Grothendieck Topology

Download as pdf or txt
Download as pdf or txt
You are on page 1of 143
At a glance
Powered by AI
The lectures aim to provide an introduction to Grothendieck's theory of the fundamental group in algebraic geometry. In particular, the lectures cover the fundamental group of an algebraic curve over an algebraically closed field of arbitrary characteristic.

The purpose of these lectures was to give an introduction to Grothendieck’s theory of the fundamental group in algebraic geometry with, as application, the study of the fundamental group of an algebraic curve over an algebraically closed field of arbitrary characteristic.

We assume that the reader is somewhat familiar with the notion and elementary properties of preschemes. Specifically chapters I and II of EGA as well as the fundamental theorems of EGA chapter III are assumed as prerequisites.

Lectures on

An Introduction to Grothendiecks Theory of


the Fundamental Group
By

J.P. Murre

Notes by

S. Anantharaman

No part of this book may be reproduced in any


form by print, microfilm or any other means without written permission from the Tata Institute of
Fundamental Research, Colaba, Bombay 5

Tata Institute of Fundamental Research, Bombay


1967

Preface
These lectures contain the material presented in a course given at the
Tata Institute during the period December 1964 - February 1965. The
purpose of these lectures was to give an introduction to Grothendiecks
theory of the fundamental group in algebraic geometry with, as application, the study of the fundamental group of an algebraic curve over
an algebraically closed field of arbitrary characteristic. All of the material (and much more) can be found in the Seminaire de geometrie
algebrique of Grothendieck, 1960-1961 Expose V, IX and X.
I thank Mr. S. Anantharaman for the careful preparation of the notes.

J.P. Murre

iii

Prerequisites
We assume that the reader is somewhat familiar with the notion and
elementary properties of preschemes. To give a rough indication: Chap.
I, 1-6 and Chap. II, pages 1-14, 100-103 and 110-114 of the EGA
ements de Geometrie Algebrique of Grothendieck and Dieudone).
(El
We have even recalled some of these required elementary properties in
Chap. I and II of the notes but this is done very concisely.
We need also all the fundamental theorems of EGA, Chap. III (first
part); these theorems are stated in the text without proof. We do not
require the reader to be familiar with them; on the contrary, we hope
that the applications which have been made will give some insight into
the meaning of, and stimulate the interest in, these theorems.

Contents
Preface

iii

Prerequisites

1 Affine Schemes
1.1
. . . . . . . . . . . . . . . .
1.2
. . . . . . . . . . . . . . . .
1.4 The Sheaf associated to Spec A
1.5
. . . . . . . . . . . . . . . .
1.6 Affine Schemes . . . . . . . .

.
.
.
.
.

1
1
1
3
4
6

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

2 Preschemes
2.2 Product of Preschemes . . . . . . . .
2.3 Fibres . . . . . . . . . . . . . . . . .
2.4 Subschemes . . . . . . . . . . . . . .
2.5 Some formal properties of morphisms
2.6 Affine morphisms . . . . . . . . . . .
2.7 The finiteness theorem . . . . . . . .

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

9
10
12
14
17
19
20

3 Etale
Morphisms and Etale
Coverings
3.2 Examples and Comments . . . . .
3.3
. . . . . . . . . . . . . . . . . .
3.4
. . . . . . . . . . . . . . . . . .

3.5 Etale
coverings . . . . . . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

23
24
25
34
39

vii

.
.
.
.

.
.
.
.

Contents

viii
4

The Fundamental Group


4.1 Properties of the category C . . . . . .
4.2
. . . . . . . . . . . . . . . . . . . . .
4.3
. . . . . . . . . . . . . . . . . . . . .
4.4 Construction of the Fundamental group

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

43
43
46
49
52

Galois Categories and Morphisms of Profinite Groups


69
5.1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.2
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

Application of the Comparison Theorem....


6.1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.2 The Stein-factorisation . . . . . . . . . . . . . . . . . .
6.3 The first homotopy exact sequence . . . . . . . . . . . .

The Technique of Descents and Applications


89
7.1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
7.2
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
7.3
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

An Application of the Existence Theorem


117
8.1 The second homotopy exact sequence . . . . . . . . . . 117

The Homomorphism of Specialisation.....


125
9.1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
9.2
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
9.3
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

Appendix to Chapter IX

75
75
77
82

127

Chapter 1

Affine Schemes
1.1
1

Let A be a commutative ring with 1 and S a multiplicatively closed


a
set in A, containing 1. We then form fractions , a A, s S ; two
s
a1 a2
,
are considered equal if there is an s3 S such that
fractions
s1 s2
s3 (a1 s2 a2 s1 ) = 0. When addition and multiplication are defined in the
obvious way, these fractions form a ring, denoted by S 1 A and called the
ring of fractions of A with respect to the multiplicatively closed set S .
There is a natural ring-homomorphism A S 1 A given by a 7 a/1.
This induces a(1 1) correspondence between prime ideals of A not
intersecting S and prime ideals of S 1 A, which is lattice-preserving. If
f A and S f is the multiplicatively closed set {1, f, f 2 , . . .}, the ring of
quotients S 1
f A is denoted by A f . If p is a prime ideal of A and S = A p,
the ring of quotients S 1 A is denoted by A p ; A p is a local ring.

1.2
Let A be a commutative ring with 1 and X the set of prime ideals of A.
For any E A, we define V(E) as the subset {p : p a prime ideal E}
of X. Then the following properties are easily verified:
T
S
(i)
V(E ) = V( E )
2

1. Affine Schemes

2
(ii) V(E1 ) V(E2 ) = V(E1 E2 )
(iii) V(1) =
(iv) V(0) = X.

Thus, the sets V(E) satisfy the axioms for closed sets in a topology
on X. The topology thus defined is called the Zariski topology on X; the
topological space X is known as Spec A.
Note. Spec A is a generalisation of the classical notion of an affine algebraic variety.
Suppose k is an algebraically closed field and let k[X1 , . . . , Xn ] =
k[X] be the polynomial ring in n variables over k. Let a be an ideal of
k[X] and V be the set in kn defined by V = {(1 , . . . , n ) : f (1 , . . . , n )
= 0 f a}. Then V is said to be an affine algebraic variety and the
Hilberts zero theorem says that the elements of V are in (1 1) correspondence with the maximal ideals of k[X]/a.
Remarks 1.3. (a) If a(E) is the ideal generated by E in A, then we
have: V(E) = V(a(E))
(b) For f A, define X f = X V( f ); then the X f form a basis for the
S
Zariski topology on X. In fact, X V(E) =
X f by (i).
f E

(c) X is not in general Hausdorff; however, it is T 0 .


S
(d) X f X f an n Z+ such that f n is in the ideal generated

by the f s.

For any ideal a of A, define a = {a A : an a for some

a is an ideal of A and we assert that a = {p :


n Z+ }.
p a prime ideal a}. To prove this, we assume, (as we may, by

passing to A/a) that a = (0). Clearly, (0) p. On the other


hand, if a A is such that an , 0n Z+ , then S a1 A = Aa is a
non-zero ring and so contains a proper prime ideal; the lift p of
this prime ideal in A is such that a < p.

1.4. The Sheaf associated to Spec A

It is thus seen that V(a) = V( a) and V( f ) V(a) f a.


Hence:
[
\
[
Xf
X f V( f )
V( f ) = V( { f })

and this proves (d).


(e) The open sets X f are quasi-compact.
In view of (b), it is enough to consider coverings by Xg s; thus, if
r
P
S
ai fi (say); again by
X f X f , then by (d), we have f n =

(d), we obtain X f = X f n

r
S

i=1

i=1

X fi .

(f) There is a (1 1) correspondence between closed sets of X and 4


roots of ideals of A; in this correspondence, closed irreducible
sets of X go to prime ideals of A and conversely. Every closed
irreducible set of X is of the form (x) for some x X; such an x is
called a generic point of that set, and is uniquely determined.
(g) If A is noetherian, Spec A is a noetherian space (i.e. satisfies the
minimum condition for closed sets).

1.4 The Sheaf associated to Spec A


We shall define a presheaf of rings on Spec A. It is enough to define
the presheaf on a basis for the topology on X, namely, on the X f s; we
set F (X f ) = A f . If Xg X f , V(g) V( f ) and so gn = a0 f for some
q
aa
n Z+ and a0 A. The homomorphism A f Ag given by af q 7 qn 0
g
is independent of the way g is expressed in terms of f and thus defines
f
a natural map g : F (X f ) F (Xg ) for Xg X f . The transitivity
conditions are readily verified and we have a presheaf of rings on X.
e = OX of rings on X. It is easy to check that the
This defines a sheaf A
stalk O p,X of OX at a point p of X is the local ring A p .

1. Affine Schemes

If M is an A-module and if we define the presheaf X f 7 M f =


e of A-modules
e
e
e
M A A f , we get a sheaf M
(in short, an A-Module
M),
whose stalk at p X is M A A p = MP .
5

Remark . The presheaf X f 7 M f is a sheaf, i.e. it satisfies the axioms


(F1 ) and (F2 ) of Godement, Theorie des faisceaux, p. 109.

1.5
In this section, we briefly recall certain sheaf-theoretic notions.
1.5.1 Let f : X Y be a continuous map of topological spaces. Suppose F is a sheaf of abelian groups on X; we define a presheaf of abelian
groups on Y by U 7 ( f 1 (U), F ) for any open U Y; for V U,
open in Y, the restriction maps of this presheaf will be the restriction homomorphisms ( f 1 (U), F ) ( f 1 (V), F ). This presheaf is already
a sheaf. The sheaf defined by this presheaf is called the direct image
f (F ) of F under f .
If U is any neighbourhood of f (x) in Y, the natural homomorphism
( f 1 (U), F ) Fx given a homomorphism (U, f(F )) Fx ; by
passing to the inductive limit as U shrinks down to f (x), we obtain a
natural homomorphism:
f x : f (F ) Fx .
f (x)

If OX is a sheaf of rings on X, f (OX ) has a natural structure of a


sheaf of rings on Y. If F is an OX -Module, f (F ) has a natural structure
of an f (OX )-Module.
The direct image f (F ) is a covariant functor on F .
6

1.5.2 Let f : X Y be a continuous map of topological spaces and g


be a sheaf of abelian groups on Y. Then it can be shown that there is a
unique sheaf F of abelian groups on X such that:
(a) there is a natural homomorphism of sheaves of abelian groups
= g : g f (F )

1.5.

5
and

(b) for any sheaf H of abelian groups on X, the homomorphism


HomX (F , H ) HomY (g, f (H )) given by 7 f () g is
an isomorphism. The unique sheaf F of abelian groups on X
with these properties is called the inverse image f 1 (g) of g under
f.
It can be shown that canonical homomorphism
(*)

fx f (x) : g f (x) f 1 (g) x

is an isomorphism, for every x X.


The inverse image f 1 (g) is a covariant functor on g and the isomorphism () shows that it is an exact functor.
If OY is a sheaf of rings on Y, f 1 (OY ) has a natural structure of a
sheaf of rings on X. If g is an OY -Module, f 1 (g) has a natural structure
of an f 1 (OY )-Module.
1.5.3 A ringed space is a pair (X, OX ) where X is a topological space 7
and OX is a sheaf of rings on X, called the structure sheaf of (X, OX ). A
morphism : (X, OX ) (Y, OY ) of ringed spaces is a pair ( f, ) such
that
(i) f : X Y is a continuous map of topological spaces, and
(ii) : OY f (OX ) is a morphism of sheaves of rings on Y.
Ringed spaces, with morphisms so defined, form a category. Observe that condition (ii) is equivalent to giving a morphism f 1 (OY )
OX of sheaves of rings on X (see (1.5.2)).
If F is a sheaf of OX -modules, we denote by (F ) the sheaf
f (F ), considered as an OY -Module through . If g is an OY -Module,
f 1 (g) is an f 1 (OY )-Module and the morphism f 1 (OY ) OX , defined by , gives an OX -Module f 1 (g) f 1 (OY ) OX ; the stalks of this
OX -Module are isomorphic to g f (x) O f (x) O x , under the identification
f 1 (g) x g f (x) . We denote this OX -Module by (g). In general, is
not an exact functor on g.

1. Affine Schemes

1.6 Affine Schemes


e A a ring, defined in (1.4) is
A ringed space of the form (Spec A, A),
called an affine scheme.
8

1.6.1 Let : B A be a ring-homomorphism; defines a map


f = a : X = Spec A Spec B = Y
p 7 1 (p).

Since a 1 (V(E)) = V((E)) for any E B, a is a continuous map.


Let s B; defines, in a natural way, a homomorphism
s : B s A(s) .
In view of the remark at the end of (1.4), this gives us a homomorphism:
e A (s) = (X(s) , A)
e = (Y s , f (A))
e
B s = (Y s , B)

e f (A).
e If x X the stalk map
and hence a homomorphism e
: B
defined by e
, namely
e
x : O f (x) B f (x) O x A x

is a local homomorphism (i.e. the image of the maximal ideal in B f (x) is


contained in the maximal ideal of A x ).
9

e (Spec B, e
Definition 1.6.2. A morphism : (Spec A, A)
B) of two
affine schemes, is a morphism of ringed spaces with the additional property that is of the form (a , e
) for a homomorphism : B A of
rings.
It can be shown that a morphism = ( f, ) of ringed spaces is
e
a morphism of affine schemes (spec A, A)(spec
B, e
B) if and only if the
stalk-maps
O f (x) O x
are local homomorphisms.

defined by

(rather, by )

1.6. Affine Schemes

e is an exact covariant
Remarks 1.6.4. (a) If M is an A-module, M
functor on M.
e N)
e is canonically isomorphic
(b) For any A-modules M, N, HomAe( M,
to HomA (M, N).
e (Y, OY ) = (spec B, e
(c) If (X, OX ) = (spec A, A),
B) are affine, there is
a natural bijection from the set Hom(X, Y) of morphisms of affine
schemes X Y onto the set Hom(B, A) of ring-homomorphisms
B A.
e be an affine scheme and F an OX (d) Let (X, OX ) = (Spec A, A)
module. Then one can show that F is quasi-coherent (i.e. for
every x X, an open neighbourhood U of x and an exact se(I)
(J)
e for an
quence (OX |U) (OX |U) F |U 0) F M
A-module M. If we assume that A is noetherian, one sees that F
e for a finite type A-module M.
is coherent F is to M

(e) Let X Y be a morphism of affine schemes and F (resp. g) 10


be a quasicoherent OX -Module (resp. OY -Module). Then one can
define a quasi-coherent OY -Module (resp. OX -Module) denoted
by (F ) (resp. (g)) just as in (1.5.3); if X = Spec A, Y =
e M an ASpec B and = (a , e
) for a : B A and if F = M,
e
module (resp. g = N, N a B-module) then (F ) (resp. (g)) is
gM , where [] M is the abelian group
canonically identified with []
M considered as a B-module through (resp. M]
B A).
(For proofs see EGA Ch. I)

Chapter 2

Preschemes
Definition 2.1. A ringed space (X, OX ) is called a prescheme if every 11
point x X has an open neighbourhood U such that (U, OX |U) is an
affine scheme.
An open set U such that (U, OX |U) is an affine scheme is called an
affine open set of X; such sets form a basis for the topology on X.
Definition 2.1.1. A morphism : (X, OX ) (Y, OY ) of preschemes
is a morphism ( f, ) of ringed spaces such that for every x X, the
stalk-map x : O f (x) O x defined by is a local homomorphism.
Preschemes then form a category (Sch). In referring to a prescheme,
we will often suppress the structure sheaf from notation and denote
(X, OX ) simply by X.
2.1.2 Suppose C is any category and S Ob C . We consider the pairs
(T, f ) where T Ob C and f HomC (T, S ).
If (T 1 , f1 ), (T 2 , f2 ) are two such pairs, we define Hom((T 1 , f1 ),
(T 2 , f2 )) to be the set of C -morphisms : T 1 T 2 , making the diagram
T1 @

@@
@@
f1 @@


S
9

/ T2
~
~
~
~~ f2
~
~

2. Preschemes

10

12

commutative.
This way we obtain a category, denoted by C |S . In the special case
C = (Sch), the category (Sch /S ) = (Sch)|S is called the category of S preschemes; its morphisms are called S -morphisms. S itself is known
as the base prescheme of the category.
Remark 2.1.3. Let Spec A be an affine scheme and Y any prescheme.
Then Hom(A, (Y, OY )) is naturally isomorphic to Hom(Y, Spec A).
In fact, let (Ui ) be an affine open covering of Y and Hom
(A, (Y, OY )). The composite maps

restriction

i : A
(Y, OY ) (Ui , OY )
give morphisms a i : Ui Spec A, for every i, since the Ui are affine.
It is easily checked that a i = a j on Ui U j , i, j. We then get
a morphism a : Y Spec A; the map 7 a is a bijection from
Hom(A, (Y, OY )) onto Hom(Y, Spec A) (cf. (1.6.4)(c)).
It follows that every prescheme X can be considered as a Spec Zprescheme in a natural way:
(Sch) = (Sch / Spec Z) = (Sch /Z).
13

Remark 2.1.4. Let (X, OX ) be a prescheme and F an OX -module. Then


it follows from (1.6.4)(d) that F is quasi-coherent for every x X
eU , for a (U, OX )
and any affine open neighbourhood U of x, F |U M
- module MU .
We may take this as our definition of a quasi-coherent OX -module.

2.2 Product of Preschemes


2.2.0 Suppose (X, f ), (Y, g) are S -preschemes. We say that a triple
(Z, p, q) is a product of X and Y over S if:
(i) Z is an S -prescheme
(ii) p : Z X, q : Z Y are S -morphisms and

2.2. Product of Preschemes

11

(iii) for any T (Sch /S ), the natural map:


HomS (T, Z) HomS (T, X) HomS (T, Y)
f 7 (p. f, q. f )

is a bijection.
The product of X and Y, being a solution to a universal problem, is
obviously unique upto an isomorphism in the category. We denote the
product (Z, p, q), if it exists by XY and call it the fibre-product of X
S

and Y over S ; p, q are called projection morphisms.


Theorem 2.2.1. If X, Y (Sch /S ), the fibre-product of X and Y over S
always exists.
We shall not prove the theorem here. However, we observe that if 14
X = Spec A, Y = Spec B and S = Spec C are all affine, then Spec(A B)
C

is a solution for our problem. In the general case, local fibre-products


are obtained from the affine case and are glued together in a suitable
manner to yield a fibre product XY.
S

(For details see EGA, Ch. I, Theorem (3.2.6)).


Remarks.

(1) The underlying set of XY is not the fibre-product of


S

the underlying sets of X and Y over that of S . However if x X,


y Y lie over the same s S , then there is a z XY lying over
S

x and y. (For a proof see Lemma (2.3.1)).

(2) An open subset U of a prescheme X can be considered as a prescheme in a natural way. Suppose S S , U X, V Y are open
sets such that f (U) S , g(V) S ; we may consider U, V as
S -preschemes. When this is done, the fibre-product U V is isoS

morphic to the open set p1 (U) q1 (V) in Z = XY, considered


S

as a prescheme.

This follows easily from the universal property of the fibre-product.

2. Preschemes

12

2.2.2 Change of base: Let X, S be S -preschemes. Then the fibeproduct XS can be considered as an S -prescheme in a natural way:
S

XS

Xo

S o
15

When this is done, we say that XS is obtained from X by the baseS

change S S and denote it by X(S ) . Note that, in the affine case, this
corresponds to the extension of scalars.
If X is any prescheme, by its reduction mod p, p Z+ , (resp.
mod p2 and so on) we mean the base-change corresponding to Z
Z/(p) (resp. Z Z/(p2 ) and so on).
If f : X X ., g : Y Y are S -morphisms f and g define, in
a natural way an S -morphism: XY X Y , which we denote by
S

f g or by ( f, g)S . When g = IS : S S , we get a morphism:


S

.
f IS = f(S ) : X(S ) X(S
)
S

2.3 Fibres
Let (X, f ) be an S -prescheme and s S be any point. Let U S be
an affine open neighbourhood of s and A = (U, OS ). If ps is the prime
ideal of A corresponding to s, O s,S is identified with A ps . Denote by k(s)
the residue field of O s,S = A ps . The composite A A ps k(s) defines
a morphism Spec k(s) Spec A = U S ; i.e. to say, Spec k(s) is an S prescheme in a natural way. Consider now the base-change Spec k(s)
S:
p
X = X Spec k(s)
Xo
S
q

S o
16

Spec k(s)

The first projection p clearly maps X into the set f 1 (s) S . We

2.3. Fibres

13

claim that p(X ) = f 1 (s) and further that, when we provide f 1 (s) with
the topology induced from X, p is a homomorphism between X and
f 1 (s).
To prove this, it suffices to show that for every open set U in a covering of X, p is a homeomorphism from p1 (U) onto U f 1 (s). In view
of the remark (2) after Theorem (2.2.1) we may then assume that X, S
are affine say X = Spec A, S = Spec C. That p(X ) = f 1 (s) will follow
as a corollary to the following more general result.
Lemma 2.3.1. Let X = Spec A, Y = Spec B be affine schemes over
S = Spec C. Suppose that x X, y Y lie over the same element s S .
Then the set E of elements z Z = XY lying over x, y is isomorphic to
S

Spec(k(x) k(y)) (as a set).


k(S )

Proof. One has a homomorphism : A B k(x) k(y) which gives


C

k(s)

a morphism a : Spec k(x) k(y) Z; clearly the image of a is


k(s)

contained in the set E = p1 (x) q1 (y). That a is injective is seen


by factoring as follows: A B A x By k(x) k(y). In order
C

Cs

k(s)

to see that a is surjective one remarks that for z E the homomorphism A A B k(z) factors through k(x), similarly for B, therefore
C

we have for AB k(z) a factorisation AB k(x) k(y) k(z).


C

k(s)

Q.E.D.

In the above lemma if we take B = k(s) it follows that in the diagram 17


X = X Spec k(s) = Spec(Ak(s))
S

q
qqq
q
q
qq
qqqp
q
q
qq
x qq
q

X = SpecNA

NNN
NNN
NNNf
NNN
NNN
NNN
'

C
MMM
MMM
MMM
M
q MMM
MMM
MM&

Y = Spec B

pp
ppp
g pppp
pp
ppp
p
p
pw pp

s S = Spec C

2. Preschemes

14
the map p : X f 1 (s) is a bijection.

2.3.2 Returning to the assertion that X is homeomorphic to the fibre


f 1 (s) (with the induced topology from X) we note that if : C k(s)
is the natural map, then p : X X is the morphism corresponding to
1 : A Ak(s). To show that p carries the topology over, it is
C

18

enough to show that any closed set of X , of the form V(E ), is also of
the form V((1 )E) for some E A.
!
P
ci
=
Now, any element of Ak(s) can be written in the form ai
C
t
i !
!
!
P
1
1
(ai ci ) 1
with ai A, ci , t C. Since 1
is a unit
t
i
Pt
of Ak(s), we can take for E A, the set of elements ai ci where
C
i
 
P
ai ci /t is an element of E .
Q.E.D.
i

Note. The fibre f 1 (s) can be given a prescheme structure through this
homeomorphism p : X f 1 (s). If, in the above proof, we had taken
O s /M sn+1 , instead of k(s) = O s /M s we
 would still have obtained homen+1
omorphisms pn : X Spec O s /M s
f 1 (s). The prescheme strucS

ture on f 1 (s) defined by means of pn , is known as the nth -infinitesimal


neighbourhood of the fibre.

