Lecture Notes in Mathematics A collection of informal reports and seminars Edited by A. Dold, Heidelberg and B. Eckmann,...
76 downloads
1462 Views
5MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
Lecture Notes in Mathematics A collection of informal reports and seminars Edited by A. Dold, Heidelberg and B. Eckmann, Z0rich
100 M. Artin Massachusetts Institute of Technology, Cambridge, Mass.
B. Mazur Harvard University, Cambridge, Mass.
Etale Homotopy 1969
Springer-Verlag Berlin. Heidelberg. New York
This research has been supported by NSF.
All rights reserved. No part of this book may be translated or reproduced in any form without written permission from Springer Verlag. 9 by Springer-Verlag Berlin" Heidelberg 1969 Library of Congress Catalog Card Number 75-88710 Printed in Germany. Title No. 3706
Contents
.
A g l o s s a r y of the categories in which we shall work, and fibre resolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.
Pro-objects
3.
Completions.
4.
Cohomologlcal
5.
Completions
6.
Homotopy
7.
Stable
in the h o m o t o p y
category. . . . . . . . . . . . . . . . . . . . . . .
.............................................. criteria
for
6 20 25
~ -isomorphism. ................
35
...............................
60
of completions . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
70
results.
...........................................
75
8.
Hypercoverlngs.
...........................................
93
9.
The V e r d l e r
functor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
111
and fibratlons.
groups
lO.
The f u n d a m e n t a l
group. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
117
ll.
A proflniteness
theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12@
12.
Comparison
theorems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
129
1.
Limits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
147
2.
Pro-objects
3.
Morphisms
of pro-objects. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.
Exactness
properties
Appendix:
and p r o - r e p r e s e n t a b l e
functors.
of the pro-category.
................ 15@ 159
. . . . . . . . . . . . . . . . . . 163
References .....................................................
167
-iThese notes are an expansion of the results announced in [2].
The material was presented in a seminar at Harvard
University during the academic year '65-'66. ~ Our aim is to study the analogues of homotopy invariants which can be obtained from varieties by using the etale topology of Grothendieck.
Using the constructions of Lubkin [21]
or Verdier [3], we associate to any locally noetherian prescheme
X
a pro-object in the homotopy category of simplicial
sets (cf. 2), which we call the etale homotopy type of the prescheme
X .
For a normal variety over the field of complex
numbers, we show that
Xet
is a certain profinite completion
of the classical homotopy type.
This comparison result, to-
gether with a number of others, is in section ]2. Much of our work consists in setting up a reasonable theory of homotopy for pro-simplicial sets.
This homotopy category
turns out to be quite amenable to the techniques of classical algebraic topology.
One may establish the analogues of Hurewicz
and Whitehead theorems (4.3), (4.4), (4.5), and also one has available the techniques of Postnikoff decomposition.
There is
*We thank G. Borkowski for her flawless typing of the manuscript.
-2 -
O.2
a direction in which this pro-category the classical homotopy category:
is more flexible
It is sufficiently
rich so
as to admit pro-flnlte and p-adic completion functors (3.4)).
than
(cf.
Many of our results about schemes are phrased in
terms of these functors. For instance, an algebraically let
X1 , X2
from
X
let
X
be a connected,
closed field
K
of
A
of characteristic
K
in
where the subscript A
cl
denotes
denotes pro-flnlte
there is an example due to Serre singular variety into
C
X
damental groups. homotopy type.
1 Xcl
and
i Xcl
Thus However,
type of a nonslngular
~ .
Then
'
the classical homotopy type, completion.
We remark that
[28] of a projective non-
over a number field
such that
independent
zero, and
A
XI 2 cl ~ Xcl
K
scheme over
be the schemes over the complex numbers obtained
via two embeddings
and where
pointed
2 Xcl
K
and Imbeddings
have non-isomorphlc
of
fun-
do not necessarily have the same
as Serre indicated
to us, the homotopy
simply connected projective
of complex imbedding.
surface is
This gives an affirmative
response in that case to a question raised in [ 2 ]. know whether the same holds for higher dimensional
We do not simply con-
nected nonsingular varieties. Atlyah has suggested analogous Let
L/K
questions for bundles:
be an extension of algebraic number fields, with
a subfield of the complex numbers.
Let
V
denote a variety
K
-3defined over
K , and
over
An imbedding
V~L
.
E
an algebraic vector bundle defined c
of
induce a complex vector bundle Vcl = in
(V~C)cl
K(Vcl )
.
depends on
schemes over a field
of characteristics
let
K
p ~ q
xl,~
X,Y
rings
will
over the topological
space
Rp,Rq
over
of
zero.
K
Ec
Suppose given
with residue fields
and that
reductions modulo
X,Y
have
each of the valua-
be the schemes over the complex numbers obof
K
in
are simply connected. A
Then
1
"
Note that in this case we may conclude X1
and of
C , and assume finally
A
Xcl
motopy groups of
~
be smooth proper connected
respectively,
tained from some imbeddings X 1,~
K
of characteristic
isomorphic non-degenerate
that
Ec
into
c .
two discrete valuation
Let
L/K
The problem is to study how tbe class of
As another example,
tions.
0.3
y1
that the classical ho-
are isomorphic
(abstractly)
since they are finitely generated abellan groups with isomorphic profinite
completions
This result is in the spirit of the question originally posed by Nashnitzer: zero with isomorphic
Given two varieties
reductions modulo p , what additional
ditions will imply that the varieties or, to go even further, stract question:
in characteristic
diffeomorphic?
How much information
from an honest simpliclal
set
K
con-
are themselves homotoplc, We are led to the abis lost when one passes
to its proflnite
completion
4 -
K ?
Suppose
CW-complex.
that
K
is of the homotopy type of a finite
Then we conjecture
number of homotopically
that there is at most a finite
distinct
completion is isomorphic
to
version of the conjecture, gebraic description
0.4
simplicial
K .
In section 7, we prove a stable
and moreover,
we give a complete al-
of the set of distinct
with isomorphic profinite there exist distinct
sets whose profinite
completions.
stable homotopy types
We show by examples
that
stable homotopy types whose completions
are
i s omo rphi c. If where
V E
is a variety over a field is the separable algebraic
on the ~-adic homotopy type of variety,
this representation
K , and closure,
~ = V~
.
G = Gal~/K) then
If
G
V
operates
is an abelian
yields no further information
the operation on ~-adic cohomology.
However,
than
for a more general
V , it may be expected that more precise information is contained in this representation. the Grassmanlan
Perhaps the special case
of m-dimensional
V = G m'n ,
planes in n-space,
K = Q , is
of interest. Indeed,
if
the comparison
^
theorem
the pro-simplicial space
m,n GC
denotes
set
~-adic completion,
(12.9) an operation of ~,n
.
one obtains using G = Gal~/Q)
In other words,
is endowed with a natural action of
~-adic completion.
on
the topological G
on its
The recent ideas of Quillen [13] (b) with
regard to the conjecture of Adams suggest that an analysis of the above action may provide fruitful results in topological K- the o ry. We have not attempted
to develop an analogue of topological
-
O.5
5-
K-theory associated to the etale topology.
Quillen's ideas
cry out for the development of such a functor, however.
Look-
ing even further ahead, one may wonder whether analogues of the Novikov-Browder invariants
(or the more recent theory of
Sullivan) describing the dlffeomorphy types representing a given tangential homotopy type are amenable to definition for an arbitrary scheme, proper and smooth over a field.
This
would bring one closer towards an understanding of the above problem of Washnitzer. We should like to signal two errors in our announcement [ 2 1.
Namely,
(5.4) of [ 2 ] is obvious nonsense.
statement is (8~8) below.
Theorem
A correct
(6.3) of [ 2 ] lacks a co-
homologlcal dimension hypothesis, as in (12.5).
6
w
i.i
-
A glossary of the catesories in which we shall work,
and fibre resolutions.
We will make free use of the classical theory of simpllcial objects with values in an arbitrary category. references are [8], [12], [17], [23], [25].
The principal If
ST
gory of simplex types (objects being ordered sets and morphisms are monotone maps), and gory generated by gory
for
[O,...,n]
is the full subcate-
i < k , then we have, for any cate-
ST
ST(C) . to
(ST/k)(C) . to
C .
C ,
It is the category of contravarlant functors
C .
Similarly we have the category of simplicial
objects truncated at level
ST/k
An=
C , the category of slmpllcial objects with values in
denoted from
Ai
ST/k
is the cate-
k ~ with values in
C , denoted
It is the category of contravarlant functors from The functor which "truncates at level
k ,"
*/k- ST(C) --> (ST/k)(C) will have a left (resp. right) adJolnt, denoted cOSkk)
provided
C
sk k
is closed under finite inductive
nlte projective) limits.
(resp. (resp. fi-
We also consider the compositions,
coskk: ST(C)--> (ST/k) (C) --> ST(C) sh: ST(C)--> (ST/k) (C) --> ST(C) and note that the adjointness properties give us functorial morphlsms, SkkX--> X ,
and
X--> coskkX .
-
Both of these functors understandable
(and for general
different).
Briefly,
that of
1.2
C
the simpllclal
set
SkkX
whose truncation at level
k .
The slmpllcial
able from
X
than
In fact, in all dimensions
k .
is the simplicial
set
is that subsimk
agrees with
simplices in dimencOSkkX
plainer language,
Hom(skkAn, x)
An
pliclal map
of
is the n-simplex,
cOSkkX
greater of
where in this formula
set, which as a functor from
The reader should interpret
n-simplices
is obtain-
n , the n-slmpllces
is represented by the object of the same name in
set).
is the cate-
by adding certain simpllces in dimensions
are given by the set
cOSkkX
C
admit
the situation is hardly any
X , and which has no nondegenerate
sions greater than
An
and coskeleton)
in the case where
gory of sets
X
-
(the skeleton,
visualizations
pllclal set of
7
ST ST
to Sets (or in
regarded as a simpllcial
the above formula for the
in this way:
Every time we have a sim-
SkkAn--> X , we are given a canonical way of extend-
ing it to a simplicial map
An --> X .
The reader should become at home with the coskeleton functor. It is the exceptionally
good functorial properties
allow one to achieve a functorial for slmplicial
sets
obtain, ultimately, pro-simplicial Since
(and functoriality
Let morphism,
cosk k
that
decomposition
is crucial if one is to
any sort of Postnikoff-type
decomposition
for
sets).
cosk k
is defined as a (finite) projective
commutes with projective inductive
Postnikoff-type
of
limits.
Similarly,
sk k
limit,
it
commutes with
limits. C
be the category of sets.
or synonymously,
fibre map
We have the notion of Kan (cf. [23], or for a more
-8general treatment, (B,b)
If
is given as a pointed
be referred
if
one-polnt
f: E--> B simpllcial
to as the fibre of
The functor = Am
[25]).
cosk k
k ~ m .
Thus
1.3 is a fibre map, where f-l(b)
set, then
f .
preserves Kan morphlsms.
Also
cOSkk(e ) = e , k ~ 0 , where
slmpllcial set, and
will
cOSkk(Am ) e
is the
cOSkk(I).. = I , k > 0 , where
I
is the one-slmplex. If
F--> E--> B P
is a sequence of pointed simpllclal morphism,
and
F
is its fibre,
cOSkkF-->
has the property that
sets, where
p
is a Kan
then
cOSkkE p--~ cOSkkB
cos~F
is the fibre of
p' , since
cosk k
commutes with fibre products. It is of course also true that for homotopies,
which suggests
simple interpretation
that
cosk k
k > 0 , cosk k (k > 0)
preserves
should have a
as a functor on the homotopy category,
and
indeed we shall see this below. Let agreeable
S
denote the category of simplicial
simpliclal
sets.
The most
sets with which to work are those which sat-
isfy the Kan condition.
(A simpllcial
set is said to satisfy the
9 -
Kan,
1.4
[17], [12] 3.1 or Extension condition if
morphism. plexes. )
K--> e
is a Kan
These simplicial sets will be referred to as Kan comLet
So
denote the full subcategory of Kan complexes.
The notion of homotopy has good properties only when restricted to the subcategory
So
(see Kan, On C.S.S. Complexes [18] for
more discussion of this), and the technique one has for extending homotopy notions to all of tain adJoint functor:
S
is the construction of a cer-
Let
So/(,,,)A> sl (~) denote the inclusion functor of the homotopy categories. (On C.S.S. Complexes) Kan constructs a left adJoint to he calls
Ex ~ .
In [18], i , which
See also [12] 2.3.1 for a more complete treat-
ment from the point of view of "Calculus of Fractions."
Explic-
itly, Homo(EX~K,L) ~ Hom(K, iL)
where the homs are taken in the homotopy categories, plicial set, and
L
a Kan complex.
Remarks:
(i)
K
a sim-
Kan had not
formulated the notion of adJointness yet when [18] was written, and no such explicit statement as above is made in [18].
However
one easily checks this adjointness, using the morphisms of functors (loc. cit. 5.2):
Ex ~.i ident.
> ident. > i .Ex ~ e @@
10-
1.5
together with homotopy commutativity
for any map
f
of the dia-
gram
K
> Ex~L
e
e
Ex~K
>
L
f
The main category in which we will do homotopy theory is the extended homotop~ K
are simpllcial
morphisms
from
category of simplicial
sets, and for simplicial
X
to
Y
in
K , denoted
sets.
sets [X,Y]
be the set of homotopy classes of morphisms
from
~ ,
The objects of
X, Y
the set of
is defined to X
to
Ex'Y.
Note that
[x,x] = [x, Ex|
[Ex X, Ex'Y]
and the natural inclusion of the homotopy category of Kan complexes into
~
is an equivalence
of categories.
For short we refer to
as the simplicial category. When we wish to speak of a simplicial map from one simplicial set to another
(rather than a morphism in the category
there is the possibility of confusion, former by referring
~) and
we signal that we mean the
to it as an actual simplicial map.
As usual, we shall have occasion to use certain closely related homotopy
categories
orate one that we need,
of diagrams.
first, and obtain
We define
the most elab-
the others as full
-
11
1.6
-
subcategories. ~sq:
The extended homotopy category of simpliclal squares:
Objects are actual commutative squares of s~mpliclal sets, and actual slmpllclal maps:
f
A
i
C
> B
> D
A morphlsm is an equivalence class of actual commutative diagrams,
Ex~A,
Ex~f I
> Ex~B ,
f S
Ex~g'
A m >
B
k
D
C m ~
J
Ex~C '
.> Ex~D , Ex~i,
The equivalence relation is homotopy for such morphlsms of diagrams. Ko,pairs:
The extended homoto~y category of p o l n t e d c o n n e c t e d
simpllclal pairs: slmpliclal sets
Objects are pointed maps of pointed connected X --> Y
regarded as a square as follows:
-
e
12
I.Z
-
> e
1: [
X
> Y
The extended homotopy category of simplicial pairs:
~pairs:
X --> Y
Objects are maps of simplicial sets
regarded as a square
as follows:
[
l
.> Y
X
K
0
:
The pointed connected simplicial category (for short):
Objects are maps
e --> X
simplicial set, and category of
X
Kpair s .
where
e
is the one point constant
is connected; regard (Thus the subscript
~o
as a full sub-
indicates more gen-
O
erally "pointed connected objects.") ~an sq:
The full subcategory of
generated by those
~o, sq
squares whose horizontal maps are fibre maps. ~anpairs:
The full subcategory of
~o,pairs
generated by
those pairs whose horizontal maps are fibre maps. The category
~
CW-complexes, denoted
is equivalent to the homotopy category of ~ .
One passes from
~
to
by
the geometric realization functor, and from
~
to
by
the singular functor (these functors being adjoint: of [19] and also
[22],
related categories to
[12]).
As with
K
II,
S ,
cf. Prop. 9.1
there are closely related
, which we will use.
13 ~o:
1.8
The pointed homotopy category of connected pointed
CW-complexes. ~pairs:
The homotopy category of CW-pairs.
continuous maps of CW-complexes, commutative
squares,
f: X --> Y
Objects are
where morphisms are
again,
f
X
> Y
,l
I
X'
..> yt
considered up to homotopy equivalence. The singular functor to
~o
and
~pairs
to
S
may be interpreted as bringing
~pairs
We may form the categories
" pro-~ , p r o - ~ o , prO-~pair s , etc.
These will be referred to as the pro-simplicial pointed pro-simplicial etc.
category,
By a pro-simplicial
We will also pass,
~o
category,
the pro-simplicial
the
pairs category,
set we will mean an object of pro-~ .
on occasion,
to the equivalent
categories pro-~,
pro-~ o . (The reader may wonder why we do not stick either to the category of CW-complexes The reason is this.
or to the category of Kan complexes.
On the one hand, we have occasion to make
very strong use of fibre resolution and coskeleton functors, ing rise to Postnikoff
decompositions
have been systematically Kan complexes.
niques of attaching
etc., and these techniques
and quite elegantly developed
On the other hand, cells,
giv-
only for
we also use the "adjoint"
to kill specific obstructions,
tech-
etc.
- 14 -
1.9
which have been systematically exposed only for CW-complexes. ) Fibre Resolutions:
The problem is to "replace a map, up to
homotopy, by a fibre map. Proposition 1.1:
,!
The inclusion functors
Kan--> Ko,pair s ~ansq --> ~o, sq
are equivalences of categories. Proof:
One may give various procedures for functorially
associating to any actual simpliclal map of pointed connected pairs
p: E --> B
actually-commutative
E
(1.2)
c
>
triangles,
Et
p~a B
such that ~o,pairs
p' "
[12], where
is Kan, and
a: p --> p'
is an isomorphism in
For example, we may take the construction a
is, in their terminology,
and therefore an isomorphism in
(5.5.1) of
an anodyne extension,
~o,pairs "
The construction is
functorial and hence yields isomorphy-inverses
to the inclusion
functors appearing in the statement of proposition i,i. that we do not need that pl : E' --> B
E
and
to be a fibre map.
B
Note
be connected in order for
All that is necessary is that
-
E --> B
15
-
l.lO
be surjective on connected components.
pointed connected pairs to diagrams
The functor from
(1.2) will be referred to as
fibre resolution. The 3 by 3 Lemma:
Given a diagram of pointed connected
simplicial sets and actual simplicial maps,
f Xll
> X12
(6) X21
such that
g,h
h
'> X22
induce surJections on fundamental groups, we will
form, in a certain sense, a double fibre resolution of (8).
What
is meant is that we will end up with a three by three square array of simplicial sets and maps, any vertical or horizontal line of which will form, up to homotopy, a fibre triple. To begin, let us fibre-resolve
(5) vertically
(i.e. make the
vertical maps fibre maps) and consider the fibre triple obtained.
Xo1
(6v)
l ,L v
> X02
L l v
Xll
-> X12
~l
> X22
- 16 -
The base point of
XI2
i.ii
actually lles in
X02 .
a surjectlon on fundamental groups, we have that
Since X02
g
induces
is connected.
Therefore we may form the horizontal fibre resolution of the above diagram, yielding:
00 ~ >
(6vh)
Lemma 1.2: mental groups.
Ol
X02
i
ih
X20 ~ >
X~I ~ >
Suppose that
g, h
__>
I
X22
induce surJections on funda-
Then the left-hand row of (6vh) is, up to homo-
topy, a fibre triple. Proof:
Denote (6vh) by
~i " and form the three-by-three
diagrams c~I c
as follows:
a2
c~2 c
is made from
m3
c
(~#
ml ' and
~#
is made from
fibre-resolving the bottom three vertical arrows, from
~2
~i "
by
is made
by fibre-resolving the right-hand three horizontal arrows.
Denote by of
aS
~3
Then
Remark:
r i, c i , the upper row and left-hand column, resp. rl, rS, c2, c4 A morphlsm
~
are flbre-triples. of fibre-triples
- 17
-
1.12
A
> A'
B
>
C
> C'
,[ [ ,l l
with connected
fibres
A,A'
B'
w h i c h induces
on two out of three of the s i m p l i c i a l induces
a homotopy
exact h o m o t o p y classical
equivalence
sequence
Whitehead
vi
and
ci
and
sets involved,
on the third.
of a fibre
triple,
This
equivalence actually
is by the long
the five lemma and the
theorem.
If, w h e n d e a l i n g with triples as the "fibre"
a homotopy
g
A --> B --> C , we refer to g
as the "proSection,"
have a common
"fibre,"
A
then we m a y say that
which we denote
Xi .
The
n a t u r a l maps
c I c c 2 c c 3 c c$
induce h o m o t o p y - i s o m o r p h i s m s our remark
,
v I c v2 c v 3 c v4
of the "projections."
We m a y a p p l y
to the i n c l u s i o n maps:
x 1 = x2 = x 3 ~ x4
where
the composite
isomorphism
since
of the first vI
and
v3
two maps must be a homotopyare fibre-triples.
of the last two maps must be an i s o m o r p h i s m
since
The composite c2
and
c#
18 -
are fibre-triples.
1.13
Thus all maps are homotopy-isomorphisms,
con-
cluding the proof of the lemma. Examples:
Let
X
be in
Kan condition.
Let
n ~ 0 .
Ko
and suppose it satisfies the
Form the object
X --> c o s ~ X
in
Ko,pairs ' and fibre resolve it, to obtain the fibre triple,
X (n) --> X ' - >
giving us the canonical .
If
X
cos~X
(n-l)-connected flbre d e c o m p o s l t l o n o f
is a general object in
the Kan condition, by applying
Ko 9 we first make it satisfy
Ex ~
to it, and then follow the
above procedure to obtain its canonical composition.
(n-1)-connected fibre de-
To indicate that we have applied
Ex ~ , we denote
the sequence obtained:
x(n) _> X' --> c o s ~ X ' p'
.
We also obtain sequences,
X (n) --> c o s ~ + i X '
where
Pn+l = c~
--> Pn+l
' using that
cos~X'
coskn+iCOSk n = cosk n .
This latter sequence, is the n-th fibre triple in the canonical Postnikoff decomposition of Maclane space o f type
X .
(~n(X),n)
Clearly
X(n)
is an Eilenberg-
.
This being functorlal, we have the identical situation in pro-K o .
Thus, for
X
in
pro-~ o
we have objects of
pro-Van ,
-
19-
1.14
x(n) _> x' -> cos~x' X(n)->
for all
cos~+IX'-->
n ~ 0 , where, again
trivial except in dimension
X(n)
cos~X'
has homotopy pro-groups
n , where
~n(X(n)) --> ~n(X) .
-
w
20
-
2.1
Pro-objects in the homotopy category.
We work in this section with
~o ' the category whose objects
are connected pointed CW-complexes and whose morphisms are pointed homotopy classes of maps between them. phisms from
X
to
Y
by
IX,Y] .
functorial constructions in
~o
Denote the set of mor-
We are going to recall some
which yield
(cf. App.) analogous
constructions in pro-~ o : To begin with, the homotopy and homology groups are functors on
~o ' hence if
X = [Xi]
is a pro-object,
its homotopy and
its homology with values in some abelian group
A E (Ab)
are nat-
urally defined as pro-groups and/or pro-abelian groups:
v n(X) = {Vn(Xi ) } (2.1) Hn(X,A ) = [Hn(Xi,A)]
,
A E (Ab) .
For our purposes, it is preferable to take the actual limit group when defining cohomology. and since
lim
Since cohomology is contravarlant,
is an exact functor,
this works well.
Thus the
cohomology is (2.2)
Hn(X,A) = lim Hn(Xi,A)
A E (Ab)
i
which is an abelian group. Homology and cohomology with values in twisted abellan groups can be defined in the obvious way: Vl(X) = [Vl(Xi ) ]
on
Suppose given an operation of
A , i.e., a map
2.2
- 21 -
Vl(X
in pro- (grps) .
) . T_> Aut(A)
Then
Hn(X,AT) =
lim -@
(i,r (2.3) ~ ( X , AT) = [Hn(Xi,A~) } (i,$)
where the index category consists of pairs is a homomorphism
~l(Xi) --> Aut(A)
(i,$)
representing
such that ~ .
