian schemes 91
16. Proper base change 94
16.1. The classical topological case 95
16.2. The étale case 96
16.3. Proof of Theorem 16.2.3 97
References 105

1. What is the “shape” of a scheme?

Starting in 1895 Poincaré [Po95] associated natural invariants to any separated locally
compact path-connected topological space X: its (co)homology groups HBetti • (X, Q) and
its fundamental group π1 (X), the group of loops in X up to homotopy. While cohomol-
ogy groups were originally defined in many different ways (simplicial cohomology, cellular
cohomology, singular cohomology...) Eilenberg and Steenrod [ES52] axiomatized in 1952
the properties of any good cohomology theory, with the effect that all these definitions
essentially coincide. From a modern point of view a unifying definition is via sheaf theory:
•
HBetti (X, Q) := H • (X, QX ) ,
where QX denotes the constant sheaf with value Q on X. These cohomology groups
are invariant under homotopy equivalence for X. Nowadays the shape of X has to be
understood as the class of X in the homotopy category of spaces.
Suppose now that X is a scheme. We would like to understand its shape, in particular
its cohomology and its fundamental group. The underlying topological space |X| with
its Zariski topology is usually not separated, in a very strong sense. Recall the
Definition 1.0.1. A topological space X is irreducible if any two non-empty open subsets
of X have non-empty intersection. A scheme X is irreducible if |X| is.
•
If we define the cohomology groups of X as HBetti (|X|, Q), this definition makes sense
but is of no interest:
Lemma 1.0.2. (Grothendieck) If Y is an irreducible topological space then H • (Y, F) = 0
for any constant sheaf F on X.
Proof. Let F := H 0 (X, F) be the group of global sections of the sheaf F on X. As F
is constant, it is the sheafification of the presheaf with value F on any connected open
subset U of X. As Y is irreducible any open subset of Y is connected. Hence F(U ) = F
for any open subset U of Y and the sheaf F is flasque, in particular acyclic. 
As a consequence, any reasonable definition of the “shape of a scheme” will depend
not only on the underlying topological space but also on the finer schematic structure.
4 BRUNO KLINGLER

1.1. Characteristic zero. In the case where X/k is a separated scheme of finite type
over a field k of characteristic char k = 0 which admits an embedding σ : k ,→ C (i.e.
k has cardinality at most the continuum), one can try using the embedding σ to define
invariants attached to X. Let us write

X σ := X ×k,σ C (:= X ×Spec k,σ Spec C) .

A natural topological invariant associated to X/k and σ is then

•
HBetti ((X σ )an , Q) ,

where (X σ )an denotes the complex analytic space associated to X. How does it depend
on σ?

Theorem 1.1.1 (Serre). Suppose that X is a smooth projective variety over k. The
i
dimension bi (X) := dimQ HBetti ((X σ )an , Q) is independent of σ. We call it the i-th
Betti number of X.

Proof.

H p ((X σ )an , Ωq(X σ )an )

M
i
HBetti ((X σ )an , C) '
p+q=i

H p (X σ , ΩqX σ ) .
M
'
p+q=i

The first isomorphism is the Hodge decomposition for the cohomology of the smooth
complex projective variety X σ (see [Vois07] for a reference on Hodge theory). The
second one is Serre’s GAGA theorem [Se56] for smooth complex projective varieties. We
conclude by noticing that H p (X σ , ΩqX σ ) = H p (X, ΩqX )⊗k,σ C has dimension independent
of σ. 

Remark 1.1.2. Serre’s theorem can be extended to quasi-projective varieties using more
Hodge theory (logarithmic complex).

Even if embedding k in C defines unambiguously the Betti numbers of the scheme

X/k, it does not define its fundamental group: in 1964 indeed, Serre constructed a
smooth projective X over a number field k and σ, τ : l ,→ C two different infinite places
of k such that
π1 ((X σ )an ) 6' π1 ((X τ )an ) .
In particular (X σ )an is not in general homotopy equivalent to (X τ )an .

1.2. Positive characteristic. The situation is worse for schemes over a field k of
positive characteristic. What do we understand as the “shape” of such a scheme? The
question is particularly relevant if k is a finite field Fq (finite field with q = pn elements, p
prime number). The basic theme of the Weil conjectures is that the shape of a separated
scheme X of finite type over Fq is, in first approximation, described by counting points
of X.
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 5

2. The Weil conjectures, first version

2.1. Reminder on finite fields. Let k be a field and consider the natural ring
homomorphism Z → k associating n · 1k to n ∈ Z. Its kernel is a prime ideal of Z hence
of the form pZ, p prime number, called the characteristic of k.
Suppose now that k is a finite field, hence necessarily of positive characteristic p. In
particular Fp ,→ k and k is a finite dimensional Fp -vector-space, so |k| = pn for some
n ∈ N∗ . Fix an algebraic closure Fp of Fp .
Theorem 2.1.1. Let p be a prime number. For any n ∈ N∗ , there exists a unique field
Fq of cardinal q = pn (up to isomorphism). If Frp : Fp → Fp is the arithmetic Frobenius
Frn =1
associating to x its p-power Frp (x) := xp , the field Fq is the fixed field Fp p .
n
The field Fpn is nothing else than the splitting field of X p − X, in particular it is
Galois over Fp with Galois group Z/nZ generated by Frq and
Gal(Fp /Fp ) = lim Gal(Fpn /Fp ) = lim Z/nZ =: Ẑ
←− ←−
n

topologically generated by Frp .

Examples 2.1.2.
F4 = F2 [X]/(X 2 + X + 1)
F8 = F2 [X]/(X 3 + X + 1)
F9 = F3 [X]/(X 2 + 1)
2.2. Schematic points. Which points of X do we want to count? Recall that we have
two different notions of points for schemes.
The first notion of point for a scheme X is the obvious one: an element x ∈ |X|. We
call such a point a schematic point of X. Define by
Z(x) := {x}
(r)
`
the associated closed subscheme of X. One obtains natural partitions of |X| =
` r∈N X =
r∈N X(r) , where

X (r) = {x ∈ |X|, codimX Z(x) = r}

X(r) = {x ∈ |X|, dim Z(x) = r}
Here dimension and codimension are understood in their topological sense: the di-
mension dim Z of a scheme Z is the maximum length n of a chain Z0 ( Z1 ( · · · ( Zn
of non-empty closed irreducible subsets of Z; the codimension codimX Y of a closed irre-
ducible subscheme Y ⊂ X is the maximal length n for a chain Z0 = Y ( Z1 ( · · · ( Zn
of closed irreducible subsets of X.
Remark 2.2.1. Recall that codimX Y is not necessarily equal to dim X − dim Y , even
for X irreducible: take X = Spec k[[t]][u], Y = V (tu − 1). Then dim Y is the Krull
dimension of the field k[[t]][u]/(tu − 1) = k[[t]][ 1t ], hence zero. On the other hand
dim X = dim k[[t]] + 1 = 2 and codimX Y is the height of the ideal (tu − 1), hence 1.
6 BRUNO KLINGLER

The set |X| is usually infinite. To count schematic points, we will need a notion of
“size” which guarantees that there are only finitely many schematic points of given size.
This can be conveniently done for any scheme X of finite type over Z if one restrains
oneself to the atomization X(0) of X.
The following lemma is straightforward:
Lemma 2.2.2. Let X be a scheme of finite type over Z and let x ∈ |X|. The following
properties are equivalent:
(a) x ∈ X(0) .
(b) the residue field k(x) is finite.
If we define the norm of x ∈ X(0) as N (x) = |k(x)|, there are only finitely many x ∈ X(0)
with given norm.
Our vague question about counting points can thus be precisely rephrased as:
Problem 2.2.3. Given a scheme X of finite type over Z, compute the number of points
of X(0) of given norm.
This problem looks even more natural if one extends it a little bit. Let us consider
the case X = Spec OK , where OK denotes the ring of integers of a number field K. We
want to count not only prime ideals of OK (i.e. points of X(0) ) but all ideals of OK . As
OK is a Dedekind ring we are in fact counting effective zero-cycles on X in the sense of
the following:
Definition 2.2.4. Let X be a scheme. The group of algebraic cycles on X is the free
abelian group Z(X) :=< |X| > generated by the points of X. Hence an element α ∈
Z(X) is a linear combination α = ri=1 ni · xi , ni ∈ Z, xi ∈ |X|. The cycle α is said to
P
be effective if all ni are positive.
The group Z(X) is naturally graded: Z p (X) =< X (p) > or Zp (X) =< X(p) >. If X
is of finite type over Z the norm N on X(0) extends to N : Z0 (X) → Q by
Xr r
Y
N( ni · xi ) = N (xi )ni .
i=1 i=1
Once more there are only finitely many effective zero-cycles of given norm and Prob-
lem 2.2.3 can be extended to:
Problem 2.2.5. Given a scheme X of finite type over Z, compute the number of effective
zero-cycles on X of given norm.
2.3. Scheme-valued points. The second notion of points come from the interpreta-
tion of a scheme as a functor.
Definition 2.3.1. Let S be a scheme and X, T two S-schemes. One defines the set of
T -points of X as:
X(T )S := HomS (T, X) .
We denote X(T ) := X(T )Z .
We are interested in the case T = Spec K, K a field.
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 7

Proposition 2.3.2. Let K be a field. Then

X(K) = {(x, i), x ∈ |X|, i : k(x) → K field homomorphism} .
More generally if k ⊂ K is a field extension:
a
X(K)k = Homk−alg (k(x), K) .
x∈|X|

Proof. Let s : Spec K → X be a morphism. It is entirely defined by the continuous map

|s| : |Spec K| → |X|, i.e. the point x image of |Spec K|, plus the morphism of sheaves of
rings s−1 OX → OSpec K over Spec K, i.e. the ring morphism
Γ(Spec K, s−1 OX ) =: OX,x → Γ(Spec K, OSpec K ) = K .
Notice that the ring morphism OX,x → k uniquely factorizes through k(x).
The proof of the generalization is similar. 
Exercice 2.3.3. Show that X(T )Fp = X(T ) and X(T )Q = X(T ) but that X(T )k 6= X(T )
for a general field k.
Let X be a scheme of finite type over Fq . Counting points of X can also be understood
as:
Problem 2.3.4. Given a scheme X of finite type over Fq , compute the number |X(Fqr )Fq |
for all positive integers r.
2.4. Counting points for schemes of finite type over Fq . Problem 2.2.3 and
Problem 2.3.4 are essentially equivalent for schemes of finite type over Fq :
Lemma 2.4.1. Let X be a scheme of finite type over Fq . Then
X
|X(Fqr )Fq | = e · {x ∈ X(0) / deg(x)(:= [k(x) : Fq ]) = e} .
e|r

Proof. By Proposition 2.3.2:

a
X(Fqr )Fq = HomFq (k(x), Fqr ) .
x∈|X|

In particular if HomFq (k(x), Fqr ) 6= 0, the field k(x) is finite hence x belongs to X(0)
thanks to Lemma 2.2.2. Moreover deg(x)|r. Hence:
a
X(Fqr )Fq = HomFq (k(x), Fqr ) .
deg(x)|r

Now Gal(Fqr /Fq ) ' Z/rZ acts transitively on HomFq (k(x), Fqr ), with stabilizer Gal(Fqr /k(x)) '
r
Z/( deg(x) ) · Z. Thus:
| HomFq (k(x), Fqr )| = deg(x) .

Counting points of a scheme X of finite type over Fq is usually a hard problem. The
following immediate corollary of Lemma 2.4.1 enable us to do it in simple cases.
8 BRUNO KLINGLER
`
Corollary 2.4.2. If a scheme X of finite type over Fq satisfies X(0) = (Xi )(0) for a
family (Xi ) of subschemes (i.e. closed subscheme of open subscheme) of X then
a
X(Fqr )Fq = Xi (Fqr )Fq .
Examples 2.4.3. (1) X = AnFq . As AnFq (Fqr )Fq = HomFq −alg (Fq [T1 , · · · , Tn ], Fqr ) '
(Fq )⊕n one obtains
|AnFq (Fqr )Fq | = q nr .
(2) X = PnFq . As (PnFq )(0) = ni=0 (AiFq )(0) one obtains
`

|PnFq (Fqr )Fq | = 1 + q r + q 2r + · · · + q nr .

(3) Let us give an example which shows that counting points is usually difficult. Let
X = {y 2 = x3 + x, y 6= 0 ⊂ A2Fq } ⊂ X = {y 2 z = x3 + xz 2 ⊂ P2Fq } .
The variety X is an affine curve, its closure X in P2Fq is an elliptic curve. Hence:
|X(Fq )Fq | = |X(Fq )Fq | + |{u ∈ Fq , u3 + u = 0}| + |{u ∈ Fq , u3 = 0}|
( √
1 + 1 if −1 6∈ Fq (i.e. q = −1 mod 4)
= |X(Fq )Fq | + √
3 + 1 if −1 ∈ Fq (i.e. q = 1 mod 4)
Recall that there are exactly (q − 1)/2 squares in F∗q .
Let us assume that q = −1 mod 4. As −1 is not a square, if c ∈ F∗q then either
c or −c is a square but not both. Hence u3 + u or −(u3 + u) = ((−u)3 + (−u)) is
a square but not both. Hence a3 + a = b2 for exactly (q − 1)/2 values of a, with
two choices for b each time. Hence
q−1
|X(Fq )Fq | = 2 × = q − 1 and |X(Fq )Fq | = (q − 1) + 2 = q + 1 .
2
I don’t know of any general procedure for q = 1 mod 4. For q = 5 writing
the table of all possibilities one obtains |X(F5 )F5 | = 0, hence |X(F5 )F5 | = 4.
2.5. Weil conjectures, first version. In [We49] Weil proposed a general set of
conjectures describing the number of points of any scheme of finite type over Fq (we will
come back later to the history of these conjectures). Recall first:
Definition 2.5.1. A q-Weil number of weight m ∈ N is an algebraic number whose
Archimedean valuations are all q m/2 .
Remark 2.5.2. In the literature Weil numbers are sometimes assumed to be algebraic
integers.
Example 2.5.3. 1 ± 2i is a 5-Weil number of weight 1.
Conjecture 2.5.4 (Weil). Let X be a scheme of finite type over Fq .
1. (Rationality) There exists a finite set of algebraic integers αi , βj such that:
X X
∀r ∈ N, |X(Fqr )Fq | = αir − βjr .
qd
2. (Functional equation) If X is smooth and proper of pure dimension d then γ 7→ γ
induces a permutation of the αi ’s and a permutation of the βj ’s.
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 9

3. (Purity) If X has dimension d, the αi ’s and βj ’s are Weil q-numbers of weights

in [0, 2d].
If moreover X is smooth and proper the weights of the αi ’s are even while the
weights of the βj ’s are odd.
4. (link with topology) Suppose that X/Fq is the smooth and proper special fiber of
a smooth and proper X /R, Fq  R ,→ C. Then
(
|{αi , weight(αi ) = m}| if m is even,
dimC H m ((XC )an , C) =
|{βj , weight(βj ) = m}| if m is odd.

3. Zeta functions
We will refine our understanding of points for a scheme X of finite type over Z by
constructing a generating function encoding the numbers N (x), x ∈ X(0) . How do we
construct such a generating function? There is no general recipe. We can only learn
through experiment, starting with Euler and Riemann.

3.1. Riemann zeta function. The Riemann zeta function is the well-known function
of one complex variable s:
∞
X 1
ζ(s) = .
ns
n=1

It encodes the “counting” of points of Spec Z, i.e. of prime numbers. It was first studied
by Euler (around 1735) for s real, then by Riemann for s complex (1859). This function
serves as a model for any other zeta or L-function. Let us prove its basic properties.

3.1.1. Convergence.

Proposition 3.1.1 (Riemann). The function ζ(s) converges absolutely (uniformly on

compacts) on the domain Re(s) > 1, where it defines a holomorphic function. It diverges
for s = 1.

Proof. Write s = u + iv, u, v ∈ R. Then |n−s | = n−u . To prove absolute convergence we

can thus assume that s belongs to R.
t−s is decreasing. Thus the series ∞ −s converges if
P
For s ∈ R the functionR ∞t 7→ 1 n
and only if the integral 1 t−s dt converges. Hence the convergence for s > 1 and the
divergence at s = 1.
The previous comparison yields:
∞ Z ∞
X
−s (N − 1)1−Re(s)
| n |≤ t−Re(s) dt = ,
N −1 Re(s) − 1
N

which proves the uniform convergence on compacts.

The function ζ(s) is a limit, uniform on compacts, of holomorphic functions. Hence
it is holomorphic. 
10 BRUNO KLINGLER

3.1.2. Euler product.

Proposition 3.1.2 (Euler).

Y 1
ζ(s) = ,
p
1 − p−s

where the product on the right is absolutely convergent for Re(s) > 1.

To prove Proposition 3.1.2, we introduce the notion of a completely multiplicative

function:

Definition 3.1.3. A function a : N∗ → C is completely multiplicative if a(mn) =

a(m)a(n) for all m, n ∈ N∗ .

Proposition 3.1.2 follows immediately from the following lemma:

Lemma 3.1.4. Let a : N∗ → C be completely multiplicative. The following are equiva-

lent:
P∞
(i) |a(n)| < +∞.
Q1 1
(ii) p 1−a(p) < +∞.
If one of these equivalent conditions is satisfied then
∞
X Y 1
a(n) = .
p
1 − a(p)
1

Proof. Assume (i). Thus for any prime p the sum m a(pm ) converges absolutely, with
P
1
sum 1−a(p) . Let E(x) ⊂ N∗ be the set of integers whose prime factors are smaller than
x. As
X YX Y 1
a(n) = a(pm ) = ,
p<x m p<x
1 − a(p)
n∈E(x)

one obtains
∞
X Y 1 X X
a(n) − = a(n) ≤ |a(n)| .
p<x
1 − a(p)
1 n6∈E(x) n≥x

1
converges to ∞
Q P
Hence p 1−a(p) 1 a(n), absolutely (replacing a by |a|).
Conversely assume (ii). Then
X X Y 1
|a(n)| ≤ |a(n)| =
n<x p<x
1 − a(p)
n∈E(x)

hence (i). 

3.1.3. Formal Dirichlet series. It will be convenient to first work with formal series,
without convergence questions:
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 11

∞
X an
Definition 3.1.5. A formal Dirichlet series is f = where n ∈ N∗ , an ∈ C. Given
ns
1
∞
X bn
another formal Dirichlet series g = one defines
ns
1
X an + bn
f +g = ;
ns
n≥1
X cn X
f ·g = with cn = ap bq .
ns pq=n
n≥1

Formal Dirichlet series form a commutative ring Dir(C), where we can perform formal
computations.
∞
X an
Definition 3.1.6. Let f = be a formal Dirichlet series. If f 6= 0 one defines the
ns
1
order ω(f ) as the smallest integer n with an 6= 0 (if f = 0 one puts ω(f ) = +∞).
Notice that the subsets {f | ω(f ) ≥ N } are ideals of Dir(C). They define a topology
on Dir(C) making Dir(C) a complete topological ring. Hence:
Corollary 3.1.7. A sequence (fn )n∈N of Dir(C) is summable if and only if lim ω(fn ) =
n→+∞
+∞; a sequence (1 + fn )n∈N , with ω(fn ) > 1 for all n, is multipliable in Dir(C) if and
only if lim ω(fn ) = +∞.
n→+∞

Lemma 3.1.4 implies immediately the following:

Proposition 3.1.8. Let a : N∗ → C be a completely multiplicative function. Consider
the formal identity
X a(n) Y  a(p) −1

= 1− s .
ns p
p
n≥1

Given a real number α, the left hand side converges absolutely for Re(s) > α if and only
if the right hand side converges absolutely for Re(s) > α.
3.1.4. Extension to Re(s) > 0.
Proposition 3.1.9 (Riemann). The function ζ(s) extends meromorphically to Re(s) > 0
with a unique simple pole at s = 1 and residue 1.
∞
X (−1)n+1
Proof. Define ζ2 (s) = . Recall:
ns
n=1

Lemma 3.1.10 (Abel). Let (an )n∈N and (bn )n∈N be two sequences of complex numbers.
Then
m0 m 0 −1 n m0
! !
X X X X
an bn = ai (bn − bn+1 ) + an bm0 .
n=m n=m i=m n=m
12 BRUNO KLINGLER
Pn
In particular if there exists ε > 0 such that | i=m ai | ≤ ε for all m ≤ n ≤ m0 and if the
sequence (bn )n∈N is real and decreasing then

m 0
X
an bn ≤ εbm .
n=m

Using Abel’s lemma for an = (−1)n and bn = n−s one checks that ζ2 (s) converges
(not absolutely!) for Re(s) > 0. Notice that
X 1 X 1
n+1
ζ(s) − ζ2 (s) = s
(1 − (−1) ) = s ks
· 2 = 21−s ζ(s)
n
n 2
n=2k

hence ζ2 (s) = (1 − 21−s )ζ(s). So (s − 1)ζ(s) extends meromorphically to Re(s) > 0.

As ζ2 (1) = log 2 one obtains lims→1 (s − 1)ζ(s) = 1.
More generally for r ∈ N \ {0, 1} we define

1 1 r−1 1
ζr (s) = + ··· + − s + + ···
1s (r − 1)s r (r + 1)s

Once more: ζr (s) is analytic for Re(s) > 0 and

 
1
1 − s−1 ζ(s) = ζr (s) .
r

Suppose that s 6= 1 is a pole of ζ(s).

√
2π −1k
• for r = 2 one obtains 2s−1 = 1 hence s = log 2 + 1 for some k ∈ N∗ .
√
2π −1l
• for r = 3 one obtains similarly s = log 3 + 1 for some l ∈ N∗ .
Hence 3k = 2l : contradiction.


3.1.5. Extension to C and functional equation. Recall the Γ function:

Z +∞
dt
Γ(s) = ts e−t ,
0 t
which converges for Re(s) > 0. It satisfies Γ(1) = 1 and the functional equation Γ(s+1) =
sΓ(s). Hence Γ(s) extends meromorphically to C with a simple pole at s = −n, n ∈ N,
with residue (−1)n /n.

Theorem 3.1.11 (Riemann). The function ζ(s) extends to a function on C, holomorphic

s
except for a single pole at s = 1. If Λ(s) := π − 2 Γ( 2s )ζ(s) then away from 0 and 1 the
function Λ(s) is bounded in any vertical strip and satisfies Λ(1 − s) = Λ(s).
In particular ζ(s) does not vanish for Re(s) > 1. In Re(s) < 0 it has simple zeroes at
−2, −4, −6, etc... All the other zeroes are inside the “critical strip” 0 ≤ Re(s) ≤ 1.
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 13

Proof. For Re(s) > 1:

s XZ ∞ dt
− 2s s s
Λ(s) = π Γ( )ζ(s) = e−t t 2 π − 2 n−s
2 t
n≥1 0
 
Z ∞ X Z ∞ s
−πun2  2s du u 2 du
=  e u = θ̃(u)
0 u 0 u
n≥1

2
X
where we made the change of variables t = πun2 and defined θ(u) := e−πun and
n∈Z
X
−πun2 θ(u) − 1
θ̃(u) := e = .
2
n≥1
Recall that the Fourier transform of a real integrable function f is fˆ(y) =
R
R f (x) exp(2πixy)dx
and that the Poisson formula states the equality:
X X
f (n) = fˆ(n) .
n∈Z n∈Z

πy 2
−πux2 e− u
Considering f (x) = e one obtains fˆ(y) = √ and the Poisson formula reads:
u
1 √
(1) θ( ) = uθ(u)
u
(in other words: the theta function θ is a modular form of half integer weight). Equa-
tion (1) implies
1 √ 1 √
θ̃( ) = uθ̃(u) + ( u − 1) .
u 2
R∞
Since 1 t−s dt = s−11
one obtains:
Z 1 s Z ∞ s
u 2 du u 2 du
Λ(s) = θ̃(u) + θ̃(u)
0 u 1 u
Z ∞ − 2s Z ∞ s
1 u du u 2 du
= θ̃( ) + θ̃(u)
1 u u 1 u
Z ∞  −s Z ∞ s
√ 1 √ u 2 du u 2 du
= uθ̃(u) + ( u − 1) + θ̃(u)
2 u u
Z1 ∞  du
1
 s 1−s 1 1
= θ̃(u) · u 2 + u 2 + − .
1 u s−1 s
The right hand side integral is a priori defined only for Re(s) > 1 but using θ̃(u) =
O(e−πu ) one easily checks it is entire. Moreover it is clearly bounded in every vertical
strip. Finally the right hand side is symmetric with respect to s 7→ 1 − s.

Theorem 3.1.12 (Hadamard- De La Vallée-Poussin, 1896). ζ(s) does not vanish on
Re(s) = 1.
14 BRUNO KLINGLER

X p−ms
Proof. log ζ(s) = hence
m
m≥1,p
X p−mu
Re (3 log ζ(u) + 4 log ζ(u + iv) + log ζ(u + 2iv)) = Re(3 + 4p−miv + p−2miv ) .
p,m
m

Using that 3 + 4 cos t + cos(2t) = 2(1 + cos t)2 ≥ 0 one obtains:

3 4
ζ(u) |ζ(u + iv)| |ζ(u + 2iv)| ≥ 1 .
The left hand side is equivalent to c(u − 1)k+4h−3 as u → 1, where c denotes a positive
constant and h, k denote the order of ζ(x) at s = u + iv and u + 2iuv respectively. Hence
k + 4h − 3 ≤ 0 hence h = 0 as h, k ≥ 0. 
This result is enough to show the prime number theorem (cf. [El75, chap.2]):
Theorem 3.1.13 (Hadamard- De La Vallée-Poussin, 1896). Define π(x) as the number
of primes smaller than x. Then
x
π(x) ∼ .
log x
Theorem 3.1.11 enables to state the famous
Conjecture 3.1.14 (Riemann hypothesis). All the zeroes of ζ(s) inside the critical strip
lie on the line Re(s) = 21 .
We refer to [El75] for a nice survey on the relation between the Riemann hypothesis
and the distribution of prime numbers.
3.2. Zeta functions for schemes of finite type over Z.
3.2.1. Arithmetic zeta function for schemes of finite type over Z. The definition of
ζ(Spec Z, s) := ζ(s) as an Euler product generalizes naturally to any scheme X of finite
type over Z:
Definition 3.2.1. Let X be a scheme of finite type over Z. One defines
Y 1
ζ(X, s) = .
1 − N (x)−s
x∈X(0)

Remarks 3.2.2. (a) By Proposition 2.3.2 there are only finitely many points x ∈ X(0)
of given norm. Hence Corollary 3.1.7 implies that ζ(X, s) is a formal Dirichlet
series.
(b) Notice that ζ(X, s) depends only on X(0) .
Y 1
Developping the product one obtains as in the case of the zeta
1 − N (x)−s
x∈X(0)
function:
Lemma 3.2.3. Let X be a scheme of finite type over Z. Then
X 1
ζ(X, s) = s
+
N (c)
c∈Z0 (X)

in Dir(C), where Zr (X)+ denotes the monoı̈d of effective r-cycles of X.

 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 15

Moreover one immediately obtains from the definition:

`
Lemma 3.2.4. Let X be a scheme of finite type over Z. If X satisfies X(0) = (Xi )(0)
for a family (Xi ) of subschemes of X then
∞
Y
ζ(X, s) = ζ(Xi , s) .
i=1
In particular Y
ζ(X, s) = ζ(Xp , s) ,
p
where Xp is the fiber of X → Spec Z over p.
3.3. Geometric zeta function for a scheme of finite type over Fq . If X is a
scheme of finite type over Fq , one can introduce a more natural generating series encoding
the points of X: its geometric zeta function.
Definition 3.3.1. Let X be a scheme of finite type over Fq . Its geometric zeta function
is defined as:
∞
!
X tn X
Z(X/Fq , t) := exp X(Fqn )Fq = tdeg c .
n
n=1 c∈Z0 (X)+
P P
Here the degree of a zero-cycle c = i ni · xi ∈ Z0 (X) is defined as deg c = i ni [k(x) :
Fq ].
Remark
P 3.3.2. Let X be a scheme P of finite type over Fq . Notice that the degree
i ni [k(x) : Fq ] of a zero-cycle c = i ni · xi ∈ Z0 (X) depends on the base field Fq , while
N (c) does not. As a corollary the geometric zeta function Z(X/Fq , t) really depends on
the base field: if X is defined over Fqr then Z(X/Fq , t) = Z(X/Fqr , tr ). On the other
hand ζ(X, s) is an absolute notion.
The following obvious formula is the basis for any calculation with the geometric zeta
function:
Lemma 3.3.3. Let X be a scheme of finite type over Fq . Then
∞
d X
t· log Z(X/Fq , t) = X(Fqn )Fq tn .
dt
n=1
The relation between ζ(X, s) and Z(X/Fq , t) is given by the following:
Lemma 3.3.4. Let X be a scheme of finite type over Fq . Then ζ(X, s) = Z(X/Fq , q −s ).
Proof.
∞
X X X N (x)−ms
log ζ(X, s) = − log(1 − N (x)−s ) =
m
x∈X(0) x∈X(0) m=1
 
X∞ X N (x)−ms X∞ X q −m deg(x)s X ∞  X  q −ns
= = = deg(x)
 
m m  n

m=1 m=1 n=1

x∈X(0) x∈X(0) x∈X(0)
deg(x)|n
16 BRUNO KLINGLER
X
By Lemma 2.4.1 |X(Fqn )Fq | = deg(x) hence the result.
x∈X(0)
deg(x)|n

3.4. Properties of zeta functions.
Lemma 3.4.1. Let X be a scheme of finite type over Fq . Then ζ(A1X , s) = ζ(X, s − 1).
Proof. Applying Lemma 3.2.4 one obtains
Y
ζ(A1X , s) = ζ(A1x , s) .
x∈X(0)

Hence we are reduced to show that ζ(A1x , s) = ζ(x, s − 1). Applying Lemma 3.3.4 and
writing k(x) = Fq one gets
∞
!
1
X
1 q −ns
ζ(Ax , s) = exp |AFq (Fqn )Fq |
n
n=1
∞
!
X q −ns 1
= exp qn · =
n 1 − q 1−s
n=1
= ζ(x, s − 1) .

Theorem 3.4.2. Let X be a scheme of finite type over Z. Then ζ(X, s) converges for
s > dim X.
Proof. Step 1: One can assume that X is irreducible. Indeed suppose that X = X1 ∪ X2 ,
Xi ⊂ X closed subscheme, i = 1, 2. It follows from proposition 3.2.4 that
ζ(X1 , s) · ζ(X2 , s)
ζ(X, s) = ,
ζ(X1 ∩ X2 , s)

where X1 ∩ X2 denotes the schematic intersection X1 ×X X2  / / X . Hence the state-
ment for X1 and X2 implies the statement for X by induction on the dimension.