2.4 Subschemes
2.4.0 Let X be a prescheme and J a quasi-coherent sheaf of ideals of
OX . Then the support Y of the OX -Module OX /J is closed in X and
(Y, OX/J |Y) has a natural structure of a prescheme. In fact, the question
is purely local and we may assume X = Spec A. Then J is defined
by an ideal I of A and Y corresponds to V(I) which is surely closed.
The ringed space (Y, OX/J |Y) has then a natural structure of an affine
scheme, namely, that of Spec(A/I).
Such a prescheme is called a closed subscheme of X. An open subscheme is, by definition, the prescheme induced by X on an open subset

2.4. Subschemes

15

in a natural way. A subscheme of X is a closed subscheme of an open


subscheme of X.
2.4.1 A subscheme may have the same base-space as X. For example, 19
one can show that there is a quasi-coherent sheaf N of ideals of OX
such that N x = nil-radical of O x . N defines a closed subscheme Y of
X, which we denote by Xred and which is reduced, in the sense that the
stalks Oy,Y of Y have no nilpotent elements. X and Y have the same base
space (A and Ared have the same prime ideals).
Consider a morphism f : X Y of preschemes. Suppose X :
Xred X, Y : Yred Y are the natural morphisms. Then a morphism
fred : Xred Yred making the diagram
XO

/Y
O
Y

Xred

/ Yred

fred

commutative. This corresponds to the fact that a homomorphism :


A B of rings defines a homomorphism red : Ared Bred such that

A
A

/B

Ared

red


/ Bred

2.4.2 A morphism f : Z X is called an immersion if it admits a 20


f

factorization Z Y
X where Y is a sub-scheme of X, j : Y X
is the canonical inclusion and f : Z Y is an isomorphism. The
immersion f is said to be closed (resp. open) if Y is a closed subscheme
(resp. open subscheme) of X.

2. Preschemes

16
f

Example . Let X
S be an S -prescheme. Then, there is a natural
S -morphism : X XX such that the diagram
S

ll X

l
l
lll


ulll

X `BB

BB


B
IX

p1 B
IX =idXllll

XX
S

BB
BB
BB

X
l
l
lll
lll
l
l
 lu lll f
p2

is commutative. is called the diagonal of f . It is an immersion.


Definition 2.4.2.1. A morphism f : X S is said to be separated (or X
is said to be an S -scheme) if the diagonal : X XX of f is a closed
S

immersion.

A prescheme X is called a scheme if the natural map X Spec Z is


separated.
21

Remark . Let Y be an affine scheme, X any prescheme and (U ) an


affine open c over for X. One can then show that a morphism f : X Y
is separated if and only if , ,
(i) U U is also affine
(ii) (U U , OX ) is generated as a ring by the canonical images of
(U , OX ) and (U , OX ).
(For a proof see EGA, Ch. I, Proposition (5.5.6)).
2.4.3 Example of a prescheme which is not a scheme. Let B = k[X],
C = k[Y] be polynomial rings over a field k. Then Spec BX and Spec CY
are affine open sets of Spec B and Spec C respectively; the isomorphism

2.5. Some formal properties of morphisms

17

f (Y)

f (X)
7
of BX onto CY defines an isomorphism Spec CY

m
m
X
Y
Spec BX . By recollement of Spec B and Spec C through this isomorphism, one gets a prescheme S , which is not a scheme; in fact, condition (ii) of the proceding remark does not hold: for, (Spec B, OS )
B = k[X] and (Spec C, OS ) C = k[Y]; the canonical maps from these
into (Spec B Spec C, OS ) k[u, u1 ] are given by X 7 u, Y 7 u and
the image in each case is precisely = k[u].

2.5 Some formal properties of morphisms


(i) every immersion is separated
(ii) f : X Y, g : Y Z separated g f : X Z separated.
(iii) f : X Y a separated S -morphism f(S ) : X(S ) Y(S ) is
separated for every base-change S S .
(iv) f : X Y, f : X Y are separated S -morphisms f f : 22
S

XX YY is separated.
S

(v) g f separated f is separated


(vi) f separated fred separated.
The above properties are not all independent. In fact, the following
more general situation holds:
Let P be a property of morphisms of preschemes.
Consider the following propositions:
(i) every closed immersion has P
(ii) f : X Y has P, g : Y Z has P g f has p
(iii) f : X Y is an S -morphism having P f(S ) : X(S ) Y(S )
has P for any base-change S S .
(iv) f : X Y has P, f : X Y has P that f f : XX
YY has P
S

2. Preschemes

18
(v) g f has P, g separated f has P.
(vi) f has P fred has P.

If we suppose that (i) and (ii) hold then (iii) (iv). Also, (v), (vi)
are consequences of (i), (ii) and (iii) (or (iv)).
Proof. Assume (ii) and (iii). The morphism f f admits a factorization:
S

ff
S

X XE

EE
EE
EE
E
f IX EE"
S

/ YY
S
z=
z
zz

zz
zz IY f
S

YX
S

23

By (iii), the morphisms f T X and IY f have P and so by (ii) f f


S

also has P.

On the other hand, assume (i) and (iv). IS being a closed immersion,
has P by (i) and so f(S ) = f IS has P by (iv).
S

Now assume (i), (ii) and (iii). If g : Y Z is separated, Y


YY
Z

is a closed immersion and has P by (i); by making a base-change X


Y
Y IX

we get a morphism X YX YYX XY which, by (iii) has


Y

property p. The projection p2 : XY Y satisfies the diagram:


Z

/ XY
Z

p2

g f




Zo

i.e. to say, p2 is obtained from g f by the base-change Y Z and so,


by (iii) has P.

2.6. Affine morphisms

19
Y IX

Finally, f : X Y is the composite of X XY and p2 :


Z

XY Y and so by (ii) has P.


Z

To prove (vi) from (i), (ii), (iii) use the diagram


Xred

fred

24

/ Yred
Y

'



/Y

and the facts that the canonical morphisms X , Y are closed immersions
and so have P, that Y fred = f g has P and that a closed immersion
is separated, then use (v).
Q.E.D.
We now remark that if we replace (i) of the above propositions by
(i) every immersion has P, then (i), (ii), (iii) imply (v) g f has P,
g has P f has P.

2.6 Affine morphisms


Definition 2.6.1. A morphism f : X S of preschemes is said to be
affine (or X affine over S ) if, for every affine open U S , f 1 (U) is
affine in X.
It is enough to check that for an affine open cover (U ) of S , the
f 1 (U ) are affine.
2.6.2 Suppose that B is a quasi-coherent OS -Algebra. Let (U ) be
an affine open cover of S ; set A = (U , OS ), B = (U , B) and
X = Spec B. The homomorphism A B defines a morphism 25
f : X U ; the X s then patch up together to give an S -prescheme
f

X
S ; this prescheme X is affine over S , is such that f (OX ) B,
and is determined, by this property, uniquely upto an isomorphism. We
denote it by Spec B. conversely, every affine S -prescheme is obtained

2. Preschemes

20

as Spec B, for some quasi-coherent OS -Algebra B. (For details, see


EGA Ch. II, Proposition (1.4.3)).
Remarks. (a) Any affine morphism is separated. (Recall the remark
at the end of (2.4.2).)
(b) If S is an affine scheme, a morphism f : X S is affine X is
an affine scheme.
(c) The formal properties (i) to (vi) of (2.5) hold, when P is the property of being affine.
h

(d) Suppose X
Y is an S -morphism. If f , g are the structural morphisms of X, Y resply, the homomorphism OY h (OX ) defined
by h, given an OS -morphism
A (h) : g (OY ) g (h (OX )) = f (OX );
then we have a natural map: HomS (X, Y) HomOS (g (OY )
f (OX )) (the latter in the sense of OS -Algebras) defined by h 7
A (h). If Y is affine over S , it can be shown that this natural
map is a bijection. (EGA Ch II, Proposition (1.2.7)). (Also, compare with remark (1.6.4) (c) for affine schemes, and with remark
(2.1.3)).

2.7 The finiteness theorem


26

Definition 2.7.0. A morphism f : X Y of preschemes is said to be of


finite type, if, for every affine open set U of Y, f 1 (U) can be written as
n
S
f 1 (U) =
V , with each V affine open in X and each (V , OX ) a
=1

finite type (U, OY )-algebra.

It is again enough to check this for an affine open cover of Y. An


affine morphism f : X Y is of finite type the quasi-coherent
OY -Algebra f (OX ) is an OY -Algebra of finite type. In particular, a
morphism f : Spec B Spec A is of finite type B is a finite type
A-algebra (i.e. is finitely generated as A-algebra).

2.7. The finiteness theorem

21

Definition 2.7.1. A morphism f : X S is universally closed if, for


every base-change S S , the morphism f(S ) : XS S is a closed
S

map in the topological sense.

Definition 2.7.2. A morphism f is proper if


(i) f is separated
(ii) f is of finite type and
(iii) f is universally closed.
The formal properties of (2.5) hold then P is the property of being
proper.
Definition 2.7.3. A prescheme Y is locally noetherian, if every y Y
has an affine open neighbourhood Spec B, with B noetherian. It is said
n
S
to be noetherian, if it can be written as Y = Yi where the Yi are affine 27
i=1

open sets such that the (Yi , OY ) are noetherian rings.

If f : X Y is a morphism of finite type and Y is locally noetherian,


then X is also locally noetherian.
2.7.4 Let (X, OX ) and (Y, OY ) be ringed spaces and f a morphism from
X to Y. Let F be an OX -Module. We then define, for every q Z+ , a
presheaf of modules on Y by defining: U 7 H q ( f 1 (U), F ) for every
open U Y (See EGA Ch 0, III 12). The sheaf that this presheaf defines on Y is called the qth -direct image of F and is denoted by Rq f (F ).
Theorem 2.7.5. Let X, Y be preschemes, Y locally noetherian, and f a
proper morphism from X to Y. Then, if F is any coherent OX -Module,
the direct images Rq f (F ) are all coherent OY -Modules.
(For a proof see EGA Ch. III, Theorem (3.2.1)). This is the theorem
of finiteness for proper morphisms.

Chapter 3

Etale
Morphisms and Etale
Coverings
Throughout this chapter, by a prescheme we will mean a locally noethe- 28
rian prescheme and by a morphism, a morphism of finite type (unless
it is clear from the context that the morphism is not of finite type, e.g.,
b A a noetherian local ring).
S Spec O s,S , Spec A Spec A,

Definition 3.1.0. A morphism f : X S is said to be unramified at a


point x X if (i) M f (x) O x = M x , (ii) k(x)/k( f (x)) is a finite separable
extension.
Definition 3.1.1. A morphism f : X S is said to be flat at a point x
X if the local homomorphism O f (x) O x is flat (i.e., O x , considered as
an O f (x) -module is flat; note that since the homomorphism O f (x) O x
is local, O x will be faithfully O f (x) -flat).
Definition 3.1.2. A morphism f : X S is said to be e tale at a point
x X if it is both unramified and flat at x.
We say that f : X S is unramified (resp. flat, e tale) if it is unramified (resp. flat, e tale) at every x X.
f

Remarks 3.1.3. (1) An unramified morphism X


S is e tale at x
bf (x) O
bx is flat.
XO
23

3. Etale
Morphisms and Etale
Coverings

24
29

(2) A morphism f : X S is unramified at


x X fSpec k( f (x)) : X Spec k( f (x)) Spec k( f (x))
S

is unramified at the corresponding point; in other words, it is


enough to look at the fibre for nonramification.

(3) If f : X S is unramified at x X and k( f (x)) = k(x) then

bf (x) O
bx is surjective; if f is e tale in addition, then O
bf (x)
O

bx .
O

3.2 Examples and Comments


(1) A morphism f : Spec A Spec k (k a field) is e tale it is
r
L
Ki , Ki/k finite separable extensions.
unramified A =
i=1

(2) Let X and S be irreducible algebraic varieties, S normal. Then it


can be shown that a dominant morphism X S is e tale = it is
unramified.
(3) If S is non-normal, an unramified dominant morphism X S
need not be e tale; for instance, let c be an irreducible curve over
an algebraically closed filed with an ordinary double point x, e
c its
normalisation and p : e
c c the natural map.

30

(i) p is unramified. One has to prove this only at the points


a, b e
c sitting over x. Now Ma is generated in Oa by
a function with a simple zero at a, but such a function we
can find already in O x , for instance a function induced by a
straight line through x not tangent to c at x.

bx
ba but we know that O
bx is
(ii) p is not e tale: otherwise, O
O
b
not a domain, while Oa is.

(4) A connected e tale variety over an irreducible algebraic variety


need not be irreducible.

3.3.

25
For instance, in the previous example we may take two copies e
c
and e
c of the normalisation of c and fuse them together in such a
way that the points a, b on e
c are identified with the points b , a
on e
c . We then get a connected but reducible variety X and the
morphism p : X c defined in the obvious manner is surely
e tale.

(5) Let X and S be irreducible algebraic varieties and suppose that 31


S is normal and f : X S a dominant morphism. Assume, in
addition, that the function field R(X) of X is a finite extension of
degree n of the function field R(S ) of S . If x X is such that the
number of points in the fibre f 1 ( f (x)) equals n, then it can be
shown that f is e tale in a neighbourhood of x.

3.3
Our aim now is to give a necessary and sufficient condition for a morphism to be unramified.

3.3.0 Some algebraic preliminaries. Let A


B be a homomorphism
of rings defining an A-algebra structure on B. Then BB is an A-algebra
A

3. Etale
Morphisms and Etale
Coverings

26
in a natural way and

p1 : B BB given by b 7 b 1
A

p2 : B BB given by b 7 1 b
A

and : BB B given by b1 b2 7 b1 b2 are all A-algebra homoA

morphisms. We may make BB a B-algebra through p1 i.e. by defining


A

b(b1 b2 ) = bb1 b2 and the kernel I of is then a B-module. Since we


have p1 = Id . and the equality b1 b2 = (b1 b2 1)b1 (b2 11b2 )
is follows that BB = (B1) I, as a B-module.
A

32

We define the space B/A of A-differentials of B as the B-module


2
I/I . We note that the B-module structure on B/A = I/I 2 is natural, in
the sense that it does not depend on whether we make BB, a B-algebra
A

through p1 or through p2 .

Some properties of B/A


(1) The map d : b 7 (1 b b 1) (mod I 2 ) from B to B/A is
A-linear. Also, since
(1 b1 b2 b1 b2 1) = b1 (1 b2 b2 1) + b2 (1 b1 b1 1)
= +(1 b1 b1 1)(1 b2 b2 1),

we have: d(b1 b2 ) = b1 db2 + b2 db1 b1 , b2 B.


(2) Since I is generated as a B-module by elements of the form (1
b1 b1 1), it follows that B/A is generated by the elements db.
(3) Let W be the B-submodule of BB, generated by elements of the
A

form (1bb bb b b); B/A is then isomorphic to (BB)/W.


A

In fact,
B/A = I/I 2  (B1 I)/(B1 I 2 )
A

3.3.

27
= (BB)/(B1 I 2 )
A

33

and clearly W B1; so W = B 1 (W I).


A

On the other hand, we have


(1 bb b b b b) = bb 1 + (1 b b 1)(1 b b 1)
and hence: W B1 I 2 . Also clearly, I 2 W. It follows that
A

W I I 2 , W = B1 I 2 and B/A  (BB)/W.


A

(4) If V is any B-module and D : B V any A-derivation of B in V,


then there is a B-linear map HD : B/A V such that
D
/V
-H
-
-
-
H
d --- $
 D
-
-- 
- 

B-

B/A

In fact, D defines an A-linear map BB V given by b1 b2 7


A

b1 Db2 . This is certainly B-linear and is trivial on W. We then obtain a B-linear map HD : B/A V such that HD (db) = Db b
B. Also the correspondence D 7 HD is a B-isomorphism from
the B-module of A-derivations B V, onto the B-module HomB
(B/A , V). (This is the universal property of B/A ). In particular, the B-module of A-derivations of B is isomorphic to HomB
(B/A , B); and even more in particular, if A = k, B = K are fields, 34
we see that the space of k-differentials K/k is the dual of the
space of k-derivations of K and therefore is trivial if K/k is separably algebraic.
(5) Consider a base-change A A and the extension of scalars B =
BA . From the universal property of the space of differentials, it
A

3. Etale
Morphisms and Etale
Coverings

28
follows that

B /A B/A B B/A A .
B

3.3.1 The sheaf of differentials of a morphism.


Let f : X S be a morphism. Our aim now is to define a sheaf
X/S on X. Assume, to start with, that X = Spec B, S = Spec A; then
e B/A on
B/A is a B-module in a natural way and we define X/S =
X. In the general case, consider the diagonal of f in XX. Then
S

is a closed subscheme of an open subscheme U of XX and X is


S

35

an isomorphism. Let IX be the sheaf of ideals of OU defining . Then


IX/I 2 is a sheaf of OU -modules whose support is contained in . By
X
means of the isomorphism X , we can lift this sheaf to a sheaf of
abelian groups on X. Clearly the lift is independent of the choice of U.
It remains to show that this sheaf is an OX -Module. To do this, we can
assume again that X S are affine and in this case, IX is the sheaf e
I on
XX (where I is the ideal of BB defined in 3.3.0) and therefore, our
S

e B/A considered above.


sheaf is
The OX -Module thus obtained will be called the sheaf of differentials of f (or of X over S ) and will be denoted X/S . The stalk x,X/S of
this sheaf at a point x X is canonically isomorphic to Ox /O f (x) . This is
seen either from the universal property or by observing that the natural
map x,X/S Ox /O f (x) given by
Xb

si

dbi 7

Xb

si

d(bi/1 ) (with X = Spec B, bi , bi , si B, si < p )


x

is an isomorphism.
Since, by assumptions, S is locally noetherian and f is of finite type
it is cler from the definition that X/S is coherent.
Proposition 3.3.2. For a morphism f : X S and a point x X the
following are equivalent:
(i) f is unramified at x
(ii) x,X/S = (0)

3.3.

29

(iii) : X XX is an open immersion in a neighbourhood of x.


S

Proof. We may assume X = Spec B, S = Spec A.

36

(i) (ii): Consider the base change:


X = XS Spec k(s)

xXo

Spec k(s) = S

s = f (x) S o

If x X is above x, then we have: x,X/S O x = x ,X /S ; this


Ox

follows from property (5) of 3.3.0; furthermore,O x = O x/Ms Ox = k(x)


since f is unramified at x. Therefore x,X/S k(x) = x ,X /S and by
Ox

Nakayama it suffices to show that the latter is (0). By the remark made
above, x ,X /S = k(x)/k(s) and this is zero as k(x)/k(s) is separably
algebraic.
(ii) (iii)
Let z be the image of x in XX under the diagonal : X XX
S

(X) is a closed subscheme of an open subscheme U of XX, and is


S

defined, therefore, by a sheaf I of OU -ideals. If x,X/S = (0), then, we


will have Iz = Iz2 = . . .; hence Iz = (0) (Krulls intersection theorem). 37
However, I is a sheaf of ideals of finite type and so I vanishes in a
neighbourhood of z i.e. to say : X XX is an open immersion in a
S

neighbourhood of x X.

(iii) (i)
The question being local, we may assume that : X XX is
S

an open immersion everywhere. An open immersion remains an open


immersion under base-change, and since we only have to look at the
fibre over f (x), for non-ramification at x, we may take X = Spec A,
S = Spec k, k a field and A a L
k-algebra of finite type.
We are to show that A =
Ki , where Ki/k are finite separable field
finite

extensions; for this, we have to show that A is artinian and, if k is the

3. Etale
Morphisms and Etale
Coverings

30

algebraic closure of k, Ak is radical-free. It is thus enough to show that


k
L
k. By making the base-change k k, we may assume k is
Ak =
k

finite

algebraically closed.
Let a X be any closed point of X. Since k is algebraically closed,
k(a) k and we have then

X
X Spec k(a) = X (a)
i

X (a)

Spec k

(say).

is canonically imbedded in XX
S

and let be the composite of the morphisms:


i1

X X (a) XX.
S

38

Then 1 () = (a) is open in X, since, by assumption, is open in


XX; i.e. any closed point of X is also open. But X = Spec A is quasiS

compact and this implies that A has only finitely many maximal ideals.
But A is a k-algebra of finite type and so the set of closed points of X
n
L
Ai
is dense in X. It follows that A is artinian and we may write A =
i=1

where the Ai are artinian local rings. We may then assume A = Ai ; the
open immersion : X XX then gives an isomorphism AA A.
S

This however means A = k.

Q.E.D.

3.3.3 Some properties of e tale morphisms.


(1) An open immersion is e tale.
(2) f , g e tale g f e tale.
(3) f e tale f(S ) e tale for any base change S S

(not necessarily of finite type), S , S locally noetherian.

This follows from condition (iii) of Proposition 3.3.2, the facts


that an open immersion remains an open immersion and that a
flat morphism remains flat under base change.

3.3.

31

(4) f1 , f2 e tale f1 f2 e tale.


S

Follows from (2), (3) and the equality


f1 f2 = ( f1 IY2 ) (IX1 f2 ).
S

(5) g f e tale, g unramified f e tale.

39

Let f : X Y and g : Y Z. The question being local, we may


assume from (iii) Proposition 3.3.2 that = (Y) : Y YY is
Z

an open immersion, hence e tale. By the base-change X


Y we
get (X) : X XY which again is e tale by (3). The diagram:
Z

XY

Xo

p2

g f




Zo

shows that the second projection p2 : XY Y is given by


Z

(g f )(Y) ; so, again by (3), p2 is also e tale and from (2) we obtain
(X)

that f = p2 (X) : X XY Y is e tale.


Z

p2

(6) g f e tale, f e tale g e tale.

The composite: Og f (x) O f (x) O x is flat and O f (x) O x


is faithfully flat, and thus Og f (x) O f (x) is flat. It is clear that
k( f (x))/k(g f (x)) is a finite separable extension. Finally M f (x) O x = 40
M x and Mg f (x) O x = M x = (Mg f (x) O f (x) )O x ; since O f (x) O x
is faithfully flat, it follows that Mg f (x) O f (x) = M f (x) .

(7) An e tale morphism is an open map.


In fact, we prove more generally:
Proposition 3.3.4. A flat morphism is an open map.

