R ENDICONTI
del
S EMINARIO M ATEMATICO
della
U NIVERSITÀ DI P ADOVA
C ARLO C ASOLO
Groups with finite conjugacy classes of subnormal subgroups
Rendiconti del Seminario Matematico della Università di Padova, tome 81 (1989), p. 107-149
<http://www.numdam.org/item?id=RSMUP_1989__81__107_0>
© Rendiconti del Seminario Matematico della Università di Padova, 1989, tous droits réservés.
L’accès aux archives de la revue « Rendiconti del Seminario Matematico della Università di Padova » (http://rendiconti.math.unipd.it/) 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 conte- nir la présente mention de copyright.
Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques
http://www.numdam.org/
Groups Conjugacy
of Subnormal Subgroups.
CARLO CASOLO
(*)
1. Introduction.
In papers
published
in 1954-55[10, 11],
B. H. Neumannproved, together
with many otherresults,
thefollowing
fundamental Theo- rems :1.1
I f
in a group G everysubgroup
has af inite
numbero f conju- gates,
then the centreZ(G)
hasf inite
index in G.1.2.
It
in a group G everysubgroup
hasf inite
index in its normalclosure,
then the derivedsubgroup
G’ isf inite.
Sometime
later,
such results were in some sensespecified by
I. D.Macdonald
[9],
as follows :1.3. There exist
f unctions
u,¡1 o f
N in1~T,
such that:(i) I f
G is a group in which everysubgroup
has at most mconjugates
then(ii) I f
G is a group in which everysubgroup
has index at most min its normal
closure,
then,u (m ) .
Subsequently,
the literature on theargument
has been enrichedby
several authors. In thepresent
paper, westudy
the case in which(*)
Indirizzo dell’A.:Dipartimento
di Matematica e Informatica, Univer- sità diUdine,
Via Zanon 6, 33100 Udine,Italy.
similar conditions are
imposed
not to allsubgroups
of a group, butjust
to subnormalsubgroups.
To this purpose let us introduce thefollowing
classes of groups.DEFINITIONS. Let G be a group, m a
positive integer.
Then:1)
G E T* if everysubnormal subgroup
of G has finite index in its normalclosure;
that is if oo for everyH sn G;
2)
ifIHG:H/::S m
for everyH sn G;
3)
V E V if every subnormalsubgroup
of G has a finite number ofconjugates;
that is if oo for every H snG;
4)
if for every H sn G.We will
occasionally
refer to two otherclasses, namely:
for every
H sn G;
for
every H
sn G.The
following diagram
illustrates the inclusions among these classes.Where all inclusions are
proper,
and no other inclusion holds.In fact:
(a)
Let ~’ be a group; If thenHence
Um
CTm
r1Ym ;
moreoverConversely,
suppose that = and r oo;if H =
... ,
H,.
are the distinctconjugates
of ~1 inG,
then:Thus
Vr C Unt
and T* r~ Y ~ U. In conclusion we have(b ) U ~ U Um :
Infact,
the infinite dihedral groupD~ _ ~x,
y ;meN
=
y2 = 2) is an U-group,
but it does notbelong
to anyUm,
n E N.
(c) U Vmg
T*. Forexample,
the standard wreathproduct
meN
wr
C9;
where is a Priifer group oftype p°°
andOf)
is acyclic
group of
order p, p
aprime, belongs
toYp
but not to T*.(d) U Tm rt
Y. The centralproduct
of an infinite number ofm E N
groups
isomorphic
to thequaternion
group of order8, belongs
toT2
but not to V.
Clearly,
everyT-group,
that is every group in which each sub- normalsubgroup
isnormal, belongs
to all the classes above defined.These classes can therefore be viewed as
generalizations
of the class ofT-groups. Also,
if we denoteby Bn
the class of groups in which every subnormalsubgroup
has defect at most n,Tn C Bn ,
for every n E N. Inclusion
Tn C B,~
isobvious,
andYn
SBn
followsfrom the fact that if H is a subnormal
subgroup
of a groupG,
andG =
Ho
R!:Hi
o ... ogn
= H is the normal closure series of g inG,
then
Na(Hn_1) .~.... ~ Na(H1)
= G(by contrast,
we observethat the infinite dihedral group
belongs
to U but not to the class ofgroups in which the intersection of any
family
of subnormalsubgroups
is
subnormal). Furthermore,
we recall that asubgroup
g of G is said to be almost normal if isfinite,
and almost subnormal if is finite for some subnormalsubgroup K
of G. Then T* isprecisely
the class of those groups in which every almost subnormalsubgroup
is almost normal.Indeed,
T* is the class of groups in which the relation of almostnormality
istransitive,
see[2].
