C
AHIERS DE
TOPOLOGIE ET GÉOMÉTRIE DIFFÉRENTIELLE
CATÉGORIQUES
B.
DE
S
MIT
Measure characteristics of complexes
Cahiers de topologie et géométrie différentielle catégoriques, tome
37, n
o1 (1996), p. 3-20
<http://www.numdam.org/item?id=CTGDC_1996__37_1_3_0>
© Andrée C. Ehresmann et les auteurs, 1996, tous droits réservés. L’accès aux archives de la revue « Cahiers de topologie et géométrie différentielle catégoriques » implique l’accord avec les conditions générales d’utilisation (http://www.numdam.org/conditions). Toute utilisation commerciale ou impression systématique est constitutive d’une infraction pénale. Toute copie ou impression de ce fichier doit contenir la présente mention de copyright.
Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques
MEASURE
CHARACTERISTICS
OF
COMPLEXES
by
B. de SMITCAHIERS DE TOPOLOGIE ET
GEOMETRIE DIFFERENTIELLE CATEGORIQUES
Volume,UffllI-1 (1996)
Résumé. Nous donnons une version du lemme du
ser-pent
en théorie de la mesure des groupes abéliensloca-lement
compacts.
Cette version utilise la notiond’ho-momorphisme topologique
strict et la notion d’exac-titude de mesure. Grâce à celemme,
les méthodesd’algèbre homologique
peuvent
êtreappliquées
dans le contexte del’analyse harmonique
abstraite. On endéduit des résultats sur les
caractéristiques
descom-plexes
de groupes abéliens localementcompacts
avecdes mesures de Haar.
1. Introduction.
For a bounded
complex
of abelian groups with finitehomology
groupsthe Euler-Poincaré characteristic is the
alternating product
of the size of thehomology
groups. Theobject
of this paper is todevelop
ananalog
of the Euler-Poincaré characteristic forcomplexes
oflocally
compact
abelian groups with Haar measures. This entailsproving
ameasure-theoretic version of the snake lemma.
The main
applications
we have in mind are inalgebraic
numbertheory. Many important
invariants one associates to numberfields,
such as the idele class group, are
locally
compact
groups with anor-malization of the Haar measure. One often
manipulates
with suchgroups
by using
exact sequences, and tokeep
track of whathappens
to the measures one needs a
theory
of measure characteristics.Re-cently
there has been more interest in the Arakelov class group[8,
III,
(1.10)]
of a number field.By using
volume characteristics ofcomplexes
of such groups one can prove class number relations[4].
All results in
4-Under the
assumption
that all groups are countable atinfinity,
Oesterlé defines a volume characteristic of bounded
complexes
withfinite
homology
groups[9].
By
dropping
theseassumptions
we can runinto certain anomalies of a
purely topological
nature,
namely
non-stricthomomorphisms.
We will address this in the nextsection,
and we willneed some
arguments
concerning
strictness that goslightly beyond
Bourbaki
[3,
chap. III,
§2.8].
In section 3 the measure-theoretic snake lemma is
proved by
apurely category-theoretic
reduction to a much easier case of a 3 x 3diagram
(a
short exact sequence of short exactséquences)
-In section 4 we define measure characteristics of
complexes
andshow that
they
aremultiplicative
over short measure exact sequencesof
complexes.
As aspecial
case we recover results usedby
Lang
in hisbook on Arakelov
geometry [7,
chap. V,
§2].
2. Strict
morphisms
oftopological
abelian groups.All
objects
in this section aretopological
abelian groups and amor-phism f :
A --+ B is a continuous grouphomomorphism.
We sayf
isstrict if the map from A onto its
image
I in B(with
relativetopology
fromB)
is an open map. In otherwords,
f
is strict if andonly
if thecontinuous
bijection
A/
Kerf
--+~ I
is ahomeomorphism.