3. Etale
Morphisms and Etale
Coverings

32

The proposition will follow as a consequence from the lemmas below. We first make the
Definition. A subset E of a noetherian topological space X is said to be
constructible if E is a finite union of locally closed sets in X.
Lemma 3.3.5. Let X be a noetherian topological space and E any set
in X. Then, E is constructible for every irreducible closed set Y of X,
E Y is either non-dense in Y or contains an open set of Y.
Proof. : Suppose E =
E Y Y

41

n
S

i=1

n
S

i=1

(Oi Fi ), Oi open, Fi closed in X. Then

(Fi Y) for any closed Y X; if Y is irreducible with

E Y dense in Y, this means Y Fi Y for some i, i.e., Fi Y; then


Y E (Oi Y) open in Y.
: We shall prove this by noetherian induction. Let E be the set of
all closed sets F in X such that E F is not constructible. If E ,
choose a minimal F0 E . By replacing X by F0 , we may assume that
for every closed set F properly contained in X, E F is constructible.
If X is reducible, say X = X1 X2 , X1 , X2 both proper subsets of X
and closed, then E X1 , E X2 are both constructible and hence so is
E = (E X1 ) (E X2 ). If X is irreducible, either
(i) E , X and so E = E E is constructible or
(ii) E U , , U open, so that E = U (E (X U)) is still
constructible. This contradiction shows that E = .
Q.E.D.

Lemma 3.3.6. Let S be a noetherian prescheme and f : X S a
morphism. Then the image, under f , of any constructible set is constructible.
Proof. Using the preceding lemma, the fact that S is noetherian and by
passing to subschemes, irreducible components and so on, we are readily reduced to proving the following assertion: if X, S are both affine,

3.3.

42

33

reduced, irreducible and noetherian and if f : X S is a morphism of


finite type such that f (X) = S , then f (X) contains an open subset of S .
If X = Spec B, S = Spec A, our assumptions mean that A, B are
noetherian integral domains and the ring-homomorphism A B defining f is then easily checked to be an injection. The lemma will then
follow from the following purely algebraic result.

Lemma 3.3.7. Let A be an integral domain and B = A[x1 , . . . , xn ] an
integral A-algebra containing A. Then there exists a g A such that for
every prime ideal p of A with g < p, there is a prime ideal P of B such
that P A = p.
Proof. Choose a transcendence base X1 , . . . , Xk of B/A. Then the extension B = A[x1 , . . . , xn ] is algebraic over A = A[X1 , . . . , Xk ]. By writing
down the minimal polynomials of the xi over A and by dividing out
these polynomials hbyi a suitably chosen g A, g , 0, we can make the
xi integral over A 1g . Consider now the tower of extensions.
B

h i

h i

h i

1
g

1
g

1
g

A
If p is a prime ideal of A such that g < p then there is a prime p of
h i
h i
A 1g lying over p. The prime ideal p + (X1 , . . . , Xk ) of A g1 lies over
h i
h i
p. Now g1 is a finite extension of A 1g and by Cohen-Seidenberg,
h i
a prime ideal P of B 1g lying over p. The restriction P of P to B
Q.E.D.

then sits over p.

3. Etale
Morphisms and Etale
Coverings

34

Lemma 3.3.8. Let f : X S be a morphism (of finite type) and T X 43


be an open neighbourhood of a point x X. Assume that for every
s1 S such that f (x) (s1 ), there is a t1 T such that f (t1 ) = s1 . Then
f (T ) is a neighbourhood of f (x).
Proof. Since the question is local, we may assume S noetherian and
then by lemma 3.3.6 it follows that f (T ) is constructible. We may then
S
write f (T ) =
(Oi Fi ); and by choosing only those i, for which
i=1

f (x) Oi , we may say that f (T ) is closed in a neighbourhood of f (x).


The hypotheses show that f (T ) is also dense in a neighbourhood of f (x).
It follows that f (T ) is itself a neighbourhood of f (x). Q.E.D.

Proof of Proposition 3.3.4.
In view of the lemma 3.3.8, it suffices to show that if U X is
an open neighbourhood of x X, f (U) contains all generisations of
f (x). The generisations of f (x) are the points of Spec O f (x) and those
of x are points of Spec O x . But O x is O f (x) -faithfully flat and then it
is well-known for any prime ideal p of O f (x) , there is a prime-ideal P
of O x contracting to p. U being open we have Spec O x U. Hence
Spec O f (x) f (U) (cf. Bourbaki, Alg. Comm. Ch. II, 2, n 5, Cor. 4
to Proposition 11).
Q.E.D.

3.4
44

A morphism f : S S is said to be an effective epimorphism if the


sequence
S = S S
S

p1
p2

i.e., if the sequence Hom(S , Y)

//

/S

is exact

/ Hom(S , Y)

exact, as a sequence of sets, Y.


(We say that a sequence of sets E1

h1

/ E2

p1

//

p2
h2
h2

is an injection and h1 (E1 ) = {x E2 : h2 (x) = h2 (x)}).

//

Hom(S , Y) is

E3 is exact if h1

3.4.

35

A morphism f : S S is faithfully flat if it is flat and surjective. Our aim now is to show that any such morphism is an effective
epimorphism.
3.4.1 Some algebraic preliminaries; the Amitsur complex. A homomorphism of rings f : A A defines a sequence:
f

/ A

p1
p2

p21
p32

// A A = A
A

p31

//

/ A A A = A
/

/
/

where p1 (a ) = a 1; p2 (a ) = 1 a , p21 (a b ) = a b 1;
p31 (a b ) = a 1 b ; p32 (a b ) = 1 a b and so on.
We may then define homomorphisms of A-modules:
45
0 = p1 p2

1 = p21 p31 + p32

2 = p321 p421 + p431 p432 . . .


and so on. One then checks that i+1 i = 0i; we thus get an augmented
cochain complex:
0

A
A A A A . . . .
f

This complex is called the Amitsur complex A /A.


f

Lemma 3.4.1.1. If A
A is faithfully flat, then

(i) A
H 0 (A /A)
(ii) H q (A /A) = (0)q > 0.
Proof. Suppose B is any faithfully flat A-algebra.
Consider then the complex
10

B /B B(A /A) B B = BA B = B B = BA
A

1 f

3. Etale
Morphisms and Etale
Coverings

36

Since B is A-faithfully flat, it is enough to prove that H 0 (B /B)


B 46
and H q (B /B) = (0)q > 0. As a particular choice we may take B = A .
f =1 f

Then the homomorphism A A A = A admits a section, i.e.,


A

: A A such that f = 1A , namely, the homomorphism:


a b 7 a b .
We may thus assume, without any loss of generality that the homof

morphism A
A admits a section : A A such that f = 1A :
f

/ A

/ A

/ A . . .

/.

We construct now a homotopy operator in this complex, as follows:

A9

/ A

99
99
9
1A 999
9 

p1
p2

p21

// A A = A
A

p31

//

/ A A
A
A

//

A A A = A
A

//

p32

11
p1
p2

// A A A
A
A

///

(i) H 0 (A /A) = ker 0


A.
47

Let a A be such that p1 (a ) = p2 (a ). Applying 1 we get


(1 )(a 1) = (1 )(1 a ) i.e. a 1 = 1 (a ) in A A.

Under the canonical identification A A


A , this means that
A

a = f ((a )) i.e.

a f (A).

On the other hand, f (A) ker 0 and f being faithfully flat, is

injective. It follows that ker 0


A.
(ii) By using the homotopy operator, one can show that H q (A /A) =
(0)q > 0; the proof is omitted. (We do not need it).


3.4.

37

3.4.2
Proposition 3.4.2.1. A faithfully flat morphism is also an effective epimorphism.
Proof. Case (a). S = Spec A , S = Spec A are affine.

From local algebra, it follows that the ring homomorphism A


A
defining f : S S , is also faithfully flat. We have to show that, for
any Y, the sequence
(*)

Hom(S , Y) Hom(S , Y) Hom(S S , Y) is exact in Ens .


S

(i) Suppose Y = Spec B is affine.


Then the sequence (*) is equivalent to a sequence:
()

48

Hom(B, Z) Hom(B, A ) Hom(B, A A ).


A

The exactness of this sequence () now follows from assertion (i)


of lemma 3.4.1.1.
(ii) Let Y be arbitrary.
Let 1 : S Y, 2 : S Y be two morphisms such that 1 f =
2 f . Since f is a surjection, it is clear that 1 (s) = 2 (s)s S .
Choose a point s S and a point s S with f (s ) = s; let
y = 1 (s) = 2 (s) Y. Choose an affine open neighbourhood
Spec B of y Y and an element A such that s Spec A and
1 (Spec A ) Spec B, 2 (Spec A ) Spec B.

Set 1 = () A . Then s Spec A , and f (Spec A )


Spec A ; and by (i) it follows that 1 and 2 , when restricted to
Spec A , define the same morphism of preschemes. Since s S
was arbitrary, this proves that Hom(S , Y) Hom(S , Y) is injective.

Now, suppose that : S Y is a morphism such that in

49

3. Etale
Morphisms and Etale
Coverings

38

S = S S
S

p1

//

p2

/S
>>
>>
>
>>


S1>

Y.

We have p2 = p1 . We want to find a morphism : S Y


such that = f .
In view of the injectivity established above, it is enough to define
locally. Let s S and s S with f (s ) = s. Choose an affine open
neighbourhood V of (s ) in Y. Then the open neighbourhood 1 (V)
of s in S is saturated under f : in fact, let x1 1 (V) and x2 S
such that f (x1 ) = f (x2 ). Then, there is an s S = S S such that
S

p1 (s ) = x1 and p2 (s ) = x2 ; so

(x1 ) = p1 (s ) = p2 (s ) = (x2 ) and

x2 1 (V).

Now, f is flat and hence an open map (Prop. (3.3.4)) and so


is an open neighbourhood of s S . Choose an element
A such that s Spec A f ( 1 (V)); if 1 = () A then
f (Spec A ) = Spec A , and Spec A = f 1 (Spec A ) 1 (V) since
1 (V) is saturated under f .
We then have a diagram
f ( 1 (V))

50

Spec(A A )
A

p1 p2

Spec A

vv
vv
v
v
vv
vv
v
v
vv
v
{v

Spec A = f (Spec A )

The problem of defining : Spec A V is now a purely affine


problem and we are back to (i).


3.5. Etale
coverings

39

Case (b). The general case.


Without loss of generality, we may assume S affine. Since f is a
morphism of finite type, there is a finite affine open covering (S )n=1
`
of S . Consider the disjoint union S = n=1 S . S is affine and the
morphism S S defined in the obvious way is again faithfuly flat.
Let Y be an arbitrary prescheme and F be the (contravariant) functor
X 7 Hom(X, Y). We have a commutative diagram:
F(S
O )

/ F(S )

//

F(S )

identity

F(S )


/ F(S )


// F(S S )
S

The lower sequence is exact by case (a); the first vertical map is the 51
identity and the second vertical map is clearly injective. Usual diagram
- chasing shows that the upper sequence is also exact. Q.E.D.


3.5 Etale
coverings
Definition . A morphism of preschemes, f : X S , is said to be finite if, for every affine open U S , f 1 (U) is also affine and the ring
( f 1 (U), OX ) is a (U, OS )-module of finite type.
It is again enough to check the conditions for an affine open cover
of S .
(1) If S is locally noetherian and f : X S is a finite morphism,
f (OX ) is a coherent OS -Module.
(2) A finite morphism remains finite under a base-change.
In particular, if f : X S is finite and s S any point, the
morphism fSpec k(s) : Xk(s) k(s) is finite and this means that
S

the fibre f 1 (s) is finite, discrete.

3. Etale
Morphisms and Etale
Coverings

40

(3) A finite morphism is proper.


In face, a finite morphism is affine and is hence separated; it remains finite under any base-change and so it is enough to show
that a finite morphism is closed.
52

By obvious reductions, we may take X, S affine reduced and


f (X) = S . If X = Spec B and S = Spec A, the corresponding
homomorphism A B is an injection and B is a finite A-module.
Cohen-Seidenberg then shows that f (X) = S .
(4) The following result holds:
Lemma 3.5.1 (Chevalley). Let S be (as always) locally noetherian and
f : X S a morphism of preschemes. Then the following conditions
are equivalent:
(a) f is finite
(b) f is proper and affine
(c) f is proper and f 1 (s) is finite s S .
(For a proof see EGA Ch.III (a) Proposition (4.4.2)).
Definition. A morphism f : X S is said to be an e tale covering if it
is both e tale and finite.
f

Let X
S be an e tale covering. Then f (OX ) is a locally free OS Algebra of finite rank. For any s S , the fibre f (OX )k(s) is a finite
direct sum

ns
P

Ki of finite separable field extensions Ki of k(s). The rank

i=1

of f (OX ) at s S is then given by

ns
P

[Ki :
i=1
fibre f 1 (s).

53

k(s)] which equals the

number of geometric points in the


This is constant in each
connected component of S ; if S is connected, this constant rank is called
the degree or the rank of the covering f . In this case, if this rank equals
1, then f is an isomorphism.


3.5. Etale
coverings

41

Note . For e tale coverings we have properties similar to (2), (3) (with
S S not necessarily of finite type) (4), (5) and (6) from 3.3.3; this
follows immediately from 3.3.3 and properties of coverings.
A word of caution:
This concept of an etale covering (French: revetement e tale)
should not be confused with the concept of a covering in the e tale
topology (French: famille couvrante). The latter concept is not treated
in this course. We also note that an e tale covering, as defined here, is
not necessarily surjective if S is not connected.

Chapter 4

The Fundamental Group


Throughout this chapter, we shall denote by S a locally noetherian con- 54
nected prescheme and by C = (E t/S ) the category of e tale coverings of
S . We note that the morphisms of C will all be e tale coverings.
(See 3.3.3 and the note at the end of Ch. 3).

4.1 Properties of the category C


(C0 ) C has an initial object (the empty prescheme) and a final object
S.
(C1 ) Finite fibre-products exist in C , i.e., if X Z and Y Z are
morphisms in C , then XY exists in C (see (3.3.3))
Z

(C2 ) If X, Y C , then the disjoint union X

Y C (obvious).

(C3 ) Any morphism u : X Y in C admits a factorisation of the form


u

X@
@

@@
@@
ui @@@
@

/Y
?




/  j

Y1

where u1 is an effective epimorphism, j is a monomorphism and


`
Y = Y1 Y2 , Y2 C .
43

4. The Fundamental Group

44
55

In fact, u is an e tale covering so u is both open and closed; if we


`
write u(X) = Y1 we have Y = Y1 Y2 and u = u1 : X Y1 is then an
effective epimorphism
(see (3.4.2.1)).
Further, this factorisation of u into an epimorphism and a monomorphism is essentially unique in the sense that if
X@
@

@@
@@
@
u1 @@
@

/Y
?




/  j

Y1

is another such factorisation, then there exists an isomorphism : Y1


Y1 such that u1 = u1 and j = j .
(Because a factorisation into a product of an effective epimorphism
and a monomorphism is unique).
(C4 ) If X C and g is a finite group of automorphisms of X acting,
say, to the right on X, then the quotient X/g of X by g exists in C

and the natural morphism X


X/g is an effective epimorphism.

56

The quotient, if it exists, is evidently unique upto a canonical isomorphism; also X is affine over S and therefore the existence of X/g has
only to be proved in the case X, S affine, say X = Spec A, S = Spec B
and G = the group of B-automorphisms of A corresponding to g. Then
Spec AG (AG is the ring of G-invariants of A) is the quotient we are looking for. (Compare with Serre, Groupes algebriques et corps de classes,
p. 57). Our aim now is to show that X/g is actually in C . The question is again local and we may assume X = Spec A, S = Spec B, with
B noetherian, and X/g = Spec AG as above. X S is finite and so
X/g S is also finite. It remains to show that this morphism is e tale.
In order to do this, we first make some simplifications.
Suppose that S S is a flat affine base-change. We have a com-

4.1. Properties of the category C

45

mutative diagram:
X = XS

/ Y = YS

/ Spec B = S


/ X/g = Spec AG = Y


/ Spec B = S .

Spec A = X

g acts on X in the obvious way, as a group of S -automorphisms of X .


We assert that Y = YS is the quotient of X with respect to this action.
S

Indeed,
we have anL
exact sequence of B-algebras: 0 AG A
L
P
(a a ). Since B
A, where A
A is the map given by A 7
G
G
G
L
is B-flat, we get an exact sequence: 0 AG B A B
(A B )
B

and this proves that the subring of invariants of A B is AG B ; hence


B

our assertion.
Let y Y = X/g and s S be its image. Take for B the local ring
O s,S . Then there is a unique point y Y = YS over y and one has
S

Oy ,Y = Oy,Y ; hence Y S is e tale at Y Y S is e tale at y . We 57


may thus assume that S = Spec B, B a noetherian, local ring. In view of
the following lemma, we may assume B complete.
f

S a faithfully flat
Lemma 4.1.1. Let X
S be a morphism and S
base-change. Then f is e tale f(S ) is e tale.
Proof. : is clear.
: flatness of f is straightforward. To prove non-ramification one
observes that in view of (5) (3.3.0), one has X /S = (X/S ); but
being faithfully flat, X /S = 0 X/S = 0; one now applies proposition 3.3.2.
Q.E.D.

Let x1 , . . . , xn be the points of X over s. By hypothesis each k(xi )/
k(s) is a finite separable extension. We choose a sufficiently large finite
galois extension K of k(s) such that each k(xi ) is imbedded in K. We
now need the

4. The Fundamental Group

46

Lemma 4.1.2. Let B be a noetherian local ring with maximal ideal M


and residue field k. Let K be an extension field of k. Then a noetherian
local ring C and a local homomorphism : B C such that (i) is
B-flat and (ii) C/M C  K. (see EGA. Ch. 0III , Prop. (10.3.1)).
58

In addition if [K : k] < , we can choose C to be a finite B-algebra.


(EGA. Ch. 0III , Cor. (10.3.2)).

By making such a base-change B


C we may assume that each
r
L
B, a fik(xi ) is trivial over k(s). Under these assumptions we get A =
i=1

nite direct sum of copies of B. Under the action of g, the set {x1 , . . . , xn }
splits into disjoint subsets {x1 , . . . , xl }, {xl+1 , . . . , xm }, . . ., on each of
which g acts transitively. The corresponding decomposition of A will
m
l
L
L L
L
B)
. . .. The action of G on each
then be given by A = ( B) (
i=1

i=l+1

block, for instance, on a (b1 , . . . , bl )

l
L

B, will then be just a permu-

i=1

tation. The subring AG will then be the direct sum 1 . . . . . .


l
L
B and so on. Each is evwhere 1 is the diagonal of the block
i=1

idently isomorphic to B. Our assertion that X/g S is e tale is now


clear.

Thus X/g C ; the natural morphism X


X/g is also then an e tale
covering. Therefore will be an open map and thus if is not surjective
one could replace Y by the image of ; this is clearly impossible. Hence
is surjective and thus an effective epimorphism 3.4.2.1.

4.2
59

We shall now define a covariant functor F from C (E t/S ) to the category of finite sets. We shall fix once and for all a point s S and an
algebraically closed field k(s).
For any X C , F(X), by definition, will be the set of geometric points of X over s S , with values in , i.e., is the set of all S -

4.2.

47

morphisms Spec X for which the diagram is commutative.


Spec

/X


/S

Spec k(s)

We observe that if x X sits above s S , then giving an S morphism Spec X whose image is x X is equivalent to giving a
k(s)-monomorphism of k(x) into . Also note that for any X C , F(X)
is a finite set whose cardinality equals the rank of X over S .
Properties of the functor F.
(F0 ) F(X) = X = .
(F1 ) F(S ) = a set with one element;
F(XY) = F(X) F(Y), X, Y, Z C .
Z

(F2 ) F(X1

F(Z)

`
u

X2 ) = F(X1 )

F(X2 ).

(F3 ) If X
Y is an effective epimorphism in C , the map F(u) : 60
F(X) F(Y) is onto.
In fact, if y Y is a point above s S and y = u(x), x
X, then any k(s)-monomorphism of k(y) to extends to a k(s)monomorphism of k(x) to .
(F4 ) Let X C and g a finite group of S -automorphisms of X (acting
to the right on X). Then g acts in a natural way (again to the right)
on F(X) as expressed by
g

Spec X X.
The natural map : X X/g (see (C3 )) defines a surjection
F() : F(X) F(X/g) (see (F3 )). In view of the commutativity

4. The Fundamental Group

48
of the diagram

g
/X
/X
AA
}
AA
}
}
AA
}
}}
AA
}
~}

Spec

X/g

it follows that F() descends to a surjection e


: F(X)/g F(X/g).
We claim that e
is actually a bijection. This follows immediately
from the
Lemma 4.2.1.
61

(i) g acts transitively on the fibres of .

(ii) Suppose y Y = X/g and x 1 (y). Let gd (x) be the subgroup


{ g : (x) = x} (called the decomposition group of x). Then
we have:
(a) k(x)/k(y) is a galois extension.
(b) the natural map gd (x) the galois group G(k(x) /k(y)) is
onto.
Proof.

(i) We may assume X = Spec A, Y = X/g = Spec AG as


before (both noetherian); also one knows that A is finite on AG .
Let P, P1 be prime ideals of A (i.e., points of X) such that P1 ,
P G, while P AG = P1 AG = p. We may assume
P and P1 maximal (otherwise apply the flat base change Y
Spec Oy,Y , where y Y corresponds to p). Then there is an a
P1 such that a < P (Chinese Remainder Theorem). Thus
Q
b =
(a) < P; but b AG , so b AG P1 = AG P

contradiction.

(ii) We have a diagram:


AO

/ A/P k(x)
O

AG

/ AG /p k(y),

4.3.

49
and we know that k(x)/k(y) is a finite, separable extension. Let
Q
A be such that k(x) = k(y)(). The polynomial f = (T )

62

is in AG [T ], has as a root and splits completely in A[T ]. The


reduction f of f mod p is in k(y)[T ], has as a root and splits
completely in k(x)[T ]. It follows that k(x)/k(y) is normal, hence
galois.
Consider now the subgroup Gd (P) of G corresponding to gd (x).
We have P , 1 P < Gd (P) and by the Chinese Remainder Theorem we can choose a 1 A such that 1 (modP)
and 1 0(mod1 P), < Gd (P).
We have k(x) = k(y)(1 ). Consider now the polynomial g =
Q
(T 1 ) k(y)[T ]. As 1 is a root of g, for every

G(k(x) /k(y)), (1 ) is also a root of g; hence (1 ) = (1 ) for


some G. But (1 ) , 0 and, by the choice of 1 , (1 ) = 0
if < Gd (P); hence (1 ) = (1 ) for some Gd (P), i.e.,
= for some Gd (P).
Q.E.D.

(F5 ) If u : X Y is a morphism in C such that F(u) : F(X) F(Y)
is a bijection, then u is an isomorphism
From the fact that F(u) is a bijection it follows (see the remark at 63
the end of Ch. 3) that the rank of u : X Y is 1 at every y Y, hence
(again by the same remark) u is an isomorphism.
A category which has the properties (C0 ), . . . , (C4 ) of 4.1 and from
which there is given a functor F into finite sets with the above properties
(F0 ), . . . , (F5 ) is called a galois category; the functor F itself is known
as a fundamental functor.

4.3
Before we start our construction of the fundamental group of a galois
category we motivate our procedure by two examples.

4. The Fundamental Group

50

Example 1. Let S be a connected, locally arcwise connected, locally


simply connected topological space and C the category of connected
p
coverings of S ; the morphisms of C are covering maps. Let X
S be
such a covering. Fix a point s S ; we define F(X) = p1 (s). Then
(X, p) 7 p1 (s) is a covariant functor F : C Ens.