Step 2. One can assume that X is affine (and integral). Indeed if Z  / / X o O ? _ U
one similarly obtains
ζ(X, s) = ζ(U, s) · ζ(Z, s) .
Hence the statements for U and X are equivalent by induction on the dimension.
Step 3. If f : X → Y is a finite morphism and if the statement holds for Y then it
holds for X. Indeed it follows from Lemma 3.2.4 that:
Y
ζ(X, s) = ζ(Xy , s) .
y∈Y(0)

Let d be the degree of f , the fiber Xy has at most d closed points. If x ∈ (Xy )(0) is such
a point then N (x) is a power of N (y) hence for Re(s) > 0:
1 1
−s
≤
1 − N (x) 1 − N (y)−s
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 17

This implies:
|ζ(Xy , s)| ≤ |ζ(y, s)|d .
It follows that |ζ(X, s)| ≤ |ζ(Y, s)|d and the result.
Step 4. We can assume that X = AdZ or X = AdFp . Indeed let X → Spec Z be affine
and integral. Recall the
Lemma 3.4.3. (Noether normalization lemma) For any field k and any finitely gen-
erated commutative k-algebra A, there exists a nonnegative integer d and algebraically
independent elements x1 , · · · , xd in A such that A is a finitely generated module over the
polynomial ring k[x1 , · · · , xd ]
Equivalently: every affine k-scheme of finite type is finite over an affine d-dimensional
space.
- if X → Spec Z is dominant, it follows from Noether normalisation lemma applied to
XQ → Spec Q that there exists a finite flat morphism XQ → AdQ . It extends to a finite
flat morphism f : XU → AdU for some open subset U ⊂ Spec Z. We can assume that
X = XU by step 2, then X = AdU by step 3, then X = AdZ by step 2 again.
- otherwise there exists some prime p so that X is of finite type over Fp . Applying
Noether normalization lemma to X → Spec Fp we are reduced to X = AdFp .
If X = AdZ then Lemma 3.4.1 gives ζ(AdZ , s) = ζ(s − d), which converges absolutely
for Re(s) > d + 1 = dim X.
1
If X = AdFp then ζ(AdFp , s) = , which converges absolutely for Re(s) > d =
1 − pd−s
dim X.


3.5. Some examples.

Example 3.5.1. X = AnFq . We computed AnFq (Fqr )Fq = q nr hence

∞
!
X t m 1
Z(AnFq , s) = exp q mn = .
m 1 − qnt
m=1

Example 3.5.2. X = PnFq . We computed PnFq (Fqr )Fq = 1 + q r + · · · + q nr hence

∞
!
X tm 1
Z(PnFq , s) = exp (1 + q m + · · · + q nm ) = .
m (1 − t) · (1 − qt) · · · (1 − q n t)
m=1

3.6. Some questions and conjectures.

Question 3.6.1. Let X be a scheme of finite type over Z. Suppose that we know ζ(X, s).
What can we say of X?
- If X = Spec OK is the ring of integers of a number field K then ζ(OK , s) = ζK (s) is
the Dedekind zeta function of K. One shows that
hK · RK
ords=0 ζK (s) = r1 + r2 − 1 =: r and lim s−r ζK (s) = − ,
s7→0 ωK
18 BRUNO KLINGLER

where r1 is the number of real places σ1 , · · · , σr1 of K, r2 is the number of complex

places σ10 , · · · , σr0 2 not conjugate two by two, hK = |Pic OK | is the class number of K,
ωK is the number of roots of unity of K and RK is its regulator i.e. the covolume of the
lattice OK∗ /torsion in Rr under the regulator map

∗ / Rr = { r+1 xi = 0} ⊂ Rr1 +r2

P
reg : OK i=1

/ log σ1 (u), . . . , log σr (u), 2 log σ 0 (u), . . . , 2 log σ 0 (u)

u 1 1 r2 .
A theorem of Mihaly Bauer (1903) says that if K, L are two number fields which are
Galois over Q then K ' L is equivalent to ζK = ζL . On the other hand Gassmann
(1936) showed that there do exist non-isomorphic number fields K, L (in fact hK 6= hL )
with ζK = ζL . The example of smallest degree occur in degree 7 over Q.
- If X is a smooth projective curve over a finite field Fq , the curve X is not determined
by its zeta function. However Tate (1966) and Turner (1978) proved that two curves
X, Y over Fq satisfy ζ(X, s) = ζ(Y, s) if and only if their Jacobians are isogeneous.
Conjecture 3.6.2. Let X be a scheme of finite type over Z. The function ζ(X, s)
extends meromorphically to all C and satisfies a functional equation with respect to s 7→
dim X − s.
This is proved for dX = 1, for some very particular cases for dX > 1 when X is flat
over Z and for all dX when X is a scheme of finite type over Fp (the so called positive
characteristic case).
It follows from the Weil conjectures that ζ(X, s) always has a meromorphic continu-
ation to Re(s) > dim X − 1/2.
3.7. Weil conjectures. We now concentrate on the positive characteristic case.
Definition 3.7.1. A q-Weil polynomial (resp. pure of weight m ∈ N) is a polynomial
r
Y
P := (1 − γi t) ∈ Z[t]
i=1
where the γi ’s are q-Weil numbers (resp. of same weight m).
The Weil conjectures can be stated as follows:
Conjecture 3.7.2 (Weil). Let X be a scheme of finite type over Fq of dimension d.
1. (Rationality) Z(X/Fq , t) ∈ Q(t).
2. (Functional equation) If X is smooth and proper of pure dimension d, let χ be
the self-intersection of the diagonal in X × X. Then
1 d·χ
Z(X/Fq , d ) = ±q 2 tχ Z(X, t) .
q t
3. (Purity) If X has dimension d then
2d
i+1
Y
Z(X/Fq , t) = Pi (t)(−1)
i=0
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 19

where the Pi ’s are q-Weil polynomials.

If moreover X is smooth and proper of pure dimension d then the Pi ’s are pure
of weight i and
1
P2d−i (t) = Ci tdeg Pi Pi ( d ) with Ci ∈ Z .
q t
P2d
4. (link with topology) χ = i=0 (−1)i deg Pi . If moreover X/Fq is the smooth and
proper special fiber of a smooth and proper X /R, R finitely generated Z-algebra,
Fq  R ,→ C, then
deg Pi = bi ((XC )an ) .
In particular χ coı̈ncides with the Euler characteristic χ(XC )an .
Remark 3.7.3. The rationality of Z(X/Fq , s) is already a highly non-trivial result. It
implies in particular that if we know |X(Fqr )| for sufficiently many (depending on X)
values of r ∈ N then we know |X(Fqr )| for all r ∈ N.
Remark 3.7.4. To prove rationality of zeta functions (Conjecture 3.7.2(1)), it is enough
to prove it for X an irreducible hypersurface in AnFq . Indeed arguing as in Lemma 3.2.4
and by induction on dimension we can assume that X is irreducible and affine. But then
(using generic projections) X is birational over Fq with a hypersurface in an affine space
and we are done by induction on dimension.

4. The Weil conjectures for curves

In this section we prove the Weil conjectures for a smooth projective, geometrically
irreducible, curve C over a finite field Fq . Recall first that the fundamental invariant of
the curve C is its genus g = h0 (C, ωC ) where ωC = Ω1C/Fq is the canonical line bundle of
C. Second, it follows from [SGA1] (as we will see later) that the curve C is always the
special fiber of some smooth projective curve C over a finitely generated Z-algebra R,
Fq  R ,→ C [SGA1], so that we are in the situation of Conjecture 3.7.2.4. Classically
the smooth projective complex curve CC satisfies b0 (CC ) = b2 (CC ) = 1 and b1 (CC ) = 2g,
hence χ(CC ) = 2 − 2g. Hence the Weil conjectures for C are the following:
Theorem 4.0.1 (E.Artin, Schmidt, Hasse, Weil). Let C be a geometrically irreducible
smooth projective curve of genus g over Fq . Then:
P (t)
Z(C/Fq , t) =
(1 − t)(1 − qt)
where P (t) = 2g
Q
i=1 (1 − αi t) ∈ Z[t] is a polynomial
√
of degree 2g and constant term 1,
with inverse roots αi of absolute value |αi | = q for any embedding of Q in C. Moreover
it satisfies the functional equation:
1
Z(C/Fq , ) = q 1−g t2−2g Z(C/Fq , t) .
qt
Corollary 4.0.2 (Riemann hypothesis for curves over finite fields). Let C be a geometri-
cally irreducible smooth projective curve of genus g over Fq . Then all the roots of ζ(C, s)
lie on the line Re(s) = 1/2.
20 BRUNO KLINGLER

Proof. As ζ(C, s) = Z(C/Fq , q −s ), the roots of ζ(C, s) are the roots of P (q −s ) = 2g

Q
√ i=1 (1−
αi q −s ). The purity condition |γi | = q for any embedding of Q in C is thus equivalent
to saying that these roots lie on the line Re(s) = 1/2. 

Let us give some short historical comments (we refer to [Aud12] for more details). In
his thesis E.Artin (1921) defined the zeta function of a quadratic extension of Fq ((t)) and
proved its rationality (in p−s ). F.K. Schmidt [Sch31] generalized Artin’s definition to
any function field over Fq in one variable. He deduced the rationality and the functional
equation for Z(C/Fq , t) from his proof of the Riemann-Roch theorem for C. In [Ha36]
H.Hasse proved the Riemann Hypothesis for elliptic curves over finite field. A.Weil
[We40] announced the proof of the Riemann Hypothesis for curves over finite fields
and gave a complete proof eight years later after a complete refoundation of algebraic
geometry.
We now indicate the general strategy for the proof of the Weil conjectures for curves
(we refer to [Sil09, Chap.5] for an elementary proof in the case of elliptic curves). While
it is difficult to understand 0-cycles on a general scheme, a zero cycle on the curve C is
nothing else than a divisor. Counting points on C is thus equivalent to counting sections
of line bundles on C. The Riemann-Roch formula provides a complete answer to this
problem for line bundles of degree big enough. The rationality of Z(C/Fq , t) follows
immediately. The functional equation for Z(C/Fq , t) is then a shadow of Serre duality
for the cohomology of line bundles on curves. As is the case in higher dimension, the
most delicate part of Theorem 4.0.1 is purity, equivalently the Riemann Hypothesis. It
is easily seen to be equivalent to proving the bounds
√
(2) C(Fqn )Fq | − q n − 1 ≤ 2g q n .
For proving these bounds, we introduce one of the main player of this entire course: the
geometric Frobenius FrX,q , a canonically defined endomorphism of any scheme X over
Fq . The set C(Fqn )Fq can be interpreted as the intersection in (C × C)Fq of the graph
of FrnC,q with the diagonal ∆. The bounds eq. (2) then follow from the Hodge index
theorem on the surface (C × C)Fq .

4.1. Heuristics.

4.2. The Riemann-Roch’s formula. Let k be a field and C be a smooth projective

curve over k. We denote by C its base change to an algebraic closure k of k. If π : C → C
is the natural projection then H i (C, F) ⊗k k ' H i (C, π ∗ F) for any quasicoherent OC -
module F. We will assume that C is geometrically irreducible, i.e. C is irreducible.
The group Z0 (C) coincide with the group of Weil divisors Z 1 (C) on C. As C is smooth
(in particular integral, separated and locally factorial) the group of Weil divisor coincide
with the group H 0 (C, FC∗ /OC ∗ ) of Cartier divisors on C. Here F denotes the function

field of C and FC the associated constant sheaf on C (as C is integral it coı̈ncides with
the sheaf of rational functions on C). Moreover principal Weil divisors and principal
Cartier divisors do coincide.
To any divisor D, seen as a Cartier divisor (Ui , fi ), we associate the line bundle
O(D) ⊂ FC on C generated as an OC -module by fi−1 on Ui . We denote by h0 (O(D))
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 21

the k-dimension of its space of global sections

H 0 (C, O(D)) = {f ∈ F ∗ | D + (f ) ∈ Z0 (C)+ } .
This defines an isomorphism between the group Z0 (C)/ ∼, where two divisors are ra-
tionally equivalent if their difference is principal, with the group Pic (C) of isomorphism
classes of line bundles on C. The degree morphism deg : Z0 (C) → Z descends to
deg : Pic (C) → Z. Moreover one has a short exact sequence:
deg
0 → Pic 0 (C) → Pic (C) → Z .
A priori the degree map is not surjective:
Definition 4.2.1. We denote by δ > 0 the index of the curve C: the unique positive
integer such that
deg(Pic (C)) = δZ .
Remark 4.2.2. Notice that δ|2g − 2 = deg ωC . For curves over Fq we will show that
δ = 1.
Given a line bundle L on C the set of effective divisors D on C with O(D) ' L is in
∗ ) via multiplication.
bijection with the quotient H 0 (C, L) \ 0 by the action of H 0 (C, OC
As C is irreducible and projective we obtain H 0 (C, OC ) = k hence H 0 (C, OC ) = k.
The Riemann-Roch formula states that for any line bundle L on C one has:
(3) h0 (L) − h0 (ωC ⊗ L−1 ) = deg(L) + 1 − g .
As a corollary:
(4) If deg(L) > 2g − 2 then h0 (L) = deg(L) + 1 − g .
4.3. Rationality.
Proposition 4.3.1. Let C be a smooth projective, geometrically irreducible, curve over
Fq . Then Z(C/Fq , t) is a rational function.
Proof. X
Z(C, t) = tdeg(D) .
D∈Z0 (C)+
It follows from our discussion of the relation between line bundles and effective divisors
that:

X X q h0 (L) − 1
Z(C/Fq , t) = PH 0 (C, L) · tdeg L = · tdeg L
q−1
L∈Pic C L∈Pic C
deg L≥0 deg L≥0
0 (L) 0 (L)
X qh − 1 deg L X qh − 1 deg L
= ·t + ·t
q−1 q−1
0≤deg L≤2g−2 2g−2<deg L
X q h0 (L)
− 1 deg L X q deg L+1−g − 1 deg L
= ·t + ·t .
q−1 q−1
0≤deg L≤2g−2 2g−2<deg L

Lemma 4.3.2. The group Pic 0 (C) is finite.

22 BRUNO KLINGLER

Proof. Fix n > 2g a multiple of δ. Then any divisor D of degree n satisfies h0 (O(D)) =
n + 1 − g > 0 hence is effective. Thus the group Pic 0 (C) has a (free) orbit in Pic (C)
consisting precisely of the rational equivalence classes of effective divisors of degree n.
As we already saw (see the discussion after Definition 2.2.4) that the number of effective
divisors of degree n on C is finite the result follows. 
Remark 4.3.3. Of course the “correct proof” is as follows: for any k-variety X there
exists a k-variety Pic 0 X whose set of k 0 -points is the group Pic 0 (X ×k k 0 ) for any field
extension k 0 of k. In our case Pic 0 (C) = (Pic 0 C)(Fq ) hence is necessarily finite.
Our computation of Z(C/Fq , s) continues as:
0 (L)
X qh − 1 deg L X q nδ+1−g − 1 nδ
Z(C/Fq , t) = ·t + |Pic 0 (C)| · ·t .
q−1 q−1
0≤deg L≤2g−2 ρ:= 2g−2 <n
δ

Notice that the first term

0 (L)
X qh − 1 deg L
f1 (t) := ·t
q−1
0≤deg L≤2g−2

is a polynomial in tδ of degree at most ρ = (2g − 2)/δ. On the other hand one computes
the second term
(5) !
X q nδ+1−g − 1 |Pic 0 (C)| (qt)δ(ρ+1) tδ(ρ+1)
0 nδ 1−g
f2 (t) := |Pic (C)| · ·t = · q · − .
ρ<n
q−1 q−1 1 − (qt)δ 1 − tδ
One concludes that one might write
P (tδ )
(6) Z(C/Fq , t) = ,
(1 − tδ )(1 − (qt)δ )
where P is a polynomial with rational coefficients, of degree less than ρ + 2.
Since Z(C/Fq , t) has integer coefficients one obtains that P has integer coefficients as
well.
This shows that Z(C/Fq , t) is a rational function.

Proposition 4.3.4. The curve C has index δ = 1: it admits a divisor of degree 1.
Proof. Looking at the expression 5 for f2 one obtains:
|Pic 0 (C)| t−1 |Pic 0 (C)|
lim(t − 1)Z(C/Fq , t) = − · lim = .
t→1 q−1 t→1 1 − tδ δ(q − 1)
In particular Z(C/Fq , t) has a pole of order one at t = 1.
Lemma 4.3.5. Let X be a variety over Fq . Then
r
Y
Z(X ×Fq Fqr /Fqr , tr ) = Z(X/Fq , ξ i t) ,
i=1
where ξ denotes a primitive root of order r of 1.
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 23

Proof.
∞
X tmr
log Z((X ×Fq Fqr )/Fqr , tr ) = (X ×Fq Fqr )(Fqmr )Fqr ·
m
m=1
∞
X tmr
= X(Fqmr )Fq · as (X ×Fq Fqr )(Fqmr )Fqr = X(Fqmr )Fq
m
m=1
∞ r r
X X tl X
= |X(Fql )Fq | · ( ξ il ) · as ξ il = δr,l · r
l
l=1 i=1 i=1
∞
r X
X (ξ i t)l
= |X(Fql )Fq | ·
l
i=1 l=1
Xr
= log Z(X/Fq , ξ i t) .
i=1

It follows from Lemma 4.3.5 and the Formula eq. (6) that Z((C ×Fq Fqδ )/Fqδ , tδ ) =
Z(C/Fq , t)δ . On the other hand we can apply our results so far to C ×Fq Fqδ : the function
Z((C ×Fq Fqδ )/Fqδ , t) has a pole of order one at 1, hence also Z((C ×Fq Fqδ )/Fqδ , tδ ).
Thus δ = 1. This finishes the proof of Proposition 4.3.4. 
Remark 4.3.6. Even if C admits a divisor of degree 1 it does not necessarily admits an
Fq -point. Consider for example the genus 2 curve on F3 with affine equation
y 2 = −(x3 − x)2 − 1 .
This curve does not have any F3 -point. However if y1 and y2 are the two roots of
y 2 = −1 the divisor D1 := (0, y1 ) + (0, y2 ) is defined over F3 . Similarly the divisor
D2 := (x1 , 1) + (x2 , 1) + (x3 , 1) is defined over F3 , where xi , 1 ≤ i ≤ 3, are the roots of
x3 − x = −1. Then D := D2 − D1 is a divisor of degree 1 defined over F3 .
Corollary 4.3.7.
P (t)
Z(C/Fq , t) = ,
(1 − t)(1 − qt)
where P ∈ Z[t] is a polynomial of degree at most 2g and constant term 1.
4.4. Functional equation.
Proposition 4.4.1.
1
Z(C/Fq , ) = q 1−g t2−2g Z(C/Fq , t) .
qt
Proof. We come back to our expression Z(C/Fq , t) obtained in the proof of Proposi-
tion 4.3.1. Rearranging this expression a little bit we write Z(C/Fq , t) = g1 (t) + g2 (t)
with (as δ = 1):
0
q h (L) deg L |Pic 0 (C)| (qt)2g−1
 
X 1
g1 (t) = ·t and g2 (t) = · q 1−g · −
q−1 q−1 1 − qt 1−t
0≤deg L≤2g−2
24 BRUNO KLINGLER

1
A direct computation shows that g2 ( qt ) = q 1−g t2−2g f2 (t).
To deal with g1 (t) notice that
L 7→ ωC ⊗ L−1
defines an involution on the set of line bundles on C of degree in [0, 2g − 2]. Hence:
 
2g−2 X q h0 (L)  i
1 X
· 1
g1 ( ) = 
qt q−1 qt
i=0 L∈Pic i (C)
 
2g−2 X q h0 (ωC ⊗L−1 )  2g−2−i
· 1
X
= 
q−1 qt
i=0 L∈Pic i (C)
 
2g−2
X X q h0 (L)
=   · (qt)i+2−2g by the Riemann-Roch’s formula
q − 1
i=0 i
L∈Pic (C)
1−g 2−2g
=q t g1 (t) .

Remark 4.4.2. Hidden in this proof is Serre duality: we identified h1 (L) with h0 (ωC ⊗
L−1 ) in the Riemann-Roch’s formula.
Corollary 4.4.3. The polynomial P is of degree exactly 2g.
Proof. This follows immediately from the functional equation. 

4.5. The geometric Frobenius. In this section we introduce the Frobenius endo-
morphism, whose role will be crucial in the proof of the Riemann Hypothesis for curves
(the most delicate part of Theorem 4.0.1) and for this course in general.
Let X be a scheme of finite type over Fq . Then one has the equality:
n
X(Fqn )Fq = (X(Fq )Fq )Frq .
There are however two conceptually different interpretations of the action of Frq on
X(Fq )Fq .
(1) We already presented the first one. Consider Frq as a topological generator of
Gal(Fq /Fq ) ' Ẑ. The action we are looking for is a special case of the natural ac-
tion of Gal(k 0 /k) over X(k 0 )k = HomSpec k (Spec k 0 , X) via its natural action on Spec k 0
over k.
(2) On the other hand we can define a Frobenius endomorphism
FrX,q : X → X
as the morphism of local ringed spaces (1X , Fr]X,q ) where Fr]X,q : OX → OX maps
f ∈ OX (U ) to f q ∈ OX (U ). If X = Spec A is affine, with A a finitely generated Fq -
algebra, then FrX,q is just given by the algebra homomorphism A → A associating f q
to f ∈ A. The existence of this Frobenius endomorphism is what makes geometry over
finite fields very different from geometry over any other field.
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 25

Now FrX,q acts on X(Fq )Fq as a particular case of the action of any endomorphism of
X on HomSpec k (Spec k 0 , X).
The morphism FrX,q : X → X induces a morphism
FrX,q = FrX,q × id : X → X .

Lemma 4.5.1. The actions of Frq on X(Fq )Fq and FrX,q on X(Fq )Fq = X(Fq )Fq do
coincide.
In other words if ∆ denotes the diagonal of S = X × X and Γr the graph of FrrX,q
then X(Fqr )Fq is in natural bijection with the closed points of Γr ∩ ∆.
4.6. The Riemann hypothesis for curves over finite fields.
Proposition 4.6.1. Write P (t) = 2g
Q
√ i=1 (1 − αi t). Then every αi is an algebraic integer
and |αi | = q for any embedding of Q in C.
Proof. First notice that the functional equation implies:
2g 2g
Y αi Y
(t − ) = q −g (1 − αi t) .
q
i=1 i=1
Q2g
As a consequence i=1 αi = qg and the multiset {α1 , · · · , α2g } is invariant under the
q
map x 7→ x .
√
Hence it is enough to prove that |αi | ≤ q for all i, 1 ≤ i ≤ 2g: by the symmetry
√
above one gets |αi | ≥ q for all i, and the result.
For n ∈ N∗ let us define an := 1 + q n − |C(Fqn )Fq |. Derivating the logarithm of the
equality P (t) = Z(C/Fq , t)(1 − t)(1 − qt) and multiplying by t one obtains:
n
d log P (t) X
(7) t· =− an tn .
dt
i=1
P2g
Hence an = n
i=1 αi for every n ∈ N∗ .
One can then rephrase the Riemann Hypothesis
for curves as an estimate for the an ’s:
√ √
Lemma 4.6.2. One has |αi | ≤ q for all i, 1 ≤ i ≤ 2g, if and only if |an | ≤ 2g q n for
every n ≥ 1.
Proof. As an = ni=1 αin one implication is trivial. For the converse: if |an | ≤ 2gq n/2
P

then the series eq. (7) converges for |t| < q −1/2 . Hence P (t) has no zeroes in this domain.
By the functional equation P (t) has no zeroes in |t| > q −1/2 . Finally all the zeroes of
P (t) have absolute value q −1/2 . 
√ n
Hence we are reduced to prove that |an | ≤ 2g q for every n ≥ 1. Notice it is enough
to show that
√
|a1 | ≤ 2g q .
Indeed applying this result to C ×Fq Fqn yields the required inequality for |an |.
We will use intersection theory on surfaces. Consider the smooth projective surface
S := C × C over Fq . We know that
N1 := |C(Fq )| = (Γ · ∆) .
26 BRUNO KLINGLER

Recall the Hodge index theorem for a smooth projective surface over an algebraically
closed field (see for example [Har77, Chap V.1]): if E is a divisor on S such that (E ·H) =
0 for H ample then (E 2 ) < 0. Fix D any divisor on S and apply the Hodge index theorem
to H = L1 +L2 (where L1 = C ×pt and L2 = pt×C) and E = D−(D·L2 )L1 −(D·L1 )L2 :
one obtains
(8) (D2 ) < 2(D · L1 )(D · L2 ) .
Let us compute the different intersection numbers. Notice that (∆ · L1 ) = (∆ · L2 ) = 1
and (Γ · L1 ) = q while (Γ · L2 ) = 1. We still have to compute (∆2 ) and (Γ2 ). As Γ and
∆ are smooth curves of genus g one can apply the adjunction formula KY = (KS + Y )|Y
for a smooth divisor Y , noting that KS = (2g − 2)(L1 + L2 ):
2
2g − 2 = (K∆ ) = (∆ · (∆ + KS )) = (∆2 ) + 2(2g − 2),
2g − 2 = (KΓ2 ) = (Γ · (Γ + KS )) = (Γ2 ) + (q + 1)(2g − 2).
Therefore (∆2 ) = −(2g − 2) and (Γ2 ) = −q(2g − 2).
We apply eq. (8) to D = a∆ + bΓ, a, b ∈ Z:
−a2 (2g − 2) − qb2 (2g − 2) + 2abN1 ≤ 2(a + bq)(a + b) .
Hence:
ga2 − ab(q + 1 − N1 ) + gqb2 ≥ 0 .
This holds for all a, b ∈ Z hence
(q + 1 − N1 )2 ≤ 4qg 2
and the result.

One deduces immediately from the estimates on a1 :
Corollary 4.6.3. Let C be a smooth projective, geometrically irreducible, curve of genus
g over Fq . Then
√ √
1 + q − 2g q ≤ |C(Fq )Fq | ≤ 1 + q + 2g q .
In particular C admits a Fq -point as soon as q ≥ 4g 2 .

5. Transition to étale cohomology

5.1. Heuristic for the Weil conjectures: about the Lefschetz trace formula.
This section is borrowed from lectures of Beilinson on the Weil conjectures [Bei07].
How can one guess the Weil conjectures, for example the rationality of the zeta function
∞
X tn
Z(X, t) = exp( X(Fnq )Fq · )
n
i=1

for a scheme X of finite type over Fq ?

Frn
As X(Fnq )Fq = X(Fq )F X,q , a more general question is the following: let φ(= FrX,q ) be
q
n =1
an automorphism of a set S(= X(Fq )Fq ) such that for all n ∈ N∗ , the set S φ is finite.
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 27

Can we compute |S φ=1 | and more generally

∞
X n =1 tn
Z((S, φ), t)M = exp |S φ |· ?
n
i=1

5.1.1. The finite case. Suppose for simplicity that the set S is finite. Notice that even
in this case the rationality of Z((S, f ), t) is not a priori obvious. Let Q[S] be the Q-vector
space generated by S. The automorphism φ of S induces a linear action of φ on Q[S].
Lemma 5.1.1.
(9) Z((S, φ), t) = det(1 − t · φ | Q[S])−1 .
Proof. Denote by (αi )i∈I the eigenvalues of φ acting on Q[S]. Hence
Y
det(1 − t · φ|Q[S]) = (1 − αi t) .
i∈I
d log
Applying t · dt to both sides of eq. (9) one obtains:
n
X XX
|S φ =1 | · tn = (αi t)n
n≥1 i∈I n≥1
X
= tr(φn | Q[S]) · tn .
n≥1
n
Hence we are reduced to proving that tr(φn | Q[S]) = |S φ |, which is obvious for any
permutation φ of the finite set S. 
How can we generalize this kind of arguments for S infinite?
5.1.2. The differentiable case. Suppose now that S is a closed C ∞ manifold and φ : S →
n
S is a diffeomorphism such that for all n ∈ N∗ , the set S φ =1 is finite. For simplicity we
will assume:
(1) the manifold S is orientable.
(2) for any fixed point s ∈ S of φn one has det(1 − φn | Ts S) > 0.
Remark 5.1.2. The assumption det(1 − φn | Ts S) 6= 0 means that the point s ∈ S is
non-degenerate for φn , i.e. that the diagonal ∆S and the graph Γ(φn ) are transverse at
s ∈ S. The condition det(1 − φn | Ts S) > 0 means moreover that the local index of f at
s is positive.
In this situation, Lefschetz [Lef26] proved:
Theorem 5.1.3. (Lefschetz trace formula)
dim
XS
n =1
(10) |S φ |= (−1)i tr(φn | H i (S, Q)).
i=0
Qdim S i+1
Equivalently: Z((S, φ), t) = i=0 det(1 − t · φ∗ | H i (S, Q))(−1) .
Remark 5.1.4. If S is finite we have H 0 (X, Q) = Q[S]∗ and the higher cohomologies
vanish, hence we recover Lemma 5.1.1.
28 BRUNO KLINGLER

Weil’s main idea is that the methods from algebraic topology should be applicable in
characteristic p > 0: if one has a “sufficiently nice” cohomology theory for schemes over
Fq then Z(X, t) can be computed through the Lefschetz trace formula for FrX,q acting
on the (compactly supported) cohomology of X and a good part of the Weil conjectures
is “formal”.

5.2. Weil cohomologies. We fix a base field k, and a coefficient field K of character-
istic zero. We define axiomatically what a “nice” cohomology theory with coefficients in
K should be, at least on the category SmProj(k) of smooth projective k-schemes (with
k-morphisms). Let VectZK be the category of graded K-vector spaces of finite dimension,
with its graded tensor product.
Definition 5.2.1. A pure Weil cohomology on k with coefficients in a field K of char-
acteristic zero is a functor:
H • : SmProj(k)op → VectZK
satisfying the following axioms:
(i) Dimension: For any X ∈ SmProj(k) of dimension dX , H i (X) = 0 for i 6∈
[0, 2dX ].
(ii) Orientability: dimK H 2 (P1k ) = 1; we denote this space by K(−1).
(iii) Additivity: For any X, Y ∈ SmProj(k) the canonical morphism
a
H • (X Y ) → H • (X) ⊕ H • (Y )
is an isomorphism.
(iv) Künneth formula: For any X, Y ∈ SmProj(k) one has an isomorphism
∼
κX,Y : H • (X) ⊗K H • (Y ) → H • (X ×k Y )
natural in X, Y , satifying obvious compatibilities. In particular we require H • (Spec k) =
K in degree 0 and H • is monoidal.
(v) Trace and Poincaré duality: For any X ∈ SmProj(k), purely of dimension dX ,
one has a canonical morphism
TrX : H 2dX (X) → K(−dX ) := K(−1)⊗dX
which is an isomorphism if X is geometrically connected, and such that TrX×k Y =
TrX ⊗ TrY modulo the Künneth formula. The Poincaré pairing
∆∗ Tr
< ., . >X : H i (X) ⊗ H 2dX −i (X) → H 2dX (X ×k X) →X H 2dX (X) →X K(−dX )
is perfect.
(vi) Cycle class: For any X ∈ SmProj(k) and i ∈ N one has a homomorphism:
γX : CH i (X) → H 2i (X)(i) := Hom(K(−i), H 2i )
where CH i (X) = Z i (X)/ ∼rat is the i-th Chow group, satisfying:
(a) for any f : X → Y, γX ◦ f ∗ = f ∗ ◦ γY .
(b) for any cycle α, β, one has γX×k X (α×k β) = γX (α)⊗γX (β) in H • (X×k X) =
H • (X) ⊗K H • (X).
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 29

(c) If X is geometrically connected of dimension dX then for any α ∈ CH dX (X)

one has:
< 1, γ(α) >= deg(α) .
Notice that any Weil cohomology is endowed with a natural ring structure, the cup
product on H • (X) being defined as:
∀α ∈ H i (X), β ∈ H j (X), α · β = ∆∗X (α ⊗ β) .
5.2.1. Digression on Chow groups. At this point we don’t want to review the theory of
Chow groups. We just recall the basic definition (see [Ful98] for details). Let X be an
arbitrary variety over k. The group CH r (X) is the quotient of Z r (X) by the rational
equivalence relation ∼rat , where the equivalence relation ∼rat is generated by forcing
DivY (ϕ) = 0 where Y is an irreducible closed subvariety of X of codimension r − 1 and
ϕ is a non-zero rational function on Y . We do not give the general definition of DivY (ϕ).
For Y normal this is the usual definition of the principal divisor corresponding to a
rational function. For any morphism f : X → Y between smooth varieties one defines
non-trivially a pull-back f ∗ : CH • (Y ) → CH • (X). In the case where Z is an irreducible
subvariety of Y such that f −1 (Z) has pure dimension P dim Z + dim X − dim Y and f is
flat in a neighbourhood of Z then f ∗ [Z] := [f −1 Z] := W nW W , where W go through
the irreducible components of Z red and nW := lOZ,W (OZ,W ) is the length of its generic
point. The product on CH • (X) is defined by [Z1 ] · [Z2 ] = ∆∗X ([Z1 × Z2 ]).
5.2.2. Basic properties of Weil cohomologies. Let f : X → Y ∈ SmProj(k). One defines
the direct image
f∗ : H i (X) → H i+2(dY −dX ) (Y )(dY − dX )
as the Poincaré dual of
f ∗ : H 2dX −i (Y )(dX ) → H 2dX −i (X)(dX ) .
Hence TrX = aX ∗ , where aX : X → Spec k is the structural morphism. One easily
checks the projection formula: f∗ (x · f ∗ y) = f∗ x · y.
Lemma 5.2.2. Let X, Y ∈ SmProj(k). One has a canonical isomorphism:
Homr (H • (X), H • (Y )) ' H 2dX +r (X × Y )(dX ) .
Proof.
Y
Homr (H • (X), H • (Y )) = Hom(H i (X), H i+r (Y ))
i≥0
Y
= H i (X)∗ ⊗ H i+r (Y )
i≥0
Y
' H 2dX −i (X)(dX ) ⊗ H i+r (Y ) (Poincaré)
i≥0

= H 2dX +r (X ×k Y )(dX ) (Künneth) .