We
finally
observe thatIT-groups,
that is groups in which every infinite subnormal is normal(see
De Giovanni and Franciosi[4])
areU groups (see
Heineken[6, Corollary 3]);
and that aspecial
subclassof has been
recently considered by
Heineken and Lennox in[7].
meN
In common with other
investigations
about the subnormal struc-ture of a group
(see
forinstance,
the treatment ofT-groups by
D. Robinson in
[12]),
it is reasonable to restrict to the soluble casethe classes under consideration. In this paper, we will
mainly
con-sider soluble groups
belonging
toT*, V,
andY~ (m e N).
Thereforewe will deal with non trivial abelian normal sections of a group
G,
over which acts
by conjugation.
Hence it is ofparticular
relevancethe
study
of the action of a group ofautomorphisms
T of an abeliangroup
~,
such thatjr:Nr(H)l
or is finite(and possibly bounded)
for everysubgroup
H of .A. This is theobject
of Section2,
from which we
quote,
as anexample,
asingle result, namely:
THEOREM 2.15.
If
A is an abelian group, andr-s. Aut (A)
suchthat m
for
every HA,
then there exists ac normal sub- groupof 7~
such normalizes everysubgroup of
A and the index¡r:r¡1 is finite
and boundedby a function of
m.On the basis of the results obtained in Section
2,
in Section 3 wefirst
study hyperabelian
p-groups in T* V Y(Theorem 3.2);
after-wards,
wegive
some structuralproperties
of soluble groups in T*and V. In
particular,
we prove(Corollaries 3,4
and3.7)
that a solubleV-group (respectively T*-group)
is metabelianby
finite(finite by metabelian). Recalling
that a solubleT-group
isalways
metabelian(Robinson [12,
Theorem2.3.1]),
onemight suspect
that every solubleV-group (or T*-group)
is a finite extension of aT-group (respectively
is finite
by T-group).
That this is not the case, not even for the smaller classU,
is shownby
someexamples
at the end of Section 3.In Section
4,
westudy
solubleV.-groups,
m e N. Our main resultis the
following.
THEOREM 4.8.
If
G is a solubleVm-group,
theny(m).
Where y
is a function of N initself,
andro(G),
the Wielandt sub- group of(~,
is the intersection of the normalizers of the subnormalsubgroups
of (~. It there follows that agroup G
has a bound on thenumber of
conjugates
of its subnormalsubgroups
if andonly
ifis finite. This result can be viewed as an
analogous
forsoluble groups, of the
quoted
Theorem 1.3(i)
of I. D.Macdonald,
with
ro(G)
instead ofZ(G). However,
thisanalogy
cannot be extendedto Theorem 1.1 of B.
Neumann;
infact,
in order to haveJ finite,
it is essential to assume thatI
is notonly finite,
butalso
bounded,
for every H sn G: if G is the infinitedihedral
group, then oo for everyH sn G,
but Less obvi-ously,
it is also notpossible
todrop
thehypothesis
ofsolubility
inTheorem 4.8.
Indeed,
there existlocally
finiteV,-groups (p
aprime)
in which the Wielandt
subgroup
has infiniteindex;
afamily
of themis constructed at the end of the paper.
NOTATIONS. Let G be a group, Then
Zn(G)
is the n-thterm of the upper central series of
G,
=U Zn ( G ) ;
is then E N
n-th term of the lower central series of
G, G N n y,,(G); n(G)
is theneN
set of
primes p,
such that G has at least one element of order p;is the Wielandt
subgroup
of G.Further,
we sayssna
that G is reduced if it does not admit non trivial normal divisible abelian
subgroups.
If G is an abelian group,by
rank of G we meanthe Prufer
rank,
that is the supremum among the cardinalities of the minimalgenerating
sets of thefinitely generated subgroups
ofG;
by
total rank of G we mean the sum of the ranks of all the distinctprimary components
of G and of thecardinality
of a maximal indi-pendent
subset of elements of infinite order of G.We denote
by
P the set ofprime numbers; P,
then n’=Let A be a
nilpotent
group, thenAn
is thea-component
ofA ;
as
usual, ~p~,
we writeAn
=A9
andA,,,
=A,,,.