64

Each member of C determines (upto conjugacy) a sub-group of the


fundamental group 1 (S , s); and to each subgroup H of 1 (S , s) there
corresponds a member of C determining H. To the subgroup {e} corresponds, what is known as, the universal covering Se of S ; and 1 (S , s) is
isomorphic to the group of S -automorphisms of Se, i.e., to the group of
covering transformations of Se over S . Further we have the isomorphism
(in Ens):

HomC (Se, X)
F(X), X C .
The functor F : C Ens is thus representable in the following
sense.

Definition 4.3.1. A covariant functor G from a category C to Ens is


representable if an object Y C such that:

HomC (Y, X)
G (X) X C .
Example 2. Let k be a field and a fixed algebraically closed field
extension of k. Set S = Spec k and C = the category of connected e tale
coverings of S ; any member of C is of the form Spec K, where K/k is a
finite separable field extension. For any X C we define F(X) = the set
of geometric points of X with values in . Then, F(X) Homk (K, s )
is X = Spec K, where s is the separable closure of k in . If s is finite
over k, we can further write F(X) HomC (Spec s , X) and the functor
F : C (finite sets), defined above, will be representable. However
this is not the case in general; out we can find an indexed, filtered family
(Ni )iI of finite galois extensions of k, (namely, the set of finite galois
extensions contained in s ) such that for any X C , we can find an
i0 = i0 (X) such that F(X) HomC (Spec Ni , X), i i0 (X). In other
words, we may write
F(X) lim HomC (Spec Ni , X),

iI

X C

4.3.

51

and the family (Spec Ni )iI is in fact a projective family of objects in C . 65


Suppose now that C is any category and G : C Ens is a covariant
functor. If X C and G (X), we write, as a matter of notation,

G
X. If G
X, and G
Y and X
Y is a C -morphism, we say
that the diagram

G ?
?

??
??
?

/X



u
 

is commutative if G (u)() = .

If G
X, then for any Z C , we have a natural map HomC (X, Z)
G (Z) defined by u 7 G (u)().
Definition 4.3.2. We say that G is pro-representable if a projective
system (S i , i j )iI of objects of C and elements i G (S i ) (called the
canonical elements of G (S i )) such that
(i) the diagrams
G ?
?

??
??
?

Sj

/ Si
?
~
~~
~
~
~~ i j ( ji)

are commutative.
(ii) for any Z C , the (natural) map
lim HomC (S i , Z) F(Z)

iI

is bijective.
In addition, if the i j are epimorphisms of C , we say that G is strictly
pro-representable.
Thus, our functor F in Example 2 is pro-representable. Example 1
and 2 show that representable and pro-representable functors arise naturally in the consideration of the fundamental group.

66

4. The Fundamental Group

52

4.4 Construction of the Fundamental group


4.4.1 Main theorem
(1) Let C be a galois category with a fundamental functor F. Then
there exists a pro-finite group (i.e., a group which is a projective limit of finite discrete groups provided with the limit topology) such that F is an equivalence between C and the category
C () of finite sets on which acts continuously.
F

(2) If C C ( ) is another such equivalence, then is continuously isomorphic to and this isomorphism between and is
canonically determined upto an inner automorphism of .
67

The profinite group , whose existence is envisaged in assertion (1)


above will be called the fundamental group of the galois category C .
The theorem is a consequence of the following series of lemmas.
Definition 4.4.1.1. A category C is artinian if any decreasing sequence
T 1 j1 T 2 j2 T 3 j3 . . .
of monomorphisms in C is stationary, i.e., the jr are isomorphisms for
large r.
A (covariant) functor F : C Ens is left-exact if it commutes with
finite products i.e., if F(X Y) = F(X) F(Y) and if, for every exact
u

sequence X

/Y

u1
u2

F(X)

//

Z in C , the sequence
F(u)

/ F(Y)

F(u1 )
F(u2 )

// F(Z)

is exact as a sequence of sets.


If Y

u1
u2

//

Z are morphisms in C , a kernel for u1 , u2 in C is a pair

(X, u) with X C and u : X Y in C such that X

/Y

u1
u2

//

4.4. Construction of the Fundamental group

53

is exact in C . Clearly a kernal is determined uniquely upto an isomorphism in C .


Lemma 4.4.1.2. Let C be a category in which finite products exists. 68
Then finite fibre-products exist in C kernels exist in C .
Proof. : Let Y

u1
u2

// Z be morphisms in C . We have a commutative

diagram:
p8 Z dIII
ppp
IIu2
p
p
II
p
p
II
p
p
I
ppp
Y eLLL
;Y
xx
LLL
x
x
LL
x
xx
p1 LLLL
xx p2
u1

YY
Z

:
uu
uu
u
u
uu
uu

(YY) Y
Z

(YY)

KKK
KKK
KKK
KK%

CC
CC(p1 ,p2 )
CC
CC
C!

Y< Y
xx
x
xx
xxdiagonal
x
xx

and it easily follows that (YY Y is a solution for the kernal of u1


Z

(YY)

and u2 .
f

: Suppose X
Z and Y
Z are morphisms in . If p and q are
p
q
the canonical projections X Y
X, X Y
Y, we have an exact
sequence:
ker( f p, gq)

/ XY

fp
gq

//

Z.

It follows that ker( f p, gq) is a solution for the fibre-product XY. 69


Z

Q.E.D.

In fact, we have shown that finite fibre products and kernels can be
expressed in terms of each other. Hence, F commutes with finite fibreproducts it is left-exact.

4. The Fundamental Group

54

Corollary . A fundamental functor is left-exact (see (F1 ))


Lemma 4.4.1.3. A galois category is artinian.
Proof. Let
T 1 j1 T 2 j2 . . . jr1 T r jr T r+1 . . .
be a decreasing sequence of monomorphisms in C . We have then:
jr

T r+1 T r

is a monomorphism

T r+1
T r+1 T r+1
Tr

F(T r+1 )
F(T r+1 ) F(T r+1 ) (by (F1 ), (F5 ))

F(T r )

F( jr )

F(T r+1 ) F(T r ) is a monomorphism.


Since the F(T r ) are finite, this implies that the F( jr ) are isomorphisms for large r; we are through by (F5 ).
Q.E.D.

70

Lemma 4.4.1.4. Let C be a galois category with a fundamental functor


F. Then F is strictly pro-representable.
Proof. With the notations of 4.3.2, consider the set E of pairs (X, ) with

F
X. We order E as follows:
(X, ) (X , ) a commutative diagram:
F@
@

@@
@@
@@

/X
~
~
~~
~~
~
~~

We claim that E is filtered for this ordering; in fact, if (X, ), (X , )

4.4. Construction of the Fundamental group

55

E , in view of (F1 ) we get a commutative diagram:


mm6= X
mmmzzz
m
m
m zz
mmmm
z
m
m
zzp
m
z
m
m
z
mm
z
mmm
zz
m(,
z
m
m
)
m
/ X X
F QQQQ
DD
QQQ
DD
QQQ
DD p

QQQ
DD
QQQ
QQQ DDD
QQQ DD
QQQ D!
Q(

where p and p are the natural projections.


We say that a pair (X, ) E is minimal in E if for any commutative 71
diagram

/X
?

??

??


???
 j

? / 

F?
?

with a monomorphism j, one necessarily has that j is an isomorphism.


(*) Every pair in E is dominated, in this ordering, by a minimal pair
in E .
Observe that C is artinian (Lemma 4.4.1.3).
(**) If (X, ) E is minimal and (Y, ) E then a u HomC (X, Y) in
a commutative diagram
F?
?

??
??
???
?

is uniquely determined.

/X





u

 

4. The Fundamental Group

56

In fact, if u1 , u2 HomC (X, Y) such that the diagrams


F?
?

??
??
?

/
?Y


u1



and

F?
?

??
??
?

/
?Y


u2



are commutative then by (C1 ) and Lemma 4.4.1.2 ker(u1 , u2 ) exists;


since F is left exact we get a commutative diagram

/Y
O O
GG
GG
GG
GG
u1 u2

GGG
GG
G#


/X
ker(u1 , u2 )

F GG

with a monomorphism j. As (X, ) is minimal j must be an isomorphism, i.e., u1 = u2 .


From (*), (**) it follows that the system I of minimal pairs of E is
directed.
If (X, ) I, (Y, ) E and u HomC (Y, X) appears in a commutative diagram

F?
?

??
??
???
?

73

/X
?




 u



then u must be an effective epimorphism.


In fact, be (C3 ) we get a factorisation

??
??
?
u1 ???
?

/ X1 ` X2 = X
:
tt
tt
t
tt
tt j
, tttt
u

Y?
?

X1

72

4.4. Construction of the Fundamental group

57

with an effective epimorphism u1 and a monomorphism j. By (F2 ) and


(F3 ) we then obtain a commutative diagram:

F. OOO

.. OOO
OOO
..
OOO
..
OO'
..
X1
.
..
~?
~
~
..
~
. ~~~ u1

/ X = X1 ` X2
9
rrr
r
r
rrr j
+ rrr

By minimality of (X, ), it follows that j is an isomorphism; thus u


is an effective epimorphism. In particular:
*** The structure morphisms occurring in the projective family I are
effective epimorphisms.
Consider now the natural map
lim HomC (S i , X) F(X),

iI

X C.

By (*) this is onto; by (**) it is injective. From (***) it thus follows 74


that F is strictly pro-representable.
Q.E.D.

Definition 4.4.1.5. Let C be a category with zero () in which disjoint
`
unions exist in C . An X C is connected in C X , X1 X2 in C
with X1 , X2 , .
Note. In (E t/S ), a prescheme is connected it is connected as a topological space.
With the notations of the preceding lemma, we have:
Lemma 4.4.1.6.

(i) (X, ) E is minimal X is connected in C .

(ii) If X is connected in C , then any u HomC (X, X) is an automorphism.

4. The Fundamental Group

58

(iii) For any X C , Aut X acts on F(X) as follows:

Aut X

F
X X. It X C is connected, then for any F(X)
the map Aut X F(X) defined by u 7 F(u)() = u , is an
injection.
`
Proof.
(i) Suppose X = X1 X2 in C , X1 , X2 , and that (X, )
`
E ; then F(X) = F(X1 ) F(X2 ) say, F(X1 ).
We then have a commutative diagram

F@
@

@@
@@
@
@@
@

/X
~?
~
~~
~~
~
/ ~~~ j

X1
75

with a monomorphism j which is not an isomorphism. Thus (X, )


is not minimal.
On the other hand, let X C be connected and (X, ) E . Suppose we have a commutative diagram:

/X
?
??

??

j
???

? / 

F?
?

with a monomorphism j. By (C3 ) we get a factorisation:


Y?
?

??
??
??
j1 ??
?

/ X = X1 ` X2
:
tt
tt
t
tt
tt j2
t
, ttt
j

X1

with an effective epimorphism j1 and a monomorphism j2 . As


j is a monomorphism, so is j1 and thus j1 is an isomorphism;
since X is connected, X2 = and one gets that j = j2 j1 is an
isomorphism.

4.4. Construction of the Fundamental group

59

(ii) As X is connected, it follows by (C3 ) that u is an effective epimorphism; by (F3 ), F(u) : F(X) F(X) is onto and thus is a
bijection (F(X) finite). By (F5 ) it follows that u Aut X.
(iii) Let u1 , u2 Aut X such that F(u1 )() = F(u2 )(), i.e., 76
ker(F(u1 ), F(u2 )) = F(ker(u1 , u2 )) (F is left-exact). We thus have
a commutative diagram

FD
D

DD
DD
DD
DD
D!

/X
z=
z
zz
zz
z
zz j
. zz

u1
u2

// X

ker(u1 , u2 )

with a monomorphism j; as (X, ) is minimal by (i), j is an isoQ.E.D.


morphism, in other words, u1 = u2 .

We briefly recall now the example 2 of 4.3. Assertion (iii) of the
above lemma simply says in this case that if K/k is a finite separable extension field and if Homk (K, ), then the map Aut K Homk (K, )
given by u 7 u is injective. We know that K/k is galois this map
is also onto. Following this, we now make the
Definition 4.4.1.7. A connected-object X C is galois if for any
F(X), the map Aut X F(X) defined by u 7 u is a bijection.
Note that this is equivalent to saying that the action of Aut X on F(X) is
transitive. Also observe that this definition is independent of F because
the cardinality of F(X) is the degree of the covering X over S ; the action
is already effective since X is connected (by (iii), Lemma 4.4.1.6).
77

Lemma 4.4.1.8. If F
Y, then there is a galois object X C , a
F(X) and a u HomC (X, Y) such that the diagram
F?
?

??
??
??
??


/X




 u
 

4. The Fundamental Group

60

is commutative. In other words, the system I1 of galois pairs of E is


cofinal in E .
Proof. Let (S i )iI be a projective system of minimal objects of C such
that

F
lim HomC (S i , ).

iI

Let 1 , . . . , r be the elements of F(Y). We can choose i large enough


such that as u varies over HomC (S i , Y), the u i give all the s (i is
the canonical element of F(S i )). We then get:
i

pj

F S i
Yr = |
Y {z
... Y
} Y
r times

78

where p j is the jth canonical projection Y r Y; the elements p j i ,


1 j r, are precisely the elements 1 , . . . , r of F(Y). By (C3 ) we get
a factorisation:

/ Yr
Si ?
?
??
~~
??
~~
~
??
~
1 ?
??  ~~~
 /~

with a monomorphism and an effective epimorphism 1 . We claim


that X is galois.
(i) X is connected.

`
Suppose X = X1 X2 , X1 , X2 in C , , ; the element 1 i
F(X1 ), say. We can then choose a j large enough for us to get a
commutative diagram:
1
/ X = X1 ` X2
? SO i
9
rrr
i 
r
r


rrr

+ rrr
i j
F?
> X1
??
~~
??
~
?
~~
j ??

~~

Sj

4.4. Construction of the Fundamental group

61

Hence = 1 i j is an epimorphism, which is absurd.


(ii) Set = 1 i F(X). We shall prove that the map Aut X F(X) 79
defined by u 7 u is onto.
Let F(X); we may assume i is so large that we get a commutative diagram:
7X
ppp@
p
p
pp
ppp
p
p
1
pp
ppp
p
p
ppp i
/ Si
F NNN
>>
NNN
>>
NNN
>> 1
NNN
N
NNN >>>

NNN >>
NNN>
'

Our aim is to find a Aut X such that 1 = 1 . Since


X is connected, 1 is also an effective epimorphism. Since the
manner in which a morphism in C is expressed as the composite
of an effective epimorphism and a monomorphism is essentially
unique, we will be through if we find a Aut Y r such that the
diagram

7@ X
p
p
pp
ppp
ppp
ppp
1
ppp
p
p
p
ppp i
/ Si
F NNN
>>
NNN
>>
NNN
>> 1
NNN
NNN >>

NNN >>
NNN>>
N'   

/ Yr


/ Yr

is commutative. By assumption the elements p1 , 1 j r, 80

4. The Fundamental Group

62

are all the distinct elements of F(Y); so the morphisms p j 1


are all distinct. This means that the p j are all distinct; since
1 is an effective epimorphism the p j 1 are all distinct; and
as X is connected and is a monomorphism, it follows that the
p j are all distinct and therefore form the set F(Y).
If we set p j = j , 1 j r, and p j = ( j) , 1 j r, we
get a permutation of the set {1, 2, . . . , r}; this permutation determines
an automorphism of Y r with the required property.
Q.E.D.

By this lemma we may clearly assume now that F is strictly prorepresented by a projective system (S i ) of galois objects of C .
Let gi = Aut S i and i be the bijection gi F(S i ) defined by u 7
u Ci where Ci is the canonical element of F(S i ). For j i, we define
i j : g j gi as the composite
j

F(i j )

i1

g j F(S j ) F(S i ) gi .
81

For any u g j , i j (u) is the uniquely determined automorphism of


S i which makes either one (and hence also the other) of the diagrams
8 Si
i qqq
q
q
qqq
F LLL
LLL
j LL&

Sj

i j (u)

/ Si
O
i j

/Sj

SO i

i j (u)

i j

Sj

/ Si
O
i j

/Sj

commutative. It follows from this easily that the i j are group homomorphisms.
We thus obtain a projective system {gi , i j }iI of finite groups with
each i j surjective. Denote by {i , i j }iI the projective system of the
opposite groups. The group = lim i with the limit topology is proiI
finite and we shall prove that it is the fundamental group of the galois
category C ; we denote it by 1 (S , s) when C and F are as in 4.1 and
4.2.

4.4. Construction of the Fundamental group

63

i acts on HomC (S i , X) to the left and hence acts continuously on

F(X), to the left. Since F(X) is finite, the


the set lim HomC (S i , X)
i
action of on F(X) comes from the action of some i on F(X).
4.4.1.9 We shall find it convenient now to introduce informally the no- 82
tion of the procategory Pro C of C . An object of Pro C = (called a
e
pro-object of C ) will be a projective system e
p = (Pi )iI in C . If P,

e = (P ) jJ are pro-objects of C , we define Hom(P,


eP
e ) as the double
P
j

limit lim lim HomC (Pi , P j ). An object of C will be considered an


jJ iI
object of Pro C in a natural way.
We may look at a pro-representable functor on C , as a functor represented in a sense by a pro-object of C . For instance, in the case of
4.4.1.4, we have:

F(X)
lim HomC (S i , X),

iI

X C

HomPro C (Se, X)

where Se is the pro-object (S i )iI of C .


Also for any i I, HomPro C (Se, S i ) HomC (S i , S i ) = gi and we
may then write:
g = lim gi = lim HomC (S i , S i )

= lim HomPro C (Se, S i ) = HomPro C (Se, Se)

and hence = AutPro C Se.


If we call Se a pro-representative of F (in the case C = (E t/S ) we 83
call it a universal covering of S ) then is the opposite of the group of
automorphisms of Se.
Lemma 4.4.1.10. Let E C (); then an object G(E) C , and a
C ()-isomorphism E : E FG(E) such that the map HomC (G(E), X)
HomC () (E, F(X)) given by u 7 F(u)E is a bijection for all X C .
The assignment E 7 G(E) can be extended to a functor G : C () C
such that F and G establish an equivalence of C and C ().

4. The Fundamental Group

64

`
Proof. If E = Ei is a decomposition of E into connected sets in C ()
`
and if G(Ei ) are defined we may define G(E) = G(Ei ). We may thus
assume that acts transitively on E. Fix an element E E and consider
the surjection E defined by 7 E . As E is finite, there is an i
such that the diagram
?

??
??
??
??
??


84

/
?E
~
~~
~~
~
~~
~~

( i is the natural projection and i E is the map 7 E ) is


commutative. Let Hi i be the isotropy group of E in i . It is easily
proved that the set i /Hi of left-cosets of i mod Hi is C ()-isomorphic
to E. We then define: G(E) = G(i /Hi ) = S i /Hi0 , the quotient of S i by
the opposite Hi0 of Hi (remark: Hi0 gi Aut S i ). By (F4 ) we have:

F(G(E)) = F(S i /Hi0 )


F(S i )/Hi0 i /Hi E, and hence a C ()isomorphism E : E FG(E). If j i and H j j is the isotropy
group of E in j , then we have a C -morphism S j /H 0j S i /Hi0 ; since
F(S j /H 0j ) F(S i /Hi0 ) is a C ()-isomorphism, it follows from (F5 )

S i /Hi0 and that G(E) is independent of the choice of i


that S j /H 0j
(upto a C -isomorphism).
Let X C . Consider the map
: HomC (G(E), X) HomC () (E, F(X))
u 7 F(u) E

(i) is an injection:
Let u1 , u2 HomC (G(E), X) be such that F(u1 )E = F(u2 )E .
But E : E FG(E) is an isomorphism and so ker(F(u1 ), F(u2 ))
FG(E), i.e., F(ker(u1 , u2 ))FG(E) (F is left-exact). It fol

lows from (F5 ) that ker(u1 , u2 )G(E), in other words, u1 = u2 .

85

(ii) is a surjection.

4.4. Construction of the Fundamental group

65

Let E
F(X) be any C ()-morphism; put = (E ) F(X).
The Pro C ()-morphism : Se X can be factored through some
S i:

Se >>

>>
>>
i >>>
>

/X
?

Si

Let Hi be the isotropy group of i in i then Hi Hi where Hi


is as before (take i large enough). By the construction of G(E) =
Si
S i /Hi0 , we have a morphism 0 X making the diagram
Hi
Se

/ Si

/X
|=
|
||
||
|
||
||

CC
CC
CC
CC
CC
C!

S i /Hi0

commutative; one easily checks that this morphism goes to under . It only remains to show that the assignment E 7 G(E)
can be extended to a functor. Let E, E C () and HomC ()
(E, E ). To the composite E : E FG(E ) there corresponds 86
a unique HomC (G(E), G(E )) such that the diagram
E

/ E
E

FG(E)

F()


/ FG(E )

is commutative. We set G() = . It follows easily that G is a


covariant functor from C () to C .
One now checks that there are functorial isomorphisms
: IC G F

4. The Fundamental Group

66

: IC () F G,

and

such that, for any X C and E C (),


F((X))

1 (F(X))

G((E))

1 (G(E))

F(X) FGF(X) F(X)


and

G(E) GFG(E) G(E)

and the identity maps.


This completes the proof of assertion (1) of Theorem (4.4.1).
87

Lemma 4.4.1.11. Let C be a galois category and F, F be two fundamental functors C (finite sets). Suppose , are the profinite groups
defined by F, F respectively as above; then and are continuously
isomorphic and this isomorphism is canonically determined upto an inner automorphism of .
Proof. We know that F : C C () is an equivalence. Replacing F
by F G (with G as before) we can assume that C = C (), that F is the
trivial functor identifying an object of C () with its underlying set and
e Pro C ()
itself is the pro-object pro-representing this functor. Let T

e
pro-represent F ; first we show that
T in Pro C .
In order to do this, let (T j ) jJ be a projective family of galois objects
(with respect to F ) of C such that Te = lim T j ; we denote the canonical
j
e
maps T j T i by qi j and T T j by q j . Let t j T j be a coherent system
of points and consider the continuous maps j : T j determined by
j (e) = t j ; there exists a continuous : Te such that
>

>>
>>
>>
j >>>
>

88

qj

Tj

/e
T

is commutative. We have (e) = e


t = (t j ) Te. The T j are connected,
hence j transitive and as is compact it follows that is onto. Let

e. (/H is the set


H be the isotropy group of e
t, then we have /H
T

4.4. Construction of the Fundamental group

67

of left-cosets of mod. H) and this is an isomorphism of topological


spaces (it is a continuous map of compact Hausdorff spaces). However
F = HomPro C (Te, ) and then clearly F (X) = X H (the points of X
invariant under H). We know that F (X) = X = ; from this it
follows that H = (e) (take for X the i for larger and larger i). Hence

:
Te is an isomorphism of topological spaces. Our aim is to show
that this is an isomorphism in Pro C .
For this it is enough to show the following: if : Te is 1 and if
e i must
pi : i are the canonical maps then every i = pi : T
factor through some T j . In other words, we must find some morphism
T j i making the diagram
e>
T
>

qj

Tj

>>
>>i
pi
>>
>>
> 
/ i

commutative. For this we must show that given i, j J such that 89


1
for every t T j , an s i such that q1
j (t) i (s). Since the
1
sets 1
i (s) are open we can find for every point x i (s) an open
1
neighbourhood U x of the form q j (t x ) for some jx J and tx T jx , such
e
that U x 1
i (s). Since T is compact, a finite covering U x1 , . . . , U xN
e
of T of this type and j > max( jx1 , . . . , jxN ) satisfies our requirements.