Remark 5.2.3. Hence an element of H 2dX +r (X ×Y )(dX ) has to be thought as a covariant
correspondance of degree r from H • (X) to H • (Y ).
30 BRUNO KLINGLER

Poincaré duality defines (via transposition) an isomorphism

Homr (H • (X), H • (Y )) ' Hom2dx −2dY +r (H • (Y ), H • (X))(dx − dY )
or equivalently
H 2dX +r (X × Y )(dX ) ' H 2dX +r (Y × X)(dX )
∗
denoted ϕ 7→ t ϕ. One easily checks that t ϕ coincides with σX,Y ϕ, where σX,Y : X × Y →
Y × X permutes the factors.
Example 5.2.4. Let f : X → Y be a morphism. Then f ∗ ∈ H 2dY (Y × X)(dY ) and by
definition t f ∗ = f∗ .
Lemma 5.2.5. (Lefschetz trace formula) Let H • : SmProj(k)op → VectZK a Weil coho-
mology. Then for any X, Y ∈ SmProj(k) pure of dimension dX , dY respectively and any
φ ∈ H 2dX +r (X ×k Y )(dX ), ψ ∈ H 2dY −r (Y ×k X)(dY ) then
2dX
X
< φ, t ψ >X×k Y = (−1)i tr(ψ ◦ φ | H i (X)) .
i=0

Proof. By the Künneth formula and bilinearity one can assume that φ = v⊗w, ψ = w0 ⊗v 0
where v ∈ H 2dX −i (X)(dX ), w ∈ H i+r (Y ), w0 ∈ H 2dY −j−r (Y )(dY ) and v 0 ∈ H j (X).
Then φ (resp. ψ) vanishes outside H i (X) (resp. H j+r (Y )).
If x ∈ H i (X) and y ∈ H j+r (Y ) then
φ(x) =< x, v >X ·w, ψ(y) =< y, w0 >Y ·v 0 .
Hence ψ ◦ φ(x) = 0 except if i = j, in which case
ψ ◦ φ(x) =< x, v >X < w, w0 >Y ·v 0
thus
tr(ψ ◦ φ) =< v 0 , v >X < w, w0 >Y .
On the other hand:
< φ, t ψ >X×k Y = (−1)j(j+r) < v ⊗ w, v 0 ⊗ w0 >X×k Y
= (−1)j(j+r) TrX×k Y (v ⊗ w · v 0 ⊗ w0 )
= (−1)j(j+r)+j(i+r) TrX×k Y (v · v 0 ⊗ w · w0 )
= δij TrX (v · v 0 ) · TrY (w · w0 ) = δij < v, v 0 >X · < w, w0 >Y
= δij (−1)j(2dx −i) < v 0 , v >X < w, w0 >Y = δij (−1)i tr(ψ ◦ φ) .

5.3. Applications to the Weil conjectures.
Corollary 5.3.1. Suppose H • : SmProj(Fq ) → VectZK is a Weil cohomology. Then for
any X ∈ SmProj(Fq ), geometrically irreducibe, one has:
2n
X
|X(Fqn )| = (−1)j tr(Fr∗X,q | H j (X)) .
i=0

where FrX,q : X → X is the Frobenius endomorphism.

 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 31

Proof. We already saw in Lemma 4.5.1 that X(Fqn )Fq is in bijection with the closed
points of ∆X ∩ ΓFrn . More precisely:
X,q
 
X(Fqn )Fq = deg ∆X · ΓFrn
X,q

=< 1, γ(∆X · ΓFrn ) >=< 1, γ(∆X ) ⊗ γ(ΓFrn ) >,

X,q X,q

where γ denotes γX× . But γ(∆X ) = (∆X )∗ and γ(ΓFrn ) = (FrnX,q )∗ = t (FrnX,q )∗
Fq X X,q
hence
∗ t
X(Fqn )Fq =< ∆X , (FrnX,q )∗ >
2dX
X
= (−1)j tr(FrnX,q ∗ | H j (X)) by the LFT .
j=0

Here on the first line the schemes ΓFrn and ∆X are understood as elements of CH dX (X×Fq
X,q
X) and their product is in CH • (X ×Fq X). The second equality follows from axioms
(vi)(c) and (vi)(b) for Weil cohomologies, the last one from Lemma 5.2.5. 
Theorem 5.3.2. Suppose that there exists a Weil cohomology H • : SmProj(Fq ) →
VectZK . Then for any X ∈ SmProj(Fq ) one has:
2dX
j+1
Y
Z(X, t) = det(1 − t · FrX,q ∗ | H i (X))(−1) .
i=0

In particular Z(X, t) is rational and has the expected functional equation.

Proof. The computation of Z(X, t) is the same as the one in the proof of Theorem 5.1.3.
As a corollary Z(X, t) ∈ K(t) ∩ Q[t] = Q(t) [B, IV.5, Ex3].
The functional equation follows from Poincaré duality. Indeed as
< (FrX,q )∗ (x), x0 >=< x, FrX,q
∗
x0 > ,

one obtains that (FrX,q )∗ |H j (X) et (FrX,q )∗ |H 2dX −j (X) have the same eigenvalues. But
∗
FrX,q ◦ FrX,q = q dX as FrX,q : X → X is finite of degree q dX . Hence if (αi )i∈I are
∗
dX
the eigenvalues of Fr∗X,q on H 2dX −j (X), then ( qαi )i∈I are the eigenvalues of FrX,q
∗
on
H j (X). The functional equation follows. 
At this point it remains to construct such a Weil cohomology on SmProj(Fq ). In fact
for any field k and for each prime l 6= chark, Grothendieck and Artin construct a Weil
cohomology on SmProj(k) with coefficients in Ql : the l-adic cohomology. In some sense
we have now too many cohomologies. In particular for each l we obtain polynomials
j
Pj,l = det(1 − tFrX,q | Hét (X)) ∈ Ql [t] which depends a priori from l. In some sense all
these cohomologies can be compared, but not canonically. This problem gives birth to
the notion of motives.
Exercice 5.3.3. Deduce from the Riemann hypothesis over finite fields (purity) that in
fact Pj,l = Pj,l0 ∈ Q[t] for l 6= l0 .
32 BRUNO KLINGLER

6. Differential calculus
In this section we introduce étale morphisms, in the most geometric way. They natu-
rally occur while studying differential calculus, namely properties of morphisms relatively
to “infinitesimally closed points”. Algebraic geometry (or more generally the geometry of
locally ringed spaces) has a particularly nice format for such a calculus. Our presentation
essentially follows [Il96], which summarizes [EGAIV, 16 and 17].

6.1. Thickenings.
Definition 6.1.1. A morphism of schemes i : X → X 0 is a thickening if this is a closed
immersion (recall this means that |i| identifies X with a closed subspace of |X 0 | and
|i|
i] : OX 0 → i∗ OX is surjective) such that |X| ' |X 0 |.
It is a thickening of order 1 if moreover the quasi-coherent ideal sheaf
I = ker(i] : OX 0 → i∗ OX )
defining the closed subscheme X of X 0 has square zero: I 2 = 0.
Remarks 6.1.2. (i) This notion generalizes in an obvious way to the notion of thick-
ening over a base.
(ii) the notion of morphism of thickenings over a base S is given by the usual com-
mutative square over S.
Let i : X → X 0 be a thickening. Any local section of I = ker i] is thus locally
nilpotent. One says that i : X → X 0 is a thickening of finite order n if I is globally
nilpotent of order n: I n =
6 0 and I n+1 = 0. In this situation one has a filtration
0 ⊂ I n ⊂ I n−1 ⊂ · · · ⊂ I ⊂ OX 0
corresponding to a filtration
X = X0 ⊂ X1 ⊂ · · · ⊂ Xn−1 ⊂ Xn ⊂ Xn+1 = X 0
where each inclusion map Xi → Xi+1 is a first order thickening. The study of finite
order thickenings is thus reduced to the study of first-order ones.

6.2. First infinitesimal neighborhood. Let j : Z → X be an immersion (i.e. j is

an isomorphism of Z with a closed subscheme j(Z) of an open subscheme U of X), of
ideal I (i.e. I is the quasi-coherent sheaf of ideals of OU defining j(Z) in U ).
Definition 6.2.1. The first infinitesimal neighborhood of Z in X is the closed subscheme
 i
Z0  / / U defined by I 2 .

Hence one has a factorization of j as

 j0
Z / Z0 /X
i
/ .

The morphism Z  / Z 0 is a thickening of order 1 and one easily checks it satisfies
i
/
the following:
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 33

Lemma 6.2.2. Let j : Z → X be an immersion. The first infinitesimal neighborhood

Z 0 of Z in X has the following universal property: for any solid commutative diagram

T
a /Z

 0 
T 0 A_ _ _/ Z 0
a
AA
AA
AA
A 
X
where T → T 0 is a thickening of order 1 over X, there exists a unique morphism
(a0 , a) : (T ⊂ T 0 ) → (Z ⊂ Z 0 )
of thickenings over X factorizing the diagram.
6.3. Conormal subsheaf of an immersion. The nice formalism of infinitesimal
neighborhoods in algebraic geometry makes it natural to first define the notion of conor-
mal sheaf and cotangent sheaf and then the dual notion of normal sheaf and tangent
sheaf (notice that in differential geometry one usually proceeds the other way round).

Let Z  / / X be a closed immersion of ideal I ⊂ OX . The following short sequence
i

of quasi-coherent sheaves on X is exact:

0 → I 2 → I → I/I 2 → 0 .
Recall the following classical fact:
Lemma 6.3.1. The functor
i∗ : QCoh(OZ ) → QCoh(OX )
is exact, fully faithful, with essential image the OX -quasi-coherent sheaves G such that
IG = 0.
Hence the sheaf I/I 2 , which is killed by I, corresponds to a sheaf on Z: the conormal
sheaf CZ/X of Z in X.
We recover the classical “differential geometric” notion: the conormal sheaf of a C ∞
submanifold Z of a C ∞ manifold X defined by equations f1 = . . . = fr = 0 is generated
by the first order part of the fi ’s: it is the subsheaf of i∗ Ω1X annihilating the subsheaf
T Z of the tangent bundle T X.
More generally if i : Z ,→ X is an immersion we define CZ/X as CZ/U , where U is the
maximal open subscheme of X such that Z is a closed subscheme of U .
Remark 6.3.2. In [EGAIV] the conormal sheaf is denoted NZ/X but we keep this notation
for the normal sheaf
NZ/X := HomOZ (CZ/X , OZ ) .
Here we assume that CZ/X has finite presentation otherwise NZ/X is not even quasi-
coherent.
34 BRUNO KLINGLER

Lemma 6.3.3. Let


Z /X
i

f g
  0 
Z0 
i /
X0
be a commutative diagram of schemes, with i and i0 immersions. There is a canonical
morphism of OZ -modules
f ∗ CZ 0 /X 0 → CZ/X .

Proof. Locally we are in the situation:

 i
Spec (R/I)  / / Spec R .
f g
  i0 
Spec (R0 /I 0 )  / / Spec R0

The required morphism I 0 /(I 0 )2 → I/I 2 is deduced from f ] : R0 → R which maps I 0

to I. 
j
Lemma 6.3.4. Let Z ,→ Y ,→ X be two immersions. Then:

j ∗ CY /X → CZ/X → CZ/Y → 0

is an exact sequence of OZ -modules.

Proof. Locally one considers

Spec A → Spec B → Spec C ,

where C  B  A. Write I := ker(B → A), J := ker(C → A) and K = ker(C → B).

We want to show that the sequence

K/K 2 ⊗B A → J/J 2 → I/I 2 → 0

is exact. This follows immediately from I = J/K. 

 i0 / 0
Lemma 6.3.5. Let Z ,→ X be an immersion and Z  / Z / X its first infini-
tesimal neighborhood. The commutative square

Z / Z0


Z /X

∼
induces an isomorphism CZ/X → CZ/Z 0 .

Proof. Follows immediately from the definition of Z 0 , or from Lemma 6.3.4. 

 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 35

6.4. Cotangent sheaf: definition. Let f : X → S be a morphism of schemes. Let

∆ : X → X ×S X be the diagonal. The map ∆ is an immersion, which is closed if and
only if f is separated. The infinitesimal neighbourhoods of ∆ parametrize couples of
points of X “infinitesimally closed” one to another.

Definition 6.4.1. Let f : X → S be a morphism of schemes. The sheaf Ω1X/S of Kähler

differential forms of degree 1 is

Ω1X/S := CX/X×S X = I/I 2

where I ⊂ OX×S X denotes the ideal sheaf of ∆ : X → X ×S X.

Let
∆

 i0 / 0 "
X / X I
IIIII p
/ X ×S X
II II2
II II
p1 IIIII$  
I$
X
be the first infinitesimal neighborhood of ∆. Consider the exact sequence of sheaves on
X:
j2
x
0 / Ω1 / OX 0 / OX /0
X/S f
j1

where ji = p]i : OX → OX 0 , i = 1, 2, is a ring morphism. Define

dX/S := j2 − j1 : OX → Ω1X/S .

Definition 6.4.2. Recall that for f : X → S and M an OX -module one defines the
abelian group of S-derivations from OX to M by
( )
D : OX → M morphism of f −1 OS − module /
DerS (OX , M) = .
D(a · b) = a · Db + b · Da ∀ a, b ∈ OX

Lemma 6.4.3. dX/S : OX → Ω1X/S is an S-derivation.

The proof of Lemma 6.4.3 is immediate from the definition of dX/S . In fact one shows
that this construction provides the universal derivation:

Lemma 6.4.4. Let f : X → S be a morphism of schemes. The functor Mod(OX ) →

Sets which to M associates DerS (OX , M) is corepresented by Ω1X/S :

HomOX (Ω1X/S , M) →DerS (OX , M)

α 7→α ◦ dX/S .
36 BRUNO KLINGLER

Recall the local description of Ω1X/S . If f : Spec B → Spec A then Ω1B/A is the quotient
of the free B-module generated by symbols db, b ∈ B, modulo the relations
d(b + b0 ) − db − db0 , b, b0 ∈ B
d(b · b0 ) − b · db0 − b0 · db
da, a ∈ A
Moreover the differential dB/A : B → Ω1B/A is just the map associating db to b ∈ B.
Definition 6.4.5. The tangent sheaf TX/S is the dual Hom(Ω1X/S , OX ) of the cotangent
sheaf Ω1X/S .
Thus for any open subset U of X the sections of TX/S over U are Γ(U, TX/S ) =
DerS (OU , OU ). For S = Spec C we recover the classical definition of vector fields as
derivations of functions.
6.5. Cotangent sheaf: basic properties. Let
f
X0 /X

 
S0 /S
be a commutative diagram of schemes. The morphism
f] f∗ dX 0 /S 0
OX −→ f∗ OX 0 −→ f∗ Ω1X 0 /S 0
is obviously an S-derivation, hence defines an OX -morphism
Ω1X/S → f∗ Ω1X 0 /S 0 ,
or equivalently by adjunction a canonical map
f ∗ Ω1X/S → Ω1X 0 /S 0 .
The following three lemmas describe the basic properties of the cotangent sheaf:
f g
Lemma 6.5.1. Let X → Y → S. Then the sequence of OX -modules
f ∗ Ω1Y /S → Ω1X/S → Ω1X/Y → 0
is exact.
Proof. This is the sheafified version of [Mat80, Th.57 p.186]. 
Lemma 6.5.2. Let

Z ? /X
i
?? ~~
?? ~~
?? ~~
? ~~
S
be an immersion over S. The the sequence of OZ -modules
dX/S
CZ/X → i∗ Ω1X/S → Ω1Z/S → 0
is exact.
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 37

Remark 6.5.3. The canonical map dX/S is defined as follows. As I is contained in OX

one can consider the restriction dX/S : I → Ω1X/S . As dX/S is a derivation it maps I 2 to
I · Ω1X/S hence induces a map

I/I 2 → Ω1X/S /I · Ω1X/S

which is OX /I-linear. This defines dX/S : CZ/X = I/I 2 → i∗ Ω1X/S by the Lemma 6.3.1.

Proof. Locally X = Spec A, Z = Spec (B = A/I), S = Spec C and one has a commuta-
tive diagram of rings:
ϕ
AO / / B = A/I
O
α β

C C.
We want to prove that the sequence

I/I 2 → Ω1A/C ⊗A B → Ω1B/C → 0

is exact.
Surjectivity on the right: A  B hence Ω1A/C  Ω1B/C by the description of Ω1 by
generators and relations. A fortiori: Ω1A/C ⊗A B  Ω1B/C .
The composite of the two arrows is zero: indeed let f ∈ I. Then the image of
df ∈ Ω1A/C in Ω1B/C is df , where f is the class of f in B = A/I, hence 0.
Exactness in the middle: this is equivalent to showing that the kernel of the natural
map Ω1A/C → Ω1B/C is generated as A-module by I · Ω1A/B and df , f ∈ I. The explicit
description of Ω1 by generators and relations implies that this kernel is < da >, where
a ∈ A satisfy ϕ(a) = β(c) for some c ∈ C. Write a = α(c) + (a − α(c)). Then
da = d(a − α(c)) as d(α(c)) = 0 ∈ Ω1A/C . But a − α(c) ∈ I as ϕ(a − α(c)) = ϕ(a) −
ϕ(α(c)) = β(c) − β(c) = 0. This shows that the kernel of the map Ω1A/C → Ω1B/C is in
fact generated as A-module by the df ’s, f ∈ I. 

Lemma 6.5.4. Let Y be a scheme and consider AnY = Y [T1 , · · · , Tn ]. Then Ω1An /Y is a
Y
free OAnY -module with basis (dTi )1≤i≤n .

6.6. Digression: the De ViRham complex. Let f : X → S be a morphism of

i
schemes. Define ΩX/S := 1
ΩX/S . One easily shows that there exists a unique family
of morphisms of f −1 (OS )-modules d : ΩiX/S → Ωi+1
X/S satisfying the following properties:

(i) d is an S-derivation of i ΩX/S : d(a ∧ b) = da ∧ b + (−1)deg a a ∧ db (a, b homo-

L i

geneous).
(ii) d2 = 0.
(iii) da = dX/S a if a is of degree 0.
The complex (Ω•X/S , d) is called the De Rham complex of f : X → S.
38 BRUNO KLINGLER

7. Smooth, net and étale morphisms

7.1. Definitions. Recall that f : X → Y is locally of finite type if for any x ∈ X there
exist U = Spec B an open affine neighborhood of x in X, V = Spec A an open affine
neighborhood of y = f (x) in Y such that f (U ) ⊂ V and A → B is of finite type (i.e.
B = A[T1 , · · · , Tn ]/I). It is locally of finite presentation if moreover I can be chosen of
finite type over A.
If Y is locally noetherian (i.e. covered by spectra of noetherian rings) then f is locally
of finite presentation if and only if it is locally of finite type.
Definition 7.1.1. Let f : X → S be a morphism of schemes. One says that f is smooth
(resp. net or unramified, resp. étale) if:
(i) f is locally of finite presentation.
(ii) for any solid diagram

o7 ? X
g0ooooo
oo g
ooo
f
 ooo 
T0 / / T /S
i

where i is a thickening of order 1, there exists, locally for the Zariski topology on
T , one (resp. at most one, resp. a unique) S-morphism g making the diagram
commute (one says that f is formally smooth, resp. net, resp. étale).
Remarks 7.1.2. (i) We could have defined smooth, net and étale morphisms right
after defining thickenings. However the cotangent sheaf is a basic tool which
enables nice characterisation for smoothness or netness, see below.
(ii) In this definition one can obviously replace order 1 by any finite order thickening.
Corollary 7.1.3. (a) the composite of two smooth morphisms (resp. net, resp.
étale) is smooth (resp. net, resp. étale).
(b) these notions are stable under base change S 0 → S.
(c) from (a) and (b) it follows that if fi : Xi → S, i = 1, 2 is smooth (resp. net,
resp. étale) then X1 ×Y X2 → S is smooth (resp. net, resp. étale).
(d) AnS → S is smooth.
7.2. Main properties.
Proposition 7.2.1. (a) The morphism f : X → S is net if and only if Ω1X/S = 0.
If f : X → S is smooth, the OX -module Ω1X is locally free of finite type and
∀ x ∈ X, rk x Ω1X/S = dimx Xf (x) .
f g
(b) Let X → Y → S (situation of Lemma 6.5.1).
If f is smooth then
(11) 0 → f ∗ Ω1Y /S → Ω1X/S → Ω1X/Y → 0
is exact and locally split. In particular if f is étale then f ∗ Ω1Y /S ' Ω1X/S .
Conversely suppose that gf is smooth. If the sequence eq. (11) is exact and
locally split then f is smooth. If f ∗ Ω1Y /S ' Ω1X/S then f is étale.
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 39

(c) Let

Z ? /X
i
?? ~~
?? ~~
? ~~ g
f ?? ~~
S
be an immersion over S (situation of Lemma 6.5.2).
If f is smooth then the sequence of OZ -modules
dX/S
(12) 0 → CZ/X → i∗ Ω1X/S → Ω1Z/S → 0
∼
is exact and locally split. In particular if f is étale then CZ/X → i∗ Ω1X/S .
Conversely assume that g is smooth. If the sequence eq. (12) is exact locally
∼
split then f is smooth. If CZ/X → i∗ Ω1X/S then f is étale.
7.3. Local coordinates. Let f : X → S be a smooth morphism. Let x ∈ X and
let s1 , · · · , sn be sections
 of OX in a neighborhood of x such that ((dsi )x )1≤i≤n is an
OX,x -basis of Ω1X/S . As Ω1X/S is OX -locally free of finite type the (dsi )1≤i≤n are an
x
OX -basis of Ω1X/S over some open neighborhood U of x in X. This defines a morphism
s = (s1 , · · · , sn ) : U → AnS = S[T1 , · · · , Tn ] .
It follows from the converse part of Proposition 7.2.1(b) that the map s is étale.
Definition 7.3.1. One says that the (si )1≤i≤n form a system of local coordinates of X
over S in a neighborhood of x ∈ X.
Corollary 7.3.2. Any smooth morphism is locally the composite of the projection of a
standard affine space with an étale morphism.
7.4. Jacobian criterion. Let

Z ? /X
i
?? ~~
?? ~~
? ~~ g
f ?? ~ ~
S
be an immersion over S (situation of Lemma 6.5.2). Suppose that g is smooth. Let
z ∈ Z. In order for f to be smooth at z it is enough by Proposition 7.2.1(c) to exhibit
sections s1 , · · · , sr of the ideal IZ in a neighborhood of z, generating IZ,z and such that
the vectors {(dsi )(z)}1≤i≤r are linearly independent in Ω1X/S (z) := Ω1X/S ⊗ k(z). This is
the classical Jacobian criterion.

7.5. Implicit functions theorem. In the situation of Lemma 6.5.2 again, assume
that f is smooth in a neighborhood of z ∈ Z. Sections (si )1≤i≤r of IZ generating IZ
around z form a minimal system of generators of IZ,z (i.e. define a base of IZ ⊗ k(z)
or equivalently define a basis of IZ /IZ2 = CZ/X in a neighborhood of z) if and only if
the (dsi (z))1≤i≤r are linearly independent in Ω1X/S (z). In this case one can complete
the (si )1≤i≤r by sections (sj )r+1≤j≤r+n of OX in a neighborhood of z so that the family
40 BRUNO KLINGLER

(dsi (z))1≤i≤r+n is a basis of Ω1X/S (z). Hence the (si )1≤i≤n+r define an étale S-morphism
s on a neighborhood U of z in X making the following diagram commutative:
U ∩Z /U

s s
 
AnS / An+r .
S

This is the classical “implicit functions theorem”.

7.6. Proof of proposition 7.2.1. We give part of the proof and refer to [EGAIV,
17.2] for more details.
Sub-lemma 7.6.1. Given a commutative diagram of schemes

oo7 ? ?
X
o o
ooo g2 f
g0
ooooo g1
oo  / 
T0  / / T S,
i
 / T is a thickening of order 1 and ideal I, the map
where T0 /
i

g2] − g1] : OX → g0 ∗ OT
factorizes trough g0 ∗ I. Moreover:

g2] − g1] ∈ DerS (OX , g0 ∗ I) = HomOX (Ω1X/S , g0 ∗ I) .

Remark 7.6.2. Notice that T0 and T have the same underlying topological space. In
particular g0 ∗ OT makes sense and coincide with gi ∗ OT , i = 1, 2.
Proof. The proof is elementary. Locally one has a commutative diagram of rings
B
nn~ O
nn
g0ngnn~~~~~~
nn 1 ~
nnn ~~~~~~g2
f
wn nn ~~
C0 o o Co A.
Clearly the map ϕ : B → C defined by ϕ(b) = (g2 − g1 )(b) takes values in I := ker(C →
C0 ).
We need to check that ϕ belongs to DerA (B, I). As g1 and g2 are ring homomorphisms
one immediately obtains ϕ(ab) = aϕ(b) for all a ∈ A and b ∈ B. Moreover for any
b, b0 ∈ B:
ϕ(b · b0 ) = g2 (b)g2 (b0 ) − g1 (b)g1 (b0 )
= g2 (b)(g2 (b0 ) − g1 (b0 )) + g1 (b0 )(g2 (b) − g1 (b))
= bϕ(b0 ) + b0 ϕ(b).

 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 41

7.6.1. Proof that f : X → S is net if and only if Ω1X/S = 0. Let us suppose that
Ω1X/S = 0. We have to show that g1 = g2 . But:

g2] − g1] ∈ Hom(Ω1X/S , g0 ∗ I) = 0 ,

hence g2] − g1] = 0 and g2 = g1 .

Conversely suppose that f : X → S is net. Consider the diagram

i9 4 X
ii iiitititttt9
Id iii p2t tt
i t
ii iiiii ttttttp1 f
 iiii / tt 
X JJ / (X ×S X)1 /S
JJi
JJ
JJ
∆ JJ% 
X ×S X
where (X ×S X)1 denotes the first infinitesimal neighborhood of ∆. As f is net one
obtains p2 = p1 hence
0 = p]2 − p]1 =: dX/S : OX → Ω1X/S .
Notice that dX/S corresponds to IdΩ1 under the canonical isomorphism
X/S

DerOX (OX , Ω1X/S ) ' HomOX (Ω1X/S , Ω1X/S ) .

Hence IdΩ1 = 0 and Ω1X/S = 0.

X/S
2

7.6.2. Proof of Proposition 7.2.1(c). Let


Z@
i /
X
@@
@@ g
@
f @@ 
S
be an immersion over S. We want to show that if f is smooth then the sequence of
OZ -modules
0 → CZ/X → i∗ Ω1X/S → Ω1Z/S → 0
is exact and locally split. Consider the commutative diagram

nn > Z
nnnnn
nn
nnn
f
r
 nnn 
Z / / Z1 /S
i1
 / Z1  
where Z  / / X is the first infinitesimal neighborhood of Z in X and the
i1
(local) retraction r of i1 is provided by the smoothness of f .
42 BRUNO KLINGLER

Define ϕ : i∗ Ω1X/S → CZ/S by ϕ(da mod I) = (IdZ1 − i1 ◦ r)∗ a mod I 2 for a ∈ OX .

One easily checks that ϕ is an inverse of the natural morphism dX/S : CZ/X → i∗ Ω1X/S
hence the result.
2
7.6.3. Extensions of schemes by quasicoherent modules. The rest of the proof require
some preliminaries.
Definition 7.6.3. Let f : X → S be a morphism of schemes and I ∈ QCoh(OX ). A
S-extension of X by I is an S-thickening X 0 of X of order 1, of ideal I:

X B
i /
X0
BB
BB g
B
f BB 
S
An isomorphism of S-extensions
i0 a i00
(X ,→ X 0 ) ' (X ,→ X 00 )
is an S-morphism a : X 0 → X 00 such that ai0 = i00 and a induces the identity map on I.
In particular the map a−1 is an isomorphism:
OX 0
{{= O EEE
{{ EE
{{ EE
{ E"
{
0 /I −1 OX /0
CC a
y<
CC y
CC yy
CC yyy
! y
OX 00
Remarks 7.6.4. (i) Notice that a priori there is no multiplicative structure on I.
(ii) As a 1-thickening of X has the same space as X, the datum of an S-extension
X 0 of X by I is equivalent to the datum of an extension
p
0 /I / OX 0 / OX /0
O :
uuu
uu
uu
uu
f −1 (OS )
where OX 0 is an f −1 (OS )-algebra and p is a homomorphism of f −1 (OS )-algebras.
Hence the problem of constructing extensions is similar to the problem of con-
structing extensions of modules over a ring.
(iii) This notion of extension plays a crucial role in deformation theory but we won’t
go there.
Definition 7.6.5. We denote by ExtS (X, I) the set of isomorphism classes of S-extensions
of X by I.
Lemma 7.6.6. ExtS (X, I) is naturally an abelian group, with neutral element the trivial
extension D(I) := OX ⊕ I (dual numbers over I)
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 43

Proof. Let us define the addition. Given two isomorphism classes ci := [0 → I → OXi →
OX ] ∈ ExtS (X, I), i = 1, 2 we first consider the pull-back diagram:

0 / I ⊕I / OX ⊗O OX / OX /0
1 X 2

∆
 
0 / I ⊕I / OX ⊕ O X / OX ⊕ O X /0
1 2

then the pushout:

0 / I ⊕I / OX ⊕ O X / OX /0
1 2

+
 
0 /I / OX / OX /0
3

and define c1 + c2 := [0 → I → OX3 → OX → 0]. One easily shows this class does not
depend on the choices of representatives for c1 and c2 . 

Lemma 7.6.7. Let f : X → S and I ∈ QCoh(OX ). Assume that f is smooth. Then

the morphism

ϕ : ExtS (X, I) → Ext1OX (Ω1X/S , I)

  i/ 0
 
 X DDDX  7→ [0 → I → i∗ Ω1 0 → Ω1 → 0]
! X /S X/S
S

is an isomorphism.