If A is ap-group,
p E P,
andn e N,
weput
Let .1~ be a group of
operators
on the groupG,
and .HG;
wewrite
Nr(H) _
H" =H}
F(and Nr(x)
instead ofif
G) . Moreover,
weput
Hr =(HlX;
andHr=
ina E T
particular,
if .1~ = G in its actionby conjugation,
then Ha andga
are,
respectively,
the normal closure and the normal core of .H in G.Further,
Paut(G)
={x
E Aut(G);
.Ha= H,
for every HG)
is thegroup of power
automorphism
of G. We willfreely
use the fact that Paut(G)
is a normal abeliansubgroup
of Aut(G), and,
if G is abe-lian,
Paut(C~) ~ Z(A.ut (Q’-)) (see [3]
and[8]
for the relevant facts onpower
automorphism.s) .
For the
properties
of abelian groups that we willneed,
we refer tothe two volumes of
Fuchs [5] ;
with the notice that we use a directproduct » and
« cartesianproduct »
insteadof, respectively,
~cdirect
sum » and « direct
product »,
and that we denoteby
the Priifer group oftype p°°.
Thestandard
reference for solubleT-groups
ispaper
[12] by
D. Robinson.We will use without any further comment the obvious fact that subnormal
subgroups
andhomomorphic images
of a groupbelonging
to any of the classes under
consideration, belong
to the same class.2.
Automorphisms
of abelian groups.In this
section,
we collect some results on the action of those par- ticulartypes
ofautomorphisms
groups of abelian groups, which are relevant in oursubsequent
discussion of soluble groupsbelonging
tothe class
V, T*,
andYm (m
2.1. LEMMA. . Let N be a
periodic nilpotent
group, .1~ Aut(N).
(a) If
oofor
every HN,
then there exists afinite
set n
of prime numbers,
such that|r:Pautr (Nn’)| I
isf inite.
(b) If
there exists such that mfor
every x EN,
then there exists a
of prime numbers,
such thatini
m, and|r:Pautr (Nn’)| s
m.(c) If
oofor
everyH ~ N,
then there exists afinite
set n
of prime numbers,
such that r = Pautr(N 31’).
PROOF. In view of the
elementarity
of theseobservations,
weonly
prove(~b ) .
We argueby
induction on m. If m =1,
thenh =
Pautr (N)
and n =0.
Let m >1;
then there exists aprime
p, such that 1’ does not fix allcyclic subgroups
ofNf).
Let besuch that r = > 1 and write
r1
=Nr(x).
If y E sincex
and y
commute and havecoprime order,
we have:Hence,
forthat is:
By
inductivehypothesis,
there exists a set ofprimes
such thatInti
m - 1 andIFt:pautr1 [m/r]. Now, setting n
={p}
U ni ,we have
and, clearly,
so :The
following
isessentially Proposition
34.1 in Fuchs[5].
2.2. Let A be a reduced abelian p-group ; let B be a sub- group
of
A such thatAIB
is divisible. Assume that a is an automor-phism of
Aleaving
B invariant andacting
as a powerautomorphism
on it. Then ex is a powerr
automorphism of
A.PROOF. We observe that exp
(B)
= exp(A).
Infact,
if exp(B)
= pn, then B and soA /Dn (A)
isdivisible ;
since A isreduced,
We
get Qn(A)
=A,
that is exp(A)
=p n.
If exp(B)
= co,clearly
exp
(A)
= oo. We can therefore define a powerautomorphism
v of Aby putting,
for a EA, v(a)
=a’’x,
where v*~ is apositive integer
suchthat
a(b)
= bvx if b is an element of B of the same orderpk
of a.Now,
the kernel K of the
endomorphism
v - a of A contains B. SinceAIB
is
divisible
and A isreduced,
thisimplies
.K = A and so a = v is apower
automorphism
ofA, m
We discuss now the case in which r is a group of
automorphisms
of an abelian group
A,
andI
is finite for every .H A.The
description
of thegeneral
case ispreceded by
someparticular
cases.
2.3. LEMMA. Let A be a divisible abelian p-group, .1~ Aut
(A)
such that
C
oofor
every H A. ThenIr:Pautr (A) I
isfinite. If
m(m
EN) for
everyx E A,
then m.PROOF.
Suppose
that there exists a sequence xl , x2 , ... of elements ofA,
7 suchthat,
9 for all i: = xi for somepositive integer
ni, andNr(xi).
Then H = i EN)
is asubgroup
of A isomor-phic
to a Priifer group oftype p°° ;
whence andi EN
the index is
infinite, contradicting
ourhypotheses. Thus,
there exists an element such that
(If
forjust take xl E A
such thatI
is
maximal).
Put
then,
is finite(and
in thesecond
case).
We prove that Paut(A).