This shows that


Te in Pro C , and hence the map
0 = AutPro C Te AutPro C = 0
7

is a group isomorphism; it remains to be shown that this map is continuous and hence a homeomorphism. Take a fixed 0 = 0 in ,
some i I and consider the commutative diagram
0

pi

pi

i0


/ i

4. The Fundamental Group

68
90

Let U be the neighbourhood of 0 consisting of all such that


pi = i0 pi . We want to find a neighbourhood V of 0 such that
U V. There is an index j J and l I and morphisms
making all the following diagrams

p1

/e
T

0j

pl

qj

qj


/ Tj

/e
T

i0


/ Tj

pi

@@
@@
@@
@@
@@
 
/ i

commutative.
e Te such that q j = 0 q j ; these form a
Consider all : T
j
neighbourhood V of 0 and U V. Hence we are through.
Finally we observe that the isomorphism is fixed as soon
as : Te is fixed and this is in turn fixed by the choice of e
t =
e
e
(t j ) T . By a different choice of t we obtain an isomorphism
which differs from the first one by an inner automorphism of (or ).
Q.E.D.

91

Remark 4.4.1.12. One can, in fact, show that Pro C () is precisely the
category of compact, totally disconnected (Hausdorff) topological spaces on which acts continuously.

Chapter 5

Galois Categories and


Morphisms of Profinite
Groups
5.1
92

Suppose and are profinite groups and u : a continuous homomorphism. Then u defines, in a natural way, a functor Hu : C ()
C ( ); and Hu being the identity functor on the underlying sets is fundamental.
On the other hand, let C , C be galois categories, a profinite group
and H : C C , F : C C ( ) be functors such that F = F H
is fundamental. Then we can choose a pro-object Se = {S i , i j }iI of C

such that, X C , F(X)


HomPro C (Se, X). Moreover we may assume
that the S i are galois objects of C , therefore the F(S i ) are principal homogeneous spaces under the action of the i (on the right) (notations
from Ch. 4) and if we identify F(S i ), by means of the canonical element i , with i then the maps F(i j ) = i j : F(S j ) F(S i ) are
group homomorphisms (see Ch. 4). However, in the present situation
the F(S i ) are not merely sets but are objects of C ( ); as such, the group
acts continuously upon the sets to the left, and this action commutes
with the right-action of the i . This gives a continuous homomorphism
69

70

93

5. Galois Categories and Morphisms of Profinite Groups

ui : i determined by the condition that for the ui ( ) is


the unique element of i such that i = i ui ( ) (i is the canonical
element of F(S i )); for j i we clearly have: i j u j = ui ; thus, we
obtain a continuous homomorphism thus, we obtain a continuous homomorphism u : lim i = and u corresponds to H if we identify
i
C with C ().
Example. We shall apply the above to the particular case C = (E t/S ),
C = (E t/S ) where S , S are, as usual locally noetherian, connected
preschemes. Suppose : S S is a morphism of finite type and
s S , s = (s ) S . Let be an algebraically closed field containing k(s ). We define functors F : C {Finite sets} and F : C
{Finite sets} by defining:
F(X) = HomS (Spec , X), X C ,

F (X ) = HomS (Spec , X ), X C .
Denote by and the fundamental groups 1 (S , s) and 1 (S , s ).
Corresponding to the morphism : S S , we obtain a functor
: C C given by X 7 XS . We then have:
S

F (XS ) = Homs (Spec , XS )


S

Homs (Spec , X) = F(X).


94

Thus F = F ; in view of the fact that F is fundamental and the


F
F
equivalences C C (), C C ( ), we obtain a continuous homomorphism .

5.2
In this section, we shall correlate the properties of a homomorphism
u : and those of the corresponding functor Hu : C () C ( ).
5.2.1 Suppose u : is onto; for any connected object X of C ()
(i.e., acts transitively on X) any -morphism X defined by, say,

5.2.

71

e 7 x is onto and therefore so is the map X defined by e 7 x; in


other words, Hu (X) is a connected object of C ( ).
Conversely, suppose that for any connected X C (), the object
Hu (X) is again connected in C ( ). Write = lim i where the i are
i
finite groups and the structure-homomorphisms j i , j i, are all
onto; this implies that the i are all onto and by our assumption
then all the i are onto. Since and are both profinite it follows
that u : is onto. Thus: u : is onto for any connected
X C (), the object Hu (X) is connected in C ( ).
5.2.2 A pointed object of C () is, by definition, a pair (X, x) with X 95
C () and x X. By the definition of the topology on , it is clear that
giving a pointed, connected object of C () is equivalent to giving an
open subgroup H of ; the object is /H = set of left-cosets of mod H
and the point is the class H. A final object of C () is a point ec on which
acts trivially. We say that an X C () has a section if there is a C ()morphism from a final object ec to X; giving a section of X is equivalent
to giving a point of X, invariant under the action of . A pointed object
(X, x) of C () admits a pointed section (i.e., ec is mapped onto x) x
is invariant under .
Suppose u : is a homomorphism and H is an open subgroup
of such that u( ) H. Let (X, x) be the pointed, connected object of
C () determined by H. Then, in the action of on Hu (X), x remains
invariant, i.e., to say, the pointed object (Hu (X), x) of C ( ) admits a
pointed section. The converse situation is clear. Thus:
For an open subgroup H of , one has u( ) H Hu (/H) admits
a pointed section in C ( ).
5.2.3 We say that an X C () is completely decomposed if X is a
finite disjoint sum of final objects of C (), i.e., if the action of on X is
trivial.
Suppose u : is trivial, then for any X C (), Hu (X) is 96
completely decomposed in C ( ). Conversely, assume that for any X
C ()Hu (X) is completely decomposed in C ( ). Write = lim i as
i

72

5. Galois Categories and Morphisms of Profinite Groups

usual; by assumption, each composite i is trivial. Hence


u : is also trivial. Thus:
u : is trivial for any X C (), Hu (X) is completely
decomposed in C ( ).

97

5.2.4 Let H be an open subgroup of and X C ( ) the connected, pointed object defined by H . Assume that ker u H ; then
u( )/u(H ) /H in C ( ). This means that u(H ) is a subgroup
of finite index of the pro-finite group u( ) and hence is open in u( ).
Since is compact and Hausdorff we can find an open subgroup H of
such that H u( ) u(H ).
Consider now the connected, pointed object X = /H of C (). Denote by Hu (X)0 the C ( )-component of the pointed object Hu (X) of
C ( ), containing the distinguished point of Hu (X). There exists then an
open subgroup H1 of such that Hu (X)0 /H1 in C ( ). We claim
that H1 H ; in fact, u(H1 ) H and so u(H1 ) H u( ) u(H ),
hence H1 u1 (u(H1 )) u1 (u(H )) = H , since, by assumption, H is
saturated under u.
Thus, there is a pointed C ( )-morphism Hu (X)0 /H1 /H
X . If, on the other hand, we assume that a pointed, connected
object X of C () such that we have a pointed C ( )-morphism Hu (X)0
/H1 X /H , then, we must have H1 H and hence ker u
H1 H . If u is surjective, then we can say that Hu (X) X (see
5.2.1). Also ker u H is a relation independent of the choice of the
distinguished point in X /H . Thus:
ker u H a connected object X of C () and a C ( )-morphism of a connected C ( )-component of Hu (X) to X = /H . If u is
onto then X = /H Hu (X) for a connected object X C ().
In particular:
u is injective for every connected X C ( ), there is a connected
X C () and a C ( )-morphism from a C ( )-component of Hu (X) to
X.
u

5.2.5 Let be a sequence of morphisms of profinite


groups. From 5.2.3 and 5.2.4 we obtain the following necessary and

5.2.

73

sufficient conditions for the sequence to be exact:


(a) u u is trivial for any X C ( ), Hu Hu (X ) is com- 98
pletely decomposed in C ( ).
(b) Im u ker u for any open subgroup H of , with H Im u ,
we also have H ker u for any connected pointed object X
of C () such that Hu (X) admits a pointed section in C ( ), there
is a connected object X C ( ) and a C ()-morphism of a
C ()-component of Hu (X ) to X.
5.2.6 Let C be a galois category with a fundamental functor F. Let
Se = (S i )iI be a pro-object of C with usual properties (in particular, S i
are galois) pro-representing F. Let be the fundamental group of C
F
determined by F. We know then that C C ().
Let now T C be a connected object and t F(T ) be fixed for
our considerations. We form the category C = C |T ; it is then readily
checked that C satisfies the axioms (C0 ), . . . , (C4 ) of Ch. IV. We have
an exact functor P : C C defined by X 7 X T . We now define
a functor F : C {Finite sets} by setting, for any X C , F (X) =
inverse image of t under the map F(X) F(T ). Again it is easily
checked that C , equipped with F , is galois. A cofinal subsystem Se
of Se is defined by the condition: S i Se (S i , i ) dominates (T, t) in
the sense of Ch. 4. An S i Se can be considered in the obvious way 99
as an object of C ; it is then easily shown that they are galois in C and
the pro-object Se of C pro-represents F (To do the checking one may
identify C and C ()).
Let H be the isotropy group of t F(T ) in ; let also Ni be the
isotropy groups of i F(S i ), S i Se . We have a diagram of the form
GG

GG
GG
GG
GG
GG
G#

/ F(T ) t
9
rrr
r
r
rr
rrr
r
r
rr

i F(S i )

which is commutative and thus Ni H, i. Since the F(S i ) and F(T )


are all connected objects of C (), we have Ni F(S i ) and /H F(T ).

74

5. Galois Categories and Morphisms of Profinite Groups

The maps F(S i ) F(T ) are then the natural maps /Ni /H and it
follows that F (S i ) H/Ni in C (). But as we have already remarked
the S i in Se form a system of galois objects with respect to F , prorepresenting F and one thus obtains:
lim F (S i ) = lim H/Ni H.

S i Se

Ni H

Finally we remark that the composite functor F P is isomorphic


with F and therefore fundamental; following the procedure of 5.1 we
see that the corresponding continuous homomorphism u : is
nothing but the canonical inclusion H .

Chapter 6

Application of the
Comparison Theorem an
Exact Sequence for
Fundamental Groups
6.1
100

As usual we make the convention that the preschemes considered are


locally noetherian and the morphisms are of finite type (with the same
remark as in the beginning of Ch. 3).
Definition 6.1.1. (a) A morphism X Spec k, k a field, is said to be
separable if, for any extension field K/k, the prescheme XK is
k

reduced.
f

(b) A morphism X
Y is separable if f is flat and for any y Y,
X Spec k(y) is separable over k(y).
Y

We shall now state a few results which we will need for our next
main theorem. Proofs can be found in EGA.
Theorem 6.1.2. Let f : X Y be a proper morphism and F a coherent
75

6. Application of the Comparison Theorem....

76
q

OX -Module. If Y1
Y is a flat base-change
Xo

q1

X1 = XY1
Y

f1

Yo

Y1

then we have the isomorphisms

Rn f1 (q1 (F ))
Rn f (F ) OY1
OY

for any n Z+ . (Prop. (1.4.15), EGA, Ch. III).


101

6.1.3 Suppose Y is a noetherian prescheme and f : X Y a proper


morphism. Let Y Y be a closed subscheme of Y, defined by a
coherent Ideal T of OY . The inverse image of Y by f , namely the
fibre-product XY = X is then a closed subscheme of X, defined by
Y

the OX -Ideal T = f (T )OX .


Let F be any coherent OX -Module; for n Z+ , consider Fn =
F OX /T n+1 (this is a coherent OX -Module, concentrated on the preOX

scheme Xn = (X , OX /T n+1 ) and may also be considered as an OXn Module). Consider Rq f (Fn ); this is a coherent OY -Module (finiteness
theorem), is concentrated on the prescheme Yn = (Y , OY /T n+1 ) and is
in fact an OYn -Module. From the homomorphism F Fn we obtain
a homomorphism Rq f (F ) Rq f (Fn ) and since the latter is an OYn Module, we get a natural homomorphism:
Rq f (F ) OY /T n+1 Rq f (Fn ).

(*)

OY

As n varies, we get a projective system of homomorphisms. With


the assumptions we have made about X, Y, f , the comparison theorem
(EGA, Ch. III, Theorem (4.1.5)) states that in the limit, this gives an
isomorphism:
(**)

lim Rq f (F ) OY /T n+1
lim Rq f (Fn ).

OY
n

6.2. The Stein-factorisation


102

77

(Note: Both sides of (**) are concentrated on Y and they are also equal
f is the morphism of the ringed
f (lim F OX /T n+1 ) where b
to Rq b
n OX
b = (X , lim OX /T n+1 ) b
Y = (Y , lim OY /T n+1 ) obtained
spaces X
n
n
from the morphisms fn : Xn Yn , induced by f ).
We shall now specialise the above comparison theorem to the case
Y affine, say Y = Spec A. Then Y is defined by an ideal I of A. The first
A
which is precisely
member of (**) corresponds to lim H q (X, F ) A I n+1
n

q
the completion of H (X, F ) under the I-adic topology, while the second
member of (**) corresponds to lim H q (X, Fn ) and thus:
n

lim H q (X, Fn ).
H q (X, F )b

(***)

6.2 The Stein-factorisation


f

Let X
Y be a proper morphism. Then the coherent OY -Algebra
q
T , finite on Y.
f (OX ) (finiteness theorem) defines a Y-prescheme Y
To the identity OY -morphism q (OY ) = f (OX ) f (OX ) corresponds
a Y-morphism f : X Y , i.e., we have a commutative diagram
f

X>

>>
>>
>>
f >>
>

/ Y = Spec f (OX )
rr
rrr
r
r
r
rrr q
r
r
y r
r

The morphism f is again proper. This factorisation f = q f is 103


known as the Stein-factorisation of f . For details the reader is referred
to EGA Ch. III.
We shall now prove a theorem, which is of great importance to us.
Theorem 6.2.1. Let f : X Y be a separable, proper morphism. Let
f

Y be the Stein-factorisation of f . Then


X Y = Spec f (OX )
q

Y is an e tale covering.
Y

6. Application of the Comparison Theorem....

78

Proof. We have only to show that q is e tale; this is a purely local problem and we may thus assume that Y is affine, say, Y = Spec A.
We shall make a few simplifications to start with. Suppose X1 Y
is a flat base-change; then from 6.1.2 one gets:
f(Y1 ) (OX OY1 ) f (OX ) OY1
OY

OY

This means that in the commutative diagram


X1 = XY1

Xo

f(Y
1)

Spec f (OX ) = Y o

Y1 = Y Y1 = Spec f(Y1 ) (OX OY1 )


Y

OY

q(Y1 )

Yo

104

Y1

the second vertical sequence is the Stein-factorisation of f(Y1 ) .


In view of this and the fact that it is enough to look at Spec Oy,Y for
e taleness over y Y we may make the base change Y Spec Oy,Y and
assume that A is a local ring. Again, in view of the same remark and
Lemma 4.4.1 we may assume that A is complete.
Consider now the functor T on the category of finite type A-modules
M, defined by M 7 T (M) = (X, OX A M). We shall show that the
assumption T is right-exact implies the theorem.
We remark that the right-exactness of T is equivalent to the assumption that the natural map
(+)

(X, OX ) M (X, OX M)
A

is an isomorphism. Denote by T the right-exact functor


M 7 (X, OX ) M.
A

6.2. The Stein-factorisation

79

Suppose T is right exact. Then for any exact sequence An Am


M 0 we have a commutative diagram of exact sequences:
/ T (Am )

T (An )

/ T (M)

/0


/ T (M)

/0

T (An )

/ T (Am )

the first two vertical maps are isomorphisms; hence the third is also an 105
isomorphism. Conversely, if (+) is an isomorphism, clearly T is rightexact.
Now from the flatness of f and the assumed right exactness of T it
follows that (X, OX M) is exact in M and thus [from the above remark]
A

(X, OX ) is A-flat. But Y = Spec (X, OX ) and therefore q : Y Y is


flat. It remains to show that q is unramified.
Let y Y be the unique closed point of Y; denote by k the residue
field A/M = k(y). Then T (k) = (X, OX k) = (Xy , OXy ) where Xy is
A

the fibre Xk(y). But as X is Y-proper, T (k) = (X, OX k) is an artinian


Y

k-algebra which, by the separability of X over Y, is radical-free for any


r
L
Ki where
base-change K/k, K an extension field of k. Hence T (k) =
i=1

the Ki are finite separable field extensions of k. Also, by our assumption,


T (k) = (X, OX )k = (X, OX )/M (X, OX ). Let y Y be a point
A
r
L
Ki , (Ki /k are
above y. From the equality (X, OX )/M (X, OX ) =
i=1

separable) it follows that the maximal ideal M of Oy . generates the


maximal ideal of Oy and moreover that Oy is unramified over Oy . 

The proof of the theorem is thus complete modulo the assumption 106
that T is right-exact. Before proceeding to prove this we may make some
more simplifications. First we may assume that T (k) = (Xy , OXy ) =
r
L
k (this can be done as in Lemma 4.1.2 by making a faithfully flat
i=1

base-change which kills the extensions Ki/k ). Then finally we may

6. Application of the Comparison Theorem....

80

assume (Xy , OXy ) = k, i.e., Xy connected, because the connected components of X over y correspond bijectively with the connected components of Xy and it is enough to prove the theorem separately for each
component over y.
Case (a). Assume A artinian.

Our aim is to show that for any finite type A-module M, T (M)

T (M). We shall first show that T (M) T (M) is surjective.


For n large, one has M n M = 0 so that T (M n M) T (M n M) is
onto. Assume now that T (M i+1 M) T (M i+1 M) is onto, and coni
sider the exact sequence: 0 M i+1 M M i M MMi+1MM 0 of
A-modules.
We get a commutative diagram


/ T M iM
/0
/ T (M i M)
T (M i+1 M)
i+1
M M


T (M i+1 M)
107


/ T (M i M)

/T

 

M iM
M i+1 M

of exact sequences (the second row is exact at the spot T (M i M) because


T is semi-exact, i.e., for 0 M M M 0 exact, the sequence
T (M ) T (M) T (M ) is exact at the spot T (M)). The first vertical
M iM
map is a surjection by assumption, the last is a surjection since
M i+1 M
is a finite direct sum of copies of k and by our assumption (Xy , OXy )

k, we have T (k)
T (k). It follows that T (M) T (M) is onto, by
a downward induction. Finally consider an exact sequence 0 R
A p M 0. We then obtain a commutative diagram
T (R)

T (R)

/ T (M)

/0

onto

/ T (A p )


/ T (A p )


/ T (M)

of exact sequences. If follows easily that T (M) T (M) is also injection. Thus, the theorem is completely proved in the case (a).

6.2. The Stein-factorisation

81

Case (b). The general case


Let A be a complete noetherian local ring. Denote by Y1 the closed
subscheme of Y defined by the maximal ideal M . As in the comparison
theorem we may define the functors
T n (Mn ) = (Xn , OX MA/M n+1 )
A

= (X, OX MA/M n+1 )


A

where Mn = MA/M n+1 . If M is an A-module of finite type, so 108


A

is (X, OX M); as A is complete under the M -adic topology, so is


A

(X, OX M) and from 6.1.3 we obtain:


A

(X, OX M)
lim T n (Mn ).

A
n

To show that T is right-exact it is enough to show that for any exact


T (u)

sequence M
N 0 of finite type A-modules, T (M) T (N) 0
is again exact. But as each A/M n+1 is a complete artinian local ring, it
T n (u)

follows from case (a) that T n (Mn ) T n (Nn ) 0 is exact; also for
each n, ker T n (u) is a module of finite length over A. Thus, it is enough
now to prove the
Lemma 6.2.2. Let (Kn , nm ), (Mn , nm ), (Nn , nm ), n Z+ , be projective systems of abelian groups and u = (un ), v = (vn ) be morphisms
un
vn
such that, for every n Z+ 0 Kn Mn Nn 0 is exact. Assume, in addition, that for each n, m0 = m0 (n) such that
nm (Km ) = nm0 (Km0 ) m m0 (n) > n (this is the so-called MittagLeffler (ML) condition; it is certainly satisfied if the ker un = Kn are of
finite length). Then the sequence of projective limits is also exact.
Proof. The only difficult point is to show that lim vn is onto. By hy- 109
n
pothesis, for each n, m0 (n) > n with nm (Km ) = nm0 (Km0 ) m
m0 (n). By passing to a cofinal subsystem we may suppose, that, for any
n Z+
nm (Km ) = n,n+1 (Kn+1 ) m n + 1.

6. Application of the Comparison Theorem....

82

Let now (yn ) lim Nn . Choose x0 M0 with v0 (x0 ) = y0 . Assume


n
inductively that we have chosen (x0 , x1 , . . . , xn1 , xn ) with (i) r,r+1 (xr+1 )
= xr for 0 r n 2, and n1,n (xn ) = xn1 . (ii) vr (xr ) = yr for
0 r n 1, and vn (xn ) = yn .
Our aim now is to find (x0 , x1 , . . . , xn1 , xn , xn+1 ) for which the above
properties (i), (ii) hold when n is replaced by n + 1. Choose x
n+1 Mn+1
) x ) = y y = 0 i.e.
such that vn+1 (x
)
=
y
;
then,
v
(
(x
n
n
n+1
n n,n+1 n+1
n
n+1
K . By assumption, we can find z
Kn+1
to say, n,n+1 (x
)

x
n
n+1
n
n+1
such that:
n1,n+1 (zn+1 ) = n1,n+1 (zn+1 )

= n1,n (n,n+1 (x
n+1 ) xn )

= n1,n+1 (x
n+1 ) n1,n (xn ).

110

We now set xn+1 = (x


n+1 zn+1 ) and xn = n,n+1 (xn+1 ).
We then have:

n1,n (xn ) = n1,n+1 (x


n+1 ) n1,n+1 (zn+1 )
= n1,n (xn ) = xn1 .

vn+1 (xn+1 ) = vn+1 (x


n+1 ) = yn+1
and finally vn (xn ) = vn (n,n+1 (xn+1 )) = yn .

Q.E.D.

6.3 The first homotopy exact sequence


6.3.1 Some properties of the Stein-factorisation of a proper morphism. Let Y be locally noetherian and f : X Y be proper. Suppose
f

Y is the Stein factorisation of f . By using the comparison


X Y
theorem one can prove the

6.3.1.1 (Zariskis connection theorem).


The morphism f : X Y is also proper. And for any y Y the
fibre f 1 (y ) is non-void and geometrically connected (i.e., for any field
k k(y ) the prescheme X Spec k is connected).
Y

6.3. The first homotopy exact sequence

83

(For a proof see EGA Ch. III (4.3)).