Proof. One easily checks that ϕ is a homomorphism of abelian groups. Using that f is
smooth one defines an inverse

ψ : Ext1OX (Ω1X/S , I) → ExtS (X, I)

u v
to ϕ as follows. Given [0 → I → E → Ω1X/S → 0] ∈ Ext1OX (Ω1X/S , I) consider the
pullback diagram

(0,u)
0 /I / OX ⊕ E Id⊕v / OX ⊕ Ω1 /0.
EE O X/S
EE O
EE
EE p Id+dX/S
EE
"
OX 0 / OX /0
q


X  D / X0
" #
Define ψ(E) = DD 
The composition ψϕ is obviously the identity. To prove that
!
S
ϕ ◦ ψ = Id, note that p − q : OX 0 → E is naturally an S-derivation hence defines a
morphism γ : i∗ Ω1X 0 /S → E. The following commutative diagram whose second line is
44 BRUNO KLINGLER

ϕψ(E)
0 /I /E / Ω1 /0
O X/S

0 /I / i∗ Ω1 0 / Ω1 /0
X /S X/S

shows that γ is an isomorphism and the result. 

7.6.4. Proof that if f : X → S is smooth then Ω1X/S is locally free. We just proved that
∀ I ∈ QCoh(OX ), ExtS (X, I) ' Ext1OX (Ω1X/S , I) .
Denoting by ExtS (X, I) the Zariski sheaf on X associated to the presheaf U 7→ ExtS (U, I|U )
one concludes that
ExtS (X, I) ' Ext1 (Ω1X/S , I) .
As f is smooth any S-extension of X by I is locally trivial (as there exists a local
retraction) hence ExtS (X, I) = 0 thus Ext1 (Ω1X/S , I) = 0. As this is true for any
I ∈ QCoh(X) and Ω1X/S is of finite type over OX ,we conclude that Ω1X/S is locally free
by the sublemma below.
2
Sub-lemma 7.6.8. Let X be a scheme, F ∈ QCoh(OX ) of finite type. Suppose that for
any I ∈ QCoh(X) the group Ext1OX (F, I) vanishes. Then F is a locally free OX -module.
Proof. As F is of finite type there exists an exact sequence of the form
n
(13) 0 → I → OX →F →0 .
In particular I ∈ QCoh(OX ). By hypothesis Ext1OX (F, I) = 0 hence the exact sequence
eq. (13) locally splits, which implies that F is locally free. 
f g
7.6.5. Proof that if X → Y → S and f is smooth then the sequence 0 → f ∗ Ω1Y /S →
Ω1X/S → Ω1X/Y → 0 is exact locally split. We start with the
f g
Lemma 7.6.9. Consider X → Y → S with f affine. Let I ∈ QCoh(OX ). Then one
has a canonical exact sequence of abelian groups
(14) 0 → DerY (OX , I) → DerS (OX , I) → DerS (OY , f∗ I)
∂
→ ExtY (X, I) → ExtS (X, I) → ExtS (Y, f∗ I),
where all the maps except ∂ are defined via the obvious functorialities and if D ∈
DerS (OY , f∗ I) one defines
p
∂(D) : 0 /I / OX 0 / OX /0
O
vv:
v
vv
vvv
v
f −1 (OS )
where the map f −1 (OY ) → OX ⊕ I corresponds to (f ] , D) : OY → f∗ OX ⊕ f∗ I.
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 45

Proof. The proof is long but easy, see [EGAIV, 0IV 20.2.3]. 
Suppose now that f is smooth. The assertion that the sequence
0 → f ∗ Ω1Y /S → Ω1X/S → Ω1X/Y → 0
is exact locally split is local. Hence we can assume that X = Spec C, Y = Spec B and
S = Spec C. In particular f is affine. Showing that
0 → C ⊗B ⊗Ω1B/A → Ω1C/A → Ω1C/B → 0
is exact locally split is equivalent to showing that for any C-module I, the sequence of
Abelian groups obtained by applying the functor HomC (·, I) is exact, equivalently that
the sequence
0 → DerB (C, I) → DerA (C, I) → DerA (B, IB) → 0
is exact. As f is smooth Ω1X/Y is locally free by the previous section, hence Ω1C/B is
projective of finite type over C. Hence Ext1C (Ω1C/B , I) = ExtY (X, I) = 0 and the result
follows from the Lemma 7.6.9.
2
7.6.6. We leave the two converse statements of 7.2.1 to the reader. He will prove them
using the techniques already developed.
7.7. A remark on smoothness. Differential calculus provides a simple characteri-
sation for a morphism f : X → S to be net: Ω1X/S = 0. If f : X → S is smooth, we
showed that Ω1X/S is OX -locally free. This is not a characterization of smoothness.
Let us indeed consider the following example. Let A be a ring and B = A[X, Y ]/(g).
Consider the diagram

Spec B  / A2
i
A
www
ww
ww
f
 w{ w
Spec A.
The associated exact sequence of B-modules
(15) CB/A[X,Y ] ' (g)/(g 2 ) → Ω1A[X,Y ]/A ⊗A[X,Y ] B → Ω1B/A → 0
can be rewritten
B → BdX ⊕ BdY → Ω1B/A → 0 ,
where one maps 1 ∈ B to the differential ∂g/∂XdX + ∂g/∂Y dY . The Jacobian criterion
shows that f : Spec B → Spec A is smooth if and only if
< ∂g/∂X, ∂g/∂Y >= B .
In this caseΩ1B/A is locally free of rank one over B.
However there are other cases where Ω1B/A is locally free. Suppose that A has charac-
teristic p and f = X p + Y p . In this case Ω1B/A is free of rank 2. Clearly Spec B is still
of relative dimension 1 over A and we don’t want to call such a map smooth!
Remark 7.7.1. Still, there is a purely differential criterion for smoothness involving the
cotangent complex and not only Ω1X/S .
46 BRUNO KLINGLER

7.8. Smoothness, flatness and regularity.

7.8.1. Smoothness and flatness. In this section we relate the smoothness of a morphism
f : X → S to the smoothness of its fibers:
Theorem 7.8.1. Let f : X → S be locally of finite presentation. The following condi-
tions are equivalent:
(i) f is smooth.
(ii) f is flat and for any s ∈ S the fiber Xs /s is smooth.
Proof. Let us show that (2) ⇒ (1). Let x ∈ X, we want to show that f : X → S is
smooth at x. Let s = f (x). The problem is local on X and we may assume that X is
embedded in some Z := An+r
S with ideal I. We have the diagram

Xs / X /
i /Z
~~
~~
f ~~
  ~ ~
s / S.

Consider the exact sequence

0 → Ix → OZ,x → OX,x → 0 .
Since f is flat one obtains an exact sequence after tensoring with k(s):
0 → Ix ⊗OS,s k(s) → OZs ,x → OXs ,x → 0 .
As fs is smooth at x one may choose (g1 , · · · , gr ) generating Ix ⊗OS,s k(s) such that
dg1 (x), · · · , dgr (x) are linearly independent in Ω1Zs /s ⊗OZs ,x k(x) = Ω1Z/S ⊗OZ,x k(x). Lift
(g1 , · · · , gr ) to (f1 , · · · , fr ) ∈ Ix . Then df1 (x), · · · , dfr (x) are linearly independent in
Ω1Z/S ⊗OZ,x k(x). By Nakayama’s lemma Ix is generated by f1 , · · · , fr . By the Jacobian
criterion f is smooth at x.
Conversely let us prove that (1) ⇒ (2). Assume that f : X → S is smooth. By
Corollary 7.1.3 smoothness is stable under base change of the target thus Xs /s is smooth
for any s ∈ S. It remains to show that f : X → S is flat. Let s ∈ S and x ∈ Xs . Locally
around x we have a commutative diagram

X / Z = An+r
i
/
S
u
u uu
u
uu
f
 uz uu
S.
Notice that Z is obviously flat over S at x. To prove that X → S is flat, we introduce
the notion of regular immersion:
Definition 7.8.2. A closed immersion i : X ,→ Z of locally Noetherian schemes is said
regular at a point x ∈ X if the ideal I of i can be locally defined by (f1 , · · · , fr ) at x,
such that (fi )x is a regular sequence in OZ,x .
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 47

Proposition 7.8.3. Let


X /Z
i
/
~~
~
~~
 ~~ ~~
S
be a closed immersion locally of finite type over a locally Noetherian scheme S. Let s ∈ S
and x ∈ Zs . The following conditions are equivalent:
(1) The closed immersion is : Xs ,→ Zs is regular at x and Z is flat over S at x (i.e.
OX,x is a flat OS,s -module).
(2) X is flat over S at x and i : X ,→ Z is regular at x.
In particular the closed immersion i : X ,→ Z is regular and Z is flat over S if and only
if X is flat over S and is : Ys ,→ Xs is regular for any s ∈ S.
Admitting Proposition 7.8.3 for a moment, we are reduced to prove that the closed
immersion Xs ,→ Zs is regular.
Let I be the ideal of i and f1 , · · · , fr local sections of I at x such that (fi )x is a
minimal system of generators of Ix , i.e. (fi ⊗ k(x))1≤i≤r is a basis of CX/Z (x). As f is
smooth the sequence of k(x)-vector spaces
/ CX/Z (x) / Ω1 ⊗ k(x) / Ω1 /0
0 Z/S X/S ⊗ k(x)

is exact. As the diagram

/ CX/Z (x) / Ω1 ⊗ k(x) / Ω1 /0
0
MMM Z/S X/S ⊗ k(x)
O
MMM
MMM dZ/S
MM&
mXs ,x /m2Xs ,x

is commutative, it follows that the k(x)-linear map CX/Z (x) → mXs ,x /m2Xs ,x is injective.
Hence the (fi )x ’s, 1 ≤ i ≤ r form a regular sequence in OXs ,x and the closed immersion
Xs ,→ Zs is regular. This finishes the proof that X → S is flat, hence the proof of
Theorem 7.8.1, assuming Proposition 7.8.3. 

Proposition 7.8.3 is the special case of the following algebraic statement for A = OS,s ,
B = OZ,x and M = OZ,x :
Proposition 7.8.4. Let (A, mA ) → (B, mB ) be a local morphism of Noetherian local
rings and k = A/mA . Let M be a finitely generated B-module and (f1 , · · · , fr ) ∈ mB .
The following conditions are equivalent:
(1) M is A-flat and (f1 ⊗ k, · · · , fr ⊗ Pk) is (M ⊗ k)-regular.
(2) (f1 , · · · , fr ) is M -regular and M/ ri=1 fi M is flat over A.
Proof. We start with a few lemmas.
Lemma 7.8.5. Let R be an Artinian local ring, with maximal ideal m and residue field
k = R/m. Let M be an R-module. Then M ⊗R k = 0 implies M = 0.
48 BRUNO KLINGLER

Proof. Since R is local Artinian there exists an integer m such that mm = 0. Then
M ⊗R k = 0 implies
M = mM = m2 M = · · · = mn M = 0 .

Lemma 7.8.6. Let R be an Artinian local ring and M an R-module. Then M is free if
and only if M is flat.
Proof. If M is free it is clearly flat. Conversely let m be the maximal ideal of R and k
its residue field. Choose (xα )α∈I a family of elements of M lifting a basis of M/mM .
Denote by F the free R-module with basis (eα )α∈I and g : F → M the homomorphism
of R-modules mapping eα to xα . Applying Lemma 7.8.5 to the Coker u shows that g is
surjective, hence provides an exact sequence of R-modules
g
0→K→F →M →0 .
Writing the beginning of the long exact sequence associated to the functor · ⊗R k one
obtains
g
TorR
1 (M, k) = 0 → K/mK → F/mF → M/mM → 0 .
Hence K/mK = 0, thus K = 0 by Lemma 7.8.5. 
Lemma 7.8.7. (A, mA ) → (B, mB ) be a local morphism of Noetherian local rings and
k = A/mA . Let E, F be finitely generated B-modules and u : E → F a morphism of
B-modules.
Suppose that F is A-flat and u ⊗ k : E ⊗ k → F ⊗ k is injective. Then u is injective
and Coker u is flat over A.
Proof. (Raynaud) For n a non-negative integer let An := A/mn+1 , En := E ⊗ An and
Fn := F ⊗ An . We first show that un : En → Fn is injective and split. Since Fn is flat
over An and An is Artinian, Fn is free over An by Lemma 7.8.6. Take a basis of En ⊗ k
and lift its image in Fn ⊗ k into a part of basis of Fn , which forms a free An -submodule
L of Fn . The diagram
L Cp
CC
CC
ϕ CC
 C!
En / Fn
commutes (where ϕ is defined in the obvious way). In particular ϕ is injective. By
Nakayama’s lemma ϕ is also surjective. Hence ϕ is an isomorphism, and the sequence
n u
0 → En → Fn → Coker (un ) → 0
is exact and split.
The fact that Fn is An -flat thus implies that Coker (un ) is also An -flat. Consider the
commutative diagram
E
u /F

 

Ê := colimEn  / F̂ := colimFn ,
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 49

where E ,→ Ê (and similarly F ,→ F̂ ) by [B, III, 5, prop.2]. So E → F is injective and

Coker (u) is A-flat by [B, III, 5, theor.1]. 

Lemma 7.8.8. Let (A, mA ) → (B, mB ) be a local morphism of Noetherian local rings.
Let M be a finitely generated B-module and f ∈ mB . If M/f n+1 M is flat over A for
any m ≥ 0 then M is flat over A.

Proof. It is enough to show that for any N ,→ N 0 finitely generated A-module the
induced morphism u : M ⊗A N 0 → M ⊗A N is injective.
Let x ∈ Ker(u). Fix n ≥ 0. As M/f n+1 M is A-flat, the morphism

M/f n+1 M ⊗A N 0 → M/f n+1 M ⊗A N

is injective. Hence x ∈ f n+1 (M ⊗A N 0 ). Finally x ∈ ∩n f n+1 (M ⊗A N 0 ).

As M ⊗A N 0 is a finitely generated B-module it is separated for the f -adic topology,
hence x = 0. So u is injective. 

We now finish the proof of Proposition 7.8.4.

Let us show (1) ⇒ (2). By induction on r we are reduced to the case r = 1. By
assumption f ⊗ k : M ⊗ k → M ⊗ k is injective and M is flat over A. Thus f is injective
and M/f M is A flat by Lemma 7.8.7.
Conversely let us show (2) ⇒ (1). Once more by induction on r we are reduced to the
case r = 1. Consider the exact sequence
f
(16) 0 → M → M → M/f M → 0 .

Applying the functor ·⊗A k to this exact sequence, one obtains that f ⊗k : M ⊗ → M ⊗k
is injective as M/f M is A-flat. It remains to show that M is flat over A. Consider the
exact sequence
fn
0 → M/f M → M/f n+1 M → M/f n M → 0 .
By induction on n we obtain that M/f n+1 M is A-flat for any n. Hence M is A-flat by
Lemma 7.8.8.


7.8.2. Smoothness and regularity. Via Theorem 7.8.1 we now relate the geometric Def-
inition 7.1.1 of smoothness and étaleness to a more algebraic one (used for example in
[SGA1]):

Theorem 7.8.9. Let f : X → S a morphism of schemes locally of finite presentation.

The following conditions are equivalent:
(i) f is smooth.
(ii) f is flat and the geometric fibers of f are regular schemes.

Corollary 7.8.10. Let f : X → S a morphism of schemes. The morphism f is étale if

and only if f is locally of finite presentation, flat and net.
50 BRUNO KLINGLER

7.8.3. Regularity. We start with classical facts on regularity.

Let A be a Noetherian local ring, with maximal ideal m and residue field k := A/m.
In general d := dim A ≤ rk k m/m2 (see [Stacks Project, Commutative Algebra, 57]).
Definition 7.8.11. A Noetherian local ring A of dimension d is said to be regular if the
following equivalent conditions are satisfied:
(i) d = rk k m/m2 .
(ii) there exist x1 , . . . , xd ∈ m generating m.
A sequence (xi )1≤i≤d as in (ii) is called a regular system of parameters for the regular
local ring A.
The regularity of a Noetherian local ring is a homological property. Recall that the
homological dimension hdim(A) of a ring A is the smallest integer n such that any A-
module M has a projective resolution of length at most n (if such an integer does not
exist one defines hdim(A) = +∞). One easily shows that hdim(A) ≤ n if and only if for
any ideal I of A and any A-module M the groups ExtiA (A/I, M ), i > n, do vanish.
Theorem 7.8.12 (Serre). A local ring A is regular if and only if is has finite homological
dimension. In this case hdim(A) = dim A.
As a corollary regularity is stable under localization:
Corollary 7.8.13. If A is a regular local ring and p ∈ Spec A then Ap is regular.
Proof. Let J be an ideal of Ap . Hence J = Ip , where I is an ideal of A. Similarly any
Ap -module is of the form Mp , M ∈ A − Mod. By localization:
ExtiAp (Ap /Ip , Mp ) ' ExtiA (A/I, M )p = 0 for i > d
as A is regular of dimension d. Hence Ap has finite homological dimension, hence is
regular by Serre’s theorem. 
We also need to understand when a quotient of a regular ring is regular.
Lemma 7.8.14. Le A be a regular local ring with maximal ideal m and dim A = d. Let
I ⊂ m, B = A/I. The following properties are equivalent:
(1) B is regular.
(2) there
Pr exists a regular system of parameters (x1 , . . . , xd ) of A such that I =
i=1 xi A.

Proof. Let us show that (2) implies (1). Assume that (x1 , · · · , xr ) is part of a regular
system of parameters of A. Then dim B = d − r (see [EGA0, IV 16.3.7]). Let n = m/I
be the maximal ideal of B, then we have an exact sequence
(17) 0 → (m2 + I)/m2 → m/m2 → n/n2 → 0 .
Since the xi , 1 ≤ i ≤ r, generate I and have linearly independent images in m/m2 ,
dimk (m2 + I)/m2 = r, hence dimk n/n2 = d − r = dim B hence B is regular.
Conversely let us show that (1) implies (2). Let d−r be the dimension of B. Assuming
that B is regular, one has the equality d − r = dimk n/n2 . The sequence eq. (17) implies
that dimk (m2 + I)/m2 = r. Choose xi , 1 ≤ i ≤ r, having linearly independent images
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 51

in m/m2 , and choose xr+1 , · · · , xdP∈ m such that (x1 , · · · , xd ) is a regular system of
parameters of A. Denote by I 0 := ri=1 xi A ⊂ A and consider the exact sequence
0 → I/I 0 → A/I 0 → A/I → 0 .
As A/I is regular, A/I is a domain. Hence I/I 0 is prime in A/I 0 .
On the other hand it follows from (2) ⇒ (1) that A/I 0 is regular and dim A/I 0 =
d − r = dim A/I. The fact that I/I 0 is prime then implies I = I 0 . 

Let us globalize the notion of regularity.

Definition 7.8.15. A scheme X is called regular if it is locally Noetherian and for any
point x in X the Noetherian local ring OX,x is regular.
Corollary 7.8.16. Let X be a Noetherian scheme. If OX,x is regular for all closed
points x of X then it is regular for all points x of X.
Proof. As X is Noetherian it is quasi-compact. Hence any point has a closed point in
its closure (see [Stacks Project, Schemes, 27.5.8]) and we can assume that X = Spec A
is affine. Let p ∈ Spec A. There exists a maximal ideal m ⊃ p. Hence Ap = (Am )p is
regular by Corollary 7.8.13. 

7.8.4. Schemes of finite type over a field. Let k be a field and X/k a scheme of finite
type. Recall that x ∈ X is a closed point if and only if [k(x) : k] < +∞ by the Hilbert
Nullstellensatz. Moreover dim X = dim OX,x in this case.
Proposition 7.8.17. Let k be a field and X/k a scheme of finite type. The following
conditions are equivalent:
(i) X/k is étale.
(ii) Ω1X/k = 0, i.e. X/k is net.
(iii) X = Spec ni=1 Ki , where Ki /k is a finite separable extension.
Q

Proof. The implication (i) ⇒ (ii) is obvious.

For (ii) ⇒ (iii): We can assume that X = Spec A is affine. We want to show
N
that if k is an algebraic closure of k then A ⊗k k ' k (this characterises separable
extensions). Let Z := Spec (A ⊗k k) and x ∈ Z a closed point (hence k(x) = k). Thus
Ω1Z/k = Ω1X/k ⊗k k = 0 by assumption. The diagram

x E /Z
EE
EE
EE
f E" 
Spec k

gives, as f = Id is obviously smooth:

0 → Cx/Z = mx /m2x → Ω1Z/k ⊗k k(x) → Ω1x/k = 0 → 0 .

Hence mx /m2x ' Ω1Z/k ⊗k k(x) = 0 thus mx = 0 and OZ,x = k(x) = k as required.
52 BRUNO KLINGLER

For (iii) ⇒ (i): without loss of generality one can assume that X = Spec K, K/k
finite separable. Write K = k[T ]/(f ) with f 0 (T ) 6= 0 in K. Consider the diagram:
 i
X = Spec K  / Spec k[T ]
o o
ooooo
o
 ow oo
Spec k.
The Jacobian criterion implies that X/k is smooth, obviously of relative dimension zero,
hence étale. 
Theorem 7.8.18. Let k be a field and X/k be a scheme of finite type.
(1) if X/k is smooth then X is regular. If moreover X is integral then rk k Ω1X/k =
dim X.
(2) If k is perfect and X is regular then X/k is smooth.
Proof. For (1): by Corollary 7.8.16 it is enough to show that for any closed point x of X
the local ring OX,x is regular. Let x ∈ X be a closed point. In particular [k(x) : k] < +∞.
Locally the following diagram holds:

X / Z := An+r
i
/
k
rrrrr
rr
 rx rr
Spec k

Pr X of ideal I in Z. As X/k is smooth one can choose (fi )1≤i≤r in OZ with Ix =

with
i=1 (fi )x OZ,x and (dZ/k fi ⊗ k(x))1≤i≤r linearly independent. Denote m := mZ,x . Then:

 dZ/k / 1
I/I 2 ⊗ k(x)  ΩZ/k ⊗ k(x)
OOO O
OOO
OOO dZ/k
OOO
'
m/m2 .
Hence the [(fi )x ]1≤i≤r mod m2 are linearly independent in m/m2 . I.e. they are part of
a regular system of parameters for OZ,x .
Hence OX,x = OZ,x /Ix is regular by the Lemma 7.8.14.
Suppose moreover X integral. As X is smooth over k the following sequence is exact:
0 → I/I 2 → Ω1Z/k ⊗ OX → Ω1X/k → 0 .
But rk (Ω1Z/k ⊗ OX ) = n + r and rk (I/I 2 ) = r hence rk Ω1X/k = n = dim X.
For (2): consider once more

X / Z := An+r
i
/
k
rrrrr
rr
 rx rr
Spec k
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 53

with X of ideal I in Z. For x ∈ X a closed point we want to show that

dZ/k ⊗ k(x) : I/I 2 ⊗ k(x) ,→ Ω1Z/k ⊗ k(x) ,
hence X/k is smooth thanks to the Jacobian criterion.
As k is perfect the extension k(x)/k is separable hence Ω1k(x)/k = 0. Consider the two
exact sequences:
(18) I/I 2 ⊗ k(x) → Ω1Z/k ⊗ k(x) → Ω1X/K ⊗ k(x) → 0
(19) mx /m2x → Ω1X/k ⊗ k(x) → Ω1k(x)/k → 0
The first one implies that
dim Ω1X/k ⊗ k(x) ≥ dim Ω1Z/k ⊗ k(x) − r = n .
On the other hand the second one implies:
dim Ω1X/k ⊗ k(x) ≤ n .
Hence dim Ω1X/k ⊗ k(x) = n and dZ/k ⊗ k(x) is injective.

Corollary 7.8.19. Let k be a field and X/k be a scheme of finite type. The following
assertions are equivalent:
(i) X/k is smooth.
(ii) For any extension k 0 /k the scheme X ⊗ k 0 is regular.
(iii) There exist a perfect extension k 0 of k such that X ⊗ k 0 is regular.
Proof. (1) ⇒ (2) and (2) ⇒ (3) are obvious. Let us show (3) ⇒ (1). As X ⊗ k 0 is regular
and k 0 is perfect, it follows from Theorem 7.8.18(c) that X ⊗ k 0 is smooth over k 0 . As
X/k is of finite type there exists a closed immersion i : X ,→ Ank , we denote by C its
conormal sheaf. By base change it induces a closed immersion i0 : X ⊗ k 0 ,→ Ank0 , with
conormal sheaf C 0 . Let x be a point of X and x0 a point of X 0 over x. As X 0 /k is smooth
the linear map
dAn /k0 ⊗ k(x0 ) : C 0 ⊗ k(x0 ) → Ω1An0 ⊗ k(x0 )
k0 k

is injective by Proposition 7.2.1(c). Consider the commutative diagram

dAn ⊗k(x)
C ⊗ k(x)
k
/ i∗ Ω1 n ⊗ k(x)
Ak /X

 dAn ⊗k0 (x) 

C0 ⊗ k(x0 )
k0
/ i0∗ Ω1 n 0 ⊗ k(x0 ) .
Ak0 /X

As k → k 0 is flat one shows that the vertical maps of this diagram are injective. Hence
dAnk ⊗k(x) is injective. It follows from Proposition 7.2.1(c, converse) that X/k is smooth.

7.8.5. Proof of Theorem 7.8.9. As any algebraically closed field is perfect, it follows
from Theorem 7.8.18 that Theorem 7.8.9 is equivalent to Theorem 7.8.1.
2
54 BRUNO KLINGLER

7.9. Examples of étale morphisms.

Example 7.9.1. We relate the notion of étale morphisms to classical facts of algebraic
number theory. Let L/K be an extension of number fields. Consider the morphism
f : Spec OL → Spec OK between their rings of integers. The ramification locus of this
morphism is an ideal of OL , called the different DL/K , which is nothing else than the
annihilator of Ω1OL /OK . The discriminant of this morphism, an ideal of OK , is the norm
of the different, i.e. f∗ DL/K . If one defines X := Spec OL \DL/K , the morphism f : X →
Spec OK is unramified. As any local homomorphism of DVR is flat, f : X → Spec OK is
in fact étale. Denote by Y the complement of the discriminant in Spec OK , the morphism
f : X → Y is finite étale.
Example 7.9.2. [Ray70, p.66]
Lemma 7.9.3. Let A be a ring and B = A[T ]/(T n − a). Then B is étale over A if and
only if n = 1 or na is invertible in A.
Proof. Let p ∈ Spec A and let k := k(p). Let B := B ⊗A k = k[T ]/(T n − α) where α
denotes the image of a in k. By the Jacobian criterion B is étale over k if and only if
nT n−1 and T n − α are relatively prime in k[T ]. This holds true if n = 1 or if nα 6= 0 in
k and is not true if n is a multiple of chark or if n 6= 1 and α = 1. 
Remark 7.9.4. For a = 1 the spectrum of B is nothing else than the finite group scheme
µn over A of n-roots of unity.
Example 7.9.5. [Ray70, p.70] Let k be a field and B = k[X, Y ] with the action of
G := Z/2Z by central symmetry mapping (X, Y ) to (−X, −Y ). Then A := B G is
generated over k by u = X 2 , v = Y 2 , w = XY . Hence A = k[u, v, w]/(uv − w2 ). The
algebra B is finite over A and B = A[X, Y ]/(X 2 − u, Y 2 − v, XY − w). The Jacobian
matrix has 2×2-minors equal to 4XY, −2X 2 , −2Y 2 . By the Jacobian criterion B is étale
over A outside the origin.

8. Etale fundamental group

We give a light introduction to the étale fundamental group, following [Mi80], and
refer to [SGA1] for much more material.

8.1. Reminder on the topological fundamental group. Let X be a connected

topological space. We assume that X is arcwise connected and locally simply connected.
Let x be a point in X. The fundamental group π1 (X, x) is the group of loops in X
through x, up to homotopy. This definition can hardly generalize to schemes and we
will use a more categorical one.
Recall that π : Y →
` X is a covering of X if any point x in X admits a neighbourhood U
such that π −1 (U ) ' i Ui with π|Ui : Ui → U a homeomorphism. Denote by Cov(X) the
category whose objects are coverings of X with a finite number of connected components
(and the obvious morphisms). The functor
Fx : Cov(X) → Sets
[π : Y → X] 7→ π −1 (x)
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 55

associating to any covering its fiber over x is representable by the universal cover X̃ → X:
∀ π : Y → X, Fx (Y ) ' HomX (X̃, Y ) .
The group π1 (X, x) := AutX (X̃) acts on X̃ on the right, hence on Hom(X̃, Y ) on the
left. This enriches the functor Fx as:
Fx : Cov(X) → π1 (X, x) − Sets
and defines an equivalence of categories between Cov(X) and the category of π1 (X, x)-
sets with a finite number of orbits.
We will generalize this picture to schemes.
8.2. The étale fundamental group. Let X be a scheme. Let FEt/X be the category
of finite étale morphisms π : Y → X (with X-morphisms). Let us fix x → X a geometric
point of X (hence x = Spec k with k separably closed) and consider the functor
Fx : FEt/X → F Sets
[π : Y → X] 7→ HomX (x, Y )
which associates to any finite étale cover of X its fiber over x (where F Sets denotes the
category of finite sets).
The functor Fx is usually not representable. Consider for example X = A1k \ {0} over
an algebraically closed field k of characteristic 0. One easily checks that the only schemes
t7→tn
in FEt/X are the Xn = X → X, n ∈ N∗ . There is no “biggest” such scheme, hence no
universal cover. Notice that if k = C the topological universal cover which dominates
all the Xn is given by exp : C → C∗ which is not an algebraic morphism.
However Fx is pro-representable: there exists a projective system X̃ = (Xi )i∈I of
objects Xi → X ∈ FEt/X indexed by a directed set I such that
FX (Y ) = HomX (X̃, Y ) := colimI HomX (Xi , Y ) .
One can always choose the Xi /X Galois, i.e. of degree equal to | AutX (Xi )|. Let us
define
π1ét (X, x) = AutX (X̃) := lim AutX (Xi ) .
I
As AutX (Xi ) is a finite group the group ét
π1 (X, x) is naturally a profinite group.
Example 8.2.1. Consider again the case X = A1k \ {0}, k = k of characteristic zero and
Xn as above. Then AutX (Xn ) = µn (k) (where ξ ∈ µn (k) acts on Xn by ξ(x) = ξ · x).
Hence
π1ét (A1k \ {0}) = lim µn (k) ' Ẑ .
n

Example 8.2.2. Let X/C be a smooth quasi projective variety. The Riemann’s existence
theorem (due in this generality to Grauert and Remmert) states that the natural functor
FEt/X → F Cov(X an )
[π : Y → X] 7→ [π an : Y an → X an ]
is an equivalence of categories (where F Cov(X an ) denotes the category of finite cover-
ings). Hence π1ét (X) and π1 (X an ) have the same finite quotients. As π1ét (X) is profinite
this implies that π1ét (X) ' π1 (X an )∧ , the profinite completion of π1 (X an ).
56 BRUNO KLINGLER

Exercice 8.2.3. Show that π1ét (P1k ) = {1} for any separably closed field k.
Example 8.2.4. Let X = Spec k, k a field. Choose Xi = Spec Ki where Ki ranges through
the finite extensions of k in k s . Thus π1ét (X) = Gal(k s /k).
Example 8.2.5. Let X be a normal irreducible scheme with generic point x. Write
x := Spec k(x)s and define Xi as the normalisation of X in Ki where Ki ranges through
the finite Galois extensions of k(x) in k(x)s such that Xi /X is unramified. Thus π1ét (X) =
Gal(k(x)ur /k(x)).
Theorem 8.2.6. Let X be a connected scheme and x → X a geometric point. Then
Fx : FEt/X → π1 (X, x) − F Sets
is an equivalence of categories, where π1 (X, x) − F Sets denotes the category of finite
sets with a continuous π1 (X, x)-action).