Thisdefini-tion of strictness can also be found in Bourbaki
[3,
chap. III,
§2.8]
and in
[11,
exp.1,
§3.1].
One word of caution: acomposition
of strictmorphisms
need not bestrict,
as one can seeby considering
the mapsZqi
C R --+R/Z.
A sequence ofmorphisms
is said to be strict if allmorphisms
in the sequence are strict.A continuous
bijection
from acompact
to a Hausdorfftopological
space is a
homeomorphism,
so anymorphism
from acompact
groupto a Hausdorff group is strict. In the context of Oesterlé
[9]
oneonly
considers Hausdorff
locally
compact
abelian groups which arecount-able at
infinity,
i.e.,
a countable union ofcompact
subsets. One canshow that a
morphism
between such groups is strict if andonly
if it(2.1)
Lemma. bemorphisms.
Iff
is strict andsurjective
thenIf g
is strict andinjective
thenProof. For the first statement one notes that the open sets of B are
exactly
theimages
of open sets of A. The second statement follows from the fact that the mapf (A)g( f (A))
is ahomeomorphism. D
(2.2)
Proposition. Suppose
we have a commutativediagram
ofmor-phisms
with strict exact rows. Let s be the group
homomorphism
Kercp"
--+Coker
cp’
from the snakelemma;
see(3.5)
below. Then thefollowing
hold:
(1)
if s issurjective
and p isstrict,
thenp"
isstrict;
(2)
if s isinjective
and cp is strict thenSp’
isstrict;
(3)
if s is the zero map andp’
andp"
arestrict,
then cp is strict.Before
giving
theproof
of(2.2),
wepoint
out anexample
of adiagram
that shows how non-strict maps can occur in this context. All vertical
maps are
injective,
and one map is non-strict. The map s is the zeromap
(which
isinjective).
Proof of
(2.2).
Thehypothesis
that the rows are strict allows us toview the
injections
A’ -3 A and B’ - B as inclusions oftopological
6.
We first make the
following
remark: if X C Y C Z are inclusionsof
topological
groups, then the canonical mapY/X --> Z 1 X
is strict.One can see this
by
writing
thepreimage
U in Y of an open subset Uof with 0 open in Z. We then have l
so that U is the intersection of
Y/X
and theimage
of 0 inZ 1 X.
We first show
(1).
Thesurjectivity
of the snake map means thatB’ C
cp(A).
Since cp is assumed to bestrict,
the map A -->cp(A)/B’
isa strict
surjection. By
the remarkabove,
the mapp(A) /B’ - BIB’
is
strict,
and with(2.1)
we see that the map A -B/B’
is strict. Themap
B/B’ --+
B" is atopologically isomorphism
and the map A --> A"is strict and
surjective.
With(2.1)
it follows thatcp"
is strict.The
hypothesis
of(2)
implies
thatKer cp
C A’.
Again using
the remarkabove,
one sees that the mapA’/ Ker cp --+ A/ Ker cp
is strict.By applying
(2.1)
twice,
andusing
that cp isstrict,
we see that thecomposition
_is strict. Since B’ --+ B is a strict
injection,
a thirdapplication
of(2.1)
shows that
cp’
is strict. This shows(2).
For
(3)
we need to show that cp is open on itsimage,
so for an openneighborhood
U of 0 in A we want thatcp(U)
is aneighborhood
of 0in
cp(A).
LetUo
be an openneighborhood
of 0 in A withUo
+Uo
C U. Sincecp’
isstrict,
we havecp(A’
nUo)
=cp(A’)
n 0 for some openneighborhood
0 of 0 in B. LetOo
be an openneighborhood
of 0in B with
Oo - Oo
C O. Let the subset X ofcp(A)
be defined asthen we have
Thus,
we are done if we can show that X is open incp(A).
Denote themap A --> Ali
by f .
SinceUo n
cp-1 (O0)
is open inA,
and the mapf
Note that Since
p"
isstrict,
the setp"(V)
is openin
cp"(A").