It is
immediately seen that this category is cofinal with the original index category
[i] , hence that twisted coefficients may for
practical purposes be thought of as obtained from a compatible family of maps
~l(Xi) --> Aut(A)
.
Although fibre sequences and related constructions are not functorial in ial. For
~o
in general,
certain important cases are functor-
For instance, as remarked in w
there is the coskeleton.
X 6 ~o ' its coskeleton is characterized by the following
universal property:
(2.4)
The homotopy groups of
cosknX
vanish in dimensions
> (n-l) , and the canonical map
x - > cOSknX is universal with respect to maps into objects whose homotopy is zero in dimension > (n-l) . Thus we can define cosknX
for
X = [Xi] 6 pro-~ o
as the
22 pro-object Let
{cos~Xl} X --> Y
2.3
-
.
be a map in
~o
which
(i) induces an isomor-
phism on homotopy in dimensions < n , and (ii) such that trivial homotopy in dimensions > n .
F
is (n-1)-connected.
The homotopy classes of maps of any (n-1)-connected 0Y
sion > (n-l) .
are trivial since
0Y
has
Then if we represent the
map up to homotopy by a fibratlon, its fibre
loop space
Y
W
to the
has zero homotopy in dimen-
Therefore the exact sequence of homotopy classes
of maps ([29]w [w, oY]-> [W,F]-> [W,X]
shows that (2.5)
F
F
is characterized in
is (n-l)-connected,
trivial, and Thus if
Xi --~ Yi
F--> X
~o
as follows:
the composed map
is
is universal for these properties.
is a filtering inverse system of maps in
each of which satisfies the above assumptions fixed
F--> X--> Y
~o
(i), (li) for some
n , then there is a canonically determined inverse system
of maps Fi --> Xi --~ Yi
where
Fi
is the fibre of the map
Xi --> Yi " etc...
This situation arises for instance when
Yi = c~
' so
that Postnikov decompositions and related constructions extend naturally to the pro-category,
with the obvious definition that
(2.6)
G-
K(G,n) = [K(Gi,n)}
[Gi] E pro-(grps)
.
23
2.#
-
A similar notion is that of covering space.
H
l(X)
is a subgroup, where for the moment
Suppose X
is a CW-
complex.
Then the covering space
group
is characterized in the homotopy category
H
~
corresponding to the sub~o
by the
following property, as is easily verified:
(2.7)
For any
W
the map
the subset of the subgroup Of course if
H
X H --> X
[W,X] H
[W,~]
identifies
of maps which carry
~l (W)
with into
(x) .
of
is a normal subgroup, then
~
is Just the
fibre (2.5) of the map X--~ K(~I(X)/H,I ) .
Now let
X = [Xj)
a sub-pro-group.
be in pro-~ , and let
Replacing
assume (App. 3.3) that
H
X, H and
X
H~-> ~I(X)
be
by isomorphic objects, we may have the same filtering cate-
gory and that we are given compatible in~ections
Hjrepresenting the map.
l(Xj)
Via (2.7), we obtain an inverse system
of maps (Xj)Hj--> Xj ,
hence a map of pro-objects X H --> X .
-
For any
W = [Wi ] E
(Xj) H
24
pro-~ o , a map
iff some representing map
-
2.5
W--> Xj W i --> Xj
factors through does, and this will
J occur (2.7) iff the map Since
~l(W) --~ ~l(Xj)
(W, XH) = lim~ (W,(Xj)Hj)
factors through
Hj .
, we obtain
J Corollary
(2.8):
Let
X E pro-~ o , and let
a monomorphlsm of pro-groups. with a map for each
X H --~ X
There is an
W E pro-~Eo , [W,X HI
Note that in particular,
together
is carried to the subset of ~l(W) --~ ~l(X)
this characterizes
of the choices we made in the description. XH
X H E pro-~ o
be
which is characterized by the property that
of maps such that the induced map
refer to
H--~ ~l(X)
[W,X]
factors through H. XH
independently
We will, naturally,
as the coverlng space determined by
H .
25 w
Completions.
(3.1):
Definition gory of
(groups)
A class
@
of groups
is a full subcate-
satisfying
(o)
0
(i)
A subgroup
Ee_.,.
0--> A--> B--> B 6 @
3.1
iff
of a C-group
C--> 0
is in
is an exact
@ .
sequence
Moreover, of groups,
if then
A, C E ~ .
A class
@
is called a complete
class if in addition
the
following axiom holds (ii)
If
A , B 6 @ , then the product
self indexed by Of course,
B
is in
the index
B
AB
of
A
with it-
@ . is just to give a bound
on the cardinal-
ity which is required. We will be primarily class of finite groups, ucts of primes
@
consists
Corollary
in the case that
@
is the
or of finite groups whose orders are prod-
coming from a given set of primes.
clearly complete that
interested
classes.
However,
These are
as we shall see,
the case
of all groups has some interst for us.
(3.9) :
A complete
class
~
has the following
properties :
(i) G
and
then
If K
K
GDNmK
are groups
is normal in
N , and if
contains
a subgroup
H
such that G/N , N / K
normal in
G
N
is normal in are ~-groups,
such that
G/H E ~ .
-
(ii) that
If
G 6 ~ , and if
26
3.2
-
A
is an abelian G-module
A E @ , then the cohomology groups
Hq(G,A)
To see (i), note that the normalizer of N .
Therefore
the set
dexed by a ~-group hence is a ~ g r o u p ~-group
too.
the standard apply
C .
of conjugates
Thus
N/nK a
by (2.1)(ii).
of
are in in
K
G in
contains
G
G/aK a
is in(N/K) c ,
is a subgroup of
By (3.1)(i),
@ .
is
a
(ii), it suffices to remark that the terms in
cochain complex are of the form
A G•215
, and to
(3.1) (ii). Let
of maps map
For
[K a]
K
such
G = G~>
~ --~ ~'
[Gi]
be a pro-group,
C , where
C
and consider the category
varies in a class
is a commutative
triangle
9
G
~ , and where a
> C
C'
It follows easily from axiom
(i) above that the opposite
gory of this category is filtering; cofinal sub-category.
Therefore
group which we denote by Clearly
~
cate-
and it obviously has a small
the range
C
determines
and call the C-completion
of
a proG .
is a pro-object in the category of ~-groups, and the A canonical map G --> G is universal with respect to maps into pro~-groups.
~
We restate this fact as
Cor011ary
(3.3):
Let
~
be a class of groups.
The inclusion
27
3.3
of p r o - @ into pro~grps) has an adjoint
A9
9 pro-(grps) --> p r o - ~ .
Equivalently,
for any
G E pro-(grps),
restricted to the category If
C
~
is pro-representable
is a class of groups, we will denote by
subcategory of
~o
~ .
in
G .
C~ o
the full
Our object in this section is to
prove the analogue of (3.2) for
pro-~ o
Hom(G,.)
consisting of pointed CW-complexes whose homo-
topy groups are all in
Theorem
the functor
(3.4) :
Let
~o
:
X E pro-~.
There is an object
A X--> X , called the th~ C-completion of
and a map
is universal with respect to maps from
X
A X E X , which
to elements of p r o - G ~ o .
Equivalently, (3.4'):
The inclusion of p r o - G ~ o
in pro-~ o
has an adjoint
A 9 p r o - ~ o --> p r o - ~
(3.4"): (X,.)
Let
o
.
X 6 pro-~/o , and consider the covariant functor
restricted to the category
representable in
~o
~o
"
This functor is pro-
"
The equivalence of the three assertions follows immediately from the definitions of pro-objects.
Of course,
pro-~ o
the object in
which represents the functor (X,.) of (3.4") is just A the Crcompletion X of (3.4). Thus we may think intuitively of A X as obtained from X by ignoring all information except that pertaining to maps to objects of
~o
"
28 -
3.#
We will prove the theorem in the form of (3.4"). to be shown category
J
(App. 1.2) is that given
X 6 pro-~ o , the opposite
to the category of maps
gory, where a morphism
~ --> ~'
What has
X ~> W
is a filtering
is a commutative
cate-
triangle
r X
> W
"
WI
and that
J
has a small cofinal subcategory.
lim~ Hom(Xi,W )
when
X = IX i} E pro-~ o
and
Since
Hom(X,W)
=
W E ~o ' one sees
immediately that it suffices to prove the theorem for each of the X i , i.e., in the case that Now it follows from
X 6 ~o "
(3.1,
@
is closed under
products
of groups.
product
W 1 x W2 .
(1.2b).
This axiom reads as follows in our situation:
(3.5)
is a diagram in
Hence if
(i)) that
Thus
J
are in
satisfies
(1.2a).
@~o " so is their It remains to show If
> Wl_7_ > w2
x
~o
W 1 , W2
with
i@ = j~
and with
there is a factorization
X
> W
WI
Wl, W 2 E @~o ' then
29
with
ik = jk
such that
Let us represent complexes
-
W E ~o
"
(3.5) by a diagram of actual maps of CW-
such that
i~
is homotopic
letters for the representing maps). h: i~ N j$ .
3.5
to
j~ .
(We use the same
Choose such a homotopy
We are going to construct inductively a sequence
of factorizations
Lqq
X
X
W1
for each jSq .
q
such that
~n(Xq)
More precisely,
we take
E ~
Xq_ 1
and the homotopy
hq: i~q ~ j~q
of
iSq-1 ~ J~q-1 "
Xq_ 1
Xq
is to
(q+l)-cells,
is to extend the homotopy
Then the direct limit
~n(X)
E ~
W
of the system
Denote by
N
, and let
on a set
S
Since
Gs: Dq+l -~ W1
Replacing
X
by
~: F - ~
(q+l)-cells
N
to
X . ~: ~q(X) --~
be a surjec~iive map of a free group
of generators
~(s)
of
be a pointed
N
onto
N .
For each
~s
"
s E S ,
continuous map representing
= 0 , we may find a continuous
extending
Xq_ 1 , we
n ( q , and we are to construct
the kernel of the homomorphism
ms: 3D q+l --~ X
~(s) .
Xq , we may by induction assume
for
by the addition of some
~q(W1)
of
is already constructed.
may assume that
let
q ~ 0 ,
i~q N
Xq 's, has the required properties.
that
F
For
by the addition of certain
To prove the existence
Xq
n ~ q , and that
X = Xo .
be obtained from
hq-l:
for
Fix such a
Gs
pointed map
for each
s E S o
-
Construction the pointed
i:
Z
Z
Z
For
which is the
the base point,
(q+l)-skeleton
= (Dq+lxo)U(SDq+ixI)O(Dq+lxl)
is homeomorphic
mo rphi sm.
denotes
of
form Dq+lxl.
as follows:
Z = 8 (Dq+ixI)
Thus
3.6
-
d E D q+l
If
CW-complex
We may describe
3 0
with
S q+l .
.
Fix such a (pointed homeo-
s 6 S , define a pointed map
fs: Z --> W 2
as
follows :
(•
On
Dq+ix0
(ii)
On
8Dq+ixI
between
i@
and
(iii) On
, let , let
Dq+lxl
~q+l(W2)
the h o m o t o p y
h
' where
fs = has
h
is the h o m o t o p y
JW . , let
After the i d e n t i f i c a t i o n class in
fs = iBs "
of
which
to
fs = J6s " Z
with
represents
XU m D q+l .
sq+l ' fs
determines
the obstruction
Since
S
a
to extending
freely generates
S
we obtain in this way a h o m o m o r p h i s m Construction Let
(A,B)
Y
2:
Let
V
f: F ->
be an integer
denote the pointed
CW
greater than
pair
Ay = D q+l V D q+l v ... v D q+Iv By = S q
v Sq
v ... v S qY
We must fix an explicit pointed map
cv:
(Dq+l'sq) ->
(A,B)y
q+l(W2) 0 .
F ,
31
-
This map should have the property represents
el + r
+ "'" + cy
than
1 , and it represents
Here
ei
is the class in
Let
d
denote
-
3.7
that the class
~c
in the case where
the product ~q(B)
r162
represented
the base point of
q
cy by
E is greater when
q = 1 .
S~ c B
. Define, in analogy Y with construction 1 , Z(y) to be the (q+l)-skeleton of A xI . Y Then we may also describe Z(y) as follows:
z(y)
where
Zi
is the
the class
is represented
v z2
(q+l)-skeleton
m a y then be interpreted represents
-- z I
by
Construction
3:
as a minlmal word in
D~+lxI
6y in
Given an element
are
.
The map .
CyXid
This
c
Y
~q+l(Z(y)) , where
6i
z 6 F , we may write
z
S , in a unique manner:
eI
r
Y
.
z = S1
where the
v z
c : Z --> Z(y) Y
61+62 + . . . +
Z i c Z(y)
v ...
of
as a map
A
~ 1 .
az: By--> X , 8z:
... sy
r
Define maps
A
Y
--> Wl ' fz:
Z(y) --> W 2
as follows: (i) 6s i
on
If
r
= 1 , let
D q+l c Ay
(ii) If
r
and let
= -i , let
~z = ms i fz = fs i ~z = ~si J
on
S q c By
on on
let
Z i c Z(y) siq , let
Bz =
. 8 z = BsiJ
-
q+l Di , and let
on
fz = fsi J
32
-
on
3.8
Z i , where
evident pointed linear involution of
denotes the
J
D~+ll , S~
or
Zi
about
a hyperplane. By construction,
the map
fzCy: Z--> W 2
represents the element
q+l(W2) 9 Choose a set of generators for the image group in
Vq(X) , where
G = ker(f: F--> Vq+l(W2))
, and let
be a set of representatives of this generating set. z
in
Z
define a map
c z . S q --> X
by
c z = ~zCy .
denote the pointed CW-complex built from cells
D q+l z
--> W 1
as the pointed map extending
on
D q§
for all
X
z E Z , via the maps ~
of
X
G
Z c G
For each Let
Xq
by attaching cz .
on
~G
(q+l)-
Define
given by
~q: Xq 6zC u
.
Z
Since Let
hz:
hq: Xq•
z E G , the map
(Dq+l•
--> W 2
--> W 2
by
X q , ~ q , hq
Then
E ~ , and since F/G
is in
K c ~q(Xq)
~q(X) c N c K .
hence is in in
.
i~q
and
D q+l z •
j~q c
satisfy the necessary properties,
F/G ~ .
~q(Xq) E ~ .
Thus
~ , and
Now since
is isomorphic to a subgroup of The group
~q(Xq)
by annihilating the elements of the image Let
Define
hz .
this group,
q .
fzCy .
h , and which is given on
and it remains only to check that
~q(X)
is homotoplc to zero.
to be the pointed homotopy between
We claim that
~q+l(W2)
Z --> W 2
be an extension of
which is an extension of Xq•
fzCy
is obtained from ~Z
of
be the normal subgroup generated by The quotient ~q(X)/N
N/K
is a quotient of
is a subgroup of
~q(Xq) = ~q(X)/K E ~ .
Z
under ~Z . F/G ,
~q(Wq) , hence
This completes the proof
-
that the category construction of
J W
33
3.9
-
is filtering,
and it is clear from the
above by spanning in some cells,
has a small cofinal subcategory,
which proves
that
J
(3.4).
We want to state explicitly the construction of the cofinal index category: Corollary
(3.6):
Let
~
be a class of groups and
The following construction yields maps
X --> W
which are coflnal among maps to elements of Suppose first that fine
X
X E ~o
Choose a normal subgroup tient group is in Let
X
q
X
ically to If
W
be the limit of the
o
: De-
= X .
N c ~q(Xq_l) S c N
such that the quo-
which generates
N . q+l Ds
by spanning a (q+l)-cell
q-i
into some sphere representing Now let
X
~ , and a set
be obtained from
~o
W E ~o
is an actual CW-complex.
recursively as follows:
q
with
X E pro-~o 9
s
in X
~q(Xq_ l)
for each
's, so that
q
X
s E S .
maps canon-
W . X E pro-~ o , say
X = {Xi} , first choose an
then apply the above construction,
replacing
From this explicit description, Corollary
(3.7):
by
Xi .
we get immediately
In the notation of (3.3, 3.4),
A
Notice that moreover for in corollary
X
and
i
(x)
X E ~o
.
the maps
X --> W
described
(3.6) induce surJections on fundamental groups and
34 -
yield a pro-object of maps,
IX --> W]
(~o,pairs)
3.10
in the category of homotopy classes
such that the range
is isomorphic
[W]
to
A
X .
Let us denote this pro-object by
to extend naturally to a functor from
X --> ~ . ~o
It is easily seen
to pro-(~o,palrs)
.
Thus we obtain Corollary
(3.8):
There is a functor
pro-~ o --> pro- (~o,pairs)
carrying an
X 6 pro-~ o
d-completion
of
Remark
where
A
to an object
x->
whose range is the
X .
(3.9):
For
X E pro-~(o , X = {Xi] I E l
denotes ~-completlon,
a sense as in App.
(4.4).
and the rig~t-hand
' we have
side is given
-
w
35
-
4.1
Cohomologlcal criteria for
It is not true that a map
~-iLsqmorphlsm.
X --> Y
of objects in pro-~ o
which induces isomorphlsms on homotopy an isomorphism.
To see this, let
various coskeletons
cosknX i
~q(X) ~ ~q(Y)
X = IXi]
is itself
be in pro-~ o .
The
form a pro-object indexed by pairs
(i,n) , with the obvious canonically induced maps between them. Let us denote this functorlally determined pro-object by
(4.1)
X ~ = [ c o s ~ X i]
when
It is clear that the canonical map on all homotopy and homology.
XW :
X = [Xi] E pro-~ o .
X --> X @
But if
X
induces isomorphisms
is an actual CW-complex,
then [X ~ ,X] = llm [cosknX, X] -@ n h X--> X"
and it follows immediately that the canonical map be invertlble unless all the homotopy groups of a certain point, i.e., unless Definition
(4.2):
A map
X = cosknX f: X --> Y
a ~ - i s o m o r p h i s m if the induced map
X
cannot
vanish above
for large n . in pro-~ o
f@ : X~
--> Y @
is called is an iso-
morphism. It is clear that this is equivalent with the assertion that for every
n , the induced map
cosknX--> cosknY
is an isomorphism.
The theorem below gives cohomological criteria for a map f: X --> Y
which determine for a complete class
~
whether or
-
36
#.2
-
A
not the induced map
f
of @-completions
We will use the following terminology: twisted coefficient M E @
and a map
group
~l(X) --> Aut M
through some subgroup of ficient groups
: Vl (~) (3.7) on
M
Aut M
X
X 6 pro-~ o , a Ct-
is an abelian group
in pro-(grps) contained in
therefore correspond M , i.e.,
on
on
If
which factors @ .
Such coef-
to operations of
~l(X)
to twisted coefficient groups
M 6
~ . Let
c Vl(~)
H : ~XAlV 1 X .
M E
is a ~ - i s o m o r p h i s m .
in
Vl(X)
be a sub-pro-group.
Its inverse image
determines a covering space (2.7)
XH
of
Such a covering space will be called a ~-covering space. A
Thus
@-covering
spaces are classified by
Th=e o r e m ( 4 . 3 ) : f: X--> Y
Let
@
~v~(X) = ~l(~)
.
be a complete class of groups, and
a morphism in pro-~ o .
The following assertions are
equivalent: (i)
9: ~--> ~
is a ~-isomorphism.
(ii) Vl(X) --~> Vl(Y) , and for every d-twisted abelian coefficient group
M 6 @ ,
Hq(Y,M) ~> Hq(X,M)
(iii)
~l(X) ---> ~l (Y)
sponding ~-covering spaces abelian
for all
q .
and for every induced map of correX' --> Y'
and every (untwisted)
A 6 @ ,
Hq(Y',A) --~> Hq(X',A)
for all
q .
37
If we let
~
-
4.3
be the class of all groups, we obtain the fol-
lowing corollaries: Corollary
(4.4):
Let
f: X--> Y
be a map in pro-~ o .
The
following are equivalent: (i)
f
is a ~-isomorphism.
(ll) cos~f oos~x -> cos~Y ( ~ ) ~n(f): ~n(X)--> ~n(Y)
is an isomorphism for each n.
is an isomorphism for each
n .
The equivalence of (4.4)(i) and (ii) is trivial, and so is the implication
(1) ==> (iii).
and we will verify condition in passing to the limit: patible maps resenting
f .
It remains to show (iil) ==>
(ii) of (4.3).
We may suppose
fi: Xi -> Ti
This is an exercise
(App. 3.3) given com-
with a single index category
Consider the morphlsms of fibre sequences
K(~n(Xl),n)
> cos~+IX i
K(~n(Yi),n )
,
> cosknX i
> cos~+lY i ~ >
Let us denote it by
_>
Ki
El
> Ei ~ >
> B1
Bi
(ii),
cos~Y i
I
rep-
-
38
4.4
-
We may assume by induction that (4.3)(ii) is proved for To prove it for cient group.
coskn+ 1 , let
M
cosk n .
be a twisted abelian coeffi-
We may assume given compatible operations of
Vl(Xi) , Vl(Yi)
on
M (2.3).
Then we obtain a filtering system
of morphisms of spectral sequences indexed by
I
H p (Bi,Hq(Ki, M ) ) ---=> HP+q(Ei,M)
H p ( B I , H q ( ~ , M )) ===> HP+q(EI,M)
which converges to a morphism of limit spectral sequences whose abutment is the map
HP+qCcoskn+iY, M )
,, ,> HP+q(coskn+iX,M ) 9
Hence it suffices to show that the limit is biJectlve on Consider the index category of maps
lim H p
J
(Bj,Hq(Kj,M))
~
(ilimJ)
lim ( lim HP(Bi,Hq ) J ili'J (Kj,M) = lim J H p (B, H q (Kj,M))
lim HP (B' ,Hq(Kj,M) ) J
i --> J
in
I .
E~ q
We have
H p (Bi, Hq (Kj, M ) ) (App. l. ll(b) )
where
B = cosknY .
where
B t = cosknX
the last isomorphism being by the induction hypothesis.
Hence
6
-
39
4.5
-
we need only show
l~m H p(B',Hq(Kj,M)) ~ i ~
H p(B',Hq(K~,M))
.
But H p(B',Hq(K(r
is a functor of
r , and the two pro-objects
Vn(Xl)
and
~n(Yi)
are isomorphic, which completes the proof. Corollary Let n
(4.5):
X E pro-AGo
(Hurewlcz Theorem in the pro-category).
and suppose that
is an integer > 1 .
Vq(X) = 0
for
q < n , where
Then the canonical map
Hn(X) is biJective. For, say
X = {Xi} , and consider the fibre sequence
F i --> X i --> c o s ~ X i 9
Put
F = IFi] .
it does for F --> X
F
Since the Hurewlcz theorem holds for each as well.
Fi ,
Thus it suffices to show that the map
is a ~ -isomorphism.
Now the pro-object
cosknX
has
trivial homotopy, hence is trivial by (4.4), and thus has trivial cohomology.
Therefore for
A E (Ab) , the cohomology spectral
sequence of the above fibre sequence for the coefficient group A
collapses to yield
4.6
- 40
Since
X
for all
~> Hq(F,A)
Hq(X,A)
is simply connected and
A E (Ab)
q .
is arbitrary, we
can apply (4.3) (li). Proof of (4.3):
(li) => (i).
Suppose (ll) holds. Notice first of all that to show A A A f: X --> Y is a ~ -isomorphism is equivalent with showing that for every
W E ~o
which is of finite homotopy dimension, i.e.,
whose homotopy vanishes in high dimensions, the map
f
the composition with
induces a blJection
Ix, w] <-~ Ix, w]
For, if we let
W
have homotopy dimension < n , then
Ix,w]
[~,w] _ [cOSknX,W] ^ A
and similarly for
Y .
Therefore,
ized by its maps to objects of
since
@~o
cos~X
of homotopy dimension < n ,
this will show ^
cOSknX ~
for each
cos
n .