Let if then
1~1
normalizes~y? by (1). Otherwise,
let
x3 E A
such that= Xl.
By (1 ), 7B
normalizesx3~ (and x2~) .
Also == and so
1’1
normalizes Inparticular, rl
actson the group
Moreover, r1
induces on V a group of powerautomorphisms;
infact,
ifa, b
E Nand a #
0(mod pk) :
whence is normalized
by
Let now then:
and,
since x2 y is normalizedby r1:
By comparing (2)
and(3),
weget: x~ ~ yt = x2 y8,
henceThis
yields
t - s(mod pk)
and u+ I
= t(mod pk) ; together
witht - s
(mod pk),
we have u = 0(mod pk).
Thus:Let now y, E d such that = y,
where
Now,
for someThis proves that Paut
(A)..
2.4. THEOREM. Let A be a
periodic
abelian group, r Aut(A)
such that
Ir:Nr(H)1 for
every H A. Then isfinite.
PROOF. A is the direct
product
of itsprimary components Åp.
By
2.1(a),
there exists asubgroup
of finite index inF,
which actsas a group of power
automorphisms
on eachcomponent Åp, except
at most a finite number of them.
Since, clearly,
it is sufficient to prove the Theorem in the case in which A is a
p-group, for some
prime
p.1) I f
A isdivisible,
then weapply
Lemma 2.3.2)
Let A be reduced. In this case,by
Lemma2.2, Pautr (A) = - Pautr (B),
where B is a basicsubgroup
of A. Put =Nr(B),
then oo. Assume that =
Pautr, (M),
for somer1-inva-
riant
subgroup
M ofB,
of finite index in B. If Y is a transversal(i.e.
a set ofrepresentatives)
of if inB,
then isfinite,
say ofexponent pr;
then~Y~r~ S2r(B). Now, Ti
=Pautr1 implies
thatis
finite,
and sor2
=Or1(Qr(M))
r~ has finiteindex in
rl (and
therefore in1~) ;
moreover,r2
acts as a group of powerautomorphisms
on B(and, therefore,
onA).
Assume now,
by contradiction,
that|r:Pautr (B)|
I = 00;then, by
what we observed
above,
nosubgroup
of finite index ofh1
acts asa group of power
automorphisms
on asubgroup
of finite index of B.Let xl E B such that
gl
= =1= 9put Ki
= and let.~1
be asubgroup
of finite index inB,
such that.K1
~1Mi
= 1 and.~1
isr1-invariant (this
ispossible
because B is inparticular residually
finiteand,
if L is asubgroup
of finite index inB,
thenZr
hasagain
finiteindex in
B). Let x2 E
such thatHI; K2 = (hence
K2 Mi)
andH2 a r1-invariant subgroup
of finiteindex
in B suchthat
KI, K2>
andMi . Continuing
in this way, with the obviousnotation,
we obtain a sequence xi (i E N) of elements ofB,
such that if X =01531, ...),
then .for every iEN
(in fact, ... ~ r1 gl «xl)M1r.K1=
and,
t_ ~xl ~ ... ~ xi/ ~
so :Finally,
y weget
whichimplies, by
our choiceof the A contradiction.
3)
T hegeneral
case. Let D be the divisible radical ofA,
C acomplement
of D in A. Then C is reduced and is finite.Hence, by
the two cases discussedbefore,
we may assume thatD #
1 and T = Pautr(D)
= Pautr( C) .
If exp
(C)
isfinite,
saypn,
thenCr(C)
r1 has finite index in T and acts as a group of powerautomorphisms
on A.If exp
(C)
= oo, then(see
Fuchs[5, 35.4])
C admits a basic sub-C, and, therefore,
asubgroup
.g such thatSet
HJB
= EN)
withbo
= 1and,
if i >1, 5fl
=bi-l.
SinceB is a pure
subgroup
ofC,
we may select a set ofrepresentatives
~b~;
i E1~T~
of the cosetsbi,
in such a way thatIbil
=Ibil
=pi (thus
n B =
1)
for every i =0, 1,
.... Let K be asubgroup
ofD, isomorphic
tog = ai ;
withand,
fori > 1, a~
= ai-l. Consider thesubgroup
L =B,
of A. ThenFo
=Nr(L)
has finite index in F. We show thatr0 Pautr (A).
Observe first that L is
isomorphic
to.g,
which isreduced;
more-over,
Now, Fo
=Pautr0 (B), hence, by
Lemma2.2, Fo
=Pautr0 (L). Thus,
if g EFo,
then G induces a power onD,
0and
Z ;
hence for any there existpositive integers di,
such that:
Since ai E
D, bi
E C and D n C =1,
weget a’i6’
=ail,
andbY,
=b!’.