Denoting
the mapp(A) - cp"(A")
by
g, we deduce thatg-’(V"(V»
= X +Kerg
is open incp(A).
By
somesimple diagram
chasing,
one checks that ourhypothesis
that s = 0 isequivalent
toKerg
=cp’(A’).
But we have X = X +cP’(A’),
so thisimplies
that Xis open in
cp(A).
D(2.3)
Proposition. Suppose
we have a commutativediagram
ofmor-phisms
in which all rows and columns are exact. If five of the six exact
se-quences are strict then so is the sixth.
Proof. The case that the sixth sequence is the middle row or middle
column follows from
(2.2)
part (3).
So assume that the middle row and column are strict. It is clear
from
(2.1)
that the maps out of A’ and the maps into C" are strict.Consider the
following diagram
with strict exact rowswhere B’ x A has the
product topology,
andf
is the sum map. Thesnake map for this
diagram
is thesurjective
zero map, soby
(2.2) part
(1)
and(3)
the strictnessof f
isequivalent
to the strictness of the mapA -
B",
and, by
symmetry,
it is alsoequivalent
to themap B’ --+
CThe fact that there is at most one non-strict row or
column,
and(2.1)
nowimplies
that A --+ B’ or B’ -> C is strict. But thenf
isstrict,
so both A - B’ and B’ --+ C are strict.Using
(2.1)
again
itfollows that all
incoming
andoutgoing
maps of C’ and A" are stricthomomorphisms.
D(2.4)
Remark.Suppose
all maps in thediagram
in(2.3)
are strict.We view
A’,
A and B’ assubgroups
of B so that A" and C’ can beidentified with
subgroups
ofB/A’.
The mapf
in theproof
above isstrict,
and one can deduce that the canonical exact sequenceis strict.
3. Haar measures and the snake lemma.
Let G be a Hausdorff
locally
compact
topological
abelian group. AHaar measure on G is a translation invariant non-zero measure on
G,
for which all open sets aremeasurable,
and allcompact
sets have finite measure. IfF(G)
denotes the real vector space of real valuedcontinuous functions on G with
support
inside acompact
subset ofG,
then the Haar measure can be viewed as an R-linear map
F(G) -+
R
sending f
tofa
f (g)dg.
The Haar measure on G isunique
up tomultiplication by
apositive
real number. We refer to[5],
section 11and
15,
and to Bourbaki[2,
chap.
VII]
for theprecise
statements andproofs.
We say G is a measured group if G is a Hausdorff
locally
compact
topological
abelian groupequipped
with a choice of Haar measure.If G is a
compact
measured group then its volume is defined as the measure of the whole space,i.e.,
vol(G)
=fG
ldg.
A strictmorphism
of measured groups has a closed kernel andimage
[3,
chap. III,
§3.3].
(3.1)
Quotient
measures.Suppose
that H and G are measuredgroups, and that H is a closed
subgroup
of G. ThenG/H
is Hausdorffmeasure. A function
f
EF(G)
on G induces a functionf
onG/H
defined
by
where x is a
representative
of x in G. Note thatf (x)
does notdepend
on the choice of x as the measure on H is translation invariant.Further-more,
f
hassupport
inside acompact
subsetof G / H
sof
eF(G/H).
The
quotient
measure onG / H
is theunique
Haar measure onG / H
for which
(3.2)
Measure characteristic of short exact sequences.Suppose
we have a strict exact sequence of measured groups
(S)
The
isomorphism
G’---+~ Ker cp
is ahomeomorphism,
so we cangive
Ker p
the Haar measure of G’.Uniqueness
of the Haar measureim-plies
that thetopological isomorphism
G/
Ker p
--+~ G"
identifies thequotient
measure onG/
Ker p with c times the measure on G" for aunique
constant c eR>0 .
This constant c is called the measurechar-acteristic of
(S),
and we denote itby
k(S) .