The problem is the following:
X
~>W,
is character-
Given a map
41 -
W
as above, to factor it through
4.[
Y , i.e., to find a commuta-
tive diagram
f
X
.> Y
(#.6)
and to show that it is unique. given a filtering category
By (App. 3.3), we may suppose
I
of diagrams in fi
Xi
(#.7) W
Since by (4.3)(ii), ~l(W)
~l(X)--~> ~I(Y) , the map from
factors uniquely through
Vl(Y ) .
~l(X)
to
Therefore we may (App.
3.3) further suppose given compatible commutative triangles for each
i ~i (Xi)
> ~i (Yi)
i(w) where of course the map
Let
J
~l(Xi) --> ~l(W)
be the category of pairs
map making the triangle
is the one induced by
(i,$)
such that
r
is a
-
42
-
~.8 fl
Xi
i
> Yi /
(4.8) W
commute a map
(up to homotopy),
(i,$) --> (i',~')
is
The following lemma will prove both the existence
and
i--> i'
in
I
where a morphism
such that
Yi t
"> Yi
(4.8') W
commutes.
the uniqueness
of (4.6), hence will complete the proof that
(li) => (i): Lemma
(4.9):
J
is filtering and is cofinal with
I .
We leave it to the reader to verify that this is sufficient. For our particular index categories,
the assertions
of (4.9)
are implied by the following two facts, as is easily seen: (4.10): such that (b) triangle
~i'
For any
i E I , there is an
factors through
Given two maps
$~'$~: Yi' -> W
i--> i'
in
I
fi': Xi' -> Yi' "
$1" ~2: Yi -> W
(4.8)', there is an
composed maps gory).
(a)
i--> i t
in
are equal
making a commutative I
such that the
(in the homotopy cate-
-
43
)4.9
-
We shall show (a) by an inductive argument on the homotopydimension of
W , and an application of obstruction theory: vanishes for
q > n .
Suppose that (#.10)
has already been demonstrated for
cos~_iW
.
Suppose
~l(W)
q
(W)
factors through
(xi) ->
Since
~l(Yi) , we may also suppose
n>2.
Thus we have a homotopy-commutative diagram,
Xi
"'
>
W
Yi
> cos~_lW
Our problem is to find a morphism a homotopy-lifting,
such that one has
as indicated:
Xi"
'
W
I.
i --> i"
>
Yi"
> coskn_iW
An elementary construction:
A continuous map
f: X --> Y
cellular ([29] page 403) if
of CW-complexes will be called
f(S~X)
c SknY
for all
One has a notion of cellularity for homotopies,
n . as well (loc.
cit.) and the well known fact (loc. cit., chap. 7, sec. 6, thm. 17) that if
f
then
is homotopic rel A to a cellular map.
f
is cellular when restricted to a subcomplex Let
A c X ,
44
#.i0
-
X I
.> yw
X
-> Y f
be homotopy-commutatlve, F: tf' N fs .
with a given homotopy-commutatlvlty,
Suppose that
f,f'
are cellular.
form the CW-complex mapping cylinders of
f,fl
Then we may to get the dia-
gram
XI
> CI fl
L
Lt
X
> C
f
where the horizontal arrows are now the natural inclusions as CW-complexes.
Using
obtain a new map
~
F , we shall describe how to modify and an actually-commutative
XI
diagram,
> C'
s[
X
~ , to
f'
~
It p
-f
Recall that the mapping cylinder
> C
C
is the fibre sum indicated
in the diagram below, in the category of CW-complexes and continuous maps, (where
Xxl
in the diagram below is given its
-
natural CW-structure
45 -
4.11
(loc. cit. p. 401, ex. i0)):
x
x "'~''tUJ > xxl
Y
>
C
J We now consider the map (i)
7: X'xl --> C
~(x',r) = (s(x'),2r-1)
(ii) T ( x ' , r ) =
jF(x',2r)
given as follows:
for
for
1/2 < r < 1 .
0
<
r
<
1/2
.
Since
X'
x{o? i~ > X tx I
ft
Y'
jt
>
is commutative, we get a unique map struction makes tion,
s
(yF)
C
tF: C' --> C
above actually-commutative.
is cellular, we might modify
retaining the actual-commutativity of lar.
If
F
(for short:
F
If, in addi-
to a cellular homotopy,
(yF) , making
tF
is a homotopy relative to the q-skeleton of
celluX'
a homotopy rel q ), the above may itself be taken
to be a homotopy relative to Remark:
which, by con-
SkqX' x l .
The homotopy-commutative diagram (~) induces no
homomorphism on relative cohomology,
-46
H*(f,G)--> but
H*(f',G)
,
certainly does:
(YF)
H* (f, g)
>
H* (f' ,G)
II
II
H*(C,X;G)
> H*(C',X';G)
Thus, a homotopy-commutatlve diagram of homotopy-commutativity
F
(Y)
together with a choice
determine a morphism on relative
cohomology.
2.
K%llln~ obstructlon classes:
Having replaced
Xi -->
W
Y•
> coskn_lW
by some actually-commutative diagram, we may apply classical obstruction theory ([29], Ch. VIII) to the fibre bundle
W-->
coskn_lW , whose fibre is isomorphic to
The obstruc-
K(~n(W),n) .
tion to the existence of the homotopy-lifting,
47 -
Xi
W is a class
>
~.>
Yi
cos~_iW
~(i) 6 En+l(Yi,Xi;]--~)
local system of coefficients, is a ~-twisted
%. 13
where
{Vn(W)} .
~-~
is the induced
By our assumptions,
system of ~-groups.
Consider the long exact sequence,
~O:i,T- D _. ~(xi,]-71 ~ ~+l(q,xi~T- D ~ ~§ By hypothesis, i -, i'
En+l(Y,~)
-, E n + l ( x , ~ )
such that the image of
But the "elementary
construction"
So there is a morphism
.
r~(i)
_. ~+l(xl,T-D,
vanishes
in
Bn+l(Yi,
allows us to represent
the choice of a homotopy-commutativity
m , after
F , by an actually commuta-
tive diagram Xi ,
> Yi i (u
Xi --
> Yi
inducing a morphism on relative cohomology. ~(i' ) , and therefore,
~(i')
Hn(Xi , , ~ )
5x' = ~(i' ) .
such that
such that the image of y" 6 Hn(Yi,,,]- ~)
.
= 0 .
x'
be
construction"
another actually-commutative
diagram,
a
since
yF~(i) =
class in
Find a morphism
x' , Mr' E Hn(Xi.,, ~ )
This is possible,
Use the "elementary
Let
We have that
i' -~ i"
comes from some Hncy,~-I)
to represent (7p,) 9
~ Hn(x,V-~)
i' -r i"
by
.
48
By construction, the obstruction class
~(1)
goes to zero
under the map, ->
9
Thus we obtain the required extension,
Xi,, ~ m >
Yi"
w
oo.,
_lw
The proof of (b) is similar, and we leave it to the reader.
(ill) --> (ll): Assume that (ill) holds, and let abelian coefficient group on
X
M 6 ~
and
be a ~-twlsted
Y , given by a map
~l(X) ~ ~l (Y) -> Aut M . We can suppose (App. 3.3) given a filtering system of maps fl
xl in
~o
representing
> Yi
f , and a compatible sequence of maps
7rl(Xi) --> ~l (Yi) -> Aut M representing the operation of arrow is the map induced by
~I(Y) fl "
denote the kernel and image of and let
X~--> X i , Y~--> Yi
and ~ i
respectively.
on
Let
~l(Xl)
Aut M , where the first Hi , Gi
(resp. ~&i '
(resp. ~l(Yi))
in
~i )
Aut M ,
be the covering spaces induced by The
X~
(resp. Y~)
form a pro-object
49 X'
4.15
(resp. Y') which is the covering space of
termined by the kernel of the map
(resp. Y) de-
X
~l(X) --> Aut M
(resp. ~l (Y)
--> Aut M). Using the covering resolution of
M
X!l
of
X i , one constructs a canonical
in the well known way:
If
X!l -> Xi
sented by an actual covering space of CW-complexes, tion is the one obtained from the simplicial
xi <
x,
(xl3• Xi
on the various fibre powers of
X i' 9
the resolu-
object
<
I)
by taking the direct images on
is repre-
<
@
@
9
of the coefficient
groups
M
Let us denote this resolu-
tion by 0--> M--> L~--> Li1 --> ...
V
Each on
Li
is the direct image of an untwisted
coefficient group
X[ , namely V
L~ v ~ M Gi
Gv
where
M
is untwisted
product of
M
on
X! l
with itself,
by construction,
and
indexed by elements of
M i
is the
V
G i = Gix...xG i.
Denote by
o
l_>
0--> M--> K i --> K i
the resolution of
M
from the covering
Y~ , so that
coefficient
group
on
Yi
...
obtained in the corresponding Ki
way
is the direct image of the
-
50-
4.16 ~V
K[ v = M i
By functorality,
there is a filtering inductive system of
morphisms of spectral sequences relating the spectral sequences of these two resolutions:
H p(Hq(Yi,K i) ) ====> HP+q(yi,M )
H p (Hq(XI,L [) ) ---~=--> HP+q(xi,M)
hence a morphism of the limit spectral sequences, whose abutment is just the map
HP+q(y,M)--> HP+q(X,M)
We are to show that this map is biJective. to show that for each
(*)
is biJective.
q
and
.
Therefore it suffices
v , the map
V
lim Hq(Yi,K.Vm) --> lira Hq(Xi, L i) -@
.-~
i
i
Now we have
and
qcxi,
qcxb Iv)
-
51
-
4.17
Consider the index category of maps
i --> J
in
I .
Clearly
we have isomorphisms
lim_, H q (Y~, K~ v )
J V
lim_, H q
(q, K~v)
(where
K I. = M ~j) J
i-,j V
= lira (lira j
(App. 1.11)
i~j V
lim (Hq (Y', M J))
J V
Now (4.3)(iii) asserts that
V
Hq(y',M ~j) ~ Hq(x',M ~j) .
Hence we
are reduced for (*) to showing that v
(**)
Gv
lim H q(X',M ~ j) --~> lira H q(X',M J) . -@ -@
j But
{~j} , {G j}
J are isomorphic as pro-groups,
since they repre-
A
sent the images of the corresponding maps Vl (~) -> Ant M a functor of
and since
v l(~)
v l(~)
~I(Y) -> Aut M .
Since
and
Hq(x ',M Fv
)
is
F , (*) follows.
(i) => (iii): This implication follows from the following result: Theorem (4.11): A
X 6 pro-~ o 9
Let
Let
~
be a complete class and let
A
H c ~I(X)
be a sub-pro-group and let
H c ~l(X)
A
be its inverse image. space
XH
of
X
Then t h e O ~ c o m p l e t i o n
determined by
H
Xtt
is canonically
of the covering ~-isomorphic to
52 -
the covering
A X~
4.18
A X .
of
To obtain the implication
(i)=>
(iii) from (#.Ii), we may
apply the theorem to the corresponding C-covering spaces X',Y' A A of X,Y respectively 9 Since X ~ Y , it implies that also A A X w = Y' Hence we are reduced to the case X = X' i e. to showing that if
f
is a
~-isomorphism,
Hq(Y,A) --~ Hq(X,A)
then
for
A E ~
.
But Hq(Y,A)
[Y,K(A,q)] =
[~,K(A,q) ]
=
[X,K(A,q)]
=
Hq (X,A)
since
K(A,q)
E C,~o
.
This completes the proof of (4.3). Proof of (4.11):
Of course, the map
A A X H -->
is induced
^
by the canonical map want to show that
~A fl
--> ~
since
E pro-~ 0 9
is the ~-completion of
~
.
Thus we
We may
(App. 3.3) represent the given data by a compatible filtering family of sub-pro-groups
where
X = {Xi] , X i E ~o " A is the inverse image of H i
Since
(2.8)
XH=
[(Xi)Hi]
Let in
Hi
be the sub-pro-group which
~l(Xi)
and (3.9)
so that
H = [Hi] .
[(Xi)Hi] = X H , it follows
- 53
-
4.19
that it is enough to show that for each
N
(xi)H i Hence we may replace Write
X
by
Vl(~) = [~j]
)A
> (xi H i
X i , i.e., suppose
X
is in
(~j E ~) as a pro-OFgroup,
are quotient groups of H
^
i
~o "
where the
Vl(X) , in such a way that (App. 5.2)
is represented by a filtering family of subgroups
^ Then if #~j
denotes the inverse image of ~ j
A Vl(X),- , it is
in
clear that
^
Thus, applying
(2.8),
(3.9) again, we may replace
H
by
j
,
i.e., assume that (4.12).
H
is a subgroup of
image of a subgroup
~
Recall (3.4) that I
of objects
W
~l(X) , and is the inverse
of some quotient group A X
#
of
Vl(X)
may be regarded as an inverse system
each of which is obtained from
X
by spanning
in cells in successive dimensions so that the resulting homotopy groups are all in
~ .
Thus (2.8)
~
will be obtained as an
inverse system of the corresponding covering spaces
W'
indexed by
so that the
map
I) of these
Vl(X) --> ~
W , if
factors through
W
is "large enough" Vl(W ) .
(again
This covering space
W'
-
is obtained from in
@~o "
X' = X H
54
-
%. 20
by spanning in some cells, and it is
Therefore the theorem in our case is equivalent with
the assertion that the category of such cofinal with the category of all maps
W'
(indexed by
X' --> Y
such that
Thus we need to verify App.(1.5) in this context.
I) is Y 6@~/0 9
The conditions
read:
(a) a map
Given a map
X--> W , W
X t --> W t
with
Y E @~o ' there is
factors through
as above, such that
.
(b) maps
X' --'> Y
Given a map
V' _i> _> Y
with
as above, and a pair of
X ~> V , u 6 ~ o Y 6 ~o
such that
ir
~ jr
(where
J Ct:
_> V'), there exists a diagram
X t
X
> W W
as above
V
such that
ik' ~ jk I .
To prove
(a), we construct
W
as the limit of a compatible
sequence of CW-complexes
Xq , where
~n(Xq) E ~
for
Given
struct
by spanning in some (q+l)-cells in such a way that the
map Xq_ I'
Xq
' I --> Y Xq_
n < q .
extends to
Xq' .
by spanning in (q+l)-cells,
the map
t I) --> Y Vq(Xq_
~
factors through
X'q
and
Xq_ 1 , the problem is to con-
Since
X'q
is also obtained from
the condition is clearly that
factor through
~q(X~)
.
Let
N c ~ q ( Xq _I l )
55 -
4.21
be the kernel of this map. We treat first the case K c H c ~l(X.) Then
K c N
be the kernel of the map
is a normal subgroup of
an injection
K/K n N c H/NO-> ~l(y)
Therefore by (B.2)(i), ~l(X)
and with
Now suppose operates on
q > 1 .
is isomorphic
in
.
, hence
We can let
Then Let
~q(Xq_l)
above.
~l(X) --> K .
X1
is a @-group,
be the intersection
(~q/N)
normal in
be obtained by
under this action.
to a subgroup of
F
F .
~l(Xq_l)
F
We have
K/K G N E @ 9
contains a subgroup
E @ 9
Let
Xo = X .
H , hence of
to kill the subgroup
~q(Xq_l) N
K 0 N
~l(X)/F
spanning in 2-cells
jugates of
q = 1 , setting
and it
of the con-
Then
~q(Xq_l)/F
, hence is a @-group.
Suppose we span a (q+l)-cell into
Xq_ 1
sents the element
The homotopy classes of
s E ~q(Xq_ l) .
~q(Xq_ l) = ~q( X'q_l ) lying over
S
in
are all translates
repr@sented
Xq_ 1 of
whose boundary
by the various
Consider
F
in
s
X
under the action of
X' q-i
~l "
Therefore if
Thus we may span in
q-1 "
(4.13)(b).
closely that of (a).
spheres of
repre-
(by choosing a path to the base point)
s E F , then so are all these other classes. cells to kill
S
The proof of this assertion follows
We again construct
W
as a limit of a se-
quence of factorizations
X
,> X
W
such that
~n(Xq)
E @
for
n < q , where
Xo = X
and
Xq
is
56 -
obtained from
X
4.22
by spanning in (q+l)-cells, and where the
q-1
homotopy
hq-l: ir
|
~j
!
r
extends to a homotopy
hq: i ~
• J~
.
Following the notation of (3.4), replace N
be the kernel of the map
r
X
ms: D q+l --> X
S c N
~: F--> N onto
N .
be a pointed continuous map representing
and extend
Cms
tion of
through
~
Xq_ 1 , let
~q(X) --> 7rq(V) , and let
be a map of a free group on a set of generators Let
by
to a map
Bs: D q+l --> V .
~(s),
We obtain a factoriza-
corresponding to
X'
D q+l = X s . The covering X~ of X s s is obtained by spanning cells into each of
the spheres
of
X'
a'
SV
X Um
lying over
represents the choice of a point Choose for each For each
v
a path from
m s , where the index
Pv
above the base point of
i(pv)
to the base point of
v , we get a map
f
SV
: Z-->Y
as in construction 1 of (3.4), replacing respectively,
v
m s , 8s
by
m~v, 6~v
and, using the chosen path above, the map
determines a class tending the homotopy
fsv
in
~q+l(Y)
h
to
X s'
("" ,?sv"'" )
.
fsv
The obstruction to ex-
is represented by the vector
( q+l(Y))P
X . Y .
57 where
P = Iv]
fs: F-->
is the index set.
4.23 Thus we obtain a homomorphism
(~q+l(Y)) P .
The important thing is that since and
~q+l(Y)
6 @ , the group
(Vq+l(Y)) P
Therefore if we replace the group (3.4) by
Example:
(a)
Vq(X)
-isomorphic m>
is again a @-group.
Vq+l(W2)
of the proof of (#.13)(b)
the proof of (4.11).
Let
vanish for to
Hence
(Uniqueness of Eilenberg-Maclane
Corollary #.i#:
if
is a @-covering,
(~q+l(Y)) P , the same argument applies.
holds, which completes
groups
X' --> X
K(v,n)
X
be in pro-~ o
q ~ n , and ; cos~X
spaces)
and suppose the pro-
~n(X) = v .
Then
is isomorphic with
X
is
K(~,n) ,
n . I
Proof: and that
An application of (4.4) gives that
coskn_lX
is contractible.
X ~ cos~X
,
(4.1#) then follows from
the fibre triple: coskn_lX-->
(b)
cosknX--> K(Tr,n)
.
(Uniqueness of the n-sphere)
Corollary #.15:
Suppose
@
is the class of finite groups
whose orders have prime divisors in a fixed set Suppose if
X
P
of primes.
is in p r o - ~ o , is simply connected, and H (X,Z) = 0 A q q ~ n , and Hq(X,Z) = Z if q = n for some fixed n > 2 .
Then there is a ~ -isomorphism, A Sn
>
X
-
Proof:
Let
58
-
4.24
~
denote a topological generator of the proA jectlve limit of the pro-abelian group Z . An application of (4.4) enables us to represent each
Xi
is
n-1
X
by a system
[Xi]
such that
connected, and for which there is a compat-
ible system
Hn(Xi'Z) which represents the element
6 Lim Hn(Xi, Z ) = Lim(~) (-. (.-
We have that
by the ordinary Hurewlcz theo-
H n(xi,Z) ~ ~n(Xi )
rein, and hence the
may be interpreted as producing
T i E ~n(Xi )
a morphlsm Sn
in pro-~ o . mology groups
~>X
.
This morphism is clearly an isomorphism on all hoHq
for
We shall show that
q ~
.
Hn(Sn) ~
Hn(X)
is an isomorphism as
well, for if we evaluate, using our assumption that we find that
Sn
is a morphlsm
such that Lira #n" Lim (--
(--
--> Lira (-.
Hn(X ) ~ ~ ,
-
is an isomorphism.
59-
This latter fact implies that
an isomorphism, and (4.15) follows from (4.3, iii).
#.25
~n
itself is
60
-
w
5.1
-
Completlons~and fibrations.
Throughout this section, we let
@
denote a class (3.1)
consisting of finite groups. Let
X E pro-~ , and consider the pro-object pair
described in (3.8).
A X-->X
We can replace this object by a canonically
isomorphic fibre resolution, as in (1.1).
Let us write it the
same way. Proposition (5.1): erty that
~
Proof: X --> W
The fibre
is ~-contractible The inverse system
A X --> X
of
X
has the prop-
(4.2). A X--> X
is represented by maps
which induce isomorphisms of fundamental groups (3.8) and
thus each object in the inverse system of X is connected. DeA note by X the universal covering of , and by ~ the induced covering of tion
X
that of
of
X .
By (#.ll),
~
is @-isomorphic with the comple-
N
~
X^, and the fibre of
X--> X
X--> X .
A
is ~-isomorphic with simply connected, and
Thus we may suppose
in fact that the inverse system comprising
consists of simply
connected objects. We need to show that cohomology groups
Vl(~) = 0
and that for
A E ~
the
vanish for all q > 0 . Let us write A A X --> X = IXi --> Xi] i , and x = {xi} 9 We
Hq(X,A)
the inverse system as
obtain a directed system of spectral sequences
(5.2)
(EPq)i = H p(~i,Hq(Xi,A)) ==> HP+q(Xi,A)
where the coefficient groups
Hq(XI,A)
are untwisted since
A Xi
-
is simply connected. A Vq(X i) E ~ for all
Now q .
6~
-
5.2
~
is a class of finite groups, and A Hence X i is homotopic to a CW-complex
which is finite in each dimension. Therefore the cohomology of A X i commutes with direct limits of coefficient groups. Passing to the limit of the
E~ q
term of (5.2) we obtain
A lira HP(xi,Hq(xi,A)) i
A Hq = lim HP(xi, (xj,A)) = i-~j
A
A
lira H p (X, Hq (xj,A))
= ~P(x, lim ~q(xj,A))
=
J
J
HP(I,~q(x,A)) .
Therefore the limit of (5.2) is a spectral sequence
E~q = HP(~,Hq(x,A)) ==> HP+q(X,A) .
Since the coefficient groups are u~twisted, Hq(X,A)
.
we have
E~q=
But by (4.3), the map
~,~o= ~P(I,A)-> ~P(x,A) is an isomorphism for all and hence
E pq = 0
for
p , which implies q > 0
It remains to show that
by induction on ~l(~) = 0 .
q .
Consider
sequence
~2 (~) -> ~i (•
E~ q = 0 , q > 0 ,
-> ~i (x) -> o
the exact
62
Let
G
be a @-group,
and suppose given a (say) surJectlve map
(*)
~i (•
Sinoe
5.3
-
-> G .
=1(x) = ~l(~) = o , and since
~l(X)
maps to the group
A
0/~(~2(~)) G
, it follows that
is abelian and the map
HI(7~G)
=2(~)
maps onto
a.
Therefore
(*) represents an element of
Hom(x,G) =
which was shown above to be zero.
Lemma
(5.3):
A pro-object
of maps of CW-complexes
X --> Y
in the homotopy
category
gives rise in a functorial way to a pro-
object X
> Y
A
.>~
(5.4) x
in the category of squares,
where the vertical arrows are isomor-
phic to the ones introduced
in (3.4).
Proof: X c y
It clearly suffices
of CW-complexes.
to treat the case of a pair
We let the pro-object
the category of diagrams
(5.5)
X
.> Y
it
.>V
U
(5.4) be given b y
63 -
where
U,V E @ ~
, U
5.4
is obtained from
X
b y spanning cells in
successive dimensions as in (3.4), and
V
is obtained from
b y spanning in cells including at least those spanned into to obtain fixed,
U .
A morphism is a map of squares leaving
Y X
X --> Y
taken up to homotopy.