This shows that g E
To
induces the same powerti
on elements of D and C of the same orderp i.
This is true for every g ETo,
whencePautr (A ), concluding
theproof..
2.5. LEMMA. Let A be a torsion
free
abelian group, 1~’ s Aut(A)
such that C oo
for
every H A. Then T isfinite.
PROOF. Let x E
A, and g
E then(xg)-
=x~,
whence(x-1xg)m
= 1. Since A is torsionfree,
we have x =0153ø,
that is g ECr(x).
HenceCr(x)
=Cr(xm)
for every x E A and 0 0Assume, by contradiction,
that 1~’ is infinite. We construct a se-quence of elements and normal
subgroups
of finite index in.1~
f suchthat,
f for everyi = 1 f 2 ~ ...
f f~ yl ~ ... f yi ~ ^·
-v X ... X
yi~
is centralizedby Fi,
butyl,
7 7 Yi is not normalizedby
Weproceed by
induction on i. Let suchthat
Cr( y1 )
~ .F’. Weput r1
= and observethat,
sinceis
finite,
is also finite andtherefore,
isfinite.
Let
2,
and assume that we havealready
constructed yi, ... ,y; E A
andsatisfying
the desiredproperties.
Let ==
0 A(ri-l)’
I thenBi-l
isr-invariant, and,
since has finite index in1’, Bi-l =1= A ;
moreover,by
what observedabove,
is torsion free.Let 9
Eand y E A,
then[y,
g,g]
E[Bi-,, g]
=1;
onthe other
hand,
thereexists, by hypothesis,
a k E N such thatgk
ECr(y).
Since[y, g] and g commute,
weget [y, g~k
=[y, gk;
=1, yielding [y, g]
=1,
as A is torsion free. This holds for every y EA, hence g
ECr(A)
= 1. Thisimplies,
inparticular,
thatwould have order at most
two, and P
would be finite.Thus,
thereexists
yi E A,
such thatYi)
is not normalizedby -Pi-, (observe
that this
implies
that ...,Yi)
is not normalizedby Now, (Yi)
nBi-l
=1,
sinceAlBi-,
is torsionfree;
inparticular:
and so
We take now
.ha
=(Cri_1(yi)) ;
then andFi
is a normalsubgroup
of finite index in7~
moreover... , y z~
=B,.
Let now
~-I = yi;
thenI
is finiteby hypothesis;
thus there exists an index n e N such that ==
Now,
if = thenBn+l
isr-invariant,
and soFn+1Nr(H)
normalizes g n
But, by construction,
g nB,,+,
== ...,In
particular, Fn+1Nr(H)
normalizesYi, ..., contradicting
our choice of Y n+l . We therefore conclude that 1~ is
finite..
2.6. THEOREM..Let A be an abelian group, T Aut
(A).
ThenIr:Nr(H)I
oof or
every H Ai f
andonly if
oneof
thefollowing
holds:
2 )
Tor(A) (the
torsionsubgroup of A)
hasinfinite exponent,
andthere exists a
r-invariant, free subgroup of A, of f inite rank,
such thatA/.F’
isperiodic,
andI
and|r:Pautr (AI F) I
are bothfinite.
PROOF. Assume that oo for every H A. Let T = Tor
(.A). By
Theorem2.4, IF: Pautr (T)~ I
is finite. If exp(T)
is
finite,
say exp(T)
= r, thenCr(T) I
is finite.Moreover,
Thas,
in this case, a
complement
U inA,
which is torsion free.Since, by assumption,
00, it follows from Lemma 2.5 thatI
is finite. Hence:
and we are in case
1).
Let
exp (A)
= oo and assume that|r:Pautr (A)
= 00.Let:F A be
the set of non trivial free
subgroups
of finite rank ofA,
then,~ ~ ~ 0,
otherwise A is
periodic,
and so,by
Theorem2.4, Ir:Pautr (A)]
00.Assume that
~’1 1~2 ...
is an infinite chain of elements of Isuch that
Cr(.Fi)
for every i =1, 2,
... ;then,
if KFi,
iEN we have = 00, which is not
possible by
Lemma2.5,
sinceoo and K is torsion free. Hence there exists an I
such
that,
for any Y EITA
I and it is C =Cr(Fo)
=Cr(V).
In
particular,
ifyFo
is an element of infinite order in then(Fo, y~ E:FA
and so C =y~) Cr( y ) .