Oesterlé[10]
callsc-’
the"Haar index." If
k(S)
= 1 then wesay
(S)
is a measure exact sequence.If G is
compact,
then so are G’ andG",
andThe
following proposition
resembles Oesterlé[10, A.4.2].
Oesterléim-poses
stronger
topological conditions,
but he also allows non-abelian
10-(3.3)
Proposition. Suppose
we have a commutativediagram
in which all rows and columns are strict short exact sequences of
mea-sured groups. Then we have
Proof. We will
identify
Ai, Ai
andA2
with theirimage
inA2.
The idea is tocompute
theintegral
of a functionf E
F(A2)
in two ways.For a e
A22
and(x, y)
EAi
xA13
weput
where ài
and y
are lifts inAi
andA2
of x and y. Notethat ga(x, y)
does not
depend
on the choice of the lifts. Now consider the strictexact sequence of
(2.4):
We have
ga(x, y)
=f (a
+ x+ y),
where/
EX(A22) /A11)
is the functionthat 9a E
F(A3
xA13).
The statements of(3.1)
now tell us that for u eA33
theintegral
where îi is any lift of u in
A22,
does notdepend
on the choice of û andthat we
have g
eF(A33).
We now rewrite the
integral fAz
f (a)da
by
firstusing
the middlecolumn,
and then the outer rows:By
Fubini’s theorem[5, (13.8)]
this isequal
toThe same is true with rows and columns switched.
By
choosing
afunction
f
whoseintegral
overA2
is not zero, weget
theequality
stated in
(3.3).
D(3.4)
Long
measure exact sequences.Suppose
we have a strictexact sequence
of measured groups, with almost all
Ai
equal
to the zero group ofvolume 1. For
each i,
choose a Haar measure on theimage
Bz
ofAi-1
in
Az
such that almost allBi
are the zero group of volume 1. We have
12-and we define the measure characteristic
K(A.)
to beThis does
not
depend
on the choice of measures on theBi,
becausechanging
the measure ofBi
by
a factor c ER>0 changes
bothK(Si)
andk(Si-1)
by
a factorc-1.
We say thatA,
is measure exact ifK(A.) =
1.It is easy to see that that
K(A.) =
lTi
vol(Ai)(-1)i
if allAi
arecompact.
(3.5)
The snake lemma. Webriefly
recall the snake lemma asgiven
in
Atiyah-Macdonald
[1,
prop.2.10];
see alsoLang
[6,
chap. II,
§9].
Suppose
we have a commutativediagram
of abelian groups, with exactrows and columns
The snake lemma asserts that we have a canonical exact sequence
where the "snake
map"
K3 - Ci
is defined as follows: take an elementof
K3,
find itsimage
inA3,
lift it toA2,
map it toB2,
lift toB1
and(3.6)
Theorem.(1)
Suppose
that the groups in thediagram
aretopological
groupsand that the maps are strict continuous
homomorphisms.
Thenthe maps in the snake sequence
(S)
are continuous and strict.(2)
Suppose
that thediagram
consists of measured groups and stricthomomorphisms,
and that the rows and columns are measureex-act. Then the snake sequence
(S)
is also measure exact.Proof. Note that for
continuity
of the mapK1--+ K2
we need strict-ness of the mapK2 - A2.
For any commutative square of continuoushomomorphisms
oftopological
groupswe have induced continuous maps Ker cp--+ Ker
cp’
and Coker cp--+Coker V’.
This showscontinuity
of all maps in(S) except
for the snakemap.
In a
purely category-theoretic
way one can build up the snakediagram
from fivediagrams
of thetype
considered in(2.3)
and(3.3).
Thus theproof will
be a reduction to(2.3)
and(3.3).
First fix some
notation: Ii
is theimage
ofAi
inBi
and for Xequal
to the letterK,
I orC,
we let KX be the kernel ofX2 -> X3,
and we let CX be the cokernel of
Xi - X2.