9. Sites and sheaves

9.1. Presheaves. Recall that a category is small if its objects and its morphisms form
sets.
Definition 9.1.1. Let C be a small category and D be any category. A presheaf on C
with value in D is a functor F : C op → D. We denote by PSh(C, D) the category of
presheaves on C with value in D.
Definition 9.1.2. We write PSh(C) := PSh(C, Sets) and PAb(C) := PSh(C, Ab). If
Λ is a ring, we denote by Λ − Mod the category of Λ-modules and by PΛ − Mod(C) :=
PSh(C, Λ − Mod).
Example 9.1.3. Let X ∈ C. Then
hX : C op → Sets
U 7→ hX (U ) := HomC (U, X)
is the presheaf represented by X.
Lemma 9.1.4. (Yoneda) Let C be a category. Then for any F ∈ PSh(C) there is a
functorial isomorphism
F (X) ' HomPSh(C) (hX , F ) .
In particular the functor C → PSh(C) mapping X to hX is fully faithful.
9.2. Sheaves on topological spaces. We recall some classical facts concerning
sheaves on topological spaces.
Let (X, τ ) be a topological space i.e. X is a set and τ is the set of open subsets of X.
Hence τ is a subset of P(X) such that:
(a) ∅ ∈ τ , X ∈ τ .
(b) If I is a set and (Ui )i∈I ∈ τ I then i∈I Ui ∈ τ .
S
(c) ∀ U, V ∈ τ, U ∩ V ∈ τ .
One associates canonically a category Xτ to (X, τ ). Its objects are the elements of τ
and HomXτ (U, V ) is empty if U 6⊂ V , the set with one element otherwise. By definition
a presheaf on (X, τ ) is a presheaf on Xτ .
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 57

Definition 9.2.1. Let F be a presheaf of sets on (X, τ ). It is a sheaf if the following

conditions are satisfied:
S
(1) For any U = i∈I Ui ∈ τ , for any s, t ∈ F(U ) such that s|Ui = t|Ui ∈ F(Ui ) for
all i ∈ I then Ss = t ∈ F(U ).
(2) For any U = i∈I Ui ∈ τ and any (si ∈ F(Ui ))i∈I such that si |Ui ∩Uj = sj |Ui ∩Uj
for all i, j ∈ I, there exists s ∈ F(U ) such that si = s|Ui .
S words: F ∈ Sh(Xτ ) if and only if for any U ∈ τ , for any decomposition
In other
U = i∈I Ui , the natural sequence of sets
/
Q // Q
F(U ) i∈I F(Ui ) i,j F(Ui ∩ Uj )

is exact.
Remark 9.2.2. In this theorem and in the rest of the text: a sequence of sets is said to
be exact if this is an equalizer.
Definition 9.2.3. One defines Sh(Xτ ), resp. Ab(Xτ ), resp. Λ − Mod(Xτ ), as the full
subcategory of sheaves in PSh(Xτ ), resp. in PAb(Xτ ), resp. in PΛ − Mod(Xτ ).
9.3. Sites.
9.3.1. Sieves.
Definition 9.3.1. Let C be a small category and S ∈ C. A sieve of S is a subfunctor
U ⊂ hS = Hom(·, S). In other words this is a collection of morphisms T → S stable by
precomposition.
Definition 9.3.2. Let U ⊂ hS be a sieve and f : T → S. The pull-back of U is
f ∗ U = U ×hS hT ⊂ hT .
9.3.2. Topology.
Definition 9.3.3. A Grothendieck topology τ on a small category C is the datum, for
every object S ∈ C, of a family Covτ (S) of sieves of S, called covering sieves of S,
satisfying the following axioms:
(GT1) ∀ S ∈ C, hS ∈ Covτ (S).
(GT2) ∀ f : T → S ∈ C, ∀ U ∈ Covτ (S), f ∗ U ∈ Covτ (T ).
(GT3) If V ∈ Covτ (S) and U ⊂ hS are such that for any g : T → S ∈ V, g ∗ (U) ∈
Covτ (T ) then U ∈ Covτ (S).
Definition 9.3.4. A site Cτ is a small category C equipped with a Grothendieck topology
τ.
Lemma 9.3.5. Let Cτ be a site.
(i) If U ⊂ V ⊂ hS and U ∈ Covτ (S) then V ∈ Covτ (S).
(ii) If U, V ∈ Covτ (S) then U ∩ V ∈ Covτ (S).
Proof. For (i): it is enough to notice that if f : T → S ∈ U the pull-back f ∗ V is the
sieve of T of arrows X → T whose composite with f is in V. As f ∈ U and U is a sieve,
f ∗ (V) = hT ∈ Covτ (T ) by (GT 1). It follows from (GT 3) that V ∈ Covτ (S).
For (ii): obviously U ∩ V is a sieve. Let g : T → S ∈ V. The sieve g ∗ (U ∩ V) of T
coincide with g ∗ U, which belongs to Covτ (T ) by (GT 2). The result follows then from
(GT 3). 
58 BRUNO KLINGLER

9.3.3. Pre-topologies.
Definition 9.3.6. Let C be a small category with fiber products. A Grothendieck pre-
topology on C is the datum of covering families (Si → S)i∈I for all objects S ∈ C such
that:
(PT1) For any S ∈ C, any isomorphism S 0 ' S in C is a covering family of S.
(PT2) If (Si → S)i∈I is a covering family and if T → S ∈ C is any morphism then the
family Si ×S T → T is a covering family of T .
(PT3) If (Si → S)i∈I and (Si,j → Si )j∈Ji are covering families for S and Si respectively
then (Si,j → S)i,j is a covering family for S.
Lemma 9.3.7. Let C be a small category with a Grothendieck pre-topology. Define a
covering sieve U ∈ hS as any sieve containing a covering family of S. Then this family
of covering sieves define a Grothendieck topology on C.
Proof. Exercice. 
Example 9.3.8. Let C be any small category. Define the collection of covering sieves for
S ∈ C as being reduced to hS . The associated topology is called the chaotic topology.
Sheaves for this topology are just presheaves.
Example 9.3.9. Let (X, τ ) be any topological space. One
S defines a covering family of
U ∈ Xτ as any family (Ui → U )i∈I in τ such that U = i∈I Ui . This makes Xτ a site.
Example 9.3.10. Let G be a group. Let TG be the category of G-sets (with G-equivariant
S
morphisms). The covering families are the (fi : Ui → U )i∈I such that U = i∈I fi (Ui ).
This makes TG a site.
9.3.4. Topologies on categories of schemes.
Definition 9.3.11. Let S be a scheme. One denotes by Sch/S the category of schemes
over S.
Lemma 9.3.12. Let C be a subcategory of Sch/S with fiber products. Let (P ) be a
property of morphisms of C satisfying:
(i) (P ) is true for isomorphisms of C.
(ii) (P ) is stable by base-change.
(iii) (P ) is stable by composition.
Define a family (fi : Ti → T )i∈I in C S to be a covering family if for any i ∈ I the arrow
fi : Ti → T satisfies (P ), and |T | = i∈I fi (|Ti |). This defines a (pre)-topology on C.
Proof. It is enough to check (P T 2). This follows from the fact that the underlying set
of a fiber product of schemes surjects onto the fiber product of the underlying sets. 
Definition 9.3.13. Being an open immersion, an étale morphism, a smooth morphism
or a faithfully flat morphism of finite presentation are properties (P ) satisfying the
conditions of Lemma 9.3.12. These properties define respectively the sites (Sch/S)Zar ,
(Sch/S)ét , (Sch/S)smooth , (Sch/S)fppf .
Lemma 9.3.14. Let τ ∈ {Zar, ét, smooth, fppf}. Let T ∈ Sch/S be an affine scheme and
let (Ti → T )i∈I be a τ -covering family. Then there exists a τ -covering (Uj → T )1≤j≤m
which is a refinement of (Ti → T )i∈I such that each Uj is open affine in some Ti .
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 59

This last property, which is crucial for reducing oneself to finite coverings, is not
automatically satisfied for more general flat families. Hence we define:

Definition 9.3.15. Let T → S be a scheme over S. An fpqc covering of T is a family

(fi : Ti → T )i∈I such that:
S
(i) each Ti → T is a flat morphism and |T | = i∈I fi (|Ti |).
(2) For each affine open US⊂ T there exists a finite set J ⊂ I and affine opens
Uj ⊂ Tj such that U = j∈J fj (Uj ).

This defines a site (Sch/S)fpqc .

Example 9.3.16. (i) If f : T 0 → T is flat surjective and quasi-compact then this is

an fpqc-covering.
(ii) For k an infinite field, the morphism ϕ : x∈An Spec (OAnk ,x ) → Ank is flat and
`
k
surjective but it is not quasicompact hence it is not an fpqc-covering.
(iii) Write A2k = Spec k[x, y]. The family (D(x) ,→ A2k , D(y) ,→ A2k , Spec k[[x, y]] →
A2k ) is an fpqc-covering (where D(x) and D(y) are the standard Zariski open
subsets).

9.4. Sheaves on a site.

9.4.1. Sections of a presheaf on a sieve. Let F ∈ PSh(C) and S ∈ C. By Yoneda’s

lemma:
F (S) ' HomPSh(C) (hS , F ) .
Hence it is natural to make the following:

Definition 9.4.1. Let F ∈ PSh(C) and U ⊂ hS a sieve of S ∈ C. One defines F (U) :=

HomPSh(C) (U, F ).

In down-to-earth terms: if U = {f : Uf → S}, a section s ∈ F (U) is a collection

Y
(sf ) ∈ F (Uf ) such that F (g)sf = sf g ,
f ∈U

for any f : Uf → S ∈ U and any g : X → Uf .

9.4.2. Sheaves: definition.

Definition 9.4.2. Let Cτ be a site. A presheaf F ∈ PSh(C) is a τ -sheaf (resp. is

τ -separated) if for any S ∈ C and any U ∈ Covτ (S) the restriction map

F (S) → F (U)

is bijective (resp. injective).

Definition 9.4.3. One defines Sh(Cτ ) as the full subcategory of PSh(C) whose objects
are τ -sheaves.
60 BRUNO KLINGLER

9.4.3. Sheafification. Let Cτ be a site.

Definition 9.4.4. Let F ∈ PSh(C) and S ∈ C. Define
F + (S) := colimU ∈Covτ (S) F (U) .
Lemma 9.4.5. For any F ∈ PSh(C), F + ∈ PSh(C).
Proof. Let f : T → S ∈ C and U ∈ Covτ (S). Taking the colimit on Covτ (S) of the
arrows
F (U) → F (f ∗ (U)) → colimV∈Covτ (T ) F (V) = F + (T )
f∗
defines a map F + (S) → F + (T ) . 
Lemma 9.4.6. For any F ∈ PSh(C) the presheaf F+ is τ -separated.
One has a natural morphism of functors F → F + in PSh(C) hence a morphism of
functors
HomPSh(C) (F + , ·) → HomPSh(C) (F, ·) .
Lemma 9.4.7. If G ∈ Shτ (C) then HomPSh(C) (F + , G) ' HomPSh(C) (F, G). In partic-
ular if F is a sheaf one has a canonical isomorphism F ' F + .
Lemma 9.4.8. If F is separated then F + is a sheaf.
Definition 9.4.9. One defines the τ -sheafification F ] ∈ Sh(Cτ ) of F ∈ PSh(C) as
F → F + → F ++ =: F ] .
Lemma 9.4.10. One has a natural adjunction
/
·] : PSh(C) o Sh(Cτ ) : i .

9.4.4. Properties of sheafification. As the sheafification functor ·] has a right adjoint it

commutes with all colimits. In particular: for any family (Fi )i∈I of Sh(Cτ ),
colimSh(Cτ ) Fi = (colimPSh(C) Fi )] .
In the category Sets the filtered colimits commute with finite limits. As the functor
·+ is defined using filtered colimits it preserves small limits. In particular it preserves
algebraic structures: the sheafification of an abelian presheaf is an abelian sheaf, etc...
9.4.5. Sheaves and pre-topologies. Suppose the site Cτ is defined by a pre-topology given
by covering families (Ui → X)i∈I . Let U be a covering sieve of X ∈ C generated by a
covering family (Ui → X)i∈I . Then for all F ∈ PSh(C) the following sequence of sets is
exact:
/
Q // Q
F (U) i∈I F (Ui ) i,j F (Ui ×X Uj )) .

The presheaf F is a sheaf if and only if the map F → F + is an isomorphism. Hence it

is enough that for any object X there exists a cofinal set of covering sieves U of Covτ (X)
such that the natural map F (X) → F (U) is an isomorphism. Hence F is a τ -sheaf if
and only if for any object X ∈ C and any covering family (Ui → X)i∈I the following
sequence of sets is exact:

/
Q // Q
F (X) i∈I F (Ui ) i,j F (Ui ×X Uj )) .
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 61

Exercice 9.4.11. Let TG be the site defined in Example 9.3.10. Show that Sh(TG ) is
naturally equivalent to the category of G-sets.
9.5. The abelian category of abelian sheaves; cohomology. In this section, for
simplicity of notations we denote by the same symbol a site and its underlying category.
Theorem 9.5.1. Let C be a site. The category Ab(C) is Abelian. Moreover:
(1) if ϕ : F → G is a morphism in Ab(C) then ker ϕ = ker i(ϕ) and Coker ϕ =
(Coker i(ϕ))] .
(2) A sequence F → G → H in Ab(C) is exact in G if and only if for any U ∈ C and
any section s ∈ G(U ) whose image in H(U ) is zero, there exists (Ui → U )i∈I ∈
Cov(U ) such that s|Ui lies in the image of F (Ui ) → G(Ui ).
Proof. We first state the following lemma in categorical algebra, whose proof is left to
the reader:
/
Lemma 9.5.2. Let b : B o A : a be an adjoint pair of categories. Suppose that:
(i) A, B are additive and a, b are additive functors.
(ii) B is abelian and b is left exact (i.e. commutes with finite limits).
(iii) ba = IdA .
Then A is abelian and if ψ : A1 → A2 ∈ A then ker ψ = b(ker(aψ)) and Coker ψ =
b(Coker (aψ)).
/
Applying this lemma to ·] : PAb(C) o Ab(C) : i we obtain that Sh(C) is abelian
and the description of Coker ϕ. For ker ϕ: notice that the kernel is a finite limite
and ·] commutes with finite limits hence the result. This finishes the proof of (1) in
Theorem 9.5.1. The assertion (2) follows immediately as Im = ker ◦ Coker . 
We state the following general result without proof:
Theorem 9.5.3. Let C be a site. The Abelian category Ab(Cτ ) has enough injectives.
Definition 9.5.4. Let Cτ be a site. Let X ∈ C and F ∈ Ab(Cτ ). One defines the coho-
mology groups of F on X as the right-derived functors of the functor of global sections
H 0 (X, ·) : Ab(Cτ ) → Ab:

∀ X ∈ C, ∀F ∈ Ab(Cτ ), H p (X, F ) := Rp H 0 (X, ·)(F ) = H p (H 0 (X, I • )) ,

where F → I • is an injective resolution in Ab(Cτ ).
9.6. Functoriality. Let u : C → D be a functor between categories. It induces
canonically a functor:
up : PSh(D) → PSh(C)
F 7→ F ◦ u .
Suppose now that C and D are sites. We would like up to map sheaves to sheaves.
Definition 9.6.1. A functor u : C → D between two sites is continuous if it preserves
coverings and fiber products: for all (Vi → V )i∈I ∈ Cov(C) then
(1) (u(Vi ) → u(V ))i∈I ∈ Cov(D).
(2) ∀ T → V ∈ C then u(T ×V Vi ) ' u(T ) ×u(V ) u(Vi ).
62 BRUNO KLINGLER

This definition is tailored so that one obtains the:

Lemma 9.6.2. Let u : C → D be a continuous functor between sites. If F ∈ Sh(D) then

up F ∈ Sh(C).

Definition 9.6.3. We denote by us : Sh(D) → Sh(C) the functor deduced from up .

On the other hand the functor up always has a left adjoint

up : PSh(C) → PSh(D)

defined as
(up F )(V ) = colimI op FV ,
V

where:
- IV is the category whose objects are pairs (U, ϕ), U ∈ C, ϕ : V → u(U ) and the
morphisms between such pairs are the obvious ones.
- FV : IVop → Sets is the functor associating F (U ) to an object (U, ϕ) of IVop .

Lemma 9.6.4. The functor

us : Sh(C) → Sh(D)
G 7→ (up G)]

is left adjoint to us .

Definition 9.6.5. A morphism of sites f : D → C is a continuous functor u : C → D

(notice the inverse direction!) such that us : Sh(C) → Sh(D) is left exact (hence exact
as it has a right adjoint). We write
/
f −1 := us : Sh(C) o Sh(D) : us =: f∗ .

9.6.1. Digression on Topoi.

Definition 9.6.6. A topos is a category Sh(C) for some site C. A morphism of topoi
from Sh(D) to Sh(C) is an adjoint pair
/
f −1 : Sh(C) o Sh(D) : f∗ .

such that f −1 is left exact (hence exact).

Example 9.6.7. C = {pt} with one object, one morphism, one covering. Then Sh({pt}) =
Sets.

Remark 9.6.8. If f : D → C is a morphism of sites then

/
f −1 := us : Sh(C) o Sh(D) : us =: f∗

is a morphism of topoi.
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 63

10. Sheaves on schemes; fpqc sheaves

10.1. Cohomology of sheaves on schemes. Let τ ∈ {Zar, ét, smooth, fppf} and
(Sch/S)τ the corresponding site. If X ∈ Sch/S and F ∈ Ab((Sch/S)τ ) Definition 9.5.4
particularizes to define H • (Xτ , F ) as the right derived functors of the functor of global
sections H 0 (X, ·) : Ab((Sch/S)τ ) → Ab.
Let τ ∈ {étale, Zariski}. It is obvious that if f : X → S and g : Y → S are open
immersions then any S-morphism from X to Y is an open immersion. It follows from
their very definition that étale morphisms satisfy a similar property:
Lemma 10.1.1. If f : X → S and g : Y → S are étale morphisms then any S-morphism
from X to Y is étale.
Hence for τ ∈ {étale, Zariski} we can consider the restriction of τ to the subcategory
of Sch/S whose objects are the étale maps f : X → S, resp. the open immersions: this
still defines a site, denoted Sτ and called the small τ -site of S.
If X ∈ Sτ , then any F ∈ Ab((Sch/S)τ is in particular an element of Ab(Sτ ). In
particular, while we defined the cohomology groups H • (X, F ) in terms of the big site
of S, an alternative definition would be to consider the derived functors of H 0 (X, ·) :
Ab(Sτ ) → Ab. However one can show that these two definitions give canonically
isomorphic groups.

10.2. A criterion to be a sheaf on (Sch/S)τ . We have the following continuous

functors of sites:
id id id id
(20) (Sch/S)Zar → (Sch/S)ét → (Sch/S)lisse → (Sch/S)fppf → (Sch/S)fpqc .
Hence any τ -sheaf, τ ∈ { fpqc, fppf, lisse, étale, Zariski} is a Zariski sheaf. The fol-
lowing lemma characterizes τ -sheaves among Zariski sheaves.
Lemma 10.2.1. Let τ ∈ { fpqc, fppf, lisse, étale, Zariski} and let C = (Sch/S)τ , or Sτ .
A presheaf F on C is a sheaf if and only if:
(i) it is a Zariski-sheaf.
(ii) For any V → U ∈ CovC (U ), with U and V affine in C, the sequence
/ F (V ) //
F (U ) F (V ×U V )
is exact.
` Q
Proof. The fact that F is a Zariski sheaf implies that F ( Ui ) = F (Ui ). Hence the
sheaf condition for the covering
` (Ui → U )i∈I ∈ CovC (U ) is equivalent to the sheaf
condition for the covering i∈I Ui → U as
a a a
( Ui ) ×U ( Uj ) = Ui ×U Uj .
i,j

is satisfied for coverings (Ui → U )i∈I

This implies in particular that the sheaf condition `
such that |I| is finite and each Ui is affine for then i∈I Ui is affine.
Let f : U 0 → U ∈ CovC (U ). Choose an open affine covering U = ∪i Ui and write
−1
f (Ui ) = ∪k Uik 0 a finite open affine covering (this is possible as f is quasi-compact).
64 BRUNO KLINGLER

Hence U 0 = ∪i,k Uik

0 is an open affine covering. Consider the commutative diagram:

/ F (U 0 ) //
F (U ) F (U 0 ×U U 0 )

  
Q /
Q 0 // Q 0 × U0 )
i F (Ui ) i F (Uik ) i,k,l F (Uik U il

   
/ 0 0
Q Q
i,j F (Ui ×U Uj ) i,j,k,l F (Uik ×U Ujl )

The two columns on the left are exact as F is a Zariski-sheaf while the second row is
exact as all the schemes considered are affine and for each i the sets of corresponding
indices j and k are finite. It follows first that F (U ) ,→ F (U 0 ) (i.e. F is separated), hence
the row on the bottom is injective and F is a sheaf by diagram chasing. 
10.3. fpqc sheaves and faithfully flat descent. Although our main object of in-
terest are étale sheaves, we start by studying a few fpqc sheaves as any such sheaf is in
particular an étale sheaf by eq. (20).
Lemma 10.3.1. Let S ∈ Sch and F ∈ QCoh(S). Then the presheaf
F : Sch/S → Ab
[f : T → S] 7→ Γ(T, f ∗ F )
is an fpqc sheaf, in particular an étale sheaf.
Proof. That F is a Zariski sheaf is a classical fact. Thanks to Lemma 10.2.1 we are
reduced to showing that for any A → B a faithfully flat ring morphism and writing the
coherent sheaf F as M̃ on Spec A, the sequence of A-module
(21) 0 → M → B ⊗A M → B ⊗A ⊗B ⊗A M
is exact. This follows from the results below on faithfully flat descent. 
Grothendieck’s topologies appeared originally as a residue of his theory of descent,
whose goal is to define locally global objects via a glueing procedure. Let us develop a
bit the problem of descent for quasi-coherent sheaves. Let X ∈ Sch/S and let U ⊂ hX be
a covering sieve for a topology τ on (Sch/S). A quasi-coherent module “given U-locally”
EU is the following set of data:
(a) for all U ∈ U, a module EU ∈ QCoh(U ).
∼
(b) for all U, V ∈ U and any X-morphism ϕ : V → U , an isomorphism ρϕ : EV →
ϕ∗ EU , such that
ψ ϕ
(c) for all W → V → U the diagram
ρψ◦ϕ
EW G / ψ ∗ ϕ∗ EU
GG 9
GG ss
G sss
ρψ GGG ss ∗
# ss ψ ρϕ
ψ ∗ EV
commutes.
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 65

Of course any E ∈ QCoh(X) defines by pull-back a quasi-coherent module EU given

U-locally. Descent theory deals with the converse problem: does every EU comes from
some E ∈ QCoh(X)? This is true by the very definition of an OX -module if τ = Zar,
but not very useful as many modules are naturally given locally for coverings defined
only in finer topologies. Notice that the local EU ’s naturally form a category QCoh(U).
The first main result of descent theory is the following:
Theorem 10.3.2. Let (Ui → X)i∈I be an fpqc-covering family and let U be the sieve
generated by (Ui → X)i∈I . Then the functor
ψ : QCoh(X) → QCoh(U)
E 7→ EU
is an equivalence of categories.
Proof. As in Lemma 10.2.1 one easily reduces to the case of a covering defined by a
faithfully flat morphism U → X, U and X both affine.
Notice that the statement is obvious if U → X admits a section. In this case X ∈ U
hence for any EU ∈ QCoh(U) the module E := EX ∈ QCoh(X) is well-defined and one
easily checks that EU ' ψ(E). We will reduce ourselves to this case.
Let EU ∈ QCoh(U). One easily checks that the datum of EU is equivalent to the
datum of a diagram
// 00 /// E 000
E0 E
cartesian over
U oo U ×X U ooo U ×X U ×X U
i.e. in terms of modules:
// /// E 000
E0 E 00
cartesian over
B // B ⊗A B /// B ⊗A B ⊗A B

(by cartesian we mean that each natural map ∂i : E 0 ⊗B,∂i (B ⊗A B) → E 00 is an

isomorphism and similarly for the other maps). In this language the functor ψ can be
described as
ψ : Mod(A) → QCoh(U)
// ///
 
E 7→ M ⊗A B M ⊗A B ⊗A B M ⊗A B ⊗A B ⊗A B .

It admits a natural right-adjoint functor, which associates to E 0 // E 00 /// E 000

// 00
the A-module ker(E 0 E ).
We are thus reduced to prove that the two adjunction arrows
//
(22) E → ker( E ⊗A B E ⊗A B ⊗A B )
and
//
(23) ker( E 0 E 00 ) ⊗A B → E 0
are isomorphisms.
Remark 10.3.3. Notice that eq. (22) being an isomorphism is equivalent to our original
claim that the sequence eq. (21) is exact.
66 BRUNO KLINGLER

It is enough to prove the result after a faithfully flat base change A → A0 , as faithfully
flat maps preserves exact sequences. Taking A0 = B the structure map A → B becomes
B → B ⊗A B mapping b to b ⊗ 1, which admits a section b ⊗ b0 7→ bb0 . Hence we are
done thanks to the previous case.

Remark 10.3.4. More generally given A → B a ring morphism one can consider the
complex
(B/A)• : B //
B ⊗A B ///
B ⊗A B ⊗A B //// · · ·

Lemma 10.3.5. If A → B is faithfully flat then for any A-module M the complex
(B/A)• ⊗A M is acyclic and H 0 ((B/A)• ⊗A M ) = M .
10.4. The fpqc sheaf defined by a scheme. Another kind of fpqc sheaf is provided
by the following:
Lemma 10.4.1. Let X ∈ Sch/S. Then hX ∈ Sh((Sch/S)fpqc ) (hence also hX ∈
Sh((Sch/S)ét ).
Proof. Clearly hX is a Zariski sheaf. We have to show that if A → B is faithfully flat
then
/ X(B) //
X(A) X(B ⊗A B)
is exact. One easily reduces to the case X = Spec C is affine, in which case we have to
show that
HomA−alg (C, A) / HomA−alg (C, B) // HomA (C, B ⊗A B)

is exact. This follows immediately from eq. (21). 

Remark 10.4.2. One can show that on any category there exists a finest topology such
that all representable presheaves are sheaves: the canonical topology. Hence the fpqc
topology (and a fortiori the étale topology) is coarser than the canonical topology: one
says it is subcanonical.
Remark 10.4.3. Let S = Spec k and F ∈ Sh(Sét ). Let E := colimF (Ki ), where Ki /k is
a finite separable extension.
` The set E has a continuous G := Gal(k s /k) action, hence
can be written E = Ei where ` Ei is finite, Ei = G/Hi with Hi an open subgroup of
G. Then F is represented by Ui where Ui = Spec Ki , Ki := (k s )Hi . Hence F is an
ind-object in (Spec k)ét .
Remark 10.4.4. On C = (Sch/S)τ or Sτ , the sheaf (OS )τ associated to the quasi-coherent
sheaf OS coincide with the sheaf Ga,S hence is representable. The presheaf OS∗ of OS is
easily seen to be an fpqc subsheaf which coincide with Gm,S .
10.4.1. Roots of unity. Let n be a positive integer. Define the fpqc sheaf
(·)n
µn,S := ker(Gm,S → Gm,S ) .
Proposition 10.4.5. If n is invertible on S then the sequence of Ab(Sét )
(·)n
0 → µn,S → Gm,S → Gm,S → 0
is exact.
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 67

Remark 10.4.6. This is not true for the Zariski topology! Usually an element of Γ(U, OU∗ )
is not Zariski-locally a n-th power.
Proof. Let U ∈ Sét and a ∈ G(U ) = Γ(U, OU∗ ). The integer n is invertible on S hence
on U , thus T n − a is separable over OU . By the Jacobian criterion this implies that
U 0 := Spec OU [T ]/(T n − a) is étale over U . As U 0 → U is surjective this is an étale
covering family. As a admits an n-th root on U 0 we conclude. 
10.5. Constant sheaf. Let C be an abelian group. The Zariski sheafification of the
constant presheaf C on SZar is the sheaf
CS : U 7→ C π0 (U ) .
As it is representable by the group scheme S × C this is also an fpqc-sheaf (hence an
étale sheaf).
We will be especially interested in (Z/nZ)S .

11. Etale sheaves

We now turn to a more detailed study of étale sheaves.
11.1. Neighborhoods and stalks.
Definition 11.1.1. Let X be a scheme and x a point of X.
(i) An étale neighborhood of (X, x) is an étale morphism (U, u) → (X, x).
(ii) If x : Spec k s → X is a geometric point of X of image x, an étale neighborhood
of x is a commutative diagram
;U
vvv
u vv
vv ϕ
vv 
Spec k s / X,
x

where ϕ : (U, u) → (X, x) is an étale neighborhood of (X, x). One writes (U, u) →
(X, x).
(iii) Morphisms of étale neigborhoods are defined in an obvious way.
Definition 11.1.2. Let F ∈ Sh(Xét ). The fiber of F at x is the set
Fx := colim(U,u) F (U ) ,
where the colimit is taken over the cofiltered category of étale neighborhoods of (X, x).
Proposition 11.1.3. Let X be a scheme.
(i) A morphism f : F → G ∈ Sh(Xét ) is a monomorphism (resp. an epimorphism)
if and only if for any geometric point x → X the morphism fx : Fx → Gx is a
monomorphism (resp. an epimorphism).
(ii) A sequence
0 → F → G → H → 0 ∈ Ab(Xét )
is exact if and only if for any geometric point x of X the sequence of abelian
groups
0 → Fx → Gx → Hx → 0
68 BRUNO KLINGLER

is exact.
Proof. Let us prove the abelian case. First the surjectivity.
Suppose that F  G. Consider the exact sequence defining the cokernel Λ:
Fx → Gx → Λ → 0 .
Let us define Λx ∈ Sh(Xét ) by
M
Λx (U ) := Λ,
HomX (x,U )

it obviously satisfies the adjunction

HomSh(Xét ) (F, Λx ) = HomAb (Fx , Λ) .
If x is closed in X this is the skyscraper sheaf at x with value Λ. The morphism Gx → Λ
defines a morphism of sheaves G → Λx . The composite
F → G → Λx
is zero as it corresponds to the composite Fx → Λ. If Λ 6= 0 this contradicts the
assumption F  G.
Conversely, suppose that Fx → Gx is surjective for all x. Let U → X ∈ Xét and
u → U a geometric point with image x → X. Clearly Fu ' Fx hence we can assume
that U = X. Let s ∈ G(X). Fix x → X a geometric point. As Fx  Gx there exists
an étale neighborhood (V, v) → (X, x) such that s|V ∈ Im (F (V ) → G(V ). Arguing this
way for sufficiently many x one can cover X by the union of the V ’s. Hence the result
by Theorem 9.5.1(2).
For the injectivity: a colimit of exact sequences is exact hence
0 → F (U ) → G(U )
implies
0 → F x → Gx .

11.2. Strict localisation.
Definition 11.2.1. The strict localization of X at x is the ring OX,x .
As any Zariski neigborhood of x is also an étale neighborhood of x one obtains a
morphism OX,x → OX,x . Similarly for any étale neighborhood (U, u) → (X, x) one
obtains a ring morphism OU,u → OX,x and clearly OX,x = colim(U,u) OU,u .
sh of O
Lemma 11.2.2. The ring OX,x is the strict henselianisation OX,x X,x .

Let us recall a few facts on henselian rings.