Hence C centralizes every element of A whose order moduloFo
is infinite. IfA/Fo
is notperiodic,
then A =x E A ;
= whence C centralizes.A,
con-tradicting
ourassumption (recall
that isfinite).
ThusA/Fo
isperiodic;
now, since.F’o
has a finite number ofconjugates
under1~,
I’ = is a r-invariant free
subgroup
of finite rank ofA,
andA/F
isperiodic. Finally, I
is finiteby
Lemma 2.5 and|r:Pautr (AIF) I
is finiteby
Theorem 2.4.Conversely,
let A be an abelian group, .l’ Aut(A).
If HA, clearly Nr(.H) > Pautr (A);
hence if1 )
holds thenIr:Nr(H)1
o0for
every H
A.Assume now that condition
2) holds,
with F a freesubgroup
offinite rank of A.
Firstly,
sinceIr:Or(F)1 I
andIr:Pautr (AIF) I
areboth
finite,
we may assume, with no loss ingenerality,
thatCr(F) _
==
Pautr
== .1~.Therefore,
ifA,
then HF and H n F are r-invariant. Inparticular .Hr
HF.Let be the torsion
subgroup
ofHF/H n I’,
thenBIH
n Fis
finite,
since F has finite rank.Now, A/I’
isperiodic,
soH/H
nF ~
is
periodic,
andRHIH
n F is the torsionsubgroup
ofHF/H
n F. Hence RH isr-invariant;
inparticular, Hr
.RH.Now,
J = is
finite,
whence isfinite,
say = r. It follows that Let now T be the tor- sion
subgroup
ofA ;
then ourhypotheses imply
that 1~ = Pautr(T),
whence T r1 H is r-invariant and so
Hr >
T r1 H.Moreover,
sinceF has finite rank and is
periodic, A/T
has finite rank.Now,
is a
homomorphic image
ofHIH
nT ~ HTIT
of finite expo-nent ;
we therefore conclude thatHfHr
is finite. Hence is finite and soNr(H) > Cr(HrlHr)
has finite index inh.
EXAMPLE 1. We show
that,
in thehypotheses
of Theorem2.6,
case
2)
canactually
occur,yet |r:Pautr (A)|
= 00. Let A = CX E,
where
C ~
for aprime
p, and g =x~
is acyclic
group of infi- nite order. Let h = Aut(A ),
where ot centralizes x and induceson C a power
automorphism
of infinite order(for
instance za = zP+l for everyz E C).
It is easy to check thatIr:Nr(H)1
oo for everyH A
(indeed
for every HA) ,
but no power of afixes every
subgroup
of A.We now turn to the case in which .1~ is a group of
automorphisms
of an abelian group
A,
such thatI
isfinite,
for every.H A.
Again,
wesplit
the discussion of thegeneral
case into a number ofsteps,
each of thosedealing
with aparticular
case.2.7. torsion
free
group, 1~ Aut(N)
such thatfor
every .HN,
then 1~ = Pautr(N).
Inparticular
1 ~ =
1 i f N is
notabelian, ;
and 1 ~ = 1 or _~a~,
where oc is the inver- sion map,if
N is abelian.PROOF. Let
1 =1= YEN,
and set .g =y?r. Then, by hypothesis,
I
is finite. Hence is finite. Now1 ~ y~K
is infi-nite
cyclic, put
=z>.
Assume that there exists t EI~,
such thatz’ ¥= z; then zt = z-1.
But,
for some n EN, tn
E~)
and so tn = ==
t-n, yielding
t2n =1;
since N is torsionfree,
it follows t= 1,
con-tradiction. Hence
z> -~5 Z(K)
and sois
finite.By
a Theoremof I. Schur
(see [13, 10.1.4]) ,
K’ isfinite;
that is g’ = 1.Thus,
.K isa
finitely generated
torsion free abelian group; since 00,we have that K is
cyclic,
whence K =y~, proving
that Paut(N).
The final claims
follow,
forinstance,
fromCooper [3, Corollary 4.2.3]. ·
2.8. LEMMA. be a divisible abelian group, .r Aut
(A)
suchthat oo
for
every H A. Then r~ Paut(A).
PROOF. We observe first that if .H is a divisible
subgroup
ofA,
then Hr = H. In
fact, if g E 1’,
thenbut H is divisble and so it does not admit any
subgroup
of finiteindex;
hence .H’ r1 gg =H, yielding
H = Hg. This is true for anyg E r,
whence H = Hr. Let now IfIxl
_. oo choose a sub-group g of
A, Q
and x E g.By
what observedabove,
g isr-invariant;
moreover, it is torsionfree, hence, by
Lemma2.7,
F =Pautr (K);
inparticular, x~r
=x>.