We can choosetopologies
(measures)
on these groups so that in thefollowing
twodiagrams
the
14-Let H be the cokernel of the map
K2 -
K3
andgive
it thetopology
(measure)
thatgives
thefollowing diagram
strict(measure exact)
rowsand columns:
We now deduce that the
top
row in the leftdiagram
below is strict(measure exact),
and that the bottom row in theright diagram
is strict(measure exact)
Our snake sequence now consists of four short strict
(measure exact)
sequences. This proves
(3.6).
D(3.7)
Remark.Suppose
that the rowsRA
andRB
and the columnsPl, P2
andP3
of the snakediagram
are strict exact sequences ofmea-sured groups, without
assuming
thatthey
are measure exact. In orderto consider measure characteristics we need to fix the
parity
of indicesand
Bl
for the rows and atKl
for the snake sequence(S).
Then onehas
To see this one can either go
through
theproof
of(3.6)
again,
or one can reduce to case weproved
already
by changing
the measures on thegroups in order to obtain measure exact rows and columns and
keeping
track of the effect of the measure
changes
on all measure characteristicsinvolved.
(3.8)
Remark.By
modifying
A,
andB3
one caneasily
generalize
this snake lemma to the
following diagram
where n 1 and m > 3. One obtains a snake sequence
Again,
strictness of thediagram implies
strictness of the snakese-quence. In the measure-theoretic
setting
one can formulate there-sult as follows: if we fix
indexing
parity
on the snake sequenceby
giving
Ki
index1,
then the characteristic of the snake sequence is
16-4. Measure characteristics on
complexes.
A
complex
is a collection of abelian groupsAi
with i eZ,
with mapsd2: Ai -+
Ai+1,
such thatdi+1 di
= 0 for all i.By
a measuredcomplex
wemean a
complex
A,
of measured groups for which alldi
are strict andalmost all
Ai
are the zero group of volume 1.Suppose
A,
is a measuredcomplex.
LetZi
be the kernel ofdi,
and letBi
be theimage
ofdi-le
The ithhomology
group ofA,
is defined to beHi(AO)
=Zif Bi,
andit is
again locally
compact.
We have strict exact sequencesIf we fix
indexing
in(S:)
by
Si
=Zi,
and choose Haar measures on allZi
andHi
(making
almost all of them zero groups of volume1)
then we can use(3.4)
to define the measure characteristic ofA.
to beNow
x(A,)
does notdepend
on the choice of measures onZi
but itdoes
depend
on the measures onHi (AO) .
(4.1)
Exact sequences ofcomplexes. Suppose
we have a short measure exact sequence of measuredcomplexes
0 -->A’O ->
A,
-AO" ->
0.By
this we mean that for each z we have a sequence 0-A’i---+
Ai -+
Ai’
--> 0 as in(3.2),
such that thediagram
(4.2)
Theorem. The sequence(HO)
is strict.Moreover,
if we fixin-dexing
parity by setting
H0
=Ho (Aé),
and we choose Haar measureson all
homology
groups(making
almost all of them zero groups ofvolume
1),
thenProof. The
proof
of(4.2)
isby
repeated
application
of the snake lemma. LetZi
be the kernel of themap Aj - Ai+1,
and defineZ’i,
andZr
similarly.
Also choose Haar measures on allZi,
Zi’
andZi",
giving
almost all of them volume 1. We showby
induction that for each i we have a strictlong
exact sequenceand we
keep
track of its measure characteristic. Forsufficiently large
i this sequence consists of
only
zero groups of volume 1 and forsuffi-ciently
small i it is identical to11..
The reader may finish theproof
18
(4.3)
Global measure characteristic. If thehomology
groups of ameasured
complex
A,
arecompact,
then we cangive
them measure1,
and thecorresponding
valueXgl(AO)
ofX(AO)
is called theglobal
measure characteristic. This characteristic is the best measure of
size,
in the sense that we have
if all
Ai
arecompact.