One can then easily draw the following corollaries.
Corollary 6.3.1.2. For any y Y, the connected components of the fibre
f 1 (y) are in a (1 1) correspondence with the set of points of the fibre
q1 (y) (which is finite and discrete).
Corollary 6.3.1.3. For any y Y, let k(y) be the algebraic closure of 111
k(y) and X y be the prescheme Xk(y). The connected components of X y
Y

(known as the geometric components of the fibre over y) are in (1 1)


correspondence with the geometric points of Y over y.
6.3.2 Now, in addition, suppose that
(i) Y is connected.
(ii) f is separable

(iii) f (OX )
OY .
f

Assumption (iii) implies that X


Y is its own Stein-factorisation.
1
From ??, it follows that the fibres f (y) are (geometrically) connected.
f is a closed map and therefore X is also connected. Similarly, for any
y Y, X y is also connected.
Fix y Y. We have a commutative diagram:
aXo

Xy a

yY o

k(y)

Let be an algebraically closed field k(y) and let a X y be a


geometric point over ; let a X be the image of a in X. We have the
fundamental groups 1 (X y , a), 1 (X, a), 1 (Y, y) and continuous homo- 112

1 (X, a), 1 (X, a)


1 (Y, y) (See 5.1). We
morphisms: 1 (X y , a)
now have the following:

6. Application of the Comparison Theorem....

84

Theorem 6.3.2.1. The sequence

1 (X y , a)
1 (X, a)
1 (Y, y) 0
is exact.
Proof.

(a) is surjective.
In view of 5.2.1 it is enough to show that, if Y /Y is connected
e tale covering, then the e tale covering X = XY over X is conY

nected.

XY = X

Xo

f (proper)

Y (connected)

Yo

But OY = f (OX ) and therefore, we have (from 6.1.2)


f (OX ) = f (OX OY ) = f (OX ) OY = OY .
OY

OY

It follows that the fibres of f are connected, and hence X = XY


Y

is also connected.
(b) is trivial.
113

In view of 5.2.3 it is enough to show that if Y /Y is any e tale covering, the e tale covering (Y X y )/X y is completely decomposed.
Y
`
`

k(y) and hence: Y X y  X(Y k(y))


Xy.
But Y k(y) =
Y

finite

finite

(c) Im ker .
g

X is
In view of 5.2.4 it is enough to prove that: Suppose X

a connected e tale covering of X and X y


X y admits a section
(over X y ). Then a connected e tale covering Y /Y such that

XY . We need, for proving this, the following


X
Y


6.3. The first homotopy exact sequence

85
g

X
Y is proper and separaLemma 6.3.2.2. The composite h : X
ble.
Proof. h is obviously proper and flat. We have only to show that the
fibres of h are reduced, and remain reduced after any base-change Y
Spec K, K a field. For this, it is enough to prove that:
X /X e tale covering, X reduced X reduced.

We may assume X = Spec A, X = Spec A ; we have A = Ared and


we want to show that, A = Ared .
Let (Pi )ni=1 be the minimal prime ideals of A. By assumption, the
n
n
Q
Q
natural map A (A/Pi ) is an injection. So, A
(A /Pi A ) 114
i=1

i=1

is also injective and it is enough then to show that each A /Pi A is


reduced. By making the base-change A A/Pi , we may then assume that A is an integralldomain. Let a X = Spec A be the generic
point of X; then k(a) = K the field of fractions of A. The fibre over
a is = Spec(A K) and since this is non-ramified over k(a) = K, we
A

have

A K
A

r
P

Ki , Ki/K finite separable field extensions. In particular,

i=1

A K is reduced and hence for each x X , O x K is reduced and


A

Og(x )

hence so is O x O x K. It follows that A = Ared .


g(x )

Coming back to the proof of the assertion, let now X Y Y


be the Stein-factorisation of h : X Y. From the above lemma 6.3.2.2
and Theorem 6.2.1 it follows that Y Y is an e tale covering. We have
a commutative diagram

X
sss
s
s
s
ss

sss
s
s
s

ss
X = XY
oys
p1

Yo

Y = Spec h (OX )

6. Application of the Comparison Theorem....

86

Our assertion will follow if we show that : X X = XY is

115

an isomorphism. We do this by showing that:


(i) is an e tale covering.
(ii) X is connected.
(iii) rank of is 1 at some point of X .
(see the remark at the end of Ch. 3).

(i) Y Y is an e tale covering and so X X is an e tale covering.


The composite X X X is the e tale covering g and so is
an e tale covering.
(ii) We know that h is onto (6.3.1.1) and X is connected (hypothesis). Hence Y is connected and (ii) follows now from (a).
(iii) We make the base-change Y k(y) and obtain:
-
oX y
o
o
ooo 
g ooo  
o
 
o
 
ooo
o

o
oo
 
woooo


Xy
Xy
p1
  h

f
 
 


Y
k(y) o

116

It is enough to show that rank = 1 at some point of X y . Since Y /Y


n
`

ki each ki = k(y) and so X y = X y Y y =


is an e tale covering, Y y =
i=1

n
`

i=1

X yi each X yi = X y . The section : X y

k(y)

Xy

is an e tale covering and

X y is connected and hence (X y ) is a component Z of X y ; (Z) must


then be some X yi . Also the projection p1 is an isomorphism from X yi

6.3. The first homotopy exact sequence


to X y ; it follows that Z o

(p1 )

87

X yi are inverses of one another. Also

we know that the number of components of X y is equal to the number


of geometric points of Y over y, i.e., is n. It follows that the number

of connected components of X y and X y are the some. is surjective


because it is both open and closed and X is connected, therefore is

surjective and e tale and we get rank = 1 at every pt. of X y .Q.E.D. 


Remark 6.3.2.3. One may drop the assumptions f (OX ) = OY and Y
connected, in the above theorem; the assertion of the theorem will then
be:
Denote by 0 (X y , a), 0 (X, a), 0 (Y, y) the (pointed) sets of (connected) components of the preschemes X y , X, Y. Then, if f is proper 117
and separable, we have the following exact sequence:
1 (X y , a) 1 (X, a) 1 (Y, y) 0 (X y , a) 0 (X, a) 0 (Y, y) (1)

Proof. Assume to start with that X, Y are connected, dropping only the

assumption f (OX )
OY . We have then the Stein-factorisation:
f

/Y
X LLL
rr8
LLL
r
r
r
LLL
rrr
L
r
r
LL
f
LLL
rr q
&
rrr
Y = Spec f (OX ).

Applying the above theorem to f we obtain an exact sequence:


1 (X y , a) 1 (X, a) 1 (Y , y ) (e) where y Y is the image
of a X. We know then that 1 (Y , y ) 1 (Y, y) is an injection (5.2.6)
and the quotient 1 (Y, y)/1 (Y , y ) (set of left cosets mod 1 (Y , y )) is
isomorphic to the set of geometric points of Y over y. By corollary
6.3.1.3 this is isomorphic to 0 (X y , a). We thus obtain the exact sequence:
(X y , a) 1 (X, a) 1 (Y, y) 0 (X y , a) (1).
Now the assumptions about the connectivity of X and Y are dropped
in turn to get the general exact sequence.
The procedure is obvious and the proof is omitted.


Chapter 7

The Technique of Descents


and Applications
7.1
Before stating the problem with which we shall be concerned in this 118
section, in its most general form, we shall look at it in three particular
cases of interest.
Example 1. Let A, A be rings and : A A be a ring-homomorphism.
Let CA (resp. CA ) be the category of A-(resp. A -) modules. defines a covariant functor : CA CA , viz., (M) = MA = M .
A

Suppose M, N CA and u : M N is an A-linear map. Then


(u) = u1A : M N is an A -linear map.
A

Problem 1. Suppose u : M = (M) (N) = N is an A -linear


map. When can we say that u = (u) for an A-linear map u : M N?
Problem 2. Suppose M CA . When can we say that M = MA for
A

an M CA ?

Example 2. Let S , S be preschemes and be a morphism : S S .


Let CS (resp. CS ) be the category of quasicoherent OS -(resp. CS -)
Modules. defines a (covariant) functor CS CS given by F 7
89

7. The Technique of Descents and Applications

90
(F ). We again have:

Problem 1. If u : F = (F ) g = (g) is an OS -morphism when


can we say that u = (u) for an OS -morphism u : F g?
119

Problem 2. If F is any quasi-coherent OS -Module, when can we say


that F = (F ) for a quasi-coherent OS -Module?
Example 3. Let S , S be preschemes and let CS = (Sch /S ), CS =
(Sch /S ). Suppose : S S is a morphism; defines a (covariant)
functor : CS CS given by
X 7 (X) = X = XS .
S

We may again pose the two problems as in the previous examples.


It is clear that the two problems posed are of the same nature in the
different examples and admit a generalisation in the following manner:
Let C be a category and suppose that for every S C , we are given
a category CS such that for every morphism : S S in C , we are
given a covariant functor : CS CS . Assume, in addition that
(1) If S
S
S is a sequence in C , there is a natural isomor

phism: ( ) .

(2) If S S S S is a sequence in C , the diagram,


3

obtained from (1),

( 2 3 )fN

1
pp8
p
p
pppp
p
p
p
p
p
p
p

3 (1 2 )

fMMM
MMM
MMM
MMM
MMM

NNN
NNN
NNN
NNN
NN

8
qqq
q
q
q
q
q
qqq
q
q
q
q

3 2 1

120

is commutative.

(2 3 ) 1

7.1.

91

(3) (Id) = Id.

( ) of (1) has the property that


(4) The isomorphism
if either or is the identity, the isomorphism also becomes the
identity.
Problem 1. Given , CS and a morphism
u : = () () =

in

CS

when can we say that u = (u) for a morphism u : in CS ?


Problem 2. Given CS , when is of the form () for a CS ?
We shall now obtain certain conditions necessary for the above two
questions to have an answer in the affirmative.
For Problem 1. Suppose u : in CS such that u = (u) :
. Assume that S S = S exists in C ; consider the sequence:
S

S o

o
S o

p1
p2

Let p1 = p2 = and denote by , the elements (),


() of CS . Consider the morphism p1 (u) : p1 ( ) p2 ( ); accord
. There is then a (dotted) morphism 121
ing to our assumption pi
making the diagram
p1 ( )

p1 (u )

/ p ( )
1

= () _ _ _ _/ () =
commutative. In the following we also denote this morphism by p1 (u ),
i.e., we identify p1 ( ) with and p1 ( ) with . We introduce similarly p2 (u ). It is clear that with these notations we must have
p1 (u ) = p2 (u );

7. The Technique of Descents and Applications

92

in fact : both equal (u).


If this necessary condition is also sufficient (as they turn out to be in
some cases) we say that : S S is a morphism of descent.
For Problem (2). Suppose CS , with () = . Then, with the
above notations, we have:

p1 ( )
()
p2 ( ),
122

i.e., an isomorphism : p1 ( ) p2 ( ) making the diagram

p1 ( )

FF
FF
FF
FF
FF
FF
#

/ p ( )
2
xx
x
xxx
xx
x
x
x| x

()

commutative. We shall now get conditions on . Assume that S =


S S S exists in C ; let qi be the ith projection S S ; also let
S

p ji ( j i) be the morphism (qi , q j )S : S S . Consider the sequence:


S o

o
S o

p1
p2

p21

o
S o
o

p31
p32

S .

Let = p1 p21 = p1 p31 = : S S . We then have


commutative diagrams of the type:
q1 ( ) o

x
xxx
x
xx
{x
x

()cFl
FF
FF
FF
FF

q2 () o

123

p21 p1 ( )

LLL
LLL
LLL
L%
p21 ()

p ()

21
rr9
rrr
rr
rrr

p21 p2 ( )

We also denote by p21 () the (dotted) morphism q1 ( ) q2 ( )

7.1.

93

which will make the diagram

q1 ( ) o


p21 p1 ( )

p21 ()





q2 ( )

p21 p2 ( )

commutative. Using this diagram and the large diagram above we see
that
q1 ( )
q

qqqq

qq
x qq
q

() fM

MMM
MMM
MMM
M

p21 ()

q2 ()

is commutative. We make similar conventions and get similar commutative diagrams for p32 () and p31 (). Then we must have
p32 ()p21 () = p31 ()
(the so-called cocycle condition). Finally, if : S S = S S is
S

the diagonal, by the assumptions made at the beginning we must have


() = identity.

If these necessary conditions are also sufficient and if is also a


morphism of descent, we say that is a morphism of effective descent.
An of the above type is then called a descent-datum on .
Remark. Let C be the category of preschemes; if we agree that we take
S and S locally noetherian but if we dont assume S S of finite type
then it may very well happen that S = S S is not locally noetherian
S

b A a noetherian local ring). We shall deal


(e.g. S = Spec A, S = Spec A,
with this difficulty in the sections 7.2.1.3, 7.2.1.4, 7.2.1.5.

124

7. The Technique of Descents and Applications

94
Example (1).

Proposition 7.1.1. A faithfully flat ring-homomorphism : A A is


a morphism of effective descent.
Proof. A (a) We shall first show that is a morphism of descent. Let M,
N be A-modules and u : MA NA be an A -linear map. Consider
A

the commutative diagram


M

p1

/ MA = M
A

// MA A = M
A

p2

p1 (u )

N
125


/ NA = N

 
A A = N
// N
A
A

p1

p2 (u )

p2

We have to show that if p1 (u ) = p2 (u ) then an A-linear map


u : M N such that u = uIA . This will follow, if we show that
A

(i) for any M CA , M MA is an injection,


A

/ N

(ii) for any N CA , N

p1
p2

// N is exact, and

(iii) u (M) ker(N , p1 , p2 ).


(i) We know that A is a direct factor of A A by means of the map
A

a 7 a 1; thus, for any M CA , MA MA A makes


A

MA a direct summand of MA A and is in particular an inA

jection. As A is faithfully A-flat it follows that M MA is an


A

injection.

(ii) We have to prove that the sequence


N

/ N

p1
p2

// N

7.1.

95
is exact.
It suffices to prove this after tensoring the sequence with a ring B
faithfully flat over A; note that we get a similar situation with the
pair B B = BA as with A A . Take for B the ring A
A

itself. But then A A A = A admits a section : A A


A

given by (a1 a2 ) = a1 a2 . We may thus assume without loss of


generality that : A A itself admits a section : A A. 126
Consider then the commutative diagram:
N Eb E

=1

p1

/ NA
A

EE
EE
EE
E
Id . EEE
EE
" 

=1

/ N AA
NA

NA
N
Let x =

// NA A

p2

(xi ai ) be an element of N = NA such that p1 (x ) =


A
P
P
p2 (x ); that is, (xi ai 1) = (xi 1 ai ). On applying 1
we obtain:
X
X
(x1 (ai ) 1) =
(xi 1 ai );

NA , this means that


under the identification NAA
A

(xi ai ) =

X
((ai )xi 1),

i.e.,

x = ((x )) (N).

(iii) By the commutativity of the diagram at the beginning of the proof,


we have: p1 (u ) p1 = p1 u , p2 (u ) p2 = p2 u and by
assumption p1 (u ) = p2 (u ) while for m M one has p1 (m) =
p2 (m) = m 1 1 M . Thus, for m M, p1 u (m) = p2 u (m)
and (iii) is proved.
(b) It remains to show that is effective. Suppose N CA and we have 127

96

7. The Technique of Descents and Applications

a commutative diagram:
N A = p1 (N )

A
7
p1 pppp

p
ppp
p
p
pp
N NNN
NNN
NNN
p2 NNN'

A N = p2 (N )
A

where is an A -isomorphism satisfying the cocycle condition. We

N and such that


want to find an N CA such that NA
A

NA A
A

p1 ()

/ N A
A

(natural)

A NA
A

p2 ()

/ A N
A

commutes.
Set N = ker( p1 p2 ) CA (this choice of N is motivated by (ii)

N . To show that
of (a)). We always have an A -linear map NA
A

128

is an A -isomorphism, it is enough to show that is an A-isomorphism;


and for this, we may assume, as in (a), that there is a section : A A
for : A A .
For the sake of clarity and ease, we now go back to the general case
and prove:

Lemma 7.1.2. If : S S admits a section : S S , then is a
morphism of effective descent.
Proof. Let CS and : p1 ( ) p2 ( ) be an isomorphism such
that p32 ()p21 () = p31 () (notations as before). Our aim is to find an

7.1.

97

CS and an isomorphism = ()
such that the diagram
r
rrrr
r
rr
y rr
r

eLL
LLL
LLL
LLL

p1 ()

p1 ( )

/ p ( )
1


/ p ( )
2

p2 ()

p2 ( )

is commutative.
We have the diagram:
S ];o

p1

o
SO o

;;
;;
;;
;;
Id. ;;
;;
;

S S
S

p2

1
S

S o

S
S S
S

with p1 ( 1) = and p2 ( 1) = identity. Hence, if = ( ) 129


we have = ( ) = (p1 (1)) ( ) = (1) p1 ( ) and ()

, namely, = ( 1) ().
( 1) p2 ( ) = . We then get a :
We obtain thus a : p1 () p2 ( ) which makes the diagram
p1 ( )
O

p1 ()

p1 ( )

/ p ( )
2O
p2 ()

/ p ( )
2

commutative. However, we do not, in general, have a commutative dia-

7. The Technique of Descents and Applications

98
gram
p1 ( )

cHH
HH
HH
HH
H
(natural) HHH
HH
H

/ p ( )
; 2
vv
v
v
vvv
v
v
vv (natural)
vv
v
v

= (p1 ) ()

130

by means
Therfore, we want to modify : to a :
of an automorphism : (i.e., = ) such that the diagram

p1 ( )
O

/ p ( )
2O

p1 ()

p2 ()

p1 ( )

p ( )

2
]<<
A
<<

<<

<<


<<

<< (natural) 
<<

<<

<<



is commutative. This will satisfy our requirements. We first observe


that : p1 ( ) p2 ( ) also satisfies the cocycle condition. We may
consider as an element of Aut( ). To find a in Aut such as above,
i.e., such that = p2 () p1 ()1 we have only to prove the following:

Lemma 7.1.3. Let CS and consider the complex
(Aut)

Aut

p1
p2

// Aut

p31
p21
p32

/
// Aut

(notations as before). If there is a section : S S then


H 1 (Aut) = (e).

7.1.
131

99

Proof. By definition, the 1-cocycles are elements Aut for which


the cocycle condition is satisfied and the 1-coboundaries are those
Aut which are of the form = p2 () p1 ()1 , for a Aut .
Consider now the corresponding complex in C ; the section : S S
defines a homotopy operator for this complex in the following manner:

o
S \:o
SO o
::
::
::
:

Id. ::
::
:
S o

p1

p21
p31

O oo

p2

S
O

p32

S oo
S S
S

11
p1
p2

S
S S S
S

with p1 ( 1) = , p2 ( 1) = Id.,
( 1) p1 = p21 ( 1 1),

( 1) p2 = p31 ( 1 1),

and
p32 ( 1 1) = Id .

If is a 1-cocycle, set = ( 1) (). We then have:


p2 () p1 ()1 = p2 ( 1) () (p1 ( 1) ())1
= ((( 1) p2 ) ()) ((( 1) p1 ) ())1
= ( 1 1) p31 () (( 1 1) p21 ())1
= ( 1 1) (p31 () p21 ()1 )
= ( 1 1) p32 () = .
Q.E.D.

Example 2. C = (Sch), S , S (Sch); CS (resp. CS ) is the category of 132


quasi-coherent OS -(resp. OS -)-Modules.
Proposition 7.1.4. If : S S is faithfully flat, quasi compact then
is a morphism of effective descent.

100

7. The Technique of Descents and Applications

(A morphism f is quasi-compact if f 1 (U) is quasi-compact for every quasi-compact: U).


(Note: We do not assume here that is a morphism of finite type-our
hypothesis is much weaker).
Proof. Case (a) S , S both affine. The proposition in this case follows
from Proposition 7.1.1.
Case (b) S affine. There is a finite affine open cover (S i )iI of S
`
( quasi-compact). Set Se = S i . Then Se is affine and the morphism
iI

Se
S (composite of the natural map : Se S and ) is also
faithfully flat. Now consider the diagram:
e oo

q1

S
q2
{{
{{
{{
{{
}{{
(natural)
S aDD
DD
DD
DD
D 
p1
o
S o

p2

133

q31
q32

SeSe o
S

S S o
o
S

q21

p31
p32
p21

SeSeSe
S

S S S
S

If , CS and u : is a CS -morphism then u defines a


CSe -morphism e
u : () (). And from the equality p1 (u ) = p2 (u )
and the commutativity of the above diagram follows q1 (e
u) = q2 (e
u). By
case (a), a u : , a CS -morphism, such that e
u = (u); it is
immediate that u = (u); in fact this holds in every open set S i .
If CS and is a descent-datum for , e
defined in the obvious
way is a descent-datum for ( ) = e
and again by case (a), CS

().
such that e

(). It is easy to see that


T j

Case (c). S , S arbitrary. Let (T j ) be an affine open cover of S and


`
Tj
= 1 (T j ). Then the T j form an open cover of S . Let T =
j

and : T S the natural map. If F and G are quasi-coherent OS Modules denote by (S ) the group HomS (F , G ), by (S ) the group

7.2.

101

HomS (F , G ) and so on. If we set T =

`
j

T j and : T S is the

natural map, we have a commutative diagram:


(S )


(T )
 

(T T )
S

p1

/ (S )

//

p2
p1


/ (T )

p2

//

(S )
(T )

 
/ (T T )
S

By case (b), the morphisms T j T j are morphisms of effective 134


`
`
descent and therefore clearly so is T = T j T j = T . It follows
j

that the second row is exact while (T T ) (T T ) is an injection.


S

As it is clear that : T S is a morphism of (effective) descent, the


first column is also exact. Usual diagram-chasing shows that the first
row is also exact, in other words, that is a morphism of descent. We
show similarly that is also effective.
Q.E.D.

Example 3. We shall not discuss the problems (1) and (2) in their general form, in this case. However, if we restrict ourselves to the case
of preschemes affine over S , S then a faithfully flat, quasi-compact
morphism S S is a morphism of effective descent. In fact, such
preschemes are defined by quasi-coherent OS -and OS -Algebras and
we are essentially back to example (2). [It is not difficult to see that
u (resp. F ) in example (2), problem (1) (resp. problem (2)) is an
OS -Algebra homomorphism (resp. an OS -Algebra) provided we start
with OS -Algebras and homomorphisms of OS -Algebras instead of OS Modules (look at the proof of Proposition 7.1.1)].

7.2
Let S be a locally noetherian prescheme and X, Y be e tale coverings of
S . Let S 0 S be a closed subscheme of S defined by a Nil-Ideal F

7. The Technique of Descents and Applications

102

135

of OS (F is a Nil-Ideal of OS Fs nil-radical of O sS , s S ;
this is equivalent to saying that S 0 and S have the same base space or

(S 0 )red = S red ). We then have an obvious functor (E t/S )


(E t/S 0 )
given by X 7 XS 0 = X0 . We assert that defines an equivalence of
S

categories in the sense of the following


7.2.1 Main Theorem

(a) HomS (X, Y)


HomS 0 (X0 , Y0 )
(b) If X0 (E t/S 0 ), then X (E t/S )

and an isomorphism XS 0
X0 .
S

The theorem will follow from the series of lemmas below.