Definition 11.2.3. Let (R, m, κ) be a local ring.
(i) The ring R is said to be henselian if for any monic f ∈ R[t] and ao ∈ κ such
that f (ao ) = 0 and f 0 (a0 ) 6= 0 then there exists a unique a ∈ R with image a0
such that f (a0 ) = 0.
(ii) The ring R is said to be strictly henselian if moreover κ is separably closed.
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 69

Lemma 11.2.4. The following statement are equivalent:

(1) the ring R is henselian.
(2) if f ∈ R[t] is monic and f = g · h with g, h ∈ κ[T ] monic satisfying g ∧ h = 1,
there exists unique relatively prime monic g, h ∈ R[t] with image g and h such
that f = gh.
(3) any finite extension of R is a product of local rings.
(4) for any étale morphism R → S and q ∈ Spec S over m with κ(q) = κ there exists
a section τ : S → R of R → S.
Example 11.2.5. Any complete local ring is henselian.
Lemma 11.2.6. Let (R, m, κ) be an henselian local ring. Then reduction mod m estab-
lishes an equivalence of categories between the category of finite étale extensions R → S
and the category of finite étale extensions κ → S.
Definition 11.2.7. Let R be a local ring. A local homomorphism R → Rh is called the
henselianization of R if it is universal among henselian extensions:
RC /S
CC O
CC
CC
!
Rh .
Similarly for the strict henselianization.
11.3. Direct image and inverse image, the étale case. Let f : Y → X be a
morphism of schemes. Let u : Xét → Yét be the corresponding functor, in fact one easily
checks this is a morphism of sites. Hence:
/
f −1 (:= us ) : Sh(Xét ) o Sh(Yét ) : (us =:)f∗
One has a canonical morphism (f∗ F )x → Fy which is neither injective nor surjective
in general.
11.3.1. Direct image.
(
 if x ∈ U,
Lemma 11.3.1. (a) If j : U o / X then (j∗ F )x = Fx
? otherwise.
(

(b) If i : Z / / X then (i∗ F )x = Fx if x ∈ Z,
0 otherwise.
L d(y)
(c) Let f : Y → X be a finite morphism. Then (f∗ F )x = y7→x Fy where d(y) is
the separable degree of the extension of residues fields κ(y)/κ(x).
Proof. For (a): by definition (j∗ F )x = colim(V,v) (j∗ F )(V ) where (V, v) ranges through
the étale neighborhoods of (X, x). If x ∈ U the image of such sufficiently small étale
neighborhoods is contained in U . Thus the étale neighborhoods of (U, x) are cofinal in
the étale neighborhoods of (X, x) hence (j∗ F )x = Fx in this case.
For (b): If x 6∈ Z the image in X of a sufficiently small étale neighborhood of (X, x)
does not meet Z hence (i∗ F )x = 0. If x ∈ Z it is enough to show that an étale
neigborhood of (Z, x) extends to an étale neighborhood of (X, x). Locally X = Spec A
70 BRUNO KLINGLER

and Z = Spec (A/a) with a an ideal in A. Let us write A = A/a and let A → B
be an étale ring homomorphism. Hence one can write B = (A[T ]/f (T ))(b) for some
b ∈ A[T ]/f (T ), where f [T ] ∈ A[T ] with f 0 [T ] invertible in B. Choose f (T ) ∈ A[T ]
lifting f and set B := (A[T ]/f (t))(b) . For an appropriate b lifting b, the extension
A → B is étale and extends A → B.
For (c): left as an exercice. 
Corollary 11.3.2. If f : Y → X is finite then f∗ : Ab(Yét ) → Ab(Xét ) is exact.
Proof. Check on stalks using Lemma 11.3.1(c). 
11.3.2. Inverse image. if f : X → Y then f p : PAb(Yét ) → PAb(Xét ) is defined by
(f p F )(U ) = colimV F (V )
where V ranges through the commutative diagrams
U /V

ét
 
X /Y
f

and f ∗ = (f p )] .
Remark 11.3.3. If f is étale then f ∗ is just the usual restriction functor.
Remark 11.3.4. If ix : x → X is a geometric point then i∗x F = Fx by definition.
Lemma 11.3.5. Let f : X → Y . Then (f ∗ F )x = Ff (x) .
Proof. Consider the commutative diagram:
ix
xA /X
AA
AA
A
if (x) AA 
f

Y .
Notice that (g ◦ f ) = f ∗ g ∗ by unicity of the left adjoint to (g ◦ f )∗ = g∗ f∗ . Hence:
(f ∗ F )x = i∗x (f ∗ F ) = if (x) F = Ff (x) .

Corollary 11.3.6. The functor f ∗ is exact.
Corollary 11.3.7. f∗ : Ab(Yét ) → Ab(Xét ) send injectives to injectives.
Proof. This is a formal consequence of the fact that f∗ admits a left adjoint functor
f ∗ which is left exact. Indeed let I be an injective in Ab(Yét ). Completing the solid
diagram

F B /G
BB
BB
BB
B 
f∗ I
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 71

is equivalent by adjunction to completing the solid diagram (the upper row remains
injective as f ∗ is left exact)

f ∗ F F / f ∗G
FF
FF
FF
FF 
#
I.
This follows from the injectivity of I. 
Remark 11.3.8. At this point one easily showa that Ab(Xét ) has sufficiently many in-
jectives (hence we prove in this particular case the Theorem 9.5.3 stated without proof
for the category of abelian sheaves on any site). Indeed consider the monomorphism
Y
F ,→ ix∗ i∗x F .
x→X

Choose a monomorphism i∗x F

,→ Ix in Ab Q
with Ix injective in Ab. This exists as Ab
has sufficiently many injectives. Thus F ,→ x→X ix∗ Ix and the term on the right is an
injective sheaf by the corollary above.
11.4. Extension by zero.
Lemma 11.4.1. Let j : U → X be an étale morphism (for example an open immersion).
Then j ∗ : Ab(Xét ) → Ab(Uét ) has a left adjoint j! : Ab(Uét ) → Ab(Xét ) which is exact
(in particular j ∗ maps injectives to injectives).
ϕ
Proof. Let F ∈ Ab(Uét ). For V → X define
M
F! (V ) = F (V ) .
α/
V B U
B j
ϕ
X

Notice that if j : U o / X is an open immersion then
(
F (V ) if ϕ(V ) ⊂ U,
F! (V ) =
0 otherwise .

Hence F! ∈ PAb(Xét ) and clearly F! is left adjoint to j p . Set j! F := (F! )] . If G ∈

Ab(Xét ) then
HomAb(Xét ) (j! F, G) = HomPAb(Xét ) (F! , G) = HomPAb(Uét ) (F, j p G) = HomAb(Uét ) (F, j ∗ G) .
This proves the existence of the left adjoint functor j! F .
One easily shows from the definition that
L
 u→U Fu if x ∈ j(U ) ,
(j! F )x = j(u)=x
0 otherwise .
This implies immediately that j! is exact. 
72 BRUNO KLINGLER

11.5. The fundamental triangle. Consider the following geometric situation:


U := X \ Z  ? _Z
j
o / X o /i .
Lemma 11.5.1. Let F ∈ Ab(Xét ). The following sequence of Ab(X)ét
0 → j! j ! F → F → i∗ i∗ F → 0
is exact (where we set j ! := j ∗ ).
Proof. Consider the corresponding stalks. If x ∈ U one obtains
id
0 → Fx → Fx → 0 → 0
which is obviously exact. If x ∈ Z then
id
0 → 0 → Fx → Fx → 0
which is also exact. 
Given F ∈ Ab(Xét ) we set
FU := j ∗ F ∈ Ab(Uét ),
FZ := i∗ F ∈ Ab(Zét ) .
By adjunction one obtains a canonical map F → j∗ j ∗ F = j∗ FU . Applying i∗ gives:
FZ → i∗ j∗ FU .
Proposition 11.5.2. Let us denote by T the category of triplets
φ
(FZ ∈ Ab(Zét ), FU ∈ Ab(Uét ), FZ → i∗ j∗ FU )
with the obvious morphisms. The functor
Ab(Xét ) → T
F 7→ (FZ , FU , FZ → i∗ j ∗ FU )
is an equivalence of categories.
φ
Proof. We construct an inverse functor as follows. Starting from (FZ , FU , FZ → i∗ j∗ FU )
let us define F̃ ∈ Ab(Xét ) as the cartesian product
F̃ / j∗ FU

 
i∗ FZ / i∗ i∗ j∗ FU .
i∗ ϕ

If now F ∈ Ab(Xét ) the natural maps F → j∗ FU and F → i∗ FZ defines a morphism

F → F̃ . We have to check this is an isomorphism. Hence we have to show that the
diagram
F / j∗ FU

 
i∗ FZ / i∗ i∗ j∗ FU
i∗ ϕ
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 73

is cartesian. As stalks at geometric points commute with fiber product and form a
conservative family we have to check that the corresponding diagrams of stalks are
Cartesian. For x ∈ U we obtain
Fx / Fx

 
0 / 0.
For x ∈ Z:
Fx / (j∗ j ∗ F )x

 
Fx / (j∗ j ∗ F )x .

Both squares are Cartesian hence we are done. 

Definition 11.5.3. Let F ∈ Ab(Xét ). If Y is any subscheme of X we say that F has
support contained in Y if Fx = 0 for any x 6∈ Y .

Corollary 11.5.4. Let Z / / X . Then
i

i∗ : Ab(Xét ) → Ab(Xét )
induces an equivalence of categories between Ab(Zét ) and the full subcategory of Ab(Zét )
of sheaves with support contained in Z.
Proof. Notice that F ∈ Ab(Xét ) has support contained in Z if and only if it is of the
form (FZ , 0, 0) in the description of Proposition 11.5.2. 
We summarize our results through the following diagram of adjunctions:
Ab(Uét )
O
j! j ! =j ∗ j∗

 Ab(Xét ) 
O
i∗ i∗ =i! i!
 Ab(Zét ) 

satisfying the following identities

∼
id → j ! j! ,
∼
j ∗ j∗ → id ,
∼
i∗ i∗ → id ,
∼
id → i! i! ,
j ∗ i∗ = 0 hence i! j∗ = i∗ j! = 0 ,
where we defined the functor of sections with support in Z:
i! : (FZ , FU , ϕ : FZ → i∗ j∗ FU ) 7→ ker ϕ .
74 BRUNO KLINGLER

12. Etale cohomology

Let X be a scheme. Consider the left exact functor
Γ(X, ·) := HomAb(Xét ) (ZX , ·) : Ab(Xét ) → Ab
F 7→ Γ(X, F ) = HomAb(Xét ) (ZX , F ) .
One considers its right derived functors
RΓ(X, ·) = R HomAb(Xét ) (ZX , ·) : D+ Ab(Xét ) → D+ Ab .
and define
H i (Xét , F ) := Ri HomAb(Xét ) (ZX , ·) .

12.1. Cohomology with support. Let Z / / X and F ∈ Ab(Xét ). Define U :=
i
X \ Z and
ΓZ (X, F ) := ker(Γ(X, F ) → Γ(U, F|U ))
the group of sections of F with support in Z. The functor ΓX (X, ·) is clearly left exact,
hence we can define its right derived functors.
Definition 12.1.1. We define the cohomology groups of F with support in Z as
HZr (X, F ) := Rr ΓZ (X, F ) .
Theorem 12.1.2. The following long sequence of abelian groups is exact:
· · · → HZr (X, F ) → H r (X, F ) → H r (Z, F ) → HZr+1 (X, F ) → · · ·
Proof. Consider the Ext-long exact sequence obtained by applying HomAb(Xét ) (·, F ) to
the exact sequence of sheaves provided by Lemma 11.5.1:
0 → j! j ! ZX → ZX → i∗ i∗ ZX → 0 .
Notice that
HomAb(Xét ) (j! j ! ZX , G) = HomAb(Uét ) (j ∗ ZX , j ∗ G) = G(U ) ,
hence by considering an injective resolution F ' I • :
ExtrAb(Xét ) (j! j ! ZX , F ) = H r (Uét , F|U ) .

Looking at the beginning of the Ext long exact sequence:

0 → HomAb(Xét ) (i∗ i∗ ZX , F ) → F (X) → F (U )

is exact hence the left hand term is necessarily ΓZ (X, F ). Applying to an injective
resolution of F we deduce:
ExtrAb(Xét ) (i∗ i∗ ZX , F ) ' HZr (Xét , F ) .

The result follows. 

 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 75

12.2. Nisnevich excision. The excision theorem for usual cohomology says that
cohomology with support in Z depends only on a neighborhood of Z in X. Similarly:
Theorem 12.2.1. Let
X0
f

 /X o o ? _ U := X \ Z
Z /
i j

with f an étale map such that f|f −1 (Z)red : f −1 (Z)red ' Z (such an étale covering of X
is called an elementary Nisnevich covering). Then:
0
HZr (Xét , F ) ' HZr (Xét , f ∗F ) .
Proof. We proved that f ∗ is exact. Moreover as f is étale f ∗ preserve injectives by
Lemma 11.4.1. Hence it is enough to prove the result for r = 0.
Consider the commutative diagram:

0 / ΓZ 0 (X 0 , f ∗ F ) / Γ(X 0 , f ∗ F ) / Γ(U 0 , f ∗ F )
O O O
ϕ

0 / ΓZ (X, F ) / Γ(X, F ) / Γ(U, F ) .

We have to show that ϕ is an isomorphism.
For the injectivity: suppose that s ∈ ΓX (X, F ) is mapped to zero in ΓZ 0 (X 0 , f ∗ F ).
Hence s, seen as an element of Γ(X, F ), maps to zero in Γ(U, F ) and Γ(X 0 , f ∗ F ). But
f
(X 0 → X, U → X) is an étale covering of X and F is a sheaf hence s = 0.
For the surjectivity: let s0 ∈ ΓZ 0 (X 0 , f ∗ F ). One easily checks that the pair (s, o) ∈
Γ(X 0 , f ∗ F ) × Γ(U, F ) maps to zero on intersections hence comes from s ∈ Γ(X, F ) as F
is an étale sheaf. 

13. Čech cohomology and étale cohomology of quasi-coherent sheaves

The goal of this section is to prove:
Theorem 13.0.1. Let S be a scheme and F ∈ QCoh(S). Then
H p (SZar , F ) = H p (Sét , F ) = H p (Sfpqc , F ) .
The basic tool will be Čech cohomology, a cohomology theory for presheaves.
13.1. Čech cohomology for coverings.
Definition 13.1.1. Let C be a category and U = (Ui → U )i∈I any family of morphism
to U ∈ C. Let F ∈ PAb(C).
The Čech complex of F with respect to U is the complex Č • (U, F ) ∈ D+ Ab:
Y Y Y
Č • (U, F ) : F (Ui0 ) → F (Ui0 ,i1 ) → F (Ui0 ,i1 ,i2 ) → · · ·
i0 i0 ,i1 i0 ,i1 ,i2
where Ui0 ,...,in := Ui0 ×U · · · ×U Uin .
The Čech cohomology of F on U is Ȟ p (U, F ) := H p (Č • (U, F )).
76 BRUNO KLINGLER

Proposition 13.1.2. Ȟ • (U, ·) : PAb(C) → Ab is a universal δ-functor.

Proof. We first show that Ȟ • : (U, ·) : is a δ-functor. Given an exact sequence of
presheaves
0 → F1 → F2 → F3 → 0
one immediately obtains that the sequence of complexes
0 → Č • (U, F1 ) → Č • (U, F2 ) → Č • (U, F3 ) → 0
is exact, hence we obtain the required long exact sequence
· · · → Ȟ r (U, F1 ) → Ȟ r (U, F2 ) → Ȟ r (U, F3 ) → Ȟ r+1 (U, F1 ) → · · · .
Recall the universality means that given any other δ-functor T • : PAb(C) → Ab and
any morphism Ȟ 0 (U, ·) → T 0 , there exist compatible morphisms Ȟ • (U, ·) → T • . We
have to show that for any i > 0 the functor Ȟ i (U, ·) is effaceable i.e. for any F ∈ PAb(C)
there exists a monomorphism F ,→ I with Ȟ i (U, I) = 0.
Given V ∈ C we denote by ZV ∈ PAb(C) the presheaf defined by ZV (W ) = Z[HomC (W, V )].
In other words Z• is the left adjoint functor to the inclusion PAb(C) → PSh(C) and
ZV := ZhV . Notice that:
 
Y Y
Č • (U, F ) =  HomPAb(C) (ZUi , F ) → HomPAb(C) (ZUi ,i , F ) → · · · 
0 0 1
i0 i0 ,i1
  
M M M
= HomPAb(C)  ZUi0 ← ZUi0 ,i1 ← ZUi0 ,i1 ,i2 ← · · ·  , F 
i0 i0 ,i1 i0 ,i1 ,i2
  
= HomPAb(C) Z`i Ui 0 ← Z`i Ui0 ,i1 ← Z`i Ui0 ,i1 ,i2 ← ··· ,F .
0 0 ,i1 0 ,i1 ,i2

Lemma 13.1.3. The complex of PAb(C)

 
Z•U : Z`i Ui0 ← Z`i ,i Ui0 ,i1 ← Z`i Ui0 ,i1 ,i2 ← ···
0 0 1 0 ,i1 ,i2

is exact in positive degrees.

Proof. Let V ∈ C. Then
 " #   
a a
Z•U (V ) = Z HomC (V, Ui0 ) ← Z  HomC (V, Ui0 ,i1 ) ← · · · 
i0 i0 ,i1
 " #   
M a a
= Z Homϕ (V, Ui0 ) ← Z  Homϕ (V, Ui0 ) × Homϕ (V, Ui1 ) ← · · · 
ϕ:V →U i0 i0 ,i1

V AA/ Ui
( )
A 
`
where Homϕ (V, Ui ) = ϕ
. Set Sϕ := i Homϕ (V, Ui ). Thus
U
M
Z•U (V ) = (Z[Sϕ ] ← Z[Sϕ × Sϕ ] ← Z[Sϕ × Sϕ × Sϕ ] ← · · · ) .
ϕ:V →U
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 77

Hence it is enough to show that for any set E the complex of abelian groups
Z[E] ← Z[E × E] ← Z[E × E × E] ← · · ·
is exact in positive degrees. This follows immediately from the contractibility of the
simplicial set ƥ . 
Lemma 13.1.4. If I ∈ PAb(C) is injective then Ȟ p (U, I) = 0 for any p > 0.
Proof. We showed that Ȟ p (U, I) = H p (HomPAb(C) (Z•U , I)). As Z•U is exact in positive
degree by the previous lemma and HomPAb(C) (·, I) is exact as I is injective, the result
follows. 
This finishes the proof of Proposition 13.1.2 
Theorem 13.1.5. Ȟ p (U, ·) = Rp Ȟ 0 (U, ·) in PAb(C).
Proof. Both functors are universal δ-functors and coincide in degree zero. 
Remark 13.1.6. Up to now we did not use the topology on C.
13.2. Čech to cohomology spectral sequence.
Theorem 13.2.1. Let C be a site. Let U ∈ C, U ∈ Cov(U ) and F ∈ Ab(C). There is
a natural spectral sequence, called the Čech to cohomology spectral sequence:
E2p,q = Ȟ p (U, Hq (F )) ⇒ H p+q (U, F ) ,
where Hq (F ) : U 7→ H q (U, F ) ∈ PAb(C).
Proof. Recall the following:
Theorem 13.2.2. (Grothendieck’s spectral sequence for composition of functors) Let
A, B, C be Abelian categories. Assume that A and B have enough injectives. Let F :
A → B and G : B → C be left exact functors and assume that F I is G-acyclic for any
injective I ∈ A. There is a canonical spectral sequence
E2p,q = Rp G(Rq F (A)) ⇒ Rp+q (G ◦ F )(A) .
We apply this result to
i / PAb(C) Ȟ 0 /4 Ab ,
Ab(C)
H0

noticing that i : Ab(C) → PAb(C) maps injectives to injectives (indeed the functor i
admits as left adjoint functor the sheafification functor ·] which is left exact) and that
(Rq iF )(V ) = Hq (F )(V ) by definition. 
Lemma 13.2.3. (locality of cohomology) Let C be a site and F ∈ Ab(C). Let U ∈ C
and ξ ∈ H p (U, F ) for some p > 0. There exists a covering family (Ui → U )i∈I of U such
that ξ|Ui = 0 for any i ∈ I.
Proof. Choose an injective resolution F ' I • in Ab(C) and ξ˜ ∈ I p (U ) lifting ξ. In
dp−1 dp
particular dp ξ˜ = 0. As the sequence I p−1 → I p → I p+1 is exact there exists a covering
family (Ui → U )i∈I of U and element ξi ∈ I p−1 (Ui ) such that ξ˜|Ui = dp−1 ξi . Hence
ξ|Ui = 0. 
78 BRUNO KLINGLER

13.3. Proof of Theorem 13.0.1. We only sketch the proof.

The result for p = 0 is equivalent to the fact that F is an fpqc sheaf.
For p > 0, the main step consists in proving the result for S affine: we want to show
in this case that H p (Sfpqc , F ) = 0 for any p > 0. The proof is by induction on p.
For p = 1: let ξ ∈ Hfpqc 1 (S, F ). By Lemma 13.2.3 there exists an fpqc-covering family

(Ui → S)i∈I such that ξ|Ui = 0 for any i ∈ I. Without loss of generality we can assume
that each U` r
i is affine (in particular H (Ui , F ) = 0 for any r > 0) and I is finite. Let
U = (V := i Ui → S). Hence ξ comes, via the Čech to Cohomology spectral sequence,
of a class ξˇ ∈ Ȟ 1 (U, F ). Write S = Spec A and V = Spec B, F = M̃ . One easily checks
that
Č • (U, F ) = (B/A)• ⊗A M ,
hence Ȟ 1 (U, F ) = 0 by faithfully flat descent. Hence ξˇ = 0 and ξ = 0.
For p > 1: Notice that each Ui0 ,...,in is affine hence E2i,j = Ȟ i (U, Hj (F )) = 0 for
0 < j < p by induction hypothesis and the same argument provides the induction.
2

13.4. Other applications of Čech cohomology.

13.4.1. Čech cohomology at the colimit. We continue with the notations of Theo-
rem 13.2.1. Let V = (Vj → U )j∈J be a refinement of U = (Ui → U )i∈I , meaning
that there exists a map τ : J → I such that for every j ∈ J one has a factorization
Vj D_ _ _/ Uτ (j)
DD
DD
DD
D! 
U.
This gives rise to a canonical restriction map:
ρV,U : Ȟ • (U, F ) → Ȟ • (V, F ) ,
and one defines:
Ȟ • (U, F ) = colimU Ȟ • (U, F )
where U ranges through all coverings of U . Taking the colimit of the Čech to Cohomology
spectral sequences for U leads to the spectral sequence:
(24) E2p,q = Ȟ p (U, Hq (F)) ⇒ H p+q (U, F ) .
In this language the locality of cohomology (Lemma 13.2.3) can be rewritten as:
Ȟ 0 (U, Hq (F )) = 0 ∀ q > 0 .
In particular the spectral sequence eq. (24) looks like:
0 NNN? ? ···
NNN
&
0 ? ? ···
E2p,q =
0 ? ? ···
? ? ? ···
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 79

hence
Ȟ 0 (U, F ) ' H 0 (U, F ) ,
Ȟ 1 (U, F ) ' H 1 (U, F ) ,
and the following sequence is exact:
0 → Ȟ 2 (U, F ) → H 2 (U, F ) → Ȟ 1 (U, H1 (F )) → Ȟ 3 (U, F ) → H 3 (U, F ) .
13.4.2. Mayer-Vietoris exact sequence in étale cohomology.
Lemma 13.4.1. (Mayer-Vietoris) Let U = U0 ∪ U1 be a Zariski-open decomposition of
U . Then for any F ∈ Ab(U ) the following long exact sequence holds:
. . . → H s (Uét , F ) → H s ((U0 )ét , F )⊕H s ((U1 )ét , F ) → H s ((U0 ∩U1 )ét , F ) → H s+1 (Uét , F ) → . . .
• (U, F ) ⊂ Č • (U, F ) of alternate cochains:
Proof. Consider the subcomplex Čalt
c(i0 , . . . , in ) = 0 if ij = ik for some j < k.
c(iσ(0) , . . . , iσ(n) = ε(σ)c(i0 , . . . , in ) .
• (U, F ) ⊂ Č • (U, F ) is a quasi-isomorphism
If U is a Zariski covering one can show that Čalt
(this is completely wrong in general!). For the covering U = (U0 → U, U1 → U ) this
implies that Ȟ s (U, ·) = 0 for any s ≥ 2. The Čech to Cohomology spectral sequence
degenerates immediately and gives rise to the Mayer-Vietoris long exact sequence. 
13.5. Flasque sheaves.
Definition 13.5.1. A sheaf F ∈ Ab(Xét ) is said to be flasque if
Hq (F ) = 0 ∀q>0 .
Theorem 13.5.2 (Verdier). The following conditions are equivalent:
(1) the sheaf F is flasque.
(2) for any U ∈ Xét , for any étale covering U of U , Ȟ q (U, F ) = 0 for all q > 0.
(3) for any U ∈ Xét , Ȟ q (U, F ) = 0 for all q > 0.
Proof. (1) ⇒ (2): consider the Čech to Cohomology spectral sequence:
E2p,q = Ȟ p (U, Hq (F )) ⇒ H p+q (Uét , F ) .
By assumption only the row q = 0 is non-zero. Hence
Ȟ p (U, F ) = E2p,0 ' E∞
p,0
' H p (Uét , F ) = 0 for p > 0 .
(2) ⇒ (3): take the colimit over all U’s.
(3) ⇒ (1): consider the Čech to Cohomology spectral sequence:
E2p,q = Ȟ p (U, Hq (F )) ⇒ H p+q (Uét , F ) .
Hence
0 NNN? ? ···
NNN
&
0 ? ? ···
E2p,q =
0 ? ? ···
? 0 0 ···
80 BRUNO KLINGLER

Let us prove by induction on n > 0 that H n (Uét , F ) = 0 for any U ∈ Xét .

For n = 1: this is OK as E20,1 = E21,0 = 0.
Suppose by induction that Hi (F ) = 0 for i ≤ n. Then E2p,q = 0 for any p + q ≤ n + 1
and the result. 
Corollary 13.5.3. Let f : X → Y . If F ∈ Ab(Xét ) is flasque then f∗ F ∈ Ab(Yét ) is
flasque.
Proof. By the previous proposition f∗ F is flasque if and only if for any covering U of
U ∈ Yét the Čech cohomology Ȟ q (U, f∗ F ) vanishes for q > 0. But Ȟ q (U, f∗ F ) =
Ȟ q (f −1 (U), F ) = 0 as F is flasque. 
13.5.1. Godement flasque resolution. Let F ∈ Ab(Xét ). We define
Y
God0 (F ) = ix∗ i∗x F .
x
Notice that any sheaf on x is obviously flasque, hence ix∗ F is flasque. It follows from
Corollary 13.5.3 that God0 (F ) is flasque. Define God1 (F ) = God0 (Coker (F ,→ God0 (F )))
and by induction:
Godi+1 (F ) = God0 (Coker (Godi−1 (F ) → Godi (F ))) .
We thus obtain a canonical flasque resolution: F ' God• (F ). For any f : X → Y it
satisfies:
f ∗ (God• (F )) = God(f ∗ F ) .
13.5.2. Flasque implies flabby.
Corollary 13.5.4. Let V ⊂ U be an open immersion in Xét and let F ∈ Ab(Xét ) be
flasque. Then F (U )  F (V ).
`
Proof. Set W = U V U . Denote by U0 , U1 the two copies of U covering W . Considering
the Mayer-Vietoris exact sequence (Lemma 13.4.1) we obtain:
−
0 → F (W ) → F (U ) ⊕ F (U ) → F (V ) → H 1 (Wét , F ) = 0
where the right hand term vanishes as F is flasque. The result follows. 
Remark 13.5.5. While the converse holds in the topological setting (any flabby sheaf is
flasque) this does not hold in the étale setting. Indeed let k be a field
`and0 set X = Spec k.
An open inclusion V ⊂ U in Xét is necessarily of the form U = V V hence any sheaf
on Xét is necessarily flabby. However it is not flasque in general as the Galois cohomology
of k is usually non-trivial.
13.6. The Leray spectral sequence. One computes the cohomology of topological
spaces by using classical dévissages (Künneth formula, Leray spectral sequence, sim-
plicial decompositions, excision...). One is reduced to compute the cohomology of the
fundamental building block in topology: the interval I = [0, 1].
In étale cohomology, the situation is similar (we use dévissage, like the Leray spectral
sequence or proper base change) but the fundamental blocks are more complicated. We
will be reduced to compute:
- the cohomology of points.
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 81

- the cohomology of curves over algebraically closed fields.

Let us start by giving one tool for dévissage: the Leray spectral sequence.
f g
Proposition 13.6.1. Let X → Y → Z and F ∈ Ab(Xét ). There is a spectral sequence:
E2p,q = Rp g∗ Rq f∗ (F ) ⇒ Rp+q (gf )∗ F .
In particular:
E2p,q = H p (Yét , Rq f∗ F ) ⇒ H p+q (Xét , F ) .
Proof. This is just the Grothendieck’s spectral sequence for a composition of functors
(noticing that f∗ maps injectives to injectives, in particular to g∗ -acyclic). 
Corollary 13.6.2. If Rq f∗ F = 0 for all q > 0 then H p (Yét , f∗ F ) = H p (Xét , F ).
To apply this corollary it will be necessary to compute the stalks of Rq f∗ F .
Proposition 13.6.3. Let f : X → Y be quasi-compact quasi-separated (recall this means
that the diagonal X → X ×Y X is quasi-compact). Let F ∈ Ab(Xét ) and y → Y a
geometric point. Then (Rq f∗ F )y = H q ((X ×Y Spec OY,y )ét , F ) (we do not indicate the
pull-back map from X to X ×Y Spec OY,y ).
Proof. By definition (Rq f∗ F )y = colim(V,v) H q ((X ×Y V )ét , F|X×Y V ), where (V, v) ranges
through the étale neighborhoods of (Y, y). By definition Spec OY,y = lim(V,v) V . As
the fiber product commutes with limits we are reduced to show that in our situation
“cohomology commutes with limits”. This follows from the following result (for details
we refer to [Stacks Project, Etale Cohomology Th.52.1]):
Theorem 13.6.4. Let X = limi∈I Xi be the limit of a directed system of schemes with
affine transition morphisms fi0 i : Xi0 → Xi . Assume that Xi is quasi-compact quasi-
separated for any i and that the following data are given:
(1) Fi ∈ Ab((Xi )ét ).
(2) for i0 ≥ i, ϕi0 ,i : fi−1 0 −1 00 0
0 i Fi → Fi such that ϕi00 i = ϕi00 i0 ◦ fi00 i0 ϕi0 i for i ≥ i ≥ i.

Set fi : X → Xi and F := colimi fi−1 Fi . Then

colimi∈I H p ((Xi )ét , Fi ) = H p (Xét , F ) for all p ≥ 0 .

Corollary 13.6.5. Let f : X → Y be a finite morphism and F ∈ Ab(Xét ). Then
Rq f∗ F = 0 for any q > 0.
Proof. By Proposition 13.6.3 one has
(Rq f∗ F )y = H q ((X ×Y Spec OY,y )ét , F ) .
As f : X → Y is finite the scheme X ×Y OY,y is a finite extension of the strictly
henselian ring OY,y , hence is a product of strictly henselian rings. The result follows
from the following:
Lemma 13.6.6. Let R be a local strictly henselian ring and S := Spec R. Then
Γ(S, F ) = Fs . In particular Γ(S, ·) is an exact functor.
82 BRUNO KLINGLER

Proof. Any étale surjective morphism onto S has a section as R is strictly henselian
hence id : (S, s) → (S, s) is cofinal among étale neighborhoods of (S, s). 