It is clear that the left vertical A arrow yields a pro-object cofinal with X--> X as in (3.4), and that moreover any map
Y--> W , W E @
factors through a
V
corn-
ing from a square (5.5). To show the right arrow cofinal with A Y--> Y and that the resulting pro-object (5.4) is functorlal, one reduces immediately to proving that given a diagram of ~'~complexes > X
Y
and a homotopy
h
> U
-
W
> V
> > W'
between the two squares
X
.>Y
W
.> W w
L
there is a square X
.>Y
U'
. .> V I
(5.6)
64
-
dominating
(5.5) such that
h
To do this, we first choose from
Y
to
extends to a homotopy on
V1
U' --> ~'
so that the homotopy extends
V 1 , as in (3.4).
Then we choose any square
X
.> Y
L I
(5.7)
ui
mapping to (5.5). the case that X--> U
h
so that
a diagram h
> vI
Replacing
(5.5) to (5.7), we are reduced to
extends to h
V .
extends to
add the cells spanned into
that
5.5
-
Choose
X--> U'
U' , as in (3.4).
X
to obtain
U'
to
mapping to Then we can V , obtaining
(5.6) mapping to (5.5); and it is clear by construction extends.
Let
X--> Y E prO(~pair s) be as in the above discussion.
we fibre resolve the square
(5.4), we obtain a pro-object
If
of dia-
grams F
.>X
F
> X
->Y
(5.8) A
where
F
and
Theorem
1~
are the fibres of the horizontal maps.
(5.9):
With the above notation,
is a pro-object such that each A F--> ~ is a ~ - i s o m o r p h i s m !
A
->Y
Yi
suppose
Y = {Yi ]
is simply connected.
Then
.
65 Proof:
5.6
Applying (1.2), we may imbed the square (5.8) in a
projective system in the homotopy category of diagrams
(5. lO )
F
.>X
>Y
I [ I where the vertical and horizontal arrows are flbratlons, up to homotopy.
Say we index the diagrams making up (5.10) by
i E I .
Now if we recall the construction (3.4) of a cofinal system of maps
Y --> W
(W E @ ~ ) by spanning in cells, it is clear that
for a simply connected
Y
we need only consider those
W
tained by spanning in cells of dimension ~ 3 , and then ~2(W)
is surJective.
=i(,)
We want to show that
~
= o .
is ~-acyclic, and since
sists of finite groups, it suffices to treat A ~ Z/p 6 @ , p
a prime.
@
con-
twisted coeffi-
Consider the inductive
system of spectral sequences
HP(@i,Hq(~i,A ) ---> HP+q(Xi,A ) .
We have
~2(Y) -->
Therefore in the diagram (5.10), we have
(5. ii )
cient groups
ob-
66
-
5.7
lim E pq = lim HP(r i = lim HP(r
Since
r
.
is simply connected, we can take the second limit over
those maps
i --> J
which induce the zero map
Then the induced action on
Hq(r
Vl(r
2Z/p .
Moreover,
cohomology with values in
r
Hq(r
homology commutes with products 9
o->
is trivial.
we want 9
are vector
(For, its co-
is trivial,
since co-
In the exact sequence
v->
o
(2~/p)s , hence induction on
q
shows what
Thus we conclude
lira E pq = o i This limit is the HP+q(X,A)
2Z/p
9
is again a sum of
Hq(~j,A)
is Cracyclic by (5.1), hence its
homology with values in a product of
V
.
A ~2Z/p , the coefficient groups
spaces over
E pq
if
p > o
term of a spectral sequence abutting to
which is zero for
.
is trivial, and so
lim E pq = lira H p(r J Since
--> ~l(r
p + q > 0
by (5.1).
lira E~ q = lira Hq(~j,A) = Hq(~,A) = O , i j
Therefore
q>O,
-
67
-
5.8
as well, as was to be shown. Now let twisted.
A E @ be a coefficient group on F , possibly A A Since X , Y are in pro-@, they are projective limits of
CW-complexes homotopic to ones finite in each dimension, and hence the same is true for
F .
Thus cohomology on
F
commutes with
direct limits, and the inductive system of spectral sequences
EPq = HP(Fi,Hq(~i,A)) ==> HP+q(Fi,A) has as limit
llm E2Pq = i
llm j
HP~,Hq(~j,A))
-
HP~,Hq(~,A))
= 0
if
q>0
.
Therefore the spectral sequence collapses and yields isomorphisms
H p (~,A) ---->H p (F,A)
for all
p .
To complete the proof that suffices by (4.3) to show that diagram
A F
and
~
are ~ - i s o m o r p h i c ,
~l~F) ---->~ i ~ ) .
Consider the
it
- 68-
5.9
> 0
7r2 (~)
Since
Y
> ~ra(~ )
l
,L
0
0
is simply connected,
exact functor
~l(X) = 0 , and
A
is a right
(meaning that it commutes with direct limits,
it has the appropriate
the rows and columns A of this diagram are exact, in an obvious sense. Since Y , Y are simply connected,
adJointness
since
property),
we have
H2(Y,A) = Hom(~r2(Y),A ) = Hom(Tr2('~),A ) = 1-12(~,A)
for
A E C , and so the map
d
b
is the zero map, and since
e
is an isomorphism,
a pointed,
fl .
a
we apply
We revert from
connected,
resolve the inclusion
is surJective,
as was to be shown.
To conclude this section, space functor
of (5.12) is an isomorphism.
~o
c
is zero.
Tbls completes
Thus
the proof.
(5.9) to study the loopto
~o " and suppose
simply connected simpliclal
set.
e --> K
If we fibre-
e --> K , we obtain a fibre triple: --> e' --> K
Hence
- 69
-
5.10
which is actually functorlal for pointed maps,
C~,I
and if pret
> ei~
f -- f' , we obtain ~
Of ~ Of'
(1.1) and thus we may inter-
as a functor from the connected
subcategory of
Ko
Corollary 5.13: pointed
.> K 1
to
Ko " If
e--> K
slmplicial set, then A
is a ~ - i s o m o r p h i s m .
simply connected full
is a connected,
simply connected
-
w
Homotopy
groups
In this section, h o m o t o p y groups for a complete
7 0
of completions.
we discuss
of an
the relationships
X E pro-~ o
class
6.1
-
~ .
between
the A X
and of its ~ - c o m p l e t i o n
Since
X
maps to
~ , we are given
a map
(6.1)
~n (X) --> ~n
for each n = 1 .
n , and we saw in More generally,
Corollary
(6.2) :
)
(3.7) that it is an isomorphism
when
we have
Suppose
that
~q(X) = 0
for
l(qKn.
Then
v n (X) --~> ~n (~)
This is analogous
to (4.5).
If say
X = ~X i] , let
the fibre of the map
~X i --~ coskn_iX i] .
(ill)) that
is a
X
by
X' --~ X
X' , i.e., we m a y suppose
and for each
i .
X--~ W
mension
where
at least
In addition,
that
It follows
W
(S.6) of
{X~]
from
(4.4
~q(Xi)
= 0
for
to treat the case ^ X--~ X , it suffices
i 4_ q ~ n X = Xi . to take
is obtained by spanning in cells of di-
(n+l) , from which the assertion one can describe
~2
in complete
is clear. generality
as follows : Proposition be abelian.
Let
(6-3): J
Let
X = {X i ] E
be
and so we may replace
Hence it suffices
Then in the construction maps
~-isomorphism
X' -
p r o - ~ o , and let
be the index category of pairs
J =
A E (i,H)
-
where
H c Vl(Xi)
Let
(Xi)H
71
-
6.2
is a normal subgroup such that
be the covering space of
Xi
Vl(Xi)/H 6 @ 9
determined by
H .
Then the groups
H 2 ( (XI)H,A)
form a directed system indexed by
J 9 and
A 9r2(X).. is characterized
by the property that
(6.4)
~o~(~ 2 (1),A) ~ lira H 2 ( (Xi)H,A) . J A
Proof:
Let
A
~
be the universal covering of
be the induced covering space of
X .
X , and let X A ~A Then by (4.11), X ~ X . We
have
li~ ~ ((xI)H,A) = ~ (~,A) J
and
~2(~)=
may assume
~2(~) . ~l(X)
Hence we may replace
X
by
has no non-zero maps to Crgroups.
9
i.e.,
we
Then (6.4)
is just the assertion that
He(~ 9
~> H2(X,A) 9
which follows from (4.3). Following Serre Definition pro-(groups). ~-module
[27]9
(6.5): G
A 6 ~ ,
Let
we introduce the notion of good group: ~
be a complete class9 and
G E
is said to be ~-good if for every twisted abelian
6.3
-72 Hq(~,A) --~> Hq(G,A)
where
~
is the d~-completion of
Corollary (6.6): G = [G i} completion of
for all
G . is d~-good if and only if the C~-
K(G,1) = [K(Gi,1)]
is ~ - i s o m o r p h i c
For, the completion of the map -isomorphism iff. Applying
to
K(G,1) --> K(~,l)
K(~,l)
.
is a
is C~-good, by (4.3).
(4.4), we can now prove
Thieorem (6.7): X 6 pro-~#o
G
q ,
Let
~
be a class of finite groups, and let
be simply connected.
Then
A ~q(X) ---->~q(~)
if
q (x)
is C-good for
Proof:
Suppose
actually
X = IXi]
~q(X)
~q(Xl) = 0
has for
descending induction on when
q < n
q < n-1 . is ~-good for
(4.3) to the fibre of the map case that
for
q ~ n-1 .
Applying
X --~ cosk2X , one reduces to the
~l(Xi) = 0 q < r
for all
and all
i .
i .
Suppose that
We proceed by
r > 2 , the theorem being true by (6.2)
r > n-1 : Let
(6.8)
G = ~r(X) , and consider the fibratlon
F--> X--> K(G,r)
.
By (5.9), the ~-completion of (6.8) is ~ - i s o m o r p h i c sequence,
and we may suppose the theorem true for
F
with a fibre by induction.
-
73
6.#
-
Thus we are done by the exact homotopy
sequence arising from the
completion of (6.8) if we prove the following: Proposition r > 1
(6.9):
an integer.
K(G,r)--> K(~,r) Proof: apply
(6.6),
@
E
r = i , this is
is contractible, (6.10):
0
is C~good iff the map
(6.1).
For
r > i , we may
is @-good.
E--> K(G,r)
and induction.
Let
sequence of pro-abelian A"
G
(5.9) to the fibration
Corollary
Suppose
group
~-isomorphism.
K(G,r-I)-->
where
be a class of finite groups and
A pro-abelian is a
For
Let
0--> A' --> A--> A" --> 0
groups,
and
~
be an exact
a class of finite groups.
Then the sequence
0--> ~'--> ~--> ~"-->
is exact. This follows from the above proposition,
and
(5.9), applied
to the fibration
K(A',2) --> K(A,2) --> K(A",2)
Example nected
(6.11):
X , even when
Theorem n = 2 .
of a 2-sphere and a 1-sphere, finite groups.
Then .
.
(6.7) is false for non-simply For example, and let
~l(X) = 2Z , and
~
let
~
corresponding
covering space is a wedge of
be a wedge
be the class of all
~2(X)
of copies of
X
con-
is a countable
For any finite quotient group of S1
~l(X)
sum
, the
and finitely many
74
S2 .
We apply (6.3).
-
6.5
It is immediately seen that while
H~
= ~-T A
the group Hom(Tr2 (~), A)
is the subgroup of this group consisting of vectors
(...,a o,a I,...)
ai E A
which are periodic, with some period. Example (6.12):
Take
@
to be the category of finite groups.
Then the additive group of rational numbers, K(Q,n)
is ~-contractible.
o->
to obtain that n > 2
K(Q/Z,n)
z->
Q , is e-good.
Thus
Apply (5.9) to
Q->
%/z-> o
is ~-isomorphic with
A K(Z,n+I) , if
.
An application of (5.13) gives the above for
n = i as well.
-
w
7 5
7.1
-
Stable results.
In this section we study the question of classification, up to stable homotopy equivalence,
of all pointed CW-complexes of a
fixed stable pro-finite homotopy type. to a purely algebraic one (Cot.
This problem is reduced
(7.15), below).
It is shown
that there are only a finite number of distinct stable CW-complexes having a given stable pro-finite completion. If
A,B
are pointed CW-complexes,
erated reduced suspension. integers
nj
let
SnA
denote nt---hit-
For any integer
(j = 1,2) , two maps,
k , and nonnegative nj+k fj: snj(A)--> S (B) ,
(j = 1,2) , will be referred to as "equivalent maps of degree k" n-n 1 if there is an integer n such that S (fl) is homotoplc to n-n e S (f2) . The set of equivalence classes of maps of degree k forms an abelian group which we denote
(A,B)k .
topy category we will mean the category pointed CW-complexes, the set
and morphisms from
(A,B) = (A,B)o .
By stable homo-
whose objects are A
to
B
are given by
As we have defined it, the stable ho-
motopy category is an additive category.
Although there is a
standard technique for enlarging this category slightly to a triangulated category,
(in the terminology of Verdier,
we shall not bother to do this.
cf. [16] )
(For general treatments of
stable theory, see [1], [24].) If
A
is a subcomplex of
X , and if we denote by
CW-complex obtained by constructing the cone over
A
in
obtain a triple (A) A --> X --> X/A where the arrows denote the natural inclusions.
Anything
X/A X
the 3 we
-76
isomorphic
tion.
to
such a triple
-
7.2
(A) w i l l
be
referred
to
as
a cofibra-
The cofibrations play the role of the "triangles" of our
category. We have the standard facts that any continuous map of CWcomplexes
f: X --> Y
may be continued to a cofibration,
Y->
x->
z
f and this cofibration is unique dependent only upon canonical) isomorphism. (V,)
f
Also, functors of the form
up to (non(,W)
and
applied to cofibrations yield three-term exact sequences
of groups, etc. In this section let
~n(A)
denote the functor
(sn,A) , the
stable homotopy group functor. L~mma (~3): If (A,B)k
A,B
are finite CW-complexes,
are finitely generated abelia~ groups for all
sequently so are Proof:
Well known, a~d easy.
as the ordinary homotopy groups of choice of
L .
k .
Con-
~n(B) . Here is one way of seeing it:
By the Freudenthal suspension theorem,
complex.
then the groups
But this
S~
S~
~n(B)
may be interpreted
for a sufficiently large
is a simply-connected finite ~ -
It therefore has finitely generated integral homology
groups, and the relative Hurewicz theorem applies, assuring that it has finitely generated homotopy groups. relating
Hm(A,~n(B))
lemma (7. i).
to
(A,B)k
The spectral sequence
allows one to then deduce
-
77
7.3
-
What is the relationship between suspension and completion? Let
A
denote completion with respect to a class
~ .
Let
be the pointed homotopy category of CW-complexes.
Let
E,A: ~o
--> ~o
be functors,
Z
left adJoint to
and suppose,
A
~o
further,
that both functors preserve the subcategory A
Lemma (7.2) : The natural map Proof:
EX --~ ~(~)
is an isomorphism. , form the com-
To obtain a two-sided inverse to
posite A
where the first map comes from adjointness
of
A
and
E .
Since
A
A(EX)
is in p r o - ~ o
--~ A ( ~ )
the above composite extends to a map,
which yields
(by adJolntness
of
Z,A
in prO-~o),
a
^
map
~: ~
--> ZX .
One checks that
a
and
~
are two-sided
inverses of one another. Corollary (7.3) : If completion, (Sk
A
denotes p-completion or profinite
the natural map
a: skx--> sk(~)
is an isomorphism.
denotes k-fold reduced suspension. ) From now on, unless explicitly indicated,
A
will denote
profinite completion. Corollary (7.4): Let --> ~n(~)
is pro-finite
Proof: skx
for some
Note that k
are ~-good where
X
be a finite CW-complex.
therefore use theorem
n(X)
completion. ~n(X)
is the ordinary homotopy groups of
(with a shift in dimension), ~
Then
and these groups
is the class of all finite groups. (6.7).
We may
78
-
7.~
y =
Consider two objects of p r o - ~ o , X = (Xl)iE I , and (Yj) j 6J "
Define the pro-abellan group
indexed over
(x,y) k - u m.(xi,Yj) k
J E J .
i Notice that
Lim(X,X) = Hom(X,Y) e-
J
the latter group being
Hom
Corollary (7.5): Let
in the category p r o - ~ o .
A,B
be finite CW-complexes.
Then the
natural homomorphism of abelian groups
(A,B) --> (~,B)
is proflnite completion. Proof:
Suppose
A 1 --> ~
--> A 3
is a cofibratlon.
Then the
Puppe sequence gives us a morphism of long exact sequences.
...-->
(AB,B)k-->
^
...-> for all
(A2,B)k-->
A
(A1,B) k --> (A3,B)k_ 1 --> ...
L?
(a3,B)k-> (a~,B)k--> (al,B)k--> (a3,B) k - 1 -> k 6 Z .
"'"
The top llne consists in finitely generated
abelian groups, and the bottom line in pro-abelian groups.
Note
that if we take the proflnlte completion of the top llne we obtain
7,5
-79-
a long exact sequence in pro-Ab . phisms for all
k .
Suppose
k .
This observation,
quance of cofibratlons us Cor.(7.5).
q.e.d.
(a)
and
~3
is an isomorphism for
coupled with Cor.(7.4)and
that any finite CW-complex
1A 1
were Isomor-
Then applying the five-lemma in the abelian
category pro-ab, we would obtain that all
~,~
A
may be built up from a finite se-
0A 1 --~ JA2--~ 0A 3 Note:
JA 3
the fact
for
J = 1,...,r
gives
"built up" means the following.
are spheres of various dimensions,
J = 1,...,r .
(b)
J+lA 1 =
(C)
rA2
--
for
J = l,...,r-i .
A.
A model for the profinite
Suppose Thus map
A = SB , where
A = B•215
U B•
fn: I/~I --~ I/3I
.
tlng
x E R/Z . fn(B,x)-
that this functor
fn
[A, ]
B
is a pointed finite CW-complex.
Consider,
for any integer
which is multiplication by
after the identification all
completion of a double suspension.
of
I/3I
with
for all
B E B
fn: A--~ A
and
x E I .
for by setNote
represents the n-th power map of the group-valued into itself.
Suppose now that
A = S2B , so that
[A, ]
values in the category of abelian groups. cofibratlons
That is,
R/Z , fn(X) -- nx
This map gives rise to a map (6,fn(X))
n .
n , the
of finite complexes,
we obtain
A-A~A/n fn ~
actually takes
Continuing the
fn
to
- 80 Since
we obtain compatible maps,
finn = fm~
fmn
A
A
n
> A
> A/ran
> A
> A/n
Thus we may interpret ~o "
7.6
A = (A/n)n6J
Applying the homology functor to
as a pro-object of (~)
finite homology groups in each dimension. connected, theorem, Thus
A If
so is A/n
A/n .
note that
Since
A
A/n
has
is simply
Consequently by the relative Hurewicz
has finite homotopy groups in each dimension.
is actually in pro-C~o, where @ is the class of finite groups. W
is in d~ o, then
[A,W]
the group of homotopy classes
of continuous maps is a finite abelian group.
Abelian since
A
is a double suspension; finite by induction on the number of cells of
B . Proposition (7.6): A--> ~ Proof:
map
is profinite completion.
We need only show for any
[A,W] --> [A,W]
is an isomorphism.
W
in
F
that the induced
This follows by consider-
ing the long exact sequences,
[SB, W]--> n
and using that If
p
[SB, W]-->
[A,W]
[A/n,W]->
[A,W]--> n
[A,W]
is finite.
is a prime the above argument gives us a very useful
model for the p-completion of
A .
It is given by the pro-object:
81
-
7.7
...--> A/pr+l--> A/pr-->
...--> A/p
Stable prohomotopy equivalence.
If
A,B
are CW-complexes we call stable prohomotopy equiA A valence between A and B an isomorphism ~: A--> B in the category pro-g~ o . Let
A
be a finite CW-complex.
Then
(A,A) = E(A)
given the structure of a ring by composition. as a Z-module, by lemma If
B
may be
It is of finite type
(7.1).
is a finite CW-complex,
module and a right- E(B)-module.
then
(A,B)
is a left
E(A)-
Of course, it is of finite type
over both of those rings, since it is in fact of finite type over Z
@
If the ring,
is the proflnite completion of Hom(~,~) = Lim(~,~)
.
Since
A , let
(~,~)
E(~)
denote
is a pro-object
in the category of finite abelian groups, we may regard the module Hom(~,~)
as a proflnite abellan group.
Thus
E(~)
is a topo-
logical ring and, after cot. (7.5), may be described as the profinite completion of
E(A) ; also
Cqrollary ~.7): Let
E(~) = E(A)@~ . A,B
be finite CW-complexes.
Then
Hom(i,B) = Hom(A,B)@z~ = HomCA, B)@E(B)E(~ ) .
Lemma (7-8): Let
r
A--> B
be a morphism of finite CW-
complexes having the property that Then
r
is an isomorphism.
~: ~--> ~
is a ~-isomorphism.
- 82 -
Proof:
7.8
Let us begin by seeing that the analogous statement
holds for finitely generated abelian groups.
Namely, if
~: C --> C'
is a homomorphism of finitely generated abelian groups such that A
is an isomorphism,
then
$
A
is, since
is a faithfully flat
functor on the category of finitely generated abelian groups. Now consider the two homomorphisms,
Hom(B,A) c~> E(A) Hom(B,A) _B> E(B)
induced by composition on the right (resp. left) by e. Since ~ is a ~-isomorphism,
and
Consequently
m
ing remarks.
If
m-l(1A)
~-i(~)
and
spectively.
A
A
is a finite complex, m and
and
O
are themselves isomorphisms by our open-
1ACE(A ) , 1BCE(B )
are the identity maps then
are left and right inverses to
This concludes lemma
Corqllary(7.9):
Let
A,B
r , re-
(7.8).
be finite CW-complexes which are
of the same stable prohomotopy type. a free module of rank one over B
~ are isomorphisms.
Suppose that
E(A) .
Then
A
Hom(A,B)
is
is isomorphic to
in the stable homotopy category. Proof:
Let
module, Hom(A,B)
e:
.
A-->
Then
B
A
r
denote a generator of the free E(A)is an isomorphism,
and lemma (7.8)
applies. Let
A
denote a ring (with unit, not necessarily commutative)
of finite type as a module over completion,
~ = A~
.
Denote by
Z .
Let
P(A)
~
denote the profinite
the set of isomorphism
-
classes of modules
M
A
property that
83
7.9
-
of finite type over
A
which have the
A
M = M|
is free of rank one over
~ .
A useful fact concerning completion of such rings
A
is the
following: Lemma (T.lO): Let
M,N
be A-modules,
with
M
of finite type.
Then
is an isomorphism.
Consequently a module
P
of finite type over A
A
is projective Proof:
over
A
if and only if
is projective
over
A.
Find a resolution of A-modules,
0--> R--> F--> M--> 0
with
F
free of finite type over
finite type over group.
A
since
F
It follows
that
R
is of
is a finitely generated free abellan
Form the diagram of exact sequences,
0
,
> HomA(M,N)|
A
> HornA
L" and since
F
Therefore
~A
general
A .
_
I"
is free of finite type, is injective.
HomA (R,N)|
Since
~F M
is an isomorphism.
was supposed to be a quite
A-module of finite type, we may conclude that
~R
is
84 -
also injective.
7. lO
Diagram-chasing then gives
~M
surjective.
The final statement of lemma ?.10 then follows in the nontrivial direction, for if
N 1 --> N 2 --> 0
is a surJective morphism
A
of A-modules then since
HomA(P, N1)@Z --> HomA(P,N2)|
is surJective
~p
is an isomorphism, and HomA(P, N1) --> HomA(P, N2) A surJective since Z is faithfully flat over Z . Let
A,B
motopy type.
is
be finite CW-complexes of the same stable prohoThen
is a module of finite type over A E(A) = A such that ~ is free of rank one over A by cor. (7.7). A Thus M is a projective module over A by lemma(7.10),and represents a class in this class c(A,B)
Hom(A,B) = M
P(A) .
c(A,B) .