Lemma 7.2.1.1. Let S be locally noetherian and f : X S a separated


morphism of finite type. Then f is an open immersion f is e tale and
universally injective.
(Note: Universally injective - radiciel = injective + radiciel residue field
extensions).

136

Proof. clear.
: f is an open map and thus by passing to an open sub-scheme
of S , we may assume that f (X) = S . Clearly f is then a homeomorphism onto. Since e tale remains e tale under base-change, f(S ) is still a
homeomorphism onto, for any base-change S S ; it follows that f
is universally closed and thus proper. By Chevalleys lemma we deduce
that f is finite. Suppose then that X = Spec A , where A is a coherent
OS -Algebra. As f is e tale and universally injective, it follows that A is

locally free of rank 1 at every point of S , hence X


S . Q.E.D.

Lemma 7.2.1.2. Let f : X S be a separated e tale morphism of finite
type (S locally noetherian). Suppoe Y S is a morphism of finite type
and Y0 = V(F ) is a closed subscheme of Y defined by a Nil-Ideal F of

7.2.

103

OY . Let 0 : Y0 X be an S -morphism. Then a unique S -morphism


: Y X making the diagram
0
X o_@
@@

Y _ 0

@@
@@
@@
@@


S o

commutative.
Proof. By making the base-change Y S and considering the morphism Y0 XY we are reduced to proving the lemma in the case
S

S = Y. That is, given a Y-morphism Y0 X, we want to extend 0 to


a section of X Y.
f
Now suppose : Y X is a section of X
Y; f = identity is
e tale and f is e tale, hence is e tale. Also f is a closed immersion
and f is separated and thus is a closed immersion. Y being locally
noetherian its connected components are open and we may then assume
Y connected. (Y) will then be a connected component of X, isomorphic
to Y under ; in view of lemma 7.2.1.1 then, the sections of f are in
(1 1) correspondence with the components Xi of X such that f |Xi is 137
surjective and universally injective on Xi to Y.
Now Y0 , Y have the same base-space and hence so have X0 = XY0
Y

and X; also, the morphism f(Y0 ) : X0 Y0 obtained from f by the


base-change Y0 Y is topologically the same map as f ; hence, f is
universally injective and surjective on Xi to Y f(Y0 ) is so on (Xi )0 to
f

Y0 , i.e., the set of sections of X


Y is the same as the set of sections of
f(Y0 )

X0 Y0 and the (1 1) correspondence is given in the obvious way.


0
The lemma now follows from the fact that Y0 X can be considered
f(Y0 )

as a section for X0 Y0 .

Q.E.D.

This lemma proves part (a) of Theorem 7.2.1. To prove part (b) we
need a generalisation of the lemma.

7. The Technique of Descents and Applications

104

Warning. In the following sections 7.2.1.37.2.1.5 until the proof of


(b) of Theorem 7.2.1, we drop the assumptions made in 6.1 that the
preschemes are locally noetherian and the morphisms are of finite type.
We begin by making the
Definition 7.2.1.3. A morphism f : X Y of preschemes is of finite
presentation if
(i) f is quasi-compact
(ii) : X XX is quasi-compact,
Y

138

(iii) for every x X, a nbd. U x of x X and a nbd. V f (x) of f (x) in


Y such that f (U x ) V f (x) and (U x , OX ) is a (V f (x) , OY )-algebra
of finite presentation; i.e., (U x , OX )  (V f (x) , OY )[T 1 , . . . , T s ]/a
where a is a finitely generated ideal of a polynomial algebra
(V f (x) , OY )[T 1 , . . . , T s ].
An example of a morphism of finite presentation is a finite type sepf

arated morphism X
Y with Y locally noetherian.
Lemma 7.2.1.4. Let U be a noetherian prescheme and (A )I an inductive family of quasi-coherent OU -Algebras. Let V be the U-affine
prescheme Spec A defined by A , I , and V = Spec A where A =
lim A .

(a) Suppose we have a diagram:

Uo

X0

//
//

//
 
// 
 
V o

00

00


00

00 
 
V o

X.

..

..
 
..
.. 
 

where X , Y are finitely presented preschemes over V I


such that , X = X V
V

Y = Y V ;
V

let

X = X V,
V

Y = Y V
V

7.2.

105

HomV (X, Y).


Then, lim HomV (X , Y )

(b) Suppose is a finitely presented prescheme over V. Then, for all 139
large I, X /V , finitely presented, such that (i) ,
X X V and (ii) X X V.
V

(c) With assumptions as in (a), if the Y s and Y are locally noetherian, X/Y e tale covering X /Y e tale covering already, for
some I.
Proof.

(a) We shall be content with merely observing that it is clear


how to prove this in case everything is affine.

(b) Since U is noetherian, in view of (a) we may assume that U =


Spec C , V = Spec A , V = Spec A where A = lim A .

Case (1). Assume X = Spec B.


X/V finitely presented B/A finitely presented; let
B  A[T 1 , . . . , T s ]/a, a
finitely generated, say, by P1 , . . . , Pn A[T 1 , . . . , T s ]. Choose so
large that the coefficients of the Pi come from A . Consider the ideal
a = (P1 , . . . , Pn ) of A [T 1 , . . . , T s ] and set B = A [T 1 , . . . , T s ]/a and
X = Spec B ; for , set X = X V .
V

Case (2). X arbitrary.


Let (Xi )ri=1 be a finite affine open cover of X. In view of case (1) and
(a), it is enough now to show that each Xi X j is quasi-compact. But
the underlying space of Xi X j is that of X (Xi X j ) as is seen from
V
(X X)
V

the commutative diagram:

140

X (Xi X j )
V
(X X)

/ Xi X j
V

(natural)


/ XX
V

7. The Technique of Descents and Applications

106

As is a quasi-compact morphism the top-morphism is also


quasi-compact; since Xi X j is quasi-compact, our result follows.
V

(c) Here again we give only some indications and leave the details
to the reader. To start with we may assume everything affine. Say U =
Spec C, V = Spec A , V = Spec A, X = Spec B1 , Y = Spec B2 ,
X = Spec B1 , Y = Spec B2 . We also note that we can assume V = Y ,

and V = Y. One checks that B1/B2


lim B1 /B2 , say. Now if X/Y is

unramified then B1 /B2 = 0. But since B0 /B0 is a finite B1 0 -module


1
2
one has B1 /B2 = 0 for large . Also, if X/Y is finite and flat then B1 is
a locally free B2 -module of finite rank but then the same is true for B1
Q.E.D.

with respect to B2 for large . Assertion (c) follows.

141

Lemma 7.2.1.5. Suppose we have a commutative diagram:

Xo

X(T ) = XT o
S

Y0 =  V(F
)
_

S o

To

with: S noetherian, f e tale and separated, T S affine, Y T


f) where F is a Nil-Ideal of OY . Then
finitely presented and Y0 = V(F
a unique : Y X(T ) keeping the diagram still commutative.

Proof. Since T S is affine, T = Spec B, where B is a quasi-coherent


OS -Algebra. Write B = lim B where the B s are OS -sub-Algebras,

of finite type, of B. Set T = Spec B . By (a) and (b) of Lemma 7.2.1.4

7.2.

107

we have, for large , a situation of the following type:


XT
S

vv
vv
v
v
vv
zvv
SX o

ok T
P L

0,

Y0,

C ?

fLLL
LLL
LLL
LL
Y0 = V(F ) = Y0, Y

XT
S

H
o
fMMM
MMM
MMM
MMM

Spec B = T

uu
uu
u
u
uu
 zuu
S S o

Spec B = T

_

;
RV 
Y

fMMM
MMM
MMM
MM

Y = Y T

For large , inductive families (Y ), (Y0, ) and morphisms 0, : 142


Y0, XT such that lim 0, = 0 (see diagram). Also it is not

S
very difficult to see (using arguments similar to those in Lemma 7.2.1.4)
that the Y0, are subschemes defined by Nil-Ideals F0, of OY . Now
T is of finite type over S and is therefore noetherian; also, since Y
T is of finite presentation so is Y T and is in particular of finite
type. Hence Lemma 7.2.1.2 applies and we get a (dooted) morphism
: Y XT keeping the diagram commutative. Passing to the
S

limit with respect to we get an : T XT making the diagram


S

commutative. The uniqueness of follows from Lemma 7.2.1.2 and (a)


of Lemma 7.2.1.4.
Q.E.D.

7.2.1.6 Proof of (b) of Theorem 7.2.1
Case 1. Assume S = Spec A, A a noetherian, complete local ring; S 0 is
f0

then given by Spec(A/J) = Spec A0 , J a nil-ideal of A. Let X0 S 0


be the given e tale covering; assume that if s0 S 0 is the unique closed
point, the residual extensions at the points of f01 (s0 ) are all trivial.
Then X0 is given byL
Spec B0 where B0 is a finite direct-product
of
Lr
r
copies of A0 , say B0 =
A
;
X
=
Spec
B,
with
B
=
A
is
then
i=1 0
i=1
a solution for our problem.
Case 2. From among the assumptions in case 1, drop completeness of A 143

7. The Technique of Descents and Applications

108

and the triviality of residual extensions along the fibre f01 (s0 ).
S = Spec
Ao
Y3

faithfully flat
quasi-compact

33
33
33
33
33
33
33
33
33
X0 o
33
33
33
33 f0
3
+K 

Spec A0 = S 0 o

Spec A = S

X0 = X0 S 0
S0

f0

S 0

= Spec

A
JA

In this case we can choose a complete, noetherian, local overring A


of A such that the following situation holds: and such that if s0 S 0
is the unique closed point, then the residual extensions along the fibre

f 1
tale covering X /S such that
0 (s0 ) are all trivial. By case 1, an e

X S 0 .
X0
S

144

We have then the following situation:

7.2.

 o
SS& o
S o
&&
&&
&&
&&
&&
&&
&&
&&
& & X0 o
&&
&&
&& f0
&
%E
S0 o

p1 (X )

FF
FF
FF
FF
FF
F"

p1
p2

X0

p2 (X )

t
tt
tt
t
tt
tt
ytt o
= S S o
o
S

p1 (X0 )

f0

S 0

oo

p1
p2

JJ
JJ
JJ
JJ
JJ
JJ
%

S 0

p1
p31
p2

S = S S S

rr
rrr
r
r
rrr
xrrro
= S 0 S 0 o
S0
o

p2 (X0 )
p21
p31
p32

S 0 = S 0 S 0 S 0
S0

S0

109

7. The Technique of Descents and Applications

110

145

Since X0 /S 0 comes from below, it clearly has a descent-datum

0 : p1 (X0 )
p2 (X0 ) satisfying the cocycle condition. We want to
lift this isomorphism 0 to an : p1 (X ) p2 (X ). For this one
may be tempted to use part (a) of Theorem 7.2.1 which we already have
proved. But observe that we are now in a type of situation we anticipated
while making the remark preceding Proposition 7.1.1 we cannot make
sure that S , S 0 are locally noetherian and hence cannot apply part (a).
It is here that Lemma 7.2.1.5 comes to our rescue. By using this lemma

in the obvious way we get an isomorphism : p1 (X )


p2 (X ); the
assertion of uniqueness in this lemma proves that the we obtained
satisfies the cocycle condition. As S S is faithfully flat and quasicompact, it is a morphism of effective descent for e tale coverings (an
easy corollary to Proposition 7.1.4) and thus an e tale covering X/S

XS . Therefore
such that X
S

X S 0 XS S 0 (XS 0 ) S 0 .
X0
S

S0

Also, by construction, the descent-datum for X /S goes down to

that for X0 /S 0 . It follows that XS 0


X0 .
S

146

Case 3. S arbitrary.
By part (a), which we have already proved, it is enough to prove the
existence of an e tale covering X/S (with the required property) locally.
Let s S and U = Spec A be an affine open neighbourhood of s. We
may assume S = Spec A, A noetherian. The local ring A s is given by
A s = lim f A A f ; we then have S Spec A f Spec A s and the given
f (s),0
e tale covering X0 over S 0 = Spec(A/J), J a nil-ideal, defines an e tale
covering X s,0 over Spec(A s /JA s ) = Spec(A/J) s and then, by case 2, an
e tale covering X s over Spec A s . By (b) and (c) of lemma 7.2.1.4, f A,

f (s) , 0 and an e tale covering X f / Spec A f such that X s


X f A s . The
Af

covering X f is a solution for our problem in the neighbourhood Spec A f


of s.
Q.E.D.
Remark. The Theorem 7.2.1 shows that if S 0 S is such that (S 0 )red

= S red then the natural functor (E t/S ) (E t/S 0 ) is an equivalence; in

7.2.

111

particular, it proves that 1 (S red , s)


1 (S , s).
f

S be a faithfully flat, quasi-compact, radiProposition 7.2.2. Let S

ciel morphism. Then the natural functor (E t/S )


(E t/S ) is an equivalence.

S = S S .
Proof. Consider the diagonal S
S

Since f is radiciel (S ) = S (Cor. (3.5.10) EGA, Ch.I), i.e. S


is a closed subscheme of S , having the same base-space.
(a) Given X, Y (E t/S ), set X = XS , Y = YS .
S

To prove HomS (X, Y)


HomS (X , Y ).
Since f : S S is faithfully flat, quasi-compact it is a morphism of descent for e tale coverings. We have thus only to show that
p1
S = S S are the canoniif u HomS (X , Y ), and if S oo
p2

cal projections, then, considered as morphisms from X = X S to


Y = Y S , p1 (u) and p2 (u) are equal. We have the diagram:

147

SG W. o

S o

 .
 ...
..

..

.


u
/

X
Y

S = S S o

];S;
A
;;

;;


;;

;;



;

p1 (u )

/

/
X
Y
p2 (u )

SG W.
 ...
..

..

..


u
/ Y
X

In view of our remarks about , and Theorem 7.2.1, it is enough to


show that (p1 (u )) = (p2 (u )). But each of (p1 (u )), (p2 (u )) is
equal to u considered as morphisms form X to Y .

(b) Given X (E t/S ), to show that X (E t/S ) such that X

XS .
S

7. The Technique of Descents and Applications

112

For this, again it is enough to find a descent-datum : p1 (X )

p2 (X ). We have the diagram:


p31

SO o S

oo

oo o
p32

j
U
S
S eLL
U
U
D Z55 UpU21
p2
2
U

LL
U
U
55
UUUU

UUUU LLdiagonal
L
5

L
5
UUUU LLL
55

1
UUUU LL

UUUL
diagonal


p1

p2 (X )

p1 (X )

SO

X = 1 p1 (X ) = 1 p2 (X )
148

in view of our remarks about 1 : S S and Theorem 7.2.1, the


identity morphism i : X = 1 p1 (X ) 1 p2 (X ) = X , lifts to an
isomorphism : p1 (X ) p2 (X ). The other diagonal morphism 2 :
S S = S S S also imbeds S as a closed subscheme of S
S

having the same base-space. Then one checks easily that satisfies the
cocycle condition again by using Theorem 7.2.1.
Q.E.D.

Proposition 7.2.2 simply says that if S S is faithfully flat, quasi
compact and radiciel, s S and s S its image, then 1 (S , s )

1 (S , s).

7.3
Let k be an algebraically closed field and X, Y be connected k-preschemes. Suppose X is k-proper and Y locally noetherian. Let a X, b Y
be geometric points with values in an algebraically closed field extension K of k. Consider a geometric point c = (a, b) XY over a and
k

b. We claim first that XY is connected. Since Y is connected and


k

7.3.

113

XY Y is proper, it is enough to show that the fibres of XY Y


k

are connected and for this one has only to show that for any field k k
(*) Xk is connected.
k

The question is purely topological and we may assume X = Xred .


Looking at the Stein-factorisation X Spec (X, OX ) k it follows 149
(by making use of 6.3.1.1 and the finiteness theorem) that (X, OX ) = k,
since X is connected and k algebraically closed. On the other hand, if
X = Xk , one has (X , OX ) = (X, OX )k (flat base-change) hence
k

= k ; again looking at the Stein-factorisation we see that X is connected


6.3.1.1.
We can then form 1 (XY, c) and form the product of the natural
k

maps 1 (XY, c) 1 (X, a), and 1 (XY, c) 1 (Y, b). We then have
k

the

Proposition 7.3.1. With the assumptions made above 1 (XY, c)

1 (X, a) 1 (Y, b).

Proof. We may assume X = Xred .


Case (1). Assume K = k.
The morphism XY Y is proper and separable and the fibre over
k

b Y is to Xk = X. We then get the exact sequence:


k

1 (X, a) 1 (XY, c) 1 (Y, b) e (Theorem 6.3.2.1).


k

But the fibre over b, namely, Xk = X is imbedded in XY and


k
p1

the composite X = Xk XY X is identity. This means that


k

continuous sections for the homomorphism 1 (X, a) 1 (XY, c).


k

Thus, we get:

(i) e 1 (X, a) 1 (XY, c) 1 (Y, b) e


k

is exact,

7. The Technique of Descents and Applications

114

(ii) and the sequence splits (also topologically).

150

Case (2). K arbitrary.


The reasoning as in case (1) gives an isomorphism

1 (XYK, c )
1 (XK, a ) 1 (YK, b )
k

where c , a , b are points of (XY)K, XK, YK respectively, above


k

c, a, b. The theorem then is a consequence of the following:

Proposition 7.3.2. Let k be an algebraically closed field and X k


a proper connected k-scheme. Let k be an algebraically closed field
extension of k and let a Xk be any geometric point. If a X be the
k

image of a , then

1 (Xk, a )
1 (X, a).
k

Proof. That Xk is connected follows from assertion (*) of 7.3; the


k

same assertion also proves that if Z is a connected e tale covering of X


then Z = Zk is a connected e tale covering of X = Xk . In other
k

words, 1 (Xk , a ) 1 (X, a) is surjective (cf. 5.2.1). We shall prove


k

that it is injective by showing that every connected e tale covering Z of


X = Xk is of the form Zk for some Z (E t/X).
k

151

By Lemma 7.2.1.4 a k-algebra A of finite type, A k , and ZA

ZAk. If Y = Spec A, y Y such that


(E t/(XA)) such that Z
A

k(y) = k [since A is of finite type over (the algebraically closed) k]. One
can apply case (1) of Theorem 7.3.1 to the k-rational point (a, y) XY
k

to obtain

1 (XY, (a, y))


1 (X, a) 1 (Y, y).
k

If ZA is defined by an open subgroup H of 1 = 1 (XY, (a, y))


k

this means that open (normal) subgroups G 1 (X) and G 1 (Y)


e
e respectively) such that H G G
(defining galois coverings X/X,
Y/Y
e Y
e
(i.e. such that ZA is obtained as a quotient of the galois covering X
k

7.3.

115

e we than
of X Y). Let Z A be the lift of the covering ZA/(X Y) to XY;
k

have the commutative diagram:

e Y
e
X

k
==
}
}
==
}
}
==
~}
}
=

eo
/ ZA
XY
ZA
k

BB

BB

BB




XY
k

We claim that Z A is connected.


152
In fact, let y Y = Spec A be the generic point of Y; then k(y) =
e lying above y Y we thus
field of fractions of A k. For any e
y Y

have k(e
y) k since k is algebraically closed. If we then apply the basee k XY
e to the e tale covering Z A XY,
e we obtain
change XY
k

e
Y

the e tale covering

Z A k = Z X
Y

which is connected. It follows that Z A is connected. Thus the more Y


e
e Y
e Z A is surjective and Z A is sandwiched between X
phism X
k

e is defined by an open subgroup


e this implies that Z A /(XY)
and XY;
k

e = 1 (X) G which contains 1 (X


e Y)
e = G G ; i.e.
of 1 (XY)
k

e1 Y
e for an X
e1 (E t/X). We now obtain
ZA = X
k

(Z A )ey = the fibre of Z A over e


y

e1 Y
e k(e
e1 k(e
=X
y) = X
y)
k

and

e
Y

Z = ZA k = (ZA )y k
Y

k(y)

e1 k
= (Z A )ey k = X
k(e
y)

Q.E.D.

Chapter 8

An Application of the
Existence Theorem
8.1 The second homotopy exact sequence
153

Let A be a complete noetherian local ring and S = Spec A; let s0 S be


the closed point of S . Let X be a proper S -scheme and set X 0 = Xk(s0 ).
S

Let a0 X 0 be a geometric point of X 0 (over some fixed algebraically closed field k(s0 )) and a0 X be its image in X. Assume
that X 0 is connected.
a0 X o

X 0 a0

proper f

s0 Spec A o

k(s0 )

Theorem 8.1.1. The sequence


e 1 (X 0 , a0 ) 1 (X, a0 ) 1 (S , s0 ) e
is exact; and we have the isomorphism
1 (S , s0 )  G(k(s0 )/k(s0 )).
117

8. An Application of the Existence Theorem

118

(Note: Compare with Theorem 6.3.2.1).


The proof of the theorem is a consequence of the following results.
Proposition 8.1.2. In the above Theorem 8.1.1 assume A is an artinian
local ring. With the same notation, the same assertions hold.
Proof. Since the introduction of nilpotent elements does not affect the
fundamental groups (Theorem 7.2.1). We may assume that A = k(s0 ).
In this case, the assertion 1 (S , s0 )  G(k(s0 )/k(s0 )) is clear. Next,

if characteristic k(s0 ) = p and if k = (k(s0 )) p , then the morphism


Spec k(s0 ) Spec k is faithfully flat, quasi-compact and radiciel and

hence by Proposition 7.2.2, we may replace k(s0 ) by (k(s0 )) p , i.e., we


may assume k(s0 ) is perfect. In this case, k(s0 ) is the inductive limit
lim ki of finite galois extensions ki of k(s0 ); set Xi = X ki and ai =
iI
k(s0 )
image of a0 in Xi .
a0 X o


k(s0 ) o
155

Xi ai o

X 0 a0


ki o

k(s0 )

By Lemma 7.2.1.4 an e tale covering of X 0 is determined by an e tale


covering of some Xi and the latter is uniquely determined modulo passage to X j , j i.
One thus gets the isomorphism:

lim 1 (Xi , ai )
1 (X 0 , a0 )

[The injectivity follows from the fact that for any open subgroup H of
= 1 (X 0 , a0 ) there exists, by Lemma 7.2.1.4 an index i and an open
subgroup H (i) of (i) = 1 (Xi , ai ) such that /H (i) /H (i) . The surjectivity follows because otherwise there would exist a set E C ((i) )
with two points a and b in E such that a and b are in the same connected
component of E with respect to the action of all ( j) ( j i) but a and
b lie in different components with respect to the action of ; again by

154

8.1. The second homotopy exact sequence

119

7.2.1.4 this is impossible because a connected component of E in C ()


can be realised in some C (( j) )].
On the other hand, we assert that each Xi /X is galois and G(ki /k(s0 ))

Aut(Xi /X).
In fact, suppose we have a situation of the following type:
a0 X o

X = Xk a
k

ko

with X universally connected over k. Then X /X is a connected e tale 156


covering. Also, we have: deg(X /X) = rank = number of geometric
points in the fibre 1 (a0 ) = deg(k /k) = number of automorphisms of
k /k number of automorphisms of X /X number of geometric points
in the fibre 1 (a0 ) (because X is connected - see the proof of Lemma

4.4.1.6). Hence Aut(k /k)


Aut(X /X) and X /X is galois. Now for
every i I we have an exact sequence:
(e) 1 (Xi , ai ) 1 (X, a0 ) Aut(Xi /X) (e)
(see 5.2.6). Since each 1 is pro-finite, by passing to the projective limit
we obtain an exact sequence:
(e) 1 (X 0 , a0 ) 1 (X, a0 ) G(k(s0 )/k(s0 )) = 1 (S , s0 ) (e)

Proposition 8.1.3. Let A be a complete, noetherian local ring and S =
Spec A. Let X be a proper S -scheme such that if s0 S is the closed
point of S the fibre X0 = Xk(s0 ) is universally connected. Let a0
S

be a geometric point of X0 with values in some algebraically closed

k(s0 ). If a0 X is the image of a0 then canonically 1 (X, a0 )

1 (X0 , a0 ).