The theorem aboveimplies
that xgl ismultiplica-tive over short measure exact sequences of measured
complexes
withcompact
homology. Moreover,
strictness of thehomology
sequenceim-plies
that if two of the threecomplexes
in the theorem havecompact
homology,
then so does the third.In
Lang
[7,
chap. V,
§2]
this characteristic occurs in thefollowing
context. Let M be a
finitely generated
abelian group with agiven
Haar measure on M oz R(with
the Euclideantopology).
Giving
M thediscrete
topology
andcounting
measure, we can consider thecomplex
CM :
0 --+ M - M xz R--+ 0. The characteristicxgl (CM )
is the ratio of the covolume of theimage
of M in M x R and the order of the torsionsubgroup
of M.Lang
also showsadditivity
over short measureexact sequences.
(4.4)
Local measure characteristic. If thehomology
groups of ameasured
complex
A,
arediscrete,
then we cangive
them thecounting
measure, and the
corresponding
valueXloc(AO)
ofX(AO)
is called thelocal measure characteristic of
A, .
This characteristickeeps
track oflocal
blow-up
factors in the measure. Forinstance,
ifA1
=A2 =
R/Z
with
d1 (x)
= 2x and all otherAi
are the zerogroup of volume
1,
thenxgl (A. )
=1,
butxlo (AO )
= 2. Note thatXi.c is
multiplicative
overshort measure exact sequences of measured
complexes
with discretehomology.
Again,
strictness of thehomology
sequenceimplies
that iftwo of the three
complexes
in the theorem have discretéhomology,
then so does the third.where x# is the usual Euler-Poincare characteristic
given
by
(4.5)
Duality.
UnderPontrjagin
duality
[5,
chap.
VI],
compact
groupsare dual to discrete groups. The dual of a strict
morphism
isagain
strict
[11,
exp.11,
§6.1].
Moreover,
the dual of a measured group has a dual measure, which is the measure for which the Fourierinver-sion formula holds. The dual of the volume-one-measure on a
compact
group is the
counting
measure on its dual. It is not hard to see thatdualizing
a measured short exact sequence inverts the measurechar-acteristic.
Taking
the dual of a measuredcomplex
A withcompact
homology
groups andreplacing
the indices i in theresulting
sequenceby
-i,
we obtain a measuredcomplex
A
with discretehomology,
and we haveIf the
homology
groups are finite we deduce thatThis statement can also be found in Oesterlé
[9, §3].
References
1. M. F.
Atiyah,
I. G.Macdonald,
Introduction to commutativeal-gebra, Addison-Wesley, Reading,
Mass.,
1969.2. N.
Bourbaki, Intégration, Hermann,
Paris,
1963.3. N.
Bourbaki, Topologie générale,
Hermann,
Paris,
1979.4. B. de
Smit,
Class group relations and Galois modulestructure,
dissertation, University
of California atBerkeley,
1993.20
6. S.
Lang, Algebra,
3rded., Addison-Wesley, Reading, Mass.,
1993. 7. S.Lang,
Introduction to Arakelovtheory,
Springer-Verlag,
NewYork,
1988.8. J.
Neukirch, Algebraische
Zahlentheorie,
Springer-Verlag, Berlin,
1992.
9. J.
Oesterlé, Compatibilité
de la suite exacte de Poitou-Tate aux mesures deHaar,
Sém. Théor. NombresBordeaux,
1982-83 no.19.
10. J.
Oesterlé,
Nombres deTamagaywa,
Invent. Math. 78(1984),
13-88.11. G. Poitou
(ed.),
Cohomologie
galoisienne
des modules finis:sémi-naire de l’Institut de
Mathématiques
de Lille1962-1963,
Dunod,
Paris,
1967.Econometrisch Instituut
Erasmus Universiteit Rotterdam Postbus