Let us write
P(A) = P(A)
The results obtained above assure us that
plays the role of an obstruction to
the same stable homotopy type. guished element
and denote
0 E P(A)
Note that
stable pro-homotopy type.
P(A)
and
B
being of
has a distln-
represented by the isomorphism class
of the free module of rank one over Corollary (7.11): Let
A
A,B Then
A .
be finite CW-complexes of the same c(A,B) = 0
if and only if
A,B
have the same stable homotopy type. It is easily seen that a module A M
is free of rank one over
jective)
A-module such that
= A@Zp , for all primes localized at the prime
Thus
P(A)
p .
M
of finite type such that
may also be described as a (proMp = M@Zp p , where
is free of rank one over Zp
denotes the integers
(An application of NakayamaWs lemma).
is exactly the set
Pl(A)
defined in [~],
According to a result of Jordan-Zassenhaus
(lO.1).
[32] there are
85 -
only a finite number of isomorphism A
satisfying (1)
M
7. ll
classes of modules
over
the conditions: is a free Z-module.
(ii) M|
is a free
A|
of finite rank.
We are grateful to Bass for calling our attention result,
M
and pointing out that this implies
that
to this
Pl(A)
is a fi-
nite set.
Twisting finite CW-complexes.
Let us consider a category
~
which is the full subcategory
of the stable homotopy category generated by simply connected finite CW-complexes. complex,
and let
T
Let
Let
CW-complex
type as
~: 8o _> Ab
~(A) = (A,X) .
be a finite,
be a class in
finite simply connected stable prohomotopy
X
X
P(X)
: ~O
> Ab
We shall describe a
Y = XT
which is of the same
and such that
the abelian group
garded in a natural manner as a right
T
.
c(X,Y) = 9 .
be the functor represented by
Of course,
a left E(X)-module
simply connected CW-
representing
(A,B)
E(X)-module.
the class
T .
X ; thus, may be reLet
M
be
Define the functor
as follows:
cpl"(A) = (A,X)@E(x)M
(we may write symbolically, Definition: one such that
~
= ~(x)M).
A finite cohomological
functor
~: ~o _> Ab
is
86 -
(i) (ii)
a(S q) if
7.12
is finitely generated for all
A I --> A 2 --> A 3
q (> 2)
is a cofibration in
@ , then
e(A1)--> ~(A 2) --> (~(A3)
iS exact. (iii) There is an integer for all N-connected objects identified
N
such that
(where in the above, we have
W
with the functor which it represents).
W
Note that if
~
is a functor representable in
is a finite cohomological functor. representing
For, if
Y
~q(Y)
is fi-
q , (li) is implied by the Puppe se-
quence, and (ill) expresses the fact that logical dimension.
@ , then
is the object
~ , then (i) expresses the fact that
nitely generated for all
let
Homfunct(a,W) = 0
Y
has finite cohomo-
Actually we have the stronger result:
Lemma (7.12 ) : Le t
~: @ o _ >
Ab
T
P(X ) .
Then
be a class in
be represented by T
X , and
is a finite cohomological
functor. Proof:
Let
M
be a left- E(X)-module representing
Then (i) follows because (ii) follows because
M
M
is of finite type over
is projective over
E(X) .
T .
E(X) , and Since
T
satisfies both (1) and (ii), using an argument similar to that of lemma ~ . l ) , o n e
sees that
group for all objects Now let We show that
N ~T
A
~ T (A) of
is a finitely generated abelian
~ .
be the integer for which satisfies
(iii) for this
~
satisfies N .
(iii).
Suppose not.
-
87
-
7.13
Then there would be an N - c o n n e c t e d of functors T
(A) =
Consider the functor A
Applying the operation /k A T A A y: --> W . Note that y
for if
A
the zero h o m o m o r p h l s m range of
y(A)
, 'r
is
either.
defined by
9 we obtain a m o r p h l s m
is not the trivial mor-
CA9
A
y(A):
then
A ~(A)-->
(A,W)
to
= Lim(X,W)
the .
is not
This uses that both d o m a i n and
are f i n i t e l y generated a b e l l a n groups.
isomorphic
= Homfunct(X,W)
~
is an object such that
is not the zero homomorphism,
that
and a n o n t r l v i a l m o r p h i s m
.
of functors phism,
v> W .
~
W
A
functor But
fO 9
(X9
Thus
Note also
Homfunct(q~1",~)
= 0 , by assumption,
c-
which g i v e s
the
contradiction
sought.
We now state and prove Ed B r o w n ' s in our context
(cf.
a
sentable in Proof.
theorem.
is a finite cohomological
functor,
We shall construct,
Xl = Y2
"""
is repre-
(z) such that
by a d J u n c t l o n of cells of d i m e n s i o n
Ne shall also find compatible m o r p h l s m s where
m
below 9 a sequence
of simply connected finite CW-complexes Yq-1
then
@ .
Yo
from
theorem
[7]).
Representability
If
representability
~q = ( 9
Yq
is built
q 9 and
of functors
q+l .
~q--> ~ ,
.
The m o r p h l s m of functors
~q--> ~
will have the p r o p e r t y
88 -
that Y
~q(S i) --~ ~(S i)
7.14
is an isomorphism for
i ~ q .
denote the direct limit of the CW-complexes
would represent the limit functor category
~ .
~ = lim(~q), were X
Y
of
Y in the
~ , ~(X) =
where the latter group is Hom in the stable
homotopy category of CW-complexes.
Note also that these limits
are actually stationary in the sense that large enough choice of
~(S i) --~ ~(S i)
~(X) = ~N(X)
N , depending of course on
tain a morphism of functors, that
Yq , then
This means that for any object
lim~ ~q(X) = (X,Y)
If we let
~ --~ m
X
X .
We ob-
which now has the property
is an isomorphism for all
from (li) and the fact that every
for a
in
~
i .
It follows
may be built up
from spheres by a finite number of cofibrations that
~ --~
is an isomorphism of f ~ c t o r s . Our assumption (iii) then assures us that homological dimension,
Y
has finite co-
and (i) says that the homotopy groups are
finitely generated in each dimension. allow us to conclude that
Y
These two facts put together
has finitely generated integral homol-
ogy groups, which vanish beyond a certain dimension.
It then fol-
lows that
Y' .
Y
prop. ~.l).)
may be replaced by a finite CW-complex It is here that we use that
This finite CW-complex
y1
Y
([30],
is simply connected.
being, of course, again simply con-
nected, is an object of our category
~ , and represents the functor
So
We shall achieve point.
Suppose
Yq-1
(Z) inductively. given.
Take
Yo = Y1
to be a
For simplicity we let an object of
stand for the contravariant functor which it represents. bl,...,b m
generate the group
a(S q) .
Let
Use the same letters to
89 -
denote the corresponding morphisms Let
7.15
of functors,
Z = Y q - l V (Sq)lv (sq)2 v... v (Sq)m .
nite CW-complex obtained from
Yq-1
Extend the functorial morphlsm by setting it to be follows that
on
Z(S q)
cj
for
j = l,...,k
satisfies
morphism
Yq--> ~ , and moreover
and cells of dimension
for
q+l
giving us also:
i < q .
is the fi-
copies of
to a morphism j = 1,...,m .
f j: S q --> Z and define
(ii), the morphism
Since we have added to
hypothesis,
for
m
Z
Let
Sq .
Z --> a It then
Cl,...,c k
be
which generate the kernel of the above homo-
Since
phism.
by wedging
is surjective.
Take continuous maps
tatives of
That is
Yq-1 -> e
(sq)j
Z(S q) --> m(S q)
elements in morphism.
bj
bj: S q --> ~ .
Z--> ~
which are represenYq = ZUfj(Dq+l)j extends to a
Yq(S q) --> ~(S q) Yq-i
9
is an isomor-
only wedges of q-spheres,
we have not disturbed the inductive Yq(S i) --> ~(S i)
is an isomorphism
q.e.d.
A consequence of the representability
theorem is that T
de-
fined above is representable by a finite simply connected CWcomplex,
Xv .
Proposition (7.lB): type as
(i)
XT
has the same stable prohomotopy
X .
(ii)
c(X,X T) =
(iii) If
Y
is a finite simply connected CW-complex having
the same stable prohomotopy type as then
M
c(X,Y) = T,
Y ~ XT . Proof:
A
X , and such that
(i) follows from cot. (7.7)together with the fact that A iS free of rank one over A . (ii) is clear:
7 916
90-
Hom(X,X T) =
(X) = ( X , X ) % ( x ) M ~
(iii) comes from this fact:
The natural homomorphism
Hom(A,X)@E(x)Hom(X,Y ) Y> Hom(A,Y)
(7.1#)
A , provided
coming from composition is an isomorphism for all X
and
Y
are of the same stable prohomotopy type.
this we need only show that true using cor.
is an isomorphism.
But this is
(7.7).
Corollary (7.14): Let complex.
~
To prove
X
be a finite simply connected CW-
The set of distinct isomorphism classes of finite ~ -
complexes of the same stable prohomotopy type as in a one:one correspondence with the set to every such
T
the class
P(X)
X
may be put
by associating
c(X,Y) E P(X) .
Of course, our methods made use of simple connectivity in only one place, namely to insure that
XT
be a finite CW-complex,
and so we may state more generally: Corollary(7.15):
Let
X
be a finite ~ - c o m p l e x .
distinct stable homotopy types of CW-complexes stable prohomotopy type as dence with the set class
P(X)
X
Y
The set of
of the same
may be put in a one:one correspon-
by associating to every such
Y
the
c(X,Y) E P(X) . Since
P(X)
is a finite set by Jordan-Zassenhaus,
Corollary (7.16): Let
X
be a finite CW-complex.
we obtain: There are
-
91
7.17
-
only a finite number of stably inequivalent same stable prohomotopy Examples: ing a class
1.
type as
n
p , where
must be greater than one, i ~ +J
.
Xi
is simply connected.
(rood p), then
Xi
[!iy] C ~ m ( S n )
~
(7.18)
A:
and to try to recover,
as an invarlant of
considering all coflbrations
Sn - > f
X i --~ S m+l g
is any cofibratlon of the above form, ~n(Xl) = Z .
Xj
are of
f
~
Xi
a priori,
the
We do this by of the form
.
of
Xi .
Now, if
A
must represent a gener-
This may be seen by applying the long exact
stable homotopy sequence to m+l
Xi .
in the category
Note that there is one by the construction
snu(iu
and
The idea is to start with the finite complex
regarded as an object of
ator of
is a prime.
stable homotopy types.
Proof:
classes
p
represent-
i , consider the CW-complex,
Lemma (7.17): If distinct
y: Sm --> S n
Fix a continuous map
X i = snu(iy)D m+l
Since
of the
X .
u E ~m(S n) , of order
For any integer
CW-complexes
A .
Thus, up to homotopy,
f: S n -->
is either the canonical inclusion or the canonical
inclusion composed with the automorphism
of
Sn
of degree -1 .
From this we may conclude that up to isomorphism in a unique cofibratlon
A
of the form
(7 .18).
~
there is
-
Continuing
this cofibration
cofibration induced by
(7.19) where
92
7.18
one stage further,
forming the
g , we obtain:
Sn _ ~ f
h = ~S(iy)
-
X i --~ S m+l _> S n+l g h
, the sign depending on the orientations
for the
Sn
and
Sm+l
classes
+iy E ~n(S
manner,"
concluding
occuring.
TM) ,
Desuspending
h
chosen
we obtain the
and this was obtained in an "a priori
the proof of the lemma.
We state without giving the proof: Lemma (7,20):If i,j are not congruent to zero mod p , then ^ ^ X i ~ Xj . Note that to show the profinite completions of X i and Xj
are the same, it suffices to show that their q-completlons
the same for all
q .
simply connected
(cf.
completions for all
of
i .
p-completlon.
Xi
(This is true since (12.13)).)
split,
Therefore
Xi,X j
But for all
in the sense that
are
are finite and
q ~ p , the q(Xi) q ~ (Sn)qv(sm+l)q
one need only concentrate
on the
- 93 -
w
8.1
Hypercoverlngs.
In this section we review, with some extensions, of Verdier
([3] SGA 4, fascicule l, expos~ 5, appendice).
will be convenient
It
to make the following preliminary definitions,
generalizing Kan's notion of a free simplicial S(n,m)
the theory
group:
denote the set of surJectlve monotone maps
Set
Let
s: An --~ ATM .
n
SCn)
=
I
I sCn, m ) =
m=l Deflnltion cial object objects
(8.1):
K.
I
I SCn,m) .
m=l
A splitting up to level
with values in a category
Nk c Kk
for every
dition: For every
k ~ n
C
n
of a simpli-
consists of sub-
satisfying the following con-
j = O,...,n , the canonical map
J J Ns->K j s S(J) is an isomorphism,
where for
N k , its mapping to object
Nk
Kj
Ns
C
Remark:
K..
denotes a copy of
s: K k --> Kj .
will be called the nondegenerate
composition of
part of
The
Kk
and the
will be called a degeneracy de-
We shall refer to a simplicial
object with
with a given splitting as a split slmplicial object. In the category of sets, any slmpllcial object has
a unique splitting.
More generally,
using
be a locally connected and distributive exists,
,
being induced by
above direct sum decomposition
values in
s E S(j,k)
it is unique.
If
C
(w
below,
category.
let
C
If a splitting
is the category of etale,
separated
94
schemes over a scheme in
C
Let
products, and let
K.
C
(We do not use this fact.)
be a category closed under finite co-
be a simpliclal object of
a splitting up to level Proof:
8.2
X , then every simplicial object with values
has a unique splitting. Lemma (8.2):
-
m .
Then
SkmK.
C
which admits
is representable.
Set m
S(n)Im = I I S(n,j)
.
j=O Then we write
(SkmK")n =
t s Im Nt "
Now for
r: Ai --> An , we seek a definition of the operator
(S~K.)n
' and we do this inductively with respect to
each direct summand
(a)
If
Nt
n
r
on
and on
separately. r
take
to be the identity map
r
is surjective,
r
is not surjective, then construct the following
N t --> Ntr 9 (b)
If
commutative square of monotone maps:
An
t
rI Ai
> Aj
!~ > k U
There is a unique such square with Since
r
is not surjective, we have
u
surjective and k ( n .
v
injective.
Thus by inductive
95 hypothesis, u
is.
the operation of
v
8.3
is defined,
and by
(a), that of
Set
rlN t = u I N ,
and one easily sees that representing
K.
Let
C
be a category closed u n d e r finite direct
a simplicial
splitting up to level ting up to level subobject
so defined is ~ simplicial object
the functor after which it is named.
Lemm.a (8.3): sums, and
SkmK.
object with values
n-1 .
in
C , given a
To extend this splitting to a split-
n , it is n e c e s s a r y and sufficient
Nn c K n
to give a
such that the natural map below is an isomor-
phism:
( S k n - I K " ) n " Nn -> Kn
Proof:
Now let U).
Evident from the c o n s t r u c t i o n
C
be a site
nite products and fibred products,
Skn_lK.
that
C
is a sheaf on
C .
A point of
(Sets) --> C , i.e.,
i.e., C
to some universe
and u n d e r finite
that for
above.
is closed under fi-
we always assume that the topology of
than the canonical topology,"
Ill, 2)
of
([3], 1 . 2 ) ( b e l o n g i n g
We will assume for convenience
Moreover,
"
C
coproducts. is "weaker
x E C , Hom(.,x)
is a m o r p h i s m of sites
a f u n c t o r on the u n d e r l y i n g
([3],
categories
p: C --> (Sets)
which carries coverings
to surjective
families,
and which is exact
-
96
8.4
-
in the sense that it commutes with finite fibred products and arbitrary direct limits. the choice of a point.
A pointed site is one together with A pointed simplicial object
values in a pointed site
C
covering in
C
X.
(8.4):
of
C
Let
C
P(Xo) .
be a (pointed) site.
Let
A hyper-
is a (pointed) slmpllclal object with values
satisfying the following conditions for all (sUrJo)
with
is a slmplicial object together
with a choice of point in the set D eflnltlon
X.
e
be the final object of
X
O
C .
n : The map
--> e
is a covering. (sUrJn)
The canonical morphlsm
Xn+l --~ (C~
is a covering. A truncated hypercoverlng of level cial object satisfying If
X
(surjq) for
is any object of
ing of the site
C/X
n
is a truncated slmpll-
q ~ n .
C , we will refer to a hypercover-
as simply a hypercovering of
X .
Note that a morphism of sites carries hypercoverlngs to hypervcoverings. Enlightenment (8.5):
The reader dismayed by the austerity of
the above should work things out carefully in these particular
-
97
8.5
-
cases:
(a)
Let
C
be the site
(sets), whe're as always coverings
are surJective families of maps. non-empty
A hypercoverlng
slmplicial set, by (sUrJo).
construction
of coskeletons
plained as follows:
(w
the condition
map of the n-slmplex
An
that such a simpliclal
to
to
is then a
Taking into account the (sUrJn) is ex-
For every map of the simplicial
the boundary of the n-slmplex,
cont rac tible.
X.
set
BA n ,
X. , there is at least one
X.
inducing it.
set satisfies
It follows easily
the Kan condition,
In fact hypercoverings
are characterized
and is by these
two properties. (b)
Let
C
be any site,
a covering morphism. cated at level
o
Y
Regarding
U
with values in
This is Just, in plainer language,
It is a hypercoverlng feature,
an object of
C/Y , we may form the canonical <
<
<
Y .
This gives the clue as to the novel
start out w i t h an old-fashioned
covering
the computation of, say, cech cohomology, object above,
determined by the original covering. coverings,
however,
mined.
Kn+l--
(C~
U--> Y
Namely,
if we
and proceed
to
then we must mechanwhose n-simplexes
In the construction
we are given much greater freedom.
choice of n-simpllces to take
COSkoU .
slmpliclal object
and the ultimate use, of hypercoverings.
ically form the simplicial
U--> Y
as a slmplicial object trun-
<__
of
C , and
are of hyper-
In the
Kn+ 1 , the mechanical procedure would be ' when say
K./n
is already deter-
Here we are given the option of "refining"
(Cos~K.)n+l
-
98
8.6
-
further
(see below),
since we require only that condition
(surJn)
holds.
It is this option of refining further in each dimension
that makes hypercoverings useful in cases where the category of coverings is too coarse.
mere
We shall now prove some elementary lemmas about hypercoverings: Lemma 8.6: (~/n___~)> (K./n)
Let
K.
be a hypercovering of
a morphism of truncated
jects,
such that
level
n .
L./n
Then there is a hypercovering
n
is the morphism
L. ~> K.
(L./n)
(pointed) slmplicial ob-
is a hypercovering of
phism of simplicial objects,
C , and
L.
C of
truncated at C , and a mor-
whose truncation at level
~/n .
In fact, we m a y take
- (COSknT.) • (COSkni" )i.
The verification that this works is easy, and we omit it. Lemma
(8.7):
Let
C
covering truncated at level
be a (pointed) site and n , split up to level
is a m o r p h i s m of hypercoverings
truncated at level
which induces an isomorphism on (n-1)-skeleta, is split up to level Proof:
We take
n . L/n-i = K./n-I , and
Ln = ( S ~ _ I K . ) n " K n
.
K.
a hyper-
n-i . n,
There
L.-->K.
and such that
L.
,
-
The boundary operators summand,
99
8.7
-
Ln--> ~n-i
and the degeneracy maps
from the simplicial object
are the natural ones on each Ln_ I --> L n
Skn_iK.
Ln_l = Kn_l A>
:
(Skn_lK.)nr__> Ln
This yields a slmplicial
(the inclusion being as first summand). object which maps to
K.
map on each summand.
So defined,
ditlon sUrJn,
are those coming
by sending
Ln
to
Kn
via the natural
clearly satisfies
L.
the con-
q.e.d.
We therefore obtain: Lemma 8.8:
Let
K.
be a hypercovering,
a slmplicial object up to level K ~. --> K.
of simpliclal objects
n .
There is a morphlsm,
such that
(a)
It induces an isomorphism
on
(b)
K|
object,
ing for
C .
which is split as
is a split slmpliclal
Sk n . which is a hypercover-
We shall call such a thing a split hy2ercoverlng.
The above lemma allows us to replace hypercoverlngs hypercoverings, Lemma level
8.9:
by split
if we wish. Let
K.
be a hypercoverlng
of
C , split up to
n : l
Kn = (S~n-IK")n" Nn '
and let L.
of
M--> N n C
be a covering morphism.
and a morphlsm
L. --> K.
There is a hypercovering
which induces an isomorphism
i00 on
n-i
skeleta,
and such that
the expected way to Proof: fined.
~
Ln =
(Skn-lK")n" M
and maps in
.
The truncation of
Now apply
8.8
L.
at level
n
is taken as de-
(8.?).
The above lemma assures us that we may "refine the nondegenerate part"
of a split hypercovering
a refinement at level
n
entails no change on the
This is a quite convenient Lemma
(8. l0 ) :
Let
arbitrarily. n-1
K.,L.
be hypercoverings,
and let
be a map.
ping to
K|/n = K./n , and an extension of
map with
with
9': KI --> L.
.
skeleton.
technical fact.
f/n: K.In--> L.In K.
Moreover,
There is a hypercovering
Kt ~/n
map to a
In particular,
if
K., L.
are hypercoverir~gs
K./n = L./n , then there is a
K!
dominating both, with
K|/n = K./n = L./n . Proof:
We will construct
suffices
to find
B y lemma
(8.6), it suffices
K.'/ (n+l)
.
Let
N'
K!
so that
K!
by induction on
$/n
n .
extends to dimension
Thus it n+l .
to find a truncated hypercovering
be the object making the following diagram cartesian
h N'
Kn+ 1
The map
f
is a covering,
>
Ln+ 1
> (Cos~L)n+ 1
since
g
is one.
Since compositions
-
i01
of coverings are coverings, the map from
f
q~_n
, and = N',, (SknK)n+l
and define the truncated object and degeneracy maps.
Now
Thus we can take for
I Sn+l
h: N' --> Ln+ 1
X
If
S
KL/n+I
~/n
Set
obtained
K~ = Kq
for
,
with the evident face
extends canonically to
S~K.
.
this canonical map together with the
above.
For the moment, let ucts.
N'--> (COSknK)n+ I
by composition is also a covering.
~+l
map
8.9
-
C
be any category with finite coprod-
is a set and
with itself indexed by
X E C , we denote the direct sum of S
by
X@S .
This is a covariant bifunctor in
X
and
S .
The unit interval will mean the 1-simplex in the category of sets, denoted by
I.
Given a simplicial object
=
K.
A .1
with values in
the simplicial object
K.@I.
whose n-simplex object is Just
.
G , we may form
- i02 ~@I
n =
If
i6in
8.10
Ki n
'
where the superscript denotes the index9 which is an increasing sequence of zeros and ones of length degeneracies
n+l .
Thus faces and
operate on these indices, vlz. dvi ~-i
dr: ~ - - >
Svi
Two m a p s
fo,fl: K. --> L.
of slmpllcial objects will be called strictly homotoplc if there is
J
a map
f: K.@I. --> L.
such that f eo _ fo
where
cv: K. ,> K.@I.
9
f el
(v = 09
fl
are the evident "inclusions."
Two maps are homotopic if they can be related by a chain of strict homotoples 9
in the usual sense.
If
fo 9
are maps of pointed
slmpllclal objects9 then the strict homotoples will be assumed to preserve the base point. This notion of homotopy is functorial in
C .
Proposition
(8.11):
be hypercoverings with maps between them. between
Let K.
103 -
8. i i
C
K.,L.
be a (pointed) site, let
split, and let
fo, fl: K. --> L.
be
Suppose given a truncated strict homotopy
fo fl : f/n: K.|
There is a refinement of
L./n .