Otherwise x has finite order. Since the torsionsubgroup
of A is the directproduct
of itsprimary components,
we may assumethat ixi
=pn,
for some p e P and n E N. Then choose asubgroup
L * ofA,
with x E L.Again
Lr = L and so =
x~.
2.9. LEMMA. Zet .A be a redneed abelian p-group, .I’ Aut
(A)
such that oo
for
every .73 A. Then there exists afinite subgroup
Nof A,
such that 1~’ = Pautr(A).
PROOF.
Assume,
for themoment,
that A isresidually
finite.Sup-
pose
further, by contradiction,
that for any finite F-invariant sub- group N ofA,
there existsx E A,
such thatN, x~
is not r-invariant.By
induction on we construct a sequence of elements xi of A suchthat,
if ... ,x£~,
then r1 = 1 andxi?
is not
r-invariant,
for every(and .go
=1).
Letxl E A
be suchthat
(Ti)
is not r-invariant. Assume now that we havealready
found x1, ... , Xn-l
satisfying
the desiredproperties.
Then M = ==
... , is finite and so, since A is
residually finite,
thereexists B A of finite index in
~,
such that M n B = 1.Suppose that,
for eachy c- B, y, M)
is.I’-invariant;
then BM is T-invariant and 1’ = Pautr(BMIX);
now, BM has finite index in~, hence,
ifY is a transversal of B~ in
.A,
then Yr is finite and .R = isfinite,
BR and I’ =Pautr contradicting
our initial assump- tion.Therefore,
there exists x~. EB,
such thatM,
is not -P-inva-riant. We have also:
In this way we construct the infinite sequence ....
Let now then
[
is finite. Onthe other hand .
Hence,
there exists n EN,
such that .gr =
KK~’.
We considerby construction,
there exists9 E r
such thathence r =
1,
that is:against
the choice of g E 1~. This contradiction shows that there existsa finite r-invariant
subgroup N
of A such that r = PautrWe now go back to the
general
case; thus let ~ be reduced(and
not
necessarily residually finite),
and let B be a basicsubgroup
of A.Now,
Br is a finite extension ofB,
and B is a directproduct
ofcyclic
groups; hence Br is
residually
finite.By
the case discussedbefore,
there exists a finite r-invariant
subgroup N
ofBr,
such that1~’ = Pautr
Now,
since N is finite and A isreduced, A/N
is also reduced. Moreover
A/Br
is divisible.Then, by applying
Lemma
2.2,
we conclude that 1~ = Pautr(A/N).
We illustrate the
previous
Lemmaby
anexample,
in which weshow that it may
happen
that nosubgroup
of finite index of 1’ actsas a group of power
automorphisms
on A.EXAMPLE 2. Let A be an infinite
elementary
abelian3-group,
with
generators
x1, x2 , ....For j
=2, 3, ...
let a3 be the automor-phism
of A definedby:
Then the
subgroup
.,T’ _ =2, 3, ...)
of Aut(A)
is also an infi-nite
elementary
abelian3-group,
and it is clear that nosubgroup
of finite index of .1~ acts as a group of power
automorphisms (in
thiscase it would centralize every a E
A)
on A. Moreover[A, 7~]
=-
(zi)
=C~(1’),
whenceH, xl~
isF-invariant,
for everyH A,
so for every .H A
(by contrast,
if thenHr = 1 ).
We observe that the natural semidirectproduct
W = AX|r
is a
nilpotent 3-group
inT3 ,
such thatW~==~(1~)
has order 3.2.10. be an abelian p-group, D the divisible radical
of A,
and r~ Aut(A )
such thatIHr:HI
oofor
every H A. Then :1)
there exists af inite
F-invariant NA,
such that2)
D hasf inite rank,
exp(A/D)
isfinite,
1~ =Pautr (D)
andthere exists a
f inite
r-invariant NA,
such thatwhere pn = exp
(A/D).
PROOF. Let A
= D X R,
where D is thedivisible
radical ofA,
Iand .1~ is a
reduced complement
of D in A.Then, by
Lemma2.8,
r =
Pautr (D).
Put L =Rr,
then L is a finite extension of a reducedgroup, and so it is reduced.
If L is
finite, put
L = N and conclude with1);
in fact isdivisible,
soby
Lemma2.8,
1~’ acts as a group of powerautomorphisms
on it.
Assume now .L to be
infinite; then, by
Lemma2.9,
there existsa r-invariant finite
subgroup N
of.L,
such that 7’ =Pautr (L/N).