13.7. Cohomology of points: Galois cohomology. Let k be a field and X = Spec k.
Denote by G the Galois group Gal(k s /k). We already proved the following:
Proposition 13.7.1. There is an equivalence of categories
{k-finite étale algebras} ' {finite sets with continuous G-action}
A 7 → Homk (A, k s ) .
Proposition 13.7.2. There is an equivalence of categories
Sh((Spec k)ét ) ' {continuous G-sets}
F 7→ Fks .
In this proposition the inverse functor associates to a continuous G-set Fks the sheaf
F defined by F (U ) = HomG−sets (U (k s ), Fks ). In particular
F (Spec k) = HomG−sets (?, Fks ) = FkGs .
By considering only abelian sheaves and taking the derived functors:
H q (Xét , F ) = Rq Γ(Xét , F ) = (Rq (·G ))(Fks ) = H q (G, Fks )
hence the étale cohomology of points coincide with their Galois cohomology.

14. Cohomology of curves over an algebraically closed field

In this section we will prove the
Theorem 14.0.1. Let k be an algebraically closed field and X a smooth curve over k.
Then: H 0 (Xét , Gm ) = H 0 (XZar , Gm ), H 1 (Xét , Gm ) = Pic (X) and H q (Xét , Gm ) = 0 for
q ≥ 2.
Remark 14.0.2. If chark = p > 0 and one only assumes that k is separably closed, the
same proof will show that H 0 (Xét , Gm ) = H 0 (XZar , Gm ), H 1 (Xét , Gm ) = Pic (X) and
for q ≥ 2 the group H q (Xét , Gm ) is p-torsion.
Corollary 14.0.3. Let k be an algebraically closed field and X a smooth projective
curve over k. Let n be a positive integer invertible in k. Then H 0 (Xét , µn ) = µn (k),
H 1 (Xét , µn ) = Pic 0 (X)n , H 2 (Xét , µn ) = Z/n and H q (Xét , µn ) = 0 for any q > 2.
Proof. The Kummer exact sequence in Ab(Xét ) is
x7→xn
(25) 1 → µn → Gm → Gm → 1 .
Writing the corresponding long exact sequence, it follows from Theorem 14.0.1 that
H q (Xét , µn ) = 0 for any q > 2. In small degree the surjectivity of the elevation to the
n-th power on k ∗ gives
x7→xn
0 → H 0 (X, µn ) → k ∗ → k ∗ → 0 .
Hence H 0 (Xét , µn ) = µn (k). The remaining part of the long exact sequence gives
×n
0 → H 1 (Xét , µn ) → Pic (X) → Pic (X) → H 2 (Xét , µn ) → 0 .
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 83

The exact sequence

deg
0 → Pic 0 (X) → Pic (X) → Z → 0
gives H 1 (Xét , µn ) = Pic 0 (X)n . Moreover Pic 0 (X) = (Pic 0 X)(k) where Pic 0 X is the
Jacobian of X. As n is invertible in k and k is algebraically closed, the multiplication
by n is surjective on the k-points of the Abelian variety Pic 0 X, hence the result. 

14.1. The divisorial exact sequence. Recall that a scheme X is said to be normal if
for any point x of X the local ring OX,x is an integrally closed domain. In particular X
is locally integral. If moreover it is Noetherian and connected then it is integral (hence
irreducible in particular). The main tool in the proof of Theorem 14.0.1 is the following:
Proposition 14.1.1. Let X be a connected Noetherian normal scheme with generic
point η. The following sequence of Ab(Xét ) is exact (surjective on the right if X is
moreover regular):
M
(26) 0 → Gm → j∗ Gm,η → ix∗ Zx 99K 0 .
x∈X (1)

Proof. We have to show that for any geometric point y → X, the corresponding sequence
of stalks M
0 → (Gm )y → (j∗ Gm,η )y → (ix∗ Zx )y 99K 0
x∈X (1)
is exact. These stalks are obtained by taking filtered colimits over the étale neighbor-
hoods (U, u) of (X, y). As filtered colimits preserve exactness, it is enough to show that
for any U → X in Xét , the restriction of the sequence eq. (26) to UZar is exact.
As X is Noetherian normal (resp. regular) the scheme U is Noetherian normal too
(resp. regular). Hence Proposition 14.1.1 follows from the analogous Zariski statement:
Lemma 14.1.2. Let X be a connected Noetherian normal scheme. The following se-
quence of Ab(XZar ) is exact (surjective on the right if X is moreover regular):
M
(27) 0 → Gm → j∗ Gm,η → ix∗ Zx 99K 0 .
x∈X (1)
×
Proof. We denote by K the function field of X, by KX the constant Zariski sheaf defined
by K × on X and by Div the Zariski sheaf on X associated to the presheaf U 7→ Div(U ),
with Div(U ) the group of Weil divisors of U . The sequence eq. (27) can be rewritten as:
∗ ×
0 → OX → KX → Div 99K 0.
Let U = Spec A be a Zariski open subset of X. Hence A is an integrally closed domain.
Consider the sequence
M
(28) 0 → A∗ → K × → Z 99K 0 ,
htp=1

where the map on the right associates to a ∈ K ∗ the collection (vp (a)). Here vp denotes
the valuation of the discrete valuation ring Ap (recall that a local ring of dimension one
is a discrete valuation ring if and only if it is integrally closed if and only if it is regular).
84 BRUNO KLINGLER

We claim that the solid sequence eq. (28) is exact if A is integrally closed. Indeed in this
case A = ∩htp=1 Ap (see [Mat80, Th.38 p.124]). This finishes the proof of Lemma 14.1.2
in the case X normal.
The surjectivity of the dashed arrow is equivalent to saying that any prime ideal p of
height 1 in A is principal, or equivalently (see [Mat80, p.141]) that the Noetherian integral
domain A is factorial. But any regular local ring is factorial. Thus the dashed sequence
eq. (28) is exact for any regular local ring A, which finishes the proof of Lemma 14.1.2
in the case X regular. 

14.2. Proof of Theorem 14.0.1. From now on X is a smooth projective curve over
an algebraically closed field k. We will compute H • (Xét , Gm ) from the exact sequence
eq. (26).
Lemma 14.2.1. H q (Xét , j∗ Gm,η ) = 0 for all q > 0.
Proof. Apply the Leray spectral sequence to j : η → X:
H q (ηét , Gm,η ) = H q (Xét , Rj∗ Gm,η ) .
Our claim then follows from the following two results:
Sub-lemma 14.2.2. Rp j∗ Gm,η = 0 for all p > 0.
Hence H q (Xét , j∗ Gm,η ) = H q (Xét , Rj∗ Gm,η ) = H q (ηét , Gm,η ).
Sub-lemma 14.2.3. H q (ηét , Gm,η ) = 0 for all q > 0.
To prove Sub-lemma 14.2.2 one argues as follows.
As X is a scheme of finite type over the algebraically closed field k, it is enough to
show that for any closed point x of X the stalk (Rq j∗ Gm,η )x vanishes.
It follows from Proposition 13.6.3 that for any closed point x ∈ X:
(Rq j∗ Gm,η )x = H q (η ×X Spec OX,x , Gm ) .
Let Spec A be some affine neighbourhood of x in X. Let K be the fraction field of A,
hence η = Spec K. Then η ×X Spec OX,x = Spec (OX,x ⊗A K). The ring OX,x ⊗A K is
sh , hence it is either O
a localisation of the discrete valuation ring OX,x = OX,x X,x or its
fraction field. As any local uniformizer of OX,x gets inverted in OX,x ⊗A K, we obtain
that η ×X Spec OX,x = Spec Frac OX,x .
As every element of OX,x = OX,x sh is algebraic over O
X,x , the extension Frac OX,x
of K is algebraic, hence an extension of k of transcendence degree 1. Thus both Sub-
lemma 14.2.2 and Sub-lemma 14.2.3, hence the proof of Lemma 14.2.1, follow from the
following:
Proposition 14.2.4. Let k be an algebraically closed field and K/k an extension of
transcendence degree 1. Then H q ((Spec K)ét , Gm ) = 0 for all q > 0.

Let us for the moment admit Proposition 14.2.4 and finish the proof of Theorem 14.0.1.
Lemma 14.2.5. H q (Xét , x∈X(0) ix∗ Zx ) = 0 for all q > 0.
L
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 85

Proof. The scheme X is quasi-compact quasi-separated hence étale cohomology on X

commutes with colimits. Hence it is enough to show the vanishing of H q (Xét , ix∗ Zx )
for all x ∈ X(0) , q > 0. As ix : x → X is a finite morphism Rq ix∗ Zx = 0 for all
q > 0 by Corollary 13.6.5. It follows from the Leray spectral sequence for ix that
H q (Xét , ix∗ Zx ) = H q (xét , Zx ), which vanishes because x is separably closed. 
We deduce from Lemma 14.2.1 and Lemma 14.2.5 and from the exact sequence eq. (26)
of étale sheaves that:
- H q (Xét , Gm ) = 0 for q ≥ 2;
- the following sequence is exact:
M
0 → H 0 (Xét , Gm ) → H 0 (Xét , j∗ Gm,η ) → H 0 (Xét , ix∗ Zx ) → H 1 (Xét , Gm ) → 0 .
x∈X(0)

Comparing this sequence with the corresponding Zariski sequence

M
0 → H 0 (XZar , Gm ) → H 0 (XZar , j∗ Gm,η ) → H 0 (XZar , ix∗ Zx ) → H 1 (XZar , Gm ) = Pic X → 0
x∈X(0)

and as the H 0 ’s coincide, we conclude that H 1 (Xét , Gm ) = Pic (X).

2
14.3. Brauer groups and the proof of proposition 14.2.4. The main tool for the
proof of Proposition 14.2.4 is the Brauer group.
14.3.1. Summary on Brauer groups. Let k be a field with algebraic closure k. In this
section an algebra over k is an associative, possibly non-commutative, unital ring A
equipped with a ring morphism from k to the center Z(A) of A mapping 1 to 1. An
A-module is a right A-module. The k-algebra A is said to be central, resp. simple,
resp. finite, if Z(A) = k, resp. A has no non-trivial two-sided ideals, resp. A is a finite
dimensional k-vector space. It is a division algebra if every element has a multiplicative
inverse.
Theorem 14.3.1. The following statements are equivalent:
(1) A is a central finite simple k-algebra.
(2) there exists a positive integer d such that A ⊗k k ' Mat(d × d, k).
(3) there exists a positive integer d and a finite extension k 0 /k such that A ⊗k k 0 '
Mat(d × d, k 0 ).
(4) A ' Mat(n × n, D) where D is a division algebra of center k.
Remark 14.3.2. The integer d in (2) and (3) is called the degree of A.
Definition 14.3.3. We define a relation on finite simple central k-algebras as follows:
A1 ∼ A2 if there exist m, n > 0 such that
Mat(n × n, A1 ) ' Mat(m × m, A2 )
Equivalently, the division algebras associated to A1 and A2 by the Theorem 14.3.1(4)
coincide.
One checks (see [Stacks Project, Brauer Groups, Lemma 5.1]) that the relation ∼ on
finite simple central k-algebras is an equivalence relation.
86 BRUNO KLINGLER

Definition 14.3.4. Let k be a field. The Brauer group of k is the set Br(k) of equivalence
classes of finite central simple algebras over k, endowed with the abelian group law [A1 ] +
[A2 ] := [A1 ⊗k A2 ].
In this definition the existence of inverses is given by
Lemma 14.3.5. Let A be a central finite simple k-algebra. Then:
A ⊗k Aop ' Endk (A)
a ⊗ a0 7→ (x 7→ axa0 )
Hence we can define −[A] := [Aop ].
Notice that Br(k) = ∪n∈N Br(n, k), where Br(n, k) denotes the torsion subgroup of
classes [A] such that there exists k 0 /k finite with Ak0 ' Mat(n × n, k 0 ). Now Br(n, k) is
easy to describe: it is the group of k-forms of Mat(n × n, k). Hence:
(29) Br(n, k) ' H 1 (G, Aut Mat(n × n, k)) = H 1 (G, P GL(n, k))
as all automorphisms of Mat(n × n, k) are interior.
The short exact sequence of G-groups
∗
1 → k → GL(n, k) → P GL(n, k) → 1
give rise to boundary maps of cohomology groups
H 1 (G, P GL(n, k)) → H 2 (G, k)
which are compatible. Composing with eq. (29) one obtains a canonical map:
δ : Br(k) → H 2 (G, k) .
Theorem 14.3.6. The map δ : Br(k) → H 2 (G, k) is an isomorphism.
Proof. Exercice, see [Stacks Project, Etale Cohomology, Th.60.6]. 

14.3.2. Brauer groups and Galois cohomology. The link between Brauer groups and
our problem lies in the following:
Proposition 14.3.7. Let K be a field with algebraic closure K and G := Gal(K/K).
Suppose that for any finite extension K 0 /K the Brauer group Br(K 0 ) vanishes. Then:
∗
(i) H q (G, K ) = 0 for all q > 0.
(ii) H q (G, F ) = 0 for any torsion G-module F and any q ≥ 2.
Proof. See [Se97, Chapter II, Section 3, Proposition 5]. 
∗
14.3.3. Tsen’s theorem. As H q ((Spec K)ét , Gm ) = H q (G, K ), Proposition 14.2.4 will
follow from Proposition 14.3.7 if we prove that Br(K) = 0 for K/k an extension of
transcendence degree 1, with k algebraically closed.
Definition 14.3.8. A field K is said to be Cr if any polynomial f ∈ K[T1 , . . . , Tn ]
homogeneous of degree d with 1 < dr < n admits a non-trivial zero.
Proposition 14.3.9. If K is C1 then Br(K) = 0.
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 87

Proof. Let D be a K-division algebra. Hence D ⊗K K ' Mat(d×d, K), the isomorphism
being uniquely defined up to interior automorphisms. In particular the determinant
det : Mat(d × d, K) → K
is G-invariant, hence descend to
Nred : D → K .
This reduced norm is a homogeneous polynomial in d2 variables of degree d over K.
Hence if d > 1 there exists x 6= 0 ∈ D satisfying Nred (X) = 0: contradiction to the
invertibility of x.
Thus d = 1 and Br(K) = 0. 
Theorem 14.3.10. (Tsen) The function field of a variety X of dimension r over an
algebraically closed field k is a Cr -field.
Proof. Without loss of generality we can assume that X is projective. Let f ∈ K[T1 , . . . , Tn ]
homogeneous of degree d, 1 < dr < n (where K = k(X)). The coefficients of f can
be assumed to lie in Γ(X, OX (H)) where H is some ample line bundle on X. Fix
a positive integer e and consider α = (α1 , . . . , αn ) in Γ(X, OX (eH)). Then f (α) ∈
Γ(X, OX ((de + 1)H)). We want to show that the equation f (α) has a non trivial zero.
The number of possible variables α is
er r
n · dimk Γ(X, OX (eH)) ∼ n · (H )
r!
by the Riemann-Roch theorem.
The number of equations is
(de + 1)r r
dimk Γ(X, OX ((de + 1)H)) ∼ (H )
r!
again by the Riemann-Roch theorem.
As n > dr there are more variables than equations hence f (α) = 0 has a non-trivial
solution. 

14.3.4. Proof of Proposition 14.2.4. Let K/k be of transcendence degree 1. We have

to show that if K 0 /K is finite then Br(K 0 ) = 0. Any such K 0 can be written as a colimit
of extensions K 00 of finite type of k, of transcendence degree 1. Any such extension K 00
is the function field of a curve over k. Hence Br(K 0 ) = colimK 00 Br(K 00 ) = 0 by Tsen’s
theorem.
2

15. Constructible sheaves

Classical topology study constant sheaves and their natural generalisation: locally
constant sheaves. These locally constant sheaves have a bad functorial behaviour: the
direct image of a locally constant sheaf is hardly ever locally constant. This leads to
the notion of constructible sheaf. We follow the same path for étale topology, with a
significant difference: one only considers torsion sheaves.
88 BRUNO KLINGLER

15.1. Pathology of the étale constant sheaf Z. The étale constant sheaf Z is
cohomologically uninteresting, as the following lemma shows:
Lemma 15.1.1. Let X be a regular scheme. Then H 1 (Xét , ZX ) = 0.
Proof.
x i
Sub-lemma 15.1.2. Let X be a scheme and x → X a (non-necessarily closed) point.
1
Then H (Xét , ix∗ Z) = 0.

Proof. The Leray spectral sequence for ix

E2p,q = H p (Xét , Rq ix∗ Z)) ⇒ H p+q (xét , Z)
implies readily H 1 (Xét , ix∗ Z)) ⊂ H 1 (xét , Z). But
H 1 (xét , Z) = H 1 (Gal(k(x)/k(x)), Z)
= Homcont (Gal(k(x)/k(x)), Z)
=0 ,
where the first equality comes from our identification of the étale cohomology of points
with Galois cohomology of their residue fields and the vanishing of Galois cohomology
follows from the fact that Gal(k(x)/k(x)) is a profinite groupe while Z has no torsion. 

Let us finish the proof of Lemma 15.1.1. As X is regular one can assume that X is
connected, hence irreducible. Let j : η → X be the generic point of X. Lemma 15.1.1
follows immediately from Sub-lemma 15.1.2 applied to j and the following:
Sub-lemma 15.1.3. The adjunction map ZX → j∗ Zη is an isomorphism.
Proof. We have to show that for any geometric point x → X the map of stalks ZX,x →
(j∗ Zη )x is an isomorphism.
On the one hand ZX,x = colim(V,v) ZX (V ) = Z, where the colimit can be taken over
connected étale neighbourhoods (V, v) of (X, x) as X is irreducible.
On the other hand (j∗ Zη )x = colim(V,v) Zη (η ×X V ) where the colimit can be taken
over the connected étale neighbourhoods (V, v) of (X, x). As V → X is étale, the scheme
η ×X V is the disjoint union of the generic points of η ×X V . As X is regular, V is regular
too. As it is connected it is irreducible. Hence η ×X V is one point, Zη (η ×X V ) = Z
and (j∗ Zη )x = Z.
One easily checks that the map ZX,x = Z → (j∗ Zη )x = Z is the identity, hence the
result. 

In view of the proof of Sub-lemma 15.1.3, it is natural to consider only torsion étale
sheaves.
Definition 15.1.4. Let X be a scheme. An étale sheaf F ∈ Ab(Xét ) is said to be a
torsion sheaf if any local section of F is killed by a positive integer n, i.e. F = colimn Fn ,
n×·
where Fn = ker(F → F ).
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 89

15.2. Locally constant constructible sheaves.

Definition 15.2.1. Let S be a scheme. An étale sheaf F ∈ Sh(Sét ) is said to be

constant constructible (or constant finite) if it the étale sheafification of the constant
presheaf associated to a finite set.

We saw that any such constant sheaf is representable by Σ × S, Σ finite set.

Definition 15.2.2. Let S be a scheme. An étale sheaf F ∈ Sh(Sét ) is said to be locally

constant constructible (lcc), or locally constant finite, if there exists an étale covering
family (Ui → S)i∈I with F|Ui ∈ Sh((Ui )ét ) constant finite.

The representability of constant sheaves generalizes to locally constant sheaves:

Lemma 15.2.3. Let S be a scheme and F ∈ Sh(Sét ). The following conditions are
equivalent:
(1) F is lcc.
(2) F ' hU where U → S is a finite étale morphism.

Proof. We start with the easy direction (2) ⇒ (1). One has to show that for any U → S
finite étale there exists an étale covering (Si → S)i∈I such that for any i ∈ I, U ×S Si is
isomorphic to a`disjoint union of copies of Si .
Write S = n∈N∗ Sn , where Sn is defined by the condition U|Sn → Sn is finite of
degree n. Without loss of generality we can thus assume that U → S is of fixed degree
n > 0.
If n = 1 the étale morphism U → S is an isomorphism and the conclusion holds true
trivially. Suppose n > 1. Consider the second projection p2 : U ×S U → U obtained
by base change to U from U → S. It is an étale`morphism of degree n and admits a
section ∆U : U → U ×S U . Hence U ×S U = ∆U U 0 where U 0 → U is étale of degree
n − 1. By induction on n there exists an étale covering (Ui → U )i∈I such that for any
i ∈ I, U 0 ×U Ui is isomorphic to a disjoint union of copies of Ui . But then U ×S Ui is
also isomorphic to such a disjoint union.

Conversely let us show that (1) ⇒ (2). This is an application of fpqc descent for
schemes.
Let F ∈ Sh(Sét ) and (fi : Si → S)i∈I be an étale covering family such that FSi '
Σi × hSi for some finite sets (Σi )i∈I . We want to show that F is representable by some
X → S finite étale.
One can work Zariski-locally on S: it is enough to prove the statement for each open
subset Sn of an open Zariski cover (S`n )n∈N of S. For i ∈ I let ni := |Σi |. For every
positive integer n let us define Un := ni =n Si and by Sn the image of Un in S. As the
fi ’s are open, Sn is an open subscheme of S. Hence without loss of generality replacing
S by Sn we can assume that ni = n for all i ∈ I.
We are thus reduced to considering the étale covering S 0 := i∈I Si → S with ξ 0 :
`
F|S 0 ' Σ × hS 0 , Σ a finite set of cardinality n.
Restricting S and replacing S 0 by a finite disjoint union of open subschemes if neces-
sary, we can assume that S and S 0 are affine, hence S 0 → S is an fpqc morphism.
90 BRUNO KLINGLER

Consider the two projections p1 , p2 : S 0 ×S S 0 → S 0 . Denoting by p∗j the corresponding

base change, one obtains an isomorphism:
p∗1 (ξ 0 )−1 p∗2 (ξ 0 )
ϕ : p∗1 (Σ × S 0 ) ' p∗1 (F|Sét
0 ) = F|(S 0 × S 0 )
S ét
= p∗2 (F|Sét
0 ) ' p∗2 (Σ × S 0 ) .
It obviously satisfies the cocycle condition
p∗23 (ϕ) p∗12 (ϕ) = p∗13 (ϕ) .
The effectivity of fpqc descent for affine morphisms implies that there exists an affine
morphism X → S such that Σ × S 0 ' X ×S S 0 (inducing an isomorphism of descent
datas).
As the morphism X ×S S 0 → S 0 is finite étale and S 0 → S is étale, the morphism
X → S is finite étale too. Hence ξ 0 : F|S 0 ' hX×S S 0 in Sh(Sét
0 ) satisfies p∗ ξ 0 = p ξ 0 in
1 2
0 0 0 0
Sh((S ×S S )ét ). Thus ξ is a section of Hom(FS 0 , hX ) on S whose two restrictions to
S 0 ×S S 0 coincide. By the sheaf condition it descends to ξ ∈ Hom(F, hX )(S). Similarly
for ξ 0−1 , hence ξ is an isomorphism. 

15.3. Constructible sheaves. One checks easily that:

- if f : X → Y is a morphism of schemes and G ∈ Sh(Yét ) is lcc then f ∗ G ∈
Sh(Xét ) is lcc.
- if f : X → Y is finite étale and F ∈ Sh(Xét ) is lcc then f∗ F ∈ Sh(Yét ) is lcc.
However, as in classical topology, the class of lcc sheaves is not stable under more general
push-forward. The class of constructible sheaves will remedy this problem.
For the sake of generality let us start with a purely topological definition.
Definition 15.3.1. Let X be a topological space. A subspace Z ⊂ X is said to be retro-
compact if the inclusion i : Z → X is quasi-compact, in other words: if the intersection
of any quasi-compact open subset of X with Z is quasi-compact.
Example 15.3.2. If X is a Noetherian scheme, any open subspace of |X| is quasi-compact,
hence retrocompact.
Definition
S 15.3.3. A subspace Z ⊂ X of a topological space X is said to be constructible
c
if Z = i∈I Ui ∩ Vi , where I is a finite set, and for any i ∈ I, Ui and Vi are retrocompact
open subsets of X.
It follows easily from this definition that if X is a Noetherian topological space then
the constructible subsets of X are exactly the finite unions of locally closed subspaces.
Definition 15.3.4. Let X be a scheme. A subscheme T ⊂ X is said to be locally closed
constructible if T is a locally closed subscheme of X such that the topological space |T |
is a constructible subspace of |X|.
Definition 15.3.5. Let X be a scheme. An étale sheaf F ∈ Sh(Xét ) is said to be
constructible
` if for any open affine subscheme U ⊂ X, there exists a decomposition
U = i UI (called a partition of U ) such that UI is a localy closed constructible subscheme
of U and F|Ui ∈ Sh(Uiét ) is lcc.
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 91

Remarks 15.3.6. (1) Notice that the condition in Definition 15.3.5 depends only on
the topological structure of the Ui ’s, not on their schematic structure. Indeed if
T, T 0 ⊂ X are two locally closed subscheme of X with |T | = |T 0 | then Tét ' Tét
0
c1 c1
to be added
(2) When X is quasi-compact quasi-separated an étale sheaf`F ∈ Sh(Xét ) is con-
structible if and only if there exists a global partition X = i Xi by locally closed
constructible subschemes Xi ⊂ X such that F|Xi is lcc.
15.4. Properties of constructible sheaves on Noetherian schemes. Let us start
by stating a few easy properties of constructible sheaves on general schemes.
S
- If X = i∈I Ui with Ui ⊂ X open subschemes and F ∈ Sh(Xét ) satisfies that for
all i ∈ I, F|Ui ∈ Sh((Ui )ét ) is constructible then F is constructible.
- If f : X → Y is a morphism of schemes and F ∈ Sh(Yet ) is constructible then
f ∗ F ∈ Sh(Xét ) is constructible.
- For Abelian sheaves the property of being constructible is stable under kernel,
cokernel, image and extension. Hence the full subcategory Abc (Xét ) of Ab(Xét )
whose objects are the constructible Abelian sheaves is an Abelian subcategory.
- If X is a locally Noetherian scheme then F is constructible if and only if for all
x ∈ X there exists an open subscheme U ⊂ {x} such that F|U is lcc.
From now on we concentrate on Noetherian schemes.
Proposition 15.4.1. Let X be a Noetherian scheme. Let F ∈ Sh(Xét ). The following
conditions are equivalent:
(1) F is lcc.
(2) F satisfies the following two properties:
(a) For any geometric point x → X the stalk Fx is finite.
(b) If y is a specialization of x (meaning that y ∈ {x} and denoted x y) the
specialization morphism Fy → Fx is a bijection.
Proof. The fact that (1) implies (2) is trivial, let us prove that (2) implies (1). Let
x → X be any geometric point of X. As Fx = colim(V,v) F (V ) is a finite set (where (V, v)
runs through the étale neighborhoods of (X, x)) there exists an étale neighborhood (V, v)
f
of (X, x) such that F (V )  Fx . Let us choose a finite set E ⊂ F (V ) with f|E : E ' Fx .
This defines a sheaf morphism EV → F|V satisfying (EV )x ' Fx .
As V is Noetherian it follows that any geometric point y of V is related to x through
a chain of specializations:
x p1 p2 p3 ··· pn y .
As (EV )y ' (EV )x the condition (b) then implies:
(EV )y ' Fy .
Hence EV ' F|V . This proves that F is lcc. 
Proposition 15.4.2. Let X be a Noetherian scheme. Let F ∈ Sh(Xét ). The following
conditions are equivalent:
(1) F is constructible.
92 BRUNO KLINGLER

(2) The function c : X → N ∪ {∞} which to x ∈ X associates the cardinality of Fx is

bounded and constructible (i.e. for all n ∈ N the preimage c1 (n) is a constructible
subset of |X|).

Proof. Once more (1) ⇒ (2) is clear, let us prove (2) ⇒ (1). As X is Noetherian,
the function c take only finitely many values. Hence without loss of generality one can
assume that c is constant.
Without loss of generality we can assume that X is irreducible. Let η be a geometric
point over the generic point η of X. As Fη is finite there exists an étale neighbourhood
(V, v) if (X, η) such that F (V )  Fη . Any geometric point x of V is a specialization of
η, hence gives rise to a commutative diagram:

F (V ) / Fx
GG
GG
GG
GG
G# 
Fη .

As Fx = Fη , it follows that Fx ' Fη . Let U be the image of V in X, this is a non-empty

open subset of X and F|U ∈ Sh(Uét ) is lcc by Proposition 15.4.1.
By Noetherian induction one can assume that F|X\U is constructible, hence the con-
clusion. 

Corollary 15.4.3. Let f : Y → X be a surjective morphism of finite type between

Noetherian schemes and F ∈ Sh(Xét ). The following conditions are equivalent:
(1) F is constructible.
(2) f ∗ F is constructible.

Proof. Let us prove the non-trivial implication (2) ⇒ (1). The result is clear if the
morphism f is moreover étale. We reduce to this case using Noetherian induction.
Without loss of generality we can assume that X is irreducible. Let η = Spec K be
the generic point of X. The base change Yη := η ×X Y is a K-scheme of finite type
hence admits a closed point, with residue field L a finite extension of K. Let E denote
the separable closure of K in L. Consider the commutative diagram:

Spec LJ / Yη /Y
JJJ
JJJ
JJJ
h %
Spec EG
GG
GG
g GGG
G#  
η /X.

The morphism h is radicial finite surjective while the morphism g is finite étale surjective
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 93

All these data are of finite presentation hence lift to an open neighbourhood V of η
in X:
VL B / YV /Y
BB
BB
B
h BB
VE B
BB
BB
g BBB
!  
V /X,
where the morphisms h and g have the same properties as above.
The fact that f ∗ F is constructible implies that h∗ g ∗ F|V is constructible. As h is
radicial h∗ : (VE )ét → (VL )ét is an equivalence of categories, hence g ∗ F|V is constructible.
But g is finite étale surjective hence F|V is constructible (easy case above).
We conclude by Noetherian induction. 
Corollary 15.4.4. Lte f : V → X be étale of finite type between Noetherian scheme.
Then hV ∈ Sh(Xét ) is constructible.
Proof. We apply Proposition 15.4.2 to the fibers of V /X. The result follows from the
fact that the cardinality of the geometric fibers of an étale separated morphism of finite
type varies lower semi-continuously on X, see [EGAIV, 18.2.8]. 
From now on we denote by Λ the ring Z/nZ.
Proposition 15.4.5. Let X be a Noetherian scheme. Let F ∈ Λ − Mod(Xét ). The
following conditions are equivalent:
(1) F is constructible.
(2) F is a Noetherian object in Λ − Mod(Xét ) (recall that an object A in an Abelian
category is Noetherian if any increasing sequence A0 ⊂ A1 ⊂ .... ⊂ A is station-
nary).
f∗
(3) There exists f : V → U in Xét of finite type over X such that F ' Coker (ΛX (V ) →
ΛX (U ).
Proof. Without loss of generality we can assume that X is irreducible.
We first show that (1) ⇒ (2) by Noetherian induction. Let F0 ⊂ F1 ⊂ · · · F be an
increasing sequence. Let U ⊂ X be a non-empty open subset such that F|U is locally
constant and consider the restriction of F0 ⊂ F1 ⊂ · · · F to U .
Let η be a geometric point over the generic point η of X. The stalk Fη is finite
hence the sequence (Fi )η is necessary stationnary. Without loss of generality we can
thus assume that the sequence (Fi )η is constant.
As F|U is lcc, the specialization map Fx → Fη is an isomorphism for any x specializa-
tion of η in U . Hence the following diagram is commutative:

(Fi )x  / (Fi )η
_


Fx / Fη .
94 BRUNO KLINGLER

Let s1 , · · · , sn be generators of (F0 )η . Hence there exists an étale neighborhood of η

such that the si ’s lift to F0 (V ). The diagram above implies that the germs of the si ’s
generate (Fi )x for any geometric point x of V .
It follows that the sequence (Fi )i∈N is stationnary on V , hence on the image U0 of V in
U . By Noetherian induction the sequence (Fi )X\U0 is stationnary. Finally the sequence
(Fi )i∈N is stationnary
Let us show (2) ⇒ (3). Any F ∈ Λ − Mod(Xét ) can be written as a quotient
M h
ΛX (Ui )  F
i∈I

for an étale covering family (Ui )i∈I . As F is Noetherian in Λ − Mod(Xét ), there exists
a finite subset I0 ∈ I such that
M h
ΛX (Ui )  F .
i∈I0
`
Let us define U = i∈I0 Ui , this is a separated étale X-scheme of finite type hence
ΛX (U ) is constructible by Corollary 15.4.4. Thus the kernel of h is constructible, hence
Noetherian in Λ − Mod(Xét ). Repeating the previous construction replacing F with
f∗
Kerh, we obtain that F can be written Coker (ΛX (V ) → ΛX (U ) as required.
Finally we show tht (3) ⇒ (1). By Corollary 15.4.4, both ΛX (V ) and ΛX (U ) are
f∗
constructible, hence also F ' Coker (ΛX (V ) → ΛX (U ).