K. , i.e., a map of hypercoverings
~: K! --> K. , and an extension of
(~|
= f'/n
to a strict
homotopy f' : Kl|
Moreover,
K!
--> L.
.
can be chosen so as to be equal to
K.
in dlmen-
sions < n-i . Proof: dimension
By induction, it suffices to extend the homotopy to n+l , applying
(8.9).
Now using the splitting of
K. ,
we have in the notation of (8.1)
(K.|
=
I I isI n
Nit
t S(n) from which one sees immediately that from that of
K.
K.|
acquires a splitting
.
The degeneracies (Skn(K.| i with the sum of those N t , i 6 In+ 1 , t 6 S(n+l) can find a degeneracy operator
i = svi'
sv: An+l --> An
,
t = svt'
1
identify
for which we
satisfying
104 for suitable N~
i',t I .
8.12
One checks easily that the non-degenerate
are of the following two types:
any
(a)
t = identity 6 S(n+l,n+l)
(b)
t = sv: An+l --> An
i' .
Thus in this case
, i.e.,
for some
N i = Nn+ 1 .
v , and
for
i ~ Svi'
t E S(n+l,n) 9 and so
N i = Nn .
We have therefore
(K. |
where
Mn+ 1
)n+l = S ~ (K. |
is the sum of the
truncated homotopy
f/n
N%
of types (a), (b).
extends uniquely to
What we have to do is to refine g: Mn+ 1 --> Ln+ 1
)n+l = ~ + l
K.
S~(K.@I.)n+ 1 .
so that there is a map
commuting with the appropriate face operators.
On the other hand 9 we also have a unique extension of map
Now the
K.@I. --> COSknL.
.
Therefore there is a map
(COSknL)n+l , and the requirement on
g-
g
f
to a
Mn+ I -->
is Just that the diagram
"'> /
(COSknL)n+ 1
commute.
Since
r
is a covering by (surJ n)
coverings of the components a map exists. proof.
N n , Nn+ 1
of
for L. , there are Mn+ 1
for which such
Thus two applications of lemma (8.9) complete the
We leave the considerations of base point to the reader.
Corollary (8.12): K./n = L./n , and that
With the notation of (8.11)9 suppose that f~
= fl/n = identity.
Then a refinement
I05
K|
of
K.
can be found with
8.13
-
K|/n = K./n , such that the trun-
cated identity homotopy at level
n
extends to a strict homotopy
fI: K|@I. --> L . . Here a straight application of (8.11) would require a refinement differing from
K.
in dimension > n .
the existence of non-degenerate n+l , a s i n
N~
the proof of (8.11).
homotopy extends trivially to are summands of
of type (b) in dimension
However, in our case the identity
SknK.@I.
(SknK.@I.)n+ 1 .
This comes from
, and the
N~
of type (b)
Thus we do not after all need to
drop back a step in order to extend the homotopy of dimension
n+l .
Now an application of (8.11) yields the result. Combining
(8.11),
Corollary (8.13):
(8.12) with lemma (8.10), we obtain (1)
For a (pointed) site
C , let
denote the category whose objects are hypercoverlngs, maps are homotopy classes of morphisms. HR(C) ~
is filtering.
underlying site.
Let
C
and whose
Then the dual category
be a pointed site, and
The stripping functor
HR(C)
U
HR(C) ~ --> HR(U) ~
its is co-
final. (ii) note by tion
Let
K./n
HR(C,K./n)
be a truncated split hypercovering,
the category of hypercoverings with tranca-
K./n , and with homotopy classes of morphisms.
HR(C,K./n) ~
and de-
Then
is filtering.
Verdier's main result ing proposition:
((8.16) below) is based on the follow-
106
-
Proposition object
(8.1#):
Let
K.
8.14
-
be a hypercovering.
U E C , we denote as in [3] by
~U
For an
the sheaf of abelian
groups characterized by the property that for any abelian sheaf A
we have Hom(~u,A)
Consider the simplicial sheaf
= A(U) .
~K.
with its augmentation
2ZK. --> ~--- = ~e (e of
the final object of ~e
C).
This augmentation is a resolution
"
We will prove the proposition under the extra assumption that the site
C
has "sufficiently many points," so that a sequence
A --> A' --> A" of sheaves is exact when for every point (sets) the sequence
p*A --> p*A' --> p*~' is exact.
enough for our applications,
p: C -->
This will be
and we leave the general case to the
reader. We have
P*( ~ U ) = ~ p ( U )
' where the abelian group on the
right is just the direct sum of set
p(U) .
For, if
~
with itself indexed by the
A E (Ab) , then
Horn( 2~p(u),A ) = A p(U) = p.A(U) = Horn(~u,P.A) = Horn(p* 2Zu, A) .
hence the functor
(*)
p*
applied to the above augmentation yields
mp(x.)
->
-
-
But
p(X.)
(8.5).
is a h y p e r c o v e r i n g
107
-
8.15
in (sets), whence is contractible
T h e r e f o r e its h o m o l o g y is that of the point,
exactness
which is the
of (*).
Corollary(8.15): a b e l i a n sheaf on
Let
C .
K.
be a hypercovering,
and
F
an
There is a spectral sequence
E~q- ~q(~p,F) --> ~P+q(c,;)
where
H
denotes
sheaf cohomology,
This is a standard
site
C .
(8.16):
H'(C,F) = H'(e,F)
spectral sequence of
tion, via the i d e n t i f i c a t i o n Theorem
and
Let
F
Ext
.
for a resolu-
HP(*,F) = ExtP(2Z.,F))
.
be an a b e l i a n sheaf on the
(pointed)
We have a canonical i s o m o r p h i s m
Hq(C,F) - lira Hq(F(K.)) -9 K. where the group
H
F(K.)
on the right denotes
c o h o m o l o g y of the cosimpliclal
, and where the limit is taken over the c a t e g o r y
HR(C) ~ Proof:
We pass to the limit over the terms
spectral sequence
E pq
of the
(8.15) :
E~q(K. ) _- HP (Hq(K.,F)) .
Let
K.
be a split hypercovering,
and
~
a class in
Hq(~,F)
.
-
Write
Kn = ~ I
classes
Nt
as in (8.1).
a t 6 Hq(Nt,F)
there are coverings Applying K!
of
.
N~
such that
-
8.16
Then
a
is the product of
Since cohomology vanishes locally, of
(8.9) repeatedly, K.
I08
Nt
such that
a t ~-> 0
in
Hq(N~,F).
one finds that there is a refinement
a ~> 0
in
Hq(KI,F)
.
This proves that
lim EPq(K.) = 0 , for all
lim Hq(K.,F) = 0 , hence that
q > 0 ,
-9
which yields the theorem. Definition
(8.17):
Let
C
be a class of groups.
of a site is said to have C-dlmension < d constant abelian sheaf stant sheaf site
C
A
for
on
if for every locally
C , locally isomorphic to a con-
A 6 C , Hq(X,F) = 0
for all
q > d .
is said to have local C-dimension < d if for every
there is a covering Theorem K./n
F
An object
X I --> X
(8.18) :
be a truncated
Let
C
such that
X'
all of C-dimension ~ d
C .
for
X 6 C,
has C-dimension < d .
have local C-dimenslon < d , let
split hypercovering,
cally constant C-sheaf on
The
and
F
an abelian lo-
Assume that the objects
q = 0,...,n .
Kq
are
Then the canonical
map --)
Hq(F(K. ))
Hq(C,F)
K. is surJective if K.
q = n+d+l , and bijective if
runs through the category Proof:
By
HR(C,K./n)
HR(C,K./n)
that as
K.
of (8.13).
(8.9), we can refine the non-degenerate parts of
a split hypercovering arbitrarily, gory
q > n+d+l , where
.
in dimension > n , in the cate-
Since cohomology vanishes locally,
runs over
HR(C,K./n)
, we have
it is clear
-
l o 9
lira HqCK.,F)
8.3.7
-
= HqCSknK.,F)
K. for all
q > 0 .
By lemma
(8.19) below, it follows that
lira HP(Hq(K.,F))
= 0
K. if
p > n , and
q > 0 .
Because the
Kq
(q = 0,...,n)
have
G-dimension <_ d , we also have by (8.2)
Hq(S~K.,F)
if
q > d .
quences
Thus if we pass to the limit over the spectral se-
(8.15), we obtain
lira K. if
= 0
or if
q > d
= lira HP (Hq(K.,;)) K.
p > n
and
q > 0 .
o
This yields the theorem
immediately. L emma category
(8.19): C
Let
K.
with finite coproducts,
a functor sending coproducts
Hp(T(S~K.))
Proof: T
be a split simplicial object in a and let
to products.
= 0
if
Consider the inclusion
carries sums to products and
K.
T: C ~ --> (Ab)
be
Then we have
p > n .
S~_IK.~-> is split,
yields an exact sequence of co-simpliclal
S~K..
Since
this inclusion
groups
-
(*)
where
i z o
8.18
-
0--> X.--~ T(SknK.) --~ T(Skn=iK.) --~ 0
X.
can be described as follows:
Let
Sn
,
denote the
minimal simpllcial pointed set representing the n-sphere, i.e., the simpllclal set with two non-degenerate simplices: O-simplex and one n-simplex. part of
~
.
Then
by the pointed set to one).
Thus
Xq Snq
Let
N E C
is the product of
one
denote the non-degenerate N
with itself indexed
(where the base point factor is set equal
Hq(X.) = Hq(s.n,T(N))
pointed n-sphere with values in
is the cohomology of the
T(N) , which is zero for
q ~ n .
The lemma now follows from the exact cohomology sequence of (*) and induction.
-iii-
9.1
w
The Verdier functor.
Let
C
be a category admitting finite fibred products.
say that
C
is distributive
and if the following
if it has an initial object
condition holds:
We
~ ,
For every set of objects
T i 6 C , i E I , such that the coproduct able in X--> S
I I Yi is representi Yi -> S , and for every m~p
C , for every set of maps in
C , the canonical morphism of functors
LiI X• i
x• s(! i I q )
is an isomorphism. E~ery site ([3] II, 4.8b). X 6 C
Let
has an underlying distributive C
be a distributive
is called connected if
and if
X
C
implies that
object.
category
category.
An object
is not the initial object
coproduct decomposition,
Xi = ~
for exactly one
~ ,
so that
i .
The cate-
is called locally connected if every object is a coprod-
uct of connected connected
X
has no non-trivial
X = Xl.. X 2 gory
C
objects.
For brevity,
if it is locally connected, Using the distributivity,
the expression of an object
X
Sects is essentially unique. an object its set of connected
C
it is immediately
seen that
as a coproduct of connected Moreover,
,
ob-
the rule associating to
components is a functor.
(Sets)
is
and has a connected final
note this functor by
~: C - >
we will say that
We de-
-
1 1 2
9.2
-
and call it the connected component functor. Now let functor
~
C
be a locally connected site.
to the category
HR(C)
Then applying the
of hypercoverings (w
we
obtain a pro-object
T-]- c = [~(K. ) ]K. ~ m~(c) in the homotopy category
~
of simplicial sets.
associating to a locally connected site
C
The rule
the pro-object
is functorlal with respect to morphisms of sites.
V-K c
We call it the
Verdier functor. Note that if functor
C
is a connected pointed site, then since the
p: C--~ (sets) is exact, it commutes with coproducts.
Hence an element ponents of
X .
x E p(X)
"lles in" one of the connected com-
Thus a pointed hypercovering yields a pointed
simplicial set via
~o ' and so the Verdier functor
pointed sites to the homotopy category sets.
~o
~-[
of pointed slmpllclal
Via the geometric realization, we can identify - ~ K
an object of
carries
C
with
~o " and thus the theory developed in sections 2-6
is applicable.
In particular, the homology pro-groups of a con-
nected (pointed) site, and the homotopy pro-groups of a connected pointed site are defined, by the rule
Hq(C,A) -- Hq(T-]- C,A) (9.1) 9 q(C) If
A
= ~qC[-]- C) .
is an abelian group, we may form the "constant'' sheaf
-
= Ae
on
Lemma X 6 C .
C , where (9.2):
9.3
-
is the final object,
e
Let
i13
C
as in (8.14).
be a locally connected site, and let
Then there is a canonical isomorphism
We leave the proof as an exercise. is a hypercoverlng,
then
It follows that if
~(K.) = A ~(K')
Hq(_A(K.))
--
A v(x) ~> A(X)
. K.
, therefore that
Hq(?z(K.),A)
where the group on the right is the cohomology of the slmpliclal set.
Applying
(8.16), we find that cohomology with constant co-
efficients factors through the Verdier functor: Corollary A
(9.3):
an abellan group.
Let
C
be a locally connected
site and
There is a canonical isomorphism
Hq(c,A) Hq<]-]-C,A) Example
(9.9):
etale site over schemes
X'
X .
Let
X
be a prescheme,
and let
C
be the
Recall that this site has as objects the
etale over
X , and as coverings surJectlve families.
It is clear that an object
X'
of
if it is a connected prescheme.
C
is connected if and only
We say that
nected if the etale site is, so that every
X X'
is locally conetale on
X
is
a coproduct of connected schemes. proposition connected. cf. EGA I 6.1.9
(9.5) :
A locally n o e t h e r i a n prescheme is locally
-
114
-
9.@
For a locally connected prescheme
(9.6)
X , we write
Xet E p r o - ~
to denote the pro-object We refer to
Xet
-[~-C
for the etale site
as the etale homotopy type of
C
X .
over
X .
By pointed
prescheme we mean a prescheme together with a geometric point. This point then gives a point for the etale site
C , in the
obvious way.
then
Thus if
pointed site, and so
X
is a pointed prescheme,
Xet
is in p r o - ~ o
in this case.
C
is a
Notations
of the type
Hq (Xet, A)
(9.7)
q(Xet) A 6 (Ab)
H q (Xet, A)
are defined as in (2. i), (2.2). Example
(9.8):
Let
X
be a pointed, connected, and locally
connected topological space, and let whose objects are coproducts X .
C
be the ordinary site
(disjoint unions) of open subsets of
Then we obtain a definition of homotopy pro-groups for
X ,
which we will denote by V
~q(X) = ~q(C)
We will discuss the relation of these pro-groups with the usual homotopy groups in section 12. Example
(9.9):
Let
G
be a group, and let
C
be the site
115
of left G-sets, where
always
as
a
-
9.5
covering is
a
surJective map.
This site is canonically pointed by the stripping functor (G-sets) --> (Sets).
Among the hypercoverlngs we have the one
(9.10)
> > GxG
GxGxG
> G=K.
We leave it as an exercise to show that since the G-set
G
is
a projective object, this hypercovering dominates every other. Thus in the category prO-Ko
we have
7r(K.) ~ - [ ~ - C
whence
77-C
is actually in
the orbit space of
K.
Ko "
Moreover,
under the action of
~(K. )
is just
G , and this orbit
space is the standard construction of the Eilenberg-MacLane space K(G,I)
.
Hence C = K(G, 1)
,
and
=l(c)
= G
=q(C) - o Example
(9. ii) :
be the site of finite
Let
G
if
.
be a pro-finite group, and let
continuous G-sets.
one sees that every truncated hypercovering by one of the form
>-G•215
q>O
,. ~
G•
>~
Then via example L./n
C (9.10),
is dominated
-
9.6
116-
m
wh e re
G
is a s u i t a b l e
finite
quotient
of
G .
Thus
we obtain
C ~ K(G,I)
and
~i(c)
- G
~q(c)
= o
if
q > o
.
- I17
The fundamental
w
Let
C
[13]) over site
i0.i
group.
be a site, and C
-
~
a flbred category
which is a category of descent
C , i.e.,
so that for
X E C
and
([15], exp. 190,
(loc.clt.) for the
a,b E $(X) , the functor
f ~-> Hom(f*a, f'b)
for maps
f: Y--> X
is a sheaf on the site
morphisms between objects of
~
"descend"
C/X , or, so that with respect to cover-
ings. Let
K.
be a hypercovering
of
C , and
x ~ ~(K o)
By descent data for
x
(lO.1)
relative to
K. , we mean an isomorphism
(~: doX ---->d~x
between the two pull-backs compatibility
.
of
x
condition that in
(10.2)
we have
"
as in (loc.cit.).
(10.3):
one correspondence
~(K l) , satisfying the
~(K 2)
d~r = % r 1 6 2
with the conventions Proposition
to
The map
K. --> COSkoK.
between descent data for
x
induces a onerelative to
K.
- 118-
and descent data for Proof:
x
relative to
First of all, since
(8.4), the compatibility iff it does on COSklK.
COSkoK.
.
K 2 --> (COSklK.)2
condition
COSklK..
is a coverin~
(10.2) holds for
~
on
K.
Thus descent data depend only on
, and so we may replace
account the description of as the subobject of
10.2
K.
Cosk I
KlXKlXK 1
by
COSklK..
(w
Taking into
we may then write
of triples
(bo,bl,b 2)
satisfy-
ing the identities
dob o -- dob I
(lO.4)
dlb o = dob 2 dlb I = dlb 2
Consider the following diagram:
(lO.5)
K I X (K ~ xK ~ )K I
d
<
0
Ko <
K1
d~
Ko < . . . .
The map
u
I'
K2
<.
do
<. KoXK o < <
is defined as follows:
K o xK o xK o
If
=
K.
I =
COSkkK.
(a,b) E KIX(KoXKo)K I
ii9 -
10.3
(here we are using a shorthand notation in which "elements" supposed to be replaced by maps from a "test object" KlX(KoXKo)K1)
, so that
dla = dib
for
is indeed in
(10.4) are immediately checked,
pull-back to
K.
is a covering
(8.4).
$
.
determines
x
relative to
it uniquely,
since
COSkoK. (do,d I)
K.
comes from one relative to
For this, we need only show that the two pull-backs For, since
the isomorphism
~
, its of (10.5)
Thus it remains to show that any descent
relative to
are equal.
u(a,b)
so that
K2 .
Now given descent data for
data
into
i = 1,2 , then
u(a,b) = (b,a, Sodla ) E K 2 c KI•215 I
The conditions
are
(do,dl)
descends to
is a covering, KoXK o .
COSkoK.. 6o~ , 6[~
this implies that
The fact that it is
descent data follows from the surjectivity of the map
r
of (10.5).
We have
~o = d o U
61 = dlU Sodl61 = d2u
Therefore
(10.2) implies
=
Hence it suffices to show with simplicial identities,
*** (61dlSor
"
d*~*~l~oV= identity. and (10.2),
But, calculating
-
1T
10.4
120-
* * (So%r
=
*d*
or
r = ~ts*d* ) O 2V Since
~
r
* * = dlSo~ * * = id , as required. Sod2~
is an isomorphism,
Thus descent data depend only on the covering the final object category in a here by
of
C , and on
well known way
~
Let
e
.
(~15],
~ E ~(e)
, and put to
ical descent data for
e*~-
x -- r
(r
~13]), which we will denote
.
=
x .
Then
Y(e)
.
d*Xo and
d~x
are
(edl)*~ , which yields canon-
If a descent data
(x,~)
is
to s u c h a canonical descent data, it is said to be
effective descent data. the categories
Y(e)
In case every descent data is effective,
and
are equivalent.
is a full subcategory of As an example,
(cf.
suppose
C
Otherwise,
~(e)
[13 ] for a detailed account).
closed under arbitrary coproducts,
and consider the fibred category of "trivial ~(X)
of
They form a
We think of it as a completion of
canonically isomorphic
isomorphic
x E ~(K o) .
K o --r e
coverings,"
so that
is the category of objects
(sets)] ,
where maps are ones compatible with the projection The category
~(e)
locall~trivlal
X@S--~ X .
is then to be thought of as the category of
coverings
of
e .
etale site of a locally noetherian
It is known that if scheme
C
is the
Y , then every object
Jl.,
of
~(Y)
is represented by an element of
data for a covering
X@S , X--~ Y
C , i.e.,
that descent
an etale surjectlve map, is
121
-
effective in
C .
-
1 0 . 5
This is proved in ([9],X,6).
Now suppose in addition that
C
is locally connected.
it is immediately seen that the group of automorphlsms of over
X
is just the permutation group
connected, and is
P(S) ~(X)
P(S)
in general.
of
S
Thus if
if K.
ing the condition P(S)
Kl@S=
is
Ko|
is satisfy-
(10.2), which in turn is given by an element of K1 .
Obviously this is
nothing but a 1-cocycle of the simplicial set
~(e)
X
do(Ko@S ) = d~(Ko@S ) *
for each connected component of
in the group
X@S
is a hyper-
covering, then descent data for the trivial covering given by a n automorphism of
Then
P(S) .
~(K.)
with values
Two such data yield isomorphic elements of
iff the cocycles are cohomologous. We recall that covering spaces of the geometric realization
of a simplicial set
L.
are in 1-1 correspondence up to canon-
ical isomorphism with simplicial covering spaces, which are maps of simplicial sets
L! --> L.
satisfying the following axiom:
For any simplex
x 6 Lq , denote by
mapping to
Then the face operators
x .
L~
the set of simplexes dv
carry
L'
iso-
X
morphically to
L~v x .
'
"
This fact is well known and elementary.
Moreover, a moment's reflection shows that isomorphism classes of simplicial covering spaces for all
x
L! --> L.
for which
are classified by elements of
card(L~) = card(S)
Hl(L.,P(S))
.
Thus we
obtain the following Corollary (i0.61:
Let
C
be a locally connected site closed
under arbitrary coproducts, and let
K.
be a hypercovering of
The category of locally trivial coverings of the final object of
C
which become
trivial on
Ko
is equivalent with the
C. e
122
-
10.6
category of simpllclal covering spaces of Now suppose that a group, and denote
C
is pointed and connected.
by
~l(c,G)
Let
G
be
the set of isomorphism classes
of locally trivial coverings of G
~(K. ) .
e
given with an operation of
making them into a principal fibre bundle.
By the above
corollary, we have Corollary (10.7):
With the above notation,
~I(c,G) = Hom(grps)(~l(C), G)
Thus the pro-group
~I(C)
has the usual property.
In case
C
is the etale site on a noetherian scheme, this is the "pro-groupe fondamentale enlargi" of ([9]X,6, [21]). As another example, we can let
~
be the fibred category of
locally constant abelian sheaves, i.e., sheaves
F
is an abelian group
such that the
restriction of C
F
A to
and a hypercovering C/K o
K.
is the constant sheaf
for which there
A .
If again
is locally connected, then a reasoning analogous to the above
shows that to give such a sheaf is equivalent with giving a local coefficient group on morphism
~(K. )
(in particular, is given by a homoif
~l(~(K. )) --~ Aut(A)
C
is connected and pointed).
Consequently we have by (9.3) Corollary (10.8):
Let
C
be a locally connected site.
category of locally constant abelian sheaves on
C
with the category of locally coefficient groups on
The
is equivalent ]--KC , and
123
-
10.7
there is a canonical isomorphism
~q(c,;) ~ Hqg[-T c,;) where
F
denotes both the corresponding sheaf and local coef-
ficient group.
124 -
w
A profiniteness
Let
X
ll. 1
theorem.
be a locally noetherian prescheme.
be geometrically unibranch if for every closure of the local ring
~X,x
x 6 X
X
is said to
the integral
is again local
(EGA IV, 6.15.1).
An equivalent
condition is that every
X'
is connected,
is in fact irreducible.
Thus the notion is local
for the etale topology.
etale over
X
which
A normal prescheme is geometrically uni-
branch. Theorem
(ll.1):
ally unibranch, Xet
Let
be a pointed,
noetherian prescheme.
is pro-finite,
~q(Z)
E @
and
@Ko
for all
To prove the theorem, hypercovering
connected,
@
denotes
denotes the category of
the Z E Ko
q .
we remark to begin with that every
K.
of
X
one in which each
Kq
is a noetherian prescheme--we
a hypercovering
geometric-
Then the etale homotopy type
i.e., is in p r o - @ K o , where
class of finite groups, such that
X
for the etale topology is dominated b y
noetherian.