Observe
that,
sinceIL n DI
= isfinite,
we may choose N in such a way that N > L r1 D. If exp(L)
= oo, there exists a basicsubgroup HjN
ofLjN
withH =1= L (see Fuchs [5, 35.4]).
Then His r-invariant and is a non trivial divisible group; moreover
is divisible and so 1^’ =
Pautr Arguing
as in theproof
of Theorem 2.4
(case 3)),
we have r =Pautr (AjN).
Hence suppose that exp
(L)
= pn is finite. Put C =0,,,(A);
thenL
C and C is reduced.By 2.9,
there exists a finite r-invariant sub- group ~ ofC,
such that 1~ =Pautr (OjM).
Assume that.M ~ S21(D),
and let a E and b E D be such that
bpn-l
= a; then(b)
nr1 M =
1,
hence Mb is an element of order pn inNow,
racts as a group of power
automorphisms
both on and~
D/D
nM,
and .Mb is an element of orderp"
in as well as inMDIM.
Thus each g E r induces the same power inCIM
and inIt follows that 1~ acts as a group of power automor-
phisms
on = and this is case1). Otherwise,
M >
S2,(D); thus,
Mbeing finite,
it follows that D has finiterank,
and this is case
2).
EXAMPLE 3. Let A = D x 1~ be an abelian group, where
D ~
podd,
and D is an infiniteelementary
abelian p-group. Let 1~ = ~Aut
(A)
where a is theautomorphism
ofA,
which maps every element of D into itsinverse,
and fixes every element of R. Then for every HA,
but there exists nosubgroup N
ofA,
such that a acts as a power
automorphism
onIn the
example above,
it is easy to checkdirectly
thatlHr:Hl
is finite for every H
A,
but this moregenerally
follows from the fact that Lemma 2.10 can be inverted. This is itself aparticular aspect
of our next result.2.11. THEOREM. Let A be a
periodic
abelian group,(A).
Then
IHr:HI
oofor
every H Aif
andonly if
there existriant
subgroups N,
Dof A,
such that ND,
N isfinite, DIN
isdivisible
of f inite
totalrank,
1~ = Pautr(AID)
= Pautr(DIN) and, if
p E
a(DIN),
then thep-component of AID
hasfinite exponent.
PROOF. Assume first that is finite for every
By
2.1
(c),
there exists a finite set ofprimes
such that 1’’ =Pautr (A,,,).
Let 7&1 be the set of those
primes
p e n such that 1’ does not act as a group of powerautomorphisms
on anyquotient
of.A~
over a finiter-invariant
subgroup.
For each p E 7&, letNp
be a finite T-invariantsubgroup
ofA,,
such that1)
or2)
of Lemma 2.10 holds(where
N =
N,) according to p 0
~l or,respectively,
p E Let(where Dp
is the obvioussubgroup
ofA~ , according
to Lemma2.10).
Then
N
Dand,
since n isfinite,
N isfinite,
and Lemma 2.10 im-plies
thatDIN
has finite total rank.Moreover,
r = Pautr(A/D)=
= Pautr
(DIN) and,
if p then thep-component
ofAID
isAJ)/Al’
r1 D = and has finiteexponent.
Conversely,
letN,
D be F-invariantsubgroups
ofA, satisfying
the conditions of the
statement,
and let H be asubgroup
of A. Wewant to show that
lHr:Hl
is finite.Now,
since N isfinite,
== is finite and Hr
(HN) r
Thus if we prove that everysubgroup
of Acontaining N
has finite index in itsr-closure,
thenthe same is true for every
subgroup
of .A.. Hence we may assume N = 1.Observe now that D and HD are
r-invariant,
since r actsas a group of power
automorphisms
on bothD/N
and in par-ticular,
Hr HD.Let n1
=;r(D)
and =(H/H
n ==
(HDIH
n Then R is P-invariant andHD/.RD
is an,,-group; but the
exponent
of theii-component
of is finite(this
follows from ourhypotheses,
since has finite total rankso, in
particular,
isfinite). Thus,
theexponent
of is finite.This
implies
that exp is finiteand,
inparticular,
exp(Hr/H)
is finite. But
DID
is a divisible group of finite totalrank;
hencelHrIHI
is finite. Thiscompletes
theproof.
The next
result, together
with Theorem2.11, completely
describesthe action of r on the abelian group A.
2.12. THEOREM. Let A be an abelian group ; Aut
(A ).
Thenoo
for
every H Aif
andonly if
oneof
thefollowing
con-ditions holds.