Proof. This will come from the fact that the natural functor (E t/X)

(E t/X0 ) is an equivalence.

8. An Application of the Existence Theorem

120

(a) If Z, Z (E t/X) then

HomX (Z, Z )
HomX0 (ZX0 , Z X0 ).
X

157

In fact, let A (Z), A (Z ) be the coherent, locally free OX -Algebras


defining Z, Z over X. If M is the maximal ideal of A, set An =
A/M n+1 , Xn = XA/M n+1 , Zn = ZXn and so on, n Z+ . We then
A

have:

HomX (Z, Z )
HomOX Alg. (A (Z ), A (Z)).

Now for the OX -Modules, we have:

HomOX (A (Z ), A (Z)) = (X, H omOX (A (Z ), A (Z)))

lim (Xn , H omOX (A (Z ), A (Z))A/M n+1 )

A
n

by the Comparison theorem, (where H omMX is the sheaf of germs of


OX -homomorphisms); since the A (Z ), A (Z) are coherent and locally
free, we have:
H omOX (A (Z ), A (Z)) A/M n+1 = H omOXn (A (Zn ), A (Zn ));
A

therefore,

lim HomOXn (A (Zn ), A (Zn )).


HomOX (A (Z ), A (Z))

158

However, this holds also for homomorphisms of OX -Algebras, because


the condition for an OX -Module homomorphism to be an OX -Algebra
homomorphism can be expressed by means of commutativity in dia
grams of OX -Modules. Therefore, HomX (Z, Z )
lim HomXn (Zn , Zn ).
n
But by Theorem 7.2.1, the natural map HomXn (Zn , Zn ) HomX0 (Z0 , Z0 )
is an isomorphism; therefore one obtains:

lim HomXn (Zn , Zn )


HomX0 (Z0 , Z0 ).

Q.E.D.

(b) If Z0 is an e tale covering over X0 , then there exists an e tale cover


ing Z/X such that Z0
ZX0 .
X

For proving this we need another powerfull theorem from the EGA.

8.1. The second homotopy exact sequence

121

8.1.4 The Existence Theorem for proper morphisms Let A be a noetherian ring and I and ideal of A such that A is complete for the I-adic
topology. Let Y = Spec A and f : X Y be a proper morphism. Set
An = A/I n+1 , n Z+ , and Yn = Spec An , Xn = XYn . Suppose, for every
Y

n, Fn is a coherent OXn -Module such that Fn1 Fn OXn1 . Then


OXn

a coherent OX -Module F such that, for each n, Fn


F O Xn .
OX

(For a proof see EGA, Ch. III, (5.1.4).)


Coming back to the proof of (b), the e tale covering Z0 X0 is
defined by a coherent locally free OX0 -Algebra B0 . By Theorem 7.2.1,

for each n, we get a coherent locally free OXn -Algebra such that Bn1

Bn OXn1 . By the existence theorem, a coherent OX -Algebra B


OXn

such that Bn
B OXn . Set Z = Spec B. We claim that Z/X is the 159
OX

e tale covering we are looking for. It is clear that ZX0


Z0 . It remains
X

to show that Z/X is e tale.


We first observe that since A is a local ring and f is closed any (open)
neighbourhood of the fibre f 1 (s0 ) is the whole of X. Also, if Z/X is
e tale over the points of the fibre X0 = f 1 (s0 ), then Z is e tale over X
at points of an open neighbourhood of X0 ; for, if x0 X0 then B x0 is
free over O x0 ,X and hence B is a free OX -Module in a neighbourhood
of x0 ; and similarly for non-ramification (use Ch. 3, Proposition 3.3.2).
Therefore it is enough to prove that Z X is e tale at points of Z lying
over the fibre X0 = f 1 (s0 ).
Let then x0 X0 ; choose an affine open neighbourhood U of x0 in
X such that B0 is OX0 -free over U0 = U X0 . Set (U, OX ) = C and

J = M C, the ideal generated by M ; then, for n Z+ , Cn = C/J n+1


(U, OXn ). Similarly, if M = (U, B) then

Mn = M/J n+1 M
(U, Bn ).

We know that M0 = (U, B0 )


(U0 , B0 ) is free as a C0 =
(U0 , OX0 )-module. Choose a basis for M0 over C0 . Lift this basis to M
and let L be the free C-module on these elements and : L M be the

8. An Application of the Existence Theorem

122

natural C-linear map; then we have an exact sequence of C-modules:


0 ker L M coker 0.
160

For n Z+ , set Ln = L/J n+1 L and let n be the Cn -linear map


Ln Mn defined by . We claim that each n is an isomorphism.
That 0 is an isomorphism is clear from the construction. Assume
that 0 , . . . , n1 are all isomorphisms and consider the exact sequence
n

0 J n /J n+1 Cn Cn1 0
of Cn -modules. Since Ln , Mn are Cn -flat (recall that Bn is OXn -flat)
we get the following commutative diagram of exact sequences of Cn modules:
0

/ J n /J n+1 Mn
C
O n

/ Mn
O

/ Mn1
O

/ J n /J n+1 Ln

n1

/ Ln

Cn

/0

/ Ln1

/ 0.

But
M
Mn

C

(U, OX0 ),
M n+1 A
i=1
r

J n /J n+1
n

J /J

and
161

J /J

n+1

Mn

Cn
n+1

Ln

Cn

r
M

(U, B0 ) =

i=1

r
M

r
M

M0 ,

i=1

L0 .

i=1

It follows that the first vertical map is an isomorphism; so is n1 by


inductive assumption. Hence n is an isomorphism.
b b
b are the J-adic completions of C, L, M and if b
If C,
L, M
is the
b
C-linear
map defined by , it follows that b
is an isomorphism. Thus,

8.1. The second homotopy exact sequence

123

one obtains: (coker )b= 0 and ker ker(L b


L). It follows then that
B is locally free at x0 X0 , i.e., Z X is flat at points over X0 . It is
also unramified at points above X0 since Z0 X0 is unramified and for
non-ramification it suffices to look at the fibre.
The proof of Proposition 8.1.3 is complete.

Proof of Theorem 8.1.1. The isomorphism 1 (S , s0 )


G(k(s0 )/k(s0 ))
follows from Proposition 8.1.3 when X = S .
Consider now the diagram
a0 X o


s0 S o

X0 a0 o

X 0 a0

k(s0 ) o

k(s0 )

By Proposition 8.1.2 we get the exact sequence:


(e) 1 (X 0 , a0 ) 1 (X0 , a0 ) G(k(s0 )/k(s0 )) (e).

One now uses the isomorphisms 1 (X0 , a0 )


1 (X, a0 ) (Propo
1 (S , s0 ) to get he required exact
sition 8.1.3) and G(k(s0 )/k(s0 ))
sequence.
Q.E.D.

Chapter 9

The Homomorphism of
Specialisation of the
Fundamental Group
9.1
162

With the notations and assumptions of Chapter 8, let s1 be an arbitrary


point of S = Spec A and X 1 = Xk(s1 ). If a1 is a geometric point of X 1
S

and a1 its image in X and if X 1 is connected we get a sequence


1 (X 1 , a1 ) 1 (X, a1 ) 1 (S , s1 ) (e)
such that the composites are trivial. We have continuous isomorphisms
: 1 (X, a1 ) 1 (X, a0 ) and : 1 (S , s1 ) 1 (S , s0 ). Consider now
the diagram:
1 (X 1 , a1 )

/ 1 (X, a1 )

/ 1 (S , s1 )

(e)

/ (X , a )
1 0 0

/ 1 (X, a0 )

/ (e)

/ 1 (S , s0 )

/ (e).

This is not necessarily commutative; but it is commutative upto an


inner automorphism of 1 (S , s0 ). (It is readily seen in the example of
125

126

163

9. The Homomorphism of Specialisation.....

section 5.1 that, with the notation therein, the continuous homomorphism there, is determined upto an inner automorphism of 1 (S , s) if we
take for s a point different from (s )-see Ch. 4.)
Now the lower row is exact. This implies that there is a continuous
homomorphism 1 (X 1 , a1 ) 1 (X 0 , a0 ) which is clearly determined
upto an inner automorphism of 1 (X). This is called the homomorphism
of specialisation of the fundamental group.

9.2
With the same assumptions as above further suppose X/S separable.
Since we have assumed X 0 connected, the condition f (OX ) = OS is
automatically satisfied and then the upper sequence in the diagram of
9.1 is also exact. In this case the homomorphism of specialisation is
surjective.

9.3
Let Y be locally noetherian and X Y be a separable, proper morphism
with fibres universally connected. Suppose y0 , y1 Y with y0 (y1 ).
Let a0 , a1 be geometric points of X 0 = Xk(y0 ) and X 1 = Xk(y1 )
Y

respectively. Then there is a (natural) homomorphism of specialisation


1 (X 1 , a1 ) 1 (X 0 , a0 ) which is surjective.
In fact, in view of Proposition 7.3.2 we may make the base-change
by ,Y and apply the above considerations to points above y0
Y Spec O
0
and y1 .
This is the so-called semi-continuity of the fundamental group.

9.3.

127

Appendix to Chapter IX

The Fundamental Group of an Algebraic


Curve
Over an Algebraically Closed Field
Let k be an algebraically closed field and X a proper, nonsingular, 164
connected algebraic curve over k (i.e., a k-scheme of dimension 1).
Case 1. Charac. k = 0.
We compute 1 (X) in this case by assuming results of transcendental
geometry.
In view of Lemma 7.2.1.4 and Proposition 7.3.2 one can assume that
k = C, the field of complex numbers. Then one has an analytic structure
on X and on every X (E t/X) (see GAGA - J.P. Serre); one can then
show that the natural functor
(
)
Finite topological coverings of the
(E t/X)
analytic space X h defined by X

is an equivalence [cf. GAGA and Espaces fibres algebriques - Seminair


Chevally 1958, (Anneaux de Chow) - See especially Prop. 19, Cor. to
Prop. 20 and Cor. to Theorem 3].
One thus obtains:
c
top
1 (X, a) 1 (X h , a), the completion of the topological fundamentop
tal group 1 (X h , a) with respect to subgroups of finite index. This
top h
group 1 (X , a) is known if X is a nonsingular, proper, connected
curve of genus g. It is a group with 2g generators ui , vi (i = 1, . . . , g)
subject to one relation, namely
1
1 1
1 1
(u1 v1 u1
1 v1 )(u2 v2 u2 v2 ) . . . (ug vg ug vg ) = 1.

Before going to the case charac. k = p , 0, we shall briefly recall 165


some evaluation-theoretic results. (Ref. S. Lang-Algebraic Numbers or
J.P. Serre-Corps Locaux).
Let V be a discrete valuation ring with maximal ideal M and field
of fractions K. Let K /K be a finite galois extension and V the integral

128

9. The Homomorphism of Specialisation.....

closure of V in K . Choose any maximal ideal M of V and consider


the decomposition group of M , i.e., the subgroup gd (M ) of G(K /K)
of elements leaving M stable. Then there is a natural homomorphism
gd (M ) G(k(M )/k(M )); the kernel gi (M ) of this homomorphism
is called the inertial group of M . The maximal ideals of V are transformed into each other by the action of the galois group G(K /K) and
the various inertial groups are conjugates to each other (Compare with
Lemma 4.2.1). For the following definition it is irrelevant which of the
M we take and therefore we shall write gi , or sometimes gi (K /K) instead of gi (M ). We say that K is tamely ramified over V if the order of
gi is prime to the characteristic of k(M ) and that K is unramified over
V if gi is trivial.
One has then the following:
(1) If K /V is tamely ramified then the inertial group gi (K /K) is
cyclic.
(2) If K K K is a tower of finite galois extensions and if V is a
localisation of the integral closure of V in K and if K /V, K /V
are tamely ramified then K /V is tamely ramified.
166

In addition, one has an exact sequence:


(e) gi (K /K ) gi (K /K) gi (K /K) (e)
(cf. Corps Locaux, Ch. I, Propostion 22).
(3) Let M be a uniformising parameter and n Z+ be prime
to charac. k(M ). If K contains the nth roots of unity, then K =
K[X]
is a finite galois extension, tamely ramified, with galois
(X n )
group = gi (K /K) Z/nZ.
Lemma (Abhyankar). Let L, K be finite galois extensions of K, tamely
ramified over V with order gi (L/K) (= n) dividing order gi (K /K)(= m).
Let L be the composite extension of L and K . Then L is unramified
over the localisations of the integral closure V of V in K .

9.3.

129

Proof. One checks that a monomorphism gi (L /K) gi (K /K)


gi (L/K) such that the projections gi (L /K) gi (K /K) and gi (L /K)
gi (L/K) are onto (see 2)). Since the orders of gi (L/K) and gi (K /K)
are prime to p = charac. k(M ), so is the order of gi (L /K), i.e., L /V
is tamely ramified; hence gi (L /K) is cyclic (1)). As n|m, it follows
that each element of gi (L /K) has order dividing m. But gi (L /K)

gi (K /K) is onto. Thus, gi (L /K)


gi (K /K). From (2) the kernel of
this map is gi (L /K ) which is therefore trivial.
Q.E.D.

Case 2. Charac. k = p , 0.

167

Definition. We say that a Y-prescheme X is smooth at a point x X (or


X Y is simple at x) if an open neighborhood U of x such that the
natural morphism U Y admits a factorisation of the form
e tale

U Spec OY [T 1 , . . . , T n ] Y
(the T i are indeterminates).
Note. Smoothness is stable under base-change.
Definition . For a Y-prescheme X, the sheaf of derivations of X over Y
is the dual of the O-Module XY ; it is denoted by gXY .
We shall assume the following
Theorem 1. (SGA, 1960, III, Theorem (7.3)).
Let A be a complete noetherian local ring with residue field k. If
X0 /k is a projective, smooth scheme such that
H 2 (X0 , gX0 |k ) = (0)
and

H 2 (X0 , OX0 ) = (0);

then a projective, smooth A-scheme X such that Xk X0 .


A

Let k be an algebraically closed field of charac. p , 0 and X0 a 168


nonsingular (= smooth in this case) connected curve, proper (hence, as

9. The Homomorphism of Specialisation.....

130

is well-known projective) over k. Consider the ring A = W(k) of Wittvectors; this is a complete, discrete valuation ring with residue field k; if
K is the field of fractions of A, then charac. K = 0, and K is complete for
the valuation defined by A. The conditions of Theorem 1 are satisfied by

X0 . Therefore a projective smooth A-scheme X such that X0


Xk.
A

The X0 is universally connected; but then it follows that the generic fibre
is also universally connected (Use Stein-factorisation; Note that Spec A
has only two points). Furthermore, as X is smooth over A, the local
rings of X are regular (one has to show that if X Y is e tale and Y is
regular then X is regular). Finally we mention the important fact that by

Proposition 8.1.3, we have: 1 (X)


1 (X0 ).
We may thus replace X0 by X. Let a0 be the closed point of Spec A
and a1 the generic point; set XK = XK, X K = XK, where as usual
A

K is the algebraic closure of K. If a0 Xk = Xk is a geometric


A

point over a0 and a1 X K , a geometric point over a1 then we have the


homomorphism of specialisation
1 (X K , a1 ) 1 (Xk , a0 ) 1 (X)
169

which is surjective (9.2).


From case 1 one already knows about 1 (X K , a1 ) (charac. K = 0).
Thus it only remains to study the kernel of the above epimorphism.
This amounts by 5.2.4 to studying the following question: Given a connected e tale covering Z of X K which is galois, when does there exist a

Z (E t/X) such that Z


ZK. [Remember: a connected e tale covA

ering is galois if the degree of the covering equals the number of automorphisms. Also note: we have integral schemes because, for instance,
our schemes are connected and regular. Therefore we may consider the
function fields R(Z) and R(X K ) of Z and X K . Clearly if Z is galois over
X K then R(Z) is a galois extension of R(X K ).].
By Lemma 7.2.1.4, a finite subextension K1 of K and a ZK1
(E t/XK1

= XK1 ) such that Z


ZK1 K. Then the above question takes the
A

K1

form: Given a connected Z (E t/X K ), which is galois when does

9.3.

131

there exist a Z (E t/X) and a finite extension K2 of K1 such that

ZK2 = ZK1 K2
ZK2 ?
K1

Now, for any finite extension K of K, let A be the integral closure


of A in K ; then by a corollary to Hensels lemma it follows that A
is again a discrete valuation ring whose residue field is again k (recall
that k is algebraically closed). Thus (X) 1 (Xk ) 1 (XA ) where
XA = XA , again by Proposition 8.1.3. Thus the question becomes:
A

Given ZK1 (E t/XK1 ), universally connected and galois, when does 170
there exist a finite extension K2 of K1 such that ZK2 = ZK1 K2 comes
K1

from a ZA2 in (E t/XA2 = XA2 ) (where A2 is the integral closure of A in


A

K2 )?
Consider any finite extension K of K1 and let A be the integral
closure of A in K . We are given a situation of the form:
ZK

+ p - generic point of the fibre over the closed point.

XK
generic point

Spec K |{z}
0 -closed point (residue field k).
Spec A

The ZK1 (resp. ZK = ZK1 K ) are connected and hence, as we have


K1

already remarked, integral. Let R(ZK1 ), R(ZK ), R(XK1 ) and R(XK ) be


the function fields of ZK1 , ZK , XK1 and XK . Then R(ZK1 ) is a finite
galois extension of R(XK1 ) and
R(ZK ) = R(ZK1 ) K = R(ZK1 ) R(XK ).
K1

R(XK1 )

Our aim now is to choose, if possible, the field K K1 such that R(ZK )
is unramified over XA . By this we mean the following: consider the
normalisation Z of X = XA in R(ZK ); we want Z to be unramified 171
over X .
(Note: Since ZK is regular, so certainly normal, we have that Z K

ZK , because the process of normalisation is unique. See EGA, II,


(6.3) and also S. Lang, Introduction to Algebraic Geometry, Ch. V).

132

9. The Homomorphism of Specialisation.....

We can also express this as follows: we want Z to be unramified over


the whole of X and not merely on the open subscheme XK = X K .
A

Consider the integral closure A1 of A in K1 . Let p be the generic


point of the fibre in XA1 = XA1 over the closed point (as a space, this is
A

nothing but X0 ). Then p is a point of codimension 1 in XA1 ; furthermore


XA1 /A1 is smooth and therefore the local ring O1 of p (as a point in XA1 )
is regular. Therefore O1 is a discrete valuation ring in R(XA1 ) = R(XK1 ).
Similarly define O in R(XK ) = R(XA ) for any K K1 . It is easily
checked that O is the integral closure of O1 in R(XK ). Now, any open
set in XA , containing XK and the generic point p of the fibre over the
closed point of Spec A is such that its complement is of codimension
2 in XA . We have the following theorem depending on the so-called
purity of the branch locus:
Theorem 2. (SGA, 1960-1961 expose X, Cor (3.3))
172

Let P be a locally noetherian regular prescheme and U the complement of a closed set of codimension 2 in P. Then the natural functor

(E t/P)
(E t/U) is an equivalence. (For a proof see SGA, X, 1962).
[For the special case of the theorem which we need, there is a direct
proof in SGA, X, 1961, p. 16. However even for that proof one needs
the following result (which we have not proved in these lectures). An
unramified covering of a normal prescheme is e tale. (SGA, I, 19601961, Theorem (9.5)).]
Thus, it is enough to prove that K K1 such that R(Z ) is unramified over X at the point p because Z is clearly flat over X = XA at the
point p (the local ring O of p in X is a discrete valuation ring).
If is a uniformising parameter of A1 , it is also a parameter of O1 .
Let then n Z+ be such that (n, p) = 1 and set K = K1 [T ]/(T n ).
Then K /K1 is a finite galois extension and R(XK ) R(XK1 )[T ]/(T n e).
Hence R(XK ) is tamely ramified over O1 , and has an inertial group of
order n.
Assume now that the degree of the galois covering Z over X K is
prime to p = charac.k. Then one may take n equal to this degree. By
Abhyankars lemma one then has that R(Z ) is unramified over X above
the point p . Thus one has proved the

9.3.

133

Proposition 3. The kernel of the (surjective) homomorphism of specialisation 1 (X K , a1 ) 1 (X0 , a0 ) is contained in the inter-section of the
kernels of continuous homomorphisms from 1 (X K ) to finite groups of
order prime to p.
Therefore if, for a profinite group G , we denote by G (p) the profinite 173
group lim Gi where the limit is taken over all quotients Gi of G which

are finite of order prime to p, then we have:


(p)

(p)

1 (X0 , a0 ).
1 (X K , a1 )
(One can also say: G (p) is the quotient of G by the closed normal subgroup generated by the p-Sylow subgroups of G ). Since we know, by
topological methods, the group 1 (X K , a1 ) we obtain:
Theorem 4. If X is a nonsingular, connected, proper curve of genus g

(p)
over an algebraically closed field k of charac. p , 0, then 1 (X)

G (p) where G is the completion with respect to subgroups of finite index


of a group with 2g generators u, ui , vi (i = 1, 2, . . . , g) with one relation:
1 1
1 1
(u1 v1 u1 v1
1 )(u2 v2 u2 v2 ) . . . (ug vg ug vg ) = 1.

Bibliography
[1] Bourbaki, N. Algebre Commutatif (Hermann, Paris)
ements de Geometrie Algebrique (EGA) [2] Grothendieck, A. El
Volumes I, II, III, IV. (I.H.E.S. Publication)
[3] Seminaire de Geometrie Algebrique (SGA) - Volumes I, II,
(1960-61) I, II (1962)
[4] Lang, S. Introduction to Algebraic Geometry (Interscience Publication)
[5] Serre, J.P. Corps locaus (Hermann, Paris)
[6] Groupes algebriques et corps de classes (Hermann, Paris)
[7] Espaces Fibres algebriques (Expose 1, Seminaire Chevalley
(1958) Anneaux de Chow et applications)
[8] Geometrie Algebrique et Geometrie Analytique (Annales de
lInstitut Fourier, t.6, 1955-56, pp.1-42)

135

You might also like