This is trivial.
will call such
Thus the theorem
will result from the following more precise assertion: Theorem noetherian
(11.2):
etale hypercovering
Proof: lent,
X .
([13]VIII, l.1), we may suppose
over
and so
of
Since the etale sites of
irreducible, Y
With the notation of (ii.i), let
X
it has a generic point will cover
Y• P
P .
If
Y
Then X
X
and
~(K.) Xre d
reduced.
K. be a
6 d2(O 9 are equivaSince
X
is
P , and every non-empty etale is connected,
will be a one point scheme.
it is irreducible,
Thus the connected
-
components
of
Y
and of
125
YXxP
-
l l . 2
are in 1-1 correspondence,
whence
we have ~r(K. ) = 7r(K. XxP )
where
K. XxP
,
is the induced etale hypercovering
Now the etale site over
Spec K ,
the site of continuous G-sets,
K
a field,
G = Gal(~/K)
G-sets.
Assume
K. Kq
a hypercoverlng
Truncating quotient
Let
G
be a pro-
q .
Then
~q(~(K.))
q .
K. , and replacing
G
by a suitable finite
~ , we are reduced to the following:
Proposition the pointed
site
covering of
C .
Proof: functor. functor
is equivalent with
of the site of continuous
a finite set for all
are finite groups for all
P = Spec k(P).
([13],VIII,2).
Hence we are reduced to proving the following: finite group, and
of
(ii.3): (w
Let
G
be a finite group,
of finite G-sets,
Then
~q(~(K. ))
The functor
~: C-->
and let
([3],VIII,2), correspond
q .
(sets) is just the orbit space
This functor is a morphism of sites C
be
be a hyper-
i.e.,
the stripping
(sets) sending a G-set to its underlying
the category of sheaves on
C
are finite groups for all
We can also consider the point, p: C-->
K.
let
(sets) --> C .
set.
If we identify
with the category of G-modules,
then the inverse and direct image functors
to the restriction of a G-module
p*,p.
to the trivial sub-
group, and the induction of a module from the trivial subgroup to G .
We have for any sheaf
F
on
C
maps as with G-modules
126
(ll.4)
whose
-
F re_~s > p,p*F
composition Lemma
Let
that the u n d e r l y i n g
L.
that
is a finite
group for every
constant
by the order
be a simplicial
simplicial
> 0 , i.e.,
Proof:
trace.> F
is m u l t i p l i c a t i o n
(ll.5):
i1.3
set
p(L. )
sheaf on
is acyclic Then
G . C
such
in dimension
Hq(v(L.),ZZ)
q > 0 .
Hq(v(L. ),~ ) = Hq(2Z(L.))
C .
of
object in
Hq(p(K.),~- ) = 0 , q > 0 .
We have
n
D e n o t i n g by
~
, where
also the constant
2Z
is the
sheaf on
(sets), we have
Hq(p.2Z (L.)) = H q ( p ( L . ) , ~
Since
p*~=
nihilated rated,
by
~
, it follows n .
from
K.
For,
that
.
Hq(~(K.))
is angene-
the lemma follows.
is a hypercovering,
suppose
this shown.
ing as simplicial in
(11.4)
q > 0
Since these groups are obviously finitely
Now we claim that to prove if
) = 0 ,
set.
C , and a map
(ll.2) it suffices
then Let
Vl(V(K.))
S. --> ~(K.)
We can construct
L. --> K.
such that
to show that
is a finite group. be the u n i v e r s a l
a simplicial
object
coverL.
v(L.) = S. , as follows:
We put symbolically
L. = K.@~(K.)S.
i.e.,
for every connected
the elements
of
Sq
component
mapping
to
x .
x
of Then
Kq , let
Sx
denote
-
Lq =
127
~ K
-
ll.#
) x@S x q
Or, viewing Lq
Kq
as a set with G-operation,
the underlying set of
is p(Lq) = p(Kq)Xv(Kq)S q
and
G
operates on the left factor.
is clear that
p(L.)
simplicial set
,
Via this description,
it
is a simplicial covering of the contractible
p(K. ) .
Therefore
p(L. )
is acyclic in positive
dimensions, and by lemma (ll.5), we find that
Hq(~(L. ),~ ) =
Hq(s.,~ )
Since
are finite groups for all
q > 0 .
S.
connected, it follows from the m o d - ~ Hurewicz theorem that = ~q(~(K.))
are finite groups for all
is simply ~q(S. )
q > 1 , which completes
the proof of the theorem. Thus it remains to show that
~l(~(K.))
is a finite group.
Since this is a finite simplicial set, it will suffice to prove Lemma (ll.6):
Let
G
be a group and let
covering of the site of G-sets.
Then
~(K.)
K.
be a hyper-
satisfies the Kan
condition in dimension 1. For, then Kan's construction of of the finite set
~l
gives it as a quotient
K1 . I
Proof: do~ .
Write
K. = ~(K.) .
We need to find
S E
Let
with
x,y E KI doW = ~
is clear that we may choose representatives ly speaking, in z E K1
with
P(K1) )
with
doX = doY .
satisfy and
x,y
doX =
dlb = ~ . in
K1
It
(strict-
By (8.4), there is a
128
*
11.5
doZ = dlX dlZ = dly
Then the triple c KlXKlXK 1 to
(x,y,z)
(cf.(lO.~)).
(x,y,z) , and
~
9
identifies with an element of By (8.4), there is a
has the required property.
b E
(co,%K.)2 ~ppCn8
129
w
-
12.1
C ompari son theo reins.
We begin with a topological result: pointed topological space. is paracompact, and that every point
x E X
neighborhoods. C
X
S.X
hypercoverlng of
X
Let
of coproducts of open subsets.
the singular complex, i.e., the slmpllcial
C
SqX = [maps Aq --> X] .
With the above notation, let
U.
be a
such that every connected component of every
is contractable.
Then the simpliclal set
ically homotopic to the singular complex 7-I C
X
is locally contractable, i.e., that
Such a space is locally arc-wise connected.
Theorem (12.1):
pro-object
be a connected,
contains arbitrarily small contractable
set of singular simplexes
Uq
X
Assume that every open subset of
be the ordinary site on
We denote by
Let
S.X
~(U.) of
is canon-
X .
Hence the
is canonically isomorphic to the element
S.X
of Ko 9 and in particular, in the notation of (9.8), ~q(X) v ~q(X) . Corollary (12.2) : scheme over
C .
Let
X
be a connected pointed algebraic
With the notation of (12.1), we define
Xcl = - ~
C
.
Then in pro-K o
Xcl ~ S.X
For, the hypotheses of (12.1) on
X
are satisfied for an algebraic
-
130
scheme, since it is triangulable Proof of (12.1): Uq .
Then
S.U.
Let
-
12.2
(cf. [20] ).
S.Uq
denote the singular complex of
is a bisimpliclal set.
We denote by
A.
its
diagonal simplicial set
Aq = S qUq
.
Then we have obvious maps of simplicial sets
s.x
and we claim that these two maps are homotopy equivalences, which will prove the theorem. First of all, to prove that the maps induce an isomorphism on
~l ' it is convenient to do this by showing that the category
of simplicial covering spaces of the three objects are equivalent via
~,8 9
Since
X
is locally arcwise connected, we can identify
simplicial covering spaces of spaces of
X .
X
with locally trivial covering
Now by the discussion of section lO, simplicial
covering spaces of space of
S.X
~(U. )
are given by gluing data for a covering
relative to the open covering
trivial that coverings of
~(U.)
and of
Uo . S.X
Thus it is
correspond.
It
remains to show for instance that every descent data for a simplicial covering of a unique way.
A.
is obtained from descent data on
~(U. )
This is a simple exercise which we leave to the
in
12.3
13I
reader. X' --> X
Now let
is obviously a hypercoverlng
u' = x' •J %
assumptions space set
say connected.
of
X'
corresponding
covering
= A| , it is in fact the diagonal
SqUq ' to
of
S.U ~ . . X'
simplicial
Moreover the covering space of
is Just
S.X'
.
Then
satisfying the
of (12.1) , and if we form the simplicial
~(Ut')• Aq' =
be a covering space,
S.X
Therefore we obtain a
diagram
)
covering
(12.3).
s.x'
To show that
~,8
are homotopy equivalences,
it suffices to show that for every such induce isomorphlsms
X' , the maps
on cohomology with arbitrary coefficients
Thus we may as well drop the By the Eilenberg-Zilber
theorem,
the two spectral sequences
"EPq = H~(H~(AS'U'))
Since
Uq
'E~ q = 0
if
(h = horizontal, v = vertical)
the graded groups of certain filtrations
is component-wise q > 0 .
'E p~ = HP(H~
A .
' .
,EPq = ~ ( ~ ( A S ' U ' ) )
have abutments
~' , 8'
contractable,
we have
of
Hq(Up,A) =
Hence the spectral sequence yields
= HP(~(U.),A)
~ HP(A.,A)
H'(A.,A).
132
This shows that
~
To show that
constant sheaf
8
induces an isomorphism on cohomology,
sheaf A
12.4
is a homotopy equivalence.
use the sheaf of singular cochains The cosimplicial
-
S'(A)
gq(.,A)
we
(cf. [6], p. 18).
is a flabby resolution of the
([16], p. ll9), and it is used to give the
usual identification
of sheaf c o h o m o l o g y w i t h
singular cohomol-
ogy (loc.cit.). Consider the surjective map ([16], p. 19) of bicomplexes
A SpUq
~ > gP(Uq,A)
> 0
.
A consideration of the first spectral sequences shows that associated quences:
~
above
induces an isomorphism on the cohomology of the
total complexes. We have
Consider the second spectral se-
Hq(gP(u.,A))
~ Hq(X,~P(A))
from the spectral sequence
(8.15) since
moreover
if
"E~ q
'E~ q
Hq(X, gP(A)) = 0
q > 0 .
.
gP(A)
This follows is flabby, and
Thus the map
r
on
yields
(12.4)
,,po_2 = Hph
and the groups sequences.
HP(x,A)
(AS.U. )) >
HP (X,A)
,
are the abutments of the two spectral
Now we are interested in the map
A S.X
given by the projection
S.Uq
> A S'U"
-> S.X .
This map induces a map
-
1 3 3
-
12.5
Hp (X,A)
> ,,~po
Clearly the composition of this map with which shows that completes
and
f* D
induces an isomorphism on cohomology,
and
the proof.
Now let back
6
(12.4) is the identity,
f
be a morphism of sites
carries hypercoverings
C --> D .
Then the pull-
to hypercoverings.
are pointed and connected,
and if
f
Thus if
C
is a pointed morphism
then we obtain a map
U-Ff--T We callanf-m_~_~ ~ C
:' I 1
D
c
in p r o - K o
between hypercoverings
L.
of
.
D
and
K.
of
a map
$: L. -> f*K.
.
Such a map induces a map
~(f,r
~ ( L . ) --~ ~(K.)
,
and it is these maps which induce the map of pro-objects Theorem let that
T-F
^
(12.5):
Let
~
denote ~-completion. C,D
class of groups,
With the above notation,
are of local ~-dimension < d
is a ~-isomorphism. Proof:
be a complete
Then
for some
~-~f
Our problem is the following:
~-Kf and
suppose
d , and that
is an isomorphism. Let
W E ~o
be a
.
- 134
-
12.6
"test object," and suppose given a map
b: ~ - [ D --> W .
to show that it factors uniquely through
7--]-f .
We have
Thus we need
to show: (i)
There is an f-map
$: L. --> f*K.
of hypercoverings,
and a homotopy commutative diagram
~(L. )
=(f,r
,
> =(K.)
(m.6) W
in
~o 9 such that (ii)
Given
8
represents
$: L. --> f*K.
is a "refinement" of
b . and two maps
mi
as above, there
~ , i.e., a homotopy commutative diagram
L|,
~,
> f~K|
L.
~
> f~.
such that the pull-back of
~i
to
~(K.)
are equal in
These two problems can be treated analogously.
~'o "
We give a
detailed treatment of (i) in steps (a)-(c) below, leaving (ii)
iS5
12.7
-
to the reader. We shall make repeated use of the "elementary construction" of (4.9), which replaces homotopy-commutative diagrams by actually commutative ones, after a choice of homotopy-commutativity. Our obstruction theory argument below is, in fact, quite similar to the argument of (4.9), except that we must now be a bit more delicate in our method of killing obstructions, to be able to pass to a limit.
so as
In fact we only see how to do
this by an inductive process in the skeleta of
~(L.),
than, as in (4.9), by induction on the coskeleta of
rather
W .
In the special context of schemes, our technique pecially the somewhat convoluted manner of expressing
(and es(12.7))
m~ght be rendered superfluous by a rigidification of the category of hypercoverings.
(a)
Killing Relative Cohomology Classes by Refinement.
We adopt the following conventions:
B
will refer to the
CW-complex which is the realization of the simplicial set ~(L. ) ;
A
will refer to the CW-complex mapping cylinder of
the geometric realization of inclusion complex of
~(~) .
Thus we have the natural
~: B--> A , allowing us to identify A .
If
L! ~ >
f'K|
B
with a sub-
is another morphism of hyper-
coverings, we refer to the CW-pair associated to it by the symbols:
B' ~I> A' .
Lemma (12.7):
Under the hypotheses of (12.5), suppose
-
x E Hq+I(A,B;G) ~-groups.
where
G
136
12.8
-
is a ~-twisted system of abellan
There is a refinement of
~ :
Lr:
> f*K':
~"
L.
> f*K.
which is a homotopy-commutatlve diagram of split hypercoverlngs such that: ; K"Is - K.16
; #"18 - #18
(i)
L"I8 -- L.16
(2)
L" , K'.w
(3)
There is a homotopy-commutativlty rel 6 , F , of the
tl B fl
> A"
B
> A
such that the induced map,
Hq+I(A,B;G) --> Hq+I(A",B '';G)
x
to zero.
8
=
q - d - i
have all objects of d~-dlmenslon < d .
diagram,
sends
where
.
-
Proof:
Hq(A)
12.9
-
Consider the diagram:
-->
Hq(B)
II H q (wK.)
137
,>
> Hq+l(A)
Hq+Z(B,A)
II >
x
in
K' --> K.
L ~q+l(II
c,G)
H q + l ~ - I C,G)
equal to
Then we have a map
(8.10), there is an of this map
K|
to
K.
with
is zero,
in dimensions <_
L|/a = L./a
and by
and an extension
L' --> f*K' .
We obtain a diagram
BI
--.-~~>
At
(~') B
~
>A
which is commutative in dimensions <_ a . L'
in dlmenslons > a-1
conmratatlvity tel(a-l)
By (8.11) we can change
to make the truncated commutative diagram
extend to a homotopy-commutative for
square.
(~') .
Then
Let F'
F'
)
L ,o.~>Hq+~IVD,o)
Hq+l(~K'.,G)
L./a--> f*KI./a = f*K./a
L' --> L.
> Hq§
is zero, hence there is a
such that the image in
and by (8.18) we m a y take a = q-d .
II
Hq+l(~K. )
~qgl-T c,a) i> ~q(I-TD, Q)
refinement
Hq+l(B)
II
H q (TtL.)
The image of
-->
be a homotopy-
induces a map
138
Hq+I(A,B;G) --> Hq+I(A',B';G) image of a class
such that the image of Hq(A",G) (8.18), ment
.
x
l0
is the
We now make a second refinement,
r,
>
f*~:
........
>
f'K|
in
Hq(B",G)
,,
y
12.
such that the image of
y 6 Hq(B ',G) .
T.'
-
is in the image of
This may be done, by our assumptions,
and invoking
(8.11) a second time, we may obtain this with a refine-
$"
which when restricted
$'/6 , and such that
(~")
to 6-skeleta is isomorphic
is homotopy-commutative
ly, for any homotopy-commutativity
F"
rel 6 9
to Clear-
of (~") the induced map
Hq+I(A',B ' ;G) --> Hq+l(A",B ";G)
sends
By = x'
to zero.
To conclude the proof of (12.7) we need
only take for (~) the "composite"
,
L .v .
.
.
.
of the diagrams
>
f*~:
>
f*K~
(~') and (~")
139
-
and take
F
to be
by "composing" F'
the
-
12. ii
homotopy-commutativity rel 8 , obtained
(in the evident sense) the homotopy-commutativity
of (~') with any homotopy-commutativity rel 6 , F" (b)
of
).
The Inductiv e Step.
Let
A,B
be as in (a)
above. Suppose we are given
a
diagram 9o,
B
> A
W
where
W 6 ~o
" such that for a given integer
q > 0 , there is
a homotopy-commutatlve triangle,
Sk qB
-
> Sk qA
W
Lemma 12.8:
There is a refinement
$'
of
~
such that
(i)
r
= @/q-d-i
(2)
LI.,Kt. have all objects of ~-dimenslon < d .
(3)
There is a homotopy-commutative diagram,
Skq+lBt
.> Skq+lAI
Skq+l~
~ q + l W
(~q+l)
140-
where
12.12
m~+l/q-d-i = mq/q-d-i . Proof:
since
Skq~
We may assume that
(~q)
is actually-commutative,
is an inclusion as a subcomplex
([29], Chap. 7.
sec. 6, thm. 12). Let Skq+ 1
Oq+ 1
denote the obstruction to finding an extension to
agreeing with
(~q)
on
a relative cohomology group,
Skq_ 1 .
The obstruction lles in
Hq+l(A,B;G)
coefficient group locally isomorphic to induced by the action of
~l(W)
on
where
G
is the twisted
~q+l(W) , the twist being
~q+l(W)
, whence it is a
~-twisted abelian C-group. We apply (12.?) with ~': L! --> f*Kl
x = Oq+ 1 , to get a refinement
and an actually-commutative
BI
~'
B
~
diagram
> AI
,.> A
satisfying the conclusions of (12.?).
Since
realization of a morphlsm of simpllcial sets, and
~'
cellular.
are clearly cellular.
Hence
tF
We may then consider the diagram
s
is the geometric s
is cellular;
may be assumed (Skq~F)
to obtain
141
SkqB I
-
]2.]3
> SkqA'
1
,[
Sk q ]3
> Sk qA
W
The above being actually commutative gives us an extension,
8~: B t U SkqA t
of
8' , whose obstruction,
>
W
o~+ 1 , is the image of
Oq+ 1
the mapping on relative cohomology induced by
(~F) .
our very choice of refinement
o'q+l = 0
sequently,
Bq_ v I
yields our diagram
(12.7) we have:
may be extended to (~+l)
~q+l ' "
via
Thus, by Con-
This latter map
, and the three properties asserted
in (12.8) come from the similarly numbered conclusions of (12.7). q.e.d. (c)
The Proof of (i):
We find a preliminary refinement
L.(~) ~ (~ f*K.(~)
that the factorization below may be found:
l--Fl(A(~
, ,
T It(w)
> l--[l(B(~
such
142
12.1#
A This is possible since Then for (D1) .
L.(~
r
~l(W)
f.K.(O)
6 @
and
~ f
is a
~-isomorphism.
we may find a diagram of the form
Applying our inductive step (c) we obtain successive re-
finements,
~)(q): L.(q)., > f*K.(q) which admit extensions,
(~+i)
'
for all
q >_ 1
.
Since
(~(q)/q- (d+l) = (~(q+l)/q_ (d+l) we may define the refinement ~(~)/n = $(q)/n
for any
~ ('): L.(")
q l n+d+l .
> f*K.(")
by
Thus admits an extension
of the form (12.6), concluding our proof of (i). Theorem Let
X
(12.9):
(generalized Riemann existence theorem)
be a connected, pointed prescheme of finite type over the
field of complex numbers. >
and
Then there is a canonical map
Xet , and it induces an isomorphism on proflnite Proof:
We introduce in addition to the sites
Cetal e
(9.4) also the site
spaces
X'
its image.
completions. Cclas s
X' --> X
x' 6 X w
(12.2)
whose objects are topological
lying over the underlying topological space of
such that the map every point
C'
e: Xcl
is a local isomorphism,
X ,
i.e., that
has a neighborhood which is isomorphic to
Since an etale map of preschemes
X' --> X
is a local
isomorphism on underlying spaces, and since an open set lles in C t , we have morphlsms of sites
143 -
12.15
C!
Cclass
Cetale
Now it is clear from the definition of local isomorphism that every hypercovering of
Ct
is dominated by a hypercoverlng of
Cclas s .
~7
C' --> Xcl
and so
Thus the map e
yields the map
r .
is a homotopy equivalence,
By the discussion of section lO
and the comparison theorems of ([3], Xl, XVI, 4), this map
r
induces an isomorphism on cohomology with twisted finite coefA
ficients, and on
91 ( ^ =
is a ~-isomorphism,
profinite completion).
by (4.3).
Therefore
A e
Now apply Theorem (12.5) and
([3] X, 4.3) to obtain an actual isomorphism. Corolla r~ (!2.10): addition that
X
With the notation of (12.9), suppose in
is geometrically unibrauch.
morphic to the profinite completion of
Then
Xet
is iso-
Xcl .
Apply (ll.1). The same considerations as used for (12.9), together with assorted comparison results from ([S], XVI) yield the following re suit s: Corollary (12.11) :
Let
k
be a field admitting two imbed-
dings el" r
k -->
into the complex numbers, and let type over
k , Xi
the schemes over
X
be a pointed scheme of finite C
obtained by the imbeddings
i~4 -
ei , i = 1,2 .
12.16
Then
XI, class ~ X2, class A
where
denotes profinite
Corollary Spec(k)
, where
and let
K m k
XK
(12.32): k
completion.
Let
X
be a proper pointed
is a separably algebraically
closed field,
be another separably closed field.
the scheme obtained from
X
scheme over
Denote by
by extension of scalars.
Then
the canonical map
A (XK) et -> Xet
is an isomorphism, Corollary
(12.13):
separably algebraically Spec(R)
A
where
denotes profinite
Let
R
completion.
be a discrete valuation
closed residue field
k , and let
ring with f: X -->
be a smooth proper scheme with connected geometric fibres
Xo,X 1 , both being assumed pointed compatibly with a chosen section of
X/Spec R .
Then there is a canonical isomorphism
XI, et ~
Xo, et
^ where
denotes completion with respect to the class of finite
groups of order prime to the characteristic In what follows, and
p~
let
Up
p
of
k .
denote the class of finite p-groups,
the class of finite groups whose order is not divisible
-
by
p .
If
Z
145
-
12.17
is in pro-K o , we let
pletion with respect to the class p@ A Let denote profinlte completion. Corollary (12.14):
Let
D
pZ
and
and
@p
Z denote comP respectively.
be a dedekind domain, and con-
sider a situation as follows
V
> Spec D Q
where
y
is a geometric point with residue field
the generic point, and where characteristics
p ~ q
proper scheme over Xy, cl
P,Q
C
lying over
are two geometric points of
respectively.
Let
X
be a smooth,
Spec D , with a given section.
is connected and simply connected.
Assume that
Then
A p(Xp, et ) x (XQ, et) p ~ X y, cl
Proof:
We apply (12.12) and (12.9) at the local rings of
the two points
P,Q
respectively,
p(Xp, cl) ~ p(Xy, cl)
,
(XQ,et) p ~ (Xy, cl) p
Thus it suffices to show that if element of
to conclude that
Z
is any simply connected
Ko " then A Z ~ pZXZp
9
.
- 146
In fact, if
Z
is in
-
12.18
the class of finite groups,
@Ko ' @
it is easily s e e n that in fact
Z
Z = p Z X Zp
itself.
of factors in
from a consideration construction
of
Z
@Ko
decomposes
X'
(12.15):
and
elements Z
By the
Z , it is a pro-object of
@Ko ' and hence
splits too.
With the notation of (12.13),
is another scheme satisfying
Xp ~ X'p
This follows for example
for a simply connected
split in the canonical way, whence Corollar~
as a product
of the Postnikov decomposition.
whose members are simply connected
XQ ~ X~ , then
then
the hypotheses
Xy, cl ~ X'y, cl
"
of
X .
suppose If
A.l.1 - 147
APPENDIX
w
Limits.