Corollary 15.4.6. The full subcategory Λ − Mod(Xét )c ⊂ Λ − Mod(Xét ) of con-
stuctible ΛX -module is a Serre subcategory.
Proof. This is true for the full subcategory of Noetherian objects in any Abelain cate-
gory> 
Corollary 15.4.7. Any F ∈ Λ − Mod(Xét ) is a filtered colimit of constructible Fi ∈
Λ − Mod(Xét )c .
Proof. The category Λ − Mod(Xét ) admits as a generating family the ΛX (U ), U → X
affine étale, hence in particular constructible. Thus any F ∈ Λ − Mod(Xét ) is a filtered
union of its constructible sub-modules. 
Corollary 15.4.8. Any torsion sheaf in Ab(Xét ) is a filtered colimit of constructible
sheaves.
×n
Proof. Let F be an étale torsion sheaf. Hence F is a filtered colimit of Fn := ker(F →
F . Each Fn belongs to Z/nZ − Mod(Xét ), hence is a filtered colimit of constructible
subsheaves by the previous corollary. Hence the result. 

16. Proper base change

The basic reference for this chapter is [SGA4, Exp. XII, XIII].
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 95

16.1. The classical topological case. Let f : X → S be a continuous map between

topological spaces and F ∈ Ab(X). Given a point s ∈ S, let us denote by i : f −1 (s) → X
the closed inclusion. Hence i∗ is exact and the morphism of functors 1 → i∗ i∗ induces a
natural morphism of groups
(30) (Rr f∗ F )s := colims∈V H r (f −1 (V ), F|f −1 (V ) ) → H r (f −1 (s), i∗ F ),
(where the colimit is taken over all open neighborhoods of s in S).
In general this morphism is not an isomorphism, even for r = 0. Suppose indeed that
f is the inclusion of an open subset X of S. For a point s ∈ X \ X the stalk (f∗ F )s is
usually non-zero while f −1 (s) = ∅ hence the right hand side H 0 (f −1 (s), i∗ F ) is zero.
Notice that if f is closed and U is a neighborhood of f −1 (s), the image f (X \ U ) is a
closed subspace of S, the point s belongs to the open subspace V := S \ f (X \ U ), and
f −1 (V ) ⊂ U . Hence the open sets f −1 (V ) of X form a neighborhood basis of f −1 (s).
Thus (Rr f∗ F )s = colimU ⊃f −1 (s) H r (U, F ). In the case where X is locally compact one
can go further thanks to the following result, whose elementary proof is left to the reader:

Lemma 16.1.1. Let X be a locally compact space and Z  / / X a compact subspace.
i

Then the natural map colimU ⊃Z H r (U, F ) → H r (Z, i−1 F ) is an isomorphism.

Recall that a continuous map f : X → S between topological spaces is said to be
proper if it is separated and universally closed. When both X and S are locally compact
(in particular Hausdorff) f : X → S is proper if and only if is universally closed, if and
only if the preimage of a compact subset is compact.
Corollary 16.1.2. Let f : X → S be a continuous proper map between topological
spaces. For any s ∈ S the natural morphism (Rr f∗ F )s → H r (Xs , f ) is an isomorphism.
More generally:
Theorem 16.1.3. (topological proper base change) Let f : X → S be a continuous
proper map between topological spaces. Consider a Cartesian base change diagram of
topological spaces:
g0
XS 0 /X

f0 f
 
S0 /S.
g

Then for any F ∈ Ab(X) the natural morphism of sheaves on S 0

∗
g ∗ (Rr f∗ F ) → Rr f∗0 (g 0 F )
is an isomorphism.

Remark 16.1.4. If g := is : s  / / S one recovers Corollary 16.1.2.

The morphism g ∗ (Rr f∗ F ) → Rr f∗0 (g 0 ∗ F ) is obtained as follows. By adjunction it is

equivalent to construct a morphism of functors Rr f∗ → g∗ (Rr f∗0 )g 0 ∗ , which we define as
the composition:
∗ ∗ ∗ ∗
Rr f∗ → Rr f∗ g∗0 g 0 → Rr (f ◦ g 0 )∗ g 0 = Rr (g ◦ f 0 )∗ g 0 → g∗ (Rr f∗0 )g 0 .
96 BRUNO KLINGLER

The first map is given by the adjunction 1 → g∗0 g 0 ∗ ; the second and the last ones are
special instance of the following: in the situation of Theorem 13.2.2 one has natural
morphisms of functors Rp G ◦ F → Rp (G ◦ F ) and Rp (G ◦ F ) → G ◦ Rp F , which are
nothing else than the “border morphisms” of Grothendieck’s spectral sequence.
16.2. The étale case. In étale topology the proper base change theorem still holds if
one restricts oneself to torsion coefficients:
Theorem 16.2.1. (étale Proper Base Change) Let S be a scheme and let f : X → S
be a proper morphism (i.e. of finite type, separated and universally closed). Consider a
Cartesian base change diagram
g0
XS 0 /X

f0 f
 
S0 /S.
g
0
Then for any Abelian torsion sheaf F on Xét the natural morphism of sheaves on Sét
∗
g ∗ (Rr f∗ F ) → Rr f∗0 (g 0 F )
is an isomorphism.
Corollary 16.2.2. Let f : X → S be a proper morphism of schemes and let F be an
Abelian torsion sheaf on Xét . For any geometric point s → S the natural map
(Rq f∗ F )s → H q ((X ×S s)ét , F|X×S s )
is an isomorphism.
Proof. Apply Theorem 16.2.1 with S 0 = s. 
Theorem 16.2.3. Let A be a strictly henselian local ring and S = Spec A. Let f :
X → S be a proper morphism of schemes and X0 the closed fiber of f . Then for any
Abelian torsion sheaf F on Xét and any non-negative integer q, the natural restriction
map H q (Xét , F ) → H q ((X0 )ét , F|X0 ) is an isomorphism.
Proposition 16.2.4. Theorem 16.2.1 and Theorem 16.2.3 are equivalent.
Proof. We first show that Theorem 16.2.1 implies Theorem 16.2.3. Let s ∈ S be its
closed point. As S is strictly henselian, s = s and (Rq f∗ F )s ' H q ((X0 )ét , F|X0 ) by
Corollary Corollary 16.2.2. On the other hand by the description of the stalks of étale
sheaf given in Proposition Proposition 13.6.3,
(Rq f∗ F )s ' H q ((X ×S OS,s )ét , F|X×S OS,s ) .
But OS,s = A hence X ×S OS,s = X and the conclusion follows.
Conversely let us show that Theorem 16.2.3 implies Theorem 16.2.1. Let s0 → S 0 be
a geometric point of S 0 mapped to a geometric point s → S of S. Then
(g ∗ (Rr f∗ F ))s0 = (Rr f∗ F )s = H r ((X ×S OS,s )ét , F )
while (Rr f∗0 (g 0 ∗ F ))s0 = H r ((X 0 ×S 0 OS 0 ,s0 )ét , g 0 ∗ F ). By Theorem 16.2.3 the natural map
H r ((X ×S OS,s )ét , F ) → H r ((X 0 ×S 0 OS 0 ,s0 )ét , g 0 ∗ F ) coincide with the identity map
∗
H r ((X ×S s)ét , F ) → H r ((X 0 ×S 0 s0 )ét , g 0 F ) ,
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 97

hence the result. 

16.3. Proof of Theorem 16.2.3. The proof has three steps:
(a) Reduction to the case where F is a constant finite étale sheaf.
(b) Explicit computation of the cases q = 0 and q = 1 for F = Z/nZ.
(c) Reduction to the case where f : X → S is of relative dimension at most one;
computation for q = 2.
16.3.1. Notations. If A denotes a local ring with maximal ideal m and spectrum S and
f : X → S a morphism, we denote by Sn the spectrum of A/mn+1 , by Xn := X ×S Sn the
n-th infinitesimal neighborhood of the closed fiber X0 in X and by X̂ := colimn Xn →
Ŝ := colimn Sn the formal scheme formal completion of X along X0 . Hence one has a
commutative diagram:
   
(31) X0  / X1  / · · · / Xn  / ···

      
S0  / S1  / · · · / Sn  / ···

16.3.2. Reduction of Theorem 16.2.3 to the excellent case. To prove Theorem 16.2.3
we will have to compare schemes over the stricty henselian ring A and schemes over its
m-adic completion Â. For general A (even Noetherian) the flat map A → Â can have a
pathological behaviour. The class of excellent rings was introduced by Grothendieck as
a remedy to this problem. We recall the definition for completeness:
Definition 16.3.1. A ring A is excellent if:
- it is Noetherian,
- for every p ∈ Spec A the map Ap → Âp is geometrically regular,
- for every finite A-algebra B the singular points of Spec B form a closed subset of
Spec B,
- A is universally catenary.
For us it will be sufficient to know that the strict henselization of a Z-algebra of finite
type is an excellent ring c1 . c1
reference?
Lemma 16.3.2. If Theorem 16.2.3 is true for A excellent then it is true for all A.
Proof. c2 As any ring is a filtering colimit of its subrings which are of finite type as c2
details!
Z-algebras and A is strictly henselian, A is a filtering colimit of Ai , i ∈ I, where Ai is the
strict henselianization of a Z-algebra of finite type. Hence S = Spec A is the projective
limit of the Si ’s, Si = Spec Ai . As f : X → S is of finite type, one can assume it is the
limit of fi : Xi → Si , i ∈ I and F is the filtering colimit of constructible Fi on Xi . In
the commutative diagram
colimi H q ((Xi )ét , Fi ) / colimi H q ((X0,i )ét , Fi )

 
H q (Xét , F ) / H q ((X0 )ét , F )

the vertical maps are isomorphism thanks to [SGA4, exp.VII, Th5.7]. As the Ai ’s are
excellent, Theorem 16.2.3 in the excellent case implies that the top horizontal map is
 98 BRUNO KLINGLER

an isomorphism. Finally the bottom horizontal map is also an isomorphism and the
result. 
Hence in the following we will be free to assume that A is excellent.
16.3.3. Reduction to the case F constant.
Proposition 16.3.3. Under the hypotheses of Theorem 16.2.3, suppose that for any n ≥
0 and any finite morphism X 0 → X, the restriction map H q (Xét
0 , Z/n) → H q ((X 0 ) , Z/n)
0 ét
is bijective for q = 0 and surjective for q > 0.
Then for any abelian torsion sheaf F on Xét and any q > 0:
∼
(32) H q (Xét , F ) → H q ((X0 )ét , F ) .
Proof. First, any torsion sheaf F is a filtered colimit of constructible sheaves by Corol-
c1
ici on utilise lary 15.4.8 c1 . As cohomology commutes with filtered colimits, it is enough to prove
A noetherien? eq. (32) for F constructible.
The proof for F constructible works as follows:
(1) H q (X, ·) : Abc (Xét ) → Ab and H q (X0 , ·) : Abc (Xét ) → Ab are cohomological
functors. Denote by ϕq : H q (X, ·) → H q (X0 , ·) the natural morphism.
(2) The functor H q (X, ·) : Abc (Xét ) → Ab Q is effaceable for q > 0. Indeed, let
F ∈ Abc (Xét ). The sheaf G0 := God0 (F ) = x∈X ix∗ Fx is an étale torsion sheaf on
X which is flasque. Writing G0 as a filtered colimit of constructible subscheaves, we see
that there exists F ⊂ G ⊂ G0 with G constructible such that H q (Xét , G) = 0 for all
q > 0.
(3) Every object of Abc (Xét ) is a sub-object of
Y
E := { pi ∗ Ci , pi : Xi → X finite, Ci constant} .
i

The result then follows from the equivalence (i) ⇔ (ii) in the following general homo-
logical lemma, whose proof by induction on q is left to the reader:
Lemma 16.3.4. Let A be an Abelian category, T • , T 0 • : A → Ab be two cohomological
functors, and E ⊂ A a full subcategory such that any object of A is a sub-object of an
object of E. Suppose T q is effaceable for all positive q.
Let ϕ : T • → T 0 • be a morphism of cohomological functors. The following conditions
are equivalent:
(i) ϕq (A) is a bijection for all q ≥ 0 and all objects A ∈ A.
(ii) ϕ0 (M ) is a bijection and ϕq (M ) is a surjection for all q > 0 and all objects
M ∈ E.
(iii) ϕ0 (A) is an isomorphism for all A ∈ A and T 0 q is effaceable for all q > 0.


16.3.4. The case q = 0, F constant (not necessarily finite). If Y is a scheme and F a

constant sheaf on Yét , H 0 (Yét , F ) = F π0 (Y ) . Hence Theorem 16.2.3 in this case follows
from Zariski’s connexity theorem:
Proposition 16.3.5. Let A be a local henselian noetherian ring, S = Spec A and f :
X → S a proper morphism. Then the natural morphism
π0 (X0 ) → π0 (X)
is an isomorphism.
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 99

Proof. Equivalently we have to show that the set OC(X) of clopen (closed and open)
subsets of X are in bijection with the set OC(X0 ) of clopen subsets of X0 . As OC(X)
(resp. OC(X0 )) is in bijection with the set Idem Γ(X, OX ) (resp. Idem Γ(X0 , OX0 )) we
have to show that the natural map
Idem Γ(X, OX ) → Idem Γ(X0 , OX0 )
is an isomorphism. Recall:
Theorem 16.3.6. (Finiteness of proper morphisms, see [EGAIII, 3.2]) Let S be a locally
Noetherian scheme and f : X → S a proper morphism. Then for any quasi-coherent
OX -module F and any non-negative integer q the sheaf Rq f∗ F is OS -coherent.
Applying this result for q = 0 gives in our case that Γ(X, OX ) is a finite A-algebra.
As A is henselian, it follows that Γ(X, OX ) is a product of local rings, equivalently that
the natural injection Idem Γ(X, OX ) → Idem (Γ(X,ˆOX )) is a bijection c1 . c1
cf. Raynaud
On the other hand f proper also implies (see [EGAIII, 4.1]) that Prop.4 p.2]
∼
Γ(X,ˆOX ) → lim Γ(Xn , OX ), n
n
hence ∼
Idem Γ(X, OX ) → lim Idem Γ(Xn , OXn ) .
n
But Xn and X0 have the same underlying topological space thus the righthandside
coincide with Idem Γ(X0 , OX0 ). 
16.3.5. Case q = 1 and F = Z/nZ. The group H 1 (Xét , Z/nZ) parametrizes isomor-
phism classes of étale Galois covers of X with Galois group Z/nZ c2 . Hence the result c2
Démontré
in this case follows from the more general. où?

Proposition 16.3.7. Let A be an henselian excellent ring with spectrum S. Let f :

X → S be a proper morphism. Then the natural functor
FEt(X) → FEt(X0 )
is an equivalence of categories (equivalently if X0 is connected: the natural morphism
π1 (X0 ) → π1 (X) is an isomorphism).
Proof. If X 0 , X 00 ∈ FEt(X), an X-morphism from X 0 to X 00 is defined by its graph,
which is clopen in X 0 ×X X 00 . Hence the full faithfullness of FEt(X) → FEt(X0 ) follows
from Proposition 16.3.5 applied to the proper morphism X 0 ×X X 00 → S.
It remains to show that FEt(X) → FEt(X0 ) is essentially surjective. Hence it is
enough to show that any étale cover h0 : Y0 → X0 extends to an étale cover h : Y → X.
Let us first assume S = Ŝ. In this case let us consider the commutative diagram:
j
X0
i / X̂ /X

  
s /S S,
where the map j is a flat morphism in the category of locally ringed spaces. We want
to show that the composite
j∗ i∗
FEt(X) → FEt(X̂) → FEt(X0 )
is essentially surjective.
As étale covers do not depend on nilpotents c3 , the finite étale cover h0 : Y0 → X0 can c3
cite reference
 100 BRUNO KLINGLER

be uniquely extended to an étale cover hn : Yn → Xn for all n ≥ 0, hence to an étale

c4
Definition? cover Y → X̂ in the category of formal schemes c4 . It remains to show that the formal
étale scheme Y → X̂ is the completion of an étale cover h : Y → X along Y0 . Recall:
Theorem 16.3.8. (Grothendieck’s algebraization theorem, [EGAIII, 5]) Let S be a com-
plete local ring and f : X → S a proper morphism. Then:
j∗
(a) The functor Coh(OX ) → Coh(OX0 ) is an equivalence of categories.
(b) The module M ∈ Coh(OX ) is locally free at any point of X0 if and only if
Mn ∈ Coh(OXn ) is locally free for any non-negative integer n.
This equivalence induces an equivalence between the category of finite X-schemes and
the category of finite X̂-schemes.
It follows from Theorem 16.3.8 that there exists a unique finite map h : Y → X such
that Y → X̂ is the completion of h. It remains to show that h : Y → X is étale.
c1
reference On the one hand the locus of Y where h : Y → X is étale is open in Y c1 . On the
other hand any open subset of Y containing Y0 is necessarily the all of Y as f is closed.
Hence it is enough to show that h : Y → X is étale (i.e. flat and unramified) at every
point y of Y0 .
The sheaf OY is OX -flat if and only if it is a colimit of locally free OX -sheaves. It
follows from Theorem 16.3.8(b) that being OX -free in restriction to X0 is equivalent to
being OX̂ -free on X0 . Hence the flatness of h : Y → X follows from the flatness of
Ŷ → X̂, which holds true as it is a formal étale morphism.
Let us show that Ω1Y /X vanishes. By [Stacks Project, Lemma 28.32.10], Ω1Y /X =
|Y0 |Y0
Ω1Y0 /X0 , hence vanishes as h0 : Y0 → X0 is étale.
In the general case, consider the commutative diagram:

X0
i ˆ j
/ X̂ = X /X /X

  

s / Ŝ Ŝ / S,

where the right hand square is Cartesian. Starting with the étale cover h0 : Y0 → X0 ,
the previous case applied to the two left squares furnishes a finite étale cover h : Y → X
extending h0 . Recall:
Theorem 16.3.9. (Artin’s approximation theorem) Let (A, m, k) be a local excellent ring
and F : A − Alg → Sets a functor locally of finite presentation. For every ξ ∈ F (Â),
there exists ξ ∈ F (A) such that ξ and ξ have the same image in F (k).
Consider the functor F : A − Alg → Sets which to an A-algebra B associates the
set FEt(X ⊗A B)/ ∼. One easily checks this is a functor of locally finite presentation
(i.e. commutes with filtering colimits). It follows from Artin’s Theorem 16.3.9 applied
to ξ := [h : Y → X] that there exists ξ = [h : Y → X] a finite étale morphism whose
restriction to Y0 is h0 . 
16.3.6. Reduction to the case f projective of relative dimension at most 1.
Proposition 16.3.10. Suppose that the Proper Base Change theorem (PBC) holds true
for f : X → S projective and S noetherian. Then it is true for general f .
Proof. We admit the following:
 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 101

Lemma 16.3.11. (Chow’s lemma, see [EGAII, 6.5.1]) Let f : X → S be a proper

morphism. Then there exists a commutative diagram
h /
X 0B X
BB
BB
B f
f ◦h BB 
S
such that h : X 0 → X is projective, surjective and an isomorphism over an open dense
subset of X, and f ◦ h is projective.
Lemma 16.3.12. Under the assumptions of Lemma 16.3.11, if (PBC) is true for h and
f ◦ h then it is true for f .
Proof. As f is projectif and surjectif let us first check that the natural adjunction mor-
phism of sheaves ϕ : F → h∗ h∗ F is injective. If x → X is a geometric point the (PBC)
for h and q = 0 implies that (h∗ h∗ F )x = Γ(Xx0 , F|Xx0 ). Hence we can assume that X = x
and h : X 0 → x is projective. In this case the identity of the abelian group F factorizes
ϕ
F → Γ(Xx0 , F|Xx0 ) → (h∗ F )x0 ' Fx ' F,
which shows that ϕ is injective.
Without loss of generality we can assume that F = h∗ L, with L an étale torsion
flasque sheaf on X 0 . Indeed, choose an injection h∗ F ,→ L0 with L0 torsion flasque.
Thus F ,→ h∗ L0 . Replacing F by Coker (F ,→ h∗ L0 ) and iterating, one obtains a
resolution
F ' L• ,
∼
where the Li ’s are étale torsion flasque sheaves on X 0 . We want to show RΓ(X, h∗ L• ) →
RΓ(X0 , (h∗ L• )|X0 ). Considering the hypercohomology spectral sequence, it is enough to
show that for each i, the morphism RΓ(X, h∗ Li ) → RΓ(X0 , (h∗ Li )|X0 ) is an isomorphism.
Consider the commutative diagram
RΓ(X, h∗ L) [0] / RΓ(X0 , (h∗ L)|X )
0

[1] [2]
 
RΓ(X 0 , L) RΓ(X0 , Rh∗ (L|X0 ))
RRR
RRR
[3]RRR ∼
RRR
( 
RΓ(X00 , L|X00 ).

To show that [0] is an isomorphism, it is enough to show that [1], [2], [3] are isomorphisms.
For [1]: as L is flasque, h∗ L = Rh∗ L hence the result.
For [3]: this is (PBC) for the projective morphism f ◦ h.
For [2]: apply (PBC) to the projective morphism h : X 0 → X. It follows that
Rh∗ (L|X00 ) ' (Rh∗ L)|X0 ' (h∗ L)|X0 ,
where the last equality follows from the fact that L is flasque. 

102 BRUNO KLINGLER

Proposition 16.3.13. If (PBC) is true for f : X → S projective of relative dimension

at most 1 then (PBC) is true.
Proof. From Proposition 16.3.10, it is enough to prove that if one has a commutative
diagram

X / P := Pn
i
/ S
pp p
pp
f
ppp
 pw pp
S = Spec A
then the morphism RΓ(X, F ) → RΓ(X0 , F|X0 ) is an isomorphism for any torsion étale
sheaf F on X.
As the diagram
RΓ(X, F ) / RΓ(X0 , F|X )
0

∼ ∼
 
RΓ(P, i∗ F ) [1] / RΓ(P0 , i∗ F |P )
0

commutes, it is enough to prove that [1] is an isomorphism, i.e we are reduced to the
case X = PnS .
For n = 1 it follows from our hypothesis.
Let n > 1 and suppose by induction hat (PBC) is true for any projective morphism
of relative dimension at most n − 1. Let t0 , · · · , tn be homogeneous coordinates on PnS .
Consider the pencil of hypersurfaces Hλ := {λt0 + (1 − λ)t1 = 0}, λ ∈ P1S with base
locus ∆ := H0 ∩ H1 and the blow-up diagram
X 0 := Bl∆ X /X

f0 f
 
P1S / S.
g

By Lemma 16.3.12 (PBC) for f will follow from (PBC) for h and f ◦ h = g ◦ f 0 .
But h is projective of relative dimension at most one hence (PBC) holds for h by
hypothesis.
For g ◦ f 0 , consider the diagram
RΓ(X 0 , F ) [0] / RΓ(X 0 , F 0 )
0 |X0

RΓ(P1S , Rf∗0 F )) 0 (F 0 ))
RΓ(P1S,0 , Rf0∗ |X0
jj4
jjjjj
jjj[2]
[1]
 jjjj
RΓ(P1S,0 , (Rf∗0 F )P1 ).
S,0

The morphism [1] is the (PBC) for g, hence is an isomorphism as g is projective of

relative dimension 1. By induction on n, the morphism [2] is an isomorphism as f 0 is
projective of relative dimension at most n − 1.

 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 103

16.3.7. The case f projective of relative dimension at most 1: end of the proof of the
Proper Base Change theorem. If follows from the previous steps it is enough to show:
Proposition 16.3.14. Let S be the spectrum of an excellent strictly henselian ring
(A, m, k) and f : X → S a proper morphism. Then for any integers n > 1 and q ≥ 0
the restriction morphism H q (Xét , Z/n) → H q ((X0 )ét , Z/n) is an isomorphism for q = 0
and a surjection for q > 0.
Proof. The cases q = 0 and q = 1 are treated in Proposition 16.3.5 and Proposition 16.3.7
respectively for any X.
Under our assumptions X0 is a point or a projective curve over the algebraically closed
field k hence H q (X0 , Z/n) = 0 by [SGA4, IX 5.7] (we proved it for X0 smooth projective
and n invertible on X0 in Corollary 14.0.3).
It remains to prove the statement for q = 2. Without loss of generality we can assume
that n = lr , l prime, then n = l. There are two cases:
Either l = p = chark, in which case H 2 (X0 , Z/p) = 0. Indeed, the Artin-Schreier
exact sequence of étale sheaves on X0
F −1
0 → Z/p → OX0 → OX0 → 0
induces an exact sequence of groups
F −1
H 1 (X0 , OX0 ) → H 1 (X0 , OX0 ) → H 2 (X0 , Z/p) → 0 .
The result follows from the semi-linear algebra lemma:
Lemma 16.3.15. c1 Let k be a separably closed field of positive characteristic p, V a c1
reference
finite dimensional k-vector space and ϕ : V → V and F -linear map. Then F −1 : V → V
is surjective.
In the case l 6= p, identify Z/n = µn . The Kummer exact sequence of étale sheaves
on X0
0 → µn → Gm → Gm → 0
induces an exact sequence
Pic (X0 ) → H 2 (X0 , µn ) → H 2 (X0 , Gm ) = 0
(once more we showed this exact sequence for X0 smooth). c2 Consider the commutative c2
reference
diagram
Pic (X) [1] / H 2 (Xét , µ )
n

 
Pic (X0 ) / / H 2 ((X0 )ét , µ ).
n

The surjectivity of [1] follows from the

Proposition 16.3.16. Let S = Spec A with A a local noetherian henselian ring and
f : X → S a proper morphism of relative dimension at most 1. Then the restriction
map Pic X → Pic X0 is surjective.
Proof. One can assume without loss of generality that S is excellent.
Consider the diagram eq. (31) Let L0 be an invertible sheaf of X0 and suppose that
L0 has been extended to an invertible sheaf Ln on Xn . The obstruction to extending
Ln to Xn+1 lies in H 2 (X0 , mn+1 /mn ) = H 2 (X0 , OX0 ) ⊗A mn+1 /mn , which vanishes has
dim X0 ≤ 1.
104 BRUNO KLINGLER

Considering again the commutative diagram

X0
i ˆ j
/ X̂ = X /X /X

  

s / Ŝ Ŝ / S,

ˆ extending L .
it follows that there exists a formal invertible sheaf L on X 0
By Grothendieck’s Theorem 16.3.8, there exists a unique invertible sheaf L on X such
that L ˆ ' L.
Consider the functor F : A − Alg → Sets which to an A-algebra B associates the
set FEt(X ⊗A B)/ ∼. One easily checks this is a functor of locally finite presentation
(i.e. commutes with filtering colimits). It follows from Artin’s Theorem 16.3.9 applied
to ξ := [h : Y → X] that there exists ξ = [h : Y → X] a finite étale morphism whose
restriction to Y0 is h0 . 

 ETALE COHOMOLOGY AND THE WEIL CONJECTURES 105

References
[Aud12] M.Audin, La guerre des recensions, Math. Semesterberichte, 2012,
http://arxiv.org/pdf/11095230.pdf
[Bei07] Lectures on the Weil conjectures, http://www.math.lsa.umich.edu/ mityab/beilinson/
[B] Bourbaki, Algèbre Commutative
[EGA0] Eléments de Géométrie Algébrique 0
[EGAII] Eléments de Géométrie Algébrique II
[EGAIII] Eléments de Géométrie Algébrique III
[EGAIV] Eléments de Géométrie Algébrique IV
[El75] W.J. Ellison, Les nombres premiers, in collaboration with Michel Mendès France, Publications de
l’Institut de Mathématiques de l’Université de Nancago, IX, Hermann, Paris, (1975)
[ES52] S.Eilenberg, N. Steenrod, Foundations of algebraic topology, Princeton University Press, Prince-
ton, New Jersey, 1952
[Ful98] W.Fulton, Intersection Theory, Second Edition, Ergebnisse der Mathematik und ihrer Grenzge-
biete, (1998)
[Har77] R. Hartshorne, Algebraic Geometry, Graduate Texts in Mathematics 52, Springer-Verlag, New
York-Heidelberg, 1977
[Ha36] H. Hasse, Zur Theorie der abstrakten elliptischen Funktionenkörper, I, II, III, J. reine angew.
Math 175 (1936), 55-62, 69-88, 193-207
[Il96] Illusie L., Frobenius et dégénérescence de Hodge, Panorama et Synthèse 3 (1996)
[Lef26] S. Lefschetz, Intersections and transformations of complexes and manifolds, Trans. Amer. Math.
Soc. 28 1-49 (1926)
[Mat80] H. Matsumura, Commutative algebra, second edition, Mathematics Lecture Note Series 56,
Benjamin/Cummings Publishing (1980)
[Mi80] J.Milne, Étale cohomology, Princeton Mathematical Series 33, Princeton University Press, (1980)
[Po95] H. Poincaré, Analysis situs in Oeuvres Tome VI, Les Grands Classiques Gauthier-Villars, Editions
Jacques Gabay, Sceaux, 1996
[Ray70] M. Raynaud, Anneaux locaux henséliens, LNM 1969 Springer-Verlag (1970)
[Sch31] F.K. Schmidt, Analytische Zahlentheorie in Körpern der Characteristik p, Math. Zeit. 33 (1931),
1-22
[Se56] J.P. Serre, Géométrie algébrique et géométrie analytique, Ann. Inst. Fourier 6 (1955-1956)
[Se97] J.P.Serre, Galois cohomology, Springer Monographs in Mathematics, Springer Verlag, 1997
[SGA1] A. Grothendieck, Revêtements étales et groupe fondamental, Séminaire de Géométrie Algébrique
I, IHES (1963)
[SGA4] A. Grothendieck, Séminaire de Géométrie Algébrique IV, IHES (1963)
[Sil09] J.H. Silverman, The arithmetic of elliptic curves, Second edition, Graduate Texts in Mathematics
106 Springer, Dordrecht 2009
[Stacks Project] Stacks Project, http://stacks.math.columbia.edu/
[Vois07] C. Voisin, Hodge theory and complex algebraic geometry, Cambridge Studies in Advanced Math-
ematics 76, Cambrige University Press, 2007
[We40] A. Weil, Sur les fonctions algébriques à corps de constantes fini, C. R. Acad. Sci. Paris 210
(1940)
[We49] A. Weil, Numbers of solutions of equations in finite fields, Bull. Amer. Math.Soc. 55 497-508
(1949)