We recall some facts about limits.
Since the results are
well known and/or routine, we will omit most of the proofs. will work in a fixed universe
U
([3], VI),
gories are assumed to be U-categories, small, i.e., a set of Let
I
meaning),
U
so that all cate-
but not necessarily U-
([3], VI, 2.1).
be an "index" category
(the word index has the empty
and
(1.1)
X: I --> C
a functor, where Xi E C
C
the value of
is another category. X
on
i .
For
i E I
denote b y
A direct limit
lim X = lim X. i is an object of i 6 I
We
C
together with maps
such that for every map
i--> i'
diagram
Xi
"> X i ,
lim X -@
X i --> lim X -@
in
I
the
for each
resulting
- 148-
commutes,
and such that
these properties. morphism,
lim X
Clearly
if it exists.
A.I.2
is universal with respect to
lim X
is unique up to unique iso-
It does always exist if
stance the category of sets, groups,
C
is for in-
or abelian groups and if
I
is a U-small category. The inverse limit
lira X
is defined in an analogous way
([3], A category (1.2) (a)
I
will be called filtering if
Every pair
of objects of
i,i'
I
can be embedded
in a diagram
i i' ~
(b) (essential uniqueness of maps in
I , there is a map
posed maps
i
When (1.2),
then
I
.... > i"
i"
of maps) i' --> i"
set of equivalence
i
, > i'
is a pair
such that the two com-
are equal.
is a U-small filtering llm X
If
category,
and
C = (sets) in
can be described in the familiar way as the classes in
U Xi
for the following equivalence
relation: (1.3)
Let
m 6 X i 9 m' E X i, 9
a diagram
i'
Then
iff there is
- 149
such that the images of
,~ I
A. 1.3
-
in
under the induced maps
X i,,
are equal. For a U-small
(1.4)
I , the functor
lim: Hom(I, (sets)) --> (sets) -9
always
commutes with arbitrary direct limits.
If
I
is filter-
ing, it is also left exact in the sense that it commutes with finite inverse limits,
i.e.,
it preserves monomorphisms
and com-
mutes with finite fibred products. If
X: I-->
(grps)
(or of abelian groups),
is a U-small filtering then the set-theoretic
system of groups limit has an
(abelian) group structure making it into the direct limit in the category of
(ab)
groups.
The functor
lim
is an exact functor
on abelian groups. Now let
I ~> J
We will call (1.5)(a)
be a functor,
if
For every
J E J , there is an
in
I
are two maps in
such that the composed maps
i E I
If
j E J , and
easily that
$
and a map
i E I , and
J , there is a map J ~>> ~(i')
The condition to be coflnal is transitive. verifies
filtering.
#(1) .
(essential uniqueness) > $(i)
I
cofinal if it has the following properties:
j->
(b)
and suppose
cofinal implies
J
i--> i'
are equal. Moreover,
filtering
one
(where we
-
always assume as above that
A.I.#
1 5 0 -
I
is filtering in any case).
dition
(b) can be restated in several equivalent ways,
if
is filtering.
J
even if both
I
Example object
But note that
and
(1.6):
J
subcategory
Let
ff = f . I
(b) does not follow from
J
(a)
be the category consisting of one
f: J --> J Then
J
other than the identity.
is filtering.
consisting of the object
map is also filtering,
especially
are filtering:
j , with one map
Suppose that
Con-
j
The one point
and the identity
and the inclusion satisfies
(a) but not
(b). A functor
X: J-->
a projection map
f
(sets)
from
X
is Just a set to a subset
X
together with
y c X .
Clearly
lim X = Y -@
J while lim Xll = X .
I (1.7)
One case in which
is filtering and that gory
I
I
(b) follows from
(a) is that
is a full subcategory of
J
The cate-
J .
is then automatically filtering.
Proposition ing, and let Denote by
X~
(1.8) :
Let
X: J --> (sets)
If-> J
be cofinal, with
be a functor.
the composed functor
i ~
nonical map lim X
I
--> lim X
J
Suppose Xr
.
I l,J
filterU-small.
Then the ca-
-
151
-
A.1.5
is biJective. Proof:
Apply (1.3).
The surjectlvity of the map follows
immediately from (l.5)(a). u' E X~(i, )
To prove inJectivlty,
and suppose their images are equal in
This means that there is a diagram in
let
~ 6 X$(i),
lim~ X .
J
(1.9)
r such that ~(i")
u = u'
in
(cf. (1.5)(a)).
that there are maps maps
Xj .
Choose
Changing
i--> i"
and
i" 6 I
i"
~(i'" )
i I --> i" .
reasoning to
Then we obtain two
By (1.5)(b), --> such that the composed maps ~(i) _>
i" --> i"'
are equal.
J -->
if necessary, we may assume
~(i) -> $(i") , one factoring through
there is a map
and a map
Replacing
i"
by
i'"
J .
and applying the same
i' , it follows that there exists a diagram
i
in
I
whose image in
J
fits into a commutative diagram
)
152
with (1.9).
Hence the images of
hence equal in
J
Let
in
J
m,~'
I
be the category of maps
f--> g
X@(i,,) '
be a filtering category, and i f> i'
f
in
I , where a morphism
> i'
i"
I .
are equal in
is a commutative diagram
i
in
A.I.6
lira X
Proposition (1.10): let
-
> i "'
Then
Ca) (b)
J
is filtering.
The functors "domain" and "range" from
J
to
I ,
taking
i
f > il
~
i i'
are cofinal. (c)
The functor "identity map"
I --> J
taking-
i ~ i = i
is cofinal. Propositlon (i.ii): and let (a)
J
Let
I
be a U-small filtering category
be the category of maps of For fixed
I
above.
i E I , the sub-category
Ji
of
J
consisting
- 153
of maps from (b) functor
Let
i
to varying X:
J --> (sets)
I --> (sets)
-
i' E I
A.1.7
is filtering.
be a functor.
sending
,l I
and a canonical isomorphism
~ . (~m xl%) z> alm. x . i i Ji J
There is a natural
154-
A.2.1
w
Pro-objects and pro-representable
Let
C
A pro-object
be a category.
(contravariant)
functors.
[15]
in
C
is a
to
C .
functor
X:
I ~
--> C
from some U-small filtering index category
I
We will
often use the notation
x = [x i]i
One thinks of a pro-object of
X
I
as an inverse
system of objects
C , and the point is that the pro-objects
of
C
can be made
in a good way into a category
pro-C
by the rule
(2.1)
Hom(X,Y) = lim(lim Hom(Xi,Yj))
j when
X = [Xi] i 6 I
reader to elucidate
and
i
Y = [YJ]J E J
' where we leave it to the
the maps involved in the directed
Note that the index categories is important to understand much more information
are not assumed equal.
systems. Also,
right away that the pro-object
than the inverse limit
lim X
it
contains
even if the
155 -
latter exists in
A.2.2
C , which we do not assume.
Clearly, a functor
F: C --> C'
induces in the obvious way
a functor
(2.2)
pro-F: pro-C--> pro-C'
C
The objects of
.
themselves are pro-objects if the index C
category is taken to be the one point category, and by (2.1), forms a full subcategory of pro-C .
If
X = IXi ] E
pro-C
and
Y E C , then
(2.3)
Hom(X,Y) = lim Hom(Xi,Y) -@ i
In this way every
X E pro-C
gives rise to a functor
Horn(X,. ): C --> (sets)
It can be shown ([25]) that morphisms between the functors associated to pro-objects elements of
Hom(X,Y)
equivalent via
X,Y
are in 1-1 correspondence with
defined as in (2.1).
(2.4) to a full subcategory of
The functors which are isomorphic to object
X
I--~> J
Horn(X,. )
is
Hom(C,(sets))
.
for some pro-
are called pro-representable functors.
Corollary let
Thus pro-C
(2.5):
Let
be a cofinal functor, with
Then the pro-object
be a pro-object and
X = [Xj]j E J
X~ = [ X ~ ( i ) ] i E i
I
filtering and U-small.
is isomorphic with
X .
- 156-
A.2.3
In fact, it follows from (2.1) that the functors represented by
X
and
X~ --> X
X$
are isomorphic.
Of course, the isomorphism
is given by the obvious element of
We will refer to the index category
X$
as obtained from I
via
X
lim(limw ~ Hom(X$(i),Xj)).
j i by "re-lndexlng" with
$ .
The conditions on a functor to be pro-representable are easily understood.
Recall that if
F: C--> (sets) is a functor,
then the morphisms of functors
Hom(Z,.) --> F
are in 1-1 correspondence with elements of seen.
Suppose we denote by
(2.6)
J
J = [(Z,~) I Z E C
F(Z).
This is easily
the category of pairs
and
~ E F(Z)]
,
where a morphism
(z,r
is a map
Z <-- Z'
is
The functor
~
.
-> (z',r
such that the induced image of
jo _> C
sending
z
~'
in
F(Z)
-
gives rise to a functor
1 5 7
Y: C - >
~(Y) =
A.2.4
-
by
(sets)
llm
Hom(Z,Y)
-@
.
J
The index category
J
is in general not a U-small category,
this limit is a priori in the next u~iverse gives a map
Hom(Z;.) --> F
U+ .
But since
as above, we obtain a map
and it is well known that this morphism is bijective particular that the limit can be taken in Proposition and let above.
J
(2.7):
(i)
Let
There is a ~kunctor
I ~> J
i ~
whe re
~i
This
is cofinal (ii)
J
is the image in (hence
Conversely,
let
(hence in
be a pro-object,
[(Z,~)I~ E Hom(X,Z)}
as
given by
(Xi,~i)
Hom(X,X i) J
of
idxi E Hom(Xi,X i) 9
is filtering).
F: C --> (sets)
be a functor and let
be the category of pairs defined as in (2.6).
pro-representable
~--> F ,
U).
X = [Xi} i E I
be the category of pairs
so
Then
F
is
iff there is a filtering U-small category
and a cofinal functor
I I> j .
pro- representable iff
J
Equivalently
(cf. [15]),
I F
is
is filtering and contains a U-small
cofinal sub-category. we omit this routine verification. Corollary
(2.8):
Suppose
c
is itself a u-small category
158-
A.2.5
which is closed under finite inverse limits, i.e., under finite products and finite fibred products.
Then a functor
F: C-->
(sets) is pro-representable iff it is left exact, i.e., commutes with finite inverse limits.
- 159
-
A.3.1
w
Morphisms of pro-objects.
Let
X = {Xi] i E I ' x = [Xj]j 6 J
be in pro-C .
If
I=J
and if i 6I
fi: Xi -> Yi
is a compatible system of maps, i.e., a morphism of functors f: X --> Y , then pro-objects
f
X --> Y
determines in an obvious way a morphism of which we denote by the same letter
f .
Of
course, a general morphism of pro-objects will not be of this form even if the index categories are equal.
However,
one can
put it into such a form by re-indexing as follows: A morphism of maps
f: X --> Y
{fj: X --> Yj]j 6 J
lim~ Hom(Xi,Yj) i
.
is by (2.1) a compatible collection " and each
fj
is an element of
We will say that a map
: X i --> Yj
represents
f
if the image of
~
in
lim_~Hom(X i,Yj) i This is the same as saying that the diagram in pro-C
iS
fj .
f
commutes.
A morphism
X
>
Xi
-> Yj
r --> r
Y
between maps representing
f
- 160 consists of a map
i --> i'
in
I
A.3.2
and a map
j --> j'
in
J
such that the diagram
I
Xil
> Yj,
Xi
.> Yj
commutes. Proposition ing a map
(3.1):
f: X --> Y
The category
M
of maps
~
represent-
of pro-objects is U-small and filterlng,
and the functors
M
,,,> I
sending
$: X i --> Yj ~
i
M
> J
sending
$: X i
j
and
> Yj ~
are cofinal. Corollary
(3.2):
A map
f: X --> Y
of pro-objects of
C
can be represented, up to isomorphism, by a (U-small) filtering inverse system of maps
[fi: Xi -> Yi]i 6 I ' i.e., by a pro-
object in the category of maps of
C .
The above results can be generalized as follows. Proppsition
(3.3):
("uniform approximation")
Let
A
be
a finite diagram with commutation relations, and suppose that
- 161 A
has no loops, i.e.,
that the beginning and end of a chain of
arrows are always distinct. the type of
Let
D
Remark
(3.4):
following type: .
X,Y
A
[Di]
Let
Let
F: C --> C'
f: F(X) --> Y
follows immediately from
A
be a functor,
C
which represent
tially to show that
i.e., of
D .
I
I
Xi
and
Then we may re-
so that
f
is repre-
fi: (xi) -> Yi"
I
This
of diagrams
D , i.e., D .
Di
each of whose maps
The problem is essen-
is filtering and has a U-small cofinal
and that for every object
that the
such that
(3.3).
represents the corresponding map of
subcategory,
C
X E pro-C
be a morphism.
(3.3) consider the category
in
There is a
is isomorphic to
by a single index category
To prove
of
One can also make related assertions of the
sented by a compatible system of maps
of type
to pro-C.
of diagrams of
{Di}i E I
the diagram in pro-C determined by
index
be a diagram in pro-C
A , i.e., a morphism of
filtering inverse sysiem
Y E pro-C'
A.3.3
X
of
D ,
X X
are coflnal among all maps of
~tXi~i E I r to objects
C . Induction on the number of vertices of
true for
(n-l)
vertices,
an "initial" v e r t e x ing to it.
v
and that
of
A
has
A . n
Suppose it
vertices.
Choose
A , i.e., one having no arrows lead-
This is possible since
A
has no loops.
A
leading out of
be the corresponding diagram in
pro-C .
Let
By induction,
{D~}s E J "
Let
D' D'
v
A'
be the diagram obtained from v .
by removing
Let
and all arrows
can be approximated uniformly,
X = [Xi) i E I
be the object of
D
say by
corresponding
'
162 -
to
v .
Let
K
pair of indices
is a map
be t h e c a t e g o r y o f o b j e c t s
(i,j)
made up out of
A.3.#
and a diagram
Dk
k
c o n s i s t i n g of a
representing
X.,D'. and some maps, where a morphism l D
(i,j) --> (i',j')
morphism of diagrams
D
k--> k'
such that the induced maps give a
Dk. --> D k .
It remains to show that
is filtering,
and that the obvious functors
are cofinal.
We leave this as an exercise for the reader.
Scholie
(3.5):
and
K
K --> I , K --> J
Suppose given a U-small filtering inverse
system of pairs of maps fi >
>Yi
Xi gi
Then by (2.1) the induced maps of pro-objects equal iff for each
i
the maps
equal, i.e., iff for each
i
X --> Yi
f,g: X --> Y
induced by
there is an
i'
are
f,g
are
such that the
composed maps fi X i, --> X i
> Yi gi
are equal.
In particular,
Then the maps is an
f: X --> Y
i --> i'
the zero map.
suppose
C
has a zero object
is the zero map iff for every
such that the composed map If we set
ject is the zero object
i--> i'
there is
Xi = Yi ' fi = id , we find that an ob(which is equivalent with the assertion
such that the structure map Xi, --> X i
is the zero map.
i
Xi' --> Xi --> Yi
that the identity map is the zero map) iff for every is an
0 .
i E I
there
163 -
w
Exactness properties of the pro-category.
Proposition and
C
(4.1):
any category.
limits.
Let
I
C
has finite direct
(inverse)
Then the functor
associating with an
--> pro-C
X: I ~ --> C
commutes with finite direct Proof: I ~ --> C .
be a U-small filtering category
Suppose that
Hom(I~
Let
We will denote by
X = [Xi} i E I
D
the corresponding pro-object
(inverse) limits.
D = [Di] i E I
making up the diagram Let
A.4.1
be a finite diagram of functors X~ =[X~i]i 6 I
the various objects
(we do not bother to label the maps).
be the object
lim D = [lim Di] i E I -@ -@
"
Then in
pro-C , Hom (X, Y) = llm lim Horn(Xi, Yj )
7 =
llm lim lim Hom -@
-@
J
i
@-
(X~i, Yj )
= lim lim lim Hom(X~i,Y ) @-
@-
commutes with finite inverse limits
--~
([3],
j
j
i
commutes with inverse limits
= llm Hom(X~,Y) @(Z
lira
because
lim (-.
-~
I, 2.8)
= lira lim lim Hom(X~i,Yj) @(-. -~ a
because
j
164 -
This shows that
X
A.#.2
has the property required of a direct limit
in the category pro-C .
The proof for an inverse l~m~t goes the
same way. Proposition (4.2):
If
C
is closed under finite direct
(inverse) limits, so is pro-C . Proof:
To show pro-C closed under arbitrary flnite direct
limits, it suffices to show it closed under coproducts and amalgamated coproducts
([3], I, 2)
Xa Y
and
X ~Z Y .
To show
for instance the second, we suppose given a diagram
X
in pro-C.
By (App. 3.3) we may represent it up to isomorphism
by a filtering system
z/xi i6I l ~
.
Yi
Now apply (#,i). Proposltion
(4,5):
If
c
is a U-small category which is
closed under finite inverse limits, then pro-C is closed under arbitrary (U-small) direct limits. Proof:
If
IX~]
is a system of objects of pro-C , the
165
-
A.#.3
-
functor lim (--
Horn(Xa, Y)
is a left exact functor from is left exact.
C
Y6C
to sets, since each
Hom(Xa,Y)
Hence the functor is pro-representable
Proposition
(4.4):
For any
(2.8).
C , pro-C is closed under
(U-small) filtering inverse limits. Proof:
Let
pro-objects of
[xJ]j 6 J
be a filtering inverse system of
C , and say
X j = [X~]
where
i 6 Ij .
be the category whose objects are pairs of indices j E J
and
of a map
i EIj j --> j'
, and where a map and a map
One verifies easily that
K
(J,i)
(j,i) --> (j',i')
X , --> X
Let
representing
K
with
consists
x J' _> x j.
is filtering (and U-small), and we
claim that the pro-object
is an inverse limit of the filtering system is
clear
that
if
Z 6 pro-C
IX j] .
, then an element
In fact, it
l i m Hom(Z,X j )
J determines a unique element of
Lim Hom(Z,X~)
, and conversely.
(j,i) Proposition gory.
from those of
C
Let
A
be an additive
Then pro-A is again additive Proof:
if
(#.5):
(abelian) cate-
(abelian).
The axioms of ([ll]) for pro-A follow immediately A
using (4.1),
(4.2).
Note that
is a U-small category, then pro-A is equivalent by (2.7)
-
with the category
Sex(A, (ab))
Proposition (4.6): monomorphism
Let
-
A.4.4
considered by Gabriel ([ll]). A
be an abelian category.
A
(epimorphism) in pro-A can be represented by an in-
verse system of monomorphisms Proof:
166
(epimorphisms).
In an abelian category, a monomorphism is a kernel
Of a map, and an epimorphism is a cokernel. finite limits, we can apply (4.1). pro-(grps) as an exercise.
Since these are
We leave the assertion for
-
167
-
R 1
REFERENCES
[i]
J. F. Adams, Stable homotopy theory , Lecture Notes in mathematics, No. 3, Springer 1966.
[2]
M. Artin, and B. Mazur, Homotopy of varieties in the etale topology, Proceedings of a conference on local fields, Springer 1967.
[B]
M. Artin, A. Grothendieck, and J.-L. Verdier, S6minaire de g~om~trie alg@briaue; Cohomologie etale des sch@mas,-Inst. Hautes Etudes Sei. (mimeographed notes) 1903-64.
[#]
H. Bass, K-theor~ and stable algebra, Pub. Math, Inst. Hautes Etudes Sc~.', No. 22, 1964.
[5]
H. Bass, M. Lazard, and J.-P. Serre, S ous-groupes d'indice fini dans SL(n, ~), Bull. Amer. Math. Sot., 70(196#), ' '
[6]
G. Bredon, Sheaf theory, McGraw-Hill 1967.
[7]
E. Brown, C ohomology 467-484.
[8]
P. Cartier, Structures simpliciales, Seminaire Bourbaki, expos6 199 (mimeographed notes), 1959-60.
[9]
M. Demazure and A. Grothendleck, S4minaire de geometrle algebrlque, Sch6masen groupes, Inst. Hautes Etudes Sci. (mimeographed notes) 1963-b4.
theories, Annals of Math. 75(1962)
/
[lO]
A. Dold and D. Puppe, The generalized ba r construction, Annales de i' Instltut de Fourier (1962).
[ll]
P. Gabriel, Des categories a_bellennes, Th~se, Paris, Gauthier-Vi~lars, 1962.
[12]
P. Gabriel and M. Zisman, Calculus of fractions and homotopy theory, Ergebnisse der Mathematik vol. 35, Springer 1967.
[13]
J. Giraud, Technique de descente , Bull. Soc. Math France, M4moire No. 2, 1964.
J
/
9
/
[l# ] R. Godement, Theorie des faisceaux, Hermann 1958. [15]
A. Grothendieck, Technique de descente et th6or~mes d' existence en ~om4trle a l~@~riq~ue, S6mlnaire Bourbaki, expos@s 190, 195, (mimeographed notes) 1959-60.
-
168-
R2
[16]
R. Hartshorne, Residues and duality, Lecture notes in mathematics, No. 20, Springer 1966.
[1T]
D. M. Kan, Abstract homotopy, I, II, III, IV, Proc. Nat. Acad. Sci. ~i(1955), 1092-1096, 42(1956), ~19-~21, 42(1956)
542-54~. [z8]
M . Kan, On C.S.S. complexes, Amer. J. Math. 79(1957) 449-476.
D.
[19]
D. M. Kan, On homotopy theory and C.S.S. groups, Annals of Math. 68(1958] 38-53.
[20]
S. Lojasewicz, Triangulation of semi-analytic sets, Annali della scuola Normale Superiore di Pisa Ser. III 1~(196#)
~49-474. [21]
~
[22]
J. W. Milnor, The geometric realization of a semi-simpllcial complex, Annals of Math., 65(1957) 357-362.
[23]
J. Moore, Semi-simplicial complexes and Postnikoff s[stems, Symposium International de Topologie" Algebralca
[24]
D. Puppe, On the formal structure of stable homotopy theory, Aarhus Colloquium in algebraic topology, i~52.
[25]
D. Quillen, Homotopical algebra, Lecture notes in mathematics No. # 3, Sp ridge r i~o7.
[26]
D. Quillen, Some remarks on etale homotopy theory and a con~ecture of Adams, Topology 7(i96~) iii-i16.
[27"]
J.-P. Serre, Cohomolo~ie galoisienne, Lecture notes in mathematics, No. 5, Springer 1965.
[28]
J.-P. Serre, Exemples de varletes ~rojectives conJugees non hom~omorphes, Comptes Rendues Acad. Scl. Paris 256(196~)
Lubkin, On a conjecture of Andr~ Well, Amer. J. Math.
9
I
I
d
~19~-~196.
[29]
E. Spanler, Algebraic topology, McGraw-Hill 1966.
[30]
C. T. C. Wall, Finiteness conditions for CW-complexes, Annals of Math. 81(1965) 56-85.
[31]
G. W. Whitehead, Homotopy theory, M.I.T., 1966.
[32]
H. Zassenhaus, Neuer Bewels de Endlichkelt der Klassenzahl, Abh. Math. Sam. Univ. Hamburg 12(1956) 276-266.
169
-
R3
/
P
/
t'
EGA
J. Dieudonn~ and A. Grothendieck, Elements de geometrle alg6brlque, Pub. Math. Inst. Hautes Etudes Sci. Nos. 41960-.
SGA
A. Grothendleck, S6mlnaire de geometrie algebrlque, Inst. des Hautes Etudes Sci. (mimeographed notes) 1960-bl and 1961-62.
s
/
Offsetdruck: Julius Beltz, Weinheim/Bergstr.
/